# THE GATING AND MAINTENANCE OF SLEEP AND WAKE: NEW CIRCUITS AND INSIGHTS

EDITED BY : Michael Lazarus, Sakiko Honjoh, Kaspar Emanuel Vogt, Ada Eban-Rothschild, Yu Hayashi, Qinghua Liu and Takeshi Sakurai PUBLISHED IN : Frontiers in Neuroscience, Frontiers in Pharmacology, Frontiers in Cellular Neuroscience and Frontiers in Molecular Neuroscience

### Frontiers eBook Copyright Statement

The copyright in the text of individual articles in this eBook is the property of their respective authors or their respective institutions or funders. The copyright in graphics and images within each article may be subject to copyright of other parties. In both cases this is subject to a license granted to Frontiers. The compilation of articles constituting this eBook is the property of Frontiers.

Each article within this eBook, and the eBook itself, are published under the most recent version of the Creative Commons CC-BY licence. The version current at the date of publication of this eBook is CC-BY 4.0. If the CC-BY licence is updated, the licence granted by Frontiers is automatically updated to the new version.

When exercising any right under the CC-BY licence, Frontiers must be attributed as the original publisher of the article or eBook, as applicable.

Authors have the responsibility of ensuring that any graphics or other materials which are the property of others may be included in the CC-BY licence, but this should be checked before relying on the CC-BY licence to reproduce those materials. Any copyright notices relating to those materials must be complied with.

Copyright and source acknowledgement notices may not be removed and must be displayed in any copy, derivative work or partial copy which includes the elements in question.

All copyright, and all rights therein, are protected by national and international copyright laws. The above represents a summary only. For further information please read Frontiers' Conditions for Website Use and Copyright Statement, and the applicable CC-BY licence.

ISSN 1664-8714 ISBN 978-2-88966-061-2 DOI 10.3389/978-2-88966-061-2

### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# THE GATING AND MAINTENANCE OF SLEEP AND WAKE: NEW CIRCUITS AND INSIGHTS

Topic Editors:

Michael Lazarus, University of Tsukuba, Japan Sakiko Honjoh, University of Tsukuba, Japan Kaspar Emanuel Vogt, University of Tsukuba, Japan Ada Eban-Rothschild, University of Michigan, United States Yu Hayashi, University of Tsukuba, Japan Qinghua Liu, National Institute of Biological Sciences (NIBS), China Takeshi Sakurai, University of Tsukuba, Japan

Citation: Lazarus, M., Honjoh, S., Vogt, K. E., Eban-Rothschild, A., Hayashi, Y., Liu, Q., Sakurai, T., eds. (2020). The Gating and Maintenance of sleep and Wake: New Circuits and Insights. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88966-061-2

# Table of Contents

*05 Editorial: The Gating and Maintenance of Sleep and Wake: New Circuits and Insights*

Michael Lazarus, Yu Hayashi, Sakiko Honjoh, Kaspar E. Vogt, Ada Eban-Rothschild, Qinghua Liu and Takeshi Sakurai

*08 Ablation of Central Serotonergic Neurons Decreased REM Sleep and Attenuated Arousal Response*

Kanako Iwasaki, Haruna Komiya, Miyo Kakizaki, Chika Miyoshi, Manabu Abe, Kenji Sakimura, Hiromasa Funato and Masashi Yanagisawa

*21 The Leptomeninges Produce Prostaglandin D2 Involved in Sleep Regulation in Mice*

Yoan Cherasse, Kosuke Aritake, Yo Oishi, Mahesh K. Kaushik, Mustafa Korkutata and Yoshihiro Urade

*29 Whole-Brain Monosynaptic Afferent Projections to the Cholecystokinin Neurons of the Suprachiasmatic Nucleus*

Xiang-Shan Yuan, Hao-Hua Wei, Wei Xu, Lu Wang, Wei-Min Qu, Rui-Xi Li and Zhi-Li Huang


Hui Dong, Juan Wang, Yan-Fei Yang, Yan Shen, Wei-Min Qu and Zhi-Li Huang


Ze-Ka Chen, Xiang-Shan Yuan, Hui Dong, Yong-Fang Wu, Gui-Hai Chen, Miao He, Wei-Min Qu and Zhi-Li Huang


Michael Lazarus, Yo Oishi, Theresa E. Bjorness and Robert W. Greene

*188 The Transition Between Slow-Wave Sleep and REM Sleep Constitutes an Independent Sleep Stage Organized by Cholinergic Mechanisms in the Rostrodorsal Pontine Tegmentum*

Carlos Carrera-Cañas, Miguel Garzón and Isabel de Andrés

*201 Differential Role of Pontomedullary Glutamatergic Neuronal Populations in Sleep-Wake Control*

Evelyn T. M. Erickson, Loris L. Ferrari, Heinrich S. Gompf and Christelle Anaclet

*217 Effects of 3 Weeks of Water Immersion and Restraint Stress on Sleep in Mice*

Shinnosuke Yasugaki, Chih-Yao Liu, Mitsuaki Kashiwagi, Mika Kanuka, Takato Honda, Shingo Miyata, Masashi Yanagisawa and Yu Hayashi

*231 Dynamic Metabolic Changes in the Human Thalamus at the Transition From Waking to Sleep - Insights From Simultaneous Functional MR Spectroscopy and Polysomnography*

Mick Lehmann, Andreas Hock, Niklaus Zoelch, Hans-Peter Landolt and Erich Seifritz

*238 Transcriptome Analysis of Pineal Glands in the Mouse Model of Alzheimer's Disease*

Kwang Il Nam, Gwangho Yoon, Young-Kook Kim and Juhyun Song


Amisha A. Patel, Niall McAlinden, Keith Mathieson and Shuzo Sakata


Tingting Lou, Jing Ma, Zhiqiang Wang, Yuka Terakoshi, Chia-Ying Lee, Greg Asher, Liqin Cao, Zhiyu Chen, Katsuyasu Sakurai and Qinghua Liu

# Editorial: The Gating and Maintenance of Sleep and Wake: New Circuits and Insights

Michael Lazarus <sup>1</sup> \*, Yu Hayashi 1,2, Sakiko Honjoh<sup>1</sup> , Kaspar E. Vogt <sup>1</sup> , Ada Eban-Rothschild<sup>3</sup> , Qinghua Liu1,4,5 and Takeshi Sakurai <sup>1</sup>

*1 International Institute for Integrative Sleep Medicine (WPI-IIIS), University of Tsukuba, Tsukuba, Japan, <sup>2</sup> Graduate School of Medicine, Kyoto University, Kyoto, Japan, <sup>3</sup> Department of Psychology, University of Michigan, Ann Arbor, MI, United States, <sup>4</sup> National Institute of Biological Sciences, Beijing, China, <sup>5</sup> Tsinghua Institute of Multidisciplinary Biomedical Research, Tsinghua University, Beijing, China*

Keywords: sleep, sleep need, arousal, sleep-wake cycle, wakefulness

### **Editorial on the Research Topic**

### **The Gating and Maintenance of Sleep and Wake: New Circuits and Insights**

Sleep is highly conserved among all organisms with a nervous system, from worms to humans, and is vital for survival (Siegel, 2008; Trojanowski and Raizen, 2016). Humans spend one-third of their lives asleep, but due to work schedules and expectations, life-style choices, or medical conditions, many people experience a wide range of sleep disturbances (Kryger et al., 2017). Sleep disturbances may take a serious social and economic toll due to an increased prevalence of psychiatric illnesses, especially anxiety and mood disorders, decreased economic productivity, and a strong link to traffic and work-related accidents. Insufficient sleep is also an established risk factor for obesity, diabetes, heart disease, and other lifestyle diseases, as well as infectious diseases and cancers.

The governing principles of sleep remain one of the biggest black boxes of neuroscience. Therefore, there is increasing urgency to gain knowledge of the neural mechanisms and molecular substrates that regulate sleep and to reveal the causal links between sleep and health. New opto-/chemo-genetic-based tools and tracing methods for neural circuits have enabled unprecedented investigation of discrete circuit elements (e.g., transmitters, pathways, cell populations) and identification of genes and signaling pathways that regulate sleep/wake behavior (Saper et al., 2010; Saper and Fuller, 2017). As a result, new cellular and molecular targets for treating sleep disorders have been identified (Saper and Scammell, 2013). In this Research Topic, we present up-to-date original articles and reviews that demonstrate the versatility and power of state-of-the-art tools in advancing our knowledge of the regulation and function of sleep.

The neurotransmitter serotonin is believed to play an important role in sleep/wake regulation, but the underlying molecular mechanisms and brain circuits remain unclear. Iwasaki et al. used diptheria toxin-mediated ablation of central serotonergic neurons to examine the role of serotonergic neurons in the sleep/wake behavior of adult mice. Their findings suggest that serotonergic neurons support wakefulness and regulate REM sleep through a biased transition from non-rapid eye movement (non-REM, NREM) sleep to REM sleep. In addition, Saito et al. examined the physiologic relevance of serotonergic regulation of orexin neurons in mice lacking inhibitory serotonin 1A receptors, specifically in orexin neurons. They found that serotonin regulates the orexinergic tone that is important for maintaining the sleep/wake architecture. Although GABAergic neurons in the parafacial zone are implicated in NREM sleep (also known as slow-wave sleep), the function of glutamatergic neurons in this brain area is unknown. Therefore, Erickson et al. utilized a chemogenetic approach to investigate the role of glutamatergic neurons in the parafacial zone in regulating NREM sleep in mice.

Edited and reviewed by: *Luis de Lecea, Stanford University, United States*

\*Correspondence: *Michael Lazarus lazarus.michael.ka@u.tsukuba.ac.jp*

### Specialty section:

*This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience*

Received: *22 June 2020* Accepted: *30 June 2020* Published: *11 August 2020*

### Citation:

*Lazarus M, Hayashi Y, Honjoh S, Vogt KE, Eban-Rothschild A, Liu Q and Sakurai T (2020) Editorial: The Gating and Maintenance of Sleep and Wake: New Circuits and Insights. Front. Neurosci. 14:773. doi: 10.3389/fnins.2020.00773*

One advantage of using transgenic animals is to simplify the identification of neuronal populations and their projections, which can be laborious using traditional techniques. Yuan et al. employed rabies virus-based retrograde tracing to map and quantitatively analyze the whole-brain monosynaptic inputs to cholecystokinin-expressing neurons in the suprachiasmatic nucleus. This research is complemented by the article by Chen et al., as they characterized the monosynaptic inputs and axonal projections of GABAergic neurons in the lateral pontine tegmentum, which play key roles in regulating sleep and locomotion, using virally mediated, cell-type-specific, retrograde and anterograde tracing systems.

In-vivo imaging used in combination with pharmacologic or genetically driven perturbation of defined sets of neurons or glial cells is useful for identifying the brain circuits underlying sleep regulation. For example, Dong et al. employed a genetically encoded dopamine indicator to track striatal dopamine levels across the sleep-wake cycle and in response to external stimuli. In addition, Patel et al. describe simultaneous electrophysiologic recording and cell-type-specific imaging for interrogating state-dependent neural circuit dynamics in vivo. To bridge the gap in our knowledge about the dynamic changes in neurotransmitter systems during the transition from wake to sleep in humans, Lehmann et al. utilized functional magnetic resonance spectroscopy in combination with polysomnography to detect naturally occurring thalamic metabolite concentrations, a methodology that may hold potential for investigating the molecular mechanisms underlying the transition between sleep and wake states and the maintenance of these states in humans. Moreover, human sleep studies may also benefit from the automated classification of sleep stages based on machinelearning algorithms, as, for example, described in the research article by Malafeev et al..

All living organisms respond to stress, i.e., perceived or actual threats, with a predictable biologic pattern in an attempt to restore body homeostasis. Stress is a risk factor for mental disorders and often leads to sleep alterations. Fujii et al. established a mouse model of acute social defeat stress based on a resident-intruder paradigm and examined its effects on the sleep/wake behavior of the mice. The authors demonstrated that NREM sleep strongly increases in response to social defeat stress. Although some of the described sleep changes can be attributed to non-specific effects of the social defeat procedure, most of the NREM sleep increase is likely a specific stress response. By contrast, Yasugaki et al. investigated how sleep changes during a long period of chronic stress in a mouse model of depression induced by water immersion and restraint stress. The authors report that chronic stress produced in their mouse model differentially affects sleep in early and subsequent periods. Likewise, Lou et al.studied changes in the sleep-wake architecture of mice exposed to a single episode of prolonged stress. The authors report a causal link between persistent activation of the mouse medial prefrontal cortex and specific short/long-term electroencephalogram alterations induced by a single episode of prolonged stress.

Insomnia is a subjective complaint of inadequate sleep and the most frequently prescribed hypnotics to treat insomnia are benzodiazepines, which are unsafe when excessively dosed or used for long-term treatment due to severe side effects such as addiction and interference with memory processes (Brett and Murnion, 2015). Benzodiazepines also increase the risk of falling in the elderly (Díaz-Gutiérrez et al., 2017). Several papers in this Research Topic evaluated alternative pharmacologic strategies for the treatment of insomnia. For example, Feng et al. reported that oral delivery of the highly selective α2-adrenergic agonist dexmedetomidine can induce sedative and hypnotic effects. In addition, Yoon et al. report that dieckol, a polyphenol from the brown alga species Ecklonia cava found off the coasts of Japan and Korea, promotes sleep via the benzodiazepine binding site of the GABA<sup>A</sup> receptor in mice. These articles are complemented by the study from Cherasse et al. examining the sleep-inducing effect of prostaglandin D<sup>2</sup> in mice.

Additional original papers examined different aspects of electrocortical activity during sleep and wakefulness. Specifically, Tavakoli et al. present evidence for the occurrence of P3a, a component of event-related potentials elicited by auditory stimuli during sleep in humans, whereas evidence provided by Carrera-Cañas et al. may indicate the existence of a transition state between NREM and REM sleep in rats. The Research Topic is further enriched by a study from Nam et al. on transcriptome analysis of the pineal gland in a mouse model of Alzheimer disease. As the hormone melatonin, which is responsible for regulating the sleep/wake cycle, is produced in the pineal gland of the brain, transcriptomic changes in the pineal glands of patients with Alzheimer disease may contribute to the sleep alterations that often accompany this neurodegenerative disease.

This Research Topic also contains a series of excellent authoritative reviews on various aspects of sleep/wake regulation and related topics. For example, a unified model accounting for the myriad of behavioral state-dependent changes in physiology, including the regulated reduction in body temperature that occurs with sleep, remains lacking. In this context, we highly recommend reading the thoughtful review by Harding et al. on the link between sleep and body temperature. Moreover, the sleep field has recently witnessed an exponential increase in the understanding of brain circuitries regulating sleep/wake behavior. On the other hand, it remains puzzling that brain circuits' switching occurs within seconds, while sleep regulation, i.e., the process of accumulation/dissipation of sleep need, takes hours. Obviously, there is a lack of sufficient experimental data for the integration of glia-neuron interactions, brain circuit activity, and intracellular signal transduction. In this context, Kaur and Saper provide an overview on the circuitries that regulate waking-up from sleep in response to hypercapnia, and Mieda discusses the different roles of multiple neuropeptides and neuropeptide-expressing neurons in the suprachiasmatic nucleus, the central circadian pacemaker in mammals. These "circuit/network" reviews are complemented by two perspective reviews on the integration of neuronal and molecular mechanisms in the regulation of sleep. Yamada and Ueda provide a fresh look at REM sleep regulation from a molecular point of view, whereas Lazarus et al. review the various roles of the classic somnogen adenosine in gating sleep and regulating sleep need.

In summary, the complexity of sleep control by the nervous system and the mystery surrounding the functions of sleep remain a subject of great fascination among scientists. In an effort to tackle these questions, sleep studies are taking advantage of transgenic technologies, in-vivo recording/imaging of neural activity, "omics" approaches, and human functional magnetic resonance imaging. Emerging technologies, such as single-cell and spatial transcriptomics (Hammond et al., 2019; Maniatis et al., 2019) or in-vivo quantification of dynamic changes in neurotransmitters and neuromodulators (Sun et al., 2018; Feng et al., 2019) will further expand the sleep scientists' toolbox. We hope the articles included in this Research Topic will spark new ideas in laboratories that are interested in the sleeping, waking, and dreaming brain.

### AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

### REFERENCES


### FUNDING

Our work was generously supported by the Japan Society for the Promotion of Science [Grants-in-Aid for Scientific Research B (Grant No. 17H02215) to ML and C (Grant No. 18K06849) to KV]; the Japan Science and Technology Agency [CREST grant (Grant No. JPMJCR1655) to ML, YH, KV, and TS]; the Ministry of Education, Culture, Sports, Science, and Technology (MEXT) of Japan [Grants-in-Aid for Scientific Research on Innovative Areas "Living in Space" (Grant Nos. 15H05935, 15K21745, and 18H04966) and "WillDynamics" (Grant No. 19H05004) to ML]; and the World Premier International Research Center Initiative (WPI) from MEXT (to ML, YH, SH, KV, QL, and TS).

### ACKNOWLEDGMENTS

We would like to thank all authors who made original contributions to this research and all reviewers who promoted the quality of the research and manuscript with their comments.


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Lazarus, Hayashi, Honjoh, Vogt, Eban-Rothschild, Liu and Sakurai. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Ablation of Central Serotonergic Neurons Decreased REM Sleep and Attenuated Arousal Response

Kanako Iwasaki <sup>1</sup> , Haruna Komiya<sup>1</sup> , Miyo Kakizaki <sup>1</sup> , Chika Miyoshi <sup>1</sup> , Manabu Abe<sup>2</sup> , Kenji Sakimura<sup>2</sup> , Hiromasa Funato1,3 \* and Masashi Yanagisawa1,4,5 \*

*1 International Institute for Integrative Sleep Medicine (WPI-IIIS), University of Tsukuba, Tsukuba, Japan, <sup>2</sup> Department of Cellular Neurobiology, Brain Research Institute, Niigata University, Niigata, Japan, <sup>3</sup> Department of Anatomy, Faculty of Medicine, Toho University, Tokyo, Japan, <sup>4</sup> Department of Molecular Genetics, University of Texas Southwestern Medical Center, Dallas, TX, United States, <sup>5</sup> Life Science Center, Tsukuba Advanced Research Alliance, University of Tsukuba, Tsukuba, Japan*

### Edited by:

*Zhi-Li Huang, Fudan University, China*

### Reviewed by:

*Michihiro Mieda, Kanazawa University, Japan Christopher S. Leonard, New York Medical College, United States*

### \*Correspondence:

*Hiromasa Funato funato.hiromasa.km@u.tsukuba.ac.jp Masashi Yanagisawa yanagisawa.masashi.fu@ u.tsukuba.ac.jp*

### Specialty section:

*This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience*

Received: *28 March 2018* Accepted: *16 July 2018* Published: *07 August 2018*

### Citation:

*Iwasaki K, Komiya H, Kakizaki M, Miyoshi C, Abe M, Sakimura K, Funato H and Yanagisawa M (2018) Ablation of Central Serotonergic Neurons Decreased REM Sleep and Attenuated Arousal Response. Front. Neurosci. 12:535. doi: 10.3389/fnins.2018.00535* Sleep/wake behavior is regulated by distinct groups of neurons, such as dopaminergic, noradrenergic, and orexinergic neurons. Although monoaminergic neurons are usually considered to be wake-promoting, the role of serotonergic neurons in sleep/wake behavior remains inconclusive because of the effect of serotonin (5-HT)-deficiency on brain development and the compensation for inborn 5-HT deficiency by other sleep/wake-regulating neurons. Here, we performed selective ablation of central 5-HT neurons in the newly developed *Rosa-diphtheria toxin receptor (DTR)-tdTomato* mouse line that was crossed with *Pet1Cre*/<sup>+</sup> mice to examine the role of 5-HT neurons in the sleep/wake behavior of adult mice. Intracerebroventricular administration of diphtheria toxin completely ablated tdTomato-positive cells in *Pet1Cre*/+*; Rosa-DTR-tdTomato* mice. Electroencephalogram/electromyogram-based sleep/wake analysis demonstrated that central 5-HT neuron ablation in adult mice decreased the time spent in rapid eye movement (REM) sleep, which was associated with fewer transitions from non-REM (NREM) sleep to REM sleep than in control mice. Central 5-HT neuron-ablated mice showed attenuated wake response to a novel environment and increased theta power during wakefulness compared to control mice. The current findings indicated that adult 5-HT neurons work to support wakefulness and regulate REM sleep time through a biased transition from NREM sleep to REM sleep.

Keywords: 5-HT neuron, serotonin, sleep, REMS, diphtheria toxin, brain, mouse model

# INTRODUCTION

Sleep/wake behavior is regulated by distinct groups of neurons located from the forebrain to the medulla (Luppi et al., 2011; Weber and Dan, 2016; Scammell et al., 2017). Among a variety of neuronal groups, monoaminergic neurons, such as dopaminergic and noradrenergic neurons, are considered to be wake-promoting neurons (Carter et al., 2010; Eban-Rothschild et al., 2016). Although serotonin (5-HT)-containing neurons are also regarded as wake-promoting neurons (Saper et al., 2005; Scammell et al., 2017), the role of serotonergic neurons on sleep/wake behavior remains inconclusive (Ursin, 2002; Monti, 2011).

In early studies conducted by Jouvet et al. dorsal raphe lesions decreased sleep time in correlation with a reduction in brain serotonin (Jouvet, 1999). The administration of pchlorophenylalanine (PCPA), which is an irreversible inhibitor of tryptophan hydroxylase (TPH), the rate-limiting enzyme in the synthesis of 5-HT, induced insomnia, suggesting a role of the ascending serotonergic system in sleep enhancement (Mouret et al., 1968; Jouvet, 1999). However, PCPA also block TPH1 and reduced peripheral 5-HT, which is involved in a variety of functions(El-Merahbi et al., 2015). Furthermore, PCPA reduces dopamine and noradrenalin levels in the brain(Dailly et al., 2006) which may alter sleep/wakefulness. Hypothermia after PCPA treatment may also lead to insomnia (Murray et al., 2015).

Subsequently, unit recordings of raphe nucleus neurons revealed that their activity was usually high during wakefulness, low during NREM sleep and almost absent during REM sleep (McGinty and Harper, 1976; Trulson and Jacobs, 1979; Jacobs and Fornal, 1999). Pharmacological studies targeting 5- HT receptors supported the wake-promoting effects of 5-HT signaling. Systemic administration of a 5-HT1A agonist (Monti and Jantos, 2003), 5-HT1B agonist (Bjorvatn and Ursin, 1994; Monti et al., 1995) and 5-HT2 agonist (Dugovic et al., 1989) increased wakefulness and decreased NREM sleep and REM sleep. Intracerebroventricular infusion of a 5-HT3 agonist also increased wakefulness and decreased NREM sleep and REM sleep (Ponzoni et al., 1993). Among the 14 5-HT receptors in mice, 5-HT1A, 1B, 2A, 2C, and 7 have been investigated through the examination of sleep/wake behavior in mice deficient in these receptors (Hannon and Hoyer, 2008). The time spent in wakefulness was greater in 5-HT2A- and 2C-deficient mice than in control mice (Frank et al., 2002; Popa et al., 2005) and was not altered in 5-HT1A- or 5-HT1B-deficient mice (Boutrel et al., 1999, 2002), which is seemingly inconsistent with the wakepromoting effect of agonists for 5-HT1A, 1B, and 2A/C. 5-HT7 deficient mice also had a normal total wake time (Hedlund et al., 2005).

Since 5-HT receptors are diverse in their regional and subcellular distribution and differ in their effect on intracellular signaling and neuron activity (Hannon and Hoyer, 2008), the role of the entire 5-HT neuron network in sleep/wake behavior cannot be characterized by investigating mice that are deficient in each 5-HT receptor but must be assessed by examining mice that are deficient in 5-HT or 5-HT neurons. Mice deficient in Tph2, a rate-limiting enzyme for brain 5-HT, showed normal time spent in wakefulness, NREM sleep and REM sleep with a longer NREM sleep duration than control mice (Solarewicz et al., 2015). The unexpectedly mild sleep phenotype of Tph2 deficient mice could be due to a compensation for the inborn deficiency by other sleep/wake-regulating neurons. In addition, loss of serotonin signaling during brain development modulates the response of developing thalamocortical axons to netrin-1 (Bonnin et al., 2007) and results in disturbances in the formation of neural circuits related to sleep/wake behavior, potentially masking the role of 5-HT in sleep/wake behaviors of wild-type mice. In fact, acute disruption of Tph2 in the adult raphe nuclei led to a reduction in time spent in behavioral sleep (Whitney et al., 2016).

In addition to their effect on wakefulness, 5-HT neurons have been proposed to play an inhibitory role in the switch from NREM sleep to REM sleep based on the cessation of firing of dorsal raphe neurons during REM sleep (McCarley, 2007). Consistent with this finding, microinjection of a 5-HT1A agonist (Monti and Jantos, 2003) and 5-HT2 agonist (Amici et al., 2004) into the laterodorsal tegmental nucleus, which contains REM sleep-active cholinergic neurons, decreased REM sleep. The increased REM sleep of 5-HT1A-deficient mice (Boutrel et al., 2002) and 5-HT1B-deficient mice (Boutrel et al., 1999) compared to that of wild-type mice furthers support the role of 5-HT1A and 1B in REM sleep suppression. 5-HT7-deficient mice have exhibited shorter REM sleep time than wild-type mice (Hedlund et al., 2005).

In addition to 5-HT, 5-HT neurons express vesicular glutamate transporter 3 (VGLUT3), which is responsible for the uptake of glutamate into synaptic vesicles (Fu et al., 2010; Hioki et al., 2010), and 5-HT neurons release glutamate to modulate reward behavior (Liu et al., 2014). 5-HT neurons in the raphe nuclei also express some neuropeptides (Okaty et al., 2015). Thus, neurotransmitters other than 5-HT may be involved in sleep/wake regulation, and thus the sleep/wake behavior of 5-HT neuron-ablated mice may be distinct from that of mice deficient in 5-HT.

To examine the role of 5-HT neurons in sleep/wake behaviors of adult mice, the effects of 5-HT deficiency on brain development and on the maturation of sleep/wake behavior during the early postnatal stage must be excluded. Here, we performed selective ablation of central 5-HT neurons by the intracerebroventricular (ICV) administration of diphtheria toxin (DT) (Murphy, 2011) in the newly developed Pet1Cre/+; Rosa-diphtheria toxin receptor (DTR)-tdTomato mouse line to examine sleep/wake behavior based on electroencephalogram (EEG)/electromyogram (EMG). Ablation of central 5-HT neurons in adult mice resulted in a decrease in time spent in REM sleep, which was associated with fewer transitions from NREM sleep to REM sleep than in control mice, and attenuated the increase in wakefulness in response to a novel environment.

# METHODS

### Animals

All of the procedures were conducted in accordance with the Guidelines for Animal Experiments of University of Tsukuba and were approved by the Institutional Animal Care and Use Committee of University of Tsukuba (Approved protocol ID # 140144). Mice were raised in our breeding colony under controlled conditions (12-h light/dark cycle, lights on at 9:00 A.M., 55 ± 5% humidity, and ad libitum access to water and food). Mice were weaned at 4 weeks of age and were housed in groups of four or five. Pet1-Cre mice (B6.Cg-Tg(Fev-cre)1Esd/<sup>J</sup> (Scott et al., 2005; Deneris, 2011) were used in this study. At any stage of this study, when mice exhibited symptoms of being severely unhealthy, such as massive weight loss or an inability to walk, we sacrificed them by cervical dislocation under deep anesthesia and did not use the data associated with that animal in this study. We housed mice and performed all experiments at 23 ± 2 ◦C.

### Production of Rosa-loxP-stop-loxP-DTR-tdTomato Mice

For the generation of Rosa26-loxP-stop-loxP-DTR-tdTomato mice (Rosa-DTR-tdTomato mice), a genomic fragment that contained the ROSA locus was isolated from C57BL/6 mouse genomic BAC clone from an RP23 mouse genomic BAC library (Advanced Geno TEchs Co). The targeting vector had a splice acceptor (SA), two loxP sequences that were inserted before the repeated SV40 polyadenylation signal and after the "stop" sequence that contained the terminator of the yeast His3 gene and SV40 polyadenylation signal. A 1.7-kb fragment of the FRT-PGK-gb2-neo-FRT-loxP cassette (Gene Bridges) was inserted after the repeated SV40 polyadenylation signal. After the second loxP sequence, the targeting vector contained the coding sequence of the DTR, simian HBEGF followed by IRES and tdTomato sequences. The targeting vector was linearized and electroporated into the C57BL/6N ES cell line RENKA. Correctly targeted clones were injected into the eight-cell stage ICR mouse embryos, which were cultured to produce blastocysts and later transferred to pseudopregnant ICR females. The resulting male chimeric mice were crossed with female C57BL/6N mice to establish the LSL-DTR-tdTomato; neo line. To remove the neomycin resistance gene with the FLP-FRT system, LSL-DTRtdTomato; neo mice were crossed with Actb-FLP knock-in mice (Kono et al., 2017). LSL-DTR-tdTomato mice were further crossed with Pet1Cre mice to establish Pet1Cre; DTR-tdTomato mice. Gene-modified mice were regularly crossed with C57BL/6J mice (CLEA Japan) to minimizing genetic drift.

### Surgery

For the infusion of DT into the lateral cerebral ventricle, a cannula was implanted. The guide cannula was prepared by cutting stainless steel wire (Ziggy's tubes and wires, 23R304- 36) into 10-mm lengths, and the dummy cannula was made by cutting a 30-gauge stainless steel tube (Ziggy's tubes and wires, #30R304-36) into a 15-mm-long tube and folding one end of the tube. An injection cannula was also made from 23- and 30-gauge stainless steel tubes.

Male mice (2.7–5.5 months old) were anesthetized with isoflurane, and the cranium was exposed. A guide cannula was inserted under stereotaxic control, which was tilted 30◦ to the sagittal plane to make room for the implantation of EEG electrodes. The tip of the cannula was targeted to the following coordinates: anteroposterior (AP): −0.50 mm, mediolateral (ML): 1.00 mm, and dorsoventral (DV): −2.25 mm. Immediately after cannula implantation, each mouse was implanted with an EEG/EMG electrode containing 4 electrode pins and two flexible stainless steel wires as reported previously (Komiya et al., 2018). The electrode pins were lowered to the dura under stereotaxic control, and the electrode socket was subsequently attached to the skull with dental cement (3M ESPE, Ketac Cem Aplicap). Two ipsilateral pins (AP: 0.5 mm, ML: 1.77 mm, DV: −1.3 mm; and AP: −4.5 mm, ML: 1.77 mm, DV: −1.3 mm) were used for the EEG recording. For the EMG recording, two stainless wires were inserted into the neck extensor muscles.

### ICV Injection of aCSF and DT

After habituation to the recording conditions, the mice were injected with 1 µL aCSF through the guide cannula using a 5 µl syringe (Hamilton #85) under anesthesia with isoflurane. The aCSF contained 125 mM NaCl, 2.5 mM KCl, 1.25 mM NaH2PO4, 26 mM NaHCO3, 10 mM glucose, 2 mM CaCl<sup>2</sup> and 1 mM MgCl2. Twelve to fourteen days after the aCSF injections, 5 ng DT was injected at a concentration of 5 ng/µL.

### EEG/EMG Recording and Analysis

EEG/EMG signaling was obtained and analyzed as previously described with some modifications (Funato et al., 2010). EEG/EMG signals were amplified, filtered (EEG: 0.3–300 Hz; EMG: 30–300 Hz) with a multi-channel amplifier (NIHON KODEN, #AB-611J), and digitized at a sampling rate of 250 Hz using an analog-to-digital converter (National Instruments #PCI-6220). The EEG/EMG data were visualized and semiautomatically analyzed by MATLAB-based software followed by visual inspection. Each 20-s epoch was staged into wakefulness, NREM sleep and REM sleep. Wakefulness was scored based on the presence of low amplitude, fast EEG activity and high amplitude, variable EMG activity. NREM sleep was characterized by high amplitude, delta (1–4 Hz)-frequency EEG waves and low EMG tonus, whereas REM sleep was staged based on theta (6– 9 Hz)-dominant EEG oscillations and EMG atonia. EEG/EMG was subsequently recorded for 11 days after injections. The recordings on the 9th and 10th day after the injections were analyzed beginning at ZT0 (9:00 A.M.). The total time spent in wakefulness, NREM sleep, and REM sleep were derived by summing the total number of 20-s epochs in each state. Mean episode durations were determined by dividing the total time spent in each state by the number of episodes of that state. Epochs that contained movement artifacts were included in the state totals but excluded from subsequent spectral analysis. EEG signals were subjected to Fourier transform analysis from 1 to 30 Hz with 1-Hz bins using the MATLAB-based custom software. The EEG power density in each frequency bin was expressed as a percentage of the mean total EEG power over all frequency bins and sleep/wake states.

### Arousal Response to a Novel Cage

A mouse that was housed in a home cage for at least 1 week was transferred to a novel cage that contained fresh bedding at ZT7. The sleep/wake behavior was assessed for 24 h from ZT0. To assess EEG power density during wakefulness from ZT7 to ZT12, delta and theta range power were normalized by the total EEG power over all frequency (1–30 Hz) during wakefulness from ZT7 to ZT12. We performed novel cage experiments using 8 mice, and the EEG/EMG data from 2 mice that contained movement artifacts were included in state totals but excluded from the spectral analysis.

### Blood Glucose and Body Temperature

Blood glucose was measured from tail blood using Glutest system (Sanwa Kagaku) at the late light phase in a fed condition. The body temperature of a mouse was monitored using a digital thermometer (BDT-100, Bio Research Center) with a rectal probe (RET-3, Bio Research Center) without anesthesia and was acquired in 15 s, including mouse restraint, probe insertion into the rectum, stabilized temperature recording, and probe removal. After full acclimatization, the mice only needed to be gently restrained for the body temperature measurement.

We measured body temperature of Pet1Cre/+; Rosa-DTRtdTomato mice at ZT5, 11, 13, and 23 on the 9th and 10th days after the aCSF injection. Subsequently, we injected DT into the same Pet1Cre/+; Rosa-DTR-tdTomato mice and measured their body temperature at ZT5, 11, 13, and 23 on the 9th and 10th days after the DT injection.

### Histological Examination

Mice were deeply anesthetized with isoflurane and transcardially perfused with PBS followed by 4% paraformaldehyde (PFA) in PBS. The brains were postfixed in 4% PFA at 4◦C overnight, cryoprotected in 30% sucrose in PBS for 2 days, embedded in OCT compound (Sakura Finetech), and stored at −80◦C until use. The brains were cryosectioned coronally at a thickness of 40µm and stored in tissue cryoprotectant solution at −20◦C.

We used anti-5-HT antibody to visualize 5-HT neurons. More than 99% of Tph2-positive cells are positive for 5- HT and vice versa (Hioki et al., 2010). The sections were rinsed with PBS and incubated in 0.4% Block Ace (Snow Brand Milk Products) in PBS with 0.1% Tween20 (0.1% PBST) for 1–2 h at room temperature. This procedure was followed by overnight incubation with goat anti-5-HT (1:5000; ImmunoStar #20079) and rabbit anti-RFP (1:2500; MBL #PM005) antibodies in 0.2% Block Ace in 0.1%PBST overnight at 4◦C. The sections were rinsed with PBS and then incubated with AlexaFluor488-conjugated donkey anti-goat IgG (1:1000; ThermoFisher #A11055) and AlexaFluor594-conjugated donkey anti-rabbit IgG (1:1000; ThermoFisher # A21207) antibodies in 0.2% Block Ace in 0.1%PBST overnight at 4◦C. Sections were rinsed with PBS and mounted with Vectashield mounting medium with 4',6-diamidino-2-phenylindole (DAPI) (Vector Lab, #H-1200). The signals were visualized using a confocal microscope (Zeiss, #LSM700) with Plan-Apochromat 20x/0.8 M27 (Zeiss). AlexaFluor488 or AlexaFluor594 was excited with 488 or 555 nm laser beams, respectively, and their fluorescence was obtained 490–587 or 585- nm emission wavelength, respectively. The sections 4.7, 6.6, and 7.1 mm posterior from bregma were used for cell counting for the dorsal/median raphe, raphe pallidus and raphe obscurus, respectively. The images of stained sections were acquired using ZEN 2010 and analyzed using ZEN Black software (Zeiss). The raphe areas were shown on the display with the scale in which 1 cm on the display represents 25µm on the section. Using a line tool of ZEN software, we marked positive cells at the single fluorescent image for AlexaFluor488 or AlexaFluor594, separately and subsequently classified positive cells into 5-HT positive, tdTomato positive and 5-HT/tdTomato positive neurons.

# High-Performance Liquid Chromatography (HPLC)

After cervical dislocation, the brain was immediately removed and sectioned coronary at 2.4 mm posterior from bregma. The anterior part of the brain was placed on ice and processed for HPLC analysis, whereas the posterior part of the brain was used for immunohistochemistry to assess the efficiency of the cell ablation. The brain was homogenized with 0.2 M perchloric acid, 100µM EDTA2·Na in MilliQ (0.5 mL/ 100 mg brain) and 200 ng isoproterenol as an internal standard.

The homogenized sample was left on ice for 30 min and then centrifuged at 20,000 × g for 15 min at 0◦C. The pH of the supernatant was adjusted to be approximately pH 3, and the supernatant was then filtered through a 0.45µm Millex filter (Millipore). The noradrenalin, 5-HT, 5-HIAA, dopamine, DOPAC, and HVA in the solutions were separated using an Eicompak SC-5ODS column (Eicom) and subsequently detected using an electrochemical detector HTEC-500 (Eicom). The concentration was calculated based on peak areas which were quantified based on the external standerd calibration employing linear regression analysis using LC solution software (Shimadzu). Norepinephrine bitartratesalt (Sigma N5785), Serotonin creatininesulfate (Sigma H7752), 5-Hydroxyindoleaceticacid (Sigma H8876), Dopamin hydrochloride (Sigma H8502), 3,4-Dihydroxyphenylacetic acid (Sigma 850217), and Homovanillicacid (Sigma H1252) dissolved in 0.1 M acetic acid, 50µM EDTA2·Na in MilliQ were measured as external standerds in each experiment.

### STATISTICAL ANALYSIS

No method of randomization was used in any of the experiments. The experimenters who staged sleep/wakefulness based on EEG/EMG were blinded to treatment assignment. Statistical analysis was performed using SPSS Statistics 22 (IBM). All of the data were tested for Gaussian distribution and variance. For group comparisons among the control group and the groups treated with 2.5 ng and 5 ng DT, the number of neurons was analyzed using one-way ANOVA followed by post hoc Tukey's test. Wake response was analyzed using two-way ANOVA followed by post hoc Tukey's test. To compare the parameters of paired groups, we performed paired t-test with Bonferroni correction. We performed Wilcoxon matched-pairs signed-rank test with Bonferroni correction to compare the parameters of paired groups when the data did not follow Gaussian distribution.

### RESULTS

### Ablation of Central 5-HT Neurons in Adult Mice

Whereas simian or human are sensitive to DT, mice are resistant to DT (Pappenheimer et al., 1982; Saito et al., 2001). The species difference in sensitivity to DT depends on the binding affinity of DT to heparin-binding epidermal growth factor (HBEGF). Simian and human HBEGF works as the receptor for DT (DTR) (Mitamura et al., 1995). To render a specific group of neurons sensitive to DT in a Cre-dependent manner, we generated mice in which the Rosa26 locus was modified by a targeted insertion of a construct that contained loxP-StoploxP-simian HBEGF or DTR, followed by internal ribosomal entry site (IRES)-tdTomato (**Figure 1A**). In these mice, DTR is expressed in the presence of Cre protein in cells that can be visualized by a fluorescent protein, tdTomato. Rosa26-DTRtdTomato mice were crossed with Pet1Cre mice (Scott et al., 2005; Liu et al., 2010), in which Cre is expressed in 5-HT neurons, intestinal epithelial cells and pancreatic islet cells in adult mice (Scott et al., 2005). In the pons of Pet1Cre ; Rosa26-DTR-tdTomato mice, the percentage of 5-HT-postive cells that were positive for tdTomato was 68.1 ± 2.7 (mean ± SEM) % and 71.2 ± 4.1% in the dorsal and median raphe, respectively (**Figures 1B,C**), whereas 4.4 ± 0.8% and 12.6 ± 1.2% of tdTomato-positive cells were negative for 5-HT in the dorsal and median raphe, respectively. In the medulla, the percentage of 5-HT-postive cells that were positive for tdTomato was 63.6 ± 5.1% and 64.1 ± 3.1% in the raphe pallidus and raphe obscurus, respectively, whereas 9.9 ± 2.8% and 11.9 ± 1.9% of tdTomato-positive cells were negative for 5-HT in

FIGURE 1 | Ablation of central 5-HT neurons. (A) Schematic diagram of the wild-type (WT) Rosa26 locus, PGK-Neo-STOP- HBEGF-tdTomato allele, PGK-Neo deleted by β*-actinFLPase*/<sup>+</sup> and the allele activated by Pet1-Cre-mediated deletion of the STOP sequence. (B) Leftmost column: Representative images of cells double positive for 5-HT (green) and tdTomato (magenta) with a merged image (white). Second column: The raphe nuclei contained many cells that were doubly positive for tdTomato and 5-HT cells in *Pet1Cre*/+; *Rosa-DTR-tdTomato* mice without diphtheria toxin (DT) administration. Third column: There were many remaining cells that were doubly positive for tdTomato and 5-HT after the administration of 2.5 ng DT. Fourth column: After the administration of 5 ng DT, no tdTomato-positive cells were observed in the raphe nuclei. Arrows indicate cells doubly positive for 5-HT and tdTomato. Arrowheads indicate 5-HT-positive, tdTomato-negative cells. Rightmost column: After the administration of 5 ng DT, the raphe nuclei contained many 5-HT cells in *Pet1*+/+; *Rosa-DTR-tdTomato* mice. Scale bars: 10µm for the leftmost column, 50µm for other columns. (C) The number of serotonergic neurons in raphe nuclei without (*n* = 4) and after the administration of 2.5 ng (*n* = 3–6) and 5 ng DT (*n* = 5) in *Pet1Cre*/+; *Rosa-DTR-tdTomato* mice, and 5 ng DT (*n* = 3–4) in *Pet1*+/+; *Rosa-DTR-tdTomato* mice. One-way ANOVA followed by Tukey's test. (D,E) Quantification of 5-HT, 5-HIAA, dopamine, HVA, DOPAC, and noradrenaline in brain homogenates of DT-administered *Pet1Cre*/+; *Rosa-DTR-tdTomato* (*n* = 12) and *Pet1*+/+; *Rosa-DTR-tdTomato* mice (*n* = 9). Two-tailed *t*-test. \**p* < 0.05, \*\**p* < 0.01, \*\*\**p* < 0.001. The data are presented as the group mean ± S.E.M.

the raphe pallidus and raphe obscurus, respectively, which is consistent with the results of a previous study (Cerpa et al., 2014) (**Figures 1B,C**).

First, we determined the amount of DT necessary to ablate all tdTomato-positive cells expressing DTR. Since DT ablated the neurons within 1 week (Wu et al., 2008), we quantified the remaining tdTomato-positive cells 12 days after ICV DT administration. When 2.5 ng of DT was administered, many tdTomato-positive cells survived in the raphe nuclei of the Pet1Cre/+; Rosa-DTR-tdTomato mice (**Figures 1B,C**). However, 5 ng of DT completely ablated the tdTomato-positive cells in the raphe nuclei (**Figures 1B,C**). After the administration of 5 ng of DT, the number of 5-HT neurons was decreased by 69% in the dorsal raphe compared with that in the non-injected mice. The number of 5-HT neurons in the median raphe, raphe pallidus and raphe obscurus decreased to 33, 35, and 30%, respectively, compared with that in the non-injected mice (**Figures 1B,C**), consistent with the percentage of all 5-HT neurons negative for tdTomato (**Figures 1B,C**). The administration of 10 ng of DT did not decrease the number of remaining 5-HT neurons (data not shown). The administration of 5 ng DT did not ablate 5-HT neurons of Pet1+/+; Rosa-DTR-tdTomato mice in which no DTR was observed (**Figure 1C**). At 9 days after the administration of 5 ng of DT, the number of central 5-HT neurons was significantly lower than that in control mice and similar to that observed 12 days after the DT administration (101.0 ± 4.6 cells in the dorsal raphe, P = 0.71 vs. the 12th day; 34.0 ± 0.8 cells in the median raphe, P = 0.11 vs. the 12th day; 11.0 ± 1.7 cells in the raphe pallidus, P = 0.82 vs. the 12th day; 14.7 ± 1.5 in the raphe obscurus, P = 0.66 vs. the 12th day, n = 3, one-way ANOVA with Tukey's post hoc test). Thus, we used 5 ng of DT to assess the effect of central 5-HT neurons on sleep/wake behaviors.

We confirmed the reduction in 5-HT levels after the ablation of 5-HT neurons using high-performance liquid chromatography (HPLC) of forebrain homogenates. DTadministered Pet1Cre/+; Rosa-DTR-tdTomato mice showed that a 55.3 and 46.1% reduction in 5-HT and its metabolite, 5-hydroxyindoleacetic acid (5-HIAA), respectively, compared with DT-administered Pet1+/+; Rosa-DTR-tdTomato mice (**Figure 1D**). To further examine whether central 5-HT neuron ablation affected the amount of other monoamines, we quantified the levels of noradrenaline, dopamine, and dopamine metabolites 3,4-dihydroxyphenylacetic acid (DOPAC) and homovanillic acid (HVA) in forebrain homogenates, which were comparable between forebrain homogenates from DT-administered Pet1Cre/+; Rosa-DTR-tdTomato and DT-administered Pet1+/+; Rosa-DTR-tdTomato mice (**Figure 1E**).

### Physical Parameters After Central 5-HT Neuron Ablation

Since central 5-HT neurons have been reported to be involved in glucose metabolism and body temperature (Cerpa et al., 2014; McGlashon et al., 2015), we examined whether DT administration affected the serum glucose level, body temperature, and body weight. Although 5 ng DT did not affect

the serum glucose level (**Figure 2A**), slightly but significantly reduced body weight [change in body weight; −1.1 ± 0.37 g (mean ± S.E.M), **Figure 2B**]. We measured body temperature of Pet1Cre/+; Rosa-DTR-tdTomato mice at ZT5, 11, 13, and 23 on the 9th and 10th days after the DT administration. Mice are most often asleep at ZT5 and are awake and most active at ZT13. ZT11 and ZT23 are 1 h before the light-off and light-on, respectively. DT administration did not alter body temperature at ZT11, 13 and 23 (change in body temperature; −0.15 ± 0.15◦C at ZT11, 0.08 ± 0.19◦C at ZT13, 0.03 ± 0.21◦C at ZT23) but decreased at ZT5 (change in body temperature; −0.63 ± 0.12◦C, **Figure 2C**).

are presented as the group mean ± S.E.M.

number (C) of aCSF- and diphtheria toxin (DT)-administered *Pet1Cre*/+; *Rosa-DTR-tdTomato* mice. (D–F) Total time spent in NREM sleep (NREMS) (D), NREMS episode duration (E) and NREMS episode number (F) of aCSF- and DT-administered *Pet1Cre*/+; *Rosa-DTR-tdTomato* mice. (G–I) Total time spent in REM sleep (REMS) (G), REMS episode duration (H) and REMS episode number (I) of aCSF- and DT-administered *Pet1Cre*/+; *Rosa-DTR-tdTomato* mice. Ten mice in each group. \**p* < 0.05. Wilcoxon matched-pairs signed-rank test with Bonferroni correction. The data are presented as the group mean ± S.E.M.

# Decreased REM Sleep in Central 5-HT Neuron-Ablated Mice

We assessed the sleep/wake behaviors of DT-administered mice 9 and 10 days after DT administration. Central 5-HT neuron ablation did not alter the time spent in wakefulness (**Figure 3A**), the wake episode duration (**Figure 3B**) or the wake episode number (**Figure 3C**). Furthermore, the daily variation in wake time was similar between artificial cerebrospinal fluid (aCSF)- and DT-administered Pet1Cre/+; Rosa-DTR-tdTomato mice (**Figure 4A**).

DT-administered Pet1Cre/+; Rosa-DTR-tdTomato mice did not alter total NREM sleep time (**Figure 3D**). The ablation of central 5-HT neurons decreased the NREM sleep episode duration during the dark phase (**Figure 3E**). There was no significant difference in the number of NREM sleep episodes (**Figure 3F**), and the daily variation in NREM sleep time was similar between aCSF- and DT-administered Pet1Cre/+; Rosa-DTR-tdTomato mice (**Figure 4B**).

The ablation of central 5-HT neurons decreased the time spent in REM sleep by approximately 24.7% daily and 53.6% in the dark phase but the difference did not reach significance during the light phase (**Figures 3G**, **4C**). Furthermore, central 5-HT neuron ablation increased the duration of REM sleep episodes during the light phase (**Figure 3H**). Finally, DT-administered Pet1Cre/+; Rosa-DTRtdTomato mice exhibited fewer REM sleep episodes during

both the light and dark phases than those administered aCSF (**Figure 3I**).

The ablation of central 5-HT neurons decreased the transitions from NREM sleep to REM sleep by 34.7% (**Figures 4D,E**). The number of transitions from REM sleep to wakefulness were also decreased in DT-administered Pet1Cre/+; Rosa-DTR-tdTomato mice compared with that in mice given aCSF. Conversely, the transitions from NREM sleep to wakefulness increased in central 5-HT neuron-ablated mice (**Figures 4D,E**). To normalize the difference in NREM sleep episode number of each mouse, we calculated the transition ratio of NREM sleep episodes. DT-administered Pet1Cre/+; Rosa-DTR-tdTomato mice showed a higher transition ratio from NREM sleep to wake and a lower transition ratio from NREM sleep to REM sleep than aCSF-administered Pet1Cre/+; Rosa-DTR-tdTomato mice (**Figure 4F**).

### Increased Theta Power During Wakefulness of Central 5-HT Neuron-Ablated Mice

The EEG spectral analysis of the wake state revealed that the theta (6–9 Hz)-range power density in Pet1Cre/+; Rosa-DTR-tdTomato mice given DT was significantly higher than that in aCSF-administered Pet1Cre/+; Rosa-DTR-tdTomato mice (**Figure 5A**) without any significant change in the power density in the delta (1–4 Hz) range. Furthermore, the ablation of central 5-HT neurons did not alter the EEG spectrum during NREM sleep (theta range, P = 0.588; delta range, P = 0.524; **Figure 5B**) or REM sleep (theta range, P = 0.818; delta range, P = 0.270; **Figure 5C**).

### Ablation of Central 5-HT Neurons Attenuated the Arousal Response to a Novel Environment

To further examine the wake-promoting effect of 5-HT neurons, we assessed the sleep/wake behavior after the mice were moved into a novel cage during the light phase, as previously performed with mice deficient in wake-promoting transmitters such as histamine, noradrenaline, and orexin (Parmentier et al., 2002; Hunsley and Palmiter, 2003; Mochizuki et al., 2004). Central 5-HT neuron-ablated mice exhibited a strong arousal response similar to control mice during the first 1 h after the cage change at ZT7 (**Figures 6A,B**), but the time spent in wakefulness of 5-HT-ablated mice was significantly less than that of aCSFadministered mice during the remaining light phase from ZT7 to ZT12 (**Figures 6A,C**), suggesting a shorter duration of the arousal response in DT-administered Pet1Cre/+; Rosa-DTRtdTomato mice. Although DT-administered Pet1Cre/+; Rosa-DTR-tdTomato mice showed shorter total wake time during the light phase after cage changes than aCSF-administered Pet1Cre/+; Rosa-DTR-tdTomato mice, both delta power and theta power during wakefulness from ZT7 to ZT12 were similar between the two groups (**Figures 6D,E**).

### DISCUSSION

The present study demonstrated that the ablation of central 5-HT neurons in adult mice resulted in a reduction in total REM sleep time that was associated with fewer transitions from NREM sleep to REM sleep. Central 5-HT neuron-ablated mice also showed an attenuated arousal response to a novel environment compared to control mice.

DT kills cells by inhibiting protein synthesis through ADPribosylation of elongation factor 2 (Murphy, 2011). Since the amount of DT that is required to ablate target cells differs depending on the route of administration and the location of the target cells, we need to determine the minimum amount of DT that reliably ablates all cells expressing DTR. The Rosa-DTR-tdTomato mouse is a useful mouse line in which the tdTomato protein allows for the identification of cells susceptible to DT and examination of the efficiency of DT-induced cell ablation.

Pet1 is necessary for the proper differentiation and function of almost all 5-HT neurons (Wyler et al., 2015, 2016). Although Pet1-deficient mice have 20–30% of 5-HT neurons compared with wild-type mice (Kiyasova et al., 2011), the remaining 5- HT neurons in Pet1-deficient mice lack the expression of genes such as Tph2, serotonin transporter, organic cation transporter 3 (Slc22a3), and alpha 1 adrenergic receptor that are required for the differentiation and function of 5-HT neurons (Wyler et al., 2015, 2016). The presence of tdTomato-negative, 5-HT-positive cells in Pet1Cre/+; Rosa-DTR-tdTomato mice was consistent with a previous study using Pet1Cre/<sup>+</sup> mice (Cerpa et al., 2014) and may be caused by imperfect recombination efficiency, which was suggested by the comparison between Sert-Cre and Pet1- Cre activity on a floxed vesicular monoamine transporter 2 gene (Narboux-Nême et al., 2011, 2013).

The current study showed decreased REM sleep after central 5-HT neuron ablation. Consistent with this finding, Pet1- Cre; Lmx1bflox/flox mice that did not produce central 5-HT exhibited less REM sleep than wild-type mice (Buchanan and Richerson, 2010). Since mice deficient in Gs-coupled 5-HT7 exhibited a decrease in the total time of REM sleep (Hedlund et al., 2005) but no change was observed in mice deficient in Gq-coupled 5-HT2A (Popa et al., 2005) and 5-HT2C (Frank et al., 2002), the loss in 5-HT7/Gs-signaling may be crucial for the observed reduction in REM sleep time after central 5-HT neuron ablation. The increased total REM sleep time in mice deficient in the Gi-coupled autoreceptors 5-HT1A (Boutrel et al., 2002) and 5-HT1B (Boutrel et al., 1999) may be explained by enhanced 5-HT7/Gs signaling subsequent to the loss of inhibitory autoreceptors. 5-HT neurons are regarded as "REM-off " neurons, which cease firing during REM sleep (McGinty and Harper, 1976; Trulson and Jacobs, 1979; Jacobs and Fornal, 1999). The observation that the loss of "REM-off " 5-HT neurons decreases REM sleep time is counterintuitive. One potential explanation for these confounding observation is that the activity of 5-HT neurons during NREM sleep enhances the tendency for a state transition from NREM sleep toward REM sleep and away from wakefulness, which could explain the reduction in the transition from NREM sleep to REM sleep after central 5-HT neuron ablation observed in the present study.

5-HT neuron-ablated mice exhibited an attenuated arousal response to a novel environment compared to the control mice, which has been reported in mice deficient in histidine

followed by Tukey's test. The data are presented as the group mean ± S.E.M.

decarboxylase, a rate-limiting enzyme for the synthesis of histamine (Parmentier et al., 2002), and in mice deficient in dopamine β-hydroxylase (Hunsley and Palmiter, 2003). This finding supports the role of 5-HT neurons as wake-promoting neurons, consistent with the findings that optogenetic activation of 5-HT neurons in the dorsal raphe enhanced wakefulness (Ito et al., 2013), activation of 5-HT neurons using the designer receptor exclusively activated by designer drugs (DREADD) system increased regional cerebral blood flow in many cortical and subcortical areas including the ventral tegmental area (Giorgi et al., 2017), and 5-HT neurons are usually more active during wakefulness (McGinty and Harper, 1976; Trulson and Jacobs, 1979; Jacobs and Fornal, 1999). However, video monitoring has shown a decrease in sleep time in adult 5- HT deficiency (Whitney et al., 2016), suggesting that the effect of 5-HT neuron ablation on sleep/wakefulness could be different from that of adult 5-HT deficiency, potentially because glutamate transmission from 5-HT neurons may contribute to wakefulness (Fu et al., 2010; Hioki et al., 2010; Liu et al., 2014).

Although the current study failed to detect any changes in daily total wake time after the ablation of 5-HT neurons, this result is consistent with previous studies on mice that were deficient in wake-promoting neurotransmitters, such as histamine, noradrenalin and orexin, which showed normal total wake time under basal conditions (Parmentier et al., 2002; Hunsley and Palmiter, 2003; Mochizuki et al., 2004). Since multiple wake-promoting circuits work together to maintain baseline wakefulness, the redundancy in the wake-promoting system may be able to compensate for the loss of 5-HT neurons in basal wakefulness but may not be sufficient for a full arousal response in a novel cage.

The current study also showed higher theta power during wakefulness of central 5-HT neuron-ablated mice, suggesting a suppressing effect of 5-HT signaling in hippocampal theta generation. The theta power during wakefulness usually increases during exploratory behaviors and may play a role in navigation (Buzsáki, 2005; Bender et al., 2015). Thus, our results indicates 5- HT neurons may negatively regulate hippocampal theta rhythm during navigation and locomotive behavior. Consistently, systemic administration of selective agonists for autoreceptors 5-HT1A and 1C enhanced the hippocampal theta of freely moving animals (Marrosu et al., 1996; Sörman et al., 2011). In addition, 5-HT fibers were found to be abundant in the hippocampus of wild-type mice but drastically decreased in the hippocampus of Pet1-deficient mice (Kiyasova et al., 2011). Although the theta power of central 5-HT neuronablated mice from ZT7-12 did not reach statistical significance (**Figure 6E**), this may be due to a smaller number of mice examined.

A limitation of the current study was that the role of 5- HT neurons in sleep/wake behavior could be underestimated due to the effect of the remaining 5-HT neurons after the ablation. Total loss of central 5-HT neurons may cause a further decrease in total REM sleep time and a weaker arousal response compared to the current results, and could result in an decrease in total wake time and a increase in total NREM sleep time. The number of 5-HT neurons was decreased by 68%, and the 5-HT content was decreased by 55%. The smaller reduction in 5-HT content than in 5-HT neuron number can be partly explained by 5-HT derived from the blood. Tph2-deficient mice have 4–7% of the 5-HT found in control mice in the brain or cerebral cortex (Savelieva et al., 2008; Alenina et al., 2009). A reduction in negative feedback via autoreceptor 5-HT1a may also work to increase 5-HT synthesis by the residual 5-HT neurons.

The raphe pallidus neurons directly connect to the sympathetic preganglionic neurons which activate the brown adipose tissue to enhance heat production (Nakamura, 2011). Thus, ablation of the raphe pallidus neuron may account for mild reduction in body temperature. The ablation of central 5-HT neurons did not reduce body temperature at ZT11, 13, and 23 but decreased at ZT5 by 0.6◦C, suggesting that central 5-HT neuron ablation resulted in a decrease in body temperature during the early-mid light phase. Since energy expenditure that is associated with locomotion and thermic effect of food is much higher during the dark phase, than during the light phase (Abreu-Vieira et al., 2015). Therefore, we think that low body temperature of central 5-HT neuron-ablated mice during the light phase suggests that preserved energy expenditure associated with locomotion and/or food digestion. It is also possible that central 5-HT neuron ablation disturbs circadian change in body temperature which is regulated by the suprachiasmatic nucleus where several 5-HT receptors are expressed (Versteeg et al., 2015). Compared with our results, larger reduction in body temperature was reported on 5-HT neuron-ablated Pet1/DTR mice through intraperitoneal administration of diphtheria toxin (Cerpa et al., 2014; McGlashon et al., 2015). However, systemic administration of diphtheria toxin ablated many Pet1-positive cells, including pancreatic islet cells (Ohta et al., 2011), which results in a severe diabetic condition (Jia et al., 2014) that is usually accompanied by hypothermia. Consistently, Tph2 deficient mice showed a body temperature that was similar to wild-type mice (Solarewicz et al., 2015).

Since central 5-HT neuron ablation decreased total REM sleep time and NREM sleep episode duration during the dark phase but not during the light phase, it is unlikely that the reduction in body temperature caused a decrease in REM sleep time and NREM sleep duration in the central 5-HT neuronablated mice. However, we cannot deny the possibility that this mild reduction in body temperature during the light phase affects the transition frequency from NREM sleep to REM sleep and the arousal response in response to a novel cage. Furthermore, the sleep/wake state and body temperature affect each other (Murray et al., 2015), making a simple conclusion difficult to draw. If distinct subgroups of 5-HT neurons separately regulate sleep/wakefulness and body temperature, a future study manipulating specific projections of 5-HT neurons may elucidate this issue by separating the behavioral effects of central 5- HT neuron ablation. Central 5-HT neuron ablation tends to decrease both body temperature and body weight. Given that decreased body temperature is closely correlated with reduced energy expenditure, central 5-HT neuron-ablated mice may show a reduction in food intake. Further study is needed to examine the role of central 5-HT neurons in energy metabolism including food intake and oxygen consumption.

In summary, the current study demonstrates that the Rosa-DTR-tdTomato mouse is a useful mouse line in which the ablation efficiency of target cells is easily evaluated and suggests a crucial role of central 5-HT neurons in regulating REM sleep time, the transition from NREM sleep to REM sleep and the arousal response.

### DATA AVAILABILITY

All data are available upon request from the corresponding author.

### AUTHOR CONTRIBUTIONS

HF and MY conceived and designed the experiments. KI, HK, MK and CM performed the experiments. KI and HF analyzed the data. KS and MA contributed reagents, materials, analysis tools. KI, HF and MY wrote the paper.

### FUNDING

This work was supported by the World Premier International Research Center Initiative from MEXT to MY, JSPS KAKENHI (Grant Number 17H06095 to MY, HF; 16K15187, 17H04023, 17H05583 to HF; 26507003 to CM, HF), MEXT KAKENHI (Grant Number; 15H05935 to HF), CREST (A3A28043 to MY). Funding Program for World-Leading Innovative R&D on Science and Technology (FIRST program) from JSPS to MY, Research grant from Uehara Memorial Foundation research grant to MY and Research grant from Takeda Science Foundation research grant to MY.

### ACKNOWLEDGMENTS

We thank all Yanagisawa/Funato lab members, especially Noriko Hotta-Hirashima, Aya Ikkyu and Satomi Kanno for technical assistance, and IIIS members for discussion and comments on this manuscript.

### REFERENCES


eye movement sleep. Proc. Natl. Acad. Sci. U.S.A. 107, 18155–18160. doi: 10.1073/pnas.1012764107


ablation in transgenic mice. Nat. Biotechnol. 19, 746–750. doi: 10.1038/ 90795


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Iwasaki, Komiya, Kakizaki, Miyoshi, Abe, Sakimura, Funato and Yanagisawa. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Leptomeninges Produce Prostaglandin D<sup>2</sup> Involved in Sleep Regulation in Mice

Yoan Cherasse<sup>1</sup> \*, Kosuke Aritake<sup>1</sup>† , Yo Oishi<sup>1</sup> , Mahesh K. Kaushik<sup>1</sup> , Mustafa Korkutata1,2 and Yoshihiro Urade<sup>3</sup> \*

1 International Institute for Integrative Sleep Medicine (WPI-IIIS), University of Tsukuba, Tsukuba, Japan, <sup>2</sup> Ph.D. Program in Human Biology, School of Integrative and Global Majors, University of Tsukuba, Tsukuba, Japan, <sup>3</sup> The University of Tokyo Hospital, The University of Tokyo, Tokyo, Japan

Injection of nanomolar amounts of prostaglandin D<sup>2</sup> (PGD2) into the rat brain has dose and time-dependent somnogenic effects, and the PGD2-induced sleep is indistinguishable from physiologic sleep. Sleep-inducing PGD<sup>2</sup> is produced in the brain by lipocalin-type PGD<sup>2</sup> synthase (LPGDS). Three potential intracranial sources of LPGDS have been identified: oligodendrocytes, choroid plexus, and leptomeninges. We aimed at the identification of the site of synthesis of somnogenic PGD<sup>2</sup> and therefore, generated a transgenic mouse line with the LPGDS gene amenable to conditional deletion using Cre recombinase (flox-LPGDS mouse). To identify the cell type responsible for producing somnogenic PGD2, we engineered animals lacking LPGDS expression specifically in oligodendrocytes (OD-LPGDS KO), choroid plexus (CP-LPGDS KO), or leptomeninges (LM-LPGDS KO). We measured prostaglandins and LPGDS concentrations together with PGD synthase activity in the brain of these mice. While the LPGDS amount and PGD synthase activity were drastically reduced in the ODand LM-LPGDS KO mice, they were unchanged in the CP-LPGDS KO mice compared with control animals. We then recorded electroencephalograms, electromyograms, and locomotor activity to measure sleep in 10-week-old mice with specific knockdown of LPGDS in each of the three targets. Using selenium tetrachloride, a specific PGDS inhibitor, we demonstrated that sleep is inhibited in OD-LPGDS and CP-LPGDS KO mice, but not in the LM-LPGDS KO mice. We concluded that somnogenic PGD<sup>2</sup> is produced primarily by the leptomeninges, and not by oligodendrocytes or choroid plexus.

Keywords: prostaglandin D2, leptomeninges, lipocalin-type prostaglandin D synthase, sleep, adeno-associated virus (AAV)

# INTRODUCTION

In 1982, it was discovered that microinjecting prostaglandin (PG) D<sup>2</sup> into the preoptic area of conscious rats induces sleep (Ueno et al., 1982). It is now widely accepted that PGD<sup>2</sup> (Urade and Hayaishi, 2010; Huang et al., 2011) is one of the endogenous chemicals proposed by Ishimori (Ishimori, 1909; Kubota, 1989) and Pieron (Legendre and Pieron, 1913) more than 100 years

### Edited by:

Qinghua Liu, National Institute of Biological Sciences (NIBS), China

### Reviewed by:

Marina Bentivoglio, Università degli Studi di Verona, Italy William Wisden, Imperial College London, United Kingdom

\*Correspondence: Yoan Cherasse Cherasse.yoan.fm@u.tsukuba.ac.jp Yoshihiro Urade uradey@gmail.com

### †Present address:

Kosuke Aritake, School of Pharmaceutical Sciences, Daiichi University of Pharmacy, Fukuoka, Japan

Received: 20 July 2018 Accepted: 24 September 2018 Published: 11 October 2018

### Citation:

Cherasse Y, Aritake K, Oishi Y, Kaushik MK, Korkutata M and Urade Y (2018) The Leptomeninges Produce Prostaglandin D<sup>2</sup> Involved in Sleep Regulation in Mice. Front. Cell. Neurosci. 12:357. doi: 10.3389/fncel.2018.00357

**Abbreviations:** AAV, adeno-associated virus; CP, choroid plexus; CSF, cerebrospinal fluid; LM, leptomeninges; LPGDS, lipocalin-type PGDS; OD, oligodendrocytes; PGD2, prostaglandin D2; PGDS, PGD synthase; REM, rapid-eye movement.

ago to induce sleep. Other substances that induce sleep are cytokines (Krueger et al., 2011), adenosine (Porkka-Heiskanen et al., 1997; McCarley, 2007; Huang et al., 2011; Porkka-Heiskanen and Kalinchuk, 2011), anandamide (Garcia-Garcia et al., 2009), and peptides such as urotensin II (Huitron-Resendiz et al., 2005). These studies independently postulated that prolonged periods of wakefulness can lead to an accumulation of hypothetical endogenous substances that induce sleep. Indeed, elevated PGD<sup>2</sup> levels are found in diseases with sleep alterations such as mastocytosis and African trypanosomiasis (Roberts et al., 1980; Pentreath et al., 1990).

Prostaglandin D<sup>2</sup> is a derivative of arachidonic acid produced by two different PGDS, hematopoietic PGDS, and LPGDS. Hematopoietic PGDS is a member of the sigma class glutathione-S-transferase family (Urade et al., 1987) and synthesizes PGD2, for example, in mast cells during allergic reactions (Tanaka et al., 2000), whereas glutathione-independent LPGDS is primarily expressed in the brain (Urade and Hayaishi, 2000). Natural sleep is inhibited in wild-type and hematopoietic PGDS knockout (KO) mice, but not LPGDS KO mice, after administering the inorganic tetravalent selenium compound selenium tetrachloride (SeCl4), a specific inhibitor of PGDS activity (Islam et al., 1991; Matsumura et al., 1991; Takahata et al., 1993), demonstrating that PGD<sup>2</sup> is involved in regulating physiologic sleep (Qu et al., 2006). A recent study from our lab showed that postictal (pathological) sleep is also regulated via LPGDS-derived PGD<sup>2</sup> (Kaushik et al., 2014).

Experiments based on in situ hybridization and immunohistochemistry in rat brains demonstrated that LPGDS is expressed in three intracranial cell populations: cells of the CP, cells of the LM, and OD (Urade et al., 1993). Which of these cell population produces the PGD<sup>2</sup> involved in sleep-wake regulation is, however, unclear.

In the present study, we generated a mouse line with a loxP-site-inserted LPGDS gene that is amenable to conditional disruption by cell type-specific expression of Cre recombinase to obtain CP-LPGDS KO mice, LM-LPGDS KO mice, and OD-LPGDS KO mice. When the mice were administered the PGDS inhibitor SeCl4, only the CP- and OD-LPGDS KO mice exhibited disrupted sleep, and not the LM-LPGDS KO mice. Our findings reveal that the LM, but not the CP and OD, produce the PGD<sup>2</sup> that induces physiologic sleep.

### MATERIALS AND METHODS

### Genetic Mouse Models

Animals were handled according to the NIH Guide for the Care and Use of Laboratory Animals and in accordance with protocols approved by animal research committees at the Osaka Bioscience Institute and the International Institute for Integrative Sleep Medicine (animal protocol #16-086). All male mice (weighing 24–28 g, 10–14 weeks old) used in the present study were housed at a constant temperature (22 ± 1 ◦C) with a relative humidity of 50 ± 2% on an automatically controlled 12:12 light/dark cycle (light on at 8:00 am). A mouse line called flox-LPGDS on a C57BL/6 background with a loxP-site-inserted LPGDS gene that is conditionally disrupted by expressing Cre recombinase was generated as previously described (Kaneko et al., 2012) and used in this study. This mouse line has not been deposited to any animal repository.

### Generation of Cell Type-Specific LPGDS Knockout Animals

Several serotypes of AAV have been identified and are commonly used in neuroscience. These serotypes differ in their tropism (the types of cells they infect), making AAV a very useful system for preferentially targeting the gene of interest in specific cell types. We tested the ability for AAV to specifically infect the LM and the CP. We tested 5 different serotypes of AAVs (serotypes 2, 5, 8, 10, and 11) expressing the reporter protein mCHERRY in wildtype mice and discovered that the AAV of serotype 5 was the most efficient to target the CP, while only the serotype 8, when injected in postnatal mice (2-day-old), could infect the LM. Therefore LM-LPGDS KO mice were obtained by micro-injecting 6 µL of serotype 8 AAV-Cre into the lateral ventricle of 2-day-old flox-LPGDS mice. Briefly, neonatal mice were anesthetized by hypothermia on ice for 5 min before fixing to the pad of a stereotaxic arm. A glass micropipette with a 10- to 20-µmdiameter tip was introduced manually into the external corner of the right eye of the animal. Light pressure was applied to pass through the eye socket bone and deliver the AAV vectors into the lateral ventricle (Kalamarides et al., 2002). Using an air pressure injection system, 6 µL of viral vector from serotype 8 was delivered into the CSF over 3 min. After the injection, the pipette was kept in place for a few seconds until the CSF pressure returned to a normal level and then removed. Following the injection, the neonatal mice were kept in a cage warmed to 37◦C until they recovered and then returned to their mother.

CP-LPGDS KO mice were obtained by micro-injecting serotype 5 AAV-Cre into the lateral ventricle of adult male flox-LPGDS mice. Briefly, mice weighing 24–28 g were anesthetized with pentobarbital (50 mg/kg, i.p.), and 50 µl AAV5-Cre was stereotaxically microinjected into the left lateral ventricle (0.45 mm caudal to bregma, 1.6 mm lateral from bregma, and 1.6 mm below the dural surface) at a flow rate of 0.6 µL/min using a 100-µL Hamilton syringe and a syringe pump. Our observations indicate that the virus can disperse across ventricles and infect the remaining CP.

OD-LPGDS KO mice were obtained by cross-breeding flox-LPGDS females with transgenic mice expressing Cre recombinase under control of the nestin promoter. The nestin promoter drives Cre expression only in neural precursor cells, leading to total KO of LPGDS in the derived cells, including the OD, but not in the LM or CP.

As controls, in the case of LM- and CP-LPGDS KO mice we used littermate Flox-LPGDS mice injected with AAVs of the same serotypes and administered following the same protocol, however, these AAVs express the fluorescent protein mCHERRY instead of Cre recombinase (hence called littermate control mice). In the case of OD-LPGDS KO mice, littermate control mice are born from the same parents but did not express Cre under the control of Nestin promoter (n = 6 for each group).

# Vigilance State Assessment Using Electroencephalogram (EEG), Electromyogram (EMG) and Locomotor Activity Recordings

Vigilance states were assessed in adult male conditional LPGDS KO mice as described earlier (Huang et al., 2001). All mice subjected to EEG recordings were injected with vehicle or drug on two consecutive days. On day 1, the mice were injected with vehicle (saline, i.p.) at 10 am, and the 24-h recordings performed on day 1 were used as baseline data. On day 2, the mice were injected with SeCl<sup>4</sup> (i.p., 5 mg/kg, 10 mL/kg body weight) and EEG/EMG signals were recorded for 24 h. The EEG/EMG signals were amplified and filtered (EEG: 0.5–30 Hz, EMG: 20–200 Hz), then digitized at a sampling rate of 128 Hz, and recorded using SLEEPSIGN software (Kohtoh et al., 2008). In addition, locomotor activity was recorded with an infrared photocell sensor (Biotex, Kyoto, Japan). The vigilance state of each 10-s epoch was automatically scored offline into three stages: waking, REM sleep, and non-REM sleep, according to standard criteria (Mizoguchi et al., 2001). As a final step, defined vigilance stages were examined visually, and corrected manually when necessary. A total of 6 to 8 animals per group were used in this experiment.

### Generation of AAV Vectors

AAV-CMV-Cre and AAV-CMV-mCherry (mCHE) vectors from serotypes 5 and 8 were obtained as described earlier (Lazarus et al., 2011). Briefly, the AAV-mCHE vector plasmid was generated by replacing the humanized Renilla reniformis green fluorescent protein (hrGFP) sequence from the AAV-hrGFP vector plasmid with the mCHE sequence. Subsequently, the gene coding for mCHE was replaced with the Cre recombinase coding sequence derived by polymerase chain reaction from the pBS185 plasmid (Sauer and Henderson, 1990). The serotype 5 AAV was generated by tripartite transfection (AAV-pXR5 capsid plasmid, adenovirus helper plasmid, and AAV-vector plasmid), as well as the serotype 8 AAV (AAV-rep2/cap8 capsid plasmid, adenovirus helper plasmid, and AAV-vector plasmid) into HEK293-derived AAV-293 cells (Stratagene, catalog #240073). Three days after transfection, the virus was extracted and then quantified by quantitative polymerase chain reaction.

### Immunohistochemistry

Following all procedures, the animals were deeply anesthetized with chloral hydrate (500 mg/kg, i.p.) and perfused through the left ventricle of the heart with saline followed by neutral buffered 10% formalin. The brains were removed and placed in 20% sucrose in phosphate-buffered saline (PBS) overnight at 4◦C for cryoprotection. The brains were then frozen on dry ice and sectioned at 30 µm on a freezing microtome. Immunohistochemistry was performed on free-floating sections with a primary antibody directed at LPGDS as described previously (Beuckmann et al., 2000). Briefly, sections were rinsed in PBS, incubated in 3% hydrogen peroxide in PBS for 30 min at room temperature, and then sequentially at room temperature in 3% normal donkey serum and 0.25% Triton X-100 in PBS (PBT) for 1 h and primary antibody diluted in PBT with 0.02% sodium azide overnight. After several washes in PBS, the sections were incubated in goat anti-rabbit horseradish peroxidase-conjugated secondary antibody for 1 h, and the presence of LPGDS was revealed by a 0.05% 3,3<sup>0</sup> -diaminobenzidine/0.015% hydrogen peroxidase reaction for 10 min. Photomicrographs were obtained using a Nikon Eclipse E600 microscope coupled to a Digital Sight DS-2Mv camera. LM, CP, and OD were characterized by their localization in the tissue sections, as well as their morphology and their potential expression for LPGDS.

### Enzymatic PGDS Assay

The effect of the conditional KO of LPGDS (n = 6 for each group) in LM, CP, and OD on the brain PGDS enzymatic activity was measured by incubating 40 µg of brain extracts at 25◦C for 1 min with [1-14C]PG H<sup>2</sup> (final concentration of 40 µM) in 50 µL of 0.1 M Tris–HCl (pH 8.0) containing 1 mM dithiothreitol

FIGURE 1 | Localization of LPGDS expression in the LPGDSflox control (CTRL) mice (A), LM-LPGDS KO mice (E), CP-LPGDS KO mice (I), and OD-LPGDS KO mice (M) by immunohistochemistry. Serial coronal sections were stained with an LPGDS antibody and representative images are presented. LPGDS was not detected in the LM of LM-LPGDS KO mice (F) while it was strongly expressed in the LM of LPGDSflox control mice (B) and other cell type-specific KO mice (J,N). LPGDS staining was absent from CP cells in CP-LPGDS KO mice (K), but was expressed in the other mice (C,G,O). No LPGDS staining was detected in the brain parenchyma of the OD-LPGDS KO mice (P) where many OD, morphologically identified, were positive for LPGDS staining in the other mice (D,H,L).

(Urade et al., 1995). [1-14C]PG H<sup>2</sup> was prepared from [1- <sup>14</sup>C]arachidonic acid (2.20 GBq/mmol; PerkinElmer, Wellesley, MA, United States) as described previously (Urade et al., 1985).

### Prostaglandin D2, Enzyme Immunoassays

The brains of LM-LPGDS KO, CP-LPGDS KO, and OD-LPGDS KO mice (n = 6 for each group) were harvested and immediately frozen in liquid nitrogen. They were then homogenized in ethanol containing 0.02% HCl at pH 2.0 and centrifuged at 500 g for 20 min. <sup>3</sup>H-Labeled PGD<sup>2</sup> (60 Bq/assay; PerkinElmer) was added to the supernatant as tracer to estimate recovery. The recovery value was approximately 60%. PGD<sup>2</sup> was extracted with ethyl acetate, which was evaporated under nitrogen. The samples were then separated by HPLC (Gilson, Middleton, WI, United States) (Pinzar et al., 2000). PGD<sup>2</sup> was quantified using a PGD<sup>2</sup> enzyme immunoassay kit for this prostanoid (Cayman Chemicals, Ann Arbor, MI, United States).

### Statistical Analysis

Data are presented as the mean ± standard error of the mean (SEM). For the sleep data analysis, an unpaired Student's t-test was used to analyze the amount of time spent in the different sleep-wake states. A one- or two-way ANOVA followed by the Fisher protected least significant difference test was used to determine whether differences in LPGDS, PGD<sup>2</sup> contents, or PGDS activity in the brain of control or KO mice were statistically significant. In all cases, p < 0.05 was considered statistically significant.

# RESULTS

# Generation of Cell Type-Specific LPGDS KO Mice

To determine which intracranial cell population produces the PGD<sup>2</sup> that regulates sleep, we generated cell type-specific LPGDS KO mice for the CP, OD, and LM based on a mouse line in which the LPGDS gene is flanked by loxP sites in intron 1 and 6 and is amenable to conditional deletion by Cre recombinase (**Supplementary Figure S1**) (Kaneko et al., 2012). To obtain the LM-LPGDS KO mice, we injected 6 µL of AAV serotype 8 expressing Cre recombinase (AAV8-Cre) into the CSF of 2-day-old mice. Two months later, immunohistochemistry confirmed the absence of LPGDS expression in the LM, but not in the CP or OD (**Figures 1E–H**). To generate CP-LPGDS KO mice, we infused 50 µL of AAV-Cre, serotype 5 (AAV5- Cre), into the lateral ventricle of adult animals and 3 weeks later immunostaining confirmed the selective absence of LPGDS in the CP (**Figures 1I–L**). Moreover, we crossed our floxed LPGDS mice with mice expressing Cre under control of the rat Nestin promoter and enhancer (Tronche et al., 1999). The Nestin promoter drives Cre recombinase expression only in cells from the brain parenchyma of neuroectodermal origin (neurons, astrocyte, and OD), and among them only the OD express LPGDS (Urade and Hayaishi, 2011). This resulted to the deletion of LPGDS in the OD (OD-LPGDS KO mice; **Figures 1M–P**). In contrast, LPGDS immunostaining in the control flox-LPGDS animals revealed intense staining for LPGDS in the LM, CP, and OD (**Figures 1A–D**).

# LPGDS and PGD<sup>2</sup> Content in the Brain of LM-, CP-, and OD-LPGDS KO Mice

First, we measured LPGDS concentration in the brain of the LM-, CP-, and OD-LPGDS KO mice. We then determined the amount of PGD<sup>2</sup> in flox-LPGDS mice compared with that in control mice. Whereas the total amount of LPGDS in the brain of the CP-LPGDS KO mice remained unchanged compared with the control mice (**Figure 2A**), the amount of LPGDS was reduced by 58.3 ± 11.1% (p < 0.001) and 75.7 ± 2.7% (p < 0.001), respectively, in the OD- and LM-LPGDS KO mice compared with the control mice. These results indicate that the contribution of the CP to the production of LPGDS in the mouse brain was

### FIGURE 3 | Continued

decreased the hourly amount of sleep during the subsequent 3 h compared with saline in LPGDSflox control mice, CP-LPGDS KO mice, and OD-LPGDS KO mice, but not in LM-LPGDS KO mice. (C) Injection of SeCl<sup>4</sup> dramatically decreased the total amount of sleep during the subsequent 4 h compared with saline in LPGDSflox control mice, CP-LPGDS KO mice, and OD-LPGDS KO mice, but not in LM-LPGDS KO mice. Values are means ± SEM (n = 6–8). ∗∗P < 0.01 compared with its own control (saline).

negligible. In contrast, one-third to one-half of the brain LPGDS was produced by the OD.

We next measured PGDS enzymatic activity of the LM-, CP-, OD-LPGDS KO mice and their respective control littermates. PGDS activity in the brain of the CP-LPGDS KO mice remained unchanged compared with the control animals (p = 0.6082), whereas it was significantly reduced by 39.7 ± 6.5% (p = 0.0028) and 78.9 ± 6.9% (p = 0.0004), respectively, in the OD- and LM-LPGDS KO mice (**Figure 2B**). These findings are in good agreement with the distribution of LPGDS in the brain of the LM-, CP-, and OD-LPGDS KO mice.

Finally, we measured the total amount of PGD<sup>2</sup> contained in the brain of the cell type-specific KO animals. In the OD- and LM-LPGDS KO mice, PGD<sup>2</sup> concentrations were significantly reduced by 44.2 ± 4.9% (p = 0.004) and 55.5 ± 1.8% (p = 0.001), respectively, compared with the control flox-LPGDS animals (**Figure 2C**). On the other hand, no change in PGD<sup>2</sup> concentration was observed in the brain of CP-LPGDS KO mice compared with the control mice (p = 0.9078).

### SeCl4-Induced Insomnia Was Abolished in LM-, but Not in OD- or CP-LPGDS KO Mice

We analyzed the sleep-wake pattern of the LM-, CP-, and OD-LPGDS KO mice by measuring their EEG and EMG activity before and after an i.p. injection of the PGDS inhibitor SeCl<sup>4</sup> at a dose of 5 mg/kg at 10:00 am when mice are mostly asleep, as previously described. SeCl<sup>4</sup> is a specific inhibitor of PGDS activity that can inhibit natural sleep in wild-type and hematopoietic PGDS KO mice, but not LPGDS KO mice (Qu et al., 2006). The basal sleep behavior of the cell type-specific LM-, CP-, and OD-LPGDS KO mice was identical to that of their control littermates, except for the LM-LPGDS KO mice, which exhibited moderately reduced REM sleep activity during the daytime (−20.4 ± 5.8%, p = 0.0487; **Supplementary Figure S2**). Furthermore, we analyzed EEG/EMG recordings obtained during a 4-h period after i.p. injections of SeCl<sup>4</sup> (5 mg/kg) and vehicle. The total amount of time spent in sleep was reduced in the CP- and OD-LPGDS KO mice as well as in the control mice during the 3-h period following the injection of SeCl<sup>4</sup> compared with vehicle, but in the LM-LPGDS KO mice, the time spent in sleep after the injection of SeCl<sup>4</sup> was indistinguishable from that following the vehicle injection (**Figure 3A** and **Supplementary Figure S3**).

The changes in the time-course of sleep in the cell typespecific CP-, OD-, and LM-LPGDS KO mice during the 4 h following the i.p. injection of 5 mg/kg SeCl<sup>4</sup> is shown in

**Figure 3B**. The total amount of sleep was drastically reduced in the CP- and OD-LPGDS KO mice during the 4 h following the SeCl<sup>4</sup> injection (−50.1 ± 11.6%, p = 0.004 and −52.7 ± 8.9%, p = 0.0022, respectively) compared with the vehicle control (**Figure 3C**). The sleep amount was similarly reduced by the injection of SeCl<sup>4</sup> (−59.3 ± 3.1%, p < 0.0001) in the control animals (flox-LPGDS). By contrast, no sleep reduction was observed in the LM-LPGDS KO mice, indicating their complete insensitivity to the SeCl<sup>4</sup> injection (−2.5 ± 9.4%, p = 0.7341). These results suggest that LPGDS expressed in the LM, but not the CP or OD, is involved in the regulation of physiologic sleep.

### DISCUSSION

The administration of an LPGDS inhibitor (SeCl4) or PGD<sup>2</sup> receptor, subtype DP1, antagonist (ONO-4127Na) inhibits sleep in rats and mice, indicating that the PGD<sup>2</sup> system is crucial for maintaining physiologic sleep (Qu et al., 2006). The specific intracranial cell population expressing LPGDS and producing the PGD<sup>2</sup> responsible for the induction of sleep, however, has been debated over the last 30 years (Narumiya et al., 1982; Hayaishi, 1991; Urade et al., 1993; Vesin et al., 1995; Yamashima et al., 1997; Qu et al., 2006; Urade and Hayaishi, 2011; Zhang et al., 2017a,b). Using cell type-specific LPGDS KO mice for the LM, OD, or CP, we here identified the LM as the primary source of LPGDS involved in the regulation of sleep. Therefore, LPGDS in the LM produces PGD2, which is secreted into the CSF, where it stimulates DP<sup>1</sup> receptors as a sleep hormone. In the same pathway, adenosine is then released as a secondary sleep-promoting messenger and activates adenosine A2A receptor-expressing neurons (Lazarus et al., 2012; Zhang et al., 2017a). Through this pathway, the sleep center in the ventrolateral preoptic nucleus is subsequently activated and the histaminergic arousal center in the tuberomammillary nucleus is reciprocally regulated by the primary sleep-promoting neurons in the ventrolateral preoptic nucleus via GABAergic inhibitory projections (Sherin et al., 1998; Chung et al., 2017). Where DP<sup>1</sup> and A2A receptors are expressed, however, remains to be clarified.

The basal sleep behavior of the cell type-specific KO mice for the CP and OD was identical to that of their control littermates, while the LM-LPGDS KO mice exhibited slightly reduced REM sleep activity during the daytime. The absence of a major sleep phenotype in the conditional LPGDS KO mice is in good agreement with previous reports of global LPGDS KO mice (Hayaishi et al., 2004). The most widely accepted explanation so far involves a compensatory mechanism during embryonic development that helps the animal to recover a normal sleep pattern, as sleep is essential for the survival of animals. In the LM-PGDS KO mice, disruption of LPGDS occurs in a late-stage of development, which may explain the incomplete compensation and the moderate REM sleep abnormality.

While using AAVs expressing Cre recombinase permitted to delete the expression of LPGDS in the CP and the LM, we cannot exclude the possibility that other cell types might also have been infected and their copy of lpgds gene deleted. Indeed, other cells such as tanycytes -a special ependymal cells found in the third ventricle of the brain and on the floor of the fourth ventricle- or ependymocytes -a type of glial cell forming the ependyma, a thin neuroepithelial lining the ventricular systemwere most likely in contact with the AAVs and might have been infected. However, it is unclear if the serotypes of AAVs used in this study can effectively infect such ependymal cells. Furthermore, even if these and other cell types have also been infected, then started to express Cre recombinase, and were deleted of their lpgds gene, we believe that the effect would have remained unnoticed. To our knowledge, these cells have never been reported to express LPGDS and therefore animals should not be impacted by the potential ectopic expression of Cre recombinase.

# CONCLUSION

Lipocalin-type PGDS is implicated in the production of PGD<sup>2</sup> that is involved in the regulation of physiologic sleep. Our findings in cell type-specific LPGDS KO mice demonstrate that the somnogenic PGD<sup>2</sup> is primarily produced by LPGDS in the LM.

### DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

### AUTHOR CONTRIBUTIONS

YC and YU designed the research and wrote the paper. YC and KA performed the research. YO contributed the transgenic mice. YC, KA, MKK, and MK analyzed the data.

# FUNDING

This work was supported by grants from JSPS KAKENHI Grant Nos. JP16K18698 (YC), JP18H02534 (YO), and JP16H01881 (YU), a grant from the Japan Foundation for Applied Enzymology (YO), and the World Premier International Research Center Initiative (WPI) from MEXT.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fncel. 2018.00357/full#supplementary-material

### REFERENCES

fncel-12-00357 October 9, 2018 Time: 19:56 # 7



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Cherasse, Aritake, Oishi, Kaushik, Korkutata and Urade. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Whole-Brain Monosynaptic Afferent Projections to the Cholecystokinin Neurons of the Suprachiasmatic Nucleus

### Xiang-Shan Yuan† , Hao-Hua Wei † , Wei Xu, Lu Wang, Wei-Min Qu, Rui-Xi Li and Zhi-Li Huang\*

*Department of Pharmacology, Department of Anatomy, School of Basic Medical Sciences, State Key Laboratory of Medical Neurobiology, Institutes of Brain Science and Collaborative Innovation Center for Brain Science, Fudan University, Shanghai, China*

### Edited by:

*Takeshi Sakurai, University of Tsukuba, Japan*

### Reviewed by:

*Yoshitaka Fukada, The University of Tokyo, Japan William Wisden, Imperial College London, United Kingdom Arisa Hirano, University of Tsukuba, Japan*

> \*Correspondence: *Zhi-Li Huang huangzl@fudan.edu.cn*

*†These authors have contributed equally to this work*

### Specialty section:

*This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience*

Received: *07 August 2018* Accepted: *16 October 2018* Published: *05 November 2018*

### Citation:

*Yuan X-S, Wei H-H, Xu W, Wang L, Qu W-M, Li R-X and Huang Z-L (2018) Whole-Brain Monosynaptic Afferent Projections to the Cholecystokinin Neurons of the Suprachiasmatic Nucleus. Front. Neurosci. 12:807. doi: 10.3389/fnins.2018.00807* The suprachiasmatic nucleus (SCN) is the principal pacemaker driving the circadian rhythms of physiological behaviors. The SCN consists of distinct neurons expressing neuropeptides, including arginine vasopressin (AVP), vasoactive intestinal polypeptide (VIP), gastrin-releasing peptide (GRP), cholecystokinin (CCK), and so on. AVP, VIP, and GRP neurons receive light stimulation from the retina to synchronize endogenous circadian clocks with the solar day, whereas CCK neurons are not directly innervated by retinal ganglion cells and may be involved in the non-photic regulation of the circadian clock. To better understand the function of CCK neurons in non-photic circadian rhythm, it is vital to clarify the direct afferent inputs to CCK neurons in the SCN. Here, we utilized a recently developed rabies virus- and Cre/loxP-based, cell type-specific, retrograde tracing system to map and quantitatively analyze the whole-brain monosynaptic inputs to SCN CCK neurons. We found that SCN CCK neurons received direct inputs from 29 brain nuclei. Among these nuclei, paraventricular nucleus of the hypothalamus (PVH), paraventricular nucleus of the thalamus (PVT), supraoptic nucleus (SON), ventromedial nucleus of the hypothalamus, and seven other nuclei sent numerous inputs to CCK neurons. Moderate inputs originated from the zona incerta, periventricular hypothalamic nucleus, and five other nuclei. A few inputs to CCK neurons originated from the orbital frontal cortex, prelimbic cortex, cingulate cortex, claustrum, and seven other nuclei. In addition, SCN CCK neurons were preferentially innervated by AVP neurons of the ipsilateral PVH and SON rather than their contralateral counterpart, whereas the contralateral PVT sent more projections to CCK neurons than to its ipsilateral counterpart. Taken together, these results expand our knowledge of the specific innervation to mouse SCN CCK neurons and provide an important indication for further investigations on the function of CCK neurons.

Keywords: cholecystokinin neuron, circadian rhythm, monosynaptic inputs, rabies viruses, suprachiasmatic nucleus

# INTRODUCTION

The suprachiasmatic nucleus (SCN) has been widely known as a central pacemaker that orchestrates physiological and behavioral circadian rhythms that need daily synchronization to stay in phase with the 24-h solar cycle. Light information is transmitted to the SCN by the retinohypothalamic tract (RHT), which is a monosynaptic projection from the retina to the SCN (Pickard, 1982; Moore et al., 2002; Hannibal and Fahrenkrug, 2006). The SCN is a small heterogeneous structure that contains diverse subpopulations of neurons expressing distinct neuropeptides, including arginine vasopressin (AVP), vasoactive intestinal polypeptide (VIP), gastrin-releasing peptide (GRP), calretinin, cholecystokinin (CCK), and so on (Abrahamson and Moore, 2001; LeSauter et al., 2002; Moore et al., 2002; Hannibal et al., 2010). The AVP, VIP, and GRP neurons in the SCN have been elucidated as retinorecipient neurons that receive direct inputs from the RHT, indicating that these neurons play an important role in mediating the effects of light on the circadian clock (Abrahamson and Moore, 2001; Antle and Silver, 2005; Fernandez et al., 2016). The CCK neuron was also identified as a distinct cell type in the SCN (Morin, 2013), but it did not receive projections from the RHT and was not activated by light (Hannibal et al., 2010). The circadian rhythm is principally synchronized by light, but it can also be regulated by non-photic signals, such as exercise, feeding, and temperature (Buxton et al., 2003; Escobar et al., 2009; Buhr et al., 2010; Burgess et al., 2010; Morris et al., 2012). Thus, the literature suggests that SCN CCK neurons may be involved in the nonphotic regulation of circadian rhythm. Previous studies have shown that feeding and temperature signals were mediated by the diencephalon, including the preoptic area, dorsomedial nucleus of the hypothalamus (DMH), arcuate nucleus of the hypothalamus (Arc), and paraventricular nucleus of the thalamus (PVT) (Tan et al., 2016; Hume et al., 2017; Zhao et al., 2017). However, it is not clear whether these nuclei send direct projections to CCK neurons of the SCN to entrain the circadian clock. Therefore, identifying the afferent circuits of SCN CCK neurons is critical to comprehensively understand the function of these neurons.

Traditional tracing approaches with non-specific tracers have successfully classified three major afferent inputs to the SCN: the RHT, geniculohypothalamic tract (GHT), and a pathway from the raphe nuclei (Hannibal and Fahrenkrug, 2006; Fernandez et al., 2016). However, a recent study used immunohistochemical staining to reveal that CCK neurons did not receive projections form these three major afferent pathways (Hannibal et al., 2010). The staining had obvious flaws in term of non-specificity and non-systematicity, making it difficult to assess the specific afferent distributions within the SCN. To overcome these limitations and detect all of the specific afferent inputs to SCN CCK neurons from the whole brain, it is necessary to adopt a specific, viral-mediated tracing system.

In recent years, a genetically modified rabies virus combined with Cre-LoxP technology, has been used to trace monosynaptic inputs (Wickersham et al., 2007; Lerner et al., 2015) and characterize the whole-brain presynaptic neurons of a specific neuron type within a complicated neural network (Do et al., 2016; Hu et al., 2016; Su et al., 2018). In our study, we utilized this viral tracing system to map the whole-brain afferent inputs to SCN CCK neurons. We found 29 afferent brain nuclei, including several important nuclei that integrated circadian, ingestive, and osmotic information to SCN CCK neurons. Our quantitative results provide numerous evidence for the structural framework of SCN CCK neurons and can guide further investigations of neuronal pathways that mediate functions of CCK neurons in the SCN.

# MATERIALS AND METHODS

### Animals

Pathogen-free, adult CCK-ires-Cre mice (Taniguchi et al., 2011) of either sex (10–12 weeks, 25–28 g) on a C57BL/6J background and wild-type littermates were used in these experiments. CCKires-Cre mice express Cre recombinase under the control of the CCK gene promoter. The animals were housed in individual cages at constant temperature (22 ± 0.5◦C) and relative humidity (60 ± 2%) on an automatically controlled 12:12 light:dark cycle (lights on at 7 a.m.; 100 lux intensity) (Zhang et al., 2017) with free access to food and water. All of the animal studies were performed in accordance with protocols approved by the Committee on the Ethics of Animal Experiments of Fudan University Shanghai Medical College (permit number: 20110307- 049). Every effort was made to minimize the number of animals used as well as any pain or discomfort experienced by the subjects.

### Viruses and Surgery

AAV-CAG-DIO-TVA-GFP (Adeno-associated virus (AAV) 2/9 serotype; titer 1.7 × 10<sup>13</sup> genomic copies/ml), AAV-CAG-DIO-RG (AAV2/9 serotype; titer 6.8 × 10<sup>12</sup> genomic copies/ml), and EnvA-pseudotyped, glycoprotein (RG)-deleted, dsRedexpressing rabies virus (RV-EvnA-DG-dsRed; RV 5.0 × 10<sup>8</sup> genomic copies/ml) were packaged and provided by BrainVTA (Wuhan, China). The detailed production and concentration procedures for the modified rabies virus were conducted as previously described (Pollak Dorocic et al., 2014; Hu et al., 2016).

Surgical procedures were carried out according to previous studies (Hu et al., 2016; Yuan et al., 2017). Briefly, naïve mice were anesthetized with chloral hydrate (360 mg/kg) and placed in a stereotaxic apparatus (RWD Life Science, Shenzhen, China). After exposing the skull and drilling a small hole, a glass micropipette was placed above the SCN [0.5 mm posterior

**Abbreviations:** AH, anterior hypothalamic area; AI, agranular insular cortex; Arc, arcuate nucleus of the hypothalamus; BNST, bed nucleus of stria terminalis; Cg, cingulate cortex; DMH, dorsomedial nucleus of the hypothalamus; DR, dorsal raphe nucleus; IGL, intergeniculate leaf; LC, locus coeruleus; LH, lateral hypothalamic nucleus; LPO, lateral preoptic nucleus; LS, lateral septum; MnR, median raphe nucleus; MPO, medial preoptic nucleus; MO, motor cortex; OFC, orbital frontal cortex; PAG, periaqueductal gray; Prl, prelimbic cortex; PVH, paraventricular nucleus of the hypothalamus; PVT, paraventricular nucleus of the thalamus; RCh, retrochiasmatic area; SCN, suprachiasmatic nucleus; SON, supraoptic nucleus; SFO, subfomical organ; TC, tuber cinereum area; VDB, nucleus of the vertical limb of the diagonal band; VMH, ventromedial nucleus of the hypothalamus; VMPO, ventromedial preoptic nucleus; ZI, zona incerta.

and 0.2 mm lateral from bregma; 5.1 mm ventral from the pial surface]. First, two helper AAVs (AAV-CAG-DIO-TVA-GFP and AAV-CAG-DIO-RG were mixed at a 1:1 ratio in 50 nL) were injected into the SCN at 0.01 µL/min using a microsyringe pump controller (WPI, Sarasota, FL). To allow diffusion of the virus, the micropipette was not retracted until 15 min after the end of the injection. Three weeks later, 100 nL RV-EnvA-DGdsRed was similarly injected into the same site. The scalp wound was closed with surgical sutures, and each mouse was kept in a warm environment until it resumed normal activity as previously described (Luo et al., 2018b).

# Single-Cell RT-PCR

At 1–2 weeks after only helper AAV injections, CCK-Cre mice were anesthetized and perfused transcardially with icecold modified aCSF saturated with 95% O<sup>2</sup> and 5% CO<sup>2</sup> and containing (in mM): 215 sucrose, 26 NaHCO3, 10 glucose, 3 MgSO4, 2.5 KCl, 1.25 NaH2PO4, 0.6 mM Na-pyruvate, 0.4 ascorbic acid, and 0.1 CaCl2. Brains were then rapidly removed, and acute coronal slices (300µm) containing the SCN were cut on a vibratome (VT1200, Leica) in ice-cold modified aCSF. Next, slices were transferred to a holding chamber containing normal recording aCSF (in mM): 125 NaCl, 26 NaHCO3, 25 glucose, 2.5 KCl, 2 CaCl2, 1.25 NaH2PO<sup>4</sup> and 1.0 MgSO4, and allowed to recover for 30 min at 32◦C. Then, slices were maintained at room temperature (RT) for 30 min before recording.

During recording, slices were submerged in a recording chamber superfused with aCSF (2 mL/min) at 30–32◦C. Slices were visualized using a fixed-stage upright microscope (BX51W1, Olympus, Japan) equipped with a 40× water immersion objective and an infrared-sensitive CCD camera. The CCK neurons were identified based on their GFP expression in the SCN and the cytosolic content was aspirated into the patch pipette, and expelled into a 200 µL PCR tube as described previously (Luo et al., 2018b). The single-cell reverse-transcription PCR (RT-PCR) protocol was designed to detect the presence of mRNA coding for CCK. Reverse transcription and PCR amplification were performed with gene-specific multiplex primer using the SuperScript III One-Step RT-PCR kit (catalog number: 12574018, ThermoFisher). The reaction was performed as follows: 30 min at 55◦C, 2 min at 94◦C; 70 cycles of 20 s at 94◦C, 30 s at 61◦C, and 25 s at 68◦C; and 5 min at 68◦C. The PCR products were visualized by electrophoresis in agarose gels (1.5%) with ethidium bromide. The expected size of each final PCR product is CCK 215 bp. The specific primers for CCK gene were custom designed and synthesized (Biosune Biotechnology, Shanghai). CCK-F primer, 5 ′ to 3′ : AAGCCATGAAGAGCGGCGTAT; CCK-R primer, 5′ to 3 ′ : GCGGACCTGCTGGATGTATCG.

### Histology and Image Analysis

One week after injection of the rabies virus, mice were perfused with 10 mL saline, followed by 100 mL of 4% paraformaldehyde in 0.1 M phosphate buffer (PB, pH 7.4). The brains were removed, post-fixed for 4–6 h at 4◦C, and then cryoprotected in 10, 20, and 30% sucrose in 0.1 M PB at 4◦C until they sank. Tissues were embedded in OCT compound, and stored at −80◦C before use. The brains were coronally sectioned at a thickness of 30 µm on a cryostat (Leica 1950) in three series and were collected in 0.01 M phosphate-buffered saline (PBS, pH 7.4). Every third section was counterstained with DAPI (1:3000, Sigma-Aldrich, USA). The stained sections were then coverslipped with Fluoromount-GTM (Southern Biotech).

Images of whole-brain sections were captured using a 20 × objective on an Olympus microscope (Olympus VS-120, Tokyo, Japan). Further imaging analyses were performed using Olympus analysis software and ImageJ software. The number of afferent neurons and area of each input regions were counted by ImageJ automatically, and we normalized the data in every input region. Quantification of the subregion boundaries was based on the mouse brain atlas of Paxinos and Franklin (Paxinos and Franklin, 2001). The proportion of cell in each nucleus was calculated as the ratio of the normalized number of dsRed-labeled cells in each nucleus to the total number of dsRed-labeled cells, and the cell density was defined as the number of dsRed-labeled cells per unit area within each nucleus. According to the proportion of cells in each nucleus, we defined three grades of afferent inputs as numerous input (over 4%), moderate input (1–4%), and a few input (<1%).

To characterize the inputs from the paraventricular nucleus of the hypothalamus (PVH), and supraoptic nucleus (SON), we immunostained the brain slices containing the two areas with AVP and oxytocin (Oxt) antibodies according to the following protocol (Yuan et al., 2017). Brain sections from the PVH and SON were incubated overnight at 4◦C in PBS containing 5% normal donkey serum (v/v), 0.3% Triton X-100 (v/v), and the following primary antibodies: rabbit anti-AVP (1:2000, catalog number: 20069, Immunostar) and rabbit anti-Oxt (1:2000, catalog number: ab212193, Abcam). After several washes in PBS, the sections were incubated with Alexa Fluor-conjugated IgG antibody (Invitrogen) at room temperature for 2 h. The sections were then incubated in PBS containing DAPI. Finally, sections were coverslipped with Fluoromount-G TM (Southern Biotech). Fluorescence images were collected using a Leica confocal system.

# RESULTS

### Mapping Monosynaptic Inputs Onto SCN CCK Neurons Using a Rabies-Based Tracing System

To identify the monosynaptic afferent inputs to SCN CCK neurons, we used rabies virus-mediated, trans-synaptic, retrograde tracing on a transgenic mouse line expressing Cre recombinase in CCK neurons (Taniguchi et al., 2011). This retrograde, viral tracing system has been shown to label monosynaptic inputs to the desired starter cells with high specificity (Lerner et al., 2015; Do et al., 2016; Hu et al., 2016). In CCK-Cre mice, we applied a genetically engineered viral system to map the whole-brain afferent inputs to SCN CCK neurons. We first co-expressed the avian receptor TVA and the rabies glycoprotein G (RG) in SCN CCK neurons, which was achieved by a unilateral injection of two AAV-DIO helper viruses (AAV-CAG-DIO-TVA-GFP and AAV-CAG-DIO-RG) into the SCN (**Figure 1A**) of CCK-Cre mice. Three weeks later, RV-EnvA-DG-dsRed was injected into the same location, where it only infected cells expressing TVA and required RG to spread retrogradely into presynaptic cells (**Figures 1A–C**). After 1 week, the starter neurons could be characterized by the co-expression of RV-EnvA-DG-dsRed and AAV-CAG-DIO-TVA-GFP. We used single-cell RT-PCR to detect the presence of CCK mRNA in GFP-positive neurons in the SCN (7 of 7, n = 3 CCK-Cre mice), and confirmed that these neurons are CCK-expressing neurons (**Figures 1D,E**). We found that the starter neurons were restricted to the ventral part of the rostral and middle SCN, ipsilateral to the injection site (**Figures 1F,H**). Moreover, we observed numerous neurons in the SCN that were dsRed-positive but did not express GFP, demonstrating the presence of direct, monosynaptic input from other types of SCN neurons to the CCK neurons (**Figure 1H**). Upon conducting the same injection protocol in naïve animals, we did not detect the expression of GFP or dsRed in these wild-type mice that did not express Cre recombinase (**Figure 1G**). Thus, this technique could be used to map the whole-brain, monosynaptic afferent inputs to SCN CCK neurons (**Figure 1C**).

### SCN CCK Neurons Received Direct Inputs From 29 Brain Nuclei

To investigate the whole-brain monosynaptic input areas to SCN CCK neurons, we examined the serial coronal wholebrain sections (**Figure 2**). In CCK-Cre mice injected with the three viruses, those neurons expressing only dsRed were the monosynaptic inputs to SCN CCK neurons. The dsRedlabeled presynaptic neurons were observed in 29 brain nuclei throughout the telencephalon, diencephalon, and brainstem (**Figure 2**). Furthermore, we measured the number of labeled neurons and the labeling density in individual brain areas to more quantitatively describe the whole-brain distribution of afferent input to SCN CCK neurons. The locations of labeled neurons were determined using a standard mouse brain atlas (Paxinos and Franklin, 2001). To correct potential bias, the number of dsRed-labeled cells in each nucleus was further normalized by the number of starter neurons. A list of wholebrain inputs was generated for SCN CCK neurons (**Figure 3**), which consisted of 29 different nuclei. Among these, CCK neurons received numerous afferent projections from 11 nuclei: PVH, PVT, SON, ventromedial nucleus of the hypothalamus (VMH), Arc, ventromedial preoptic nucleus (VMPO), medial preoptic nucleus (MPO), tuber cinereum area (TC), DMH, anterior hypothalamic area (AH), retrochiasmatic area (RCh). In addition, we found seven nuclei with moderate input nuclei contained the zona incerta, periventricular nucleus of the hypothalamus, lateral septum (LS), bed nucleus of stria terminalis (BNST), lateral hypothalamus (LH), lateral preoptic nucleus (LPO), and the nucleus of the vertical limb of the diagonal band (VDB). Finally, we observed 11 a few inputs to CCK neurons, which originated from the orbital frontal cortex (OFC), prelimbic cortex (Prl), cingulate cortex (Cg), claustrum, posterior hypothalamic area (PH), dorsal raphe nucleus (DR), intergeniculate leaf (IGL), periaqueductal gray (PAG), locus coeruleus (LC), subfornical organ (SFO), and median raphe nucleus (MnR). Although most of the dsRed-labeled presynaptic neurons were located in the hemisphere ipsilateral to the starter neurons, the contralateral PVT had stronger projections to SCN CCK neurons than the ipsilateral PVT (712.4 ± 565.9 vs. 410.5 ± 304.5 cells/mm<sup>2</sup> , **Figure 4B**). Furthermore, in comparison with previous studies, our study revealed five novel projections to CCK neurons, namely, the VDB, LPO, Prl, OFC, and Cg (**Supplemental Figure 1**) (Moga and Moore, 1997; Krout et al., 2002).

### Preferential Inputs to SCN CCK Neurons Were From Diencephalon, Not From the Three Major SCN Afferent Pathways

We found that SCN CCK neurons integrated the majority of their inputs (91.97%) from the diencephalon, specifically from the VMH (16.47%), PVH (11.78%), Arc (9.37%), VMPO (7.66%), PVT (6.16%), MPO (6.23%), and SON (4.67%) (**Figures 3**, **4**). Although the VMH provided the largest number of inputs to CCK neurons, the SON was the most densely labeled area (**Figure 3**). Interestingly, we detected no dsRed-labeled neurons in both ipsilateral and contralateral retina (data not shown), indicating that SCN CCK neurons did not receive direct input from the major photic input pathway, the RHT (Morin, 2013). Furthermore, we observed only sparse dsRed labeling in the IGL (0.26%) and MnR (0.05%) (**Figures 3**, **4F**), which are part of the GHT and a serotonin pathway from the median raphe nuclei, respectively. In contrast with our results, these two major pathways were previously shown to send robust afferent inputs to the SCN (Morin, 2013). Our results also differed from the findings of Hannibal and colleagues, in which immunostaining did not reveal any direct innervation from the IGL and raphe nucleus to the CCK neurons of the SCN (Hannibal et al., 2010).

# SCN CCK Neurons Received Inputs From AVP Neurons in the PVH and SON

The PVH and SON of the hypothalamus are two important integrative brain structures that coordinate responses to perturbations in water balance and regulate maternal physiology through the release of the neuropeptide hormones AVP and Oxt (Qiu et al., 2011). In this study, we found that while SCN CCK neurons received significant afferent inputs from bilateral PVH (**Figures 2**, **4E**), the ipsilateral PVH had more projections to SCN CCK neurons than the contralateral PVH (2058.0 ± 418.4 vs. 841.3 ± 341.0 cells/mm<sup>2</sup> , n = 4, **Figures 5A,B**). Immunostaining for AVP showed that many PVH AVP neurons sent direct inputs onto SCN CCK neurons, whereas Oxt immunostaining demonstrated that few PVH Oxt neurons sent direct projections to CCK neurons (**Figures 5C,D**). Moreover, our results revealed that bilateral SON sent direct projections to the CCK neurons in the SCN, with more monosynaptic inputs from the ipsilateral SON than the contralateral SON (2598.7 ± 369.5 vs. 645.6 ± 104.1 cells/mm<sup>2</sup> , n = 4), similar to the afferent pattern from the PVH. The immunostaining data for AVP and Oxt in the SON showed that many SON AVP neurons sent direct inputs onto SCN CCK neurons, whereas few

FIGURE 1 | Monosynaptic afferent tracing on SCN CCK neurons with a rabies virus-based, retrograde tracing system. (A) Design of viral vectors for RV-mediated trans-synaptic retrograde tracing and experimental timeline for unilateral injections of AAV and RV in the SCN of CCK-Cre mice. (B) Schematic illustration of the starter neuron (yellow, B) after AAV helper virus (green) and rabies virus (red) injection into SCN CCK neurons. (C) Schematic illustration of whole-brain, monosynaptic input (red) to CCK starter neurons (yellow). (D) A typical section of an CCK-Cre mouse injected with helper AAVs into the SCN for patch-clamp electrophysiology shows an GFP-expressing neuron for recording, the patch pipette attached to the membrane of the recorded neuron in phase contrast (upper panel), and the recorded neuron with fluorescent contrast (lower panel). Scale bar, 15µm. (E) Representative result from a single-cell RT-PCR reaction confirming the CCK phenotype in the GFP labeled neuron of the SCN. (F,G) Fluorescence images showing RV-labeled neurons (red) in CCK-Cre mice (F), but not in wild-type mice (G). Scale bar, 500µm. (H) Fluorescence images showing that starter neurons (yellow) infected with AAV helper virus and RV were restricted to the unilateral SCN. Scale bar, 50µm. Data were obtained from independent experiments.

FIGURE 2 | Representative coronal sections showing labeling of monosynaptic inputs to SCN CCK neurons. For some sections, only one hemisphere is shown. Scale bar, 500µm. Data were obtained from four independent experiments. Abbreviations of the brain regions used are the following: AH, anterior hypothalamic area; Arc, arcuate nucleus of the hypothalamus; BNST, bed nucleus of stria terminalis; Cg, cingulate cortex; Cl, claustrum; DR, dorsal raphe nucleus; IGL, intergeniculate leaf; LPO, lateral preoptic nucleus; LS, lateral septum; MPO, medial preoptic nucleus; PAG, periaqueductal gray; Pe, periventricular nucleus of the hypothalamus; Prl, prelimbic cortex; PVH, paraventricular nucleus of the hypothalamus; PVT, paraventricular nucleus of the hypothalamus; RCh, retrochiasmatic area; SCN, suprachiasmatic nucleus; SON, supraoptic nucleus; TC, tuber cinereum area; VDB, nucleus of the vertical limb of the diagonal band; VMH, ventromedial nucleus of the hypothalamus; VMPO, ventromedial preoptic nucleus; ZI, zona incerta.

SON Oxt neurons sent direct projections to SCN CCK neurons (**Figures 6A–D**). These results can provide a foundations for further investigations on links between AVP- and Oxt-mediated ingestive behavior or osmotic stability and circadian rhythm in the SCN.

### DISCUSSION

To understand how SCN CCK neurons modulate nonphotic behaviors, it is crucial to explore the afferent inputs that influence the activity of CCK neurons. In the present study, we clarified the whole-brain, direct, monosynaptic inputs to CCK neurons in the SCN using cell-type specific infection and retrograde spread of a modified rabies virus. We efficiently characterized the distribution of whole-brain input to SCN CCK neurons, which preferentially originated from a wide range of nuclei in the diencephalon, such as the VMH, Arc, MPO, PVT, PVH, SON, TC, and DMH. Moreover, the afferent pattern to CCK neurons in the SCN was strongly ipsilateral, with few contralateral projections other than the PVT. In addition, our results revealed specific inputs to the SCN and provided a comprehensive map of the presynaptic patterns that may control SCN CCK neuron activity.

### Comparison Between Specific Trans-Synaptic Tracing and Traditional Tracing

The neural connectivity of the SCN has been extensively investigated due to its critical role in the circadian rhythm. Previous investigations used conventional tracing techniques, such as non-specific tracers, multisynaptic pseudorabies viruses, and immunohistochemistry (Moga and Moore, 1997; Abrahamson and Moore, 2001; Krout et al., 2002; Morin, 2013; Fernandez et al., 2016). Previous tract-tracing studies have consistently revealed that the SCN received input from three major pathways: direct visual input from the retina through the RHT, secondary visual input from the IGL of the lateral geniculate complex through the GHT, and a pathway from the midbrain raphe nuclei (Hannibal and Fahrenkrug, 2006). However, these tracing methods do not allow for the identification of specific afferents of SCN CCK neurons. Using the rabies virus-based approach of trans-synaptic retrograde tracing, we found that the retina did not provide direct input to SCN CCK neurons. These results can explain why CCK neurons were not light-responsive as evaluated by the induction of c-Fos (Hannibal et al., 2010). Moreover, we revealed that SCN CCK neurons received minor projections from the IGL and raphe nuclei, unlike previous reports showing robust input projections from these brain nuclei (Krout et al., 2002) or no inputs at all (Hannibal and Fahrenkrug, 2006). In addition, we detected retrograde labeling in several brain areas, namely the Prl, OFC, Cg, VDB, and LPO, which had not been previously identified as SCN inputs. These novel observations were likely due to the greater sensitivity of the trans-synaptic tracing method using the modified rabies virus. Traditional tracttracing studies tended to inject small amounts of tracer into only a part of the SCN, in an attempt to avoid the potential confounding factors of tracer spillover and tracer pickup by fibers-of-passage. Different research groups have reported inconsistent labeling patterns, especially from brain areas that

FIGURE 4 | Representative images of selected regions with monosynaptic inputs to SCN CCK neurons. Data were obtained from four independent experiments. (A) MPO; (B) PVT; (C) SCN; (D) SON; (E) PVH; (F) IGL; (G) Arc, and VMH. Scale bar, 50µm.

contributed a few and moderate inputs to the SCN (Moga and Moore, 1997; Krout et al., 2002). Using the more precise and efficient viral-mediated tracing method, our findings should provide a comprehensive map of the presynaptic patterns that control SCN CCK neurons.

# Implications for SCN CCK Neurons in Non-photic Circadian Rhythm

The SCN is widely considered to be the master circadian pacemaker necessary for physiological behaviors. Although light is the most potent zeitgeber to this master circadian oscillator, circadian clocks are entrained by both photic and non-photic signals. Previous studies found that AVP, VIP, and GRP neurons in the SCN received direct input from the retina, as well as indirect photic input from the IGL to integrate the non-imageforming photic information for circadian photo-entrainment (Abrahamson and Moore, 2001; Fernandez et al., 2016). As mentioned above, CCK neurons did not receive projections from the retina and were not directly entrained by light. However, our results revealed that there were robust dsRed-labeled neurons in the SCN, which did not co-express TVA-GFP, indicating that SCN CCK neurons received local innervations from intrinsic neurons in the SCN. It has been reported that SCN CCK neurons were innervated by the processes of SCN VIP neurons in mice and received contacts from SCN calbindin neurons in the ventral part of the central SCN in hamster (LeSauter et al., 2002, 2009; Hannibal et al., 2010). Therefore, SCN CCK neurons may relay optic or circadian signals from the VIP and calbindin neurons in the SCN and transmit these signals to mediate clock information.

Circadian entrainment can also occur in the absence of light, suggesting that non-photic signals can phase-shift and synchronize circadian clocks (Morris et al., 2012). Such non-photic signals include exercise, feeding, and temperature (McArthur et al., 1991; Damiola et al., 2000; Buxton et al., 2003; Escobar et al., 2009; Buhr et al., 2010). Previous findings showed that a majority of the preoptic neurons were activated by heat exposure and were located in the VMPO (Tan et al., 2016; Abbott and Saper, 2018). Our results revealed that the VMPO region had a robust projection to SCN CCK neurons; the interconnection

between the two areas supports the idea of circadian entrainment by temperature. Moreover, hypothalamic nuclei, including PVH, Arc, DMH, VMH, and the lateral hypothalamus, were shown to govern energy balance via both metabolic and behavioral responses (Myers and Olson, 2012). Particularly, the VMH and Arc were thought to play a role in controlling food intake and peripheral metabolism (Dhillon et al., 2006; Zhang and van den Pol, 2016). Here, we revealed that SCN CCK neurons received numerous afferent inputs from the PVH, VMH, and Arc; thus, our results may provide a circuit-based explanation of how feeding can entrain circadian rhythm via SCN CCK neurons.

FIGURE 6 | Immunofluorescence images showing dsRed-labeled afferent neurons with AVP and Oxt in the bilateral SON. Immunofluorescence images showing that some dsRed-labeled neurons were co-localized with AVP neurons in the ipsilateral (A) and the contralateral SON (B). Immunofluorescence images showing that dsRed-labeled neurons rarely co-localized with Oxt neurons in the ipsilateral (C) and the contralateral SON (D). Scale bar, 50µm. Data were obtained from four independent experiments.

# Implications for SCN CCK Neurons in Ingestive Behavior

Ingestive behavior in a natural environment is essential for survival, and is dependent on hunger signals, which involve interoceptive sensory neurons that monitor metabolic level and consequently regulate food-seeking and consumption behaviors (Trivedi, 2014; Cheng et al., 2018). Previous research has revealed many nuclei in the diencephalon as important sites that respond to feed restriction and regulate of ingestive behaviors, including the PVT, PVH, Arc, and VMH (Dhillon et al., 2006; Atasoy et al., 2012; Zhang and van den Pol, 2016; Jarvie et al., 2017; Ong et al., 2017; Luo et al., 2018a). Food deprivation incresed the synthesis of neuropeptide Y, one of the most potent orexigenic peptides, in neurons of the Arc (Sahu et al., 1988; Beck et al., 1992), and acute and chronic calorie restriction increased the activity of Arc neurons (Jarvie et al., 2017). Moreover, neurons in the VMH were inhibited by food deprivation (Kosta et al., 1987; Sternson et al., 2005; Flanagan-Cato et al., 2008). For the PVT, the effect of food deprivation are controversial. Nakahara et al. (2004) observed a remarkable increase in c-Fos expression in the PVT after restricted feeding for 2 h, suggesting increased neuronal activity by food deprivation. However, Zhang and van den Pol (2017) showed increased activity of ZI GABA neurons after food deprivation for 24 h. Since ZI GABA neurons inhibit PVT glutamatergic neurons, the findings by Zhang and van den Pol suggest that the activity of PVT neurons was inhibited by food deprivation. Together, these data indicate that feed restriction can change the activity of neurons in the Arc, VMH, and PVT. The Arc receives the hunger signals via the neural circuits that integrate visceral signals of energetic state and consequently regulate physiology and behavior via efferent outputs of the Arc neurons, consisting of the agouti-related peptide (AGRP) neurons and pro-opiomelanocortin neurons. Our results showed that the Arc had abundant projections to SCN CCK neurons; this finding is consistent with previous investigations using nonspecific retrograde tracers that also showed that the SCN received major afferent inputs from the Arc. It has been speculated that the signal inputs from the Arc to CCK neurons could explain the circadian rhythm of feeding behavior, because CCK knockout mice eat more food than control animals during the light period and less food during the dark period (Lo et al., 2008). In addition, increasing evidence has shown that the PVT played a pivotal role in integrating information related to appetitive motivation, feeding, aversion, and anxiety (Hsu et al., 2014; Millan et al., 2017; Cheng et al., 2018). Our results, showing monosynaptic input from the PVT to SCN CCK neurons, suggest that CCK neurons could be involved in energy metabolism and anxietylike behavior (Lo et al., 2008). Interestingly, we observed a preferential input from the contralateral PVT to CCK neurons, in contrast to the primarily ipsilateral pattern of innervation from the Arc, PVH, and VMH. The function of this contralateraldominant innervation pattern between SCN CCK neurons and the PVT needs to be further studied.

### Implications for SCN CCK Neurons in Osmotic Stability

Systemic osmoregulation is a vital homeostatic process because acute deviations in extracellular fluid osmolality can cause significant cellular shrinking or swelling and subsequent tissue and organ damage (Bedford and Leader, 1993). Thus, osmotic stability is controlled by centrally-mediated adjustments in the release of AVP from the hypothalamo-neurohypophyseal system. AVP is synthesized in the somata of magnocellular neurosecretory neurons located in the SON and PVH, the crucial integrative brain structures that coordinate responses to perturbations in water balance (Bargmann, 1966; Qiu et al., 2011). It has been reported that the firing activity of neurons in the PVH and SON was enhanced during water deprivation (Reis et al., 2012). Moreover, Landgraf and Ludwig (1991) have reported an increase in AVP release within the PVH and SON in response to local hypertonic artificial cerebrospinal fluid delivery. In addition, Miyata et al. (1994) have observed that the soma size of both AVP neurons in the SON in rat was enlarged in chronic osmotic stimulation. These data suggest that neurons in the PVH and SON, AVP-positive neurons, show response to osmotic disturbance.

In addition, in many species of mammals (Forsling, 2000; Moon et al., 2004), a progressive increase in circulating AVP concentration was observed during the sleep period and a trough around the wake period. This circadian rhythm in AVP levels is functionally important, because the absence of an AVP surge during the late-sleep stage resulted in polyuria and disrupted sleep (Miller, 2000). Recent findings revealed that clock neurons in the SCN sent direct, functional axonal projections to modulate the strength of the connection between the organum vasculosum lamina terminalis (the central osmosensory nucleus) and AVP neurons in the SON and PVH. This indicates that clock neurons can modulate circadian changes in osmotic and AVP regulation (Cui et al., 1997; Abrahamson and Moore, 2001; Kalsbeek et al., 2006; Trudel and Bourque, 2010, 2012). Our results suggest that SCN CCK neurons receive the osmotic signals from AVP neurons in the SON and PVH and may integrate the information of osmotic regulation with the principal pacemaker to generate the strongly reciprocal feedback circuit. When suffering from dehydration, this feedback circuit can help maintain circulating AVP at a predetermined level to exclude circadian fluctuations.

Notably, in addition to AVP expressing neurons, the PVH also contain another neuron populations, such as Oxt neurons in the magnocellular neuroendocrine group, neurons that synthesize and release corticotropin-releasing hormone, thyrotropinreleasing hormone, dopamine, somatostatin, or growth hormone-releasing hormone in the parvicellular neuroendocrine group, and neurons that project to the brainstem and spinal cord for the central autonomic regulation in the descending group (Kombian et al., 2002). In total, more than 30 neurotransmitters have been localized to neurons within the PVH (Pyner, 2009). Here, we found that a part of non-AVPand non-Oxt-positive neurons in the PVH also sent direct inputs to SCN CCK neurons, suggesting that SCN CCK neurons may have other functions in addition to osmotic regulation.

Together, our viral tracing results provide a whole-brain map of neurons that convey reinforcement signals to SCN CCK neurons. Our findings offer a new perspective for future explorations of circuit mechanisms mediating SCN functions, such as circadian rhythm, ingestive behavior, and osmotic stability. Therefore, it is critical to gather further functional and behavioral data on SCN CCK neurons.

# AUTHOR CONTRIBUTIONS

X-SY and H-HW designed and performed the experiments, analyzed data, and wrote the paper. WX performed the experiments. LW analyzed data and wrote the paper. W-MQ designed the experiments and analyzed data. R-XL and Z-LH designed the experiments, analyzed data, and wrote the paper.

### FUNDING

This work was supported by the National Basic Research Program of China (Grant NO. 31530035, 81420108015 to Z-LH, Grant NO. 31671099, 31471064, 31871072 to W-MQ, and Grant NO. 31571103 to LW); the National Basic Research Program of China (Grant NO. 2015CB856401 to Z-LH).

### REFERENCES


### ACKNOWLEDGMENTS

We thank Dr. Miao He (Fudan University) for kindly supplying us with CCK-ires-Cre mice.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2018.00807/full#supplementary-material


cholecystokinin. Am. J. Physiol. Regul. Integr. Comp. Physiol. 294, R803–R810. doi: 10.1152/ajpregu.00682.2007


effects in oxytocin and vasopressin neurones during water deprivation. J. Neuroendocrinol. 24, 653–663. doi: 10.1111/j.1365-2826.2011.02249.x


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer AH and the handling Editor declared their shared affiliation, at the time of review.

Copyright © 2018 Yuan, Wei, Xu, Wang, Qu, Li and Huang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Automatic Human Sleep Stage Scoring Using Deep Neural Networks

Alexander Malafeev1,2,3, Dmitry Laptev<sup>4</sup> , Stefan Bauer4,5, Ximena Omlin2,6 , Aleksandra Wierzbicka<sup>7</sup> , Adam Wichniak<sup>8</sup> , Wojciech Jernajczyk<sup>7</sup> , Robert Riener2,3,6,9 , Joachim Buhmann<sup>4</sup> and Peter Achermann1,2,3 \*

<sup>1</sup> Chronobiology and Sleep Research, Institute of Pharmacology and Toxicology, University of Zurich, Zurich, Switzerland, <sup>2</sup> Neuroscience Center Zurich, University of Zurich and ETH Zurich, Zurich, Switzerland, <sup>3</sup> Center for Interdisciplinary Sleep Research, University of Zurich, Zurich, Switzerland, <sup>4</sup> Information Science and Engineering, Institute for Machine Learning, ETH Zurich, Zurich, Switzerland, <sup>5</sup> Max Planck Institute for Intelligent Systems, Tübingen, Germany, <sup>6</sup> Sensory-Motor Systems Lab, ETH Zurich, Zurich, Switzerland, <sup>7</sup> Sleep Disorders Center, Department of Clinical Neurophysiology, Institute of Psychiatry and Neurology in Warsaw, Warsaw, Poland, <sup>8</sup> Third Department of Psychiatry and Sleep Disorders Center, Institute of Psychiatry and Neurology in Warsaw, Warsaw, Poland, <sup>9</sup> University Hospital Balgrist (SCI Center), Medical Faculty, University of Zurich, Zurich, Switzerland

### Edited by:

Michael Lazarus, University of Tsukuba, Japan

### Reviewed by:

Ivana Rosenzweig, King's College London, United Kingdom Jussi Virkkala, Finnish Institute of Occupational Health, Finland Alejandro Bassi, Universidad de Chile, Chile

### \*Correspondence:

Peter Achermann acherman@pharma.uzh.ch

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 18 July 2018 Accepted: 09 October 2018 Published: 06 November 2018

### Citation:

Malafeev A, Laptev D, Bauer S, Omlin X, Wierzbicka A, Wichniak A, Jernajczyk W, Riener R, Buhmann J and Achermann P (2018) Automatic Human Sleep Stage Scoring Using Deep Neural Networks. Front. Neurosci. 12:781. doi: 10.3389/fnins.2018.00781 The classification of sleep stages is the first and an important step in the quantitative analysis of polysomnographic recordings. Sleep stage scoring relies heavily on visual pattern recognition by a human expert and is time consuming and subjective. Thus, there is a need for automatic classification. In this work we developed machine learning algorithms for sleep classification: random forest (RF) classification based on features and artificial neural networks (ANNs) working both with features and raw data. We tested our methods in healthy subjects and in patients. Most algorithms yielded good results comparable to human interrater agreement. Our study revealed that deep neural networks (DNNs) working with raw data performed better than feature-based methods. We also demonstrated that taking the local temporal structure of sleep into account a priori is important. Our results demonstrate the utility of neural network architectures for the classification of sleep.

Keywords: deep learning, sleep, EEG, automatic scoring, random forest, artificial neural networks, features, raw data

# INTRODUCTION

### Problem Statement

Visual scoring of the sleep stages is the gold standard in sleep research and medicine. Sleep scoring is performed visually based on the following signals: (1) electrical activity of the brain – electroencephalogram (EEG), (2) electrical activity resulting from the movement of the eyes and eyelids – electrooculogram (EOG) and (3) muscle tone recorded under the chin (submental) – electromyogram (EMG).

Sleep scoring is usually performed according to standardized scoring rules: Rechtschaffen and Kales (1968) or the American Association of Sleep Medicine (AASM) (Iber et al., 2007). According to the AASM rules (Iber et al., 2007) an expert visually classifies consecutive 30-s epochs of polysomnographic (PSG) data (EEG, EOG and EMG) into wake, rapid eye movement (REM) sleep, and non-REM (NREM) sleep (stages N1–N3). If scoring is performed according to Rechtschaffen and Kales (1968), 20- or 30-s epochs are scored and NREM sleep is subdivided into stages 1–4 with stages 3–4 considered as slow wave sleep (SWS, deep sleep, corresponding to N3). Furthermore, Rechtschaffen and Kales (1968) defined movement time as a separate stage.

**42**

The plot of a sequence of sleep stages is called a hypnogram (see **Figure 1**). Human sleep starts generally with a stage 1 (N1), which usually lasts only up to a few min and is a very light sleep. Slow rolling eye movements are a feature of stage 1 and contractions of the muscles, hypnagogic jerks may occur.

Next follows stage 2 (N2), a deeper state of sleep than stage 1, characterized by the occurrence of sleep spindles and K-complexes and an intermediate muscle tone.

Stage 2 usually precedes deep sleep – stages 3 and 4 (SWS, N3). The main characteristic of deep sleep is the presence of slow oscillations (<1 Hz) and delta waves (1–4 Hz) in the EEG for at least 20% of the epoch duration. The muscle tone is low.

Rapid eye movement sleep occurs periodically throughout the night and is characterized by rapid eye movements, fast lowamplitude EEG activity like the wake EEG, and a low muscle tone (atonia).

The progression of the different stages is not random, but rather follows a cyclic alternation of NREM and REM sleep (Achermann and Tarokh, 2014) with a cycle duration of approximately 90 min (see **Figure 1** for a typical structure). Healthy sleep consists of approximately 3–5 sleep cycles.

Visual scoring by an expert is time consuming and subjective. Several studies addressed the interrater reliability and revealed that correspondence between scorers is far from ideal (Danker-Hopfe et al., 2004; Penzel et al., 2013; Rosenberg and Van Hout, 2013; Younes et al., 2016, 2018).

Several measures can be used to compare two experts or an algorithm with an expert. The simplest one is accuracy, the proportion of epochs which were assigned the same sleep stage. The F1 score (Dice, 1945; Sørensen, 1948) is a measure computed per class and it is widely used in the field of machine learning, and was also applied to assess performance in automatic sleep scoring (Tsinalis et al., 2016; Supratak et al., 2017; Chambon et al., 2018).

It was argued that F1 score has certain disadvantages by Powers (2014). Cohen's kappa (Cohen, 1960) is a metric accounting for the agreement by chance and thus for imbalanced proportions of different classes and is commonly used in biology and in sleep research. Values higher than 0.8 are considered to reflect excellent agreement (Mchugh, 2012). We also applied this metric in our study.

Cohen's kappa values in the study by Danker-Hopfe et al. (2009) showed good agreement for REM sleep, minimal agreement for stage 1 and moderate agreement for the other stages.

Shortly after a sleep scoring standard was established in 1968 (Rechtschaffen and Kales, 1968), attempts were made to develop algorithms for automated sleep staging (Itil et al., 1969; Larsen and Walter, 1970; Smith and Karacan, 1971; Martin et al., 1972; Gaillard and Tissot, 1973; Gevins and Rémond, 1987).

### Related Work

Martin et al. (1972) applied a simple decision tree using EEG and EOG data for scoring. A decision tree like algorithm was also used by Louis et al. (2004). Stanus et al. (1987) developed and compared two methods for automatic sleep scoring: one based on an autoregressive model and another one based on spectral bands and Bayesian decision theory. Both methods used one EEG, two EOG and an EMG channel. The EOG was needed to detect eye movements and the EMG to assess the muscle tone. Fell et al. (1996) examined automatic sleep scoring using additional nonlinear features (correlation dimension, Kolmogorov entropy, Lyapunov exponent) and concluded that such measures carry additional information not captured with spectral features. Park et al. (2000) built a hybrid rule- and case- based system and reported high agreement with human scorers. They also claimed that such a system works well to score patients with sleep disorders.

One of the commercially successful attempts to perform automatic scoring evolved from the SIESTA project (Klosh et al., 2001). The corresponding software of the SIESTA group was named Somnolyzer 24x7. It includes a quality check of the data based on histograms. The software extracts features based on a single EEG channel, two EOG channels and one EMG channel and predicts sleep stages using a decision tree (Anderer et al., 2005). The software was validated on a database containing 90 patients with various sleep disorders and ∼200 controls. Several experts scored sleep in the database and Somnolyzer 24x7 showed good agreement with consent scoring (Anderer et al., 2005).

Newer and more sophisticated approaches were based on artificial neural networks (ANNs). Schaltenbrand et al. (1993) for example applied ANNs for sleep stage classification using 17 features extracted from PSG signals and reported an accuracy close to 90%. Pardey et al. (1996) combined ANNs with fuzzy logic and Längkvist et al. (2012) applied restricted Boltzmann machines to solve the sleep classification problem, to mention just a few approaches.

The methods mentioned above require carefully engineered features. It is possible to avoid this step using novel deep learning methods. ANNs in the form of convolutional neural networks (CNNs) were recently applied to the raw sleep EEG by Tsinalis et al. (2016). CNNs are especially promising because they can learn complex patterns and 'look' at the data in a similar way as a 'real brain' (Fukushima and Miyake, 1982). However, working with raw data requires a huge amount of training data and computational resources.

Sequences of epochs are considered by a human expert according to the scoring manuals. Therefore, we assume that learning local temporal structures are an important aspect in automatic sleep scoring. Temporal patterns have previously been addressed by applying a hidden Markov model (HMM) (Doroshenkov et al., 2007; Pan et al., 2012). In the last few years, recurrent neural networks (RNNs) have demonstrated better performance than "classical" machine learning methods on datasets with a temporal structure (Mikolov et al., 2010; Graves et al., 2013; Karpathy and Fei-Fei, 2015). One of the most common and well-studied RNNs is the Long-Short Term Memory (LSTM) neural network (Hochreiter and Schmidhuber, 1997). Such networks have been successfully applied to EEG data in general (Davidson et al., 2006) as well as to sleep data (Supratak et al., 2017).

Artificial neural networks using raw data revealed comparable performance as the best ANNs using engineered features and the best classical machine learning methods (Davidson et al., 2006; Tsinalis et al., 2016; Supratak et al., 2017; Chambon et al., 2018;

Phan et al., 2018; Sors et al., 2018). See Section "Discussion" for more details.

The above-mentioned approaches were based on supervised learning. There have also been several attempts to perform unsupervised automatic sleep scoring in humans (Gath and Geva, 1989; Agarwal and Gotman, 2001; Grube et al., 2002) and in animals (Sunagawa et al., 2013; Libourel et al., 2015).

### Our Contribution

naming conventions of the algorithms.

We implemented different machine learning algorithms, random forests (RF), feature based networks (LSTM networks) and rawdata based networks (CNN-LSTM networks) and trained and tested them in healthy participants and patients. We report all the Cohen's kappa values (Cohen, 1960) of the different stages for the comparison of the performance the algorithms.

All our algorithms yielded high values of Cohen's kappa of the data of healthy subjects. Performance on data recorded in patients was lower, but less so for ANNs. Including part of the patient data into the training improved performance on the patient data. This suggests that we would need even larger and diverse datasets to train an algorithm which can be applied reliably in practice. DNNs performed well even using only a single EEG channel, an interesting observation of our work.

# MATERIALS AND METHODS

### Polysomnographic (PSG) Data

We trained and tested automatic sleep stage scoring algorithms on two datasets from two different laboratories.

The first dataset was comprised of 54 whole night sleep recordings of healthy participants. The second dataset consisted of 22 whole night sleep recordings and 21 recordings of a multiple sleep latency test (MSLT) in patients. The MSLT is routinely used to evaluate daytime sleepiness of patients. During this test a subject has four or five 20-min nap opportunities, which are separated by 1.5-h long intervals. An example of an MSLT hypnogram can be seen in **Figure 2**. Usually, only naps are recorded, but in our dataset, recordings were continuous over approximately 9 h and occasionally we observed sleep episodes in addition to the scheduled naps. In a standard setting these sleep episodes would have been missed. EEG channel C3A2,

one myographic and two oculographic channels were used for analysis and classification.

### Dataset 1: Healthy Subjects

Polysomnographic (PSG) recordings from a study investigating the effect of vestibular stimulation (Omlin et al., 2018). In total 18 healthy young males (20–28 years; mean: 23.7 years) were recorded. Three nights of sleep (8 h) were recorded in each subject. Two nights with motion (bed was rocked till sleep onset or for the first 2 h after lights off), and a control night without movement. Data were composed of 12 EEG channels, applied according to the 10–20 system, 2 EOG derivations, 1 submental EMG derivation, 1 ECG derivation and respiration signals (chest and abdomen). Recordings were performed with a polygraphic amplifier (Artisan, Micromed, Mogliano, Veneto, Italy). Sampling rate was equal to 256 Hz (Rembrandt DataLab; Version 8.0; Embla Systems, Broom Field, CO, United States). A high pass filter (EEG: −3 dB at 0.16 Hz; EMG: 10 Hz; ECG: 1 Hz) and an anti-aliasing filter (−3 dB at 67.4 Hz) were applied to the analog signals. The EEG derivations were re-referenced to the contra-lateral mastoids (A1, A2). Sleep stages (20-s epochs) were scored according to the AASM criteria (Iber et al., 2007). The study was performed in the sleep laboratory of the Institute of Pharmacology and Toxicology at the University of Zurich and was approved by the Institutional Review Board of the Swiss Federal Institute of Technology in Zurich (ETH Zurich).

### Dataset 2: Patients

Data were recorded in patients with narcolepsy (23 patients) and hypersomnia (five patients) during a night of sleep (approximately 8 h) and during a MSLT (continuous recordings over approximately 9 h). We had to exclude some recordings due to bad signal quality. Thus, some patients contributed only with a night or a MSLT recording (Hypersomnia: 5 MSLT, 4 nights; Narcolepsy: 16 MSLT, 18 nights). Data were comprised of 6 EEG, 2 EMG, 2 EOG derivations and 1 ECG. Signals were recorded at a sampling rate of 200 Hz (polygraphic amplifier Grass Technologies AURA PSG). A high pass filter (EEG: −3 dB at 0.5 Hz) and an anti-aliasing filter (−3 dB at 50 Hz) were applied to the analog signals. Sleep stages (30-s epochs) were scored according to Rechtschaffen and Kales (1968). Movement time was not scored. To make sleep stages compatible with the first dataset, we merged sleep stages 3 and 4. Recordings were performed at the Sleep Disorders Center, Department of Clinical Neurophysiology, Institute of Psychiatry and Neurology in Warsaw, Warsaw, Poland. The study was approved by the Institutional Review Board of Institute of Psychiatry and Neurology.

Data in the two laboratories were recorded with different recording devices which resulted in different sampling rates and filter settings. Signals were resampled at 128 Hz (with applying appropriate anti-aliasing filters thus, leading to a similar lowpass filtering of the data) to accommodate data recorded at different sampling rates. We did not adjust the high-pass filtering

because we did not expect it to have a big impact on classification performance. Another reason was that we consider it is important that our methods work with data recorded with equipment that differs between laboratories.

### Machine Learning: Classification

Machine Learning is a branch of computer science, which allows to learn properties of the data and solve problems without direct programming of the decision rules. The main approaches in machine learning are supervised and unsupervised learning (Bishop, 2016). In this work we used a supervised approach in order to solve the problem of classification (Bishop, 2016). Classification algorithms solve the problem of assigning labels to the data. They are trained with labeled data, the training set, to learn properties of the data and the corresponding labels [supervised machine learning (Bishop, 2016)].

In this work, we solved the classification problem by applying supervised machine learning algorithms. We followed two approaches, (1) classification based on features (RF and ANNs) and (2) classification based on raw data (ANNs).

### Classification Based on Features

Polysomnographic signals are very complex, but they reveal certain patterns crucial for scoring by an expert. For example, waves of certain frequencies: sleep spindles (12–14 Hz), slow waves (0.5–4 Hz), alpha waves (8–12 Hz), theta oscillations (4–8 Hz) are very important to distinguish the different sleep stages. These measures can be easily quantified in the frequency domain. We applied classical spectral analysis (Welch, 1967) but also a multi-taper approach (Babadi and Brown, 2014) might be considered in particular when spectrograms are used as features. Other important markers of sleep stages such as rapid and slow eye movements, eye blinks and muscle tone can also be quantified. Such measures are called features and the process of their definition is called feature engineering. Using carefully engineered domain-specific features for machine learning systems has a lot of advantages: it requires a small amount of training data, is fast and the results are interpretable. Another approach based on deep learning, working with raw data, is described later.

### **Preprocessing and feature extraction**

In a first step, we used spectrograms of the EEG instead of using the raw signal. It is well known that spectra capture the major properties of the sleep EEG and this way we were able to significantly reduce the dimensionality of our data. Power density spectra were calculated for 20-s epochs (30-s for patient data) using the Welch function in MATLAB (FFT; average of four or six 5-s windows; Hanning windows; no overlap; frequency resolution 0.2 Hz). Spectra were plotted and color-coded on a logarithmic scale (**Figures 1**, **2**). Spectrograms were limited to the range of 0.8–40 Hz to reduce the dimensionality of the data matrix.

We used a set of 20 engineered features for the classification (see **Supplementary Material** for their definitions). They include among others power in different frequency bands and their ratios, eye movements, and muscle tone. We did not exclude any epochs (i.e., included artifacts), because we wanted to have a system, which is ready to work with the data with a minimal requirement of manual pre-processing. Moreover, epochs with artifacts contain useful information: wakefulness is almost always accompanied by movement artifacts and a movement is often followed by a transition into stage 1. Quantitative analysis, however, such as the calculation of average power density spectra requires exclusion of artifacts which can be achieved using simple algorithms (Malafeev et al., 2018).

We used two different approaches for the classification based on features: RF and ANNs.

### **Random forest (RF)**

One of the classical methods to solve classification problems is based on decision trees (Morgan and Sonquist, 1963; Hunt et al., 1966; Breiman et al., 1984). Every node of a tree corresponds to a feature and a corresponding a threshold value. For a data vector which has to be classified, we traverse the tree by comparing a corresponding feature to the threshold of the node. Depending on the outcome of the comparison, we go to the left or to the right branch. Once we have traversed the tree, we end up in a leaf that determines to which class the data point belongs to.

Decision trees have certain limitations (e.g., overfitting) (Safavian and Landgrebe, 1991; Mitchell, 1997). Overfitting means that an algorithm learns something very specific of the training data and the classifier can no longer predict new data.

A way to overcome these limitations is to create an ensemble of trees: i.e., to build many trees, each based on a random subset of the training data (Ho, 1995; Breiman, 2001). A data point is classified by all trees and we can compute the probability of a data point belonging to a particular class by the fraction of trees which "voted" for this class. RF classifiers and similar recent tree-based technique demonstrated state-of-the-art results on a variety of problems (Laptev and Buhmann, 2014, 2015; Chen and Guestrin, 2016).

We implemented the RF to classify sleep stages based on feature vectors (20 components). We computed probability vectors for every epoch (20 or 30 s). Further we considered the local temporal structure of sleep as described above about time course learning. We applied a HMM (see **Supplementary Material**) and a median filter (MF) with a window of three 20-s or 30-s epochs to smooth the data.

### **Artificial neural networks (ANNs)**

For a long time, researchers have been trying to build a computer model of a neuron (Farley and Clark, 1954; Rochester et al., 1956) and use such models for data classification (Rosenblatt, 1958). This research resulted in the development of multilayer neural networks (Ivakhnenko and Lapa, 1967) which are now denoted ANNs.

Artificial neural networks consist of interconnected neurons. Every neuron performs multiplication of input signals with parameters called weights, summed up and sent to the output. One can train ANNs by adjusting (updating) the weights (Goodfellow et al., 2016). This process of training is also called optimization. ANN training requires a function which quantifies the quality of the classification. Such a function is called the loss function or cost function. The loss function must be differentiable, otherwise it is not possible to compute the gradients. An example of a loss function is the mean square error. In our work, we used the cross-entropy loss function (De Boer et al., 2005). Cross-entropy loss is a good measure of errors of networks with discrete targets. Targets are the ground truth values given by an expert, in our case the sleep stages.

### Deep Learning With Raw Data

fnins-12-00781 November 2, 2018 Time: 17:7 # 6

Deep neural networks (DNNs), a specific type of ANNs, can learn complex models. Moreover, DNNs can automatically learn features and the feature engineering step can be omitted. Features can be learned using, for example CNNs (Fukushima and Miyake, 1982; Lecun et al., 1989; Waibel et al., 1989). DNNs usually show better performance than feature-based methods, but it comes at the price of an increased computational demand and such networks require more training data. However, DNNs require much less manual adjustments than feature-based methods and thus are easier to implement and maintain.

### **Convolutional neural networks (CNNs)**

A particular type of DNNs are CNNs. They were initially developed for image recognition (Fukushima and Miyake, 1982; Lecun et al., 1989; Waibel et al., 1989). The main property of CNNs is that they perform a convolution of an input with a set of filters, which have to be learned. They were successfully applied not only for image recognition, but also in speech recognition (Abdel-Hamid et al., 2014), text analysis (Dos Santos and Gatti, 2014) and many other areas. Moreover, CNNs have already been successfully applied to various types of physiological signals, including wake EEG recordings (Cecotti and Graeser, 2008; Mirowski et al., 2008). The filters have a certain size. Given the one-dimensional nature of our data, a filter is a vector of a specific length. The filter slides with certain step called a stride across the input data.

Another specific type of layers we used was max-pooling. It takes the maximal value of the sliding window and helps to achieve local invariance. The max-pooling layer also has a specific filter size and a stride.

### **Residual networks**

Residual networks (He et al., 2016) are a special kind of ANNs where layers are connected not only in sequential order but also with so-called skip or residual connections which jump over one or multiple layers. Gradients can vanish when networks have a lot of layers. Residual connections prevent this problem and make the training of networks more efficient and make it possible to train very deep networks with large numbers of layers.

### Learning Time Dependencies

Common machine learning algorithms consider every data sample independent from the previous ones. This is the case for RF classification and common ANNs. However, experts take information about previous epochs into account when they perform sleep scoring. Thus, it would be useful to consider some temporal information (structure) in the sleep classification algorithm.

As was mentioned in the introduction, sleep has not only a local but also a global structure, such as sleep cycles (Achermann and Tarokh, 2014). However, this global structure should not be taken into account while scoring (visual or automatic), as it might be different in pathology or during naps. Therefore, we limited the temporal memory of our models (see below), but the information of several previous epochs is still important to consider for sleep scoring. We assume that if we learn long sequences, it would bias the algorithm and such models would perform poorly on recordings where such patterns are not present, e.g., in the MSLT recordings (short naps of 20 min) or disturbed sleep.

We implemented the learning of temporal structures of sleep in two ways. First, we applied a HMM (Stratonovich, 1960) to smooth the output of the RF classification (see **Supplementary Material** for details) and by a MF with a window size of three epochs, a very simple yet efficient approach to smooth the data (see **Supplementary Material**).

As a second approach we implemented RNNs. RNNs receive their own output of the previous step as additional input in combination with the new data vector. Thus, RNNs take into account the temporal structure of the data. One of the most successful RNNs is the LSTM network (Hochreiter and Schmidhuber, 1997). RNNs can also use information about future epochs; in such a case they are called bidirectional RNNs. One of the main advantages of LSTM networks is its property to avoid vanishing gradients.

As mentioned above, the length of the input sequences should be limited to reasonably short time intervals. We limited our algorithms to learn patterns not longer than 8 (2.8 or 4 min), 32 (10.7 or 16 min), and 128 epochs (42.6 or 64 min). We dynamically formed batches of sequences: the beginning of each sequence was chosen randomly (i.e., sequences may intersect). This way more sequences may be used for training than by just taking them sequentially. For details about batches and their processing see **Supplementary Material**.

# Study Setup

### Network Architectures

We considered two types of networks:


### LSTM Networks

We implemented a network with three hidden layers (**Figure 3**). Each layer consisted of 8, 16, 32, or 128 LSTM units, and we also applied one- and bi-directional layers resulting in a total of six network configurations.

### CNN-LSTM Networks

We realized networks with 11 convolutional layers followed by two LSTM layers with 32 units (**Figure 4**).

We also used residual convolutional networks (19 layers) as outlined before, worked with different input signals (EEG, EOG, and EMG) and created separate CNN networks (CNN blocks in **Figure 4**) for every input (EEG, 2 EOG). The outputs of all

blocks were concatenated and fed into the LSTM layers. There were two bidirectional LSTM layers. Each layer contained 32 LSTM units. There were batch normalization layers (Ioffe and Szegedy, 2015) before, between and after LSTM layers. Batch normalization layer rescales the input to make sure that all the values belong to the same range. We used separate CNN blocks for the two EOG channels because correlations between the EOG signals are important to distinguish the different types of eye movements. In case the EMG was included, only a single value (EMG power in the 15–30 Hz range) per 20- or 30-s epoch was considered. Thus, three input configurations were implemented: EEG only, EEG and EOGs, and EEG, EOGs and EMG (**Figure 4**) resulting in a total of seven network configurations.

### Optimization

Networks require training which is achieved by optimization. Optimization procedures have to find minima (in case of ANN local minima) of a loss function over the parameter space (weights of the network). Weights are commonly adjusted according gradients (backpropagation, see **Supplementary Material** for details about optimization and regularization).

Networks were implemented using the Keras package (Chollet, 2015) with Theano (Al-Rfou et al., 2016) and Tensorflow (Abadi et al., 2016) backends. The Theano backend was used to train our feature-based LSTM networks and the Tensorflow backend to train the raw data based CNN-LSTM networks. We worked with different backends because we first developed the feature based networks and running on a desktop computer and later with raw data based networks. These networks had to be trained on GPUs and for this only the Tensorflow (Abadi et al., 2016) backend was available.

### Training, Validation, and Testing

To avoid overfitting, we randomly split dataset 1 (healthy participants) into three parts: training (36 recordings, 70%), validation (9 recordings, 15%) and testing (9 recordings, 15%). The data were split according to participants, i.e., all three recordings of one participant were either in the training, validation or test set. We computed the crossentropy loss and accuracy (De Boer et al., 2005) to assess convergence of the algorithms. These measures were computed on every training iteration for training and validation sets.

The idea was to train all our models using the training part of the data, then classify the data of the validation part and select only the best models for further confirmation of their performance on the test part. However, validation revealed that performance of the different models was very similar, thus, it was unclear whether their performance was really different. Therefore, we estimated the final performance of all algorithms with both the validation and test set. In addition, we used the whole second dataset (patients) as a test set, thus, assessing generalization of the approaches to datasets from another laboratory and to a different subject population (patients).

Further, we wanted to study how performance of the algorithms would benefit from the inclusion of patient data into the training set. We took the same training set of healthy subjects (36 recordings) and added patient data (19 recordings) to it, resulting in a training set of 73 recordings. The remaining patient data (10 MSLT recordings and 14 sleep recordings) were used for performance evaluation together with the test set of the healthy participants (9 recordings; a total of 33 recordings). Again, all data of one patient were assigned to the training or test set. For further details see **Supplementary Material**.

### Performance Evaluation

To assess performance of our algorithms, we used Cohen's kappa (Cohen, 1960) a metric accounting for the agreement by chance and thus for imbalanced proportions of different classes. Kappa is a number ≤ 1 (can be negative), with one reflecting ideal

Unit, it is an activation function to transform the activation of a neuron.

classification. Values higher than 0.8 are considered to reflect excellent agreement (Mchugh, 2012).

### RESULTS

### Convergence of the ANNs

During the process of the training of ANNs we can observe an increase in the quality of the classification. To ensure that the network was sufficiently trained and further training would not bring additional benefit we computed cross-entropy loss and accuracy (proportion of correctly classified examples; see section "Materials and Methods" for details). Usually these metrics show an exponential saturation with increasing training time. After the accuracy or loss function have reached a plateau we can say that a network has converged. These types of curves are called learning curves (Pedregosa et al., 2011). We computed these curves on the training and validation datasets (50 training iterations in total).

stages; R, REM sleep.

fnins-12-00781 November 2, 2018 Time: 17:7 # 9

All our feature-based LSTM networks showed good convergence when they were trained on the data of healthy participants (**Supplementary Figure S3**) and on a mixture of both datasets (**Supplementary Figure S4**; see **Supplementary Material** for the naming convention of the networks).

Learning curves for the ANN based on the raw data as input are depicted in **Supplementary Figures S5**, **S6**. Most of the networks showed good convergence (loss monotonously decreased, and accuracy increased to saturation). Some networks showed large fluctuations of loss and accuracy on the validation set: the network which has only a single EEG channel as input (1p\_32u\_8ep), the network which had EEG and EOG as input and eight epoch long sequences (1p2\_32u\_8ep), and the network with input comprised of EEG, EOG and EMG and 128 epoch long sequences (1p2p1\_32u\_128ep). The least smooth learning curves were observed in the network with residual connections. This network had the largest number of parameters and thus, more data and iterations might be needed to reach convergence. We expect that such networks to perform better if trained on an extended dataset.

### Classification Performance

The crucial information is how well the algorithms perform. As mentioned above we used Cohen's kappa to measure the quality of the automatic scoring.

**Figure 1** illustrates the hypnograms obtained with three selected algorithms (RF, LSTM, and CNN-LSTM) in comparison with the expert scoring. In general, performance of all algorithms was good capturing the cyclic structure of sleep. Slight differences to the human scorer were observed, e.g., longer REM sleep episodes with the 3-layer bidirectional LSTM network (**Figure 1**, panel 3).

Performance of our algorithms was initially assessed with the F1 score (Dice, 1945; Sørensen, 1948). But afterwards we switched to Cohen's kappa (Cohen, 1960) because F1 scores are a biased measure of classification quality (Powers, 2014), which is a problem when comparing recordings with a different prevalence of the classes (sleep and MSLT).

### Scoring of Healthy Participants

The Cohen's kappa computed on the validation part of the dataset 1 (healthy participants) are illustrated in the **Figure 5** (only four selected methods; see **Supplementary Tables S1**, **S2** for kappa of all algorithms, validation and test data): RF classification smoothed using HMM, one LSTM network trained on features, and two CNN-LSTM networks with raw data input, one of them included residual connections.

All four methods showed high performance for all stages except for the stage 1 (N1). Kappa of stage 1 was around 0.4 which we still consider a good result because it is comparable to the low human interscorer agreement of stage 1 (Danker-Hopfe et al., 2004; Danker-Hopfe et al., 2009; Penzel et al., 2013; Rosenberg and Van Hout, 2013).

The Cohen's kappa of all methods evaluated on the validation part of dataset 1 are depicted in **Supplementary Figure S7** (features) and **Supplementary Figure S8** (raw data). Most networks performed similarly well on the validation set; those which included only a single EEG derivation as an input (**Supplementary Figure S7**, s\_8u\_8ep, spectrogram as input and **Supplementary Figure S8**, 1p\_32u\_8ep, raw EEG as input) showed slightly lower performance, probably since the EEG spectrogram or the raw EEG do not contain information about eye movements and muscle tone. However, this was the case in some recordings only, for other recordings the performance was very good. Interestingly, performance of these networks on the test set was much better (**Supplementary Tables S1**, **S2**). We assume that the validation set contained some recordings which were difficult to score using only a single EEG channel.

The network with input comprised of EEG, EOG, and EMG and 128 epoch long sequences (1p2p1\_32u\_128ep) had a low performance on both, the validation and the test set because of large random fluctuation of accuracy in the last training iteration. Ideally, we should have stopped training of this network earlier or trained it longer.

Networks with 16 and 32 units in a layer were inferior for the scoring of stage 1 than the network with only 8 units probably due to overfitting, although the difference was very small. These networks may show a better performance if trained with larger datasets. One-directional network predicted REM sleep a bit worse than bidirectional ones. The advantage of one-directional network is the possibility to work online. Surprisingly, classification with RF smoothed with simple MF or HMM worked almost

FIGURE 6 | Cohen's kappa scores for the same methods as in Figure 5 applied to the patient dataset. Note that the training did not include patient data. Top panel represents sleep recordings and the lower one MSLT recordings. Note that during MSLT recordings stage N3 is not always reached; such recordings were not taken into account when computing average Cohen's kappa and standard deviations of N3. For details see Figure 5.

as good as classification with ANNs (features and raw data).

### Generalization to the Patient Data

We validated our methods on dataset 2 (patients). The kappa values for selected methods are presented in **Figure 6** (only 4 selected methods; see **Supplementary Figures S9**, **S10** and **Supplementary Tables S3**, **S4** for kappa of all algorithms used to classify patient data). Note that the data of the patient dataset were not used for the training at all.

The performance was somewhat lower for all classifiers applied to the sleep data of patients than in healthy participants and again lower for the MSLT data and kappa showed a large variance. Classification performance of stage 1 was worst for the RF classification in this dataset. Methods using only a single EEG signal as input (spectrogram or raw EEG channel as input) performed worse on the patient data.

We observed very low kappa scores in some recordings, mostly for stages 2, 3 and REM sleep in patients when the training data did not include patient data. Stage 2 was often confused with stage 1. We can explain it by different properties of sleep in patients. Their sleep was much more fragmented and disturbed. Thus, algorithms not trained with patient recordings may confuse stages 2 and 1. Kappa of stage 3 was very low mostly due to the low occurrence of deep sleep in patients, or its complete absence. Thus, small discrepancies led to low kappa values. Further, REM sleep was sometimes missed due to differences between patients and healthy participants. Sometimes REM sleep was falsely discovered. It happened because patients sometimes had a low muscle tone in wakefulness. Some of the

false discovered REM sleep turned out to be true REM sleep missed by an expert (confirmed by visual inspection).

Algorithms based on the EEG only made most mistakes. Adding ocular channels to the input resulted in less mistakes and including included muscle tone also revealed the best performance.

When the networks were trained also using patient data the result have improved.

### Networks Trained on the Data From Both Datasets

Next, we trained two networks and RF classification with a mixed training data consisting of healthy subjects (36 recordings) and part of the patient data (19 recordings; both sleep and MSLT data). We validated the models on the test part of the mixed dataset (healthy participant: 9 recordings; patients, 14 sleep and 10 MSLT recordings).

**Figure 2** illustrates the hypnograms of a MSLT recording obtained with three selected algorithms in comparison with the expert scoring. In general, performance of all algorithms was good capturing the naps. Performance of four selected methods are illustrated in **Figure 7**, and of the other methods applied in **Supplementary Figures S11, S12** and in **Supplementary Tables S5, S6**. Note, that we trained only two feature-based networks with the mixture of the two datasets. Training on the mixed data resulted in an improved performance on both patient data and data of healthy participants, kappa values increased and the variance got smaller.

As mentioned above the performance after training on both datasets was better than training only on data of healthy participants, which is not surprising. It is well known that sleep of the patients (narcolepsy and hypersomnia in our case) is quite different compared to healthy participants. Again, algorithms based on EEG, EOG, and EMG revealed reasonable kappa values for all recordings except for circumstances where some stages were not present in a recording, or only in very small amounts. This was often the case for stage 3 in almost all MSLT recordings and in some sleep recordings of the patients. Often discrepancies occurred at stage/state transitions. However, that is where mostly experts also disagree. Multiple expert scoring of the same recording would be needed to establish a "ground truth." Occasionally, the EEG was contaminated by ECG artifacts leading to a lower classification performance. Thus, removal of ECG artifacts prior to classification might improve the performance.

### DISCUSSION

### Comparison With Human Experts and Automatic Scoring of Other Groups

All our implemented methods yielded high Cohen's kappa values (kappa around 0.8) for all stages when they were trained and validated on data of the same type of subjects, except for stage 1 (N1; kappa < 0.5). Stage 1 is known as a difficult stage to score.

Common measures of interrater agreement are accuracy and Cohen's Kappa (Danker-Hopfe et al., 2004, 2009;Penzel et al., 2013; Rosenberg and Van Hout, 2013). Kappa values obtained with our models were comparable to the performance of human experts. Stage 1 was most difficult to score automatically and compares with the low interrater agreement (Danker-Hopfe et al., 2004, 2009; Penzel et al., 2013; Rosenberg and Van Hout, 2013).

Performance of the LSTM networks in our experiments were similar to the one of a recent study where a CNN was applied to EEG features (Tsinalis et al., 2016) and to Phan et al. (2018) who applied CNNs to spectral features of EEG, EOG, and EMG channels.

Our CNN-LSTM networks performed similar to the ones of recent studies which employed CNNs for sleep scoring based on a single EEG derivation (Sors et al., 2018) and on six EEG channels in conjunction with two EOG and three EMG channels (Chambon et al., 2018). Sors et al. (2018) used a large database to train their network. They reported Cohen's kappa computed over all classes equal to 0.81. Our values were close to it, however, it is not possible to compare directly because we looked at each class separately. We consider it very important to know the kappa values for wake, NREM and REM sleep separately due to their unbalanced contribution.

Supratak et al. (2017) used a technique known as residual sequence learning which might improve the performance. We did not apply this approach but used residual connections and different signals as independent inputs in the convolutional part of the network which were concatenated as input to the LSTM part. We think this was beneficial for the performance.

Even though automatic scoring algorithms have shown reasonably high performance there is no consensus yet in the sleep community that they perform well enough to replace human scorers.

### Automatic Scoring Using Different Channels

Our study showed that it is possible to score sleep data with high classification accuracy using only a single EEG channel. We got slightly better results using 1 EEG, 2 EOG, and 1 EMG channel.

It is difficult to conclude which method works best due to the small differences in performance. We assume that four channels (1 EEG, 2 EOG, and 1 EMG) contain more information, but the risk of the data being noisy is also higher. We observed that a bad EMG signal reduced the performance of the algorithms. This was also observed by SIESTA team (Anderer et al., 2005). The authors reported that in some cases the use of the EMG was not optimal due to a bad signal quality, and in certain cases they substituted the EMG with the high frequency content of the EEG and EOG which increased the performance of their algorithm. Also Phan et al. (2018) showed that the use of EOG and EMG channels was beneficial and Chambon et al. (2018) reported that the use of multiple EEG channels increased the performance of automatic sleep scoring.

It was surprising to observe that neural networks can classify sleep, especially REM sleep, with high quality using only a single EEG channel. It is a very difficult task for a human scorer to distinguish REM sleep based on the EEG only. Experts rely on eye movements and muscle tone (Rechtschaffen and Kales, 1968; Iber et al., 2007). We think that presence of patterns such as sawtooth

waves (Jouvet et al., 1960; Takahara et al., 2009) are important markers of REM sleep which help neural network to recognize this stage.

A note of caution regarding EEG channels: the signal amplitude is strongly dependent on the referencing and the scoring of SWS (N3) is dependent on an amplitude criterion (75 µV peak-to-peak) (Rechtschaffen and Kales, 1968; Iber et al., 2007). We and others (Chambon et al., 2018; Phan et al., 2018; Sors et al., 2018) used, as it is standard in the sleep field, EEG derivations referenced to the contralateral mastoid whereas a different referencing was used in other studies (Tsinalis et al., 2016; Supratak et al., 2017). Networks trained with specific referencing should not be applied to data recorded with a different reference system as in particular the amount of SWS (N3) will be affected due to the difference in signal amplitude.

### How to Measure Scoring Quality?

It is difficult to determine which method was superior based on our results. We think this is because most of our methods showed high performance based on the chosen evaluation metric and produced results comparable to human experts.

An issue is the fact that with F1 scores (Dice, 1945; Sørensen, 1948) and Cohen's kappa (Cohen, 1960) we treat epochs independently not taking the temporal structure of sleep into account. Thus, we think such metrics are not the optimal way to assess different aspects of the quality of scoring. For example, visual inspection of our results has shown that quiet wakefulness at the beginning of sleep might be confused with REM sleep and sometimes the first often very subtle and short REM sleep episodes might be missed. Such misclassification often occurred when the EMG or EOG signals were corrupt or of bad quality. It almost does not affect F1 scores and kappa values but affects the structure of sleep. Thus, novel metrics to quantify the scoring quality shall be developed that take the temporal structure into account but not overestimating differences at transitions, e.g., the start or end of REM sleep episodes.

### Which Method Is the Best?

Despite the difficulties to select the best method as the performance was very similar, we see some trends. Neural networks of all types detected stage 1 better than RF classifiers. This was especially evident when we applied the methods to the second dataset (patients), which indicates a better generalization of neural networks.

The RF classification with HMM and MF smoothing was superior to the RF classification without smoothing, and the networks based on the raw data input tended to be superior to features based networks, in particular when they were applied to the data of another laboratory and to a different subject population.

Given that our results and those of other groups (Tsinalis et al., 2016; Supratak et al., 2017; Chambon et al., 2018; Phan et al., 2018; Sors et al., 2018) are very close to the performance of the human expert we think that future evaluations of automatic scoring shall be performed using the multiple expert scoring and some other metric than F1 score or Cohen's kappa. An ideal metric shall take the temporal structure of sleep into account and treat sleep not as a set of epochs but as a set of sleep episodes. For example, a short REM sleep episode at the sleep onset does not affect F1 scores and kappa much, but might be of clinical significance. A good metric shall penalize such mistakes in scoring.

### Importance of the Training Data

An improvement of performance was achieved when the training was performed on a mixture of the two datasets, which suggests that one should train on as diverse data as possible to reach best performance. However, the models trained only on the first dataset (healthy participants) performed reasonably well on the second "unfamiliar" dataset (patients) showing a good generalization.

In case an electrode has high impedance, the signal might become very noisy. For example, as neural nets learned that a low muscle tone is required to score REM sleep, noisy, or bad EMG signals may deteriorate the performance considerably. The same holds for the EOG: if the signal quality is bad, then the algorithms may not be able to detect eye movements properly. These problems can be addressed by visual inspection of the signals before applying an algorithm and selecting the one working best with the available signals. It is also possible to develop tools for automatic examination of data quality and the subsequent selection of a corresponding algorithm.

Sometimes our models mistakenly classified epochs close to sleep onset as REM sleep, which is unlikely to occur in healthy subjects. A human expert most likely would not make such a mistake. This can be partially explained by the fact that we never presented the whole night to our neural networks and they could thus not learn that REM sleep is unlikely to occur at the beginning of sleep. Human scorers, however, have this knowledge. Some groups of patients, for example, those suffering from narcolepsy, often have REM sleep at the sleep onset, called sleep onset REM (SOREM) sleep episodes. Thus, it is important to be able to detect SOREM sleep episodes. They may occur also in healthy people in the early morning due to the circadian regulation of REM sleep (Sharpley et al., 1996; Mayers and Baldwin, 2005; Mccarley, 2007) or by experimental manipulation (Tinguely et al., 2014). They further may occur in sleep-deprived subjects, and in depressed patients, which are withdrawn from selective serotonin reuptake inhibitor (SSRI) medication (Sharpley et al., 1996; Mayers and Baldwin, 2005; Mccarley, 2007). Therefore, we did not introduce any priors preventing our algorithms from classifying epochs at sleep onset as REM sleep.

The main question, however, is how representative are the training data. We trained on healthy young participants and specific patients (narcolepsy, hypersomnia). Thus, it does not represent the entire spectrum of healthy subjects (form infancy to old age) and the patient population.

### Effect of the Length of the Sequence

We limited the length of the training sequences to 8 epochs but also tested the effect of 32 and 128-epoch long sequences. Networks trained on 128 epoch long sequences did not perform well when presented with unfamiliar datasets, i.e., they generalized worse. It might be the networks learned more global

structures of sleep and thus did not perform well on recordings with different structures (MSLT, disturbed sleep, patients, etc.). We noticed that longer sequences led to less stage changes, i.e., more consolidated sleep stages than scored by experts. Thus, we think it is better to keep the length of the training sequence short (eight epochs).

### Room for Further Improvement

We see a lot of room for further improvement. The sleep scoring manual was first developed for the scoring of healthy sleep, and is also being used for sleep in different kind of patients and people under the influence of medication or drugs. The wake EEG can also be affected by substances (Von Rotz et al., 2017). Thus, we recommend extending the training data including data from different laboratories, different pathologies, age groups and so on. One can also try to use data augmentation to increase the robustness of neural networks.

A major limitation of our study was the expert scoring: it was performed by a single expert although different ones. We suppose, that performance would have increased if several scorers would have scored the same data and consensus scoring would have been used for the training of the models. Also, human scorers have difficulties with ambiguous data and interscorer variability results in part due epochs that are difficult to score with confidence (Younes et al., 2016).

We showed that our algorithms had a good generalization capability to the patient population, but the performance was not as good as with healthy subjects. One possible reason might be the different scoring epoch length. We used the conversion procedure which worked well for most epochs, but certain discrepancies may show up at transition phases. We think this might have limited the performance, especially when these data were used for training. It was a compromise we had to make. Ideally all the data would be scored with the same epoch length. Phan et al. (2018) used a different approach and converted 20-s epochs to 30-s epochs by including the 5 s before and after a 20-s epoch.

Another aspect concerns movement time resulting in an artifact. In our datasets it was not scored, and in the AASM manual (Iber et al., 2007) scoring of movement time was abolished, which in our opinion is not optimal. Movement time basically results in EEG artifacts and it is thus difficult to assign a particular sleep stage. We suspect that the performance of the algorithms would improve if movement artifacts would have been scored as a separate class. Similarly, every artifact scored as some stage of sleep causes problems as artifacts do not look like sleep and thus such issues are equivalent to mistakes in the labels presented to the machine learning algorithm.

Recent work with automatic scoring on a large dataset (Sun et al., 2017) revealed that increasing the size of the dataset improved the performance. In the case of Sun et al. (2017) saturation occurred at approximately 300 recordings in the training set. However, their approach was feature based. We expect that saturation will occur at much larger numbers of recordings in the training set in case of DNNs working with raw data.

We demonstrated that it is possible to reliably score sleep automatically in polysomnographic recordings using modern deep learning approaches. It was also possible to identify stage 1 and REM sleep as reliable as human experts. In general, our models provided high quality of scoring, comparable to human experts, and worked with data of different laboratories and in healthy participants and patients. Furthermore, it was possible to successfully score MSLT recordings with a different structure than night time sleep recordings. We demonstrated that the local temporal structure in the data is important for sleep scoring. Some of our methods may also be applied for the on-line detection of sleep and could thus be used with mobile devices or to detect sleep in a driving simulator.

# ETHICS STATEMENT

This study was carried out in accordance with the recommendations of Review Board of the Swiss Federal Institute of Technology in Zurich, Switzerland (Dataset 1) and of the Institutional Review Board of Institute of Psychiatry and Neurology, Warsaw, Poland (Dataset 2), with written informed consent from all subjects in accordance with the Declaration of Helsinki. The protocol was approved by the Institutional Review Board of the Swiss Federal Institute of Technology in Zurich, Switzerland (Dataset 1) and by the Institutional Review Board of Institute of Psychiatry and Neurology, Warsaw, Poland (Dataset 2).

# AUTHOR CONTRIBUTIONS

AM, DL, and PA designed the analyses. AM conducted the analyses. XO, RR, AM, PA, AlW, AdW, and WJ collected the data and performed initial analyses. SB and JB provided computational resources and consultations on the methods. AM and PA wrote the manuscript. All authors commented and accepted the final version.

# FUNDING

This study was supported by nano-tera.ch (Grant 20NA21\_145929) and the Swiss National Science Foundation (Grant 32003B\_146643). The project was partially supported by the Max Planck ETH Center for Learning Systems. We are thankful to the NVIDIA Corporation for providing a GPU in the frame of academic GPU seeding.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins.2018. 00781/full#supplementary-material

### REFERENCES

fnins-12-00781 November 2, 2018 Time: 17:7 # 14



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Malafeev, Laptev, Bauer, Omlin, Wierzbicka, Wichniak, Jernajczyk, Riener, Buhmann and Achermann. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Serotonergic Input to Orexin Neurons Plays a Role in Maintaining Wakefulness and REM Sleep Architecture

Yuki C. Saito1,2, Natsuko Tsujino1,2, Manabu Abe3,4, Maya Yamazaki<sup>3</sup> , Kenji Sakimura<sup>3</sup> and Takeshi Sakurai2,5 \*

<sup>1</sup> Department of Molecular Neuroscience and Integrative Physiology, Faculty of Medicine, Institute of Medical, Pharmaceutical and Health Sciences, Kanazawa University, Kanazawa, Japan, <sup>2</sup> International Institute for Integrative Sleep Medicine, University of Tsukuba, Tsukuba, Japan, <sup>3</sup> Department of Cellular Neurobiology, Brain Research Institute, Niigata University, Niigata, Japan, <sup>4</sup> Department of Animal Model Development, Brain Research Institute, Niigata University, Niigata, Japan, <sup>5</sup> Faculty of Medicine, University of Tsukuba, Tsukuba, Japan

### Edited by:

Ritchie Edward Brown, VA Boston Healthcare System, United States

### Reviewed by:

Christopher S. Leonard, New York Medical College, United States Matt Carter, Williams College, United States

\*Correspondence:

Takeshi Sakurai sakurai.takeshi.gf@u.tsukuba.ac.jp; takeshi.sakurai@gmail.com

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 06 August 2018 Accepted: 14 November 2018 Published: 30 November 2018

### Citation:

Saito YC, Tsujino N, Abe M, Yamazaki M, Sakimura K and Sakurai T (2018) Serotonergic Input to Orexin Neurons Plays a Role in Maintaining Wakefulness and REM Sleep Architecture. Front. Neurosci. 12:892. doi: 10.3389/fnins.2018.00892 Neurons expressing neuropeptide orexins (hypocretins) in the lateral hypothalamus (LH) and serotonergic neurons in the dorsal raphe nucleus (DR) both play important roles in the regulation of sleep/wakefulness states, and show similar firing patterns across sleep/wakefulness states. Orexin neurons send excitatory projections to serotonergic neurons in the DR, which express both subtypes of orexin receptors (Mieda et al., 2011), while serotonin (5-HT) potently inhibits orexin neurons through activation of 5HT1A receptors (5HT1ARs). In this study, we examined the physiological importance of serotonergic inhibitory regulation of orexin neurons by studying the phenotypes of mice lacking the 5HT1A receptor gene (Htr1a) specifically in orexin neurons (ox5HT1ARKO mice). ox5HT1ARKO mice exhibited longer NREM sleep time along with decreased wakefulness time in the later phase of the dark period. We also found that restraint stress induced a larger impact on REM sleep architecture in ox5HT1ARKO mice than in controls, with a larger delayed increase in REM sleep amount as compared with that in controls, indicating abnormality of REM sleep homeostasis in the mutants. These results suggest that 5HT1ARs in orexin neurons are essential in the regulation of sleep/wakefulness states, and that serotonergic regulation of orexin neurons plays a crucial role in the appropriate control of orexinergic tone to maintain normal sleep/wake architecture.

Keywords: sleep, wakefulness, orexin, serotonin, stress

### INTRODUCTION

Orexin A and orexin B, also known as hypocretin-1 and hypocretin-2, respectively, are important regulators of sleep/wakefulness states (Sakurai, 2007, 2014). Orexin/hypocretin-producing neurons (orexin neurons) are located in the lateral hypothalamus (LH) and send axonal projections widely to the entire brain, except the cerebellum, with especially abundant projections to the monoaminergic/cholinergic nuclei in the hypothalamus and brainstem regions (Peyron et al., 1998; Date et al., 1999; Nambu et al., 1999). The actions of orexins are mediated via two G-protein

coupled receptors (GPCRs), orexin 1 receptor (OX1R) and orexin 2 receptor (OX2R) (Sakurai et al., 1998). OX1R and OX2R mRNAs exhibit a markedly different and basically complementary distribution, suggesting that these receptors have distinct physiological roles through different neuronal pathways (Sakurai, 2007). The importance of orexins in the maintenance of consolidated sleep/wakefulness states is highlighted by the finding that the sleep disorder narcolepsy is caused by orexin deficiency in several mammalian species, including mice, dogs, and humans (Peyron et al., 1998; Chemelli et al., 1999; Lin et al., 1999; Thannickal et al., 2000). Recent investigations have suggested additional functions of orexins in the regulation of emotions, energy homeostasis, reward systems, drug addiction, and autonomic function (Sakurai, 2014; Soya et al., 2017).

Elucidation of the regulatory mechanisms of orexin neurons is important for understanding the physiological mechanisms of sleep/wakefulness control and the roles of orexin neurons in these functions. Orexin neurons receive innervations from many brain regions, including the limbic system, preoptic area, and monoaminergic neurons (Sakurai et al., 2005; Yoshida et al., 2006; Saito et al., 2018). Among the factors that influence the activity of orexin neurons, serotonin (5-HT) was shown to strongly inhibit most orexin neurons through 5HT1A receptors (5HT1ARs) (Muraki et al., 2004; Tabuchi et al., 2014; Saito et al., 2018). At the same time, GABAergic neurons in the preoptic area that send direct inhibitory projections to orexin neurons are inhibited by 5- HT (Saito et al., 2018), implying that the serotonergic regulation of orexin neurons is complex. A retrograde tracing study of DR 5-HT neurons using rabies virus also showed that orexin neurons send innervations to serotonergic neurons in the DR, suggesting a close relationship between orexin and 5-HT (Pollak Dorocic et al., 2014). However, the function of direct serotonergic regulation of orexin neurons has remained unclear.

In the present study, we generated mice with a selective deletion of the 5HT1A receptor gene (Htr1a) specifically in orexin neurons (Htr1a<sup>f</sup> /<sup>f</sup> ;orexin-Cre mice, referred to as ox5HT1ARKO mice) to study the consequences of deletion of Htr1a in orexin neurons in mice to remove the serotonergic regulation of orexin neurons. ox5HT1ARKO mice showed shorter wakefulness time during the later phase of the dark period as compared with wild type and control mice. Alteration in the architecture of REM sleep induced by acute restraint stress was more prominent in ox5HT1ARKO mice than in controls. These observations suggest that serotonergic regulation of orexin neurons plays an important role in normal maintenance of wakefulness and homeostasis of REM sleep amount.

### MATERIALS AND METHODS

### Animals

All experimental procedures involving animals were approved by the Animal Experiment and Use Committee of Kanazawa University (AP-132649) and the University of Tsukuba (18-173), and thus were in accordance with NIH guidelines. Htr1a<sup>f</sup> /<sup>f</sup> mice with inclusion of lox-P sequences in the htr1a gene were generated by homologous recombination in C57BL/6N embryonic stem cells and implantation in 8-cell-stage embryos (ICR). A targeting vector was designed to flank the coding region of the Htr1a gene by lox-P sites (**Figure 1A**). Chimeric mice were crossed to C57BL/6J females (Jackson Labs). The pgk-Neo cassette was deleted by crossing them with FLP66 mice, which had been backcrossed to C57BL/6J mice more than 10 times. Initially, F1 hybrids from heterozygous × heterozygous mating were generated. We backcrossed them to C57BL/6J mice more than eight times. Genotypes were determined by PCR of mouse tail DNA.

Orexin-Cre transgenic mice (Matsuki et al., 2009) were crossed with wild type C57BL6/J mice more than 10 times. Orexin-Cre mice were mated with Htr1a<sup>f</sup> /<sup>f</sup> mice, and a breeding colony for producing orexin-Cre; Htr1a<sup>f</sup> /<sup>f</sup> was maintained by mating Ht1ar<sup>f</sup> /<sup>f</sup> with orexin/Cre; Ht1ar<sup>f</sup> /<sup>+</sup> mice. We used littermates as controls, and all experiments were performed in a C57BL/6J background. We confirmed that wild type mice, orexin-Cre transgenic mice, Ht1ar <sup>f</sup> /+;orexin-Cre, and Htr1a<sup>f</sup> /<sup>f</sup> (orexin-Cre-negative) mice all show the same sleep/wakefulness characteristics (data not shown). In this study, Htr1a<sup>f</sup> /<sup>f</sup> ;orexin-Cre transgenic mice were used as orexin neuron-selective 5HT1AR knockout mice (ox5HT1ARKO), and Htr1a<sup>f</sup> /+;orexin-Cre transgenic mice were used as hetero orexin neuronselective 5HT1AR knockout mice (ox5HT1ARHKO). Their Htr1a f /f ;orexin/Cre-negative littermates (control mice), which show comparable sleep/wakefulness phenotypes to those of wild type C57BL/6J mice, were used as controls. Mice were housed in a temperature- and humidity-controlled, 12 h light/12 h dark room (light on at 9:00), and access water and food ad libitum.

### Electrophysiological Study

For electrophysiological analysis, ox5HT1ARKO mice were mated with orexin-eGFP transgenic mice (Yamanaka et al., 2003) (C57BL/6J background) to obtain ox5HT1ARKO;orexin-eGFP mice, which express eGFP as a marker for detecting orexin neurons, and their brain slice preparations were subjected to patch-clamp recordings as previously described (Yamanaka et al., 2003). In brief, brains were extracted and cooled in ice-cold cutting solution consisting of (mM); 87 NaCl, 75 sucrose, 25 NaHCO3, 10 D(+)-glucose, 7 MgCl2, 2.5 KCl, 1.25 NaH2PO4, and 0.5 CaCl<sup>2</sup> bubbled with O<sup>2</sup> 95% and CO<sup>2</sup> 5%. Coronal brain slices (250 µm thick) including LHA were prepared with a vibratome (Leica, VT1200S) and incubated for 1 h at room temperature in artificial cerebrospinal fluid (ACSF) containing (mM); 125 NaCl, 26 NaHCO3, 10 D(+)-glucose, 2.5 KCl, 2 CaCl2, and 1 MgSO<sup>4</sup> bubbled with O<sup>2</sup> 95% and CO<sup>2</sup> 5%. The slices were transferred to a recording chamber (RC-27L, Warner Instrument Corp., Hamden, CT, United States) at room temperature on a fluorescence microscope stage (BX51WI, Olympus, Tokyo, Japan). Neurons that showed GFP fluorescence were used for recordings. The fluorescence microscope was equipped with an infrared camera (C-3077, Hamamatsu Photonics, Hamamatsu, Japan) for infrared differential interference contrast (IR-DIC) imaging and a CCD camera (JK-TU53H, Olympus) for fluorescent imaging. Each image was displayed separately on a monitor. Recordings were carried out with an Axopatch 700B amplifier (Axon Instruments, Foster City, CA, United States)

mice. (A) Targeting strategy for making Htr1af/<sup>f</sup> mice. (B) Matched brain sections from mice with various genotypes were double-stained with anti-orexin (red) and anti-5HT1AR (green) serum. Double-positive cells are shown by white arrowheads. Orexin neurons lacking 5HT1AR expression are shown by arrows. WT, wild type; 5HT1ARf/<sup>f</sup> , Htr1af/<sup>f</sup> (orexin-Cre-negative); ox5HT1ARHKO, Htr1af/<sup>+</sup> (hetero) mice (orexin-Cre transgene-positive); ox5HT1ARKO, Htr1af/<sup>f</sup> mice with orexin-Cre transgene. (C) Matched brain sections containing the DR from ox5HT1ARKO and 5HT1ARf/<sup>f</sup> were doubly stained with anti-TPH (red) and anti-5HT1AR (green) antibodies, showing similar staining patterns. Left, control (5HT1ARf/f); Right ox5HT1ARKO, lower panels show high power images of corresponding rectangle regions in upper panels.

using a borosilicate pipette (GC150-10, Harvard Apparatus, Holliston, MA, United States) prepared by a micropipette puller (P-97, Sutter Instruments, Pangbourne, United Kingdom) and filled with intracellular solution (4–10 M), consisting of (mM): 125 K-gluconate, 5 KCl, 1 MgCl2, 10 HEPES, 1.1 EGTA-Na3, 5 MgATP, 0.5 Na2GTP, pH7.3 with KOH. Osmolarity of the solution was checked by a vapor pressure osmometer (model 5520, Wescor, Logan, UT, United States). The osmolarity of the internal and external solutions was 280–290 and 320– 330 mOsm/l, respectively. The liquid junction potential of the patch pipette and perfused extracellular solution was estimated to be −16.2 mV and was applied to the data. The recording pipette was under positive pressure while it was advanced toward individual cells in the slice. Tight seals of 0.5–1.0 G were made by negative pressure. The membrane patch was then ruptured by suction. The series resistance during recording was 10–25 M and was compensated. The reference electrode was an Ag–AgCl pellet immersed in bath solution. Spontaneous action potentials of GFP-expressing neurons were recorded with whole cell modes at 30◦C kept by perfusion of preheated ACSF. During recordings, cells were superfused with extracellular solution at a rate of 1.0–2.0 ml/min using a peristaltic pump (K.T. Lab, Japan) at room temperature.

### Histological Analyses

Mice were anesthetized with sodium pentobarbital and perfused transcardially with phosphate-buffered saline (PBS) followed by ice-cold 4% paraformaldehyde. Then, the perfused brains were postfixed overnight in the same fixative buffer and placed in 30% sucrose buffer for 2 days. Cryostat sections (30-µm thick) of the brains were incubated for 1 h in 0.1 M phosphate buffer containing 1% bovine serum albumin and 0.25% Triton-X-100, and incubated overnight at 4◦C with the primary antibodies. The primary antibodies used were guinea pig anti-orexin (provided by Prof. Watanabe of Hokkaido University, 1:1000); mouse anti-TPH (Aldrich, T0678, 1:200), and rabbit anti-5HT1AR (Acris, AP06769PU-N, 1:500). Secondary antibodies used were Alexa 488-conjugated donkey anti-rabbit IgG (Invitrogen, A21206, 1:800), Alexa 594-conjugated goat anti-guinea pig IgG (Invitrogen, A11076, 1:800) and Alexa 647-conjugated donkey anti-mouse IgG (Invitrogen, A31571, 1:800). Images were obtained with a fluorescence microscopes (Keyence BZ-9000) or confocal laser scanning microscopes (Leica SP8 and Olympus FV10i).

### Sleep Recordings

Male wild type mice (n = 8), ox5HT1ARKO mice (n = 6), ox5HT1ARHKO mice (n = 3) and their weight- and age-matched male control littermates: Htr1a <sup>f</sup> /<sup>f</sup> (orexin/Cre-negative) mice (n = 9) were implanted with electrodes at 12 weeks of age for EEG/EMG recording, as described previously (Hara et al., 2001). An electrode for EEG and EMG recording was implanted in the skull of each mouse. The three arms of the electrode for EEG recording were placed approximately 2 mm anterior and 2 mm to the right, 2 mm posterior and 2 mm to the right, and 2 mm posterior and 2 mm to the left of the bregma. Stainless steel wires for EMG recording were sutured to the neck muscles of each mouse bilaterally, and each electrode was glued solidly to the skull. After the recovery period, animals were moved to

a recording cage placed in an electrically shielded and sound attenuated room. A cable for signal output was connected to the implanted electrode, and animals were allowed to move freely. Mice are singly housed during recordings. Signals were amplified through an amplifier (AB-611J, Nihon Koden, Tokyo) and digitally recorded on a computer using EEG/EMG recording software (Vital recorder, Kissei Comtec). Animals were housed in a 12-h light/dark cycle and allowed to habituate to recording conditions at least 7 days. Each mouse was then recorded for three consecutive 24-h periods, beginning at lights-on at 09:00. Lights were turned off at 21:00. Food and water were replenished at 08:00, and mice were not otherwise disturbed in any way. Averages of each time point on these three recording days were used as raw data, and data from all individual animals used in these studies were used to determine their sleep/wakefulness characteristics.

### Restraint Stress

Stress was applied through restraining mice with a sleeve of nylon-mesh for 30 or 90 min until 21:00, just before the start of the dark period. We folded a nylon mesh sheet (150 mm × 200 mm) in two and hold a mouse in between folds with clips. Mice were released from the restraint, and returned to their cage for sleep-wakefulness recordings.

### Statistical Analysis

Data are presented as mean ± SEM and were analyzed by one-way ANOVA followed by post hoc analysis of significance by Fisher's Protected Least Significant Difference test or paired t-test using the GraphPad Prism6 software package (GraphPad Software).

# RESULTS

### Generation of Mice Lacking 5HT1<sup>A</sup> Receptors Exclusively in Orexin Neurons

To decipher the physiological role of serotonergic regulation of orexin neurons, we examined the phenotype of mice in which orexin neurons selectively lack expression of 5HT1AR, a sole subtype of serotonin receptor in orexin neurons (Muraki et al., 2004). To generate these mice, we first made mice in which the coding region of the gene encoding 5HT1AR (Htr1a) is flanked by two loxP sites (Htr1a <sup>f</sup> /<sup>f</sup> mice). Since Htr1a has a single exon, we introduced two loxP sites in regions corresponding to the 3<sup>0</sup> - and 5<sup>0</sup> -non-coding regions in exon 1 of this gene so that two loxP sequences flank the coding region (**Figure 1A**).

To obtain mice with a deletion of Htr1a restricted to orexin-producing neurons, we mated Htr1a <sup>f</sup> /<sup>f</sup> mice with orexin-Cre transgenic mice (Matsuki et al., 2009), in which orexin neurons specifically express Cre recombinase, and obtained ox5HT1ARKO mice (Htr1a <sup>f</sup> /<sup>f</sup> ;orexin-Cre mice). Double immunofluorescence studies in wild-type control mice showed that many neurons in the LH possessed 5HT1AR-like immunoreactivity (ir) in their soma and dendrites (**Figure 1B**). Among these, most orexin-ir cells (orexin-neurons) were positive for 5HT1AR-ir, consistent with our previous observation that all orexin neurons examined showed 5HT1AR-ir and strong hyperpolarization when administered 5-HT (Muraki et al., 2004). Similarly, we also observed that all orexin neurons in Htr1a<sup>f</sup> /<sup>f</sup> mice (Cre-negative) expressed 5HT1AR-ir in their soma and dendrites (**Figure 1B**). In stark contrast, neurons positive for both 5HT1AR-ir and orexin-ir were barely detectable in the brain of ox5HT1ARKO mice, although we observed many 5HT1AR-ir-positive neurons in the LHA that were negative for orexin (**Figure 1B**). Less than 10% of orexin neurons in ox5HT1ARKO mice were double-labeled for orexin and 5HT1AR (**Figure 1B**). The remaining expression of 5HT1AR in orexin neurons may be due to incomplete penetrance of Cre expression in orexin-Cre transgenic mice and/or incomplete deletion of the gene fragment between loxP sites in ox5HT1ARKO mice. These histological observations confirmed that in ox5HT1ARKO mice, 92.3 ± 8% (n = 5) of orexin neurons lack expression of 5HT1AR.

Gross anatomical and histological studies failed to detect any structural abnormalities in the brain of ox5HT1ARKO mice. Specifically, the number of orexin neurons in the LH remained normal; the number of immunoreactive cells (located from 0.76 mm to 2.52 mm posterior to the bregma) was 3,424 ± 154 and 3,462 ± 68 for control littermates and ox5HT1ARKO mice, respectively (n = 8 each).

**Figure 1C** shows expression of 5HT1ARs in the raphe nuclei in ox5HT1ARKO and control mice. Expression of 5HT1AR in these regions in ox5HT1ARKO mice was indistinguishable from that in transgene-negative Htr1a <sup>f</sup> /<sup>f</sup> controls, further supporting that 5HT1AR remained intact in other neurons in ox5HT1ARKO mice.

### Electrophysiological Characteristics of Orexin Neurons Lacking 5HT1A Receptors

We next examined responses of orexin neurons to 5-HT by whole cell patch-clamp recording using brain slice preparations from ox5HT1ARKO and control mice (**Figure 2**). To make identification of orexin neurons easier, we used ox5HT1ARKO and control mice possessing the orexin-eGFP transgene (Yamanaka et al., 2003). Bath application of 5-HT produced potent, dose-dependent hyperpolarization in all orexin neurons in wild type and control (Htr1a<sup>f</sup> /<sup>f</sup> , orexin-Cre negative) mice, consistent with our previous report (Muraki et al., 2004) (**Figures 2A–C**). These 5-HT-induced inhibitory effects rapidly reversed upon removal of 5-HT from the external solution (**Figure 2A**). 5-HT failed to induce hyperpolarization in most orexin neurons in ox5HT1ARKO mice (**Figure 2B**). Very faint hyperpolarization was observed in four out of 19 orexin neurons tested in slices from eight ox5HT1ARKO mice, while we observed strong 5-HT-induced hyperpolarization in all neurons examined in wild type and control mice. 5-HT induced a less potent hyperpolarizing effect on orexin neurons of hetero Htr1a<sup>f</sup> /+;orexin-Cre (ox5HT1ARHKO) mice as compared with controls, suggesting that the number of 5HT1ARs in these

FIGURE 2 | Lack of functional 5HT1A receptors in orexin neurons of ox5HT1ARKO mice. (A, upper) Representative trace showing the hyperpolarizing effect of 5-HT (10 mM) on orexin neurons in a slice preparation from a control mouse (5HT1ARf/<sup>f</sup> mouse without Cre expression) in a current-clamp recording. Drugs were applied during the periods indicated by bars. (A, lower) Representative trace showing the effect of 5-HT (10 mM) on orexin neurons in ox5HT1ARKO slice. 5-HT does not induce any response, but GABA (10 mM) induces a strong inhibitory response in the same neuron, showing the neuron is viable. (B) Hyperpolarization of orexin neurons from mice with various genotypes, induced by 5-HT (10 mM). WT, wild type; 5HT1ARf/<sup>f</sup> , Htr1af/<sup>f</sup> (orexin-Cre-negative); ox5HT1ARHKO, Htr1af/<sup>+</sup> (hetero) mice (orexin-Cre transgene-positive); ox5HT1ARKO, Htr1af/<sup>f</sup> mice with orexin-Cre transgene. (C) Dose–response curves of 5-HT-induced hyperpolarization of orexin neurons prepared from mice with various genotypes. Horizontal axis show -log dose of 5HT.

ox5HT1ARHKO mice is lower than that in wild type mice, due to haploinsufficiency (**Figure 2C**).

Orexin neurons lacking 5HT1AR showed a resting membrane potential similar to that of orexin neurons in control littermates (ox5HT1ARKO, −42.2 ± 3.2 mV, n = 19; control, −42.3 ± 2.7 mV, n = 12, p = 0.98).

### ox5HT1ARKO Mice Exhibited Decreased Wakefulness Time in Later Phase of Dark Period

Sleep/wakefulness state patterns of ox5HT1ARKO mice, ox5HT1ARHKO mice, control littermate mice (Htr1a<sup>f</sup> /<sup>f</sup> mice, negative for Cre) and wild type mice were examined by EEG/EMG recording. We found that the total wakefulness time was significantly shorter and NREM sleep was longer in ox5HT1ARKO mice than in control mice in the latter half of the dark period (**Figure 3** and **Table 1**). We did not find any difference in episode duration of each state in ox5HT1ARKO mice as compared with any other genotypes

ox5HT1ARHKO and control littermate mice (Htr1af/<sup>f</sup> mice, negative for Cre) were examined by simultaneous EEG/EMG recordings. Total time of wakefulness (WAKE) (A), NREM sleep (B) and REM sleep (C) in 2-h periods. <sup>∗</sup>p < 0.05, ∗∗p < 0.01, significantly different from control mice. Bonferroni test after two-way ANOVA.


### 5HT1A Receptors in Orexin Neurons Play an Important Role in Homeostatic Regulation of REM Sleep After Exposure to Acute Stress

Physical stress has been shown to influence the sleep architecture in rats, and especially lead to alteration of the REM sleep structure mediated by serotonergic mechanisms (Rampin et al., 1991). Orexin neurons were shown to be activated by stress (Chang et al., 2007), and orexins potently inhibit REM sleep (Mieda et al., 2011). These observations suggest the possibility that the 5HTDR → orexin pathway plays a role in stress-induced alterations in REM sleep architecture. To evaluate this hypothesis, we applied a restraint stress to ox5HT1ARKO and control mice to observe alterations in the sleep architecture. Mice were subjected to mild restraint with nylon mesh for 30 or 90 min just before the onset of the dark period, and subjected to sleep recording. Both control mice and ox5HT1AKO mice showed a similar tendency for an increase in NREM sleep amount during the dark period after release from 30- or 90-min restraint stress, which is likely to be attributable to sleep deprivation during stress application. However, the stress did not alter the amount of REM sleep in control mice, while ox5HT1ARKO mice responded by an increase in REM sleep amount as compared with the basal condition (**Figure 4A**). This change was due to an increase in the number of REM sleep episodes, and not to lengthening of REM sleep duration (**Figure 4B**). ox5HT1ARKO mice had a tendency for decreased REM sleep time during the dark period in the basal condition (**Figure 3C**), but after applying stress, they exhibited a similar amount of REM sleep as compared with control mice. Altogether, these data show that ox5HT1ARKO mice exhibited augmented REM sleep enhancement responses to acute physical stress.

# DISCUSSION

Orexin neurons play highly important roles in the maintenance of sleep/wakefulness states (Sakurai, 2007). The regulatory mechanisms of orexin neurons have been examined by many electrophysiological studies in vitro, and many factors that can modulate the activity of these neurons have been identified. However, the physiological relevance of the regulatory mechanisms or inputs affecting the activity of orexin neurons remains largely unknown. In our previous study, we examined the phenotypes of mice in which GABA<sup>B</sup> receptors were selectively removed in orexin neurons (Matsuki et al., 2009). These mice showed highly fragmented sleep/wakefulness states during both the dark and light periods, suggesting the importance of GABA-mediated control of orexin neurons in physiological

fnins-12-00892 November 28, 2018 Time: 20:57 # 6

(control conditions). Bonferroni test after two-way ANOVA.

sleep/wakefulness regulation. Here, we addressed the importance of 5-HT-mediated inhibitory regulation of orexin neurons, since we found 5-HT shows a very potent inhibitory effect on almost all orexin neurons, and 5-HT also plays an important role in the regulation of sleep/wakefulness states. Firing rates of serotonergic neurons are lower during NREM sleep than during wakefulness and lowest during REM sleep, displaying a similar firing pattern to that of orexin neurons across sleep/wakefulness

states (Lee et al., 2005; Mileykovskiy et al., 2005; Takahashi et al., 2008). Orexin neurons send excitatory innervation to serotonergic neurons in the raphe nuclei, which abundantly express both OX1R and OX2R (Brown et al., 2002; Liu et al., 2002; Kohlmeier et al., 2013), suggesting that activity of serotonergic neurons is partly regulated by orexin neurons, and this function was shown to be important for inhibition of cataplexy in narcoleptic mice (Mieda et al., 2011; Hasegawa et al., 2017). Direct serotonergic input to orexin neurons might tonically inhibit orexin neurons during wakefulness states, constituting negative feedback regulation. We also confirmed that optogenetic activation of serotonergic fibers in the LH inhibited orexin neurons in vitro (Saito et al., 2018). Although chemogenetic activation of 5-HT neurons in the DR showed minimal effects on activity of orexin neurons in vivo (Saito et al., 2018), the physiological relevance of direct serotonergic influences on orexin neurons in the regulation of sleep/wakefulness states has been unclear.

Here, we made mice in which orexin neurons selectively lack 5HT1AR, the sole subtype of serotonin receptors expressed in orexin neurons (ox5HT1ARKO mice) (**Figure 1**). ox5HT1ARKO mice showed decreased wakefulness time and increased NREM sleep time in the dark period, especially in the later half, as compared with control mice, suggesting that these mice suffer from hypersomnolence. This observation suggests that 5HT1ARmediated regulation of orexin neurons plays a role in sleep regulation.

Initially, we hypothesized that deletion of serotonergic input to orexin neurons by disrupting the Htr1a gene would results in increased wakefulness, because 5-HT strongly inhibits orexin neurons. However, orexin neurons receive complex regulation by the serotonergic system. 5-HT neurons directly inhibit orexin neurons, but indirectly disinhibit these cells through inhibition of GABAergic neurons in the ventrolateral preoptic area (VLPO) that send innervation to orexin neurons (Saito et al., 2018). Therefore, total 5-HT output might have a net excitatory effect on orexin neurons. However, contrary to our initial hypothesis, ox5HT1ARKO mice showed increased NREM sleep time in the later half of the dark period as compared with control mice, suggesting that these mice suffer from hypersomnolence (**Figure 3**). It has been shown that serotonergic tone peaked in the earlier half of the dark period, but decreased in the later half of the dark period in rats (Liu and Borjigin, 2006). This suggests the possibility that decreased serotonergic activity in the later half of the dark period results in a decrease in 5HT1AR-mediated inhibition of orexin neurons, leading to activation of these cells to increase wakefulness. In accordance with this hypothesis, wild type mice show a temporal increase in wakefulness time during the later half of the dark period, which peaks at CT7, but this increase was not observed in ox5HT1ARKO mice (**Figure 3B**). This difference mainly contributes to the decreased amount of wakefulness in ox5HT1ARKO mice as compared with control mice. In addition, chronic deficiency of 5HT1AR-mediated inhibition of orexin neurons might result in plastic changes of GABAergic/glutamatergic regulations of orexin neurons. Compensatory increase in GABAergic input and/or decrease in glutamatergic input to orexin neurons might play a role in maintaining wakefulness amount within normal range in the earlier phase of the dark period, but this change might simultaneously decease the activity of these cells in the later half of the dark period, when serotonergic tone is decreased.

The phenotype of ox5HT1ARKO mice demonstrates the importance of serotonergic activity within a defined neuronal circuit; in particular, that expression of these 5HT1ARs in orexin neurons is necessary for appropriate maintenance of wakefulness especially during the later half of the dark period. It was shown that overexpression of 5HT1AR in mice rather affected sleep/wakefulness states in the earlier half of the dark period, with fragmentation of sleep/wakefulness (Tabuchi et al., 2014). This also suggests the possibility that the serotonergic influence on orexin neurons is maximal in the earlier half of the dark period, while a decrease in serotonergic influence rather disinhibits orexin neurons to support wakefulness.

We also examined the effect of acute stress on sleep/wakefulness states in ox5HT1ARKO mice, because stress might affect serotonergic regulation of orexin neurons. ox5HT1ARKO mice showed greater REM sleep rebound after release from restraint, as compared with wild type mice (**Figure 4**). This result suggests that serotonergic regulation of orexin neurons plays an important role in REM sleep homeostasis. These results suggest the possibility that stress activates orexin neurons, leading to transient activation of 5-HT neurons. Orexin neurons might be disinhibited by decreased serotonergic tone after release from stress, to inhibit REM sleep in wild type mice. REM was gradually increased after the release from the stress, peaking at 8 h after the stress, suggesting that DR serotonergic activity might gradually decrease after the stress over several hours.

This study also provides an insight into the mechanism by which drugs influencing serotonergic tone act on sleep physiology. Compounds such as SSRIs and anti-depressants may increase serotonergic tone in orexin neurons, which could affect sleep/wakefulness states at least partly through dysregulation of orexin neurons. Furthermore, this study suggests that decreased serotonergic influence on orexin neurons might be one of the possible mechanisms of sleep disturbance in patients with depression.

# AUTHOR CONTRIBUTIONS

YS performed maintenance of mice, sleep analyses, and immunostaining. NT performed electrophysiological studies. MA, MY, and KS generated 5HT1A floxed mice. TS designed the experiments, wrote the manuscript, and performed analyses.

# FUNDING

This study was supported by a JSPS KAKENHI Grant-in-Aid for Scientific Research (B) (JP 15H03122, 18H02595) (TS), a KAKENHI Grant-in-Aid for Scientific Research on Innovative Areas, "Willdynamics" (16H06401) (TS), and a Grant-in-Aid for Young Scientists (B) (16K21057) (YS).

### SUPPLEMENTARY MATERIAL

fnins-12-00892 November 28, 2018 Time: 20:57 # 9

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2018.00892/full#supplementary-material

### REFERENCES


FIGURE S1 | EEG power density of NREM sleep (A) and REM sleep (B) are shown as the mean percentage of total EEG power ± SEM in control littermate (n = 7) and ox5HT1ARKO mice (n = 6) for 0.25-Hz frequency bins between 0.25 and 20 Hz. The delta range (0.75–4 Hz) is indicated by the black bar and the theta range (6–9 Hz) is indicated by the gray bar. Bonferroni test after two-way ANOVA.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Saito, Tsujino, Abe, Yamazaki, Sakimura and Sakurai. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Oral Delivered Dexmedetomidine Promotes and Consolidates Non-rapid Eye Movement Sleep via Sleep–Wake Regulation Systems in Mice

### Zhen-Xin Feng<sup>1</sup>† , Hui Dong<sup>2</sup>† , Wei-Min Qu<sup>2</sup> \* and Wei Zhang<sup>1</sup> \*

<sup>1</sup> Department of Anesthesiology, The First Affiliated Hospital of Zhengzhou University, Zhengzhou, China, <sup>2</sup> Department of Pharmacology, School of Basic Medical Sciences, Fudan University, Shanghai, China

### Edited by:

Kaspar Emanuel Vogt, University of Tsukuba, Japan

### Reviewed by:

Christine Dugovic, Janssen Research & Development, Belgium Robert Warren Gould, Vanderbilt University, United States

### \*Correspondence:

Wei-Min Qu quweimin@fudan.edu.cn Wei Zhang zhangwei\_571012@163.com †These authors have contributed equally to this work

### Specialty section:

This article was submitted to Neuropharmacology, a section of the journal Frontiers in Pharmacology

Received: 19 July 2018 Accepted: 28 September 2018 Published: 05 December 2018

### Citation:

Feng Z-X, Dong H, Qu W-M and Zhang W (2018) Oral Delivered Dexmedetomidine Promotes and Consolidates Non-rapid Eye Movement Sleep via Sleep–Wake Regulation Systems in Mice. Front. Pharmacol. 9:1196. doi: 10.3389/fphar.2018.01196 Dexmedetomidine, a highly selective α2-adrenergic agonist, is widely used in clinical anesthesia and ICU sedation. Recent studies have found that dexmedetomidine-induced sedation resembles the recovery sleep that follows sleep deprivation, but whether orally delivered dexmedetomidine can be a candidate for the treatment of insomnia remains unclear. In this study, we estimated the sedative effects of orally delivered dexmedetomidine by spontaneous locomotor activity (LMA), and then evaluated the hypnotic effects of dexmedetomidine on sleep–wake profiles during the dark and light phase using electroencephalography/electromyogram (EEG/EMG), respectively. Using c-Fos staining, we explored the effects of dexmedetomidine on the cerebral cortex and the sub-cortical sleep–wake regulation systems. The results showed that orally delivered dexmedetomidine at 2 h into the dark cycle reduced LMA and wakefulness in a dose-dependent manner, which was consistent with the increase in non-rapid eye movement sleep (NREM sleep). However, dexmedetomidine also induced a rebound in LMA, wake and rapid eye movement sleep (REM sleep) in the later stage. In addition, orally delivered dexmedetomidine 100 µg/kg at 2 h into the light cycle shortened the latency to NREM sleep and increased the duration of NREM sleep for 6 h, while decreased REM sleep for 6 h. Sleep architecture analysis showed that dexmedetomidine stabilized the sleep structure during the light phase by decreasing sleep–wake transition and increasing long-term NREM sleep (durations of 1024–2024 s and >2024 s) while reducing short-term wakefulness (duration of 4– 16 s). Unlike the classic hypnotic diazepam, dexmedetomidine also increased the delta power in the EEG spectra of NREM sleep, especially at the frequency of 1.75–3.25 Hz, while ranges of 0.5–1.0 Hz were decreased. Immunohistochemical analysis showed that orally delivered dexmedetomidine 100 µg/kg at 2 h into the dark cycle decreased c-Fos expression in the cerebral cortex and sub-cortical arousal systems, while it increased c-Fos expression in the neurons of the ventrolateral preoptic nucleus. These results indicate that orally delivered dexmedetomidine can induce sedative and hypnotic effects by exciting the sleep-promoting nucleus and inhibiting the wake-promoting areas.

Keywords: dexmedetomidine, LMA, EEG/EMG, sleep–wake, c-Fos

# INTRODUCTION

fphar-09-01196 December 3, 2018 Time: 11:7 # 2

Sleep exists in the overwhelming majority of organisms, from humans to worms (Allada and Siegel, 2008). Sleep has an essential role in many vital physiologic functions, including energy conservation, brain waste clearance, cognition, memory, and modulation of the immune system (Zielinski et al., 2016). Insomnia is a common sleep disorder, defined as a complaint of prolonged sleep latency, difficulties in maintaining sleep, and subsequent impairments in daytime functioning (Riemann et al., 2010). It has been identified as a critical and growing cause of concern for public health (Nomura et al., 2010).

Treatment for insomnia is still mainly drug-based therapy. The pharmacologic approach primarily includes benzodiazepines and non-benzodiazepines. These classic hypnotics all act on gamma-aminobutyric acid type A receptors, mediating inhibition of the CNS and improving sleep related problems (Richey and Krystal, 2011; Uygun et al., 2016). However, these drugs reduce the depth of NREM sleep, and thus do not mimic physiological sleep (Akeju and Brown, 2017). In addition, these "sleeping pills" are associated with neurocognitive dysfunction (Manconi et al., 2017; de Zambotti et al., 2018). Other hypnotics, such as antihistamines, antipsychotics, and melatonin, are not recommended for insomnia treatment due to side-effects according to the European Guideline for the diagnosis and treatment of insomnia (Riemann et al., 2017).

Accumulating experimental evidence indicates that anesthesia and sedatives to a lesser or greater extent exert their function in natural sleep-wake circuits (Franks, 2008; Han et al., 2014). Dexmedetomidine, a α2-adrenoceptor agonist, has been widely used in clinical anesthesia and ICU sedation (Mantz et al., 2011). Patients receiving dexmedetomidine for effective sedation are still easily aroused from a state similar to sleep that is not observed in other clinical anesthetics (Venn et al., 1999; Huupponen et al., 2008). According to a previous study, the locus coeruleus (LC) is pivotal in inducing the hypnotic response to dexmedetomidine (Correa-Sales et al., 1992). Electrophysiology studies confirm that dexmedetomidine opens inwardly, rectifying potassium channels to hyperpolarize the discharge of LC neurons and reduce norepinephrine (NA) release (Chiu et al., 1995). Decreased NA levels release the inhibition of the preoptic area, resulting in excitement of the sleep-promoting nucleus, thereby inhibiting the wake-promoting areas through mediating gamma-aminobutyric acid (GABA) and galanin (Nelson et al., 2003; Zhang et al., 2015). These advances suggest that dexmedetomidine may converge on the endogenous sleep pathway to introduce sedation.

Although dexmedetomidine induced sedation resembles the recovery sleep that follows sleep deprivation (Zhang et al., 2015), it remains unclear whether it could be an ideal candidate for the treatment of insomnia. An ideal agent should effectively shorten sleep latency, increase the amount of sleep time, stabilize the sleep structure, and insure sleep depth. In addition, convenient routes of administration, such as oral delivery, can improve the compliance of patients. Intravenous administration is not only inconvenient to self-operate, but it also has the risk of bacterial infection and pain. However, in clinical practice, dexmedetomidine is always given as a continuous intravenous pump infusion (Ohtani et al., 2011) or nasal drip in pediatric sedation (Li et al., 2016). In preclinical research, intraventricular injection or intraperitoneal administration is the most common route of administration. Until now, the sedative and hypnotic effects of orally delivered dexmedetomidine remain unclear and the impacts of dexmedetomidine on sleep structure are not yet fully understood. Moreover, dexmedetomidine can still induce hypnotic effect in mice unable to synthesize NA (Gilsbach et al., 2009; Hu et al., 2012; Sanders and Maze, 2012; Garrity et al., 2015) or with selective knockdown of alpha-2A adrenergic receptors in the LC (Zhang et al., 2015), suggesting that dexmedetomidineinduced hypnosis may depend on other brain areas in addition to the LC.

Here, we used LMA and EEG/EMG to investigate whether orally delivered dexmedetomidine has sedative and hypnotic effects during the dark phase when mice are active. We then, evaluated the effects of dexmedetomidine on the time of NREM sleep, NREM sleep latency, number of sleep-wake transitions, structure of sleep-wake profiles, and the delta power density of NREM sleep during the light phase when mice were sleepy. Finally, the effects of dexmedetomidine on the expression of c-Fos protein in the brain were explored by immunohistochemistry.

# MATERIALS AND METHODS

### Animals

Male SPF C57BL/6J mice aged 8–10 weeks (weighing 20–24 g) were purchased from the Laboratory Animal Center, Chinese Academy of Sciences (Shanghai, China). The mice were housed in an insulated and soundproofed room maintained at an ambient temperature of 22 ± 0.5◦C with a relative humidity of 60 ± 2% under an automatically controlled 12-h light/12-h dark (L/D) cycle [light on at 7:00 A.M, illumination intensity≈10 lx(Zhang et al., 2016)]. Food and water were available ad libitum. Experimental protocols were approved by the Medical Experimental Animal Administrative Committee of Shanghai and strictly followed the Guidelines from the National Institute of Health (U.S.) regarding the care and use of animals for experimental procedures. Every effort was made to minimize the number of animals for experiments and any pain or discomfort they experienced.

### Chemicals

Dexmedetomidine was obtained from the Jiangsu HengRui Pharmaceutical Co. Ltd. (Jiangsu, China). Rabbit polyclonal antic-Fos antibody was purchased from Abcam (Cambridge, MA, United States). Biotinylated donkey anti-rabbit IgG and avidinbiotin-peroxidase were purchased from Vector Laboratories (Burlingame, CA, United States). Finally, 3, 3-diaminobenzidinetetra-hydrochloride (DAB) was purchased from Sigma-Aldrich (St. Louis, MO, United States). Dexmedetomidine was dissolved in sterile saline before use.

# Spontaneous Locomotor Activity in an Accustomed Environment

Spontaneous locomotor activity was measured according to the method described previously (Inoue et al., 1996). The recording system mainly consisted of soundproof cabinets, a transparent recording cage (L 280 mm, W 230 mm, H 210 mm), a continuous infrared detector, recording software, and a monitor (Biotex, Kyoto). The bottom of each recording cage was divided into an average of 256 unit areas. Mouse movements in each unit area were identified by the infrared detector, and recorded as one activity. Activity was automatically monitored and calculated every 5 min. Before the start of recording, mice were housed individually in a transparent recording cage with food and water available, and habituated over three consecutive days. On the experimental day, the control group received saline while treatment groups received dexmedetomidine.

### Polygraphic Recordings and Vigilance State Analysis

Under pentobarbital anesthesia at 50 mg/kg (i.p.), mice were chronically implanted with EEG/EMG electrodes for polysomnographic recordings (Huang et al., 2005). The implant consisted of two stainless-steel screws (1 mm in diameter) and EEG electrodes inserted through the skull of the cortex (+1.0 mm anteroposterior; -1.5 mm mediolateral from the bregma or lambda) according to the atlas of Paxinos and Franklin (2013) and served as EEG electrodes. Two insulated stainless steel wires were bilaterally placed into both trapezius muscles and served as EMG electrodes. All of the electrodes were linked to a mini-connector and fixed to the skull with dental cement. After a 7-day recovery period, the mice were housed individually in transparent barrels and habituated to the recording cable for 7 days before recording was started. The uninterrupted synchronous recordings of EEG and EMG were performed by means of a slip ring, which was designed for letting the mice move freely.

As previously described (Huang et al., 2005), cortical EEG and EMG signals were amplified and filtered (EEG, 0.5–30Hz; EMG, 20–200Hz) and then digitized at a sampling rate of 128 Hz and recorded with SleepSign (Kissei Comtec, Nagano, Japan). After the experiment was completed, the EEG/EMG data were automatically classified off-line using 4 s epochs for wakefulness, REM sleep, and NREM sleep using SleepSign software according to standard criteria. These automatically defined classifications were checked manually and corrected if necessary.

EEG power spectra were calculated through fast Fourier transform at the frequency range 0–25 Hz, with a resolution of 0.25 Hz (Qu et al., 2010), and relative power bands were summed as: delta, 0.5–4 Hz; theta, 6–10 Hz; alpha, 12–14 Hz; and beta, 15–25 Hz. Every 4 s epochs of EEG power spectra were calculated through FFT. State-dependent spectral power was averaged by pick corresponding state epoch in a state-dependent manner. The power spectra was normalized by calculation of the percentage of each 0.25 Hz bin from the total power of each individual animal. The power of each 0.25 Hz bin was first averaged for each specific stage (NREM sleep, REM sleep, Wake) individually, and then normalized as a group by calculation of the percentage of each bin from the total power (0–24.75 Hz) of the individual animal.

# Pharmacological Treatments

Dexmedetomidine was dissolved in sterile saline immediately before use and implemented by intragastrical administration (i.g.) in a volume of approximately 10 ml/kg at doses of 25, 50, 100, and 200 µg/kg. To study the sedative effect of dexmedetomidine in mice, spontaneous locomotor activity was tested consecutively for 24 h from 19:00. The mice received saline or dexmedetomidine at 21:00.

To estimate possible drowsiness or hypnotic effects resulting from dexmedetomidine in the dark phase for mice, EEG/EMG was tested consecutively for 48 h from 19:00. On the first day of the experiment, all of the groups of mice received saline at 21:00 (in the early phase of the dark period), and the recordings made on that day served as self-controls. On the second day, the same mice were administered with (i.g.) saline or dexmedetomidine (25, 50, 100, and 200 µg/kg,) at the same time, and the recordings made on the second day served as the experimental data.

To evaluate the effects of oral delivery dexmedetomidine on sleep-wake regulation in mice during the light phase (sleep stage for mice), EEG/EMG was tested consecutively for 48 h from 07:00. At the same time, we compared the route of i.g. with intraperitoneal injection (i.p.) dexmedetomidine, and diazepam was the positive control drug. On the first day of the experiment, mice received saline at 09:00 (in the early phase of the light period), and the recordings made on that day served as self-controls. On the second day, the same mice were administered with dexmedetomidine (100 µg/kg i.g. or i.p.) or diazepam (6 mg/kg, i.g.) at the same time, and the recordings made on the second day served as the experimental data.

### c-Fos Immunochemistry and Cell Counting

In order to evaluate the effects of dexmedetomidine on c-Fos expression in the cerebral cortex and sleep–wake control pathway, animals were divided into two groups. Each group was given either saline or dexmedetomidine 100 µg/kg i.g. at 21:00, and animals were then sacrificed after 120 min for immunohistochemical experiments as described previously (Qiu et al., 2014). The mice were anesthetized using 5% chloral hydrate (500 mg/kg) and perfused with saline solution followed by 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer (PB, pH 7.0) through the heart. The brains were immediately removed and post-fixed in 4% PFA in 0.1 M PB (pH 7.4) for 4 h. The brains were then transferred to 20% sucrose in phosphate-buffered saline (PBS) and kept in the solution until they sank to the bottom. Frozen sections were cut at 30 µm in coronal planes by using freezing microtome (Jung Histocut, model 820-II, Leica, Nussloch, Germany). Sections were washed in 0.01 M PBS and then incubation with c-Fos antibody. The antibody was diluted 1:10000 in antiserum solution 2 (1% normal BSA, 0.2% Triton X-100, and 0.4% sodium azide in 0.01 M PBS at pH 7.2) at room temperature overnight. On the second day, the sections were incubated with a 1:1000 dilution of biotinylated

goat anti-rabbit secondary antibodies for 2 h followed by a 1:1000 dilution of avidin–biotin–peroxidase for 1 h at 37◦C. The peroxidase reaction was visualized with 0.05% DAB in 0.1 M phosphate buffer and 0.01% H2O2. Sections were mounted, dehydrated, and cover slipped. The sections were then examined under bright-field illumination using a Leica DMLB2 microscope (Leica Microsystems, Wetzlar, Germany). Images were captured by a Cool SNAP-Proof digital camera (SPOT RTKE Diagnostic Instruments, Sterling Heights, MI, United States). For the cerebral cortex, the 200 µm × 200 µm counting box was placed in the center of cortex of three adjacent sections, and only black-stained large neurons (likely pyramidal neurons) counted as c-Fos positive neurons. For the ventrolateral preoptic nucleus (VLPO), lateral hypothalamus (LH), tuberomammillary nucleus (TMN), laterodorsal tegmental nucleus (LDT), medial parabrachial nucleus (MPB), lateral parabrachial nucleus (LPB), and LC, we counted all of the c-Fos positive neurons in the entire region of three adjacent sections. The c-Fos counts were represented by average counts per section and per side.

### Statistical Analysis

All data subjected to statistical analysis in SPSS 19.0. All of the results are expressed as means ± SEM. For the time course data, the hourly amounts of LMA and each sleep–wake stage profiles in mice treated with saline or dexmedetomidine were compared using two-way ANOVA followed by Fisher's least-significant difference test. Histograms of the amounts of LMA, sleep, and wakefulness were assessed by one-way ANOVA followed by Bonferroni tests. Histograms of sleep latency were analyzed using the two-tailed paired t-tests, with each animal serving as its own control. Comparisons of sleep counts, as well as the number of sleep/wake events, duration, and transition, and the number of c-Fos immunoreactivity neurons were performed using unpaired, two-tailed Student's t-tests. Graphs of the power density of NREM sleep and the quantitative changes in power for the delta (0.5–4.0 Hz) frequency bands were analyzed using the two-tailed paired Student's t-test. Graphs of the distribution frequency of delta (0.5–4 Hz) density of NREM sleep were assessed by one-way ANOVA followed by Bonferroni tests. In all cases, p < 0.05 was considered to be statistically significant.

### RESULTS

### Dexmedetomidine Reduced Spontaneous Locomotor Activity in Mice During the Dark Phase

To investigate the sedative effects of orally delivered dexmedetomidine, dexmedetomidine at a dose of 25, 50, 100, or 200 µg/kg was administered (i.g.) at 21:00, and LMA was recorded continuously for 24 h from 19:00. The circadian rhythm of mice means that they spend most of their time sleeping and with less LMA during the light phase. So it is more difficult to assess the effects of drugs on LMA in the light phase than in the dark phase. Therefore, the experiments were performed during the dark phase when animals were active.

As shown in **Figures 1B–D**, time course changes revealed that dexmedetomidine at 50, 100, and 200 µg/kg significantly decreased LMA in mice during the night phase, which lasted for 2 (F1,<sup>158</sup> = 17.45, P < 0.01), 4 (F1,<sup>158</sup> = 8.81, P < 0.01), and 8 (F1,<sup>158</sup> = 111.31, P < 0.01) hours, respectively. However, dexmedetomidine at 25 µg/kg did not differ significantly from the control group with respect to the LMA cycle (**Figure 1A**). When dexmedetomidine was increased to 50 µg/kg, LMA was reduced by 87% (P < 0.01) and 84% (P < 0.01) compared with control group during the second and third hour after administration (**Figure 1B**). Dexmedetomidine at 100 µg/kg significantly decreased LMA during the first, second, third, and fourth hours by 68, 92, 91, and 73%, respectively compared with the control group. However, LMA rebounded significantly during the first hour of the light period, increasing 1.7-fold (P < 0.01) compared with the control group. Despite this, there was no further disruption of LMA architecture during the subsequent period (**Figure 1C**). In addition, dexmedetomidine at the highest dose of 200 µg/kg decreased LMA for 8 consecutive hours, with significant differences found at the first, second, third, fourth, sixth, and eighth hour and rebounded significantly at the second hour of the light period (**Figure 1D**).

The total counts of LMA during the 4 h following administration of dexmedetomidine are summarized in **Figure 1E**. Dexmedetomidine at 50, 100, and 200 µg/kg decreased the total counts of LMA by 67% (P < 0.01), 77% (P < 0.01), and 82% (P < 0.01), respectively, during the 4-h period, compared with the control group. However, dexmedetomidine at 25 µg/kg did not affect the cumulative amount of LMA when measured for 4 h after administration (**Figure 1E**). These results clearly indicate that dexmedetomidine decreases LMA in a dose-dependent manner.

### Dexmedetomidine Increased NREM Sleep and Decreased Wakefulness in Mice During the Dark Phase

To investigate the hypnotic effects of oral delivery dexmedetomidine during the dark phase, EEG/EMG were recorded for 2 consecutive days. Typical examples of a compressed spectral array (0–25 Hz) EEG, polygraphic recording, and corresponding hypnograms from a mouse given saline or 100 µg/kg dexmedetomidine are shown in **Figure 2A**. Mice treated with dexmedetomidine (100 µg/kg) quickly went to the sleep state during which EMG disappeared and they spent more time in NREM sleep compared with their own control.

As shown in **Figure 2B**, time course changes revealed that dexmedetomidine at 100 µg/kg significantly increased NREM sleep (F1,<sup>118</sup> = 11.24, P < 0.01) for 4 h following administration, which is consistent with a reduction in wakefulness (F1,<sup>118</sup> = 6.05, P < 0.05) during the same period compared with their own control. Dexmedetomidine at 100 µg/kg increased the hourly NREM sleep time by 2.7- (P < 0.01), 3.5- (P < 0.01), 1.7- (P < 0.05), and 4.8- (P < 0.01) fold relative to saline control during the first, second, third, and fourth hour after administration, respectively. The duration of wakefulness was

and∗∗P < 0.01 indicate significant differences compared with the saline control as assessed by two-way repeated measures ANOVA followed by Bonferroni testing. (E) Total counts of LMA during the 4-h period following saline or dexmedetomidine (25, 50, 100, and 200 µg/kg, i.g.). Values are means ± SEM (n = 8). ∗∗P < 0.01 indicates a significant difference compared with the control group, ##P < 0.01 indicates a significant difference between the different doses of dexmedetomidine as assessed by one-way ANOVA followed by Bonferroni tests.

decreased at the same time by 69% (P < 0.01), 80% (P < 0.01), 47% (P < 0.05), and 39% (P < 0.01) during the first, second, third and fourth hour after administration, respectively. However, wakefulness rebounded on the eighth hour after administration, which is consistent with a reduction in NREM sleep during the same time. Although time course changes on REM sleep failed to show a significant decrease during the dark phase, REM sleep rebounded 3.4-fold relative to saline control during the 16th hour after administration (P < 0.05). There was no further disruption of sleep architecture.

The total time spent in NREM sleep, REM sleep, and wakefulness were measured for 4 h after dexmedetomidine administration because the time course data revealed that 100 µg/kg dexmedetomidine increased NREM sleep for this duration. Dexmedetomidine dose-dependently increased NREM sleep (F4,<sup>28</sup> = 74.22, P < 0.01) and reduced wakefulness

FIGURE 2 | Effects of dexmedetomidine on sleep-wake profiles in mice during the dark phase. (A) Typical examples of compressed spectral array (0–25 Hz) EEG, EMG, and hypnograms over 8 h (19:00–03:00) following saline (upper panel) or dexmedetomidine 100 µg/kg (lower panel) administration. (B) Time course changes in NREM sleep, REM sleep, and wakefulness following saline (open circle) and dexmedetomidine (100 µg/kg, i.g.; closed red circle) administration during the dark phase. Each circle represents the hourly mean amount of each stage. The horizontal filled and open bars on the X-axes indicate the 12-h dark and 12-h light periods, respectively. Values are means ± SEM (n = 6); <sup>∗</sup>P < 0.05 and ∗∗P < 0.01 indicate significant differences compared with their own control as assessed by two-way repeated measures ANOVA followed by Bonferroni tests. (C) Dose-response effects on total time spent in NREM sleep, REM sleep, and wakefulness for 4 h after administration of saline and dexmedetomidine in mice. Open and red filled bars show the profiles of saline and dexmedetomidine treatment, respectively. Values are the means ± SEM (n = 6). <sup>∗</sup>P < 0.05 and ∗∗P < 0.01 indicate significant differences compared with their own control as assessed by two-tailed paired Student's t-test. #P < 0.05 and ##P < 0.01 indicate significant differences between saline and different doses of dexmedetomidine as assessed by one-way ANOVA followed by Bonferroni tests.

(F4,<sup>28</sup> = 72.88, P < 0.01) (**Figure 2C**). Dexmedetomidine at 25, 50, 100, and 200 µg/kg increased the total duration of NREM sleep 1.9-, 2.3-, 2.7-, and 3.9-fold, respectively, which was consistent with the reduction in wakefulness by 21, 39, 58, and 78%, respectively, compared with their own control in each group. Due to the small amount of REM sleep in mice during the early dark phase, there was no significant difference in REM sleep within 4 h after dexmedetomidine administration at any dose. Therefore, it reveals that orally delivered dexmedetomidine increases NREM sleep and decreases wakefulness in a dose-dependent manner.

### Dexmedetomidine Shortened Sleep Latency, Altered Sleep–Wake Architecture and EEG Power Density During the Dark Phase in Mice

To assess the initiation of the sleep state after treatment, we measured the latencies to NREM sleep, which were defined as the time from saline or dexmedetomidine treatment to the first appearance of a NREM sleep episode that lasted for at least 60 s. As shown in **Figure 3A**, dexmedetomidine i.g. remarkably shortened NREM sleep latency. The latencies to NREM sleep in mice treated with dexmedetomidine (50, 100, and 200 µg/kg, i.g.) were 13.7 (P < 0.05), 11.5 (P < 0.01), and 7.7 (P < 0.05) min, respectively, which were markedly shorter than the latencies of 31.8, 31.3, and 30.3 min after saline injection. However, 25 µg/kg dexmedetomidine failed to change the latencies to NREM sleep in mince. The short NREM sleep latency following dexmedetomidine (>25 µg/kg, i.g.) clearly indicates that dexmedetomidine accelerates the initiation of NREM sleep.

To better understand the changes in sleep-wake architecture caused by 100 µg/kg dexmedetomidine, we determined the number of episodes and mean duration of wakefulness, NREM sleep, and REM sleep, as well as transitions between the three vigilance stages after dexmedetomidine at a dose of 100 µg/kg. As shown in **Figure 3B**, dexmedetomidine at 100 µg/kg increased the total number of episodes of NREM sleep 1.6-fold (P < 0.05), but there was no significant difference in the number of episodes of wakefulness and REM sleep. In addition, the mean duration of NREM sleep increased by 81% with a concomitant 64% decrease in wakefulness (**Figure 3C**, P < 0.05). As shown in **Figure 3D**, dexmedetomidine (100 µg/kg) increased the number of state transitions from NREM sleep to wakefulness and wakefulness to NREM sleep (P < 0.05) during the 4 h following administration. Neither a change in the number of transitions from NREM sleep to REM sleep nor in that from REM sleep to wakefulness were observed. Distributions of bouts of different durations of NREM sleep and wakefulness are shown in **Figure 3E**. Dexmedetomidine (100 µg/kg) increased the number of bouts of NREM sleep with durations of 4–32, 128–256, and 512–2048 s. There was no difference in the number of bounds of wakefulness that were observed. These results suggest that dexmedetomidine increases the number of episodes and mean duration of NREM sleep, which extend the overall duration of NREM sleep.

The delta activity (0.5–4 Hz) during NREM sleep is not only a symbol of NREM sleep, but it also reflects the depth of sleep (Anaclet et al., 2014). To better understand the depth of sleep caused by dexmedetomidine, we evaluated the EEG power spectra and compared the power densities of saline and 100 µg/kg dexmedetomidine in mice during NREM sleep. As shown in **Figure 3F**, the frequency ranges of 1.25 and 2.5–3.5 Hz were increased, with a decrease in the frequency ranges of 4.25–4.5 and 8–11.5 Hz following the administration of 100 µg/kg dexmedetomidine compared with their own control. Then the quantitative changes in power for delta (0.5–4.0 Hz) frequency bands during the 4-h period after saline and dexmedetomidine (100 µg/kg; i.g.) administration were measured. As shown in the insertion part of the diagram in **Figure 3F**, dexmedetomidine increased the quantitative delta power 1.14-fold (P < 0.01) compared with the self-controls. These results suggest that dexmedetomidine increased the duration of NREM sleep and also increased the depth of sleep.

### Dexmedetomidine Increased NREM Sleep, and Decreased REM Sleep and Wakefulness in Mice During the Light Phase

The light phase in mice is equivalent to the nighttime sleep stage in humans, so we assessed the effects of dexmedetomidine on sleep-wake profiles during the light period. On day 1, the mice were treated with saline at 09:00 in the early phase of the light period, and the recordings made on that day served as each animal's own control. The animals were then treated with dexmedetomidine (100 µg/kg, i.g. or i.p.) and diazepam (6 mg/kg, i.g.) 24 h later. Typical examples of compressed spectral array (0–25 Hz) EEG, polygraphic recording, and corresponding hypnograms from a mouse given saline or dexmedetomidine i.g. at a dose of 100 µg/kg are shown in **Figure 4A**. Mice treated with dexmedetomidine quickly went to into a sleep state with EMG disappearing and more continuous NREM sleep observed compared with each animal's own control.

As shown in **Figure 4B**, time course changes revealed that dexmedetomidine (100 µg/kg, i.g.) significantly increased NREM sleep (F1,<sup>118</sup> = 75, P < 0.01) and decreased wakefulness (F1,<sup>118</sup> = 21.48, P < 0.01) in mice compared with their own control. Dexmedetomidine increased the hourly NREM sleep time 1.66- (P < 0.01), 1.56- (P < 0.01), 1.29- (P < 0.05), 1.31- (P < 0.01), and 1.59- (P < 0.01) fold relative to saline control during the first, second, third, fourth, and sixth hours after administration, respectively. The enhancement of NREM sleep was concomitant with a decrease in wakefulness during the first, second, and sixth hours after the administration of dexmedetomidine. In addition, dexmedetomidine (100 µg/kg, i.g.) decreased REM sleep for 6 h from the second hour after administration (F1,<sup>118</sup> = 83.52, P < 0.01). The effects began within the first hour and lasted for 6 h. Although the route of administration is different, the effect of intraperitoneal administration (i.p.) is almost the same as that of oral delivery. Dexmedetomidine (100 µg/kg, i.p.) increased NREM sleep for 6 h, and this was significant during the first, second, third, fourth, and sixth hours, respectively, compared with their own control (F1,<sup>118</sup> = 54.79, P < 0.01). There was no further disruption of

FIGURE 3 | Changes in sleep latency, architecture, and EEG power density of NREM sleep produced by administration of dexmedetomidine. (A) Effect of different doses of dexmedetomidine on NREM sleep latency. Values are mean ± SEM (n = 6). <sup>∗</sup>P < 0.05 and ∗∗P < 0.01 indicate significant differences assessed by two-tailed paired Student's t-test. (B) Total episode number, (C) mean duration, (D) stage transition, and (E) number of NREM sleep and wakefulness bouts during the first 4 h following administration of dexmedetomidine 100 µg/kg. Values are mean ± SEM (n = 6). <sup>∗</sup>P < 0.05 and ∗∗P < 0.01 indicate significant differences when using two-tailed unpaired Student's t-test. (F) EEG power density curves of NREM sleep and quantitative changes in power for delta (0.5–4.0 Hz) frequency bands (insert) during the 4 h period after saline and dexmedetomidine (100 µg/kg; i.g.) administration. Red horizontal bars indicate location of a statistically significant difference (P < 0.05, two-tailed paired t-test). Y-axes (insert) indicate the percentage of delta frequency on the EEG power density of NREM sleep. Data (quantitative of delta frequency) were standardized and expressed as the percentage of the mean delta power of NREM sleep. Values are mean ± SEM (n = 6). ∗∗P < 0.01 indicates significant differences compared with their own control as assessed by two-tailed paired Student's t-test.

sleep architecture during the subsequent period with the two different routes of administration of dexmedetomidine. However, high dose of the classical hypnotic drug diazepam at 6 mg/kg oral delivery only increased NREM sleep by 2 h compared to its own control (F1,<sup>118</sup> = 17.68, P < 0.01). There were no time course changes during the dark phase.

To better understand the total time spent in NREM sleep, REM sleep, and wakefulness, each stage was measured for 6 h after dexmedetomidine administration. As shown in **Figure 4C**, dexmedetomidine (100 µg/kg, i.g.) increased the total duration of NREM sleep 1.4-fold (P < 0.01) than saline control, which was consistent with a reduction in wakefulness by 56% (P < 0.01) and REM sleep by 97% (P < 0.01), respectively. These results suggest that dexmedetomidine increases NREM sleep partially by reducing REM sleep during the light phase. In addition, dexmedetomidine (100 µg/kg, i.p.) and diazepam (6 mg/kg, i.g.) increased the total duration of NREM sleep 1.51-fold (P < 0.01) and 1.23-fold (P < 0.05) than its own control, respectively. There was no significant difference when the total time spent in NREM sleep following dexmedetomidine i.g. with

FIGURE 4 | Sleep–wake profiles produced by administration of dexmedetomidine in mice during the light phase. (A) Typical examples of compressed spectral array (0–25 Hz) EEG, EMG, and hypnograms over 8 h (07:00–15:00) following saline (upper panel) and dexmedetomidine 100 µg/kg (lower panel) administration. (B) Time course changes in NREM sleep, REM sleep, and wakefulness following saline (open circle), dexmedetomidine i.g. (100 µg/kg; closed red circle), dexmedetomidine i.p. (100 µg/kg; closed rose red circle), and diazepam i.g. (6 mg/kg; closed green circle) administration during the light phase. The horizontal open and filled bars on the X-axes indicate the 12-h light and 12-h dark periods, respectively. Each circle represents the hourly mean ± SEM of NREM sleep, REM sleep, and wakefulness (n = 6). <sup>∗</sup>P < 0.05 and ∗∗P < 0.01 indicate significant differences compared with their own control as assessed by two-way repeated measures ANOVA followed by Bonferroni tests. (C) Total time spent in NREM sleep, REM sleep, and wakefulness for 6 h after administration. Open, red, rose red, and green filled bars show the profiles of saline, dexmedetomidine (i.g. or i.p.), and diazepam i.g. treatments, respectively. Values are mean ± SEM (n = 6). <sup>∗</sup>P < 0.05 and ∗∗P < 0.01 indicate significant differences compared with their own control as assessed by two-tailed paired Student's t-test. ##P < 0.01 indicate significant differences compared dexmedetomidine i.g. with dexmedetomidine i.p. and diazepam i.g. as assessed by one-way ANOVA followed by Bonferroni tests.

i.p. was compared. However, both routes of dexmedetomidine administration increased the total amount of NREM sleep significantly more than diazepam, suggesting a strong hypnotic effect of dexmedetomidine.

### Dexmedetomidine Shortened Sleep Latency, Consolidated Sleep Structure, and Increased EEG Power Density of NREM Sleep

As shown in **Figure 5A**, dexmedetomidine i.g. remarkably shortened NREM sleep latency during the light phase. The latencies to NREM sleep in mice treated with dexmedetomidine (100 µg/kg) were 13 min, thus markedly shorter than the latency of 26 min after saline administration (P < 0.05). As shown in **Figure 5B**, the total number of episodes of wakefulness, REM sleep, and NREM sleep decreased by 39% (P < 0.01), 95% (P < 0.01), and 44% (P < 0.01), respectively, within 6 h of administration. However, the mean duration of NREM sleep increased 2.7-fold (P < 0.01) with a concomitant 90% decrease in REM sleep (P < 0.01) (**Figure 5C**). As mice were asleep during the light phase, there was no difference in the mean duration of wakefulness (**Figure 5C**). As shown in **Figure 5D**, dexmedetomidine (100 µg/kg) decreased the number of state transitions from NREM sleep to wakefulness (P > 0.05), wakefulness to NREM sleep (P < 0.01), NREM sleep to REM sleep (P < 0.01), and REM sleep to wakefulness (P < 0.01). These findings indicate that dexmedetomidine reduced NREM sleep fragmentation and improved the continuity of NREM sleep by decreasing the transition between each stage.

To better understand the changes in sleep architecture caused by dexmedetomidine (100 µg/kg) during the light phase, the distribution of bouts of each stage was determined as a function of duration of the bout (**Figure 5E**). Dexmedetomidine (100 µg/kg) decreased the number of bouts of NREM sleep that had durations of 4–16 (P < 0.05), 32–64 (P < 0.05), 64–128 (P < 0.05), and 256–512 s (P < 0.05), but increased the number of bouts of NREM sleep that had durations of 1024—2024 (P < 0.01) and >2024 s (P < 0.01). Simultaneously, the number of bouts of REM sleep that had durations of 16–32 (P < 0.01), 32–64 (P < 0.01), 64–128 (P < 0.01), and 128–256 s (P < 0.01) were significantly decreased. However, only the number of bouts of wakefulness of durations of 4–16 s decreased (P < 0.01). These results suggest that the increased total time in NREM sleep with dexmedetomidine (100 µg/kg, i.g.) during the light phase was based on the reduction of the quantity of short-term NREM sleep and increase in long-term NREM sleep volume. At the same time, the frequency of awakening and short-term wakefulness, especially for 4–16 s, during the sleep stage was reduced. These occurred along with REM sleep reduction.

Changes in the depth of sleep during the light phase were examined by statically comparing the EEG power density of NREM sleep between dexmedetomidine treatment and saline control. As shown in **Figure 5F**, frequency ranges of 1.75–3.25 Hz increased and frequency ranges of 7.25–11.75 Hz decreased between dexmedetomidine treatment and their own control. As shown in the insertion part of the diagram in **Figure 5F**, dexmedetomidine increased the quantitative delta power 1.13-fold (P < 0.01) compared with own control. To clarify the effects of the hypnotic diazepam, the delta power of NREM sleep, EEG power spectra, and power densities during NREM sleep for 6 h after oral delivery 6 mg/kg diazepam in mice were compared with their own control. As shown in **Figure 5G**, the frequency ranges of 0.5 and 1–4.5 Hz decreased while those of 8.75–24.75 Hz increased between diazepam treatment and their own control. Quantitative delta power also decreased by 18% (P < 0.01) compared with their own control.

To better understand the distribution changes of wave frequency of delta (0.5–4 Hz) density of NREM sleep, the distribution of percentage of delta frequency in NREM sleep over self-control during the 6 h after saline, dexmedetomidine 100 µg/kg, and diazepam 6 mg/kg were analyzed. As shown in **Figure 5H**, there was no change in the distribution of delta frequency following saline administration. The frequency ranges of 1.75–3.25 Hz (P < 0.01) increased, while ranges of 0.5–1.0 Hz (P < 0.01) decreased dexmedetomidine was compared with the saline group. However, diazepam was compared with the saline group, and the frequency ranges of 1.25–4.0 Hz were decreased (P < 0.01; P < 0.05) and only 0.5 Hz increased (P < 0.01). The increased delta density of dexmedetomidine was mainly concentrated in the frequency ranges of 1.75–3.25 Hz, while diazepam reduced most of the delta density spectrum, particularly at the frequency ranges of 1.25–4.0 Hz. In addition, the most common adverse effect of classic hypnotics is residual 'hangover' effects, such as drowsiness and impaired psychomotor and cognitive function, and it may persist into the whole day and interfere with work following nighttime administration (Vermeeren, 2004). To clarify the effects of dexmedetomidine on the spectral power of wake during the dark phase following administration, EEG power spectra, and power densities of wake for the first 6 h and the whole 12 h of dark phase were compared with their own control. The results indicate that the administration of dexmedetomidine at 2 h into the light cycle does not affect the wake of mice during the dark phase (**Supplementary Figure S1**). These results imply that dexmedetomidine does not seem to affect the quality of wake during the dark phase.

### Effects of Dexmedetomidine on c-Fos Expression in the Cerebral Cortex and Sleep–Wake Control Pathway

The number of c-Fos positive neurons in the cortex and sleep-wake control pathway of the brain was counted to investigate the effects of dexmedetomidine. C-Fos protein as a marker for neuronal activity shows changes in different brain regions during spontaneous sleep-wake episodes (Dentico et al., 2009). Here, we administered dexmedetomidine 100 µg/kg at 21:00, at which time mice are spontaneously active and awake. The control group was administered an equal volume of sterile saline and the whole process was operated under 10 lx red light. Animals were then put back into their own cage and perfused after 120 min.

FIGURE 5 | Changes in sleep latency, architecture, and EEG power density of NREM sleep produced by administration of dexmedetomidine during the light phase. (A) Effect of dexmedetomidine 100 µg/kg on NREM sleep latency. Values are mean ± SEM (n = 6). <sup>∗</sup>P < 0.05 indicates significant differences assessed by two-tailed paired Student's t-test. (B) Total episode number, (C) mean duration, (D) stage transition, and (E) number of NREM, REM, wakefulness bouts during the first 6 h following administration of saline or dexmedetomidine 100 µg/kg. Values are mean ± SEM (n = 6). <sup>∗</sup>P < 0.05 and ∗∗P < 0.01 indicate significant differences performed using two-tailed unpaired Student's t-test. (F,G) EEG power density curves during NREM sleep and quantitative changes in power for delta (0.5–4.0 Hz) frequency bands (insert) during the 6-h period after saline, dexmedetomidine 100 µg/kg, or diazepam 6 mg/kg administrations. Red or green horizontal bars indicate the location of a statistically significant difference (∗P < 0.05, two-tailed paired t-test). Black, red, and green open circles in the inserted scatter plot represent saline control, administration of dexmedetomidine, and diazepam, respectively. Y-axes (insert) indicate the percentage of delta frequency on the EEG power density of NREM sleep. Data were standardized and expressed as the percentage of the mean delta power of NREM sleep. Values are mean ± SEM (n = 5, 6). ∗∗P < 0.01 indicates significant differences compared with their own control as assessed by two-tailed paired Student's t-test. (H) The distribution of percentage of delta frequency in NREM sleep over baseline during the 6-h period after saline, dexmedetomidine 100 µg/kg, or diazepam 6 mg/kg. Open and closed red and green circles indicate the profiles of saline, dexmedetomidine, and diazepam treatment, respectively. The value of power spectrum after the first saline administration was defined as baseline. Data were standardized and expressed as percentages of baseline. Values are mean ± SEM (n = 5, 6). <sup>∗</sup>P < 0.05 and ∗∗P < 0.01 indicate significant differences performed by one-way ANOVA followed by Bonferroni tests.

Previous research has found that c-Fos expression is significantly higher in the cerebral cortex during wakefulness compared to sleep, especially in the prefrontal and frontal, motor, and sensory cortices (Qiu et al., 2014). To determine the effects of dexmedetomidine on c-Fos expression in the cerebral cortex, we selected the motor cortex for comparison, because dexmedetomidine clearly decreases LMA. As shown in **Figures 6A,B,a,b**, the representative photomicrographs of c-Fos expression in the motor cortices of saline and dexmedetomidine clearly indicate that c-Fos expression in the cortex was very low compared to saline control. Analysis of the number of c-Fos immunoreactive nuclei showed that dexmedetomidine 100 µg/kg decreased the number in the motor cortex by 71% (P < 0.01; **Figure 6C**). These results indicate that dexmedetomidine inhibits the activity of the cerebral cortex, which is consistent with the observations of decreased LMA and increased NREN sleep induced by dexmedetomidine during the dark phase.

**Figure 7** shows representative photomicrographs of c-Fos expression in different brain regions of saline and dexmedetomidine treated mice. As shown in **Figures 7A,B,a,b,K**, compared with the control group, the administration of dexmedetomidine significantly increased c-Fos expression in the VLPO (P < 0.05). Thus, dexmedetomidine exerted hypnotic effects partially by exciting the sleep-promoting nuclear VLPO. However, in contrast with sub-cortical arousal regions such as the LH (P < 0.01; **Figures 7C,D**), TMN (P < 0.01; **Figures 7E,F,e,f**), LDT (P < 0.01; **Figures 7G,H,g1,h1**), LPB and MPB (P < 0.05; **Figures 7G,H,g2,h2**), and LC (P < 0.05; **Figures 7I,J,i,j**), c-Fos expression was significantly lower after the administration of dexmedetomidine compared with the control group (**Figure 7K**). These subcortical nucleuses are not only wake-initiating but also wake-maintaining components. Results are consistent with the observation that dexmedetomidine significantly decreased wakefulness and increased NREM sleep during the dark phase.

# DISCUSSION

Before dexmedetomidine became readily available around the world, propofol and benzodiazepines were the most frequently used agents for continuous sedation and clinical surgery anesthesia. Since dexmedetomidine was approved by the U.S. Food and Drug Administration in 1999, it has been widely used due to its lack of suppression of respiratory function, protective effect on the nervous system, anesthetic-sparing activity, and attenuation of immunosuppression (Mantz et al., 2011). Both propofol and GABA-receptor agonists have been used to treat insomnia, but they have not been popularized due to respiratory depression, drug resistance, and effect of decreasing sleep depth (NIH, 2004). Whether or not dexmedetomidine can be a drug for treating sleep disorders remains poorly specified. Decreasing LMA is usually considered to reflect sedative effects. In the present study, to verify the sedative effects of orally delivered dexmedetomidine, we initially evaluated the drug by LMA. We found that orally delivered dexmedetomidine dose-dependently decreased LMA during the dark phase, and with a rebound increased in LMA at the later stage. This is consistent with the results of intraperitoneal injection (Seidel et al., 1995).

However, changes in LMA do not directly confirm that dexmedetomidine is increasing sleep. By analyzing EEG/EMG data, we found that the oral delivery of dexmedetomidine dose-dependently increased NREM sleep by improving the mean duration and number of episodes of NREM sleep during the dark phase. Shortening of sleep latency is one of the indicators used to evaluate hypnotic drugs. In the present study, dexmedetomidine reduced the latency to NREM sleep, consistent with the behavior of other hypnotics (Huedo-Medina et al., 2012; Gerashchenko et al., 2017). Fragmented sleep, like short duration sleep bouts and frequent sleep-wake state transitions during sleep are associated with a variety of disorders, including psychiatric disease, memory impairment, and deposition of toxic proteins

in cerebrospinal fluid (Wulff et al., 2010; Krause et al., 2017). Our study demonstrates that dexmedetomidine can reduce the quantity of short-term NREM sleep and increase long-term NREM sleep volume through the elimination of the bouts of awakenings during the light phase. This is similar to principal hypnotics (Xu et al., 2014; Manconi et al., 2017; de Zambotti et al., 2018). However, it is worth noting that in healthy mice, sleep fragmentation is not detrimental, which can help mice to raise vigilance and minimize risk of predation during sleep. In present study, we found that dexmedetomidine reduced sleep fragmentation in healthy mice during the light phase, which seems to imply that dexmedetomidine may provide new

treatment for sleep-disrupted sleep disorders. To validate this conjecture, researches on animal models of sleep disorders must be conducted.

In addition, these classic hypnotics always decrease depth of sleep, especially the delta activity within NREM sleep (Huedo-Medina et al., 2012; Xu et al., 2014; Gerashchenko et al., 2017). However, dexmedetomidine-induced sleep with increased the delta spectrum and promoted sleep depth. Sleep depth is essential for the brain function maintenance such as eliminating metabolic wastes, promoting learning and memory, cognitive function, and recovery from physical impairments (Xie et al., 2013). And slow wave sleep (deep NREM sleep) disruption increases cerebrospinal fluid amyloid-β levels, an early and necessary step in Alzheimer's disease pathogenesis (Ju et al., 2017). From this point of view, although dexmedetomidine-induced sleep is different from physiological sleep (increase the delta spectrum), it does not reduce sleep depth compared with traditional hypnotics. In addition, study have found that dexmedetomidine-induced sleep is very similar to sleep rebound after sleep deprivation, both with an increase in delta spectrum (Zhang et al., 2015). In humans, consistent results have found that intravenous injection of dexmedetomidine can promote biomimetic N3 sleep (slow wave sleep) (Akeju et al., 2018).

In this study, dexmedetomidine induced an increase in the high delta frequency range (1.75–3.25 Hz) while a significant decrease in the low delta frequency range (0.5-1 Hz). Although, the delta wave is mainly composed of a wave frequency of 0.5–4 Hz, there is no literature reports the functions of different frequency ranges. In addition, the increase in delta spectrum of NREM sleep during the recovery sleep after sleep deprivation has long been recognized (Zhang et al., 2015), but the effects of sleep deprivation on the distribution of delta wave frequency during the recovery sleep still waiting to study. At present, the understanding of the distribution of delta wave frequency during NREM sleep is also limited. If later studies can find the function of different delta wave frequencies, it will help to further explain the regulation of dexmedetomidine on NREM sleep.

Other research has demonstrated that dexmedetomidine (0.3 mg/kg i.p.) causes sedation and is not a substitute for normal physiologic sleep, with a rebound in both NREM sleep and REM sleep in rats (Garrity et al., 2015). However, whether the animal is in a sleep or anesthetized stage must be distinguished. Intravenous dexmedetomidine at 50 µg/kg or subcutaneous dexmedetomidine at 150 µg/kg can effectively cause loss of the righting reflex (LORR) in rats (Nelson et al., 2003; Hu et al., 2012), and higher than sedative concentrations (400 µg/kg i.p.) were available to cause LORR in the C57BL/6 mice (Gelegen et al., 2014; Zhang et al., 2015). Although the clinical practice guidelines from the American Society of Anesthesiologists classification never refers to sedation or anesthesia as sleep, high doses of dexmedetomidine intraperitoneal injection can put animals in a deeper state than sedation or sleep. The doses given in the present experiment (100 µg/kg) and the route of administration (i.g.) were not sufficient to introduce anesthesia in the animals (**Supplementary Figure S2**). In addition, clinical trials have also confirmed that low-dose dexmedetomidine can improve postoperative sleep quality (Wu et al., 2016; Lu et al., 2017). In this study, we also found that both intraperitoneal injection and orally delivered dexmedetomidine can both increase NREM sleep for 6 h during the light phase, suggesting that oral delivery dexmedetomidine could offer a good hypnotic effect.

Classic research studies have shown that the neural circuits involved in generating sleep contribute to the loss of wakefulness caused by anesthetic agents and hypnotics (Lu et al., 2008; Hudetz, 2012; Leung et al., 2014). Consistent with this hypothesis, data suggest that dexmedetomidine-induced sleep is caused, in part, by activating sleep-promoting nuclei and inhibiting wake-promoting nuclei (Nelson et al., 2003). It has been proposed that the VLPO may have a key role in sleep-related processes on the basis of the results of unit recordings (Suntsova et al., 2002, 2007), electrophysiological recordings (Gallopin et al., 2000), anatomical tracing (Uschakov et al., 2007; Chung et al., 2017), and c-Fos immunohistochemistry (Sherin et al., 1996; Dentico et al., 2009). In the present study, we also found c-Fos expression increased in the VLPO, and this is consistent with previous research (Nelson et al., 2003). Despite some studies supporting the argument that the LC is the locus of the hypnotic action of dexmedetomidine (Correa-Sales et al., 1992; Chiu et al., 1995), contradictory findings have shown that dexmedetomidine can still induce sedation in mice unable to synthesize NA (Gilsbach et al., 2009; Hu et al., 2012; Sanders and Maze, 2012) or with selective knockdown of α2<sup>A</sup> adrenergic receptors in the LC (Zhang et al., 2015). In addition, inhibition of LC neurons does not produce sustained sleep (Carter et al., 2010). These results appear to indicate that dexmedetomidine-induced sedative-hypnotic effects may also act on other wake-promoting regions.

Consistent with other research (Nelson et al., 2003; Luo and Leung, 2011), in the present study, dexmedetomidine also decreased c-Fos expression in the TMN. Silencing of histaminergic neurons during wakefulness promotes slow-wave sleep, but not REM sleep (Williams et al., 2014; Fujita et al., 2017); and this is similar to dexmedetomidine-induced NREM sleep. The LH is an important part of the forebrain that contains orexin neurons, melanin-concentrating hormone containing neurons, GABAergic, and glutamatergic neurons (Yamashita and Yamanaka, 2017). These different types of neurons have been identified to be wake-promoting and project heavily into the VLPO, and are thus well-placed to inhibit sleep-promoting neurons (Schone et al., 2014; Venner et al., 2016; Yamashita and Yamanaka, 2017). In the present study, c-Fos expression also decreased in the LH area after dexmedetomidine, and this is consistent with decreasing neuronal activity in this area during sleep (Yamashita and Yamanaka, 2017). According to the "flip-flop switch" hypothesis (Scammell et al., 2017), NREM sleep induced by dexmedetomidine may depend on the inhibition of TMN and LH by VLPO GABAergic neurons, which are excited by dexmedetomidine.

Inconsistent with our results, Garrity et al. (2015) found that dexmedetomidine did not cause significant changes in c-Fos expression in various regions of the hypothalamus after dexmedetomidine (0.1–0.5 mg/kg; i.p.) in rats. However, same results in mice study from Nelson, et al were found that dexmedetomidine (0.4 mg/kg; i.p.) induced a qualitatively similar pattern of c-Fos expression as our results with decrease in the

TMN, LH and an increase in the VLPO (Nelson et al., 2003). In addition, another experiment with mice as subject also found c-Fos expression changes in the VLPO, LPO, and MPO during dexmedetomidine-induced sedation (Zhang et al., 2015). As we all known, the c-Fos protein expresses at 30 min but reaches peak at around 90-120 min after stimulation (Zhong et al., 2014; Zhang et al., 2018). In the Garrity's paper, rats were decapitated 65 min after administration of dexmedetomidine, while mice were decapitated 2 h after administration in Nelson's work. In our study, we also selected 2 h as the time point. Taking together, the difference in these results may be caused by the different species of the experimental animals, different dose of dexmedetomidine and the time of decapitating of the animals after administration of dexmedetomidine.

In the present study, we explored the effect of dexmedetomidine on c-Fos expression in the parabrachial nucleus (PB). The PB sends glutamatergic projections to a variety of forebrain structures (basal forebrain, LH, and midline thalamus) and cerebral cortex to promote arousal (Kaur et al., 2013). C-Fos immunohistochemical data also reveal that PB neurons are active during passive emergence from isoflurane general anesthesia (Muindi et al., 2016). However, previous studies did not distinguish different regions of PB influence. It has been found that the LPB is necessary for arousal from sleep in response to CO2, while the MPB plays an important role in promoting spontaneous waking (Kaur et al., 2013). Therefore, we compared c-Fos expression levels in the LPB and MPB. Consistent with previous research, dexmedetomidine induced NREM sleep along with decrease extinction in the LPB and MPB.

Electrophysiological studies have found that the noradrenergic neurons in the LC are active during wakefulness, less active during NREM sleep, and quiet during REM sleep (Aston-Jones and Bloom, 1981). Based on the evidence that the locus coeruleus as the main site of action for the sedating effects of dexmedetomidine (Chiu et al., 1995), orally delivered dexmedetomidine might be supposed to cause an increase in REM sleep. In contrast to this prediction, our behavioral outcomes show that dexmedetomidine caused a long-lasting elimination of REM sleep during the light phase. However, c-Fos staining results confirm the truth of that dexmedetomidine inhibits the LC neurons activity with decreased c-Fos expression. Although, the current study does not explain the mechanisms by which oral delivered dexmedetomidine caused a significant decrease in REM sleep, lots of works have found that the LDT plays an important role in the induction and maintenance of REM sleep (Verret et al., 2005; Clement et al., 2011; Sakai, 2015; Cox et al., 2016). Cells in the region of the LDT discharge during waking, decrease firing during NREM sleep and increase firing during REM sleep (Boucetta et al., 2014). And selective optogenetic activation of cholinergic neurons in the LDT during NREM sleep could increase the probability of REM sleep and the number of REM sleep episodes but not the duration of REM sleep episodes (Van Dort et al., 2015). Immunohistochemically identified cholinergic neurons in the LDT express c-Fos in the highest numbers in association with REM sleep (Maloney et al., 1999). In the present study, c-Fos expression in the LDT was decreased, which is consistent with decreased REM sleep after dexmedetomidine. However, studies have found the balance between Ach- and NA-mediated neurotransmission in LDT or PPT may play an important role in the regulation of REM sleep (Jones, 2005). And when acetylcholinesterase inhibitors administered into LDT or PPT alone, the wakefulness were happened; but when administered following depletion of the catecholamine by previous treatment of reserpine, REM sleep would be occurred (Jones, 2004). This is consistent with the result of the LC nucleus stopping discharge during REN sleep. In addition, LDT neurons receive projections from LC, and the cholinergic neurons with α2A-adrenergic receptors represented approximately one-half of the LDT ChAT<sup>+</sup> neurons (Hou et al., 2002), indicating that part of cholinergic neurons in LDT are under an inhibitory influence of NA from the LC. Therefore, the occurrence of REM sleep needs to be based on the reduction excitability of the LC, and relative excitement of the LDT. Taking together, decreased in REM sleep by dexmedetomidine may take place through the inhibition not only the excitability of the LC, but also the REM-ON neuronal activity in the LDT by α2AR.

Contrary to dexmedetomidine, studies have found wake-promoting agents such as modafini, caffeine and amphetamine can increase c-Fos expression in the TMN and orexin neurons in the LH (Scammell et al., 2000; Deurveilher et al., 2006; Rotllant et al., 2010). Among them, caffeine and amphetamine can also increase c-Fos expression in the LC (Bennett and Semba, 1998; Rotllant et al., 2010), while, functional MRI found modafinil can improve the high-phasic activity of humans' locus coeruleus (Minzenberg et al., 2008). In addition, those drugs can all increase c-Fos expression in the ventral tegmental area (VTA), the dorsal raphe nucleus, and cerebral cortex. However, the present study showed that the above brain regions and cortex excited by the wake-promoting drugs were inhibited by dexmedetomidine. Equally, electrophysiological study found that the VLPO excited by dexmedetomidine was also inhibited by modafinil (Gallopin et al., 2004).

Clinical research and systematic reviews have shown that dexmedetomidine can improve postoperative sleep quality (Lu et al., 2017). Recently, a clinical pilot study found that dexmedetomidine promoted N3 sleep in a dose-dependent manner and did not impair performance on a psychomotor vigilance test on the next day (Akeju et al., 2018). However, the route of administration was via continuous intravenous pump and a detailed analysis of changes in the sleep architecture was not provided. In the present study, we also found that the oral delivery of dexmedetomidine can increase the amount of NREM sleep time, shorten sleep latency, stabilize NREM sleep structure, and increase the delta power of NREM sleep. These effects are not solely based solely on the suppression of the LC but also on the suppression of other wake-promoting nuclei, such as TMN, LH, PB, and LDT.

### CONCLUSION

In conclusion, our results indicate that the oral delivery of dexmedetomidine has sedative and hypnotic effects, and it dose-dependently promotes NREM sleep in mice. These effects may be attributed to the excitation of sleep-promoting nuclei and inhibition of wake-promoting nuclei.

### AUTHOR CONTRIBUTIONS

fphar-09-01196 December 3, 2018 Time: 11:7 # 16

Z-XF and HD designed and performed the experiments, analyzed the data, and wrote the paper. W-MQ and WZ designed the experiments, analyzed the data, and wrote the paper.

### FUNDING

This work was supported by the National Basic Research Program of China (Grant No. 81571082 to WZ)

### ACKNOWLEDGMENTS

We sincerely thank Professor Zhi-Li Huang of Fudan University for his kind comments. And we also would like to thank Mr. Ze Zhang, Ze-Ka Chen, Jian Ni, and all of the staff members at the Department of Pharmacology, School of Basic Medical Sciences of Fudan University.

### REFERENCES


### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar. 2018.01196/full#supplementary-material

FIGURE S1 | Changes in EEG power density of wake produced by administration of dexmedetomidine during the light phase. (A,B) EEG power density curves during wake and quantitative changes in power for delta (0.5–4.0 Hz) frequency bands (insert) during the 6-h period after saline, dexmedetomidine 100 µg/kg, or diazepam 6 mg/kg administrations. (C,D) EEG power density curves during wake and quantitative changes in power for delta (0.5–4.0 Hz) frequency bands (insert) during the 12-h period of dark phase after administrations. (E,F) EEG power density curves during wake and quantitative changes in power for delta (0.5–4.0 Hz) frequency bands (insert) during the 6-h period of dark phase after administrations. Red or green horizontal bars indicate the location of a statistically significant difference (∗P < 0.05, two-tailed paired t-test). Black, red, and green open circles in the inserted scatter plot represent saline control, administration of dexmedetomidine, and diazepam, respectively. Y-axes (insert) indicate the percentage of delta frequency on the EEG power density of NREM sleep. Data were standardized and expressed as the percentage of the mean delta power of wake. Values are mean ± SEM (n = 5, 6). ∗∗P < 0.01 indicates significant differences compared with their own control as assessed by two-tailed paired Student's t-test.

FIGURE S2 | External stimulation can wake up the mouse. Typical examples of raw EEG/EMG and following dexmedetomidine 100 µg/kg (upper panel) or dexmedetomidine 200 µg/kg (lower panel) administration in a mouse during the dark phase. (The initial screenshot of NREM sleep transition to wake stage from a mouse; n = 6).


Gallopin, T., Luppi, P. H., Rambert, F. A., Frydman, A., and Fort, P. (2004). Effect of the wake-promoting agent modafinil on sleep-promoting neurons from the ventrolateral preoptic nucleus: an in vitro pharmacologic study. Sleep 27, 19–25.

Garrity, A. G., Botta, S., Lazar, S. B., Swor, E., Vanini, G., Baghdoyan, H. A., et al. (2015). Dexmedetomidine-induced sedation does not mimic the

neurobehavioral phenotypes of sleep in Sprague Dawley rat. Sleep 38, 73–84. doi: 10.5665/sleep.4328


NIH (2004). What's wrong with prescribing hypnotics? Drug Ther. Bull. 42, 89–93.



autoregulatory mechanisms for controlling arousal. J. Neurosci. 34, 6023–6029. doi: 10.1523/JNEUROSCI.4838-13.2014


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Feng, Dong, Qu and Zhang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Evidence of P3a During Sleep, a Process Associated With Intrusions Into Consciousness in the Waking State

### Paniz Tavakoli<sup>1</sup> \*, Allyson Dale<sup>2</sup> , Addo Boafo1,3 and Kenneth Campbell<sup>2</sup>

<sup>1</sup> Children's Hospital of Eastern Ontario, Ottawa, ON, Canada, <sup>2</sup> School of Psychology, University of Ottawa, Ottawa, ON, Canada, <sup>3</sup> Department of Psychiatry, University of Ottawa, Ottawa, ON, Canada

The present study examines processes associated with intrusions into consciousness during an unconscious state, natural sleep. The definition of sleep is still much debated. Almost all researchers agree that sleep onset represents a gradual loss of consciousness of the external environment. For sleep to be beneficial, it needs to remain as undisturbed as possible. Nevertheless, unlike other unconsciousness states, sleep is reversible. For purposes of survival, it is critical that the sleeper be able to "detect" and perhaps become conscious of highly relevant biological or personal information. Therefore, even in sleep, the brain must decide whether a new incoming stimulus is relevant and if so, may require an arousal to wakefulness, or whether it is irrelevant and can be gated to prevent disruption of sleep. Event-related potentials (ERPs) were used to measure the extent processing of auditory stimuli some of which elicited an ERP component, the P3a, in the waking state. The P3a is associated with processes resulting in the interruption of frontal central executive, leading to conscious awareness. Very little research has focused on the occurrence of the P3a during sleep. A multi-feature paradigm was used to examine the processing of a frequently occurring "standard" stimulus and six rarely occurring different "deviant" stimuli during wakefulness, NREM, and REM sleep. A P3a was elicited by novel environmental sounds and white noise bursts in the waking state, replicating previous studies. Other deviant stimuli (changes in pitch, intensity, duration) failed to do so. The ERPs indicated that processing of the stimuli that did not elicit a P3a in wakefulness were much inhibited during both NREM and REM sleep. Surprisingly, those deviants that did elicit a P3a in wakefulness continued to do so in stage N2 and REM sleep. The subject did not, however, awaken. These results suggest processes leading to consciousness in wakefulness may still remain active during sleep possibly allowing subjects to act on potentially highly relevant input. This may also explain how sleep can be reversed if the stimulus input is sufficiently critical.

Keywords: conscious awareness, sleep, gating, event-related potentials, P3a, multi-feature paradigm

# INTRODUCTION

Natural sleep is a period of profound unconsciousness. Unlike other unconscious states, natural sleep is, however, rapidly reversible. For purposes of survival, the sleeper must have the ability to detect highly relevant external information, whether biological or personal, and if necessary, to awaken to a conscious state and take appropriate action. Nevertheless, for sleep to be beneficial,

### Edited by:

Sakiko Honjoh, University of Tsukuba, Japan

### Reviewed by:

Giulio Bernardi, University of Pisa, Italy Madoka Takahara, Fukushima University, Japan

> \*Correspondence: Paniz Tavakoli ptava069@uottawa.ca

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 03 October 2018 Accepted: 19 December 2018 Published: 10 January 2019

### Citation:

Tavakoli P, Dale A, Boafo A and Campbell K (2019) Evidence of P3a During Sleep, a Process Associated With Intrusions Into Consciousness in the Waking State. Front. Neurosci. 12:1028. doi: 10.3389/fnins.2018.01028

**84**

all but the most relevant external stimulation is inhibited. Therefore, even in sleep, the brain must decide whether incoming stimuli require further processing and possible arousal to wakefulness, or whether they can be inhibited (Campbell and Colrain, 2002). The present study examines processes associated with intrusions into consciousness during natural sleep.

Sleep is not a uniform state. It consists of a series of substages that can be divided into non-REM (NREM; stages N2 and N3) and REM. The processing of external stimulus input can be differentially affected during the various stages of sleep (for reviews, see Campbell and Colrain, 2002; Massimini et al., 2009; Scammell et al., 2017). A major problem with examining the extent of information processing during sleep is that the subject cannot signal awareness of the external stimuli. In the present study, event-related potentials (ERPs) were used to measure the extent of information processing in the brain in the absence of overt behavioral responses (for reviews, see Colrain and Campbell, 2007; Koch et al., 2016; Dykstra et al., 2017). ERPs consist of a series of negative or positive components thought to reflect different stages of information processing. Some of these components can be elicited independent of attention and consciousness while others are highly affected by whether the subject attends to the stimulus input and becomes conscious of it.

The ability to detect acoustic changes in the environment is particularly important. The detection of acoustic change and possible subsequent intrusions into consciousness is often studied using an auditory oddball paradigm, consisting of a frequently occurring "standard" stimulus and rarely occurring "deviants." The presentation of the standard stimulus elicits an obligatory negativity peaking at about 100 ms, N1, and a positivity peaking at about 200 ms, P2. The N1 and P2 are mainly associated with sensory processing of the auditory stimuli. They increase in amplitude to loud and infrequently presented auditory stimuli. The N1 gradually decreases in amplitude during sleep onset until it reaches near baseline levels during stage N2 sleep. On the other hand, P2 has been observed to increase in amplitude during sleep (Harsh et al., 1994; Cote et al., 2001; Campbell and Colrain, 2002; Crowley et al., 2002).

The deviant stimulus also elicits the N1 and P2, but in addition elicits a negative component, the mismatch negativity (MMN) occurring at about 100–200 ms after stimulus onset (Näätänen, 1990, 1992). It is maximum over fronto-central areas of the scalp inverting in polarity at the mastoids. The MMN is thought to reflect the automatic detection of acoustic change. Certain types of deviants will also elicit a larger N1 than the standard. For example, a deviant that represents an increase in intensity from the standard will elicit a larger N1 and MMN. This composite negativity following this deviant represents the spatial and temporal summation of the N1+MMN. As a result, this negativity is often called a deviant-related negativity (DRN). In this article, the negativity that is elicited by the deviants will be described as a DRN.

Results from studies examining the DRN during sleep have not always been consistent. Many studies have failed to observe a distinct DRN during NREM (Nielsen-Bohlman et al., 1991; Niiyama et al., 1994; Winter et al., 1995; Loewy et al., 1996; Nordby et al., 1996; Loewy et al., 2000; Nashida et al., 2000; Macdonald et al., 2008; Sculthorpe et al., 2009; Strauss et al., 2015) and REM sleep (Niiyama et al., 1994; Sallinen et al., 1996; Loewy et al., 2000; Strauss et al., 2015). On the other hand, some authors have reported that a DRN can be elicited by many different types of deviants during NREM (Sabri et al., 2000, 2003; Sabri and Campbell, 2005; Ruby et al., 2008) and REM sleep (Loewy et al., 1996; Atienza et al., 1997, 2000; Nashida et al., 2000; Atienza and Cantero, 2001; Cote et al., 2001; Sabri and Campbell, 2005; Macdonald et al., 2008; Ruby et al., 2008; Sculthorpe et al., 2009), although with reduced amplitudes compared to the waking state. It is possible that the nature or extent of the deviance change (frequency, intensity, duration) could account for these differences.

The output of the change detection system is claimed to vary directly with the extent of change. In the waking state, if the extent of change is large enough, an involuntary switch of attention from the demands of ongoing cognitive tasks and to the unattended auditory input may occur. This involuntary capture of attention or "intrusion into consciousness" is associated with a later positivity, P3a (Escera et al., 1998). The P3a occurs at about 200–300 ms following stimulus onset and is largest over centro-frontal areas of the scalp. There is much debate about the extent to which the P3a reflects actual consciousness of the potentially highly relevant auditory input. Some authors suggest that it reflects a precursory process occurring prior to the switch of cognitive resources to the auditory channel that may only then subsequently lead to conscious awareness (Wetzel et al., 2013; Parmentier, 2014). There is, however, general agreement that the P3a does reflect processes associated with the involuntary capture of attention. In the waking state, while any perceptible change in stimulation will elicit a DRN, only a small number will also elicit a P3a.

Few studies have examined the P3a during sleep. Most MMN/DRN studies did not report a P3a during sleep, possibly because the extent of deviance was not sufficiently large. Cote and Campbell (1999) observed a frontal positivity at about 250 ms to rarely occurring deviant stimuli that represented a large increase in intensity (90 dB SPL) from the standard (70 dB SPL) during REM sleep. The latency and scalp distribution of this positivity did correspond to that of the P3a. This positivity was followed by a second large centro-parietal positivity, the P3b, occurring at about 320 ms, similar to that which was recorded in wakefulness. Macdonald et al. (2008) also observed a P3a-like positivity during REM sleep, occurring at about 270 ms, following a 10 dB increase in intensity from the 80 dB SPL standard. They did not observe a P3a following a 20 dB decrease in intensity deviant during REM. The P3a can also be elicited by deviants that are not obtrusive. Ruby et al. (2008) employed a duration deviant and reported the presence of both a P3a-like positivity and a later P3b-like positivity occurring at about 200 and 300 ms, respectively, in both stage N2 and REM. Tavakoli et al. (2018) recorded the P3a during stage N1 and the first 30 min of stage N2 sleep. They observed a significant P3a-like positivity at about 240 ms during early stage N2 sleep following their environmental sound deviants. A P3a-like positivity was also observed following their white noise deviant during this stage at about 220 ms, although its amplitude did not attain significance.

Most studies employ oddball sequences. A problem with this paradigm is that only one or perhaps two deviants can be presented within the sequence. A reason for the somewhat contradictory findings during sleep may be the use of different deviants across studies. The oddball paradigm does not easily allow for the presentation of many types of deviants, limiting the extent to which the effect of deviants can be compared in a single study. On the other hand, a more recent multifeature paradigm (Näätänen et al., 2004), as its name implies, allows for the presentation of several deviants each representing a change of a different feature of the standard. The present study employs a multi-feature paradigm used by Tavakoli and Campbell (2016) and Tavakoli et al. (2018) to examine those deviants most likely to elicit a P3a during sleep. In the multifeature paradigm, six different deviant stimuli were presented in a single auditory sequence, consisting of an alternating standarddeviant pattern. The DRN and P3a that are elicited within a multi-feature paradigm are very similar to those elicited within the traditional oddball paradigm (Näätänen et al., 2004; Tavakoli and Campbell, 2016). Previous studies have indicated that a change in stimulus duration or an increase in stimulus intensity might elicit a P3a-like response during sleep (Macdonald et al., 2008; Ruby et al., 2008), although they did not do so when presented within a multi-feature paradigm in fully awake young adults (Tavakoli and Campbell, 2016). The determination of what stimulus features are deemed to be so potentially relevant to warrant an interrupt of executive functions may differ between the waking and sleep states. White noise and environmental sounds do elicit a large P3a in waking subjects but have not been presented as deviants in previous sleep studies. To control for the possibility that any stimulus change might elicit a P3a within sleep, deviants representing a change in frequency and a decrease in intensity were also embedded within the multifeature paradigm. The multi-feature paradigm was presented passively to subjects during the waking state while the subject's attention was engaged in watching a silent, subtitled movie, and subsequently within the different stages of all-night sleep.

### MATERIALS AND METHODS

### Subjects

Sixteen self-reported good sleepers (14 women) between the ages of 20–28 years (mean = 23.12 years, SD = 2.68 years) spent a single night in the sleep laboratory. All subjects reported normal sleep time to be between 23:00 and 24:00. None reported any history of hearing, neurological, or sleep disorders. Written informed consent was obtained prior to the start of the study and subjects received an honorarium for their participation. Subjects refrained from caffeine and alcohol in the 24 h prior to the start of the study. The study was conducted according to the Canadian Tri-Council guidelines (Medical, Natural, and Social Sciences) on ethical conduct involving human subjects.

### Physiological Recordings

EEG and electrooculography (EOG) activity were recorded using Grass gold-cup electrodes, filled with electrolytic paste, and affixed to the skin by surgical tape and to the scalp by gauze. Brain Products Recorder software and BrainAmp amplifiers were used for the EEG recording. The EEG was recorded from 18 electrodes across frontal, central, parietal, and occipital sites (FP1, FP2, FT9, FT10, F3, Fz, F4, FC3, FCz, FC4, C3, Cz, C4, P3, Pz, P4, O1, O2) placed according to the 10/10 system of electrode placement. Two additional electrodes were placed on the left and right mastoids (M1 and M2). A vertical EOG was recorded from electrodes placed at the supra-orbital and infra-orbital ridges of the left eye. A horizontal EOG was recorded from electrodes placed at the outer canthus of each eye. A defining characteristic of the MMN/DRN is that it inverts in polarity at the mastoids when a nose reference is used. For this reason, the tip of the nose served as a reference for all channels, including the EOG channels. Interelectrode impedances were kept below 5 k. The high frequency cut-off filter was set at 75 Hz and the time constant was set at 2 s (i.e., a low-frequency cut-off filter of 0.08 Hz). The physiological data were digitized continuously at a 500 Hz sampling rate.

# Procedure and Stimuli

Auditory stimuli were presented monaurally to the right ear using EAR 3A insert earphones. Ear of presentation has been shown to have very little effect on either the DRN or P3a (Grimm et al., 2008). The subject was thus able to sleep on the side where the earphone was not inserted. A multi-feature auditory paradigm was presented. The general stimulus procedure is illustrated in **Figure 1**. It consisted of six deviant stimuli presented in the same sequence so that every other tone was an 80 dB SPL 1,000 Hz "standard" tone burst (p = 0.5) and every other was one of six deviants (each with a p = 0.083). Thus, the standards and deviants alternated. Deviants in the multi-feature sequence included (a) a 90 dB SPL "increment" pure tone, (b) a 60 dB SPL "decrement" pure tone, (c) an 80 dB peak SPL white noise burst, (d) different environmental sounds (with an average intensity of 80 dB SPL), (e) a higher frequency, 1,100 Hz, pure tone, and (f) a shorter duration, 100 ms, pure tone. The order of presentation of the deviants were pseudo-randomized so that in an array of six deviants, each deviant was presented only once, and that the same deviant was never presented two times in a row. A different environmental sound was presented on each trial so that none of the environmental sounds were repeated. The features of the

TABLE 1 | The intensity, frequency, duration, and probability of the standard stimulus and the six deviant stimuli in the multi-feature paradigm.


Information in italics represents the feature of the deviant that has been changed.

FIGURE 1 | Multi-feature paradigm. In the example, in the first line, 12 stimuli are presented in an array. The sequence begins with a standard ("Stan") and then alternating with the standards: a white noise ("Noise") deviant, a frequency ("Freq") deviant, a decrement ("Dec") deviant, an environmental sound ("Env") deviant, an increment ("Inc") deviant, and a duration ("Dur") deviant. In the second line, the six deviants are presented again but in a different randomized order. Note that a different environmental sound was presented throughout the sequence. Image adapted from Tavakoli and Campbell (2016).

environmental sounds have been described in detail by Fabiani et al. (1996). They included animal, bird, human vocalizations, musical instruments, environmental, video-game sounds, and mechanically produced sounds. Their duration was, however, manipulated to be 200 ms. The first 10 tones in the sequence consisted of only standards in order to establish a memory trace for the standard stimulus. All stimuli had a duration of 200 ms and a rise-and-fall time of 5 ms, with the exception of the duration deviant. **Table 1** lists the properties of the various auditory stimuli. The stimulus onset asynchrony (SOA; onset-toonset) was 600 ms. A total of 932 stimuli were presented in a single sequence, consisting of 470 trials of standards and 77 trials of each deviant, lasting 9.5 min.

The waking data were collected between 22:00 and 23:00 while subjects were seated in a sound-attenuated room and asked to watch a silent, subtitled movie of their choice and thus to ignore the auditory stimuli. Three blocks of auditory stimuli were presented during wakefulness. A brief break was provided between blocks. Waking testing lasted for approximately 40 min, including break periods. Subjects were then permitted to fall asleep. Once subjects entered stage N2 sleep (verified by the presence of K-complexes and spindles), auditory stimulation was commenced and continued throughout the whole night. Again, stimuli were presented in a 9.5 min block. A brief 5–10 min silent period was provided between blocks. Sleep stages were classified in real-time by an experienced sleep researcher using standard scoring procedures. Stimulation was halted if there was evidence of arousal or awakenings and the entire block of data was rejected. Time permitted the inclusion of three blocks of data within each sleep stage. When more than three blocks were recorded, only the first three blocks were included in the analyses.

### Sleep Stage Scoring

Sleep staging was later confirmed on each 9.5-min sequence by two experienced sleep scorers using the American Academy of Medicine (AASM) task force criteria (Silber et al., 2007). Both scorers were blind to the real-time staging of the EEG. A 20 s epoch was employed rather than the usual 30 s to increase scoring precision. In cases of sleep stage change within the 9.5 min sequence or scoring ambiguity, the entire data within these sequences were rejected.

# ERP Analyses

The data were then reconstructed using Brain Products' Analyzer2 software. A 20 Hz (24 dB/octave) low-pass digital filter was subsequently applied to the continuous EEG data. The use of a high-pass filter is more problematic. This is because the amplitude of the background EEG during sleep is very high relative to the ERP signals of interest. This is particularly the case for low-frequency slow wave delta activity during NREM sleep. A means to attenuate the amplitude of the slow wave activity is through the use of high-pass filtering. The frequency spectrum of the P3a might, however, overlap with that of the lowfrequency delta activity. Another late positive ERP component, the much-studied P3b (or "P300"), has been demonstrated to be distorted by excessive high-pass filtering (Duncan-Johnson and Donchin, 1979; Acunzo et al., 2012). Some compromise must therefore be exercised to avoid over filtering the ERP signals of interest, the DRN and P3a, while still attenuating the energy of the background delta activity. A waking pilot study was thus initially run with seven subjects to determine the effects of different highpass filters on the P3a. Stimulus parameters were as described in the present study and subjects were asked to watch a silent video while ignoring the auditory sequence. Two different high-pass filters, 0.50 and 1.0 Hz were applied to the EEG and compared to the on-line 0.08 Hz high-pass filter. The 0.08 Hz filter would have minimal effect on the low-frequency delta activity in NREM sleep; the 1 Hz filter would markedly attenuate it. Following the filtering of the EEG, averaging procedures described below were applied to the data. The environmental sound deviant in the pilot data did elicit a large P3a. The effects of high-pass filtering on the pilot deviant-standard difference wave for the environmental sound deviant is illustrated in **Figure 2**. As can be observed, the 0.5 Hz high-pass filter had a minimal effect on the amplitude and latency of the P3a. The 1 Hz filter did have an effect on the waveform following the occurrence of the P3a. The 0.5 Hz highpass filter was therefore employed in subsequent analyses of the data. Sabri and Campbell (2001) also noted that a 0.5 Hz filter had minimal effect on the morphology of the DRN. This is also apparent in **Figure 2**.

A vertical EOG channel was computed by subtracting activity recorded at the supra- and infra-orbital ridges of the left eye. A horizontal EOG channel was computed by subtracting activity recorded at the outer canthus of each eye. In the waking state

and during stage REM, eye movements and blinks could cause artifact in the EEG recordings. Independent component analysis (Makeig et al., 1996; Chaumon et al., 2015) was used to identify the ocular activity that was statistically independent of the EEG activity. These artifacts were then partialled out of the EEG trace. The continuous data were subsequently reconstructed into discrete single trial 700 ms segments, beginning 100 ms before stimulus onset. A 0–50 ms post-stimulus baseline correction was applied to the waking data and all sleep stages because the prestimulus interval was not stable and varied between the waking state and sleep stages. It is possible that subjects might have been able to predict the onset of the stimulus during wakefulness thus affecting the pre-stimulus baseline. It would be expected that the first 50 ms following stimulus onset would mainly reflect stimulus-related sensory processing and thus should not have varied as a function of stimulus type. During wakefulness, segments in which EEG activity exceeded ±100 µV relative to the baseline were excluded from further analyses. No more than 5% of segments were rejected for any individual subject because eye movement had already been corrected. During stages N2, N3, and REM, this threshold was changed to ±200 µV to accommodate the large amplitude slow wave activity that is common to sleep. Because the amplitude of these slow waves would have been attenuated by the high-pass filter, fewer than 5% of segments were also rejected during sleep. There was no variation in the rejection of single segments across deviants or stages of sleep. The first 10 standard stimuli were excluded from the averaging procedure. The single trial segments were then sorted and averaged on the basis of state (waking, stage N2, stage N3, stage REM), stimulus type (standard and six deviants), and electrode site.

### Quantification and Statistical Analyses

The auditory deviant stimuli elicited a series of deflections that were not apparent in the standard ERP waveform. These deflections are best observed in a difference wave computed by subtracting, point-by-point, the standard from the deviant averaged waveforms at each electrode site. This process removes the commonalities in processing between the standard and the deviant, leaving only processing unique to the deviant. From this difference wave, the DRN and P3a were initially identified using the grand averaged data (the average of all subjects' averages) within waking and stages N2, N3, REM. They were then quantified for each individual subject using the mean of all the data points within ±25 ms of the peak in amplitude that was identified in the grand average.

Previous studies have indicated that not all deviants will elicit a P3a during wakefulness. It was also possible that the P3a would be absent to some or all deviants during sleep. Thus, confidence intervals were computed to determine whether a deflection was significantly less or greater than the baseline setting (in the case of the DRN and P3a, respectively). The procedure was run on the Fz electrode site for the DRN and at Cz for the P3a where each tends to be at maximum amplitude. Because a directionality was predicted (negativity in the case of the DRN and positivity in the case of the P3a), one-tailed tests of significance (p < 0.05) were applied to the confidence intervals. To restrict the likelihood of chance findings, the negativity had to conform to the usual latency (100–250 ms) and scalp distribution (fronto-central maximum, inversion in polarity at the mastoids) of the DRN, while the positivity had to conform to the usual latency (180– 350 ms) and scalp distribution (centro-frontal maximum) of the P3a.

Electrode sites were grouped into regions of interest (ROIs), to include nine electrode sites where the ERP components of interest were largest. The ROIs allowed for an analysis of an anterior–posterior and an inter-hemisphere factor. Specifically, for the anterior–posterior electrode factor, three electrodes for frontal (F3, Fz, F4), fronto-central (FC3, FCz, FC4), and central (C3, Cz, C4) sites were chosen for analysis. The DRN and P3a components were thus quantified at each of these sites within the latency range identified at Fz and Cz, respectively. For the inter-hemisphere factor, three electrodes for left (F3, FC3, C3), midline (Fz, FCz, Cz), and right (F4, FC4, C4) sites were chosen for analysis.

ANOVA procedures were then run to compare the amplitude of the DRN and P3a across waking and sleep stages. ANOVA testing was only run when a significant DRN or P3a was observed for a deviant during any of the sleep stages. Specific ANOVA procedures thus varied among the different deviants and will be described in the Section "Results." Significant main effects and interactions were followed up with Fisher's LSD post hoc testing. For all statistical analyses, a Geisser and Greenhouse (1958) correction was used when appropriate.

### RESULTS

### Standard ERP

Because the DRN and P3a were measured in a difference wave, an assumption is made that the processing of the standards is constant across all conditions. Any differences observed in

the difference wave could therefore only be attributed to the additional processing of the deviant. This assumption was tested. **Figure 3** depicts the ERPs elicited by the standard auditory stimuli during waking, stage N2, stage N3, and stage REM sleep. The amplitude of N1 and P2 were very small because of the fast rate of stimulus presentation. An ANOVA with repeatedmeasures on stage (waking, stage N2, stage N3, stage REM) was run at Cz to determine the effects of stage on the amplitudes of the standard N1 and P2. For the N1, there was a significant effect of stage, F(3, 45) = 15.42, MSE = 1.73, p < 0.0001, η 2 <sup>p</sup> = 0.51. Fisher's LSD revealed that the N1 during stage N3 was significantly smaller (more positive) than in waking, stage N2, and stage REM. As can be observed in **Figure 3**, this is due to the positive drift in the ERP waveform during stage N3. There was no significant difference in the amplitude of the N1 across waking, stage N2, and stage REM.

Similarly, for the P2, there was a significant effect of stage, F(3, 45) = 15.76, MSE = 2.64, p < 0.0001, η 2 <sup>p</sup> = 0.51. Fisher's LSD again revealed that the P2 during stage N3 was significantly larger than in waking, stage N2, and stage REM. There was no significant difference in the P2 between waking, stage N2, and stage REM.

An additional negativity at about 250 ms was also observed in the ERPs following the standard stimulus. There was a significant effect of stage for this negativity, F(3, 45) = 25.53, MSE = 3.73, p < 0.0001, η 2 <sup>p</sup> = 0.63. Fisher's LSD revealed that this negativity was significantly smaller (more positive) in stage N3 compared to waking, stage N2, and stage REM. The negativity was also significantly smaller in stage N2 compared to stage REM. There were no significant differences in its amplitude between waking and stage N2, and between waking and stage REM.

### DRN in the Waking State

The difference waves for the various deviants across the waking and sleep states are illustrated in **Figure 4**. In the waking state, a negativity peaking at about 150 ms was observed following all deviants. It was maximum over frontal areas of the scalp and inverted in polarity at the mastoids, and thus corresponded to a DRN. Confidence interval testing in the waking state revealed that all deviants, except the decrement deviant, elicited a DRN significantly different from the zero-voltage baseline (p < 0.01 in all cases).

An ANOVA with repeated measures on deviant type (frequency, duration, decrement, increment, white noise, environmental sounds), frontality (frontal, fronto-central, central), and laterality (left, midline, right) was run on the waking DRN data. There was a significant main effect of deviant type, F(5, 75) = 5.76, MSE = 12.77, p < 0.0001, η 2 <sup>p</sup> = 0.28. The DRN was largest to the increment deviant and smallest to the decrement deviant. The overall amplitude of the DRN was largest over frontal areas of the scalp, F(2, 30) = 12.24, MSE = 1.35, p < 0.0001, η 2 <sup>p</sup> = 0.45 and was significantly reduced at central sites. The DRN was also largest over midline electrode sites, F(2, 30) = 6.33, MSE = 0.70, p < 0.01, η 2 <sup>p</sup> = 0.30. Fisher's LSD revealed no inter-hemispheric differences in the amplitude of the overall DRN between left and right areas of the scalp. Interactions involving deviant and electrode site were not significant (F < 1).

### DRN During Sleep

During stage N2 sleep, confidence interval testing indicated that none of the deviants elicited a DRN that was significantly different from the baseline (p > 0.05 in all cases). Surprisingly, during stage N3 sleep, the frequency and the decrement deviant did elicit a significant negativity during the time interval of the DRN (p < 0.05 in both cases). As can be observed in **Figure 4**, this could be due to an overall long-lasting negative drift that began prior to stimulus onset which is often observed during slow wave stage N3 sleep. The amplitude of the frequency DRN was nevertheless significantly larger during the waking state compared to stage N3, t(16) = 4.21, p < 0.01, but the decrement DRN did not significantly vary (t < 1) between the waking state and stage N3. All other deviants failed to elicit a significant DRN during stage N3. During REM sleep, none of the deviants elicited a significant DRN (p > 0.05 in all cases). A small negativity could be observed for the increment, but it was not significantly different from the baseline (p > 0.05).

### P3a in the Waking State

Not all of the deviants elicited a significant P3a in the waking state (**Figure 4**). In the difference waves, a large amplitude frontocentral maximum P3a was apparent only for the environmental sound and white noise deviants, peaking at about 240 and 215 ms, respectively. Confidence interval testing (**Table 2**) of the Cz data indicated that its amplitude was significantly different from



Confidence intervals (95%) are also reported for the Cz electrode site.

FIGURE 4 | Grand average difference waveforms of the six deviants. ERPs shown at the Fz, Cz, and Pz electrode sites. The waveforms for the four different stages are superimposed for each deviant stimulus.

REM it was largest over centro-parietal areas.

baseline levels only for these deviants (p < 0.001 in both cases). An initial ANOVA with repeated measures on deviant type (white noise, environmental sounds), frontality (frontal, fronto-central, central), and laterality (left, midline, right) was run on the waking P3a data. The amplitude of the P3a to the environmental sounds and white noise deviants did not significantly differ (F < 1). The overall amplitude of these P3a data were largest over frontocentral areas of the scalp, F(2, 30) = 3.78, MSE = 0.60, p < 0.05, η 2 <sup>p</sup> = 0.20, and significantly reduced at frontal sites. The P3a was largest over midline areas of the scalp, F(2, 30) = 15.46, MSE = 0.85, p < 0.0001, η 2 <sup>p</sup> = 0.51. Fisher's LSD again revealed no inter-hemispheric difference between left and right scalp regions. Interactions involving electrode site were not significant (F < 1).

### P3a During Sleep

Again, during sleep, not all of the deviants elicited a significant P3a-like positivity. For some deviants, a positivity at the time of the P3a was observed when it was not apparent in the waking state. **Figures 5**–**7** illustrate the deviants at multiple scalp sites for which a P3a-like positivity was observed either during the waking or sleep states. During stage N2, confidence interval testing at Cz revealed that the environmental sound, white noise, and increment deviants elicited significant positivities at about 250, 190, and 200 ms, respectively (p < 0.01 in all cases). All other deviants failed to elicit a significant P3a during stage N2. The amplitude of the P3a was also compared in stage N2 occurring within the first and second halves of the night to determine possible time-of night-differences. The P3a did not significantly differ as function of time-of-night for either the environmental sound, white noise, and increment deviants (p > 0.05 in all cases). The stage N2 data were therefore collapsed across early and late halves of the night. During stage N3, only the increment deviant elicited a positivity at about 200 ms (p < 0.01). All other deviants, including the environmental sounds and white noise, failed to elicit a significant P3a-like positivity in stage N3 (p > 0.05). During stage REM, the environmental sounds again elicited a significant P3a-like positivity at about 220 ms (p < 0.01). A positivity was also observed following the white noise deviant, at about 190 ms, and increment deviant at about

200 ms. The positivity to both these deviants failed to attain significance (p > 0.05).

For the environmental sounds (**Figure 5**), an ANOVA with repeated measures on stage (waking, stages N2, N3, REM), frontality (frontal, fronto-central, central), and laterality (left, midline, right) was run. The amplitude of waking-P3a and the P3a-like positivities in during stages N2, N3, and REM were not significantly different (F < 1). The overall ANOVA revealed that the P3a was largest at fronto-central sites, F(2, 30) = 4.13, MSE = 0.79, p < 0.05, η 2 <sup>p</sup> = 0.21. Spline maps (bottom portion of **Figure 2**) confirmed this for the waking, stage N2, and REM P3a-like positivities. During stage N3, however, the distribution was more central. Nonetheless, the interaction between stage × frontality just failed to reach significance, F(6, 90) = 1.95, MSE = 0.63, p = 0.08, η 2 <sup>p</sup> = 0.11.

For the white noise deviant (**Figure 6**), the positivity observed during stages N2, N3, and REM occurred earlier (190 ms) than for the waking-P3a and its scalp distribution (bottom portion of **Figure 3**) was centro-parietally maximum, unlike the usual fronto-central P3a. Given the centro-parietal maximum distribution, the ROI analysis was adjusted with an ANOVA being run with repeated measures on stage (waking, stages N2, N3, REM) and electrode site (Fz, Cz, Pz). The amplitude of the waking P3a and the positivity in stages N2, N3, and REM were not significantly different, F(3, 45) = 1.83, MSE = 8.94, p = 0.16, η 2 <sup>p</sup> = 0.11. The stage × electrode site interaction was, however, significant, F(6, 90) = 4.36, MSE = 0.92, p < 0.001, η 2 <sup>p</sup> = 0.22. During waking and stage N2, the positivity was largest at Cz, while during stages N3 and REM, the positivity was largest at Pz.

In the case of the increment deviant (**Figure 7**), a large positivity was observed during stages N2 and N3. Similar to the positivity following white noise deviant during sleep, this positivity occurred earlier (200 ms) than the usual waking-P3a and its scalp distribution (bottom portion of **Figure 4**) was centro-parietally maximum, unlike the usual fronto-central P3a. An ANOVA was thus again run with repeated measures on stage (waking, stages N2, N3, REM) and electrode site (Fz, Cz, Pz). Overall, the positivity was significantly reduced during wakefulness compared to all sleep stages, F(3, 45) = 9.67, MSE = 10.10, p < 0.001, η 2 <sup>p</sup> = 0.40. Additionally, Fisher's LSD revealed the positivity in stage N3 was also significantly larger than during stage REM. The stage × electrode site interaction was also significant, F(6, 90) = 4.16, MSE = 1.24, p < 0.001, η 2 <sup>p</sup> = 0.22. During stages N2 and REM, the positivity was largest at Cz, while during stage N3 it was largest at Pz. Spline maps of the positivity during stages N2 and N3 emphasize this centroparietal distribution. During REM, this distribution was very broad, spanning from frontal to parietal regions.

### Late Positivity During REM

A later positivity, at about 300–400 ms, occurring at about the same time as a positivity observed by Ruby et al. (2008), was also observed following some of the deviants, but only during REM sleep. This positivity was large over fronto-central areas of the scalp (**Figures 5**–**8**). When measured at Fz, it was significantly different from the baseline following the duration deviant (**Figure 8**), (p < 0.01) and the increment deviant (**Figure 4**), (p < 0.05). It did not reach statistical significance following either the environmental sound or white noise deviants (p > 0.05 in both cases). Previous studies have also reported a positivity at about this time, but with a parietal distribution (Bastuji et al., 1995; Cote and Campbell, 1999; Cote et al., 2001; Ruby et al., 2008). For this reason, it was measured at Fz, Cz and Pz. For the stage REM data, a two-way ANOVA with repeated-measure on deviant (environmental sounds, white noise, increment, and duration) and electrode site (Fz, Cz, Pz) was run. The amplitude of the positivity did not significantly differ among deviant types (F < 1). The overall amplitude of the positivity was slightly larger at Fz compared to Cz and Pz, although this difference was not significant (F < 1). The deviant type × electrode site interaction was also not significant (F < 1).

# DISCUSSION

### DRN in Wakefulness and Sleep

The multi-feature paradigm proved to be successful in permitting a significant DRN to be elicited by all deviants in the waking state, except for the decrement. A small DRN was, in fact, observed following the presentation of the decrement, but it did not attain significance. The latency and scalp distribution of the waking DRNs essentially replicated the DRN findings observed by Tavakoli and Campbell (2016) and Tavakoli et al. (2018). In the present study, subjects were tested late in the evening compared to during the day in the Tavakoli and Campbell study. Thus, provided subjects are awake, the DRN appears to be well-preserved, even if sleep is very imminent. When subjects are no longer awake, during NREM (stages N2 and N3) and REM sleep, a discernible DRN was not apparent following most of the deviants. A significant negativity was observed in the difference wave following the frequency and decrement deviants during stage N3. These results should be interpreted with caution, however, as the processing of the standard ERP was significantly different during the time interval of the DRN during stage N3. Because of the difference in processing of the standard during stage N3, possible differences observed in the difference waves during this stage could be a result of the differences in processing of the standard. The lack of a DRN during NREM is in agreement with most previous studies that have used either frequency or intensity deviants (Nielsen-Bohlman et al., 1991; Winter et al., 1995; Loewy et al., 1996, 2000; Nashida et al., 2000). During REM, some studies have reported the presence of a DRN (Loewy et al., 1996; Atienza et al., 1997; Sabri and Campbell, 2005) while others have not (Sallinen et al., 1996; Loewy et al., 2000; Macdonald et al., 2008). Ruby et al. (2008) did report a significant MMN during both NREM and REM when a 30 ms deviant signaled a change from the duration of a 75 ms standard. In the present study, the duration deviant did not elicit a DRN. The duration of the standard and deviant stimuli (200 and 100 ms, respectively) were, however, much longer than those used by Ruby et al. (2008). It is possible that the sleeping brain can only detect brief and abrupt changes in stimulus duration. Additionally, the relative

positivity at about 350 ms can also be observed unique to REM sleep. Spline maps (bottom portion of Figure) are also presented for stages N2, N3, and REM at the time interval of the initial positivity. This positivity during stages N2 and N3 has a centro-parietal distribution, while during REM, this distribution was very broad, spanning from frontal to parietal regions.

change in duration of the standard was larger in the Ruby et al. (2008) study than in the present one. Sabri et al. (2003) and Sabri and Campbell (2005) have also noted that a very large frequency change (1,000 Hz standard; 2,000 Hz deviant) elicited a DRN in stage N2 and REM but a smaller change (1,100 Hz) did not. The frequency deviant in the present study represented the same extent of change as the small frequency deviant used by Sabri et al. (2003; Sabri and Campbell, 2005); and also did not elicit a significant DRN during stage N2, N3, or REM.

It is also possible that the use of the multi-feature paradigm may account for the failure to observe a DRN during sleep. While the presentation of multiple deviants in the multi-feature paradigm permit the reliable recording of the DRN in the waking state, this may not be the case during sleep. The paradigm employed six different deviants each occurring quite rarely. Nevertheless, the overall probability of occurrence of deviants was 0.50, the standards and deviants being presented in an alternating pattern. Each different deviant does, however, share many features with the standard and as such, strengthens the sensory memory for the standard. Sabri and Campbell (2005) suggested that the failure to elicit an MMN during NREM sleep might be because of a rapidly fading sensory memory for the standard. In the case of the multi-feature paradigm, it is possible that the sensory memory for the standard is also poorly formed during sleep because it is only presented on 50% of trials. In studies employing oddball paradigms, the standard is presented on at least 80% of trials.

### P3a During Wakefulness

During wakefulness, a significant P3a was only observed following the environmental sounds (at 240 ms) and white noise deviants (at 215 ms). This replicates the results observed

by Tavakoli and Campbell (2016). The amplitude of the P3a recorded during wakefulness in the present study was nevertheless somewhat reduced compared to that observed in the Tavakoli and Campbell study (about 1 µV). It is possible that this simply reflects between-group differences. A more likely explanation is that Tavakoli and Campbell (2016) tested subjects during the day and were thus presumably well-rested. In the present study, subjects were tested late in the evening when the demand for sleep was high. Similarly, in the Tavakoli et al. (2018) study, subjects were also tested late in the evening, and the waking P3a amplitudes are comparable to the present study.

Previous studies that have presented oddball paradigms have also observed a large P3a to environmental sound and white noise deviants compared to other pure tone deviants (Combs and Polich, 2006; Cahn and Polich, 2009; Berti, 2012; Frank et al., 2012; Wetzel et al., 2013). Tavakoli and Campbell (2016) and Tavakoli et al. (2018) also noted that the frequency, duration, decrement, and increment deviants did not elicit a P3a in the waking state. When a passive paradigm is used and subjects ignore the auditory sequence, it does appear that a large extent of change from the standard is required to elicit the P3a. On the other hand, when subjects are attending to the auditory sequence containing the deviant, much smaller deviants have been reported to elicit a P3a (Rinne et al., 2006; Grimm et al., 2008; Berti et al., 2013). The environmental sound and white noise deviants varied widely in terms of their frequency spectrum and stimulus energy (e.g., intensity) from the pure tone standards. All other deviants represented only a change in a single feature from the standard. Increment deviants in previous waking studies have repeatedly been shown to also elicit a P3a. The duration of the stimulus might again explain the discrepancies. In these studies, the duration of the increment deviants was very brief, typically 50 ms (Muller-Gass et al., 2006, 2007; Macdonald et al., 2008) compared to the much longer 200 ms used in the present study. Research has shown that the perceived intensity of a stimulus is directly affected by the duration of the stimulus (Zwislocki, 1960; Stévens and Hall, 1966). In other words, the perceived intensity of a stimulus increases as the duration of the stimulus is increased. However, in the case of oddball and multifeature paradigms, the perceived intensity of both the deviant and standard increases. This may make the increment deviant to be less obtrusive.

### P3a During Sleep

No previous studies have used environmental sound deviants during all-night sleep. Remarkably, following the presentation of the environmental sounds, a positivity at the time of the waking P3a (230 ms) continued to be elicited throughout the entire night of sleep (i.e., during stages N2 and N3 of NREM sleep, and REM sleep). The scalp distribution of this P3a did not differ between waking and sleep states and the spline maps were very similar. As a result, there is little evidence to suggest that the intracranial sources of the P3a were different between the waking and sleep states. The positivity recorded to the environmental sound deviants during sleep thus appears to be a "true" P3a. Similarly, Tavakoli et al. (2018) also observed a significant P3alike positivity during the sleep onset period during stage N1 and the first 30 min of stage N2. The environmental sounds, unlike any other deviant, differed on every trial, each containing a unique spectral content. By comparison, the spectral content of the other deviants was the same on each presentation. These environmental sounds are also more ecologically valid as most

are sounds that are frequently experienced. These sounds could be sub-grouped from different categories of sound (musical instruments, birds, etc.). It is possible that certain highly salient and relevant sounds, such as human voices (Perrin et al., 1999; Pratt et al., 1999) or sounds having an emotional context (Pinheiro et al., 2016, 2017) might be processed more extensively than others. Unfortunately, it was not possible to average the various sub-categories of environmental sounds. Overall, the environmental sounds were presented on only about 8% of trials. As a result, an insufficient number of each sub-category of environmental sounds was presented to allow for a reduction of the background noise of the large amplitude EEG during sleep.

During both NREM and REM sleep, the increment and white noise deviants also elicited a positivity. This positivity, however, occurred around 190–200 ms, unusually early for a P3a. Tavakoli et al. (2018) also observed a larger positivity following the same increment deviant during the sleep onset period. Ruby et al. (2008) also observed a positivity in stages N2 and N3 that peaked earlier than the P3a in the waking state. In their case, the scalp distribution maps of the positivities in waking and sleep were similar. In the present study, the scalp distributions were different. For both deviants, it was maximum over centroparietal regions of the scalp during stages N2 and N3 compared to the fronto-central scalp distribution of the P3a. During REM, the distribution was much broader. ERP components that have different scalp distributions must have different intra-cranial sources (Picton et al., 1995). It is, therefore, unlikely that the positivity to the increment and white noise deviants reflect the same P3a source during the waking and sleeping states. Additionally, the increment deviant did not elicit a P3a in the waking state. Some studies have reported a P3a-like positivity during stage REM following the presentation of a particularly large increase in stimulus intensity (Cote and Campbell, 1999; Macdonald et al., 2008) but this increment deviant also elicited large P3a in wakefulness. It is possible that these positivities reflect the earlier P2. The peak latency and scalp distribution of this positivity following both increment and white noise is more consistent with the more usual posterior scalp distribution of the P2 than the more anterior P3a. Many sleep studies have observed that the amplitude of the P2 may increase during NREM sleep (Nielsen-Bohlman et al., 1991; Winter et al., 1995; Nittono et al., 2001; Crowley et al., 2002; Macdonald et al., 2008; Campbell and Muller-Gass, 2011). In the present study, the amplitude of the increment positivity was much also much larger during NREM sleep than in wakefulness. In their review, Crowley and Colrain (2004) have suggested that the amplitude of P2 reflects an inhibition of processing. Thus, the appearance of a large P2 during NREM reflects a need for the protection of sleep. The P2 was larger in stage N3 than N2. This is also consistent with a greater for sleep to remain undisturbed during deep "slow wave sleep" (stage N3) than in the lighter stage N2.

### Late Positivity During REM

A later small amplitude positivity, at about 300–400 ms, was also observed but was unique to REM sleep. What was especially surprising was that it was elicited by many different deviants and its amplitude did not significantly differ among them. It had a fronto-central distribution. It is possible that it simply reflects residual background noise. However, it was found to be consistently elicited at about the same latency and to have a similar scalp topography across multiple deviants. Moreover, other studies have also reported a late positivity (around 300– 500 ms) during REM (Bastuji et al., 1995; Cote and Campbell, 1999; Perrin et al., 1999; Cote et al., 2001; Ruby et al., 2008). In these studies, the scalp distribution was different, the late positivity having a distinctly centro-parietal maximum. This scalp distribution is consistent with the much-studied P3b.

It is possible that the early and late positivities during stage REM reflect a delayed P2 and P3a. The early positivity did have a centro-frontal scalp distribution, consistent with an actual P3a. As mentioned above, the scalp distribution of the P2 is more posterior. Escera et al. (1998) observed that their waking P3a to environmental sounds had two distinct subcomponents. The early portion of the P3a, peaking at around 230 ms, had a centrally dominant scalp distribution. The late portion of the P3a, on the other hand, peaked at around 330 ms and displayed a frontally maximum scalp distribution. The late positivity observed during stage REM in the present study appears to be similar to the late portion of the P3a described by Escera et al. Nevertheless, it is surprising that similar early and late subcomponents were not observed in the waking state. The functional significance of late positivity in the present study occurring only in REM sleep is thus difficult to interpret.

### Disassociation of the DRN and P3a

A DRN was not observed during either NREM or REM sleep, yet a P3a-like positivity was still elicited following some deviants. Based on the classic Näätänen model, the probability of eliciting a P3a is expected to decrease as the amplitude of the MMN/DRN is reduced. More recent studies have now suggested that the amplitude of the MMN/DRN and P3a are not necessarily linked (Sussman et al., 2003; Fischer et al., 2008; Horváth et al., 2008). Therefore, a deviant might elicit a P3a in the absence of an MMN. Näätänen also notes that other factors may affect the P3a, suggesting that the threshold for its elicitation is flexible. It is possible that during sleep the threshold to elicit the P3a is significantly lowered in order to alert the individual to potentially highly relevant information in the environment. Thus, even though the amplitude of the MMN/DRN is reduced, the P3a may still be elicited.

# CONCLUSION

The purpose of this study was to determine whether potentially highly relevant stimulus change can intrude into conscious awareness during all-night natural sleep using an auditory multi-feature paradigm. In the present study, a fronto-central maximum P3a was passively elicited in the waking state by environmental sound deviants. Importantly, the environmental sounds continued to elicit a similar positivity occurring at about the same time during both NREM and REM sleep. Subjects

rarely were awakened by the auditory stimuli, even by the environmental sounds that did elicit a large P3a. Thus, while these stimuli might be considered to be potentially highly relevant and meriting additional processing as reflected by the presence of the P3a, in the end, they turn out to be not so critical as to warrant sleep to be reversed.

Other deviants failed to elicit a definitive P3a during sleep. Another earlier positivity having a different scalp distribution than the P3a was elicited by the white noise and increment deviants during stages N2, N3, and REM, although it did not attain significance in REM following either deviant. This positivity may be an earlier P2 often reported to increase in amplitude during sleep. The environmental sounds, white noise, increment, and duration deviants also elicited a later fronto-central maximum positivity unique to REM which has not been reported in previous studies. What this late positivity reflects is largely unknown. Again, awakening during REM sleep was relatively rare.

It should be noted that auditory stimuli may trigger evoked K-Complexes during NREM sleep which may affect the shape of the ERP waveform (Czisch et al., 2009). The presence of evoked K-Complexes were not accounted for in the present study. Nonetheless, K-Complexes are only elicited when stimuli are presented very slowly (>10 s; Bastien and Campbell, 1994). Thus, very few K-Complexes would have been elicited, even by the deviants that did elicit a P3a. Moreover, the large negativity associated with the K-Complex (the N550) occurs well after the P3a and, therefore, should not affect its morphology.

The current study also has implications for other unconscious states such as general anesthesia and coma. Researchers are currently employing ERP techniques to probe the extent of information processing and consciousness in patients who may be covertly conscious (for reviews, see Gawryluk et al., 2010; Harrison and Connolly, 2013; Morlet and Fischer, 2014). The multi-feature paradigm

### REFERENCES


may be a useful tool in these clinical and applied studies.

### ETHICS STATEMENT

The study was approved by the University of Ottawa's Health Sciences and Science Research Ethics Board. Written informed consent was obtained prior to the start of the study and subjects received an honorarium for their participation. The study was conducted according to the Canadian Tri-Council guidelines (Medical, Natural, and Social Sciences) on ethical conduct involving human subjects. These guidelines are similar to those used with the Declaration of Helsinki.

### AUTHOR CONTRIBUTIONS

PT, KC, AB, and AD contributed to the rationale and the design of the study and read and approved the final manuscript. The manuscript was written by PT. PT and AD assisted with the collection and analysis of the EEG data. KC, AD, and AB provided feedback and revisions on written drafts of the manuscript.

### FUNDING

Financial support for this research was provided by an operating grant (8242) to KC by the Natural Sciences and Engineering Research Council of Canada (NSERC) and by a doctoral scholarship to PT by NSERC.

### ACKNOWLEDGMENTS

The authors wish to thank Sonia Varma and Samantha Kenny for their assistance with the collection of data.

auditory evoked responses using the oddball paradigm. J. Clin. Neurophysiol. 12, 155–167. doi: 10.1097/00004691-199503000-00006


Colrain, I. M., and Campbell, K. B. (2007). The use of evoked potentials in sleep research. Sleep Med. Rev. 11, 277–293. doi: 10.1016/j.smrv.2007.05.001


Rapid Eye Movement sleep. Neuroreport 19, 309–313. doi: 10.1097/WNR. 0b013e3282f4ede8



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Tavakoli, Dale, Boafo and Campbell. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Network Mechanism of the Central Circadian Pacemaker of the SCN: Do AVP Neurons Play a More Critical Role Than Expected?

### Michihiro Mieda\*

Department of Integrative Neurophysiology, Graduate School of Medical Sciences, Kanazawa University, Kanazawa, Japan

The suprachiasmatic nucleus (SCN) functions as the central circadian pacemaker in mammals and entrains to the environmental light/dark cycle. It is composed of multiple types of GABAergic neurons, and interneuronal communications among these neurons are essential for the circadian pacemaking of the SCN. However, the mechanisms underlying the SCN neuronal network remain unknown. This review will provide a brief overview of the current knowledge concerning the differential roles of multiple neuropeptides and neuropeptide-expressing neurons in the SCN, especially focusing on the emerging roles of arginine vasopressin-producing neurons uncovered by recent studies utilizing neuron type-specific genetic manipulations in mice.

### Edited by:

Michael Lazarus, University of Tsukuba, Japan

### Reviewed by:

Tomoko Yoshikawa, Kindai University, Japan Yoshiaki Yamaguchi, Kyoto University, Japan

\*Correspondence: Michihiro Mieda mieda@med.kanazawa-u.ac.jp

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 08 December 2018 Accepted: 07 February 2019 Published: 25 February 2019

### Citation:

Mieda M (2019) The Network Mechanism of the Central Circadian Pacemaker of the SCN: Do AVP Neurons Play a More Critical Role Than Expected? Front. Neurosci. 13:139. doi: 10.3389/fnins.2019.00139 Keywords: circadian rhythm, suprachiasmatic nucleus, vasopressin, vasoactive intestinal peptide, neural network

# INTRODUCTION

The circadian oscillator of the hypothalamic suprachiasmatic nucleus (SCN) is the central pacemaker in mammals, orchestrating multiple circadian biological rhythms in the organism and being regulated according to the external light/dark conditions conveyed from the eye (Reppert and Weaver, 2002). The SCN contains ∼20,000 neurons, most of which are able to oscillate autonomously. Individual cellular oscillators (cellular clocks) are driven by autoregulatory transcriptional/translational feedback loops (TTFLs) of clock genes in concert with cytosolic signaling molecules, including cAMP and Ca2<sup>+</sup> (Welsh et al., 2010; Herzog et al., 2017; Takahashi, 2017). Surprisingly, these intracellular molecular mechanisms are not unique to SCN cells but are shared with peripheral cells (Balsalobre et al., 1998). Rather, intercellular communications among SCN cells through the neuronal and diffusible network are the unique feature of the SCN that is responsible for the generation of highly robust and coherent oscillations as an ensemble (Welsh et al., 2010).

### STRUCTURE OF THE SCN

The SCN is a heterogeneous structure that consists of multiple types of GABAergic neurons (Antle and Silver, 2005). Many of them co-express neuropeptides, represented by. arginine vasopressin (AVP)-producing neurons located in the shell, the dorsomedial part, of the SCN, as well as by vasoactive intestinal peptide (VIP)-producing neurons and gastrin releasing peptide (GRP) producing neurons in the core, the ventrolateral part, of the SCN (**Figure 1A**). Rhythmic Period

(Per) expression in constant darkness (DD) is highest in the SCN shell (Hamada et al., 2004). In contrast, the SCN core contains retinorecipient neurons that respond immediately to the environmental light stimuli and communicate this information to the shell (Silver et al., 1996; Shigeyoshi et al., 1997). Two other input pathways from the median raphe and intergeniculate leaflet also terminate mainly in the SCN core, while afferents from the hypothalamus and limbic areas terminate mainly in the SCN shell (Moga and Moore, 1997). SCN neurons project principally to areas within the diencephalon, especially to the subparaventricular zone, the area just dorsal to the SCN. The majority of efferent projections originates in the shell, but the core neurons also send efferent projections in a manner different from the shell neurons (Leak and Moore, 2001). Within the SCN, core neurons send projections densely to the shell, while fibers of shell neurons in the core are sparse (Leak et al., 1999).

# VIP: A CRITICAL SYNCHRONIZER OF SCN NEURONS

Vasoactive intestinal peptide has been demonstrated to be especially important for the maintenance and synchronization of cellular clocks in the individual SCN neurons (Herzog et al., 2017). Thus, mice lacking Vip or VIP receptor Vipr2 gene demonstrate drastically weakened behavioral rhythms, often with multiple period components (Harmar et al., 2002; Colwell et al., 2003; Aton et al., 2005). At the cellular level, circadian oscillations of electrical firing and clock gene expression of individual SCN neurons (cellular clocks) are desynchronized in slices (Aton et al., 2005; Maywood et al., 2006; Brown et al., 2007). Furthermore, the numbers of rhythmic neurons are drastically reduced (Aton et al., 2005; Maywood et al., 2006; Brown et al., 2007). These observations suggest that VIP functions as a master synchronizer in the SCN.

Consistent with these observations, optogenetic activation of VIP neurons in the SCN phase-shifts the cellular clock (circadian PER2::LUC oscillation) in explants and entrains the behavior rhythm in vivo (Jones et al., 2015; Mazuski et al., 2018). In addition, chemogenetic inhibition of these neurons attenuates the light-induced phase-shift of circadian behavior rhythm (Jones et al., 2018). In explants, prolonged chemogenetic stimulation of these neurons further reprograms the global spatiotemporal dynamics of the SCN cellular clocks (Brancaccio et al., 2013).

On the other hand, VIP neurons may not play a significant role in the pacemaking of circadian rhythms by the SCN network, that is, determination of the circadian period, as discussed again later. Lee et al. demonstrated that overexpression of the Clock∆<sup>19</sup> transgene in VIP neurons had no effect on the behavioral free-running period, although such a Clock∆<sup>19</sup> overexpression lengthens the intrinsic period of cellular clocks in the manipulated cells (**Table 1**; Lee et al., 2015). These results contrasted clearly with the observations that the same genetic manipulations in SCN neurons expressing a neuropeptide neuromedin-S (NMS) lengthens the period of behavior rhythm: NMS neurons include VIP neurons, AVP neurons, and some other types of neurons. In conjunction with the observation that genetic ablation of cellar clocks specifically in NMS neurons disrupts circadian rhythms, Lee et al. (2015) concluded that NMS neurons act as essential pacemakers in the SCN.

### AVP PEPTIDE MAY ALSO BE INVOLVED IN THE COUPLING OF THE SCN NEURONS

The concentration of AVP in the cerebrospinal fluid (CSF) daily fluctuates with a peak in the morning (Stark and Daniel, 1989; Kalsbeek et al., 2010). Such a circadian variation of CSF AVP level has been shown to originate from the AVP content in the SCN (Södersten et al., 1985). Indeed, the transcription of Avp gene in the SCN is under the control of cellular clocks, the molecular machinery of the circadian clock composed of TTFL of clock genes (Jin et al., 1999). In conjunction with the fact that AVP-deficient Brattleboro rats display attenuated circadian rhythms but little abnormality in circadian pacemaking, AVP has been considered to function as an SCN output (Groblewski et al., 1981; Brown and Nunez, 1989; Kalsbeek et al., 2010). A recent optogenetic study directly demonstrated clock-driven AVP neurotransmission that mediates anticipatory thirst prior to sleep (Gizowski et al., 2016).

Nevertheless, AVP may also play a minor but significant role in the coupling of SCN neurons. In coculture experiments of SCN explants, the requirement of AVP signaling for the synchronization of SCN neurons becomes manifest in the absence of VIP signaling (Maywood et al., 2011; Edwards et al., 2016; Ono et al., 2016). Deletion of V1a receptor, the principal AVP receptor of the SCN, lengthens the activity time in DD by approximately 100 min in mice, suggesting the attenuated coupling among SCN neurons (Li et al., 2009). A small number of these mice even show arrhythmicity. In another study, V1a−/−; V1b−/<sup>−</sup> mice were reported to immediately reentrain to phaseshifted LD cycles whereas their free-running rhythms are intact, indicating that interneuronal communication mediated by AVP make the SCN resistant to environmental perturbations such as jet lag (Yamaguchi et al., 2013). In contrast to VIP, Avp knockout mice are not available for the examination of circadian behavior rhythm, because they do not survive beyond postnatal day 7 (Yoshikawa et al., 2015). Therefore, development and study of SCN-specific knockout mice would further elucidate the physiological role of AVP signaling in the central circadian clock.

### AVP-PRODUCING "NEURONS" ARE CRITICAL FOR THE COUPLING OF SCN NEURONS

Arginine vasopressin neurons express neurotransmitters other than AVP, such as GABA and prokineticin 2 (Antle and Silver, 2005; Masumoto et al., 2006; Welsh et al., 2010). Multiple transmitters in one neuronal type may transmit differential information, as reported in orexigenic AgRP neurons in the hypothalamic arcuate nucleus and wake-stabilizing orexin neurons in the lateral and perifornical hypothalamus

(Krashes et al., 2013; Muschamp et al., 2014; Schöne et al., 2014). Therefore, neurons producing AVP may play a more fundamental role in the circadian pacemaking of the SCN than the AVP molecule does. This hypothesis was testable by genetically manipulating AVP neurons using the Cre-loxP system. When Bmal1, an essential transcription factor of cellular clocks (**Figure 1B**; Bunger et al., 2000), was deleted specifically in AVP neurons (Avp-Bmal1−/<sup>−</sup> mice), mice demonstrated a significant impairment of the locomotor activity rhythm in DD (**Figure 1C**; Mieda et al., 2015). When released into DD, the interval between the activity onset and offset (activity time) gradually became expanded by approximately 5 h as compared with that in controls. Their free-running period was approximately 50 min longer on average than that of control mice. Furthermore, a small number of Avp-Bmal1−/<sup>−</sup> mice even demonstrated arrhythmicity. Importantly, Bmal1 restoration in AVP neurons of the SCN with the aid of a recombinant AAV vector reversed the circadian impairment of Avp-Bmal1−/<sup>−</sup> mice almost completely. These results indicate that the cellular circadian oscillation persists, but the mutual coupling between clock neurons regulating the onset and offset components of activity may be severely impaired in the SCN of Avp-Bmal1−/<sup>−</sup> mice.

In these mice, the circadian expression of factors involved in intercellular communications, including Avp, Prokineticin 2, and Rgs16, was considerably decreased in the SCN shell, where AVP neurons are located. In SCN explants, PER2::LUC oscillations in the shell cells were attenuated with highly variable and lengthened periods. Collectively, Bmal1-based cellular clocks of AVP neurons are likely to enhance the coupling of the SCN cells to generate robust circadian rhythms by regulating expression of multiple factors involved in interneuronal communications (Mieda et al., 2015).

### AVP NEURONS ARE INVOLVED IN THE CIRCADIAN PERIOD DETERMINATION AT THE NETWORK LEVEL

By artificially manipulating the period of cellular clocks specifically in AVP neurons, the possibility that AVP neurons actively work as pacemaker cells to determine the period of circadian rhythm generated by the SCN network was examined (Mieda et al., 2016). It has been shown that the phosphorylation of PER proteins by casein kinase 1δ (CK1δ) regulates the speed of cellular clocks (**Figure 1B**; Herzog et al., 2017). Artificial lengthening of the cellular circadian period specifically in AVP neurons, achieved by deleting CK1δ in AVP neurons (Avp-CK1δ <sup>−</sup>/−), also lengthened the free-running period of behavior rhythm by approximately 50 min, while their activity time remained normal (**Figure 1C** and **Table 1**). Conversely, artificial shortening of the AVP neuronal circadian period, archived by overexpression of CK1δ1 in these neurons via focal injection of a Cre-dependent AAV expression vector, shortened the period of behavior rhythm (**Figure 1C**). Thus, the manipulation of CK1δ expression levels in AVP neurons of the


SCN bidirectionally changed the free-running period of behavior rhythm, suggesting that AVP neurons do indeed regulate SCN pacemaking (Mieda et al., 2016).

How much do AVP neurons contribute to the period determination? Mice in which CK1δ was deleted in the entire SCN, using GABAergic neuron-specific Vgat-Cre driver mice, also showed a lengthened free-running period of behavior rhythm by approximately 40 min (**Table 1**; van der Vinne et al., 2018), which was comparable to that in Avp-CK1δ −/− mice. These data indicate that AVP neurons are the principal determinant of circadian period generated by the SCN network in vivo.

Because of the coherently lengthened free-running period of Avp-CK1δ <sup>−</sup>/<sup>−</sup> mice (**Figure 1C**), the cellular clocks (PER2::LUC oscillations) of the entire SCN were also expected to oscillate with a longer period in slices. Contrary to such an expectation, however, the SCN shell and core of Avp-CK1δ <sup>−</sup>/−; Per2::Luc mice transiently demonstrated different cellular periods in explants (Mieda et al., 2016). The period of the shell was longer, but this lengthening did not last into the subsequent cycles. These data suggest that the core modulated the shell in the prolonged SCN culture. Indeed, the lengthening of shell's period in Avp-CK1δ <sup>−</sup>/−; Per2::Luc mice lasted for a longer duration when slices were surgically cut between the shell and core. A similar dissociation of behavior rhythm and PER2::LUC rhythm has also been observed in Avp-Bmal1−/<sup>−</sup> mice (Mieda et al., 2015). Thus, the intact structure of the SCN and/or its connections with other brain areas might be important for the coupling between SCN shell and core in vivo.

In rodents, core neurons communicate with those in the shell, while there is less communication in the reverse direction (Leak et al., 1999). A recent study of mouse SCN connectome reported that, although the direct connection from AVP neurons to VIP neurons is extremely sparse, AVP neurons make plenty of contacts onto other types of neurons in the SCN core, such as GRP neurons, raising the possibility that AVP neurons communicate well with VIP neurons indirectly via those non-VIP core neurons (Varadarajan et al., 2018). Such asymmetric anatomical interactions between the SCN core and shell may make shell-to-core interaction more fragile in slices.

### THE ROLES OF AVP NEURONS IN THE SCN NETWORK

Lee et al. (2015) demonstrated that lengthening the cellular circadian period of NMS-producing SCN neurons by overexpression of Clock∆<sup>19</sup> lengthened the free-running period of behavior (**Table 1**). Nevertheless, NMS neurons are still a heterogeneous population that contains AVP neurons, VIP neurons, and other types of neurons (Lee et al., 2015), leaving the long-standing debate on the differential roles of the shell and core of the SCN on its pacemaking unresolved.

More recently, Smyllie et al. (2016) created chimeric mice by crossing Drd1a-Cre mice to floxed CK1 Tau/Tau mice, whose SCN contained dopamine 1a receptor (Drd1a) cells (CK1 <sup>−</sup>/<sup>−</sup> cells) with an intrinsic cellular period of 24 h alongside non-Drd1a cells

 a

fnins-13-00139 February 21, 2019 Time: 17:44 # 4

(CK1 Tau/Tau cells) with a period of 20 h (**Table 1**). Remarkably, 60% of these mice showed 24 h periods of behavior and SCN PER2::LUC rhythms, whereas 33% showed 20 h periods. Drd1a cells contain 63% of all SCN cells, including 62% of AVP neurons and 81% of VIP neurons (Smyllie et al., 2016). The fact that the behavioral period did not necessarily follow the cellular period of 80% of VIP neurons is consistent with the earlier finding by Lee et al. (2015) that lengthening the VIP neuronal cellular period had no effect on the behavioral free-running period. Collectively, these observations suggest that VIP neurons may not be directly involved in the pacemaking of the SCN, although VIP signaling plays a principal role in the synchronization of SCN neurons.

Taken in conjunction with data indicating that AVP neurons are involved in the SCN pacemaking (Mieda et al., 2016), as described earlier, the slight difference in the ratio of 24 h AVP neurons to 20 h AVP neurons in Smyllie et al. (2016) could exert a substantial impact on the period in which the chimeric SCN oscillates. In addition, the extent of lengthening in free-running period in mice lacking CK1δ in the entire SCN is comparable to that in mice with AVP neuron-specific CK1δ deletion (van der Vinne et al., 2018), as discussed earlier. These observations suggest that AVP neurons are the primary determinant of the period of circadian rhythm generated by the SCN network. VIP neurons may play a dominant role in the synchronization and phase regulation of SCN neurons, but their contribution in period determination may be little.

Recently, striking contributions of astrocytes of the SCN in the circadian pacemaking was reported (Barca-Mayo et al., 2017; Brancaccio et al., 2017; Tso et al., 2017). SCN astrocytes and neurons are likely to act as two arms of the central circadian pacemaker network, which shows oscillations anti-phasic to each other (Brancaccio et al., 2017). These neuronal and astrocytic oscillators are coupled via glutamate released from astrocytes, which increases presynaptic GABA release and subsequently suppresses neuronal activity of postsynaptic neurons during night. Floxed CK1 Tau/Tau mice that originally had a free-running behavior period of 20 h changed the period to 24 h when CK1 Tau alleles were deleted specifically in SCN astrocytes via viral Cre delivery, suggesting that SCN astrocytes can control the period of circadian behavior rhythms (**Table 1**). Interestingly, the same reversal of free-running period was observed when CK1 Tau alleles were deleted specifically in SCN neurons of the same mice. Therefore, both SCN astrocytes and neurons are equally able to impart timekeeping information to the rest of the body (Brancaccio et al., 2017). However, these results may appear a little strange and difficult to interpret. Although astrocyte- and neuron-specific deletions of CK1 Tau in floxed CK1 Tau/Tau mice resulted in reversed temporal misalignments of the SCN – that is, the 20 h neuronal clock and the 24 h astrocytic clock, and vice versa – the chimeric mice always showed a free-running period of 24 h. One explanation for these observations may be that cellular clocks and the SCN network are optimized to work at 24 h and therefore would be advantaged in the chimeras over the 20 h cells, regardless of which cell type has been targeted (Brancaccio et al., 2017). It would be very interesting to examine whether artificial lengthening (by CK1δ deletion or Clock∆<sup>19</sup> overexpression) or shortening (by CK1δ1 overexpression) of the astrocytic cellular period from 24 h alters the free-running period of behavior rhythm as much as neuronal manipulations do. In any case, comprehensive understanding of the network principle of the SCN central circadian clock needs further study.

### CONCLUDING REMARKS

A previous pioneering study utilized chimera mice of wild type and long-period Clock∆19/∆<sup>19</sup> mutant cells to address the network mechanism of the circadian period determination by the SCN (Low-Zeddies and Takahashi, 2001). In these mice, random subsets of wild type SCN cells were replaced with Clock∆19/∆<sup>19</sup> cells. The proportion of Clock∆19/∆<sup>19</sup> versus wild type cells largely determined circadian behavior in chimeric individuals. However, the intermediate periods were observed in some but not evident in all balanced chimeras. This fact indicates that the emergence of intermediate periods is dependent on not only the proportion but also the distribution of wild type and Clock∆19/∆<sup>19</sup> cells (Low-Zeddies and Takahashi, 2001), suggesting unequal contributions among SCN cells to the period determination. Cell type-specific manipulations of the cellular circadian period described earlier in this review further support such an idea that there exist cells that function as the dominant pacemaking elements in the SCN network, a likely candidate of which may be AVP neurons.

Thus, as the cellular clocks have molecular mechanisms to determine their period, amplitude, and phase within the individual cells, the SCN may have multicellular and network mechanisms to determine the period, amplitude, and phase of the circadian rhythm it generates, which is not a simple summation of multiple cellular clocks. In other words, there exists functional localization within the SCN. The characterization of Avp-Bmal1−/<sup>−</sup> mice and Avp-CK1δ <sup>−</sup>/<sup>−</sup> mice definitively demonstrated that cellular clocks of SCN AVP neurons play a critical role in the generation of robust circadian behavior rhythm through the regulation of the coupling of SCN neurons, as well as in the determination of the circadian period. Additional manipulations of cellular clocks and neuronal properties in various combinations of neuron types and geneticengineering techniques would provide further information to comprehensively understand the principle of the SCN neural network as the central circadian pacemaker.

# AUTHOR CONTRIBUTIONS

The author confirms being the sole contributor of this work and has approved it for publication.

# FUNDING

This work was supported by MEXT/JSPS KAKENHI Grant Numbers JP16H05120, JP18H04941, JP18K19421, and JP18H04972.

### REFERENCES

fnins-13-00139 February 21, 2019 Time: 17:44 # 6


with rapid induction of the mPer1 transcript. Cell 91, 1043–1053. doi: 10.1016/ S0092-8674(00)80494-8


**Conflict of Interest Statement:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Mieda. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

fnins-13-00139 February 21, 2019 Time: 17:44 # 7

# Dorsal Striatum Dopamine Levels Fluctuate Across the Sleep–Wake Cycle and Respond to Salient Stimuli in Mice

### Hui Dong<sup>1</sup> , Juan Wang<sup>1</sup> , Yan-Fei Yang<sup>1</sup> , Yan Shen1,2, Wei-Min Qu<sup>1</sup> \* and Zhi-Li Huang<sup>1</sup> \*

<sup>1</sup> Department of Pharmacology, School of Basic Medical Sciences, State Key Laboratory of Medical Neurobiology and MOE Frontiers Center for Brain Science, Institutes of Brain Science, Fudan University, Shanghai, China, <sup>2</sup> Department of Neurology and National Research Center for Aging and Medicine, Huashan Hospital, Fudan University, Shanghai, China

### Edited by:

Ada Eban-Rothschild, University of Michigan, United States

### Reviewed by:

Christian R. Burgess, Michigan Medicine, University of Michigan, United States Jimmy Fraigne, University of Toronto, Canada

### \*Correspondence:

Wei-Min Qu quweimin@fudan.edu.cn Zhi-Li Huang huangzl@fudan.edu.cn

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 08 November 2018 Accepted: 01 March 2019 Published: 19 March 2019

### Citation:

Dong H, Wang J, Yang Y-F, Shen Y, Qu W-M and Huang Z-L (2019) Dorsal Striatum Dopamine Levels Fluctuate Across the Sleep–Wake Cycle and Respond to Salient Stimuli in Mice. Front. Neurosci. 13:242. doi: 10.3389/fnins.2019.00242 Dopamine is involved in numerous neurological processes, and its deficiency has been implicated in Parkinson's disease, whose patients suffer from severe sleep disorders. Destruction of nigrostriatal dopaminergic neurons or dorsal striatum disrupts the sleep– wake cycle. However, whether striatal dopamine levels correlate with vigilance states still remains to be elucidated. Here, we employed an intensity-based genetically encoded dopamine indicator, dLight1.1, to track striatal dopamine levels across the spontaneous sleep–wake cycle and the dopaminergic response to external stimuli. We found that the striatal dLight1.1 signal was at its highest during wakefulness, lower during non-rapid eye movement (non-REM or NREM) sleep, and lowest during REM sleep. Moreover, the striatal dLight1.1 signal increased significantly during NREM sleep-to-wake transitions, while it decreased during wake-to-NREM sleep transitions. Furthermore, different external stimuli, such as sudden door-opening of the home cage or cage-change to a new environment, caused striatal dopamine release, whereas an unexpected auditory tone did not. Finally, despite both modafinil and caffeine being wake-promoting agents that increased wakefulness, modafinil increased striatal dopamine levels while caffeine did not. Taken together, our findings demonstrated that striatal dopamine levels correlated with the spontaneous sleep–wake cycle and responded to specific external stimuli as well as the stimulant modafinil.

### Keywords: dopamine, dorsal striatum, sleep–wake, dLight, modafinil

# INTRODUCTION

Dopamine is involved in numerous behavioral and psychological processes, including motor behavior, attention, motivation, reward, and feeding (Palmiter, 2007; Berke, 2018), all of which operate on the basis of wakefulness (Lazarus et al., 2012, 2013). Dysregulation of the striatum and nigrostriatal dopamine are considered to be responsible for Parkinson's disease (PD). Patients with PD have been reported to suffer from severe sleep disorders including insomnia, sleep fragmentation, excessive daytime sleepiness (EDS), and rapid eye movement (REM) sleep behavior

disorders (Adler and Thorpy, 2005). Lesioning the dorsal striatum decreases and destabilizes wakefulness in rats (Qiu et al., 2010). The dorsal striatum expresses dopamine D<sup>1</sup> and D<sup>2</sup> receptors (D1Rs, D2Rs) at high levels (Weiner et al., 1991; Levey et al., 1993). D1R and D2R agonists have been shown to dramatically promote wakefulness (Ongini et al., 1985; Monti et al., 1989). Moreover, our previous study showed that genetic deletion of D2Rs significantly decreases wakefulness in mice (Qu et al., 2010). These findings suggest that nigrostriatal dopamine is crucial for wakefulness.

The striatum receives dense dopaminergic inputs from the substantia nigra pars compacta (SNc), and partially from the ventral tegmental area (VTA) and dorsal raphe nucleus (DRN) (Stratford and Wirtshafter, 1990; Bjorklund and Dunnett, 2007; Wall et al., 2013; Poulin et al., 2018). Recent evidence reveals that dopaminergic neurons in the SNc, VTA, and DRN are pivotal for the initiation and maintenance of wakefulness (Eban-Rothschild et al., 2016; Cho et al., 2017; Oishi et al., 2017a; Yang et al., 2018). Optogenetic or chemogenetic stimulation of dopaminergic neurons in the SNc, VTA, or DRN induces robust wakefulness (Eban-Rothschild et al., 2016; Cho et al., 2017; Oishi et al., 2017a; Yang et al., 2018). The calcium activity of dopaminergic neurons is demonstrated to be high during wakefulness and correlates with state transitions (Eban-Rothschild et al., 2016; Cho et al., 2017). However, previous studies showed that dopaminergic neurons in the SNc and VTA not only release dopamine but also co-release either glutamate or γ-aminobutyric acid (GABA) (Chuhma et al., 2004; Hnasko et al., 2010; Tritsch et al., 2012; Kim et al., 2015). In addition, activation of dopaminergic fibers in striatal slices rapidly inhibits the action potential firing of striatal medium spiny neurons (MSNs) via the release of the inhibitory transmitter GABA (Tritsch et al., 2012). Early electrophysiological findings suggest that dopaminergic neurons in the VTA and SNc do not change their mean firing rate and pattern across sleep– wake states in rats and cats (Trulson et al., 1981; Miller et al., 1983; Steinfels et al., 1983; Trulson and Preussler, 1984). The lesion of VTA and SNc dopaminergic neurons in cats results in a lack of behavioral arousal but not the alteration of electrocortical waking (Jones et al., 1973). Despite numerous studies devoted to how dopaminergic neurons and dopamine receptors are vital for wakefulness, the field still lacks straightforward and detailed evidence to support that dopamine itself plays a role in the sleep–wake cycle. To address this question, methods with high temporal resolution are needed to monitor the variation of striatal dopamine levels across the sleep–wake cycle.

Classical analytical approaches such as intracerebral microdialysis and electro-chemical voltammetry have been used for the quantitative measurement of extracellular dopamine concentrations, but they provide poor temporal resolution. Using intracerebral microdialysis with a 2-min temporal resolution, a previous study found that dopamine concentrations in the nucleus accumbens (NAc) and prefrontal cortex (PFC) are higher during both the awake state and REM sleep compared to non-REM (NREM) sleep in rats (Lena et al., 2005). Another study using voltammetry at a 5-min resolution showed that the striatal dopamine voltammetric peak is higher in cats while awake than asleep (Trulson, 1985). In addition, extracellular dopamine levels in mouse striatal slices oscillates across the light/dark cycle (Ferris et al., 2014). The above methods have provided useful insights about the release of dopamine transmitter, but poor temporal resolution in freely moving animals still presents a significant limitation. Recently, Patriarchi et al. (2018) engineered a genetically encoded fluorescent dopamine sensor, dLight1.1, which is capable of tracking dopamine transients with high temporal resolution in freely moving animals. The dLight1.1 sensor is developed by replacing the third intracellular loop on D1R with a circularly permuted GFP (cpGFP) and permits the tracking of dopamine levels by detecting cpGFP fluorescence without activating D1Rs signaling cascades downstream.

In our current study, we employed an optimized variant of this dopamine sensor called dLight1.1, which is suitable for in vivo studies. We detected the dLight1.1 fluorescent signals using fiber photometry, while simultaneously collecting polysomnographic recordings in freely behaving mice after environmental or pharmacological manipulations. We found that striatal dopamine levels were at their highest during wakefulness, lower during NREM sleep, and lowest during REM sleep. We also revealed that striatal dopamine levels were correlated with sleep-state transitions. Furthermore, dopamine levels were enhanced in the striatum following the sudden opening of the home-cage door but did not respond to a high-frequency auditory stimulus whether asleep or awake. Moving the mice from their home cage to a new cage also caused striatal dopamine release. Finally, the wake-promoting agent modafinil, but not caffeine, induced the release of striatal dopamine. Taken together, our results provided strong evidence that striatal dopamine levels correlated with wakefulness and could respond to defined stimuli and stimulants.

# MATERIALS AND METHODS

# Ethics Statement

This study was carried out in accordance with the principles of China Regulations on the Administration of Laboratory Animals, the Decree NO.2 of National Science and Technology Commission of the People's Republic of China. The protocol was approved by the Committee on the Ethics of Animal Experiments of Fudan University (permit number: 20140226-024).

### Animals

Male, specific pathogen-free (SPF), inbred C57BL/6 mice (10– 14 weeks old weighing 20–25 g) were obtained from the Shanghai Laboratory Animal Center, Chinese Academy of Sciences (SLAC, Shanghai, China). The mice were housed at a constant temperature (22 ± 0.5◦C and humidity (55 ± 5%), under an automatically controlled 12/12 h light/dark cycle (lights on at 7:00 a.m., illumination intensity ≈ 100 lux) (Zhang et al., 2017). Food and water were available ad libitum. Every effort was made to minimize animal suffering, and the minimum number of animals required to generate reliable scientific data was used.

# Virus Preparation

fnins-13-00242 March 15, 2019 Time: 16:15 # 3

The adeno-associated virus (AAV) plasmid pAAV-CAGdLight1.1 was a gift from Lin Tian (Addgene plasmid # 111067) (Patriarchi et al., 2018). A recombinant AAV vector carrying the dLight1.1 element (AAV-CAG-dLight1.1) was serotyped with AAV9 coat proteins and packaged by Taitool Bioscience Company (Shanghai, China). The final viral concentration was 5 × 10<sup>12</sup> genome copies per mL. Aliquots of virus were stored at −80◦C until stereotaxic injection.

# Viral Microinjection and Optical-Fiber Cannula Implantation

Adult mice were anesthetized with pentobarbital (intraperitoneal, 80 mg/kg) and 1% lidocaine hydrochloride (subcutaneous, under the scalp). After shaving the fur on the head and sterilizing the skin with 75% ethanol, the mice were placed on a stereotaxic frame (RWD Life Science, China). The skull surface was cleaned with sterile saline on a sterilized cotton swab. Small craniotomy burr holes were made and 100 nL of the AAV-CAG-dLight1.1 virus was unilaterally microinjected through a fine glass pipette into the dorsal striatum (anteroposterior (AP): 0.80 mm, mediolateral (ML): +1.5 mm, dorsoventral (DV): −2.5 mm), according to the Allen Mouse Brain Atlas (Dong, 2008). The virus injection was administered over a 5-min period using nitrogen-gas pulses of 20 psi delivered through an air compression system (Picospritzer III, Parker Hannifin Corp.) as previously described (Yuan et al., 2017; Luo et al., 2018). At the end of the infusion, the pipette was kept in situ for at least 5 min and then withdrawn slowly. After injections, the mice used for in vivo fiber photometry experiments were implanted with an optical fiber cannula (Fiber core 200 µm, 0.37 numerical aperture (NA), Newdoon, China) into the dorsal striatum. The fiber cannula was implanted 0.2 mm above the virus injection site. After injection and implantation, the mice were placed on a heating pad for post-operative recovery. Mice were housed for at least 2 weeks after injections for complete recovery and to allow viral expression prior to any experiments.

# Electroencephalogram/Electromyogram (EEG/EMG) Electrode Implantation

As previously described (Yuan et al., 2017), the EEG/EMG electrode consists of two stainless steel screws with wire leads for EEG recording and two Teflon-coated stainless-steel wires (Cooner Wire, United States) for EMG recording. To implant the electrode, two small craniotomy holes were made in the frontal (AP: +1.5 mm, ML: −0.7 mm) and parietal (AP: −1.5 mm, ML: −1.0 mm) regions with a cranial drill. The EEG electrodes were screwed into the craniotomy holes and the EMG wires were bilaterally placed into the trapezius muscles. All the electrodes were attached to a mini-connector and fixed to the skull with dental cement.

# Polysomnographic Recording and Analysis

After 2 weeks of post-operative recovery, each animal was connected to an EEG/EMG recording cable in a recording apparatus (transparent barrel) and habituated for 3 days before polysomnographic recordings were conducted. The uninterrupted, synchronous recordings of EEG and EMG were performed by means of a slip ring, which was designed to let the mice move freely. Cortical EEG and neck EMG signals were amplified and filtered (Biotex Kyoto, Japan. EEG, 0.5– 30 Hz; EMG, 20–200 Hz), digitized at a sampling rate of 512 Hz, and recorded by a Power 1401 digitizer and Spike2 software (CED, Cambridge, United Kingdom). The Spike2 data were then converted to text format for the analysis of vigilance states using SleepSign software (Kissei Comtec, Nagano, Japan). After the experiment was completed, the EEG/EMG data were automatically classified off-line using 4 s epochs for wakefulness, REM sleep, and NREM sleep using SleepSign software according to standard criteria (Huang et al., 2005). These automatically defined classifications were checked manually and corrected if necessary. Wakefulness was defined as periods of desynchronized, low-amplitude EEG and heightened EMG activity with phasic bursts; NREM sleep was defined as periods of synchronized, high-amplitude, low-frequency (delta band: 0.5–4 Hz) EEG and low EMG activity (compared with wakefulness) without phasic bursts; REM sleep was defined as periods with a pronounced theta rhythm (6–10 Hz) and no EMG activity.

### Fiber Photometry

Following the 2-week recovery period from the virus injection and implantation surgery, dLight1.1 fluorescence emission was recorded with a fiber photometry system (Thinkerbiotech, Nanjing, China) using methods similar to previous studies (Li et al., 2016; Luo et al., 2018). The fiber photometry was performed at 8:00–18:00. Briefly, to record fluorescent signals, the beam from a 488-nm laser (OBIS 488LS, Coherent, United States) was reflected by a dichroic mirror (MD498; Thorlabs), focused by a 10× objective lens (NA = 0.3, Olympus), and then coupled to an optical commutator (Doric Lenses, Canada). An optical fiber (230 mm optical density [O.D.], NA = 0.37, 1 m long) guided the light between the commutator and the implanted optical fiber. The laser power was adjusted at the tip of the optical fiber to a low level of 10–20 µW, to minimize bleaching. The dLight1.1 fluorescent signal was bandpass-filtered (MF525- 39, Thorlabs) and collected by a photomultiplier tube (R3896, Hamamatsu). An amplifier (C7319, Hamamatsu) was used to convert the photomultiplier tube current output into voltage signals, which was further filtered through a low-pass filter (40 Hz cut-off; Brownlee 440). The photometry analog voltage signals were digitalized at 512 Hz and recorded by a Power 1401 digitizer and Spike2 software (CED, Cambridge, United Kingdom) simultaneously with polysomnographic recordings.

Photometry data were analyzed by customized Matlab software (Matlab, 2016a, MathWorks, United States) as described in our previous study (Luo et al., 2018). In brief, the photometry data were exported from Spike2 software in Matlab format for further analysis. The signal data were smoothed with a moving average filter (0.2 s span). For each session, the photometry signal F was converted to 1F/F by calculating 1F/F = (F − Fmean)/Fmean, where Fmean is the average

fluorescence in recording episode. We recorded data for 4–10 h per session and calculated the averaged 1F/F during periods of wakefulness, NREM, and REM sleep. For the analysis sleep-state transitions, we identified each state transition and aligned 1F/F with a ±60 s window before and after the switch point. For stimuli analysis, the photometry signal was aligned with a ±20 s window before and after the event onset. For the modafinil, caffeine, and cage-change experiments, we recorded signals for 6 h (from 1 h before to 5 h after the administration of drugs or the cage change) and calculated the averaged 1F/F value pre- and post-treatment.

### Auditory Tone and Door-Opening Test

To examine whether striatal dopamine levels respond to external stimuli, a high-frequency auditory tonal stimulus (70 dB, 2–4 kHz, 10 s duration) or a sudden door-opening stimulus was applied to mice during NREM sleep or wake periods as previously described (Cho et al., 2017). The loudspeaker was placed on top of the recording cage about 50 cm above the mouse and the intensity of the auditory tone inside the cage was calibrated with a sound meter (Uni-Trend UT350, Dongguan, China). The auditory stimulus and door-opening were both performed suddenly (without warning) when the mouse was either asleep or awake. Then the mice were allowed to rest without disturbance for 5–10 min before the next stimulation. Each type of stimulus was repeated at least three times for each mouse.

### Pharmacological Treatments

One hour following the onset of the photometry recording, modafinil (Sigma-Aldrich, United States) was dissolved in sterile saline containing 10% DMSO and 2% (w/v) cremophor and administered intraperitoneally at doses of 45 and 90 mg/kg. Caffeine (Alfa Aesar, United Kingdom) was dissolved in sterile saline and given intraperitoneally at a dose of 15 mg/kg. Both drugs were prepared fresh, immediately before use.

# Histology

Histological verification of viral expression was performed as described previously (Luo et al., 2018). After all the experiments were completed, the mice were deeply anesthetized with an overdose of pentobarbital and transcardially perfused with phosphate-buffered saline (PBS) followed by 4% paraformaldehyde (PFA) in PBS. Then, the brains were post-fixed in 4% PFA in 0.1 M phosphate buffer (PB; pH 7.4) for 6 h. Next, the brains were then transferred to 20% sucrose in PBS until they sank to the bottom, followed by an incubation in 30% sucrose until they sank to the bottom. Then, the tissue was embedded in optimum cutting temperature (OCT) compound, frozen, and coronal sections were cut at 30 µm by a freezing microtome (Leica, Germany). Since dLight1.1 cannot be detected directly by native fluorescence, a further immunohistochemistry was required (Patriarchi et al., 2018). The brain slices were washed in PBS and incubated in chicken anti-GFP primary antibody (1:5000 dilution; GFP-1020, Aves Labs, United States) at 4◦C overnight. The next day, the sections were incubated in Alexa Fluor 488-conjugated donkey anti-chicken secondary antibody (1:1000 dilution, Cat. # 703-545-155, Jackson ImmunoResearch, United States) for 2 h at room temperature. Finally, slices were washed in PBS and mounted on glass slides using DAPI Fluoromount-G (Southern Biotech, Cat. # 0100–20). Images were captured by a fluorescence microscope (IX71, Olympus).

### Statistical Analyses

Data are presented as the means ± SEM. Paired or unpaired Student's t-tests were used for two-group comparisons and oneway analysis of variance (ANOVA) was used for multiple-group comparisons. A two-way ANOVA was used to analyze the experiments with modafinil and caffeine treatment. Following the ANOVA, Sidak or Bonferroni's post hoc tests were used to make pairwise comparisons. All the statistical tests were two-tailed and P-values less than 0.05 were considered significant. All the statistical analyses were performed using Prism 7.0 (GraphPad Software, United States) and MATLAB R2016a software.

# RESULTS

### Striatal Dopamine Levels Across Spontaneous Sleep–Wake Cycle

Patriarchi et al. (2018) showed that the dLight1.1 plasmid was silent in the absence of dopamine. When dopamine is released from presynaptic terminals, it binds to the dLight1.1 sensor and dramatically increases its fluorescence (**Figure 1A**). In order to ensure the efficient and precise expression of dLight1.1 in the dorsal striatum, an AAV encoding dLight1.1 under the control of a CAG promoter was unilaterally injected into the dorsal striatum of mice (**Figure 1B**). The mice were also chronically implanted with (1) a fiberoptic probe upon the virus-injection site for subsequent delivery of light excitation and collection of dLight1.1 fluorescence and (2) EEG/EMG electrodes for simultaneous polysomnographic recordings (**Figure 1C**). The success of the virus infection and the appropriate location of the fiberoptic implant were verified in each mouse after all experiments were completed. As shown in **Figure 1D** and **Supplementary Figure S1**), dLight1.1 fluorescence was robustly expressed at the injection site in the dorsal striatum.

To examine whether the striatal dopamine levels correlated with distinct vigilance states, we recorded striatal dLight1.1 fluorescent signals across spontaneous sleep–wake cycle. As shown in **Figures 1E,F**, the fluctuations of dLight1.1 fluorescence were correlated with the EEG/EMG signals. To compare the dLight1.1 signal amplitude during distinct vigilance states, the fluorescent signals were averaged in a state-dependent manner. We found that the mean striatal dLight1.1 signal was significantly higher during wakefulness (0.952% ± 0.128%) than during NREM sleep (−0.6% ± 0.114%) or REM sleep (−2.129% ± 0.179%), which exhibited the lowest fluorescence (**Figure 1G**; n = 7 mice; F2,<sup>18</sup> = 116.1, P < 0.01; post hoc Tukey test: wake vs. NREM P < 0.01, wake vs. REM P < 0.01, NREM vs. REM P < 0.01). Although the peak value of REM sleep is higher and the trough value is lower (**Supplementary Figure S3**), the mean fluorescent signal is the lowest in REM sleep. Moreover, Dahan et al. (2007) found that the dopamine neuronal activity during REM sleep showed a pronounced bursting pattern with decreased amplitude. This firing pattern of dopaminergic

FIGURE 1 | Striatal dLight1.1 signals at distinct spontaneous sleep–wake states. (A) Schematic diagram of dLight1.1 with the dopamine D<sup>1</sup> receptor and circularly permuted GFP (cpGFP) module (upper panel) and the working principle of dLight1.1. (B) Schematic showing the injection of AAV-CAG-dLight1.1 into the dorsal striatum. (C) Schematic showing the setup for fiber photometry used to assess dLight1.1 fluorescence with simultaneous polysomnographic recordings. (D) Expression of dLight1.1 in the dorsal striatum. The scale bar is 200 µm. (E,F) Representative EEG, EMG, and fluorescent photometry signal traces of striatal dLight1.1. during distinct sleep–wake states (green, wake; blue, NREM sleep; magenta, REM sleep). (G) Quantification of the average striatal dLight1.1 signal at distinct sleep–wake states. One-way ANOVA: F2,<sup>18</sup> = 116.1, P < 0.0001; Tukey's post hoc test: wake vs. NREM sleep ∗∗P < 0.0001, wake vs. REM sleep ∗∗P < 0.0001, NREM vs. REM sleep ∗∗P < 0.0001; n = 7 mice. (H) Temporal dynamics of the striatal dLight1.1 signal during long-term wake (green), NREM sleep (blue), and REM sleep (magenta) episodes within normalized time. (I) Striatal dLight1.1 signal at the early wake period (first 20% of wake period) and the late wake period (last 20% of wake period) (t<sup>6</sup> = 5.058, ∗∗P = 0.0023, n = 7 mice).

FIGURE 2 | Striatal dLight1.1 signal dynamics across spontaneous sleep-state transitions. (A) Striatal dLight1.1 signals across NREM sleep-to-wake transition. (Left) The time course of the dLight1.1 signal across the NREM sleep-to-wake transition. (Right) The average amplitude of the dLight1.1 signal 60 s pre- and post-transition (t<sup>6</sup> = 5.441, ∗∗P < 0.01, n = 7 mice). (B) Striatal dLight1.1 signal at the wake-to-NREM sleep transition. (Left) The time course of the dLight1.1 signal across the wake-to-NREM sleep transition. (Right) The average amplitude of the dLight1.1 signal 60 s pre- and post-transition (t<sup>6</sup> = 2.528, <sup>∗</sup>P = 0.044). (C) Striatal dLight1.1 signal at the REM sleep-to-wake transition. (Left) The time course of the dLight1.1 signal across the REM sleep-to-wake transition. (Right) The average amplitude of the dLight1.1 signal 60 s pre- and post-transition the (t<sup>6</sup> = 1.053, P = 0.333). (D) Striatal dLight1.1 signal at the NREM-to-REM sleep transition. (Left) The time course of the dLight1.1 signal across the NREM-to-REM sleep transition. (Right) The average amplitude of the dLight1.1 signal 60 s pre- and post-transition (t<sup>6</sup> = 1.509, P = 0.182). (E) Time courses of the striatal dLight1.1 signals across NREM sleep-to-long wake or NREM sleep-to-short wake periods. (F) The change in the dLight1.1 signal after long or short wake periods compared with the 60 s pre-wake period (t<sup>12</sup> = 2.388, <sup>∗</sup>P = 0.034). Data are presented as the mean (black trace) ± SEM (gray shading) in (A–D) and as the mean (long wake period, green; short wake period, gray) ± SEM (shading) in (E).

neurons may be the reason that striatal dopamine level during REM sleep was more divergent with higher peak and lower trough. To examine the temporal dynamics of striatal dopamine during long-term sleep–wake states (duration longer than 30 s), we normalized the variable duration of sleep–wake states to a unit-less time window from 0 (state onset) to 1 (state offset). During the wakefulness period, dLight1.1 fluorescence peaked soon after the onset of wakefulness and gradually attenuated

(**Figure 1H**). We calculated the mean fluorescence during the first 20% and the last 20% of the wake episode. We found that dLight1.1 fluorescence in the early 20% of the wake episode was significantly higher than that in the late 20% (**Figure 1I**; n = 7 mice; t = 5.058, P = 0.0023). These findings demonstrated that striatal dopamine levels not only varied across spontaneous sleep–wake states but also showed dynamic changes within wakefulness episodes. Next, we assessed the striatal dopamine levels during state transitions. We found that the striatal dLight1.1 signal increased significantly during NREM sleep-towake transitions (**Figure 2A**; t = 5.441, P < 0.01), whereas it decreased during wake-to-NREM sleep transitions (**Figure 2B**; t = 2.528, P = 0.044). However, there was no significant dLight1.1 fluorescence change during REM sleep-to-wake transitions and NREM-to-REM sleep transitions (**Figure 2C**, P = 0.333 and **Figure 2D**, P = 0.182). Since the diversity of animal behaviors depend on the duration of the wakefulness episode, we further examined whether striatal dopamine levels fluctuated with the duration of wakefulness episodes; we calculated the dLight1.1 signal at longer wake episodes (duration > 30 s) and brief wake episodes (duration < 30 s). Interestingly, the net growth of dLight1.1 fluorescence was significantly higher when mice were awake for longer periods than for brief wake periods (**Figures 2E,F**; t = 2.388, P = 0.0343). These results indicated that striatal dopamine levels at wake onset were correlated with the duration of the following wake episode. Taken together, these findings demonstrated that striatal dopamine levels were highest during wakefulness and that they fluctuate dynamically across spontaneous state transitions.

### Striatal Dopamine Levels in Response to Acute External Stimuli

To investigate the dynamics of dopamine levels in the dorsal striatum in response to external stimuli, we recorded striatal dLight1.1 fluorescence while simultaneously conducting polysomnographic recordings in animals exposed to diverse salient stimuli and stimulants. The unexpected presentation of an auditory tone stimulus (70 dB, 2–4 kHz, 10 s duration) was employed as previously described (Cho et al., 2017). We exposed the mice to an auditory tone while asleep or awake (**Figures 3A,C**) and observed that the mice were immediately awakened when the tone was applied during the sleep period (**Figure 3A**). However, there were no detectable changes in striatal dLight1.1 fluorescence in response to the auditory stimulus either when mice were asleep or awake (**Figure 3B**: P = 0.260 and **Figure 3D**: P = 0.127). These results indicated that acute, short auditory tone stimuli did not elevate striatal dopamine release. Surprisingly, the striatal dLight1.1 signal ascended when the door of the mouse's home cage was suddenly opened at the end of the trial session. This observation prompted us to systematically investigate whether unexpected door-opening during sleep or wake states induced striatal dLight1.1 signal changes. We discovered that the striatal dLight1.1 signal rapidly increased whenever the home-cage door was opened suddenly during the sleep period (**Figures 3E,F**; t = 12.15, P < 0.01) and the wake period (**Figures 3G,H**; t = 11.18, P < 0.01). In addition, the amplitude of the striatal dLight1.1 signal induced by sudden door-opening was higher than the amplitude during spontaneous awake periods. Although both auditory tone stimulation and dooropening were able to wake up sleeping mice, the door-opening stimulus elevated dopamine release while auditory stimulation did not. The behavioral paradigm door-opening test maybe mix with visual and olfactory stimuli. In order to explore whether visual and olfactory stimuli enhance striatal dopamine tone, predator odor TMT or light flash were employed. We used light flash at 1 Hz for 10 s when mice were sleeping or awaking (**Supplementary Figure S2A**), and found that flash during awaking increased striatal dLight1.1 fluorescence (**Supplementary Figure S2C**, t = 4.486, P = 0.0463), but failed to enhance during sleep (**Supplementary Figure S2B**, t = 0.9181, P = 0.4555). These results indicated that awareness of visual stimuli enhanced the striatal dopamine tone. Application of air or predator odor TMT didn't enhance striatal dLight1.1 fluorescence (air, t = 0.4803, P = 0.6784; TMT, t = 0.2858, P = 0.8019. **Supplementary Figures S2D–F**). These results indicated that striatal dopamine levels responded only to specific acute stimuli.

### Striatal Dopamine Levels in Response to Cage Change

Our previous study showed that dopamine receptors were necessary for arousal when mice were exposed to new environments (Qu et al., 2010; Xu et al., 2014). However, whether exposure to a new environment augmented the striatal dopamine level was still unclear. To address this question, the cage-change model was employed to mimic a new environment (**Figure 4A**). We found that mice exhibited continuous wakefulness for almost 2 h after tail handled followed by moving to a new cage (**Figure 4B**), while mice kept awake for about 30 min after tail handled followed by returning to their home cages, coincident with our previous results (Qu et al., 2010; Xu et al., 2014). Striatal dLight1.1 fluorescence sharply increased when the mice were tail handled, then gradually attenuated to baseline about 2 h after moved to a new cage, whereas quickly decreased to baseline about 30 min after returned to home cages after tail handled (**Figures 4B,C**). We calculated the mean fluorescence 30 min before (serving as the baseline), 30 min and 30–120 min after returned home cage or moved to a new cage, and found that the mean striatal dLight1.1 fluorescence was significantly higher than baseline for post 30 min when mice were tail handled followed by returned to home cage, and but there was no statistical significance between post 30–120 min and baseline (**Figure 4D**. F2,<sup>4</sup> = 7.284, P = 0.0464), However, when mice were tail handled followed by moved to new cage, the mean striatal dLight1.1 fluorescence was significantly higher for post 30 min and the following 90 min than baseline (**Figure 4D**. F2,<sup>12</sup> = 10.1832, P = 0.0026). These findings revealed that moving the mice to home cage or new cage induced wakefulness and enhanced striatal dopamine release which sustained for 30 min for mice returned to home cage and for at least 2 h in mice exposed to a new cage.

FIGURE 3 | Striatal dLight1.1 fluorescence in response to acute stimuli. (A, Left) Schematic showing high-frequency auditory tones applied in the sleep state simultaneously with fiber photometry and EEG/EMG recording. (Right) Example traces of the fluorescence signal, EEG, and EMG before and after the onset of the auditory tone during sleep. (B, Left) The time course of the striatal dLight1.1 signal in response to auditory tones when mice were sleeping. (Right) Average fluorescence before and after onset of the auditory tone (t<sup>6</sup> = 1.243, P = 0.260). (C, Left) Schematic showing auditory tones applied in the awake state simultaneously with fiber photometry and EEG/EMG recording. (Right) Example traces of the fluorescence signal, EEG, and EMG before and after the onset of the auditory tone during the awake state. (D, Left) The time course of the striatal dLight1.1 signal in response to auditory tones when mice were awake. (Right) Average fluorescence before and after onset of the auditory tone (t<sup>6</sup> = 1.771, P = 0.127). (E, Left) Schematic showing the sudden opening of the recording-cage door during the sleep state. (Right) Typical traces of the fluorescence signal, EEG, and EMG before and after the onset of the door-opening stimulus during sleep. (F, Left) The time course of the striatal dLight1.1 signal in response to the door-opening stimulus while the mice were sleeping. (Right) Average fluorescence before and after onset of the door-opening stimulus (t<sup>6</sup> = 12.15, ∗∗P < 0.0001). (G, Left) Schematic showing the sudden opening of the recording-cage door while mice were in the awake state. (Right) Typical traces of the fluorescence signal, EEG, and EMG before and after the onset of the door-opening stimulus during the wake period. (H, Left) The time course of the striatal dLight1.1 signal in response to the door-opening stimulus while the mice were awake. (Right) Average fluorescence before and after onset of the door-opening stimulus (t<sup>6</sup> = 11.18, ∗∗P < 0.0001).

FIGURE 5 | Effects of caffeine (15 mg/kg) and modafinil (45 and 90 mg/kg) on the striatal dLight1.1 fluorescent signal. (A) Typical examples of the striatal dLight1.1 fluorescent signal, EEG, and EMG traces following vehicle (10% DMSO), caffeine, and modafinil administration. (B) Time courses of the striatal dLight1.1 fluorescent signal following vehicle, caffeine, and modafinil administration. (C) Average striatal dLight1.1 fluorescent signal in the 1 h before (gray circle and bar) and 2 h after each administration. Two-way ANOVA between time and stimuli: F1,<sup>20</sup> (time) = 147.5142, P < 0.0001, pre–post comparisons followed by Bonferroni's test: ∗∗P < 0.0001; F3,<sup>20</sup> (stimuli) = 29.5910, P < 0.0001. Stimuli comparisons followed by Sidak's test: ##P < 0.0001).

# Stimulants Induced Striatal Dopamine Release

Stimulants such as caffeine and modafinil are universally used to stay awake and to boost mental performance. Our previous studies have shown that caffeine promoted wakefulness via adenosine A2A receptors (A2ARs) (Huang et al., 2005; Lazarus et al., 2011), whereas modafinil induced wakefulness via D1Rs and D2Rs (Qu et al., 2008). A2ARs were reported to be densely co-expressed with D2Rs in the dorsal striatum (Svenningsson et al., 1999; Lazarus et al., 2012, 2013). however, it is still unknown whether modafinil and caffeine alter striatal dopamine levels. We found that caffeine (15 mg/kg) or modafinil (45, 90 mg/kg) promoted continuous wakefulness for about 2, 3, or 5 h, respectively (**Figure 5A**), consistent with our previous results (Huang et al., 2005; Qu et al., 2008). Administration of modafinil strongly increased striatal dLight1.1 fluorescence, with dLight1.1 signals rapidly reaching a peak and then gradually attenuating (**Figures 5A,B**). For better comparison of the effects of each drug on striatal dopamine tone, we calculated the average fluorescence 1 h before (serving as the baseline) and over the 2 h after each administration. We chose 2 h for comparisons because caffeine 15 mg/kg induced wakefulness for 2 h, although modafinil promoted longer effects. As shown in **Figure 5C**, administration of modafinil significantly enhanced the striatal dLight1.1 fluorescence, but vehicle and caffeine didn't, compared with their respective baseline (F1,<sup>20</sup> = 147.5142, P < 0.0001, pre–post comparisons followed by Bonferroni's test: modafinil 90 mg/kg vs. baseline, P < 0.0001; modafinil 45 mg/kg vs. baseline, P < 0.0001; caffeine vs. baseline, P > 0.9999; vehicle vs. baseline, P > 0.9999). Moreover, the mean dLight1.1 signals for 2 h after administration of modafinil 90 mg/kg was 13.81% ± 1.55%, significantly higher than 6.83% ± 1.27% for modafinil 45 mg/kg, −0.86% ± 0.55% for caffeine 15 mg/kg, and −0.02% ± 0.40% for vehicle. The dLight signal for modafinil at 45 mg/kg was higher than that for caffeine 15 mg/kg or vehicle, but there was no statistical significance between caffeine 15 mg/kg and vehicle (F3,<sup>20</sup> = 29.5910, P < 0.0001. Comparisons followed by Sidak's test: modafinil 90 mg/kg vs. modafinil 45 mg/kg, P < 0.0001; modafinil 45 mg/kg vs. caffeine, P < 0.0001; caffeine vs. vehicle, P = 0.4198). Taken together, these findings indicated that modafinil increased striatal dopamine levels but caffeine did not.

### DISCUSSION

Using a dopamine sensor and simultaneous polysomnographic recordings, we demonstrated that striatal dopamine levels were highest during wakefulness and dopamine fluctuations correlated with spontaneous sleep–wake transitions. Furthermore, we revealed that some external salient stimuli and certain wakepromoting stimulants elicited striatal dopamine release. These findings provide strong evidence that dopamine in the dorsal striatum is important for wakefulness under baseline conditions, induced by cage change or wake-promoting drug modafinil but not caffeine.

The dorsal striatum receives robust dopaminergic inputs from the SNc, as well as some input from the VTA and DRN (Beckstead et al., 1979; Stratford and Wirtshafter, 1990; Poulin et al., 2018). Although the single-unit firing rate of SNc and VTA dopaminergic neurons in cats and rats shows no changes across the stages of sleep or waking (Trulson et al., 1981; Miller et al., 1983; Steinfels et al., 1983), the specific enhancement of VTA and SNc dopaminergic neuron activity by optogenetic or chemogenetic approaches dramatically induces wakefulness (Eban-Rothschild et al., 2016; Oishi et al., 2017a; Yang et al., 2018). Furthermore, our recent work demonstrates that the inhibition of striatal D2R/A2AR-containing neurons, mimicking the action of dopamine on D2Rs, promotes wakefulness (Yuan et al., 2017). Axonal dopamine release not only depends on the firing rate and pattern of dopaminergic neurons but also on the concentration of calcium (Ca2+) (Kawagoe et al., 1992; Chen et al., 2011). Moreover, dopamine release is assumed to reflect a global response to the activity of midbrain dopaminergic neurons at a population level (Rice et al., 2011). Recent photometry data demonstrate that the population-level calcium signal of VTA and DRN dopaminergic neurons are correlated with the sleep– wake cycle (Eban-Rothschild et al., 2016; Cho et al., 2017). Using intracerebral microdialysis, Lena et al. (2005) elaborated that the dopamine concentrations in the NAc, downstream of the VTA, are higher during both wakefulness and REM sleep compared with NREM sleep in rats. Trulson (1985) used voltammetry to measure the release of dopamine in the dorsal striatum of cats across their sleep–wake cycle at a 5-min temporal resolution. During the 45-min recording consisting of consecutive 15-min periods of each sleep stage, the striatal dopamine voltammetric peak decrease from wake to NREM sleep, and from NREM to REM sleep in cats. Consistent with this finding, our present study showed that striatal dopamine levels were at their highest during wakefulness and their lowest during REM sleep. Taking advantage of the high temporal resolution of dLight1.1, we further analyzed the dynamic variation of striatal dopamine during short-term state transitions and found that striatal dopamine increased significantly during NREM sleepto-wake transitions and decreased during wake-to-NREM sleep transitions. Moreover, extracellular dopamine levels in mice striatal slices were reported to oscillate across the light/dark cycle (Ferris et al., 2014). These pieces of evidence suggest that recording the electrophysiological activity of single dopaminergic neurons does not accurately reflect the functional state of the central dopaminergic system. Collectively, the above findings solidly demonstrate that striatal dopamine release correlates with the sleep–wake cycle, despite the fact that dopaminergic neuron firing is uncorrelated with vigilance states.

The dopamine level in the dorsal striatum is a signal for prolonged time in wakefulness and crucial for the maintenance of wakefulness. Chemogenetic or optogenetic activation of SNc, VTA, or DRN dopaminergic neurons induce a long-lasting period of wakefulness (Eban-Rothschild et al., 2016; Cho et al., 2017; Oishi et al., 2017a; Yang et al., 2018). The fluctuation in DRN dopaminergic activity across sleep-to-wake transitions is significantly larger when mice are awake for a longer period than when they are briefly awake (Cho et al., 2017). Consistent

with this, our present study showed that striatal dopamine levels were higher when mice had longer periods of wakefulness (>30 s) than when they had brief periods of wakefulness (<30 s). These findings suggest that the striatal dopamine level can predict the length of the following wake episode. Chemogenetic inhibition of SNc or VTA dopaminergic neurons promotes sleep at the expense of wakefulness (Eban-Rothschild et al., 2016; Yang et al., 2018). Inhibition of dopaminergic neurons can, in theory, reduce striatal dopamine release and then lead to striatal D1R/D2R inactivation. Our previous study revealed that the genetic deletion of D2Rs destabilizes the wake stage and shortens the duration of wakefulness episodes (Qu et al., 2010). What's more, chemogenetic activation of D2Rs-containing neurons in the dorsal striatum promoted sleep (Yuan et al., 2017). Our present study showed that striatal dopamine levels peaked soon after wake onset and gradually reduced during the wake period. These pieces of evidence collectively suggest that decreasing dopamine levels can facilitate sleep initiation and lessens alertness. The multiplicity of arousal systems guarantees diverse behaviors in the normal individual. It has been reported that histaminergic tuberomammillary neurons are crucial for brief wakefulness (Huang et al., 2006). We can conclude from the literature that distinct arousal mechanisms govern different levels or types of alertness.

Our study showed that striatal dopamine levels not only correlated with the spontaneous sleep–wake cycle, but also responded to salient environmental stimuli. In addition to homeostatic and circadian drives as well as emotion, a good sleep also requires a quiet and safe environment (Saper et al., 2005). An unexpected sound or predator invasion can disrupt the quality of sleep (Saper et al., 2005; Eban-Rothschild et al., 2016; Cho et al., 2017). Consistently, the acute auditory tone and sudden dooropening of the home cage immediately awoke mice from NREM sleep. We found that door-opening and light flash induced striatal dopamine release, whereas the auditory tone failed to do so. An early study found that opening the door of a cat's housing chamber or the presence of the experimenter in the cat's field of vision is associated with the bursting activity of singleunit SNc dopaminergic neurons (Steinfels et al., 1983). While dopaminergic neurons fire in a slow, irregular fashion under baseline conditions, resulting in a tonic release of dopamine, they fire in bursts in response to salient environmental stimuli, which lead to phasic increases in dopamine release (Overton and Clark, 1997). Consistent with this, our data showed that more the striatal dopamine was released following exposure to salient stimuli than during spontaneous wakefulness. Mapping the inputs to midbrain dopaminergic neurons may help us understand their different responses to auditory and visual stimuli. Monosynaptic tracing studies demonstrate that SNc and VTA dopaminergic neurons receive dense input from the superior colliculus, a key structure processing visual information, but received hardly any input from the auditory system (Watabe-Uchida et al., 2012; Lerner et al., 2015). Moreover, previous studies have demonstrated that the superior colliculus is necessary to relay short-latency visual information to dopamine-containing regions of the ventral midbrain in rats (Comoli et al., 2003; Dommett et al., 2005). The door-opening stimulus combined both auditory and visual stimuli, making it sufficient to elicit striatal dopamine release. However, an auditory tone as a conditioned stimulus combined with a reward unconditioned stimulus induces a large dopamine release upon repeated cue-reward pairing but not in the first training session (Patriarchi et al., 2018). Collectively, the above findings suggest that striatal dopamine responds to specific stimuli.

Cage-change is a mouse model that mimics the human first-night effect, which can be observed in unfamiliar sleeping environments. We previously found that the genetic deletion or pharmacological blockade of D2Rs (densely expressed in the dorsal striatum) reduce the duration of wake episodes in mice following being moved to a new cage. The plasma corticosterone levels are elevated after cage change, suggesting that cage change or new environment induces an elevating arousal level (Qu et al., 2010; Xu et al., 2014). Eban-Rothschild et al. (2016) found that transferring the mice to a new environment or introducing novel objects to their home space enhance calcium activity in VTA dopaminergic neurons. Moreover, chemogenetic inhibition of VTA dopaminergic neurons prompts nest-building behavior and promotes sleep. Consistently, we found that cage change induced a significant increase in striatal dopaminergic tone. Therefore, we supposed that cage change induced an elevating arousal level with more increases in the striatal dopaminergic tone, suggesting that dopamine tone may relate to arousal level and the dopaminergic system may be a target for treating insomnia caused by environmental stimuli.

In the present study, we employed auditory tone, dooropening, light flash, predator odor, and cage change paradigm combined with striatal dopamine tone recording and found that striatal dopamine responds to specific stimuli. Striatal dopamine activities also are associated with lots of behaviors, such as locomotion, motivation, reward, and stress, all of which operate on the basis of wakefulness. The role of dopamine in motor behavior is extensively concerned. Rapid phasic signal in striatum-targeting dopaminergic axons is associated with triggering and locomotion in mice (Howe and Dombeck, 2016). Large proportion of SNc dopaminergic neurons transiently increased their activities before self-paced movement initiation in mice (da Silva et al., 2018). The activity of VTA dopaminergic neurons are increased during itch-induced scratching behavior in freely moving mice (Yuan et al., 2018). Dopamine is also involved in negative emotion. Intense exteroceptive stimuli, such as an electric shock on the tail or placing animals into an ice-water bath, provoke large and abrupt rises in the striatal dopamine signal (Keller et al., 1983). Dopamine neurons projecting to the anterior striatum display patterns of activity consistent with the reward value, while those projecting to the posterior tail of the striatum are activated by aversive and neutral stimuli, such as unexpected tone and air puff (Menegas et al., 2018). Taken together, striatal dopamine activities are associated with lots of behaviors, operating on the basis of wakefulness.

Modafinil is a wake-promoting drug used to treat daytime sleepiness. Numerous studies have suggested that modafinil promotes wakefulness by acting on the dopaminergic system Consistently, modafinil is found to bind dopamine uptake transporters (DATs) with low affinity (Mignot et al., 1994) and

the deletion of the DAT gene in mice blocks the wake-promoting effects of modafinil (Wisor et al., 2001). We previously found that the blockade of D1Rs and D2Rs abolishes the arousal effects of modafinil (Qu et al., 2008). In addition, modafinil has been reported to enhance extracellular levels of dopamine in the NAc, PFC, and medial hypothalamus of rats (de Saint Hilaire et al., 2001; Murillo-Rodriguez et al., 2007). Moreover, optogenetic stimulation of dopaminergic terminals in the NAc and dorsal striatum induce wakefulness, whereas the same conditions in the PFC fail to induce wakefulness. This result suggests that the NAc and dorsal striatum could be specific targets of modafinil. In line with this, our present study found that modafinil robustly raised striatal dopamine levels. Another widely used stimulant, caffeine, is a psychoactive compound that is found to promote wakefulness via A2ARs (Huang et al., 2005). A2ARs are densely co-expressed with D2Rs in the striatum (Schiffmann et al., 1991). Previous study revealed that the genetic deletion of striatal A2ARs abolishes arousal effect of caffeine (Huang et al., 2005; Lazarus et al., 2011). Chemogenetic inhibition of dorsal or ventral striatal A2AR positive neurons promote arousal, that mimic arousal effects of caffeine (Oishi et al., 2017b; Yuan et al., 2017). The external globus pallidus mediates the effect of dorsal striatal A2AR positive neurons on sleep, while ventral pallidum, but not VTA, mediates the effect of ventral striatal A2AR positive neurons on sleep. Our current study showed that caffeine did not enhance striatal dopamine levels. These results are consistent with previous studies that caffeine doesn't increase the c-fos expression in the SNc (Bennett and Semba, 1998). The differential effects of modafinil and caffeine on striatal dopamine levels suggest that despite them both being wake-promoting compounds that target the basal ganglia, their arousal effects have different underlying mechanisms, dopaminergic system for modafinil and adenosine system for caffeine. Patients with PD suffer from severe EDS and nigrostriatal dopamine deficiency has been proposed to be responsible for PD (Adler and Thorpy, 2005). In fact, most PD therapeutic agents act by increasing dopaminergic activity. In this study, we found that modafinil dramatically elicited striatal dopamine release. Hence, we propose that modafinil may be a potential agent to treat EDS in PD patients with motor symptoms. In addition, the adenosine system, especially the A2AR, has emerged as an attractive non-dopaminergic target in the pursuit of improved therapy for PD (Antonini and Poewe, 2014). Our study showed that caffeine, a non-specific antagonist of adenosine receptors, did not increase striatal dopamine, suggesting that caffeine promotes arousal but does not depend on dopaminergic systems. Moreover, large clinic studies showed that caffeine or coffee consumption has been associated with a reduced risk of PD (Ross et al., 2000; Ascherio et al., 2001). Hence, we propose that caffeine or A2AR antagonism could be a prospective agent for EDS therapy in PD.

Pharmacological, genetic, and clinical studies have demonstrated that striatal dopamine is involved in numerous behavioral and psychological processes that operate on the basis of wakefulness, including motor behaviors, attention, motivation, reward, and feeding. Dysregulation of nigrostriatal dopamine results in severe neurological disorders such as PD and Huntington's disease with diversified sleep disturbances. Our study demonstrated that striatal dopamine levels fluctuated across the spontaneous sleep–wake cycle and responded to external stimuli and wake-promoting stimulants. By understanding the dynamics of striatal dopamine under various conditions, our findings provide insight into the role of striatal dopamine in sleep regulation and suggest a potential treatment alternative for sleep disturbances in PD.

### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the **Supplementary Files**.

# AUTHOR CONTRIBUTIONS

HD, JW, Y-FY, W-MQ, and Z-LH: conceptualization. HD and Y-FY: methodology, investigation, and formal analysis. HD, Y-FY, YS, W-MQ, and Z-LH: writing original draft. HD, JW, Y-FY, YS, W-MQ, and Z-LH: revised the manuscript. W-MQ and Z-LH: supervision and funding acquisition.

# FUNDING

This work was supported by the National Basic Research Program of China (Grant No. 2015CB856401 to Z-LH) and the National Natural Science Foundation of China (Grant Nos. 31530035 and 81420108015 to Z-LH, Grant Nos. 31871072, 31671099, and 31471064 to W-MQ).

# ACKNOWLEDGMENTS

We thank Dr. Ling Gong of Fudan University for technical assistance.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins.2019. 00242/full#supplementary-material

FIGURE S1 | Expression of dLight1.1 in the dorsal striatum in all mice except 2# which is in Figure 1D. The blue is DAPI, the green is dLight1.1. The scale bar is 200 µm.

FIGURE S2 | Striatal dLight1.1 fluorescence in response to light flash and predator odor. (A,D) Schematic showing light flash (A) or predator odor (D) was applied simultaneously with fiber photometry. (B) Up panel: Example traces of the fluorescence signal, EEG, and EMG before and after the onset of the light flash during sleeping. Down panel: (Left) The time course of the striatal dLight1.1 signal in response to light flash when mice were sleeping. (Right) Average fluorescence before and after onset of the light flash (t = 0.9181, P = 0.4555). (C) Up panel: Example traces of the fluorescence signal, EEG, and EMG before and after the onset of the light flash during awake. Down panel: (Left) The time course of the striatal dLight1.1 signal in response to light flash when mice were awake. (Right)

Average fluorescence before and after onset of the light flash (t = 4.486, P = 0.0463). (E) Up panel: example traces of the fluorescence signal, EEG, and EMG before and after the onset of air release during awake. Down panel: (Left) The time course of the striatal dLight1.1 signal in response to air delivery. (Right) Average fluorescence before and after onset of the air delivery (t = 0.4803, P = 0.6784). (F) Up panel: example traces of the fluorescence signal, EEG, and EMG before and after the onset of TMT delivery during awake. Down panel: (Left) The time course of the striatal dLight1.1 signal in response to TMT delivery. (Right) Average fluorescence before and after onset of the TMT delivery (t = 0.2858, P = 0.8019).

### REFERENCES


FIGURE S3 | The distributions of dLight fluorescence values in each state. The red presents REM sleep, the blue for NREM sleep, and the green for wake. The fluorescence in REM sleep was more divergent, but more convergent in NREM sleep. There was 3.14% dLight fluorescence values in REM sleep higher than 0.1, 10.26% values higher than 0.05, 71.48% values lower than 0, and 7.51% values lower than −0.1, whereas there was 0.57% dLight fluorescence values in NREM sleep higher than 0.1, 4.42% values higher than 0.05, 62.18% values lower than 0, 0.04% values lower −0.1. Although it looks more higher in REM sleep because there a few higher values, the mean of dLight fluorescence was the lowest in REM sleep.

extracellular dopamine tone. Proc. Natl. Acad. Sci. U.S.A. 111, E2751–E2759. doi: 10.1073/pnas.1407935111



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Dong, Wang, Yang, Shen, Qu and Huang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Acute Social Defeat Stress Increases Sleep in Mice

Shinya Fujii<sup>1</sup> , Mahesh K. Kaushik<sup>1</sup> , Xuzhao Zhou<sup>1</sup> , Mustafa Korkutata1,2 and Michael Lazarus<sup>1</sup> \*

1 International Institute for Integrative Sleep Medicine (WPI-IIIS), University of Tsukuba, Tsukuba, Japan, <sup>2</sup> Ph.D. Program in Human Biology, School of Integrative and Global Majors, University of Tsukuba, Tsukuba, Japan

Social conflict is a major source of stress in humans. Animals also experience social conflicts and cope with them by stress responses that facilitate arousal and activate sympathetic and neuroendocrine systems. The effect of acute social defeat (SoD) stress on the sleep/wake behavior of mice has been reported in several models based on a resident-intruder paradigm. However, the post-SoD stress sleep/wake effects vary between the studies and the contribution of specific effects in response to SoD or nonspecific effects of the SoD procedure (e.g., sleep deprivation) is not well established. In this study, we established a mouse model of acute SoD stress based on strong aggressive mouse behavior toward unfamiliar intruders. In our model, we prevented severe attacks of resident mice on submissive intruder mice to minimize behavioral variations during SoD. In response to SoD, slow-wave sleep (SWS) strongly increased during 9 h. Although some sleep changes after SoD stress can be attributed to nonspecific effects of the SoD procedure, most of the SWS increase is likely a specific response to SoD. Slow-wave activity was only enhanced for a short period after SoD and dissipated long before the SWS returned to baseline. Moreover, SoD evoked a strong corticosterone response that may indicate a high stress level in the intruder mice after SoD. Our SoD model may be useful for studying the mechanisms and functions of sleep in response to social stress.

### Edited by:

Chiara Cirelli, University of Wisconsin–Madison, United States

### Reviewed by:

Janne Grønli, University of Bergen, Norway Peter Meerlo, University of Groningen, Netherlands

\*Correspondence:

Michael Lazarus lazarus.michael.ka@u.tsukuba.ac.jp

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 26 October 2018 Accepted: 20 March 2019 Published: 03 April 2019

### Citation:

Fujii S, Kaushik MK, Zhou X, Korkutata M and Lazarus M (2019) Acute Social Defeat Stress Increases Sleep in Mice. Front. Neurosci. 13:322. doi: 10.3389/fnins.2019.00322 Keywords: social defeat stress, slow-wave sleep, homeostatic sleep need, mouse model, slow-wave activity

### INTRODUCTION

All living organisms respond to any external biological source of stress, i.e., perceived or actual threats, with a predictable biological pattern in an attempt to restore the internal homeostasis of the body (Chrousos, 2009). Acute stress typically involves transient responses of biological systems; however, prolonged activation of these initially beneficial reactions by excessive or chronically repeated stressors may lead to stress-related disorders. Social conflicts are a major source of stress for humans and play a major role in the pathogenesis of affective disorders like anxiety and depression (Charney and Manji, 2004; Huhman, 2006). Social conflicts also occur in animals (Toyoda, 2017). A social animal with the inability to dominate its opponent shows submissive behavior and accepts a lower social rank to avoid injury and death that could occur if the animal continues to act in an aggressive manner (Huhman, 2006). Rodent models of social defeat (SoD) stress are often based on a resident-intruder paradigm where the resident attempts to defend its home cage from the intruder who eventually subordinates itself to the unfamiliar territorial

resident conspecific. Acute SoD stress in rats has been shown to increase the heart rate and body temperature by sympathetic stimulation and blood corticosterone levels by activation of the hypothalamic–pituitary–adrenal axis (Koolhaas et al., 1997; Kataoka et al., 2014). These stress responses quickly dissipate and return to baseline levels within hours after the end of the stress period. SoD stress, however, also induces longer-lasting behavioral and physiologic changes. For example, several days may be required to normalize the circadian rhythmicity of body temperature, open-field behavior, and body mass growth after acute SoD (Meerlo et al., 1996a,b,c). Moreover, social animals are known to develop depression-like behaviors in response to chronically repeated SoD stress (Von Frijtag et al., 2000; van Kampen et al., 2002; Rygula et al., 2005; Huhman, 2006). For example, chronic SoD stress in mice induces anhedonia, anxiety, and social avoidance (Golden et al., 2011; Henriques-Alves and Queiroz, 2015). The anxiety and social avoidance behaviors in mice after chronic SoD stress can last several weeks (Berton et al., 2006; Tsankova et al., 2006; Krishnan et al., 2007). These longlasting behavioral changes in animals that experience repeated SoD stress are commonly associated with alterations in gene expression patterns in the brain (Tsankova et al., 2007).

The sleep/wake cycle is regulated by multiple factors, including homeostasis, circadian rhythm, and external environment, that are integrated by specific neuronal circuits controlling the sleep/wake behavior (Weber and Dan, 2016; Eban-Rothschild et al., 2017). Stress promotes wakefulness and inhibits sleep during the period when enhanced arousal is necessary to cope with external challenges. Alterations in the sleep architecture are often observed after an acute stress exposure (Sanford et al., 2015). A sleep rebound may be induced in animals to compensate for the sleep loss during stressful situations, but it is believed that sleep alterations are not only a homeostatic response to the sleep loss but also created by the stress. The type of sleep that is induced and the extent to which sleep stages are enhanced are highly variable between the types of stress or even the study design. For example, acute immobilization or restraint stress is followed by a selective increase in rapid eye movement sleep (REMS) (Rampin et al., 1991; Meerlo et al., 2001). Moreover, mice experiencing escapable foot shock stress showed an increase in REMS, whereas mice treated with inescapable foot shock stress showed a decrease in REMS (Sanford et al., 2010). Sleep is considered to have an important role in coping with stressful situations (Goldstein and Walker, 2014).

The acute effect of SoD stress on sleep has been reported in several rodent models. Although slow-wave activity (SWA) is consistently enhanced during sleep after SoD stress in mice and rats (Meerlo et al., 1997; Meerlo and Turek, 2001; Kamphuis et al., 2015; Henderson et al., 2017), the extent to which sleep stages are promoted varies between SoD studies and protocols. Male C57BL/6j mice when defeated by aggressive mice of the same strain during a 1-h interaction period showed a strong increase of slow-wave sleep (SWS) (Meerlo and Turek, 2001), whereas only a small increase of the SWS amount was observed in another study using the same SoD procedure (Vaanholt et al., 2003). This difference may be explained by variations in the aggressive behavior of the resident mice or prior stress experiences of the intruder mice in the laboratory environment. When highly aggressive male CD-1 mice were used in another study for the SoD of C57BL/6j mice during a 5-min interaction period followed by a 20-min period of olfactory, visual and auditory contact between the resident and intruder mice, a SWS increase in the intruder mice was preceded by an increase of wakefulness (Henderson et al., 2017). A fixed interaction period likely leads to painful attacks of the CD-1 mouse on the intruder mouse and thus, the sleep-wake behavior of the intruder mice may also be affected by pain. Overall, the extent to which the SoD procedure (e.g., pain or sleep deprivation) contributes to the post-SoD stress sleep/wake effects is not well established.

In the present study, we developed a mouse model of SoD stress based on a resident-intruder paradigm to evoke sleep alterations in the intruder C57BL/6j mouse after acute SoD stress by the resident CD-1 mouse trained to display persistent aggression against the intruder mouse. To prevent painful attacks to the submissive intruder mouse, the intruder and resident mice were separated by a partition when the intruder mouse showed clear submissive behavior and was attacked by the resident mouse. We found that SWS strongly increased in response to SoD, whereas REMS was only moderately increased for a limited period several hours after the SoD session.

### MATERIALS AND METHODS

### Animals

Male C57BL/6j mice (13–20 weeks old, and weighing 26– 33 g), maintained at the International Institute of Integrative Sleep Medicine of the University of Tsukuba, were used in the experiments. Male CD-1 (retired breeders) mice were obtained from Japan SLC (Hamamatsu, Japan). The animals were housed in an insulated and soundproof recording chamber maintained at an ambient temperature of 23 ± 0.5◦C with a relative humidity of 50 ± 5% and an automatically controlled 12 h light/12 h dark cycle (illumination intensity ≈ 100 lux). All animals had free access to food and water. This study was performed in strict accordance with the Guide for the Care and Use of Laboratory Animals of the US National Institutes of Health (2011). Experimental protocols were in compliance with relevant Japanese and institutional laws and guidelines, and approved by the University of Tsukuba animal ethics committee (protocol #14-322). Every effort was made to minimize the number of animals used as well as any pain and discomfort experienced by the animals.

### Stereotaxic Surgery for Placement of the EEG/EMG Electrodes

Mice were anesthetized with pentobarbital (50 mg/kg, intraperitoneal [i.p.]) and then placed in a stereotaxic apparatus. Electroencephalogram (EEG) and electromyogram (EMG) electrodes for polysomnographic recordings were chronically implanted in the mice (Oishi et al., 2016). The implant comprised two stainless steel screws (1 mm in diameter) that served as the EEG electrodes inserted through the skull above the cortex (anteroposterior, +1.0 mm; left-right, −1.5 mm from bregma or lambda) according to the atlas of Paxinos and Franklin (2004). Two insulated, stainless steel Teflon-coated wires, serving as the EMG electrodes, were placed bilaterally into both trapezius muscles. All electrodes were attached to a micro connector and fixed to the skull with dental cement.

### Stress Procedures

fnins-13-00322 April 1, 2019 Time: 18:4 # 3

The SoD stress protocol was designed based on the residentintruder paradigm. To prepare aggressive resident mice, male CD-1 mice (body weight > 40 g) were singly housed and trained to display persistent aggression against male C57BL/6j mice by placing a male C57BL/6j mouse in the cage of a CD-1 mouse one or two times a week. During the training, the C57BL/6j mouse was immediately removed from the cage when attacked and defeated by the CD-1 mouse. Unsuitable CD-1 mice that showed little aggression or extreme violent behavior (i.e., mice that inflict serious injuries to their opponent) were excluded from the study. Only CD-1 mice showing aggression against C57BL/6j mice within 1 min were used in the experiments to ensure successful SoD. The C57BL/6j mice used for aggression training were not used for any SoD experiments.

For SoD stress, we used a transparent acrylic partition placed diagonally in a rectangular cage to separate the intruder and resident mice. The partition has a wire-mesh opening in the lower part for olfactory, visual, and auditory contact (**Figures 1B,C**). A SoD session lasted 60 min starting at zeitgeber time 11. The C57BL/6j mouse (intruder mouse) was placed behind the partition in the resident mouse's cage. After 5 min, the partition was removed and re-inserted when the intruder mouse was attacked by the resident mouse and showed clear submissive behavior, including submissive posture, escaping, or freezing behavior. During one SoD session, the removal of the partition was repeated two, four, or eight times at 25-, 15-, or 7-min intervals, respectively. After the SoD session, the intruder mouse was returned to its home cage (**Figure 1D**).

The following experiments were also conducted to differentiate the specific effects of SoD and non-specific effects of the SoD procedure (**Figure 1E**): (1) C57BL/6j mice were placed in an unused cage with bedding, food pellets, and a partition for 60 min and sleep deprived by cage tapping ("Sleep deprivation" session). (2) C57BL/6j mice were placed in a cage with a partition previously used by a CD-1 mouse for more than 5 days ("No resident" session). (3) The intruder mouse was placed in the resident cage while the intruder and resident mice were separated by the partition for 60 min ("No contact" session). Sleep deprivation by cage tapping was not conducted during the "No resident" and "No contact" sessions.

### Vigilance State Assessment Based on EEG/EMG Polygraphic Recordings

One week after surgery, the mice were individually housed in cages in an insulated soundproof recording chamber and connected to the EEG-EMG recording cables for 2–3 days of habituation before starting the polygraphic recordings. Before the SoD session, some mice were subjected to a "Sleep deprivation" or "No resident" session, followed by a "No contact" session. Sessions were run every 3 or 4 days. Baseline was recorded on the day prior to each session (**Figure 1A**). The EEG/EMG signals were amplified, filtered (EEG 0.5–30 Hz; EMG 20–200 Hz), digitized at a sampling rate of 128 Hz, and recorded using the data acquisition software SleepSign <sup>R</sup> (Kissei Comtec, Matsumoto, Japan). The vigilance states were classified offline in 10-s epochs of three stages, i.e., wakefulness, REMS, and SWS by SleepSign <sup>R</sup> (ver 3.4) according to standard criteria (Oishi et al., 2016). As a final step, the software-defined vigilance stage of each 10 s epoch was visually examined, and manually corrected when necessary. Spectral analysis of EEG by fast Fourier transform was performed, and the EEG power densities were calculated in the range of 0.5–25 Hz in 0.5-Hz bins. SWA during SWS was calculated based on EEG power in the range of 0.5– 4 Hz. In a 3-hourly plot of SWA, the data were presented as percentages of the mean of the 12-h baseline SWS during the dark period.

### Behavioral Analysis

Animal behaviors during "No resident" and "No contact" sessions were analyzed using video recordings. Vigilance states and behaviors during wakefulness were scored in 4-s epochs. Behaviors were scored as grooming, exploration (including ambulation, rearing, digging, and sniffing), consumption (eating and drinking), or quiet waking, when the behavior accounted for more than 50% of the epoch.

### Blood Sampling and Corticosterone Measurement

Blood sampling was performed by cardiac puncture with the mice under deep isoflurane anesthesia immediately after the SoD sessions with 2, 4, or 8 defeats. Blood samples were obtained within 2 min, which is rapid enough to ensure that the stress imposed by the blood-sampling procedure did not affect the corticosterone levels in the plasma (Riley, 1960). For the control group, mouse blood was collected immediately after the "No resident" session. For basal corticosterone levels, blood was collected from undisturbed mice in their home cages at zeitgeber time 12. All the mice were used only once and not subjected to any behavioral tests. Blood samples were collected into EDTAcoated syringes and immediately centrifuged at 10,000 rpm for 15 min at 4◦C. Plasma samples were collected and stored at −80◦C until the assay was performed. Plasma corticosterone was measured in duplicate wells using the DetectX <sup>R</sup> ELISA kit (Arbor Assays, Ann Arbor, MI, United States). The mean of intraassay coefficients of variations (n = 31) calculated from duplicate wells was 2.82%.

### Statistical Analysis

Statistical analyses were carried out using GraphPad Prism software (GraphPad Software, La Jolla, CA, United States). All results are presented as mean ± standard error of the mean (SEM). A paired two-tailed Student's t-test was used for statistical comparisons between paired groups (**Figures 2B,C**, **3A**, **4A–C,E** and **Tables 2**, **4**). Two-way repeated-measures analysis of variance

(ANOVA) followed by Bonferroni's post hoc comparisons was used to analyze the sleep/wake profile (**Figure 2A**) and EEG power (**Figures 4D,F**). One-way ANOVA followed by Bonferroni's (**Figures 2D–F**) or Dunnett's (**Figures 3B**, **4E**) post hoc comparisons were performed to compare groups of three or more. The Kruskal–Wallis test was used to analyze the data with significant variance heterogeneity assessed by Levene's test (**Figure 2D**). In all cases, P < 0.05 was considered significant.

### RESULTS

### SoD Stress Strongly Increases Sleep

We established a mouse model of sleep alterations after acute SoD stress based on a resident-intruder paradigm, whereby a C57BL/6j male mouse was introduced as an intruder to a resident CD-1 male mouse selected for its high level of aggression (**Figure 1**). To prevent injury to the intruder mouse, the intruder and resident mice were separated by a partition. During a 1-h SoD session, the partition was removed multiple times and reinserted every time when the intruder mouse was attacked and showed submissive behavior. After the SoD session, the intruder mouse was returned to the home cage and EEG and EMG of the intruder mice were recorded.

First, we investigated the extent to which sleep was affected by different numbers of defeats during a 1-h session. Intruder mice were exposed to two, four, or eight successive defeats. In each successive defeat, mice were in contact with the resident mice for 15.5 ± 1.5 s. We analyzed EEG and EMG recordings made after the SoD session or on the previous day (baseline). SWS strongly increased from the first 3 h after the sessions with two, four, and eight defeats. The strong increase in SWS lasted until 6 or 9 h after the sessions with two or four defeats, respectively. Accordingly, wakefulness was decreased during the corresponding periods. A REMS increase was observed only between 6 and 9 h after the SoD session with four successive defeats (**Figure 2A** and **Table 1**). When the first 3 h following the SoD sessions were analyzed in more detail, an increase in wakefulness only occurred during the first 10 min after the SoD sessions and SWS started to increase afterward (**Figure 2B**). Because most of the effects on sleep stages were observed within 9 h after the sessions, we compared the total amounts of 9 h following the SoD sessions (**Figure 2C** and **Table 2**). The total baseline amounts of the vigilance states during the same time period were not significantly different among groups [SWS: P = 0.4914, F(2,20) = 0.7364, REMS: P = 0.0761, F(2,20) = 2.938, Wake: P = 0.3578, F(2,20) = 1.083]. SWS significantly increased after SoD sessions, whereas wakefulness decreased accordingly. By contrast, REMS was only increased significantly after four defeats.

FIGURE 2 | Slow-wave sleep (SWS) increased after social defeat (SoD) stress. (A) Time-course of SWS, rapid eye movement sleep (REMS), and wakefulness for 21 h after SoD sessions with two (n = 8), four (n = 8), or eight (n = 7) defeats. <sup>∗</sup>P < 0.05, ∗∗P < 0.01, and ∗∗∗P < 0.001 compared with baseline, assessed by two-way repeated measures ANOVA followed by Bonferroni's post hoc comparisons. (B) Time-course of SWS, REMS, and wakefulness for 3 h after SoD sessions. <sup>∗</sup>P < 0.05 compared with baseline, assessed by paired two-tailed Student's t-test. (C) Total amounts of SWS, REMS, and wakefulness for 9 h after SoD sessions. <sup>∗</sup>P < 0.05, ∗∗P < 0.01, and ∗∗∗P < 0.001 compared with baseline, assessed by paired two-tailed Student's t-test. (D) Changes in total amounts of SWS, REMS, and wakefulness for 9 h between baseline and after SoD sessions. <sup>∗</sup>P < 0.05, compared between groups, assessed by the Kruskal–Wallis test (SWS and wakefulness) or one-way ANOVA (REMS) followed by Bonferroni's post hoc comparisons. (E) Deviation of changes in total amounts of SWS and wakefulness. The deviation was calculated by subtracting the mean from the experimental values of each animal for changes in the total amounts of SWS and wakefulness for 9 h in each condition. <sup>∗</sup>P < 0.05 and ∗∗P < 0.01, compared between groups, assessed by one-way ANOVA followed by Bonferroni's post hoc comparisons. (F) Blood plasma corticosterone levels from undisturbed mice (n = 6) at zeitgeber time 12 of the baseline day or mice after a "No resident" (n = 7) or SoD sessions (n = 6). <sup>∗</sup>P < 0.05, ∗∗P < 0.01, and ∗∗∗P < 0.001 compared between groups, assessed by one-way ANOVA followed by Bonferroni's post hoc comparisons. Data are presented as means ± SEM. NS, not significant.

Moreover, the calculated changes in the total amounts of SWS, REMS, and wakefulness between baseline and the post-SoD 9-h periods revealed a significant difference in REMS only between four and eight times (**Figure 2D** and **Table 1**). Because variance in SWS or wakefulness was highly significantly different among groups, we performed Bonferroni's post hoc comparisons on the absolute deviations from the mean (**Figure 2E**). The deviations were significantly lower only in the intruder group with four defeats. Finally, we studied corticosterone responses in the intruder mice after SoD as an indicator of stress due to activation of the hypothalamic-pituitary-adrenal axis, the central stress response system (Mormède et al., 2007). Blood samples for measuring corticosterone levels were collected from undisturbed mice (to establish the baseline) and intruder mice immediately after the SoD sessions. A control session in a cage previously used by the resident mouse ("No resident") was also conducted. SoD sessions significantly increased corticosterone levels and it was significantly stronger compared with "No resident" mice [P < 0.0001, F(4,26) = 24.5, one-way ANOVA followed by Bonferroni's post hoc comparisons, **Figure 2F**]. These results may indicate that SoD causes stress in mice. However, there was no significant difference between SoD sessions with different

FIGURE 4 | Sleep architecture after social defeat (SoD) stress. (A–C) Episode numbers (A), mean episode durations (B), and numbers of stage transitions (C) for 9 h after SoD session with four defeats. <sup>∗</sup>P < 0.05 and ∗∗P < 0.01 compared with baseline, assessed by paired two-tailed Student's t-test. (D) Power density of EEG during SWS for 30 min after SoD session with four defeats or onset of the dark phase on the baseline day. <sup>∗</sup>P < 0.05 compared with baseline, assessed by two-way repeated measures ANOVA followed by Bonferroni's post hoc comparisons. (E) Changes in the SWA (0.5–4 Hz) relative to the baseline for 30 min after a "Sleep deprivation," "No resident," "No contact," or SoD session. <sup>∗</sup>P < 0.05 compared with baseline, assessed by paired two-tailed Student's t-test, #P < 0.05 compared with SoD, assessed by one-way ANOVA followed by Dunnett's post hoc comparisons. (F) Time-course of SWA changes after SoD session. ∗∗∗P < 0.001 compared with baseline, assessed by two-way repeated measures ANOVA followed by Bonferroni's post hoc comparisons.

numbers of successive defeats. These results suggested that the SWS increase observed after SoD stress is mostly independent of the numbers of defeats.

### Specific Effects of SoD Versus Non-specific Effects of the SoD Procedure

To differentiate between the specific effects of SoD and non-specific effects of the SoD procedure (e.g., novelty or sleep deprivation), the sleep/wake behavior of mice after SoD stress was compared with that following other conditions, including sleep deprivation in a clean unused cage ("Sleep deprivation"), a cage previously used by the resident mouse ("No resident"), or the presence of the resident mouse separated by a partition ("No contact"). Data of the control conditions were compared with those of four successive defeats, because SWS consolidation occurred with significantly less variation after this SoD procedure than the ones with two or eight defeats.

During the "No resident" or "No contact" session, the animals were spontaneously active and awake during most of the 1-h session (**Table 3**). "Sleep deprivation" and "No contact" sessions significantly increased SWS and decreased wakefulness during a 9-h period after the sessions (**Figure 3A** and **Table 4**); however, the SWS increase was smaller than that after SoD stress (**Figure 3B**). SWS was not changed after the "No resident" session (**Figure 3A**). REMS also increased after the SoD and control procedures; however, there was only a significant difference between the baseline and experimental

### TABLE 1 | Statistics of data shown in Figures 2A,B, 3B.

fnins-13-00322 April 1, 2019 Time: 18:4 # 8


periods after SoD stress (**Figure 3A**). These results suggest that a significant part of the sleep changes after SoD can be attributed to non-specific sleep deprivation during the SoD session, i.e., about half of the increase of SWS and all of the increase of REM sleep may be a homeostatic sleep response to non-specific sleep deprivation.

### SoD Stress Increases the Number of SWS Episodes and Slow-Wave Activity

The sleep architecture and SWA were then analyzed after SoD session with four successive defeats. SoD stress significantly affected the episode number of SWS and wakefulness for 9 h after the SoD session (SWS: P = 0.0079, Wake: P = 0.0077, paired Student's t-test, **Figure 4A**). The mean duration of wake episodes decreased by 48.4 ± 6.9% (P = 0.0035, paired Student's t-test) compared with the baseline, whereas the duration of the SWS or REMS episodes was not significantly different or slightly increased, respectively (**Figure 4B**). The number of stage transitions from SWS to wake and wake to SWS was significantly increased (SWS to Wake: P = 0.0131, Wake to SWS: P = 0.0076, paired Student's t-test), whereas other stage transitions were not affected by SoD stress (**Figure 4C**).

To assess whether EEG activity was altered by SoD stress, we compared the normalized EEG power spectrum of SWS for 30 min following the SoD session with baseline SWS. EEG activity was significantly increased in the frequency range of 1–4 Hz and decreased in the frequency range of 4.5–10 Hz during SWS [interaction P < 0.0001, F(49,350) = 25.47, two-way repeated-measures ANOVA followed by Bonferroni's post hoc comparisons, **Figure 4D**]. SWA in the frequency range of 0.5–4 Hz was then calculated during the same period and compared with the "Sleep deprivation," "No resident," and "No contact" conditions. SWA was significantly increased in "No resident," "No contact," and SoD mice, although the increase of SWA in the SoD mice was significantly stronger than that in the "No resident" mice, but not in the "No contact" mice [P = 0.0176, F(3,20) = 4.258, oneway ANOVA followed by Dunnett's post hoc comparisons, **Figure 4E**]. Finally, we calculated the 3-h SWAs in intruder mice for 12 h following SoD stress and found that the first 3-h SWA was significantly higher than the baseline SWA on the previous day [main effect P = 0.0005, F(1,28) = 15.66, interaction P = 0.0002, F(3,28) = 8.991, two-way repeatedmeasures ANOVA followed by Bonferroni's post hoc comparisons, **Figure 4F**].

### DISCUSSION

The present study explored the effect of acute SoD stress on the sleep/wake cycle in mice. We observed that SoD stress strongly promoted SWS over 9 h. Corticosterone, as an indicator of stress, was significantly increased in the blood plasma by SoD, confirming that SoD induced stress in the mice. The increase in SWS was associated with more transitions from wakefulness to SWS without changing the mean episode duration of SWS.

The effect of acute SoD stress on subsequent sleep amount was reported in several studies. One study showed strong SWS increase in mice after acute SoD stress by interaction for 1 h with mice from the same strain (Meerlo and Turek, 2001). By contrast, another study using the same procedure showed little



<sup>∗</sup>P < 0.05, ∗∗P < 0.01, and ∗∗∗P < 0.001 compared with baseline, assessed by paired two-tailed Student's t-test.

TABLE 3 | Percentage of animal behaviors (during wakefulness), slow-wave sleep (SWS), and rapid eye movement sleep (REMS) during the "No resident" and "No contact" sessions.


effect of SoD on the SWS amount (Vaanholt et al., 2003). Due to the lack of information about the number and duration of attacks of the resident mice against the intruder mice, it is hard to explain the difference between the studies. Moreover, it was also reported that wakefulness is increased in the first 3 h after SoD followed by a sleep increase during later hours (Henderson et al., 2017). Highly aggressive male CD-1 mice were used in this study for the SoD of C57BL/6 mice during a 5-min interaction period followed by a 20-min period of olfactory, visual, and auditory contact between the resident and intruder mice. The initial increase of wakefulness in this study may be attributed to repeated painful attacks of the CD-1 mice against smaller C57BL/6 mice. Alternatively, the durations of physical and sensory stress (25 min) may be too short to induce sleep. Although variations in the aggressive behavior of the resident mice may account to some extent for the differences in the sleep/wake responses between the studies, sleep architecture, SWA, and corticosterone responses in socially defeated mice may also be influenced by other factors. For example, laboratory environment and time-of-theday when the conflict was introduced may affect sleep and corticosterone parameters. Moreover, prior stress experiences of the intruder mice such as early life stress during maternal care, fighting between littermates, or low social ranking in the litter can also be important factors for outcomes after SoD stress (Wang et al., 2014; Peña et al., 2017). We designed a SoD stress protocol to minimize behavioral variations during SoD. In our protocol, intruder mice were separated from aggressive resident mice (CD-1 mice) when they exhibited submissive behaviors rather than by setting a fixed duration of interaction with the resident mice, preventing repeated and painful attacks against the submissive mouse. Moreover, physical interaction was conducted two, four or eight times to maintain a high level of SoD stress in the intruders during the session. We found that acute SoD stress, introduced at the end of the light period, increases SWS shortly after the SoD session. Moreover, the comparison of the specific effects of SoD and non-specific effects of the SoD procedure showed that a large portion of the sleep changes after SoD stress can be attributed to a homeostatic response to sleep deprivation. On the other hand, about half of the SWS increase is likely a specific response to SoD.

We also observed a strong increase in SWA after the SoD session (**Figures 4E,F**), consistent with previous reports in mice and rats (Meerlo et al., 1997; Meerlo and Turek, 2001; Kamphuis et al., 2015; Henderson et al., 2017). An increase in SWA is considered the hallmark of homeostatic sleep need in response to sleep deprivation (Suzuki et al., 2013; Dispersyn et al., 2017; Wang et al., 2018). The SWA increase, however, was stronger after the SoD condition than under sleep-disrupting control conditions without a CD-1 mouse. Moreover, SWS was still enhanced long after the SWA returned to the baseline level. These observations suggest that SWA is also affected by social stress and at least part of the SWS in response to SoD is independent of SWA.

Moreover, we observed large variations in the SWS increase after the SoD sessions, especially the ones with two and eight successive defeats (**Figures 2D,E**). It is well known that grouphoused male mice establish a social hierarchy (Wang et al., 2014). As we did not control the social ranking of the intruder mice before the SoD session, it is possible that the SWS response of the intruder mice is affected by their SoD resilience.

The neuronal mechanisms of the sleep response to SoD stress are unknown. The mesolimbic dopamine system plays an important role in the development of depression-related

TABLE 4 | Total amounts (minutes, mean ± SEM) of slow-wave sleep (SWS), rapid eye movement sleep (REMS), and wakefulness for 6, 9, 12, and 24 h after "Sleep deprivation," "No resident," and "No contact" sessions.


<sup>∗</sup>P < 0.05, ∗∗P < 0.01, and ∗∗∗P < 0.001 compared with baseline, assessed by paired two-tailed Student's t-test.

behaviors after repeated SoD stress (Russo and Nestler, 2013). The mesolimbic system comprises dopaminergic projections from the ventral tegmental area to the nucleus accumbens, an area critical for reward and motivation (Graybiel, 2008; Russo and Nestler, 2013), and is also connected to the medial prefrontal cortex, amygdala, and hippocampus, allowing the system to integrate cognition, emotion, and action (Floresco, 2015). We and others recently described the role of the mesolimbic system in the gating of sleep by motivated behavior (Eban-Rothschild et al., 2016; Oishi et al., 2017a,b; Luo et al., 2018). Therefore, it is plausible that the mesolimbic system also mediates sleep responses to SoD stress.

Acute stress responses induce behavioral, sympathetic and neuroendocrine changes in animals to facilitate a fight or flight response (Chrousos, 2009). Sympathetic and neuroendocrine responses, including the increase of body temperature and heart rate and secretion of noradrenaline, adrenaline, and corticosteroids, usually dissipate within hours after acute SoD in rodents (Koolhaas et al., 1997, 2017). Sleep is known to inhibit glucocorticoid secretion in healthy men (Weitzman et al., 1983), whereas REMS disinhibition and dysregulation of hypothalamic-pituitary-adrenal axis are hallmarks of depression and chronic stress (Nollet et al., 2019; Steiger and Pawlowski, 2019). The increased SWS after SoD stress in our study may facilitate the termination of stress responses. Moreover, SWS induction may also be beneficial for restoration of brain homeostasis after the stress response (Xie et al., 2013; Ding et al., 2016). By contrast, REMS is considered to have an important function in the processing of emotional memory of adverse experiences to promote emotional and mental recovery (Suchecki et al., 2012; Goldstein and Walker, 2014). Only a moderate REMS increase was observed during a period 6–9 h after SoD sessions with four defeats (**Figure 2A**). Most, if not all, of the REMS increase, however, may be attributed to non-specific sleep deprivation during the SoD session and thus, a beneficial effect of REMS after SoD remains unclear.

In summary, our study revealed strong SWS responses to acute SoD stress in submissive male mice attempting to appease their aggressive counterpart, when painful attacks on the submissive mice were prevented. Our SoD stress model may be useful for studying the mechanisms and functions of sleep in response to social stress.

### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the supplementary files.

### AUTHOR CONTRIBUTIONS

SF and ML designed the experiments. SF, MKK, XZ, and MK collected and analyzed the data. SF and ML wrote the manuscript.

# FUNDING

This work was supported by the Japan Society for the Promotion of Science [Grant in-Aid for Scientific Research B (Grant No. 17H02215) to ML]; the Japan Science and Technology Agency [CREST grant (Grant No. JPMJCR1655) to ML]; the Ministry of Education, Culture, Sports, Science, and Technology (MEXT) of Japan [Grants-in-Aid for Scientific Research on Innovative Areas "Living in Space" (Grant Nos. 15H05935, 15K21745, and 18H04966) and "WillDynamics" (Grant No. 17H06047) to ML]; the World Premier International Research Center Initiative (WPI) from MEXT (to ML); and a research grant of the Astellas Foundation for Research on Metabolic Disorders (to ML).

### ACKNOWLEDGMENTS

We thank all of the laboratory members for technical assistance, discussions, and comments.

# REFERENCES

fnins-13-00322 April 1, 2019 Time: 18:4 # 11



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Fujii, Kaushik, Zhou, Korkutata and Lazarus. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Whole-Brain Neural Connectivity to Lateral Pontine Tegmentum GABAergic Neurons in Mice

Ze-Ka Chen<sup>1</sup>† , Xiang-Shan Yuan<sup>1</sup>† , Hui Dong<sup>1</sup>† , Yong-Fang Wu<sup>2</sup> , Gui-Hai Chen<sup>2</sup> , Miao He<sup>1</sup> , Wei-Min Qu<sup>1</sup> and Zhi-Li Huang<sup>1</sup> \*

<sup>1</sup> State Key Laboratory of Medical Neurobiology and MOE Frontiers Center for Brain Science, Institutes of Brain Science, Department of Pharmacology, School of Basic Medical Sciences, Fudan University, Shanghai, China, <sup>2</sup> Department of Neurology (Sleep Disorders), Chaohu Hospital of Anhui Medical University, Hefei, China

The GABAergic neurons in the lateral pontine tegmentum (LPT) play key roles in the regulation of sleep and locomotion. The dysfunction of the LPT is related to neurological disorders such as rapid eye movement sleep behavior disorder and ocular flutter. However, the whole-brain neural connectivity to LPT GABAergic neurons remains poorly understood. Using virus-based, cell-type-specific, retrograde and anterograde tracing systems, we mapped the monosynaptic inputs and axonal projections of LPT GABAergic neurons in mice. We found that LPT GABAergic neurons received inputs mainly from the superior colliculus, substantia nigra pars reticulata, dorsal raphe nucleus (DR), lateral hypothalamic area (LHA), parasubthalamic nucleus, and periaqueductal gray (PAG), as well as the limbic system (e.g., central nucleus of the amygdala). Further immunofluorescence assays revealed that the inputs to LPT GABAergic neurons were colocalized with several markers associated with important neural functions, especially the sleep-wake cycle. Moreover, numerous LPT GABAergic neuronal varicosities were observed in the medial and midline part of the thalamus, the LHA, PAG, DR, and parabrachial nuclei. Interestingly, LPT GABAergic neurons formed reciprocal connections with areas related to sleep-wake and motor control, including the LHA, PAG, DR, parabrachial nuclei, and superior colliculus, only the LPT-DR connections were in an equally bidirectional manner. These results provide a structural framework to understand the underlying neural mechanisms of rapid eye movement sleep behavior disorder and disorders of saccades.

Keywords: GABA, lateral pontine tegmentum, viral neuronal tracing, reciprocal connections, sleep-wake cycle, locomotion

# INTRODUCTION

The pontine tegmentum, a brainstem structure, are all the regions from the basilar pons to the fourth ventricle. It plays important roles in perception, movement, vigilance, respiration, and the sleep-wake regulation (Plazzi et al., 1996; Alheid et al., 2004; Tziridis et al., 2012). The lateral pontine tegmentum (LPT) is the dorsolateral part of the pontine tegmentum, which is included in the deep mesencephalic nucleus (DpMe). The LPT is identified as the site between the ventral lateral periaqueductal gray (vlPAG) and the pedunculopontine tegmental nucleus (PPT) on the

### Edited by:

Takeshi Sakurai, University of Tsukuba, Japan

### Reviewed by:

Christelle Peyron, INSERM U1028 Centre de Recherche en Neurosciences de Lyon, France Michihiro Mieda, Kanazawa University, Japan

> \*Correspondence: Zhi-Li Huang huangzl@fudan.edu.cn †These authors have contributed

equally to this work

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 08 January 2019 Accepted: 01 April 2019 Published: 24 April 2019

### Citation:

Chen Z-K, Yuan X-S, Dong H, Wu Y-F, Chen G-H, He M, Qu W-M and Huang Z-L (2019) Whole-Brain Neural Connectivity to Lateral Pontine Tegmentum GABAergic Neurons in Mice. Front. Neurosci. 13:375. doi: 10.3389/fnins.2019.00375

ventro-dorsal localization, and is a trip lying the region medial to the PPT on the rostro-caudal extent in rodents (approximately −4.04 to −5.02 mm corresponding to Bregma in the mouse brain atlas) (Franklin and Paxinos, 2001; Lu et al., 2006; Sherman et al., 2015). The LPT mainly contains glutamatergic and GABAergic neurons (Liang et al., 2014).

The LPT is involved in voluntary locomotion, general anesthesia, and even incentive learning (Sukhotinsky et al., 2005; Fogel et al., 2010; Le Ray et al., 2011). Dysfunction of the LPT is associated with neurological disorders such as rapid eye movement (REM) sleep behavior disorder (RBD), saccade disorders and temporal lobe epilepsy (TLE). Using magnetic resonance imaging, diffusion changes of visualized structural abnormalities were found in the areas including the LPT in patients with idiopathic RBD (Scherfler et al., 2011). Ocular flutter was found to be related to the right upper pontine tegmentum damage, which contained LPT neurons (Tsutsumi et al., 2009). LPT dysfunction in patients with TLE was found to contribute to neurocognitive disturbances (Englot et al., 2018). In addition to the above, LPT neurons are especially known to have a vital function in REM sleep regulation in rodents (Sapin et al., 2009) and demyelination of LPT neurons led to dream-like hallucinations during wake, relating to the impairment of REM sleep inhibitory mechanisms (Vita et al., 2008). Furthermore, c-fos, a marker of activated neurons, was strongly expressed in the LPT after REM sleep deprivation and colocalized with glutamate decarboxylase, a key enzyme for the production of GABA, suggesting that LPT GABAergic neurons are involved in REM sleep regulation (Lu et al., 2006; Sapin et al., 2009).

The neuronal connectivity of LPT neurons has previously been investigated using classic anterograde tracers, such as biotinylated dextran amine (BDA), and retrograde tracers, such as horseradish peroxidase or cholera toxin b-subunit (CTB). Consequently, the LPT was shown to receive inputs from the lateral hypothalamic area (LHA) (Clement et al., 2012), the pallidum (Gorbachevskaia, 2011), and the sublaterodorsal nucleus (SLD) (Boissard et al., 2003), while projecting to the LHA (Ford et al., 1995), superior colliculus (SC) (Appell and Behan, 1990), and SLD (Liang and Marks, 2014). However, specific afferent inputs and efferent outputs of LPT GABAergic neurons remain unknown.

Recent genetic tracing methods using the retrograde trans-synaptic rabies virus (RV) allow us to identify the presynaptic connections of a genetically well-defined neuronal population (Wickersham et al., 2007; Yuan et al., 2018). Moreover, the use of adeno-associated virus (AAV) vectors allows cell-type-specific, non-toxic, long-term expression of transgenes for anterograde tracing of neuronal pathways (Kuhlman and Huang, 2008; Holloway et al., 2013; Zhang et al., 2013). Here, we employed a modified RV and AAV in combination with the Cre/LoxP system to map presynaptic inputs and output projections of LPT GABAergic neurons. We identified several monosynaptic inputs of LPT GABAergic neurons by double labeling with antibodies for molecular markers related to sleep-wake regulation. Our results revealed the wide-ranging connectivity of LPT GABAergic neurons and provided a structural framework to understand the underlying neural circuits of important physiological functions.

# MATERIALS AND METHODS

### Animals

The viral-based tracing experiments were all performed in male GAD2-IRES-Cre mice obtained from Miao He and Z. Josh Huang's laboratory, which generated the strain on a mixed genetic background (129SVj/B6; with black coat color) to genetically target almost whole-brain GABAergic neurons with a high level of specificity (Taniguchi et al., 2011). The mice were bred and housed under an automatically controlled 12-h light/12-h dark cycle (lights on at 7 a.m.; 100 lux intensity) (Zhang et al., 2017). After surgery, the mice were allowed to recover for at least 2 weeks before further experiments. This study was carried out in accordance with the recommendations of the China Regulations on the Administration of Laboratory Animals (the Decree No. 2 of National Science and Technology Commission of the People's Republic of China) and all animal procedures were approved by the Medical Experimental Animal Administrative Committee of the School of Basic Medical Sciences (Permit No. 20140226-024), Fudan University (Shanghai, China).

# Virus

All the viral vectors were packaged by BrainVTA (BrainVTA, Co., Ltd., Wuhan, China). The titer of the EnvA-pseudotyped, glycoprotein (RG)-deleted, DsRed-expressing rabies virus (RV-EnvA-1RG-DsRed) was about 2 × 10<sup>8</sup> infecting units per mL. The three AAV vectors, namely AAV-EF1α-DIO-ChR2-mCherry, AAV-EF1α-DIO-TVA-GFP, and AAV-EF1α-DIO-RV-G, were all packaged into the 2/9 AAV serotype and titred at about 3 × 10<sup>12</sup> genome copies per mL (Yuan et al., 2018).

# Surgery and Viral Injections

Totally, 28 mice were injected with viruses. Only eight animals were correctly injected after histological confirmation and were included in the results. For virus injection, adult GAD2-IRES-Cre mice were anesthetized with ∼1.5% isoflurane in oxygen (flow rate of 1 L/min). Using the Nanoject II (Drummond Scientific, Broomall, PA), 23 nL AAV9-EF1α-DIO-TVA-GFP and AAV9-EF1α-DIO-RvG was stereotaxically injected (4.6 nL/s) via a micropipette into the unilateral LPT (anterior–posterior: −4.2 mm, medial-lateral: +1.0 mm, dorsal-ventral: −3.5 mm) (Krenzer et al., 2011). Following the injection, the pipette was held in place for an additional 10 min to allow diffusion of viral particles away from the injection site before being slowly withdrawn. Two weeks later, 46 nL volume of RV-EnvA-1RG-DsRed was prepared to inject into the previous site of the LPT (n = 4 mice). For anterograde tracing, AAV-EF1α-FLEX-ChR2-mCherry was injected into the LPT following the same procedures described above. After 3 weeks, these mice were perfused (n = 4 mice).

### Histology and Immunostaining

fnins-13-00375 April 17, 2019 Time: 16:25 # 3

After being deeply anesthetized, adult mice were transcardially perfused with cold normal saline followed by 4% paraformaldehyde in 0.1 M phosphate buffer (PB). The brains were then extracted, post-fixed for 6 h, and then incubated in 20% sucrose in PB at 4◦C until they sank. Coronal sections (30 µm) were cut from a fixed brain on a freezing microtome (CM1950, Leica, Germany) into four series to obtain groups of tissue for multiple manipulations, and the distance between sections in each group was 90 µm. Tissue groups were restored in a cryopreservative solution.

For immunofluorescence, sections were rinsed with 0.3% Triton X-100 in 0.01 M PBS, and then they were incubated by primary antibodies [goat anti-choline acetyltransferase (ChAT), 1: 1000, Millipore AB144P; polyclonal rabbit anti-GABA, 1:1000, Acris Antibodies 20094; polyclonal rabbit anti-5-hydroxytryptamine (5-HT), 1:3000, Sigma S5545; polyclonal rabbit anti-calretinin, 1:2000, Invitrogen MA5-14540; polyclonal goat anti-parvalbumin, 1:3000, Swant PVG213; polyclonal goat antibody against orexin A, SCB sc8070, 1:2000; polyclonal goat antibody against melanin-concentrating hormone (MCH) sc-14509, SCB,1:1000] in PBS containing 0.3% Triton X-100 (PBST) overnight at 4◦C. Then the sections were washed and incubated by secondary antibodies (AlexaFluor 488 donkey antibody against goat IgG, 1: 500, Jacksonimmuno, Inc.; AlexaFluor 488 donkey antibody against rabbit IgG, 1: 1000, Jacksonimmuno, Inc.; AlexaFluor 647 donkey antibody against goat IgG, 1: 1000, Jacksonimmuno, Inc.) for 1 h at room temperature. Next, the sections were incubated for 10 min to stain nuclei by 4<sup>0</sup> ,6-diamidino-2 phenylindole (DAPI, 1: 10000, Sigma-Aldrich D9542) and rinse washed three times. Finally, the sections were mounted on glass slides and cover-slipped. For the immunostaining of mCherry, free-floating sections were rinsed in PBS and incubated by primary antibodies (polyclonal rabbit anti-mCherry 1:3000, Clontech 632496) in PBST and waggled slowly overnight at 4 ◦C. After washing, sections were incubated with secondary antibodies (biotinylated goat anti-rabbit IgG, 1:1000, Vector Laboratories BA-1000) and then incubated with an avidin-biotin peroxidase complex (ABC) solution (1:1000, Vector Laboratories PK-6100) for 1 h. After washing, the sections were immersed in a 3,3-diaminobenzidine-4 HCl (DAB, Vector Laboratories SK-4100) for 5–10 min at room temperature, after which mCherry-immunoreactive neurons could be identified by the presence of brown reaction product. Finally, the sections were mounted on glass slides, dried, dehydrated, and cover-slipped (Chen et al., 2016; Yuan et al., 2017; Luo et al., 2018).

### Imaging and Analysis

Whole-brain sections were imaged by a 10 or 20× objective on the VS120 virtual microscopy slide scanning system (Olympus) and magnified images of brain sections were captured using a 20 or 40× objective by a confocal microscope, to obtain more detail (Olympus Fluoview 1000, Tokyo, Japan).

For cell mapping of neurons, neural bodies were quantified semi-automatically using ImageJ software. Based on mouse brain map, ImageJ was used to manually depict the boundaries of specific brain regions (Franklin and Paxinos, 2001). For starter cell mapping, we first applied ImageJ to distinguish the cells coexpressing DsRed and GFP as starter cells, then generated cell representation by applying the automatic wand (tracing) tool and bicubic interpolation to maximize neuronal fidelity. Next, we inverted the colorless areas to white and matched the cellular outlines to the appropriate brain regions based on the mouse brain atlas (**Figure 2A**). Starter cells were binned at 0.12 mm along the anterior–posterior axis, centered at the brain slice coordinate for each coronal section image.

For axonal varicosity counting, the images were captured by a 20× objective on the Olympus VS120 system. The axonal varicosity values of overall brain were calculated semi-automatically by particle analyzing plugin in ImageJ. If the axons had a transverse diameter greater than 0.5 µm, then varicosities were defined (Mechawar et al., 2000; Li et al., 2002). The outlines of brain regions were also depicted by ImageJ based on the reference brain atlas.

All data values were presented as the mean ± standard error of the mean (SEM). The strength and direction of the linear relationship between subregions and cells or varicosity proportion were measured by the Pearson product-moment correlation coefficient (CC). Two-tailed Student's t-tests were used to compare the monosynaptic inputs and axonal projections of LPT GABAergic neurons.

### RESULTS

### Applying the Rabies Virus System to Identify Monosynaptic Inputs to LPT GABAergic Neurons

In order to identify monosynaptic inputs to LPT GABAergic neurons, we used a cell-type-specific, RG-deleted RV strategy in GAD2-IRES-Cre mice. The approach has been reported to mark monosynaptic inputs to specifically selected starter cells and has been successfully used in previous studies by our group (Yuan et al., 2018). GAD2-IRES-Cre mice were injected in the LPT with AAV helper viruses fused to green fluorescent protein (GFP). Two weeks later, the modified RV was injected in the same area. One week after this second injection, the mice were perfused, and the brains were processed (**Figures 1A,B**).

Rabies virus infection was clearly detectable at the injection site in GAD2-IRES-Cre mice compared with wild-type littermates, indicating no leakage of viral infection (**Figure 1C**). The location of the injection site was verified by identifying virus-infected neurons in the dorsomedial corner of the PPT and the ventrolateral corner of the PAG. The yellow staining in neurons indicated that the starter cells were co-infected by AAV helper viruses and RV, and the purple staining in neurons presented ChAT immunoreactive signals for the PPT, which was the border of the LPT (**Figures 1D,E**). There were also some DsRed-positive neurons not expressing GFP in the LPT, revealing the existence of local input directly to LPT GABAergic neurons (**Figures 1D,E**, **2A**).

In addition, we mapped the GABAergic starter cells as described in Section "Materials and Methods" (**Figure 2A**) and revealed that they were mostly restricted to the LPT. Most of the starter neurons were found to be located in the coronal planes between −4.16 and −4.84 mm from bregma, especially between −4.36 and −4.60 mm from bregma (**Figure 2B**).

### Input Patterns of LPT GABAergic Neurons

To determine the presynaptic connections of LPT GABAergic neurons, we analyzed serial coronal sections from mice injected with TVA-based trans-synaptic tracer. Sections from a representative GAD2-IRES-Cre brain revealed RV-DsRed-labeled presynaptic input neurons located only in specific brain nuclei in a unilateral manner (**Figure 3**). **Figure 4** depicts input tracing to LPT GABA neurons in greater detail, with enlarged images of inputs from typical subregions such as the bed nucleus of the stria terminalis (BNST), central nucleus of the amygdala (CeA), parasubthalamic nucleus (PSTN), substantia nigra pars reticulata (SNr), SC, dorsal raphe nucleus (DR), PAG, and tuberomammillary nucleus (TMN). Further immunofluorescence assays showed that the inputs to LPT GABAergic neurons were colocalized with several markers associated with important neural functions, especially the sleep-wake cycle. In the hypothalamus, the inputs to LPT GABAergic neurons were found to be partly colocalized with orexin (0.14 ± 0.04 in **Figure 5A**) or MCH (0.11 ± 0.02 in **Figure 5B**) in the LHA, which was shown

to inhibit or promote REM sleep (Adamantidis et al., 2007; Sasaki et al., 2011; Vetrivelan et al., 2016). The input neurons from the PSTN to LPT GABAergic neurons were mostly colocalized with calretinin (CR, 0.44 ± 0.04 in **Figure 5C**), where the inputs might to be related to the sleep-wake cycle (Steininger et al., 1999). In addition, the inputs from the limbic system to the LPT were partly colocalized with GABA neurons, such as the BNST (0.32 ± 0.03 in **Figure 5D**) and CeA (0.41 ± 0.03 in **Figure 5E**), which may be associated with sleep-wake regulation (Kodani et al., 2017; Mahoney et al., 2017) The inputs from the DR to the LPT were found partly colocalized with serotonin (0.20 ± 0.04 in **Figure 5F**), which was proved to promote wakefulness (Ito et al., 2013). Beyond our expectation, numerous inputs from the LPT were colocalized with parvalbumin in the SNr (0.65 ± 0.03 in **Figure 5G**), a region that was reported to mostly contain GABAergic neurons and to be possibly involved in REM sleep (Pal and Mallick, 2009).

injection site is marked by red solid circles. Only the hemisphere ipsilateral to the injection area is shown. Scale bar: 500 µm. MO, momatomotor cortex; SS, somatosensory cortex; STRv, stratum ventral region; PALv, pallidum, ventral region; PALd, pallidum, dorsal region; BMA, basomedial amygdalar nucleus;CeA, central amygdaloid nucleus; MEA, medial amygdalar nucleus; BNST, bed nucleus of the stria terminalis; PF, parafascicular thalamic nucleus; POA, preoptic area; LHA, lateral hypothalamic area; MHA, medial hypothalamic area; PSTN, parasubthalamic nucleus; TMN, tuberomammillary nucleus; ZI, zona incerta; SNc, substantia nigra pars compacta; SNr, substantia nigra pars reticulata; DpMe, deep mesencephalic nucleus; VTA, ventral tegmental area; SC, superior colliculus; RRF, retrorubral field; DR, dorsal raphe nucleus; PAG, periaqueductal gray; PnR, pontine reticular nucleus; NLL, the nuclei of the lateral lemniscus; PBN, parabrachial nucleus; SLD, sublateral dorsal nucleus; LDT, laterodorsal tegmental nucleus; LC, locus coeruleus; Gi, gigantocellular reticular nucleus; CGPn, central gray of the pons; RMg, raphe magnus nucleus; Ve, vestibular nucleus; DCN, deep cerebellar nucleus.

After identifying the brain regions with monosynaptic input to LPT GABAergic neurons, the distribution of these DsRed-labeled afferent neurons was analyzed based on detailed statistics. We divided each brain into six general structures, namely the midbrain, diencephalon, telencephalon, cerebellum, medulla, and pons, which together encompassed 48 specific brain regions containing the DsRed-labeled neurons in the whole brain (n = 4 mice). Then we calculated the proportion of input from each region against the total number of input neurons (**Figure 6**, left). We found that the midbrain provided the highest numbers

retroflexus; ns, nigrostriatal bundle; cp, cerebral peduncle, basal part; mt, mammillothalamic tract; ml, medial lemniscus; InG, intermediate gray layer of the superior colliculus; InWh, intermediate white layer of the superior colliculus; DpG,; mlf, medial longitudinal fasciculus; Aq, aqueduct; scp, superior cerebellar peduncle.

of inputs to the LPT, while few labeled neurons were found in the medulla or cerebellum. The largest number of inputs to LPT GABAergic neurons was found to arrive from the SC (15.10 ± 3.44%). The brain regions sending the second and third largest number of inputs to LPT GABAergic neurons were the SNr (10.00 ± 1.62%) and the DR (6.13 ± 1.07%) in the midbrain. In addition, the DpMe (5.17 ± 0.67%), PAG (3.64 ± 0.40%), and the substantia nigra pars compacta (SNc, 2.83 ± 0.68%) in the midbrain, the LHA (4.41 ± 1.05%) and PSTN (3.12 ± 1.08%) in the diencephalon, and the CeA (4.62 ± 1.71%) and the BNST (3.29 ± 0.48%) in the telencephalon also had strong projections to LPT GABAergic neurons.

### Output Patterns of LPT GABAergic Neurons

We next mapped the output of LPT GABAergic neurons using an AAV-mediated virus, similar to our previous work with viral anterograde tracing (Zhang et al., 2013; Chen et al., 2016; Oishi et al., 2017; Wang et al., 2017). To label the

axonal projections, we injected AAV expressing Cre-dependent ChR2-mCherry (**Figure 7A**) into the LPT of GAD2-IRES-Cre mice. After 3 weeks, mCherry expression was observed to be restricted to the LPT, between the PPT and vlPAG (**Figure 7B**).

SNr, substantia nigra pars reticulata; MCH, melanin-concentrating-hormone; 5-HT, 5-hydroxytryptamine.

In order to more precisely detect axonal varicosities, we combined viral tracing with DAB immunostaining to label mCherry (Campos et al., 2014). Images were compared with the reference brain atlas, and mCherry labeled axons were

parvalbumin (PV)-positive neurons in the SNr (G). Rightmost columns, quantification of DsRed+ cells that are positive for special biomarkers. n = 4, each data point represents one experimental animal. LHA, lateral hypothalamic area; PSTN, parasubthalamic nucleus; CeA, central amygdaloid nucleus; DR, dorsal raphe nucleus;

area; ZI, zona incerta; STN, subthalamic nucleus; SNc, substantia nigra pars compacta; VTA, ventral tegmental area; SNr, substantia nigra pars reticulata; DpMe, deep mesencephalic nucleus; RRF, retrorubral field; RMC, red nucleus; CnF, cuneiform nucleus; PPT, pedunculopontine tegmental nucleus; IC, inferior colliculus; PRT, pretectal region; SC, superior colliculus; DR, dorsal raphe nucleus; PAG, periaqueductal gray; PnR, pontine reticular nuclues; NLL, the nuclei of the lateral lemniscus; LDT, laterodorsal tegmental nucleus; PBN, parabrachial nucleus; SLD, sublateral dorsal nucleus; LC, locus coeruleus; Mo5, motor trigeminal nucleus; MnR, median raphe nucleus; Gi, gigantocellular reticular nucleus; Ve, vestibular nucleus; PCRt, parvicellular reticular nucleus, alpha part; Pr5, principal sensory trigeminal nucleus; CGPn, central gray of the pons; RMg, raphe magnus nucleus; IRT, intermediate reticular nucleus; DCN, deep cerebellar nuclues.

detected (**Figure 7C**). After the neurons at the injection site (identified by the existence of mCherry labeled cell bodies) were excluded, the projection to each brain area was quantified by the pixels of the axonal varicosities (**Figure 8**). Among the five major brain subdivisions that received LPT GABAergic projections (the cerebellum received almost no projection), the midbrain, diencephalon, and pons received the most LPT GABAergic projections, while few axons were detected in the telencephalon or medulla (**Figure 9**, left). The largest proportion of projections from LPT GABAergic neurons was found in the vlPAG (11.70 ± 0.86%), whereas the proportion of the LPT GABAergic varicosities in other subregions of the PAG were 1.42 ± 0.20% (lPAG) and 0.49 ± 0.11% (dPAG). The DR (6.39 ± 0.50%) and SC (4.60 ± 1.16%) in the midbrain, the parabrachial nuclei (PBN, 8.79 ± 1.04%) in the pons, and the LHA (8.08 ± 0.94%), medial hypothalamic area (5.09 ± 1.20%), and mediodorsal/central medial thalamic nucleus (MD/CM, 3.94 ± 0.55%) in the diencephalon also received strong projections from LPT GABAergic neurons.

### Reciprocal Connections Between the LPT and Other Nuclei

Comparing the broad distribution of the inputs and outputs of the LPT, we found that there was a high correlation between LPT afferents and efferents on a large scale of the whole brain. We found the CC was high between the afferent and efferent spatial distributions in all six major brain subdivisions (R = 0.867). Among the six major brain subdivisions, the diencephalon, midbrain, and pons revealed strong reciprocal connections with the LPT; they heavily received projections from and sent axonal fibers to LPT GABAergic neurons. Specifically, the midbrain provided a significantly higher proportion of inputs to the LPT, compared with the proportion of projections from the LPT to the midbrain (53.31 ± 6.33 versus 34.87 ± 2.84%, P = 0.038; Individual P-values indicate two-tailed Student's t-test comparisons of inputs and outputs for each brain region. In contrast, the pons and diencephalon had higher proportions of fibers arriving from the LPT than inputs going into the LPT (**Figure 10A**, pons: 19.94 ± 1.36 versus 12.62 ± 1.17%, P = 0.006; diencephalon: 24.03 ± 2.59 versus 14.22 ± 2.13%, P = 0.027 by two-tailed Student's t-test).

Our analysis of the proportion of inputs versus outputs in each region revealed that LPT GABAergic neurons had equally interactive connections to structures within the DR in the midbrain (**Figure 10B**; the proportion of the inputs to the LPT versus the outputs from the LPT, 6.04 ± 1.07 versus 6.39 ± 0.50%). However, there were many brain regions where connections with the LPT dominated in a unidirectional manner, such as the PAG and SC (**Figure 10B**). **Figures 10C,D** summarizes the afferents and efferents of LPT GABAergic

neurons in schematic form, reflecting the anatomical distribution of these innervations.

# DISCUSSION

By applying advanced viral tracing methods, we identified for the first time the whole-brain distribution of input and output projections of LPT GABAergic neurons. The afferents and efferents of LPT GABAergic neurons were mostly asymmetric in major nuclei such as the LHA, SC, PAG, PBN, and SNr. Only the DR had equally reciprocal connections with the LPT. In addition, LPT GABAergic neurons received projections from several brain areas related to sleep-wake regulation and motor activity. Finally, we found a strong connection between LPT GABAergic neurons and the limbic system, suggesting that emotional processing may be associated with LPT-mediated REM sleep and motor regulation.

### Technical Advances and Considerations

Genetically engineered RV has been widely used in neuroscience for its cell type-specific infection of neurons that does not affect passing neuronal tracts and its effective passage across known synaptic connections (Wickersham et al., 2007; Weissbourd et al., 2014). Accordingly, our data obtained using RV-mediated retrograde tracing in GAD2-IRES-Cre mice demonstrated that there are, in fact, direct projections from the SLD and LHA to LPT GABAergic neurons. This result not only corroborated a similar finding using CTB retrograde tracing (Lu et al., 2006; Clement et al., 2012), but also firstly identified GABAergic neurons in the LPT that received the inputs from the SLD and LHA. In addition, there are some differences between the distributions of the inputs to LPT GABAergic neurons compared with the inputs to LPT non-specific neurons. Clement and colleagues report that the ZI sends numerus projections to the LPT by CTB tracing (Clement et al., 2012). However, we found that the specific connection from ZI to LPT GABAergic neurons by RV-based tracing was weak, suggesting that the ZI may send most projections to glutamatergic neurons in the LPT.

On the other side, we built a detailed and quantitative map of LPT GABAergic neuron projections to the entire brain by AAV-mediated anterograde tracing, which was more specific and more restricted than the traditional methods. For example, although the non-specific projections from the LPT to the subthalamic nucleus (STN) were found by biotinylated dextran amine (BDA) (Bevan and Bolam, 1995), they didn't give the statistics of GABAergic projections from the LPT or full view of any other projecting nuclei. In our results, we quantitatively showed that LPT GABAergic neurons sent GABAergic varicosities in the STN (<0.1% of whole-brain varicosities). By visualizing the AAV reporter mCherry with DAB immunostaining, we avoided false positive

anterior commissure, posterior; scp, superior cerebellar peduncle.

signals, which are commonly observed after fluorescence immunohistochemistry (Oh et al., 2014). We chose to quantify projections by counting axonal varicosities instead of measuring fluorescence intensity, in order to avoid differences between thickness of brain sections or light exposure by the microscope.

### Morphological Findings

Due to the technical shortcomings of classical tracing methods, the complete neuronal connectivity of GABAergic neurons in the LPT was previously unknown. We used cell-type-specific anterograde and retrograde viral tracing to map the connectivity of LPT GABAergic neurons and found that their main inputs (≥20%) originated from the midbrain, whereas the brain regions that received projections from LPT GABAergic neurons were mainly distributed (≥20%) between the telencephalon, midbrain, and pons.

Among major nuclei connected to LPT GABAergic neurons, the DR had equally interactive connections to LPT GABAergic neurons (**Figure 10B**). Weissbourd and his colleagues found that the LPT projects to DR 5-HT neurons (Weissbourd et al., 2014). Here, we identified that the neuronal type in the LPT projected to the DR was GABAergic (**Figure 8**). The DR 5-HT neurons innervated LPT GABAergic neurons to form bidirectional connections (**Figure 5F**). This equally reciprocal LPT-DR connection may reveal a pathway for bidirectional feedback regulation of behavioral functions such as the antidepressant effect (Briley and Moret, 1993).

tegmental area; RRF, retrorubral field; RMC, red nucleus; CnF, cuneiform nucleus; PPT, pedunculopontine tegmental nucleus; DR, dorsal raphe nucleus; dPAG, dorsal periaqueductal gray; lPAG, lateral periaqueductal gray; vlPAG, ventrolateral periaqueductal gray; PnR, pontine reticular nuclues; LDT, laterodorsal tegmental nucleus; NLL, the nuclei of the lateral lemniscus; PBN, parabrachial nucleus; SLD, sublateral dorsal nucleus; LC, locus coeruleus; Mo5, motor trigeminal nucleus; MnR, median raphe nucleus; Gi, gigantocellular reticular nucleus; Ve, vestibular nucleus; PCRt, parvicellular reticular nucleus, alpha part; Pr5, principal sensory trigeminal nucleus; CGPn, central gray of the pons; RMg, raphe magnus nucleus; IRT, intermediate reticular nucleus.

From comparison of distribution among the major nuclei more unidirectionally connecting to LPT GABAergic neurons, we found that afferents to LPT GABAergic neurons concentrated on the SC and SNr, whereas GABAergic efferents from the LPT were mostly distributed in the PAG, PBN, and LHA. Most afferents to LPT GABAergic neurons originated from the SC, particularly the intermediate and deep layers of the SC (**Figures 3**, **6**). Although the projections from the SC to the LPT have been demonstrated in monkeys by anterograde tracing with BDA (Wang et al., 2013), the fact that these projections are predominantly GABAergic was unknown. Although previous studies indicated that the LPT received inhibitory projections from the SNr (Sherman et al., 2015), we identified the SNr as the second largest input to LPT GABAergic neurons. In addition, we showed that the main outputs of LPT GABAergic neurons project to the PAG, especially the vlPAG (**Figures 8**, **9**), while LPT-projecting PAG neurons were already known (Weissbourd et al., 2014). It was previously suggested that the majority of LPT cells provide GABAergic inhibition of PPT cholinergic neurons, including those cholinergic neurons that provide a major ascending pathway into the posterior LHA (Ford et al., 1995). However, our data revealed that the LHA directly received strong projections from LPT GABAergic neurons.

The proportion of inputs to the LPT among whole-brain inputs is different from that of the LPT. A pathway from the PSTN to the LPT was previously reported using phaseolus vulgaris leucoagglutinin anterograde tracing (Goto and Swanson, 2004), and by this method it is not possible to determine whether a majority or a minority of PSTN neurons project to LPT. Here, we reported the proportion of the inputs from the PSTN projecting to LPT GABAergic neurons among whole-brain inputs was 3.02 ± 1.09%, which was about 44% colocalized with CR neurons (**Figure 5C**). Combined with dense inputs from the CeA in the forebrain, our results suggested a strong ability of the LPT to integrate behavioral information from the PSTN and CeA such as feeding and fear behaviors (Zimmerman et al., 2007; Chometton et al., 2016). Although the LPT has been shown to provide strong GABAergic projections to the dorsocaudal region of the pontine tegmentum, including the PBN, LC, and SLD (Hayashi et al., 2015), our finding of dense, LPT-originating, GABAergic axonal varicosities in the SNc of the midbrain suggested that the LPT also transmitted strong information to the SNc.

# Functional Implications

LPT GABAergic neurons have strong connections with brain regions related to sleep-wake regulation. Firstly, the output to the vlPAG is known to gate REM sleep and the ultradian rhythm of REM/NREM alternation (Weber et al., 2018). In addition, vlPAG/DpMe GABAergic neurons were shown to be important in sleep regulation (Hayashi et al., 2015; Weber and Dan, 2016) and thus, bidirectional projections between the LPT and

subdivisions of the whole brain. LPT GABAergic inputs are colored red and the outputs are shaded in orange. The data represent the mean ± SEM, with n = 4 for each group. Two-tailed Student's t-tests indicated significant (∗P < 0.05, ∗∗P < 0.01) differences between the inputs and outputs of LPT GABAergic neurons in the same subdivision. (B) The percentage of input versus percentage of output in each region of typical nuclei that had strong connections with the LPT. The proportion of inputs or outputs ≥ 5% are indicated with open circles. (C,D) Schematic diagrams presenting the afferents and efferents of LPT GABAergic neurons. BNST, bed nucleus of the stria terminalis; STRv, stratum ventral region; PALd, pallidum, dorsal region; PALv, pallidum, ventral region; CeA, central amygdaloid nucleus; BMA, basomedial amygdalar nucleus; POA, preoptic area; LHA, lateral hypothalamic area; MHA, medial hypothalamic area; PSTN, parasubthalamic nucleus; MD/CM, mediodorsal/central media thalamic nucleus; PVT, paraventricular thalamic nucleus; PF/Po, parafascicular/ posterior thalamic nucleus; SNc, substantia nigra pars compacta; ZI, zona incerta; SNr, substantia nigra pars reticulata; VTA, ventral tegmental area; DpMe, deep mesencephalic nucleus; SC, superior colliculus; RRF, retrorubral field; CnF, cuneiform nucleus; PPT, pedunculopontine tegmental nucleus; DR, dorsal raphe nucleus; PAG, periaqueductal gray; IC, inferior colliculus; PnR, pontine reticular nuclues; NLL, the nuclei of the lateral lemniscus; PBN, parabrachial nucleus; SLD, sublateral dorsal nucleus; LC, locus coeruleus; LDT, laterodorsal tegmental nucleus; IRT, intermediate reticular nucleus; Gi, gigantocellular reticular nucleus; PCRt, parvicellular reticular nucleus, alpha part.

vlPAG may be important for sleep regulation. Strong inhibitory projections from the LPT to the arousal-promoting PBN may also indicate a role of the LPT in sleep/wake control and thermoregulation (Fuller et al., 2011; Kaur et al., 2013; Nakamura, 2018; Wang et al., 2018). Furthermore, we and others identified SLD-projecting LPT GABAergic neurons (Boissard et al., 2003), which may play important roles in REM sleep regulation and related sleep disorders, such as narcolepsy or REM sleep behavior disorder. The direct connections we observed between the LPT and the hypothalamus suggested that LPT GABAergic neurons

might be involved in regulation of hypothalamus functions, such as REM sleep, arousal, feeding behaviors, body temperature, and sense of pain (Clement et al., 2012; Li et al., 2014). Finally, another study showed that the neuronal activity of the posterior lateral hypothalamus containing the PSTN was related to the wake and REM sleep (Steininger et al., 1999). In the present study, the PSTN sent abundant projections to the LPT GABAergic neurons, suggesting that the pathway of the PSTN to LPT GABA neurons might be involved in sleep-wake regulation. Besides that, chemical activation of PSTN neurons elicited depressor as well as bradycardia (Ciriello et al., 2008), and hedonic tastes increased c-Fos expression of the PSTN in rats (Chometton et al., 2016), suggesting that cardiovascular regulation and food intake were probably mediated by the PSTN innervated LPT GABAergic neurons.

Anatomical connectivity of LPT GABAergic neurons with the SC, SN, and PAG in the midbrain may implicate a role of LPT in the regulation of motor behaviors. Our data revealed the existence of asymmetrically reciprocal circuits between the SC and LPT GABAergic neurons for the transmission of saccadic information. The LPT was suggested to be involved in the control of eye movements, due to a causal link that was reported between ocular flutter and small lesions in the right upper pontine tegmentum containing the LPT (Tsutsumi et al., 2009). Moreover, strong connections between the LPT and SN suggested a role in regulating locomotor behavior (Deng et al., 2016; Lintz and Felsen, 2016; Barchini et al., 2018; Weber et al., 2018). A study previously proposed that projections of SNr neurons to PPT glutamatergic neurons primarily regulated motor activity (Roseberry et al., 2016); however, the strong projections from the SNr to LPT GABAergic neurons may also implicate inhibitory LPT neurons in some aspects of motor control such as saccadic eye movement (Sato and Hikosaka, 2002). In contrast, SNcprojecting LPT GABAergic neurons were shown to be involved in motor and reward control (Pioli et al., 2008). Our finding that SNc neurons also modestly projected to LPT GABAergic neurons could indicate a sophisticated feedback mechanism between SNc and LPT GABAergic neurons.

Finally, the inputs from the limbic system, including the CeA and BNST, suggested that LPT GABAergic neurons may be regulated by neural circuits involved in stress and fear processing (Zhao and Davis, 2004). Cataplexy, a common symptom of narcolepsy, is known to be triggered by strong emotions associated with GABAergic neurons in the limbic system and may also result from the disturbance of REM sleep atonia (related to the LPT/vlPAG/LC) into wakefulness (Snow et al., 2017). Our data provided evidence that LPT GABAergic neurons directly received projections from BNST GABAergic neurons (**Figures 4**, **5D**) and CeA GABAergic neurons (**Figures 4**, **5E**). Therefore, we confirmed the anatomical connectivity of the limbic system with LPT GABAergic neurons that control REM sleep atonia, as hypothesized by Fuller (Peever and Fuller, 2017).

### Comparison With Areas Near the LPT

Although the vlPAG and PPT are in close proximity to the LPT, the connectivity of these GABAergic neurons is different from that of the LPT. Whereas vlPAG GABAergic neurons primarily receive inputs from the ventral medulla, the SLD, and the LHA (Boissard et al., 2003; Clement et al., 2012; Weber et al., 2015), the LPT is mainly innervated by the SC and SNr. In contrast to LPT GABAergic neurons, the vlPAG has major projections to the RMg and nearby areas, commonly known as the rostral ventral medulla (Bowman et al., 2013). Moreover, PPT GABAergic neurons are rarely innervated by the SNr (Roseberry et al., 2016), while LPT GABAergic neurons are strongly innervated by this area.

There are also clear differences in the physiological and pathological functions of the LPT, vlPAG, and PPT. GABAergic neurons in the vlPAG are important for the sleep-wake cycle, micturition, and nociception. In addition, optogenetic studies revealed that vlPAG GABAergic neurons regulated non-REM and REM sleep (Weber et al., 2018). Other studies based on genetically engineered systems showed that activation of vlPAG GABAergic neurons delayed detrusor contraction and inhibited voiding (Zare et al., 2018). In contrast, PPT GABAergic neurons are mostly involved in gait and balance regulation, a function that is not identified in LPT GABAergic neurons. For example, loss of rostral PPT GABAergic neurons was reported in Parkinson's disease patients with gait and balance disorders (Sebille et al., 2018). In spite of the close proximity of the vlPAG, PPT, and LPT GABAergic neurons, it is likely that these areas have distinguishing physiological and pathological roles.

# CONCLUSION

We mapped, for the first time, the afferents and efferents of LPT GABAergic neurons and found them to be extensively interconnected with other brain areas. This suggested a vital role of LPT neurons in a wide range of physiological and pathological functions, especially sleep-wake regulation and locomotor control. Our anatomical data could be useful for future functional studies of the brain and also provide a structural basis to understand neurological disorders.

# ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the China Regulations on the Administration of Laboratory Animals, the Decree No. 2 of National Science and Technology Commission of the People's Republic of China and all animal procedures were approved by the Medical Experimental Animal Administrative Committee of the School of Basic Medical Sciences (Permit No. 20140226-024), Fudan University (Shanghai, China).

# AUTHOR CONTRIBUTIONS

Z-KC, X-SY, and HD designed and performed the experiments, analyzed the data, and wrote the manuscript. Y-FW performed the experiments and analyzed the data.

W-MQ and G-HC conceived the experiments and wrote the manuscript. MH provided GAD2-IRES-Cre mice. Z-LH conceived the experiments, analyzed the data, and wrote the manuscript.

### FUNDING

This study was supported in part by grants-in-aid for scientific research from the National Natural Science Foundation of China

### REFERENCES


(Grant Nos. 81420108015, 31530035, 31871072, 31671099, and 31571103) and the National Basic Research Program of China (Grant No. 2015CB856401).

### ACKNOWLEDGMENTS

We thank Michael Lazarus [International Institute for Integrative Sleep Medicine (WPI-IIIS), University of Tsukuba, Tsukuba, Japan] for editing this manuscript.

to disease measures in epilepsy. Neurology 91, e67–e77. doi: 10.1212/WNL. 0000000000005733


double-labeling immunoelectron microscopy. Synapse 43, 42–50. doi: 10.1002/ syn.10017



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Chen, Yuan, Dong, Wu, Chen, He, Qu and Huang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Temperature Dependence of Sleep

Edward C. Harding<sup>1</sup> , Nicholas P. Franks1,2,3 \* and William Wisden1,2,3 \*

<sup>1</sup> Department of Life Sciences, Imperial College London, London, United Kingdom, <sup>2</sup> Centre for Neurotechnology, Imperial College London, London, United Kingdom, <sup>3</sup> UK Dementia Research Institute, Imperial College London, London, United Kingdom

Mammals have evolved a range of behavioural and neurological mechanisms that coordinate cycles of thermoregulation and sleep. Whether diurnal or nocturnal, sleep onset and a reduction in core temperature occur together. Non-rapid eye movement (NREM) sleep episodes are also accompanied by core and brain cooling. Thermoregulatory behaviours, like nest building and curling up, accompany this circadian temperature decline in preparation for sleeping. This could be a matter of simply comfort as animals seek warmth to compensate for lower temperatures. However, in both humans and other mammals, direct skin warming can shorten sleeplatency and promote NREM sleep. We discuss the evidence that body cooling and sleep are more fundamentally connected and that thermoregulatory behaviours, prior to sleep, form warm microclimates that accelerate NREM directly through neuronal circuits. Paradoxically, this warmth might also induce vasodilation and body cooling. In this way, warmth seeking and nesting behaviour might enhance the circadian cycle by activating specific circuits that link NREM initiation to body cooling. We suggest that these circuits explain why NREM onset is most likely when core temperature is at its steepest rate of decline and why transitions to NREM are accompanied by a decrease in brain temperature. This connection may have implications for energy homeostasis and the function of sleep.

Keywords: sleep-wake cycle, thermoregulation, thermoregulatory behaviour, circadian, preoptic area, anterior hypothalamus, energy balance, nesting

# INTRODUCTION

In all mammals, sleep appears to be indispensable and coincides with a conserved circadian temperature rhythm. When our core and brain temperatures are in rapid decline we are most likely to choose to sleep, and if we dissociate from this cycle of body cooling we experience insomnia (Hayward, 1968; Campbell and Broughton, 1994; Lack et al., 2008). Here, we review the evidence that thermoregulatory mechanisms are fundamental to sleep and consider the neuronal circuits that connect these two physiologies. These circuits use the presence of warm microclimates to gate sleep and may enhance circadian body cooling as our first non-rapid eye movement (NREM) bout approaches. The same neurons directly link NREM initiation to body cooling and may explain why transitions from wakefulness to NREM sleep, across the sleep cycle, are immediately followed by a decrease in brain temperature, whilst transitions back to REM or WAKE are accompanied by rewarming (Alföldi et al., 1990; Landolt et al., 1995). The partitioning of brain cooling during NREM sleep and the coordination of the circadian core temperature rhythm are important for effective sleep.

### Edited by:

Yu Hayashi, University of Tsukuba, Japan

### Reviewed by:

Genshiro A. Sunagawa, RIKEN Center for Biosystems Dynamics Research, Japan Ramalingam Vetrivelan, Beth Israel Deaconess Medical Center, Harvard Medical School, United States

### \*Correspondence:

Nicholas P. Franks n.franks@imperial.ac.uk William Wisden w.wisden@imperial.ac.uk

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 15 January 2019 Accepted: 22 March 2019 Published: 24 April 2019

### Citation:

Harding EC, Franks NP and Wisden W (2019) The Temperature Dependence of Sleep. Front. Neurosci. 13:336. doi: 10.3389/fnins.2019.00336

**149**

This may have particular consequences for energy homeostasis and could open a window on sleep function.

### PREPARATION FOR SLEEP IS A THERMOREGULATORY BEHAVIOUR

Mammals have a range of thermoregulatory behaviours that allow adaptation to environmental temperature fluctuations throughout the day, but these are most visible in the preparations for sleep (Peever, 2018). These behaviours include warmth and shelter seeking, nest building, curling up and huddling (see **Figure 1A**). Mice that are inactive or sleeping are much more likely to do so in contact with nesting material (Gaskill et al., 2011). As small rodents they demonstrate surprisingly sophisticated thermal adaptations. As environmental temperature decreases, nest quality rises to compensate and results in measurable improvements in insulation (Gaskill et al., 2013a). When they can, mice engage in huddling with group members (Gaskill et al., 2011, 2012; Gordon et al., 2014). They also have a clear thermal preference during the sleep phase (lights on), choosing warmer environments approaching thermoneutrality (27–30◦C) and minimising energy expenditure (Gordon et al., 1998; Gaskill et al., 2012). These behaviours align the circadian temperature decline with the light and dark cycle and sleep onset. An example of the circadian core temperature cycle over several days can be seen in **Figure 1B**. The decline in core temperature intersects the light–dark cycle and changes over a range of about 2◦C at the transition from the active phase of the mice (lights off) to the sleep phase (lights on) (**Figures 1C,D**).

Ambient temperature is a critical determinant of energy expenditure, and failure to carry out thermal defence behaviours has consequences for energy homeostasis (Yu et al., 2018). Fur removal in Siberian hamsters, for example, increases food consumption by almost a quarter; whereas, in cold conditions, group huddling or the provision of nesting material can reduce food consumption by 15–20%, respectively (Kauffman et al., 2003; Batavia et al., 2010). Similarly, provision of nesting material at temperatures below thermoneutrality increases breeding efficiency giving larger litters, higher pup weight and reduced pup mortality (Gaskill et al., 2013b).

Thermoregulatory behaviour prior to sleeping is a core part of maintaining energy balance, at least in smaller mammals, where the consequence of thermal inefficiency is an increased

FIGURE 1 | Sleep preparation is a thermoregulatory behaviour (A) shows typical nesting behaviour in four species. Mouse nesting (Mus musculus, C57Bl6/J), house cat (Felis catus) curling up, nest building in the chimpanzee (Pan troglodytes verus) and bedding (Homo sapiens). (B) Example of circadian temperature cycle over 6 days in a male C57Bl6/J mouse. (C) Average of transitions from the same mouse over 16 consecutive days 2 h before and after the light change. (D) Minimum temperature (n = 21) during light phase compared to minimum (n = 21) and maximum temperature (n = 16) in the dark phase, plotted as change from zero for a group of male C57Bl6/J mice. Data shown in (B–D) is from (Harding et al., unpublished). All images used with permission or copyright clearance. The nesting chimpanzee photo credit: Kathelijne Koops. The nesting cat photo credit: Isobel Harding, the sleeping human is available under CC0-1.0 universal and the nesting mouse is adapted from Deacon (2006).

need for food. In larger mammals, however, the drive towards thermal preparation for sleep is no weaker. Chimpanzees and other primates select their arboreal sleeping sites (**Figure 1A**), at least partly, on thermal characteristics, and during colder weather even adjust their nest sites to be more insulating (Koops et al., 2012; Samson and Hunt, 2012; Stewart et al., 2018). In addition, humans actively regulate temperature during sleep by unconsciously increasing their exposed surface area as ambient temperatures rise. In optimal room temperatures, approximately 19–21◦C, we attempt to establish skin microclimates between 31 and 35◦C and deviation from this range has a negative influence on sleep (**Figure 2A**) (Muzet et al., 1984; Okamoto-Mizuno et al., 2003; Raymann et al., 2005). A key factor in using microclimates is that, at least in humans, it cannot be replaced by ambient warming at the same temperature, perhaps because it disrupts the self-adjustment required over the course of the night (Muzet et al., 1984; Raymann et al., 2008).

In summary, thermoregulatory behaviours prior to sleep are conserved across mammalian species suggesting they are not just a matter of comfort and may have a more functional role in sleep initiation and maintenance.

### THE "WARM BATH EFFECT"

In humans, immersion in hot water prior to, but not immediately before, the sleep period decreases sleep latency and increases sleep depth. This is the known as the 'Warm Bath Effect' (Horne and Reid, 1985; Parmeggiani, 1987; Bunnell et al., 1988; Shapiro et al., 1989; Jordan et al., 1990; Dorsey et al., 1999). In fact, warming for up to 4 h, between 1 and 8 h before to going to bed, increases slow wave sleep (SWS), increases NREM consolidation and decreases REM sleep. This effect embodies a key connection between temperature and sleep. Warming, at the right time, is causatively associated with sleep initiation. However, sleep initiation occurs within the decline of circadian temperature and NREM is associated with further reductions in the temperature of both core and the brain (Alföldi et al., 1990; Landolt et al., 1995; Kräuchi and Wirz-Justice, 2001). Many sleep studies have attempted to reconcile this counter-intuitive relationship to explain two conditions: How warming might initiate sleep and be compatible with body cooling, and how we might encounter this warming under 'everyday' conditions.

Optimal ambient temperatures, in combination with bedding, appear to be crucial for efficient sleep onset in humans (Haskell et al., 1981; Okamoto-Mizuno et al., 2003; Raymann et al., 2008). Responses to external temperature also appear to be important as the extent of vasodilation, particularly in the hands and feet (distal-skin), is a good predictor of sleep initiation (Krauchi et al., 1999). This vasodilation is usually considered part of the circadian temperature decline and is observed up to 2 h prior to the start of the first sleep episode, during the wake phase (Krauchi et al., 2000). As core temperature falls it coincides with a decrease in self-assessed alertness (Czeisler et al., 1980; van den Heuvel et al., 1998). In experiments where participants 'selfselected' their bed time, subjects were most likely to select a moment when the body temperature was at its maximum rate of decline (Campbell and Broughton, 1994). As sleep approaches, core temperature and heart rate drop, and their steepest decline intersects 'lights-off' and sleep onset (**Figure 2B**). At this point the proximal-to-distal temperature gradient is as much as 1.5◦C, but as core temperature falls the gradient reduces to about 0.5◦C; a new cooler set-point is reached just after sleep onset. The lowest core temperature is observed about 2 h after 'lights-off' and sleep onset in Homo sapiens (Krauchi et al., 2000). Under natural conditions, increased circulating melatonin also coincides with declining core temperature prior to sleep onset (Krauchi et al., 1997; Krauchi et al., 2006; Logan and McClung, 2019).

Studying the temperature dependence of sleep in people has always been confounded by our ability to manipulate

mechanisms to induce sleep and peripheral vasodilation. (B) Distal-to-proximal gradient and core temperature decline predict sleep onset (adapted from Krauchi et al., 2000).

our environment and escape daily fluctuations in light and temperature. To get around this, Yetish et al. (2015) looked at sleep in three geographically distinct pre-industrial societies. They found that sleep onset coincided most strongly with a reduction in environmental temperature. Sleep was most often initiated after dark and the entire sleep period took place as ambient temperature was declining. Awakening also occurred before dawn, as ambient temperature reached its lowest point, and coincided with vasoconstriction, as measured by finger temperature (Yetish et al., 2015). A change in temperature in the fingers is a good measure of change in blood flow, and so it seems likely that these subjects started sleep in a state of vasodilation that was progressively replaced by vasoconstriction until waking (Rubinstein and Sessler, 1990; van Marken Lichtenbelt et al., 2006). A similar result was also observed by Han et al. (2018), under sleep-laboratory conditions, with high numbers of skintemperature sensors distributed across the body. These indicated progressive vasodilation from sleep onset to waking. However, this was mostly represented in the torso, and the hands and feet were not recorded (Han et al., 2018).

The circadian cycle and the onset of the first NREM episode are strongly linked. If sleep onset is postponed by sleep deprivation, then the circadian temperature rhythm is disrupted. Similarly, a delay in core temperature decline of more than 2 h, is observed in patients with delayed sleep phase disorders (DSPDs) (Ozaki et al., 1996; van den Heuvel et al., 1998; Watanabe et al., 2003). Disruption of the peripheral vasodilatory response is sufficient to disrupt sleep. For instance, those with difficulty in peripheral vasodilation (vasospastic disorders) have longer sleep latencies than healthy controls (Pache et al., 2001). Narcoleptic patients also have a strongly altered proximalto-distal skin temperature gradient during daytime waking (Fronczek et al., 2006). But manipulating the proximal-todistal relationship can change sleep propensity. Warming of the core (proximal-skin) by less than 1◦C, easily within the range encountered within the circadian day, is sufficient to shorten sleep latency (Raymann et al., 2005). Temperature manipulation can also selectively and predictably alter vigilance states in patients with narcolepsy (Fronczek et al., 2008a,b). Additional work in the clinic has shown that neonates are three times more likely to fall asleep within 30 min, if their distal-to-proximal skin gradient is greater than 2.5◦C (Abe and Kodama, 2015). Distal vasodilation and higher foot temperature in preterm neonates is also correlated with shorter wake bouts (Barcat et al., 2017).

Understanding how warmth might be encountered on a daily basis to precipitate these changes that initiate sleep and vasodilation is crucial. But it seems that the 'warm bath effect' is more subtle than previously thought. Raymann et al. (2008) have extended the warming paradigm with the aid of a custom-made 'thermosuit' for the manipulation of skin temperature. Small changes in skin temperature of only 0.4◦C (in the 31–35 range), can shorten sleep latencies without altering core temperature. They can even encourage deeper sleep in more challenging patient groups, such as elderly insomniacs (Raymann et al., 2008). This latter group was particularly susceptible to this thermal management, supporting the hypothesis that sleep difficulties in the elderly relate to deficits in normal thermoregulation (Raymann and Van Someren, 2008).

In summary, humans and other mammals show thermoregulatory behaviour in preparation for sleep, including curling up, using bedding and nest building. This may generate a microclimate of warmth around the skin that enables entry into sleep while facilitating vasodilation in the 'distal' hands and feet. This vasodilation may prepare the 'proximal' core for the cooler and inactive phase of the circadian cycle. This warming persists through the night to maintain a sleep-permissive state that also permits selective vasodilation in NREM and constriction in REM and wake. It does so whilst maximising thermal efficiency of the core. The reasons that body cooling and sleep onset coincide are not clear. Body and brain cooling per se has not been shown to initiate NREM but is instead a consequence of vasodilation. We might expect that an upstream mechanism in the brain coordinates both NREM onset and vasodilation and in the next section we discuss how this might function (Van Someren, 2000).

### NEURONAL CONTROL OF THERMOGENESIS AND ITS INFLUENCE ON SLEEP

Sleep is a fundamental physiological process that is widely believed to be essential for life but its vital function has yet to be identified. The neuronal circuits that control sleep need to integrate information from at least two distinct inputs. According to current thinking, these are known as Process C and Process S, the circadian and the homeostatic input, respectively, and are part of the two-process model (Borbély, 1982). Transitions from wake to NREM and REM sleep are carried out by neurons that respond to cues from the homeostatic drive, that tracks the time spent awake, as well as more salient cues from the circadian zeitgeber, via the suprachiasmatic nucleus (SCN). The homeostatic process tracks the duration of the waking period and dissipates this load during sleep. However, as we have seen, sleep onset is also gated by other inputs: ambient temperature, as well as levels of satiety, mating opportunities and the need to escape predators all determine the appropriate moment for NREM onset (Borbély, 1982; Borbély et al., 2016; Eban-Rothschild et al., 2017; Logan and McClung, 2019). Neurons that influence sleep are widely distributed throughout the brain. This may allow the integration of behavioural and autonomic inputs onto the classical homeostatic and circadian sleep drive. For example, inhibition of ventral tegmental area (VTA) dopamine neurons promotes both nesting behaviour and sleep initiation (Eban-Rothschild et al., 2016). While the homeostatic drive promotes sleep following prolonged wakefulness, the circadian, the behavioural and the autonomic inputs are permissive conditions for sleep onset (see **Figure 3**). These four inputs work together to gate sleep.

Although sleep onset and the regulation of sleep transitions may involve multiple nuclei in the brain, one area has been historically associated with NREM onset. The preoptic hypothalamus (PO) is a key site for NREM initiation but is also considered an integrator for thermoregulatory information,

FIGURE 3 | Sensory and homeostatic inputs that gate sleep. Sleep onset is determined by four competing inputs: the homeostatic drive to sleep and three permissive conditions that relate to sleep timing, the behavioural input, the circadian input and the autonomic input. Endocrine inputs are also a key part of each category. Ghrelin and leptin are important for sensing of hunger/satiety, respectively, while melatonin is a key component of the circadian rhythm. Adenosine and NO may form part of the homeostatic input. (Top - Factors promoting wakefulness) Circadian cues are permissive for wake and homeostatic pressure to sleep is low. Behavioural factors also promote wakefulness and autonomic inputs are not permissive for sleep. Wake promoting nuclei drive cortical and thalamic excitability, whilst inhibiting sleep-prompting areas such as PO and vPAG. Behavioural needs of food and reproduction overcome those of sleep and thermal comfort. Behavioural inputs are also wake-promoting and may integrate this information in the VTA. Hormonal inputs, such as ghrelin, are detected in the ARC and are sleep-permissive. Autonomic signals, such as ambient temperature, are relayed via the spinal cord and pass through the LPb to the PO for integration. Circuits detecting environmental warmth are not active, vasoconstriction dominates and BAT is active. AgRP neurons signal hunger and inhibit sleep. (Bottom – Factors promoting NREM sleep) Circadian cues are now permissive for sleep and homeostatic pressure to sleep is high. Behavioural factors also promote sleep and autonomic inputs are permissive for sleep. On seeking shelter and warmth and having eaten, sleep is permitted. Autonomic signals, such as ambient temperature, are relayed via the spine and pass through the LPb to the PO for integration. NOS1-glutamate neurons are activated by skin warmth and initiate both NREM and body cooling. Activation of vasodilatory and BAT downregulation circuits is via NOS1 projections to LPO GABAergic neurons or via direct projections to DMH and rRPA/RVLM. Behavioural inputs are now sleep promoting and may integrate this information in the VTA. Hormonal inputs, such as leptin, are detected in the ARC and are sleep permissive. POMC neurons detect satiety and are permissive for sleep. NO, nitric oxide; NOS1, nitric oxide synthase-1; PO, preoptic area; LPO, lateral preoptic area; vPAG, ventral periaqueductal grey; TMN, tuberomammillary nucleus; VTA, ventral tegmental area; ARC, arcuate nucleus; LPb, lateral parabrachial; LC, locus coeruleus; DR, dorsal raphe; BAT, brown adipose tissue; AgRP, agouti-related peptide; DMH, dorsal medial hypothalamus; rRPA, rostral raphe pallidus; RVLM, rostral ventrolateral medulla; POMC, pro-opiomelanocortin (Leshan et al., 2012; Eban-Rothschild et al., 2016; Weber and Dan, 2016; Yu et al., 2016; Goldstein et al., 2018; Harding et al., 2018; Yu et al., 2019).

including cold and warm-defence (Szymusiak et al., 2007). It consists primarily of the median (MnPO), the medial (MPO) and lateral (LPO) areas that are associated with a large array of functions from sleep to parental behaviour.

Preoptic circuits have been proposed as the mechanistic connection between whole body warming and sleep induction (Morairty et al., 1993). The simplest version of this idea is that warming induces activity in sleep-promoting neurons. Indeed, warm stimuli are well known to increase activity in the PO (e.g., as seen by c-FOS expression) (Scammell et al., 1993; Gong et al., 2000). Consistent with this idea, lesions in the PO of the cat disrupt both warm-defence behaviour and reduce total sleep (Szymusiak et al., 1991). Only significant warming of these cats was able to rescue normal sleep amounts, possibly through compensation or mechanisms outside the PO (Szymusiak and McGinty, 1986). In rats, PO lesions alter thermal preference behaviour which subsequently converges on warmer temperatures (∼30◦C) that promote sleep recovery (Ray et al., 2005). In crucial experiments, using a 'thermode' implanted into the PO, warming, but not cooling, increases delta power in the EEG (Roberts and Robinson, 1969; Glotzbach and Heller, 1976; McGinty et al., 1994). To characterise preoptic neurons in this role Alam et al. (1995) repeated this protocol using an implanted microdrive and recorded the properties of preoptic neurons. Remarkably, 21% were thermosensitive and these could be further divided into two groups – cold-sensitive neurons (CSNs) and warm-sensitive neurons (WSNs). About 60% of WSN also increased their activity during NREM (Alam et al., 1995). During warming in the rat brain, they could inhibit important arousal nuclei including dorsal raphe and posterior hypothalamic neurons (Krilowicz et al., 1994; Guzmán-Marín et al., 2000; Steininger et al., 2001). In a detailed analysis of MnPO neurons by Suntsova et al. (2002), more than 75% demonstrated properties that may facilitate NREM sleep induction. This included a gradual increase in firing into, and peaking during, NREM sleep and, unexpectedly, even higher firing rates during REM sleep (Suntsova et al., 2002). Mapping of neuronal projections using retrograde and anterograde tracers has confirmed that the MnPO sends dense innervations to wake-promoting regions and is well placed to influence wake-to-sleep transitions by modulation of the lateral preoptic, lateral hypothalamus and dorsal raphe (Uschakov et al., 2007). Lastly, some MnPO neurons express c-FOS in response to sleep deprivation and may also send projections to the LPO (Chou et al., 2002; Zhang et al., 2015).

WSNs can directly sense brain temperature and are proposed to be modulated by pyrogens such as prostaglandin E2 (Scammell et al., 1996; Lazarus et al., 2007). A population of glutamatergic neurons in the midline PO express the transient receptor potential member 2 (TRPM2) channel, enabling direct warmsensing of local brain temperature. These could function to carry out heat defence but can also modulate the response to fever (Song et al., 2016).

With the exception of fever, it is unclear if skin warming could induce an increase in brain temperature that could be sensed by WSNs (Tan et al., 2016; Siemens and Kamm, 2018). Instead, a synaptic pathway is more likely. Neurons that receive afferent temperature information, but are not direct temperature 'sensors', have been distinguished by the term 'warm-activated' neurons (Tan and Knight, 2018). The MnPO and MPO hypothalamus receives sensory afferents conveying thermal information from the skin (Hammel, 1968; Boulant and Gonzalez, 1977; Morrison and Nakamura, 2011) (**Figure 4**). Sensory neurons relay ambient temperature information via the spinal cord to glutamatergic relay neurons, and on to sub-regions of the lateral parabrachial nucleus (LPb). The LPb may also receive information from other parts of the body such as the viscera, and then pass these signals onto the MnPO and MPO regions (Nakamura and Morrison, 2008, 2010). At the first point of integration, the glutamatergic neurons conveying excessive warmth synapse at glutamatergic neurons in the MnPO, whose output initiates cooling by promoting vasodilation and switching off brown fat thermogenesis (Morrison and Nakamura, 2011) (**Figure 4**). What are these neurons in the PO that respond to warming? Recent work using GCaMP6 photometry has shown that these neurons can response to external warm challenges between 30

ion channel. NO, nitric oxide; NOS1, nitric oxide synthase-1; PO, preoptic area; LPO, lateral preoptic area; vPAG, ventral periaqueductal grey; TMN, tuberomammillary nucleus; ARC, arcuate nucleus; LPb, lateral parabrachial; BAT, brown adipose tissue; AgRP, agouti-related peptide; POMC, pro-opiomelanocortin; DMH, dorsal medial hypothalamus; rRPA, rostral raphe pallidus; RVLM, rostral ventrolateral medulla; TRPM2, transient receptor potential cation channel; TRPV1, transient receptor potential cation channel vallinoid-1; GAL, Galanin (Leshan et al., 2012; Weber and Dan, 2016; Yu et al., 2016; Goldstein et al., 2018; Harding et al., 2018; Jeong et al., 2018; Tan and Knight, 2018; Yu et al., 2019).

through the leptin Rb, as do AgRP and POMC neurons. They, or a separate local population, may also respond to changes in brain temperature through the TRPM2

and 40◦C and RNA sequencing has identified them as expressing pituitary adenylate cyclase-activating polypeptide (PACAP) and brain-derived neurotrophic factor (BDNF) (Tan et al., 2016). These neurons are predominantly GABAeric and can induce hypothermia when activated. They function, at least in part, by inhibiting dorsal medial hypothalamus (DMH) glutamatergic neurons that stimulate BAT thermogenesis (Tan et al., 2016). A further population of GABAergic neurons, that act through

the DMH, have also been discovered in the nearby ventral LPO (Zhao et al., 2017).

The PO is highly diverse region with many overlapping populations, but only some of these neurons have been functionally characterised (Moffitt et al., 2018). For example, GABAergic-galanin neurons are associated with both sleep and parental behaviour, but populations of galanin-glutamate neurons also exist (Sherin et al., 1998; Wu et al., 2014; Moffitt et al., 2018). PACAP/BDNF, TRPM2-glutamate and nitrergicglutamate neurons are associated with warm-defence and fever but many other subpopulations exist (Song et al., 2016; Tan et al., 2016; Harding et al., 2018). Whilst the latter is also associated with sleep induction, populations of GABAergic-nitrergic neurons have been found but are uncharacterised (Harding et al., 2018; Moffitt et al., 2018). Given extensive diversity in PO neuronal subtypes (Moffitt et al., 2018), methods such as c-FOS-dependent activity tagging, that allow functional dissection of specific circuits from the surrounding milieu, are particularly important (Zhang et al., 2015). The PO area, including both MPO and LPO responds to recovery sleep, the sleep following sleep deprivation, by expressing c-FOS. The same regions are excited by the, α2A-adrenergic agonist and sedative, dexmedetomidine (DEX) (Zhang et al., 2015). To understand whether these physiologies shared the same circuitry Zhang et al. (2015) used c-FOSdependent activity-tagging to separate the neurons activated by recovery sleep or DEX from other PO neurons that respond to a variety of environmental and homeostatic stimuli. These neurons expressed an excitatory hM3d<sup>q</sup> DREADD receptor such that, when these mice were given clozapine N-oxide, only this unique ensemble was activated. This resulted in consolidated NREM sleep, consistent with recovery sleep or sedation. However, PO ensembles, tagged by either recovery sleep or DEX, also induced hypothermia (Zhang et al., 2015). In fact, essentially all sedatives and general anaesthetics used clinically induce coreto-peripheral heat redistribution from vasodilation and, without warming, hypothermia (Díaz and Becker, 2010; Sessler, 2016). This suggests that core PO circuitry could link natural sleep induction, the induction of body cooling and the mechanisms of sedative class drugs.

We hypothesised that thermoregulatory circuits themselves might have an important role in facilitating sleep. This would also explain the propensity of either external or direct PO warming to induce NREM sleep. We again used activitytagging, but this time labelled only those preoptic ensembles that received warm sensory information. Reactivation of these 'warmtagged' neurons produced simultaneous NREM and body cooling (Harding et al., 2018). These neurons expressed a mixture of cell type markers including the vesicular glutamate transporter 2 (VGLUT2), glutamate decarboxylase (GAD67), and nitric oxide synthase 1 (NOS1). When activity-tagging was repeated in NOS1- CRE mice, these also experienced simultaneous NREM and body cooling. However, when it was repeated in vGAT-CRE mice only NREM and a little body cooling was observed. As these NOS1 neurons express VGLUT2, our data suggest a distinct nitrergicglutamatergic circuit for linking thermal sensory information to NREM-onset that may reside upstream of a GABAergic sleep 'switch' (Harding et al., 2018). In this circuitry, external warmth is a permissive state for NREM initiation. Without this sensory input NREM onset is inhibited. We think this is consistent with data showing that external warming promotes sleep in humans and animals, whilst also providing a possible mechanism for why mammals seek nesting sites: to produce microclimates of skin warmth that permit sleep. We don't yet know if NOS1 neurons utilise nitric oxide (NO) in synaptic transmission. However, NO is implicated in modulating arousal in other areas of the brain (Gerashchenko et al., 2008; Kalinchuk et al., 2010; Cespuglio et al., 2012; Morairty et al., 2013; Yu et al., 2019).

What is downstream of the MnPO/MPO nitrergic-glutamate neurons? The local preoptic area contains multiple populations of galanin neurons both excitatory and inhibitory (Moffitt et al., 2018). Recently, experiments have demonstrated that activating galanin neurons in the ventrolateral preoptic area (VLPO) can induce both NREM and hypothermia (Kroeger et al., 2018). Similarly, activation of galanin neurons in LPO can also induce NREM and hypothermia (Ma et al., 2019). The latter has parallels to the activation of GABAergic neurons activity-tagged during recovery sleep (Zhang et al., 2015). As the MnPO is known to send projections to both LPO and VLPO we have hypothesised that GABAergic-galanin neurons may be targets for nitrergicglutamate neurons (Uschakov et al., 2007). In VLPO, activation of galanin neurons using DREADD receptors facilitated more NREM sleep when the mice were closer to thermoneutrality (29◦C) and when hypothermia was blunted by warming at 36◦C. Thermoneutrality appears to allow optimum recovery of REM sleep, compared to either the ambient (22◦C) or warmed (36◦C) conditions (Kroeger et al., 2018). This is consistent with the idea of a narrow temperature range for optimised REM sleep (Czeisler et al., 1980; Szymusiak and Satinoff, 1981). Galanin neurons in LPO appear to be necessary for the activation of homeostatic mechanisms that trigger recovery sleep. Deletion of these neurons using caspase expression ablates rebound delta power following sleep deprivation (Ma et al., 2019). More data is needed to confirm if LPO-galanin are genuine targets of nitrergic-glutamate neurons. The latter may also have other long-range projections.

### THERMOGENESIS LINKS SLEEP TO ENERGY HOMEOSTASIS

Thermoregulation, in particular thermal inefficiency, impacts energy homeostasis and changes feeding requirements. This is an additional homeostatic drive that adds its own 'pressure' to modulate sleep networks (**Figure 3**). Following a meal, adipocytes secrete the hormone leptin. This hormone is indicative of excess energy intake and discourages feeding. Leptin works through well-established pathways in the arcuate hypothalamic nucleus, where it inhibits NPY expressing AgRP neurons (Williams et al., 2009). However, there are also leptin receptors elsewhere in the CNS, including in the PO hypothalamus. PO glutamatergic neurons expressing the leptin receptor (leptinRb) are excited (they express c-FOS) when ambient temperature rises (Yu et al., 2016). This results in a reduced energy expenditure, through inhibited thermogenesis, and a decrease in food consumption (Zhang et al., 2011; Yu et al., 2016). Neurons that coexpress

NOS1 and leptinRb have been identified in other parts of the hypothalamus and these can also inhibit thermogenesis (Leshan et al., 2012). Hence, it seems likely that there is some overlap between preoptic glutamatergic-leptinRb neurons and the NOS1 populations identified in Harding et al. (2018). Similarly, many BDNF/PACAP in the PO express VGLUT2 and a subpopulation of these neurons that express c-FOS in response to a warm stimulus has recently been shown to coexpress the leptinRb suggesting yet further overlap between these populations (Moffitt et al., 2018). As well as energy regulation, leptin signalling appears to have a more direct role in sleep. Systemic administration of recombinant leptin in food-deprived mice increases both NREM and REM sleep durations, while mice deficient in leptin (ob/ob) have fragmented sleep as well as lower average core temperature (Sinton et al., 1999; Laposky et al., 2006). A key remaining question is whether NOS1 neurons that initiate NREM sleep and body cooling provide a wider link between sleep and energy homeostasis (Harding et al., 2018). These circuits are summarised in **Figure 4**.

Recent data have provided a new insight into how energy balance might influence sleep. Goldstein et al. (2018) have directly assessed the impact of AgRP/POMC neuronal activity in the arcuate nucleus on the drive to sleep. AgRP neurons can detect energy intake and are considered 'hunger sensors', inhibited by both circulating leptin and insulin. POMC oppose the action of AgRP neurons and are activated by leptin (Cowley et al., 2001). AgRP activity promotes food seeking behaviour, even at the expense of sleep. But, if mice are food deprived, inhibition of these neurons rescues sleeping behaviour at the expense of eating (Goldstein et al., 2018). As thermal inefficiency results in increased feeding, we would expect circuitry that links thermal sensation to appetite control. Consistent with this idea, Jeong et al. (2018) have shown that POMC neurons express the transient receptor potential vanilloid-1 (TRPV1) channel. Optogenetic activation of these neurons produces feeding inhibition (Jeong et al., 2018). Although this study did not assess sleep, activation of POMC neurons by Goldstein et al. (2018) could rescue sleep in animals deprived of food (Goldstein et al., 2018). This may be because POMC neurons densely innervate sleep-promoting areas including the PO and may inhibit local GABAergic interneurons (Elias et al., 1999; Wang et al., 2015; Weber and Dan, 2016). POMC neurons also inhibit AgRP neurons which project to several sleeppromoting regions including the PO, the ventral periaqueductal grey (vPAG) and parabrachial nucleus (Pb) (Betley et al., 2013; Wang et al., 2015; Weber and Dan, 2016; Weber et al., 2018) (see **Figure 4**). Arcuate NPY neurons that normally stimulate eating also downregulate BAT thermogenesis and so may have similar roles in thermoregulatory connections to sleep (Shi et al., 2013). These autonomic signals are interpreted as strong behavioural drives, for instance to find food.

In summary, it seems likely that there is significant overlap between neuronal populations that regulate sleep onset, thermogenesis and energy homeostasis. Sleep onset may be controlled, in part, by integrating these sensory inputs, including ambient temperature and energy status. It is less clear why gating sleep with these inputs would be beneficial.

# SLEEP DEPRIVATION DISRUPTS THERMOREGULATION AND ENERGY BALANCE

Sleep architecture is highly dependent on thermal factors, but the consequence of total sleep loss is a radical alteration of thermoregulation and energy balance. In rats, chronic total-sleep deprivation and selective REM deprivation, using the disk-overwater method for many days, leads to profound physiological effects and eventually death (Everson et al., 1989). In the early stages, an increase in metabolic function was observed in these rats, including core body temperature, and with it an increase in food consumption. However, the temperature rise was quickly reversed and the rats progressively developed hypothermia. They also moved to warmer parts of a temperature gradient as their sleep deprivation deepened (Prete et al., 1991). This may be an energy conservation strategy, reducing thermal load, increasing appetite and simultaneously cooling the body (Rechtschaffen and Bergmann, 1995). Similar strategies are seen in torpid animals when food is scarce (Ruf and Geiser, 2015). In the sleep-deprived animals this strategy ultimately failed, as the rats rapidly lost weight (Bergmann et al., 1989; Everson et al., 1989; Rechtschaffen and Bergmann, 1995). Sleep deprivation appears to either increase the metabolic requirements of the animal, or by other means facilitates excessive heat loss, perhaps through over activation of NREM-initiating circuits that induce vasodilation.

One mechanism by which mammals, and small rodents in particular, generate heat is through brown adipose tissue thermogenesis (BAT). This is also a key mechanism in regulating energy homeostasis. Uncoupling protein 1 (UCP-1) is a key component of thermogenesis in brown adipose tissue (BAT). It decouples the electron transport chain from the ATP-synthase, facilitating heat production through proton gradient dissipation, without ATP production, and compensatory metabolism (Cannon and Nedergaard, 2004). UCP-1 knockout mice have weakened homeostatic rebound following sleep deprivation. They also show a blunted sleep induction effect of warmer temperatures observed in control mice (Szentirmai and Kapas, 2014). Similarly, the β3-adrenergic receptor agonists, which activate BAT thermogenesis, usually induce sleep in control mice but this response is ablated in mice with chemical deafferentation of the BAT (Szentirmai and Kapás, 2017). These data suggest that UCP-1 mediated BAT thermogenesis is helpful in both recovery sleep (sleep following sleep deprivation) and NREM sleep induction. UCP-1 may also have a role facilitating NREM sleep during systemic inflammation (Szentirmai and Kapas, 2018). The heat generated by these mechanisms could activate the sensory receptors in the skin and so trigger NREM sleep (Harding et al., 2018).

### TORPOR AND HIBERNATION: TOO COLD TO SLEEP?

The cooperation of body cooling and NREM sleep suggest that energy homeostasis is an important factor for sleep, but

it is natural to ask if there is any link to more extreme states (**Figure 5**). When the need to save energy is sufficiently high many mammals sacrifice sleep to adopt an alternative thermoregulatory strategy of daily torpor or seasonal hibernation (Ruf and Geiser, 2015). Daily torpor is a state of hypothermia triggered by food scarcity. Mammals that use daily torpor, such as the Djungarian hamsters (Phodopus sungorus), typically drop their core temperature to 15–20◦C for many hours, but, in many species, daily torpor can range between 10◦C or as high as 30◦C (Ruf and Geiser, 2015) (**Figure 5**).

A range of mammals from ground squirrels to brown bears also use annual hibernation strategies for winter survival and reproduction (Carey et al., 2003; Ruf and Geiser, 2015). In hibernators core temperatures typically fall between 0 and 10◦C and are maintained for weeks or months. Exposed ambient temperatures can fall below 0◦C and core temperatures are maintained only a few degrees higher at 1% of euthermic (normal temperature) metabolic rates (Carey et al., 2003). In extreme cases, such as the arctic ground squirrel (Urocitellus parryii), abdominal and peripheral temperatures can be stable at around −2 ◦C whilst head and neck are just above 0◦C (Barnes, 1989; Boyer and Barnes, 1999). Only at body temperatures around 0 ◦C do metabolic rates rise to defend the animal from freezing, suggesting an extremely low temperature set point (Buck and Barnes, 2000) (**Figure 5**).

Animals in either daily torpor or hibernation enter a state of inactivity or quiescence, but the power of the EEG signal observed in these animals scales with temperature: the lower the body temperature, the lower the power of the EEG (Deboer, 1998). At a core temperatures of approximately 22◦C or above, the frequency component remains within the delta band of 1–4 Hz and can be classified as sleep, although the EEG power is much

37◦C. Artificial hypothermia, sometimes known as synthetic torpor, can be induced by 5-AMP (0.5 g/kg IP). This also induces delta oscillations that are suppressed by hypothermia. The example is shown in blue, 1.5 h after injection with core temperature at approximately 23◦C. Below approximately 10◦C the EEG is isoelectric and no oscillation can be discerned. Hibernators have periods have interbout euthermia with normal EEG power and wake-NREM and wake-REM transitions are detected. Example species are labelled with the temperature that they have been observed in for either daily torpor or hibernation. This reflects ambient environmental conditions important for EEG measurements but is not a strict hierarchy. EEG examples are from (Harding et al., unpublished) except for the hibernation example which is adapted from Frerichs et al. (1994). IP, intraperitoneal; 5-AMP, adenosine monophosphate. Djungarian hamster (Phodopus sungorus), Golden-mantled ground squirrel (Callospermophilus lateralis), Fat-tailed dwarf lemur (Cheirogaleus medius), Arctic ground squirrel (Urocitellus parryii). Data adapted from Frerichs et al. (1994); Ruf and Geiser (2015), and Vyazovskiy et al. (2017).

reduced (Walker et al., 1981; Daan et al., 1991). However, at lower temperatures this is not the case. If core and brain temperature is sufficiently low, then the EEG power falls below the threshold for attribution of sleep states (**Figure 5**). At brain temperatures between 10◦C and 25◦C reduced-power delta oscillations can still be identified in the EEG signal, whereas below about 10◦C the signal is isoelectric (Walker et al., 1977). It is not clear to what extent the power of these oscillations is important for natural sleep function. In dwarf lemurs (Cheirogaleus medius), for example, hibernating at low ambient temperatures of only 5 ◦C, EEG recordings are isoelectric and evidence of NREM or REM sleep are absent. In this case, only on rewarming was sleep (REM) observed (Krystal et al., 2013). It has been suggested that torpor states may be a form of sleep deprivation. For instance, the recovery of Djungarian hamsters from daily torpor, with core and brain temperatures of around 23◦C, results in a period of recovery sleep with an increased power in the delta band (Deboer, 1998). Similarly, this sleep could be deferred by sleep deprivation suggesting the accumulation of a sleep debt during torpor (Deboer and Tobler, 1994; Palchykova et al., 2002). While this sleep debt, as measured by delta power, accumulates during torpor it does so almost three times slower at brain temperature below 27◦C, compared with time awake (Deboer and Tobler, 2003). However, comparisons of recovery sleep EEG, following sleep deprivation or torpor, revealed differences in cortical network activity suggesting that torpor is not entirely equivalent to either sleep deprivation or natural sleep (Vyazovskiy et al., 2017). Thus, a critical temperature may exist below which sleep function is impaired.

To understand the relationship between sleep and temperature in hibernators, researchers have compared animals that hibernate at different ambient temperatures. Animals that hibernate at low temperatures, such as the arctic ground squirrels (Urocitellus parryii), briefly warm up to levels comparable to waking (36–37◦C). These are periods of interbout euthermia (Boyer and Barnes, 1999; Carey et al., 2003). In these periods of warming, squirrels transition from wake to NREM and then REM sleep before returning to hibernation (Daan et al., 1991). Hibernation of the golden-mantled ground squirrel (Callospermophilus lateralis) under warmer laboratory conditions (22◦C) produced continuous NREM sleep (Walker et al., 1981). During hibernation at colder temperatures, of 11◦C ambient, the minimum brain temperature, not the hibernation bout length, was the best predictor of rebound delta power during subsequent interbout euthermia. The same authors observed that, at this temperature (11◦C), the euthermic (36–37◦C) period following hibernation consisted of almost 70–80% NREM sleep, whereas animals hibernating at 21◦C spent only 40% of their euthermic period in NREM (Larkin and Heller, 1996). This indicates that the temperature at which hibernation takes places influences the degree to which sleep debt accumulates (see **Figure 5**). Of course, there are variations between species. When European ground squirrels enter a euthermic period, following hibernation at 5.5◦C, the time spent in NREM sleep is proportional to the hibernation bout length, not temperature per se (Strijkstra and Daan, 1997). Collectively, these data suggest that the restorative component of sleep is temperature-dependent.

The same temperature dependence of sleep is seen in hibernating primates. When dwarf lemurs (Cheirogaleus medius) choose a hibernaculum at warmer temperatures, their EEG resembles NREM and REM sleep and the episodes of euthermia disappear (Dausmann et al., 2004; Krystal et al., 2013). In other species of lemur (C. crossleyi and C. sibreei) that occupy a cooler environmental niche, sleep is consistently absent during the torpor phase, but returns during periods of interbout euthermia (Blanco et al., 2016). Black bears (Ursus americanus), which always hibernate at warmer temperatures of 32–34◦C, and actively defend this temperature set point, also do not show the inter-bout arousals seen in smaller mammals (Tøien et al., 2011). Like the dwarf lemurs, these higher temperatures appear to allow brown bears to spend large amount of time in NREM sleep (Tøien et al., 2015). The defending of lower temperature set points in larger mammals has remarkable parallels with people under anaesthesia. In people given either the sedative DEX or the anaesthetic propofol, shivering thresholds reduce to between 32 and 34◦C, respectively (Matsukawa et al., 1995; Talke et al., 1997; Sessler, 2016). Hence, hibernation may be too cold to facilitate sleep and episodes of interbout euthermia, lasting 12–24 h, may allow sleep processes to be recovered (Carey et al., 2003).

The neuronal circuitry that induces torpor and/or hibernation is not known. However, it is possible that it uses components of the natural sleep–wake circuitry. For example, NOS1 glutamate neurons in PO that induce NREM sleep and sustained hypothermia (Harding et al., 2018), could in colder climates, have a role in torpor or hibernation induction, but with different behavioural and environmental triggers.

In summary, NREM sleep in a state of mild body cooling may be the preferred biological condition, but clearly in extreme environments, winter survival or times of food scarcity the restorative effects of sleep are, at least in part, sacrificed for energy conservation. As sleep can only be maintained at higher temperatures, it is energetically more expensive than torpor or hibernation. At these colder brain and core temperatures, sleep debts accumulate almost three times slower than during waking (Deboer and Tobler, 2003).

# WHY LINK NREM SLEEP AND BODY COOLING?

To recap, sleep in rodents is associated with temperaturecycling: wake to NREM sleep transitions coincide with a cooler body and brain facilitated by tail vasodilation. Indeed, effective thermoregulation and nesting behaviour produce warm microclimates that have a role in stimulating NREM sleep and body cooling. We have suggested that PO neurons both receive warm thermal information from the skin and simultaneously coordinate NREM sleep initiation and body cooling (Harding et al., 2018). Transitions to wakefulness or REM sleep are accompanied by vasoconstriction and brain warming (Alföldi et al., 1990; Imeri and Opp, 2009). The absolute change in brain temperature at each NREM transition is small, about 0.2–0.4◦C, but may reach larger values, comparable with the total diurnal variation in temperature (approximately 2◦C), during extended

bouts of sleep. In humans, core temperature reliably falls about 2 h prior to sleep onset and the first NREM episode is more likely to occur at the steepest point of temperature decline. Brain temperature appears to do the same (Landolt et al., 1995). This rate of decline may be highest when PO circuitry is maximally activated, facilitating NREM sleep to the greatest extent. Other sensory inputs, such as satiety, are also permissive for sleep and their inputs are integrated to determine the precise moment of NREM onset. These temperature changes may have a direct role in the restorative functions of sleep.

One of the first hypothesis regarding the lower temperatures coinciding with NREM sleep was that it existed specifically to cool the brain (McGinty and Szymusiak, 1990). It was proposed that a lower brain temperature would reduce cerebral metabolism, conserve energy and assist other functions from immune regulation to circadian coordination (McGinty and Szymusiak, 1990). Conservation of energy for sleep in its entirety has also been proposed (Berger and Phillips, 1995). However, we have seen that when mammals of all sizes prioritise conservation of energy, extremes of hypometabolism in torpor and hibernation are selected at the expense of sleep (Ruf and Geiser, 2015). This suggests that energy conservation alone is not the primary function of sleep. Indeed, estimates of energy use over 24 h put the cost of sleep as high as 85–95% of the metabolic cost of waking (Jung et al., 2011; Abreu-Vieira et al., 2015; Hibi et al., 2017).

It is feasible that reduced temperatures have a more direct function in the brain. At temperatures of 20◦C or less, during which sleep debt is accumulated, morphological changes have been observed in dendritic spines (Peretti et al., 2015). Hibernators can undergo synaptic remodelling while cold, as do animals in artificial torpor induced by 5<sup>0</sup> -adenosine monophosphate (Popov and Bocharova, 1992; Magariños et al., 2006; Popov et al., 2007). In the latter condition, the total number of synapses is reduced (GM). The presence of these process may explain why sleep, as a restorative process, is inhibited at lower temperatures. Large changes in gene expression are also observed both in the brain and across the body in hibernators (Williams et al., 2005). Colder temperatures, particularly in the brain, can induce expression of so called 'cold-shock' proteins including cold-inducible RNA binding protein (CIRP) and RNAbinding motif protein 3 (RBM3) (Morf et al., 2012; Peretti et al., 2015; Hoekstra et al., 2019). Body and brain cooling during natural sleep are small, both from the reduction in diurnal core temperature and reductions in brain temperature at each NREM

# REFERENCES


transition, but recent data suggest they are sufficient to increase CIRP expression and so influence the expression of other genes, including the circadian genes Period and Clock (Morf et al., 2012; Hoekstra et al., 2019). This is important as cortical temperature changes are heavily influenced by the sleep–wake transitions and entry into NREM sleep can then influence clock gene expression to drive further transcriptional changes. In mice without CIRP, sleep deprivation results in 50% less REM sleep, illustrating the strength of this mechanism (Hoekstra et al., 2019). This also provides one possible mechanism that the brain may keep track of the time spent in NREM sleep.

The extensive neuronal inter-connections that cross-regulate energy use, sleep induction and body temperature (see **Figure 3**) hint that the function of sleep plays an important role in energy homeostasis. The temperature-dependence of sleep debt accumulation, which is slowed at cooler temperatures, suggests that this debt is inherently a metabolic processes. Lastly, the synchronised changes in brain temperature during sleep may coordinate gene expression important for the functions of sleep, whilst contributing to a mechanism that measures the time spent sleeping.

### AUTHOR CONTRIBUTIONS

EH wrote the manuscript and designed the figures. All authors have discussed and edited the manuscript.

# FUNDING

This work was funded by the UK Dementia Research Institute, which receives its funding from UK DRI, funded by the UK Medical Research Council, Alzheimer's Society, and Alzheimer's Research UK (NF and WW) and also by the Wellcome Trust (107839/Z/15/Z to NF and 107841/Z/15/Z to WW).

### ACKNOWLEDGMENTS

All images are used with permission or copyright clearance. The sleeping human (**Figure 1A**) is available under CC0-1.0 universal. We are grateful to Kathelijne Koops for the nesting chimpanzee picture, Isobel Harding for the picture of a sleeping cat, and to Ália dos Santos for proofreading the manuscript.


of juvenile Siberian hamsters. Physiol. Behav. 101, 376–380. doi: 10.1016/j. physbeh.2010.07.001




schweinfurthii) at the Toro-Semliki Wildlife Reserve. Uganda. Am. J. Primatol. 74, 811–818. doi: 10.1002/ajp.22031



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Harding, Franks and Wisden. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Neural Circuitry Underlying Waking Up to Hypercapnia

### Satvinder Kaur\* and Clifford B. Saper

Department of Neurology, Program in Neuroscience, Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, MA, United States

Obstructive sleep apnea is a sleep and breathing disorder, in which, patients suffer from cycles of atonia of airway dilator muscles during sleep, resulting in airway collapse, followed by brief arousals that help re-establish the airway patency. These repetitive arousals which can occur hundreds of times during the course of a night are the cause of the sleep-disruption, which in turn causes cognitive impairment as well as cardiovascular and metabolic morbidities. To prevent this potential outcome, it is important to target preventing the arousal from sleep while preserving or augmenting the increase in respiratory drive that reinitiates breathing, but will require understanding of the neural circuits that regulate the cortical and respiratory responses to apnea. The parabrachial nucleus (PB) is located in rostral pons. It receives chemosensory information from medullary nuclei that sense increase in CO2 (hypercapnia), decrease in O2 (hypoxia) and mechanosensory inputs from airway negative pressure during apneas. The PB area also exerts powerful control over cortical arousal and respiration, and therefore, is an excellent candidate for mediating the EEG arousal and restoration of the airway during sleep apneas. Using various genetic tools, we dissected the neuronal sub-types responsible for relaying the stimulus for cortical arousal to forebrain arousal circuits. The present review will focus on the circuitries that regulate waking-up from sleep in response to hypercapnia.

### Edited by:

Michael Lazarus, University of Tsukuba, Japan

### Reviewed by:

Hiromasa Funato, Toho University, Japan Zhi-Li Huang, Fudan University, China

### \*Correspondence:

Satvinder Kaur skaur@bidmc.harvard.edu

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 11 February 2019 Accepted: 08 April 2019 Published: 26 April 2019

### Citation:

Kaur S and Saper CB (2019) Neural Circuitry Underlying Waking Up to Hypercapnia. Front. Neurosci. 13:401. doi: 10.3389/fnins.2019.00401 Keywords: obstructive sleep apnea, arousal, parabrachial nucleus, calcitonin gene related peptide, hypercapnia

### INTRODUCTION

Obstructive sleep apnea (OSA) is caused by a sleep state-dependent reduction in the pharyngeal dilator muscle activity that leads to the closure of the upper airway in the susceptible individuals. These recurrent episodes of complete or partial obstruction of the upper airway lead to the airway collapse, which causes periodic hypoxia and hypercapnia during sleep, causing brief arousals that restore airway patency (Schulz, 2010; Mannarino et al., 2012; White and Younes, 2012; White, 2017; Darquenne et al., 2018; Pham et al., 2018). These repeated arousals result in sleep disruption, which in turn causes cognitive impairment as well as cardiovascular and metabolic morbidities (Bonnet, 1985; Fletcher, 1996; Bennett et al., 1998; Malhotra and White, 2002; Jun and Polotsky, 2009; Malhotra and Loscalzo, 2009; Drager et al., 2010; Bonsignore et al., 2012). The transient cortical arousals and sleep fragmentation are associated with the autonomic dysregulation, increased oxidative stress and hemodynamic changes during sleep. In patients with OSA, these consequences have been linked to increased daytime sleepiness, cardiovascular and metabolic morbidities. Due to

this, many OSA patients are also at risk of developing arterial hypertension, coronary heart disease, stroke, type 2 diabetes and mortality.

Although OSA can be treated effectively with continuous positive airway pressure, many patients do not tolerate it and compliance is often poor. One alternative therapeutic approach in OSA may involve modifying the arousal threshold that may augment respiratory drive during apnea and recruiting the upper air way muscles to reestablish stable breathing (Horner et al., 1991; Loredo et al., 1999; Horner et al., 2017; Sands et al., 2018; Zinchuk et al., 2018). However, designing drugs that can selectively reduce cortical arousals while maintaining or augmenting the respiratory drive during these respiratory events would require understanding the circuits that mediate cortical EEG and respiratory responses to apnea. This review will focus on recent attempts to identify that circuitry, in particular, using newer methods such as optogenetics and chemogenetics that allow selective, genetically directed targeting of the neuronal nodes that mediate cortical EEG and respiratory responses to apnea.

The brain circuitry that underlies waking up to hypercapnia (increased CO2) that can occur in apnea is not clearly understood. Briefly, three main sensory stimuli that alert the brain during apnea are hypoxia, hypercapnia and negative air pressure in the airways created due to increased respiratory efforts (sensed by mechanoreceptor fibers in vagus) during apneas (White, 2006, 2017). The carotid body primarily senses the hypoxia and to a lesser extent the hypercapnia, and transmits that information to the nucleus of the solitary tract (NTS) via the carotid sinus branch of the glossopharyngeal nerve (Massari et al., 1996; Lindsey et al., 2018). In addition, the chemosensory neurons in the retrotrapezoid nucleus (RTN) directly sense the CO2, and these project in parallel to the NTS to the ventrolateral medulla (VLM- pattern generator for breathing), and parabrachial nucleus (PB- relay node for visceral sensory information from the brainstem to the forebrain) (Dean et al., 1989; Herbert et al., 1990; Finley and Katz, 1992; Massari et al., 1996; Guyenet et al., 2010a; Bochorishvili et al., 2012; Guyenet and Bayliss, 2015; Lindsey et al., 2018; **Figure 1A**). The serotonergic raphe system in the brainstem (Richerson et al., 2001; Depuy et al., 2011) and orexin neurons in the lateral hypothalamus (Hunt et al., 2016; Rodrigues et al., 2019), are other CO2 sensing neurons, which also project to the NTS, the VLM, and the PB. In patients with OSA, arousal correlates closely with the airway negative pressure, to a lesser degree to the level of hypercapnia, and least with the level of hypoxia (Gleeson et al., 1990). However, all the three stimuli converge in the same brain locations; therefore studying these areas and their connections is important to understand the brain response to apnea.

### PARABRACHIAL NUCLEUS AND CORTICAL AROUSAL

The PB, a relay node for sensory visceral information, that surrounds the superior cerebellar peduncle, is referred to as the "Pontine taste area" (Saper, 2016) and the same region as early as 1920s was also identified as the "pneumotaxic center" or the "pontine respiratory group" by the workers on respiratory control (Feldman et al., 1976; Dobbins and Feldman, 1994). Recent studies have associated the PB with cortical arousal. In addition to the canonical, cholinergic (Semba and Fibiger, 1992; Steriade et al., 1993; Kleiner and Bringmann, 1996; Datta and Siwek, 1997; Cape and Jones, 2000) and monoaminergic (Aston-Jones and Bloom, 1981; Berridge and Wifler, 2000) arousal pathways from the upper brainstem, the PB projects to the thalamus, hypothalamus, and cerebral cortex (Saper and Loewy, 1980; Steriade et al., 1993; Jones, 2005). Surprisingly, cell specific lesions of the cholinergic and monoaminergic neurons, either alone or in combination, in these pontine areas have been found to cause little alteration in wake in both cats and rats (Jones et al., 1973; Holmes and Jones, 1994; Lu et al., 2006; Blanco-Centurion et al., 2007; Fuller et al., 2007), whereas large PB lesions cause profound coma (Fuller et al., 2011). Studies from our lab and others have shown that cortical arousal can be induced by activation of PB (Hayashi et al., 2015; Qiu et al., 2016; Kaur et al., 2017) and the deletion of glutamatergic signaling in PB neurons increases sleep and causes EEG slowing (Fuller et al., 2011; Kaur et al., 2013). Therefore, ascending projections of the PB through a ventral forebrain pathway via the hypothalamus and BF may play a key role in mediating cortical arousal. As the PB consists of different diverse sub nuclei, each with its distinct input and output targets, and these are often associated with different neuromodulators (Fulwiler and Saper, 1984), it is therefore, necessary to further dissect it using newer genetically specified tools to understand the roles of different cell types in this functionally heterogeneous population.

### PARABRACHIAL NUCLEUS AND BREATHING

PB receives chemosensory information from RTN and NTS, that sense hypercapnia and hypoxia and also from the upper airway afferents that respond to pulmonary negative pressure associated with apneas (Panneton and Loewy, 1980; Finley and Katz, 1992; Mizusawa et al., 1995; Berquin et al., 2000; Pete et al., 2002; Izumizaki et al., 2004; Rosin et al., 2006; Corcoran et al., 2009; Song and Poon, 2009; Gonzalez et al., 2010; Guyenet et al., 2010b; Topchiy et al., 2010; Bochorishvili et al., 2012; Yokota et al., 2012, 2015; Guyenet and Bayliss, 2015; Roman et al., 2016). As mentioned above, the PB area not only exerts powerful control over cortical arousal (Fuller et al., 2011; Kaur et al., 2013, 2017; Hayashi et al., 2015; Qiu et al., 2016) it also regulates respiration (Miura and Takayama, 1991; Chamberlin and Saper, 1994; Mizusawa et al., 1995; Chamberlin, 2004; Bonis et al., 2010a,b; Diaz-Casares et al., 2012; Damasceno et al., 2014; Kaur et al., 2017; Yang and Feldman, 2018). The ascending projections of the PB mediate cortical arousal (Saper and Loewy, 1980; Saper, 1982; Kaur et al., 2013, 2017; Saper, 2016), while its descending projections to the respiratory areas such as ventral lateral medulla, hypoglossal motor nucleus and phrenic motor nucleus, may regulate respiration (Yokota et al., 2001, 2012, 2015). Thus, the PB is an excellent candidate for a site that

can augment the airway dilator muscles, particularly following EEG arousals during sleep apneas. However, the precise brain circuitries that can be selectively targeted to prevent cortical arousal but augment respiration and maintain air-way patency during apneas need to be investigated.

### MOUSE MODEL OF APNEA

We designed a mouse model of apnea (Kaur et al., 2013) to simulate breathing during apneas and understand the brain circuitry underlying the repetitive arousals during apnea. Briefly, a mouse is kept in a plethysmograph chamber, and every 300 s, the gas mixture entering the chamber is switched for 30 s to one that contains either increased CO2 (10%), reduced O2 (10%) or both. The gasses mix in the chamber and approach the new steady state after about 10–15 s. At the end of the 30 s period, the source is switched back to air, and the gas levels return to baseline. We continuously record the EEG and EMG, the plethysmograph (which gives us tidal volume and respiratory rate) and percentage of CO2 and O2 in the chamber (**Figure 1**). In our earlier study, the arousal kinetics to CO2 (hypercapnia alone) and to the combined hypercapnia and hypoxia were identical (Kaur et al., 2013), therefore we continue to test EEG arousals with hypercapnia in our model (**Figures 1B,C**), and we will refer here only to the trials with elevated CO2. We used this paradigm of repetitive hypercapnia as a model of sleep apnea as the duration of the gas disturbance, its frequency, and the length of the arousals, were similar to those seen in a patient with mild sleep apnea. Also hypercapnia is mechanistically more relevant than hypoxia in sleep disordered breathing related neuro-impairment (Wang et al., 2014, 2016), even though the effect of hypoxia only is more extensively studied by most groups.

### GENETIC TOOLS AND TECHNOLOGIES FOR CIRCUIT ANALYSIS

In the past, researchers have used a wide range of electrophysiological and molecular tools, either individually or in combination, to probe and manipulate neural circuits. Although, these had helped us understand some basic pathways,

FIGURE 1 | (A), Three main stimuli related to apnea converge on the parabrachial area: Increased CO<sup>2</sup> (Hypercapnia), Hypoxia and negative air-way pressure cause activation of both central and peripheral chemoreceptors whose signals are integrated in the nucleus of the solitary tract (NTS) and retrotrapezoid nucleus (RTN). The NTS and RTN activate neurons in the lateral parabrachial nucleus, major node in the brain stem that relay visceral sensory information to the forebrain areas. Mouse model of apnea: (B), shows the "repetitive CO2 arousal (RCA) protocol" where a mouse is recorded in the plethysmograph chamber for the EEG, EMG and breathing, while exposed to repeated bouts of CO2 (hypercapnia). Mice undergo spontaneous periods of sleep and wake, however, only trials where the mouse is in NREM sleep for at least 30 s prior to onset of the CO2 are used to examine arousal. During these trials, the arousals are judged by EEG arousal (loss of delta waves and appearance of low voltage fast EEG), which is usually accompanied by EMG activation. Scale = 45 s (C), is a schematic of the plethysmography chamber used to model apnea in mice, while they are exposed to CO2 and recorded for EEG/EMG and breathing responses, with and without laser light that is transmitted through the pre-implanted optical glass fibers. [Adapted and modified from Kaur et al. (2013)]. Kaur et al. (2013), is published under Creative Commons Attribution-Non-commercial-Share License, and therefore no permission is required reproducing this modified version.

the inherent heterogeneity of brain cells and the lack of target specificity of these earlier tools make the interpretation of such data difficult. For example, pharmacological approaches using receptor agonists and antagonists have been confounded by poor blood-brain permeability when given systemically; low solubility when given directly into the brain; and "off-target" side effects when they engage unintended targets. Other approaches that involve the use of the global gene knockout, sometimes suffer from low temporal and spatial resolution and such approaches can also be confounded by ontogenetic and ectopic expression of the gene of interest. Similarly, both acute and chronic lesions produce collateral damage to adjacent brain structures making it difficult to interpret the effects which could be secondary to the lesion itself. Now, the emergence of newer conditional genomic models and viral-vectors approaches allow us to precisely target a selective cell population in the brain area of interest, and this is helping to link specific group of neurons and neural pathways to specific behaviors (Carter et al., 2010, 2013b; Carr and Zachariou, 2014; Fuller et al., 2015; Han et al., 2015; Campos et al., 2016, 2018; Qiu et al., 2016; Whissell et al., 2016; Saper and Fuller, 2017; Wu et al., 2018). The introduction of the Cre transgenes through gene delivery methods using the Cre/ lox system provides better temporal and spatial control over Cre-mediated excision of a selective gene sequence encoding the protein of interest (Kaur et al., 2013; Abbott et al., 2014; Todd et al., 2018), in a selective brain area (**Figure 2C**).

In recent years, the use of chemo- and opto-genetic tools had equipped us with an unparalleled ability to probe the neural circuitry that underlies behavioral state. The genetically engineered receptors are successfully used as tools for targeting chemo- and opto-genetics to selective cell types. Because they are activated either by injectable synthetic ligands that specifically bind to these receptors on the targeted cells and excite or inhibit them (chemogenetics) or through the delivery of specific wavelengths of laser-light via an implanted optical fiber (optogenetics), investigators retain temporal control over particular subsets of neurons (Fuller et al., 2015; Park and Carmel, 2016; Vlasov et al., 2018). In addition, because the opto- or chemogenetic tool is expressed in a conditional manner, it is only expressed by cells of a specific genotype, thus giving the investigator both neuroanatomical and neurochemical control over the response. Because the receptor transcript is packaged within a Flip-Excision-Switch (FLEX) cassette, the functional receptor can be expressed only in the presence of cre-recombinase (Schnutgen et al., 2003; Fuller et al., 2015; Plummer et al., 2017). The use of the Cre-driver mouse lines,

FIGURE 2 | Testing the role of glutamatergic signaling in hypercapnia induced arousal: (A,B), are the two representative trials from a control mouse (A), where cortcial EEG arousal in response to hypercapnia occurs in 15 s after onset of CO2, while the mouse with deletion of Vglut2 gene in the LPB (B), fails to wake wake up to hypercapnia. (C), Photomicrograph of the Nissl-stained coronal section of the mouse brain, showing different sub divisions of the parabrachial (PB) nucleus, Cre-immunoreactivity (brown) against a Nissl-stained background (blue) in the neurons in the lateral parabrachial (LPB) region after injection of AAV-Cre in Vglut2 flox/flox mice and last panel shows the shows a photomicrograph of a brain section immunostained for Neu-N, a neuronal marker after bilateral injection of AAV-DTA killed Vglut2+ neurons into the LPB. (D), Show graphs of the latency of arousal during and after a hypercapnic stimulus of 30 s in mice injected bilaterally with AAV-DTA (green) compared to the control (black, gray, and striped green) and LPB group from which Vglut2 was deleted in the LPB including the PBel (red). scp – superior cerebellar peduncle; dl – dorso-lateral; cl – centro-lateral; el – external lateral; vl – ventrolateral PB subnucleus; MPB – medial and MPB-ext – medial external-lateral parabrachial nucleus; KF – Kolliker Fuse; vsct – ventral cerebro-spinal tract; Scale = 100 µm. ∗∗represents p < 0.01 compared to the control group (AAV-GFP) and #p < 0.05, compared to the AAVCreWT group. [Adapted and modified from Kaur et al. (2013)]. Kaur et al. (2013), is published under Creative Commons Attribution-Non-commercial-Share License, and therefore no permission is required reproducing this modified version.

where cre is expressed downstream of a selected promoter, ensures that the designer receptors when injected in these mice are expressed in a Cre-dependent manner, specifically in neurons that express a select protein. Finally, because the viral vectors for the opto- and chemogenetic tools can be injected locally in the brain, the investigator also has spatial control over the part(s) of the brain involved in the experiment. These modified receptors can act as effective tools that allow us to manipulate a selective neural circuit and we can then evaluate the effect upon the behavior in direct relation to either the excitation or inhibition of a specific neuronal node.

### GLUTAMATERGIC SIGNALING IN THE PB AND WAKING UP TO CO2

To examine the role of glutamatergic signaling in the PB, in one set of animals we deleted the vesicular glutamate transporter-2 (Vglut2) gene in various PB sub-nuclei (by injecting an AAV-Cre into the PB of Vglut2flox/flox mice). In another set of mice, we killed the cells in the lateral PB using injections of AAV that had Cre-dependent expression of diphtheria toxin subunit A (DTA) in Vglut2-Cre mice (**Figure 2C**). We tested these mice for arousal to hypercapnia using the mouse model of apnea (**Figure 1**). Our results indicated that deletion of glutamatergic signaling from neurons in the external lateral PB (PBel), or killing the Vglut2 neurons in the PBel produced the same prolongation of the latency of waking up to CO2 (**Figures 2A–D**), suggesting that glutamate alone in PBel neurons is the operative neurotransmitter for relaying the signal for waking up from sleep in response to hypercapnia (Kaur et al., 2013). Many neurons in the PBel express calcitonin gene related peptide [CGRP, (Yasui et al., 1991); PBelCGRP], and we tested whether these are activated (cFos expression) in response to the hypercapnia (Yokota et al., 2015). Most of the cFos positive neurons in the PBel contained CGRP, while many along the lateral edge of the nucleus did not. Because most PBelCGRP neurons project to the forebrain, whereas most neurons lateral to them project to the brainstem, we hypothesized that the PBelCGRP neurons might be selectively responsible for forebrain arousal during hypercapnia.

# ROLE OF THE PBELCGRP NEURONS IN CORTICAL AROUSAL

Using optogenetic and chemogenetic tools in CGRP-CreER mice (Kaur et al., 2017), we could selectively activate and inhibit the

(2017) is published in a Cell Press journal "Neuron," and no permissions needed to reproduce the modified versions of the published figure.

PBelCGRP neurons. Optogenetic activation of PBelCGRP neurons at 10 and 20 Hz by 10 ms blue laser light pulses (**Figure 3B**) caused short latency arousals and their chemogenetic activation significantly increased wakefulness (**Figure 3A**; Kaur et al., 2017). Targeting yellow (593 nm) laser light to the archaerhodposin – TP009 (ArchT) expressing PBelCGRP neurons during the CO2 trials, silences them. Mice with inhibition of PBelCGRP neurons (Laser-ON) failed to wake up to CO2 in 50% of the trials and increased the latency to arousal by four fold in response to the CO2 (Figre 4A,C). These results were similar to those we obtained with killing most of the lateral PB neurons, or deleting their Vglut2 gene. In other words, the PBelCGRP neurons appear to provide most if not all of the arousal response to CO2, by using glutamate as their neurotransmitter.

Interestingly, the silencing of the PBelCGRP neurons preserved the respiratory drive (**Figure 4B**) during the hypercapnia, with no differences in the tidal volume and respiratory rate (Kaur et al., 2017). Also these laser-induced inhibitions did not affect the arousal thresholds to acoustic stimuli or somatosensory and vestibular stimulation (Kaur et al., 2017). Recent work from Palmiter and colleagues suggests that the CGRP neurons may respond to pain and to other visceral stimuli (e.g., gastrointestinal upset or conditioned taste aversion) (Carter et al., 2015; Han et al., 2015; Campos et al., 2018) which has led to the suggestion that they may serve a more generalized central alarm function, waking up the brain when aversive visceral or noxious stimuli arise (Saper, 2016; Palmiter, 2018).

To further investigate the arousal regulating circuitry targeted by PBelCGRP neurons for causing arousals during apnea, we inhibited terminal fields at three major forebrain arousal nodes: the substantia innominata in the basal forebrain (BF); the central nucleus of the amygdala (CeA); and the lateral hypothalamus (LH). Optogenetic silencing of these terminals fields also increased the latency for arousal, with differential responses at multiple target sites. Our data suggested that PBelCGRP neurons act most potently through their direct projections to the BF, whose neurons have direct projections to the cerebral cortex. The CeA also participates in the arousal, but has no ascending projections to the cortex or thalamus. Because it projects intensely to the BF, this is likely to be its mechanism of function. Lastly inhibition of the PBelCGRP terminals in the LH field had the least effect on the arousal in response to CO2. Although some LH

neurons directly project to cerebral cortex and others send axons to the BF, this region appears to play at most a minor role in arousal to CO2 (**Figure 4C**). Thus, the PBelCGRP neurons are a critical node in the network that receives input from neural pathways activated in apneas in response to hypercapnia, hypoxia and airway stretching, and in relaying that influence to the forebrain sites to cause awakening during apneas (**Figure 5**).

PBelCGRP neurons did not contribute to the respiratory component of apnea, as their inhibition did not diminish respiratory drive to CO2. Also, these neurons did not show any descending outputs to the respiratory areas, but adjacent neurons that showed a cFos response to CO2 in the lateral crescent (PBlc) and Kolliker Fuse (KF), do project to respiratory areas (Yokota et al., 2015). Of note, many of the neurons in this area express the transcription factor Forkhead-homeobox protein-2 (FoxP2), which is distributed throughout the respiratory column in both rats and mice (Geerling et al., 2017; Stanic et al., 2018; **Figure 5**). The glutamatergic FoxP2 neurons in the PBlc and KF have descending projections to respiratory areas such as the ventrolateral medulla including the pre-Bötzinger complex and retroambiguus area, the hypoglossal nucleus, the ventrolateral and commissural subdivisions of NTS, and the intermedio-lateral cell column (IML) and phrenic motor nucleus in the spinal cord (Geerling et al., 2017). These projections, much of which are likely to come from the PBFoxP2 neurons, may influence the respiratory efforts during apneas (**Figure 5**). However, it remains to be seen if selectively manipulating the PBFoxP2 neurons in the PBlc and KF can augment ventilatory efforts during hypercapnia. We are now investigating such a role of this population of PBFoxP2 neurons in augmenting respiration in response to hypercapnia and their possible interactions with PBCGRP neurons.

Other brainstem cell groups, such as the serotonergic dorsal raphe (Richerson et al., 2005; Buchanan and Richerson, 2010; Ray et al., 2011, 2013; Smith et al., 2012) have also been shown to regulate hypercapnia induced arousals. A recent study showed that serotonergic dorsal raphe regulates waking up to CO2, and this is mediated through 5HT2<sup>A</sup> receptors (Smith et al., 2018). However, mice deficient in 5HT neurons are responsive to the CO2 when injected with a 5HT2<sup>A</sup> agonist (Buchanan et al., 2015), suggesting that DR serotonergic neurons are modulatory and maybe acting through a non-serotonergic area, e.g., PBelCGRP neurons (Kaur et al., 2018). This possibility is also the subject of our current investigations.

# CONCLUSION

Effective pharmacotherapy for OSA will depend on identifying the sites that can selectively regulate the brain response to hypercapnia (Horner et al., 2017) and therefore be used as druggable targets. Another line of investigation, with the goal of providing more personalized therapeutic interventions for patients with a low arousal threshold (Sands et al., 2018), seeks to quantify and manipulate the "arousal threshold" in patients with OSA. The knowledge of selective neuro-circuitries that comprise functionally connected specific neurons, such as PBelCGRP and PBFoxP2 neurons for regulation of cortical arousal and respiratory efforts during apnea can help with such interventions. Importantly, PBelCGRP neurons are not only activated by hypercapnia, but are also responsive to various potentially

dangerous or aversive stimuli (Carter et al., 2013b,a, 2015; Han et al., 2015; Campos et al., 2016, 2018; Saper, 2016; Palmiter, 2018). As such, it is plausible that different subpopulations of PBelCGRP neurons encode and process different classes of aversive stimuli (Bernard et al., 1994; Campos et al., 2018) resulting in amplification of the hypercapnia-arousal response. Therefore, to treat a low arousal threshold in sleep apnea, there is a need for more precise understanding about the afferents that selectively modulate the PBelCGRP neurons and therefore likely help tune the hypercapnia-arousal response. Thus, a deeper understanding of the PBelCGRP neurocircuitry and it's connections to other neuronal subpopulations in the PB (for e.g., PBFoxP2 neurons) and each of their distinct projection targets, will help yield valuable therapeutic targets, that can help prevent cortical arousal during apneas while preserving the respiratory drive important for restoring the airway patency. This will eventually help in preventing OSA and its negative secondary health consequences.

### ETHICS STATEMENT

All animal procedures met National Institutes of Health standards, as described in the Guide for the Care and Use of

### REFERENCES


Laboratory Animals, and all protocols were approved by the Beth Israel Deaconess Medical Center Institutional Animal Care and Use Committee.

### AUTHOR CONTRIBUTIONS

SK conceptualized, designed the experiments, collected and analyzed the data, and wrote the manuscript. CS contributed to the experimental concept and design, and wrote the manuscript.

### FUNDING

This research work was supported by funding from USPHS grants 2P01 HL095491 and NS085477.

### ACKNOWLEDGMENTS

We thank Quan Ha and Minh Ha for their excellent technical support, and Sathyajit Bandaru for maintaining the mouse breeding program.


locus coeruleus neurons. Nat. Neurosci. 13, 1526–1533. doi: 10.1038/nn. 2682


the carotid body chemoreceptors in the context of reactive oxygen species biology. Respir. Physiol. Neurobiol. 174, 317–330. doi: 10.1016/j.resp.2010. 09.002




**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Kaur and Saper. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Gating and the Need for Sleep: Dissociable Effects of Adenosine A<sup>1</sup> and A2A Receptors

Michael Lazarus<sup>1</sup> \*, Yo Oishi<sup>1</sup> , Theresa E. Bjorness2,3 and Robert W. Greene1,3,4 \*

1 International Institute for Integrative Sleep Medicine (WPI-IIIS), University of Tsukuba, Tsukuba, Japan, <sup>2</sup> Research and Development, VA North Texas Health Care System, Dallas, TX, United States, <sup>3</sup> Department of Psychiatry, The University of Texas Southwestern Medical Center, Dallas, TX, United States, <sup>4</sup> Department of Neuroscience, The University of Texas Southwestern Medical Center, Dallas, TX, United States

Roughly one-third of the human lifetime is spent in sleep, yet the reason for sleep

remains unclear. Understanding the physiologic function of sleep is crucial toward establishing optimal health. Several proposed concepts address different aspects of sleep physiology, including humoral and circuit-based theories of sleep-wake regulation, the homeostatic two-process model of sleep regulation, the theory of sleep as a state of adaptive inactivity, and observations that arousal state and sleep homeostasis can be dissociated in pathologic disorders. Currently, there is no model that places the regulation of arousal and sleep homeostasis in a unified conceptual framework. Adenosine is well known as a somnogenic substance that affects normal sleep-wake patterns through several mechanisms in various brain locations via A<sup>1</sup> or A2A receptors (A1Rs or A2ARs). Many cells and processes appear to play a role in modulating the extracellular concentration of adenosine at neuronal A1R or A2AR sites. Emerging evidence suggests that A1Rs and A2ARs have different roles in the regulation of sleep. In this review, we propose a model in which A2ARs allow the brain to sleep, i.e., these receptors provide sleep gating, whereas A1Rs modulate the function of sleep, i.e., these receptors are essential for the expression and resolution of sleep need. In this model, sleep is considered a brain state established in the absence of arousing inputs.

Keywords: adenosine, slow-wave sleep, A2A receptor, A<sup>1</sup> receptor, slow-wave activity, sleep homeostasis, dopamine, motivation

# INTRODUCTION

Sleep is a highly conserved behavior that is vital to survival among all living organisms with a nervous system, from worms to humans. Chronic sleep loss is linked to a wide range of deleterious physiologic changes, such as altered food intake, weight loss or gain, skin lesions, compromised thermoregulation, and even death (Rechtschaffen et al., 1989; Siegel, 2008). Humans spend roughly one-third of their lives asleep. While we know why we eat, drink, and mate, we do not yet know why we sleep. The neuroscience community has therefore increased efforts to gain knowledge of the physiologic function of sleep.

During sleep, cortical neurons alternate between periods of firing and periods of silence. The switching between the two states, also known as ON and OFF states, is widely synchronized across neurons and represented by slow wave activity (SWA) in encephalography. SWA is observed

### Edited by:

Ritchie Edward Brown, VA Boston Healthcare System, United States

### Reviewed by:

Marcos G. Frank, Washington State University Health Sciences Spokane, United States David Elmenhorst, Jülich Research Centre, Germany

\*Correspondence:

Michael Lazarus lazarus.michael.ka@u.tsukuba.ac.jp Robert W. Greene RobertW.Greene@ UTSouthwestern.edu

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 31 March 2019 Accepted: 02 July 2019 Published: 17 July 2019

### Citation:

Lazarus M, Oishi Y, Bjorness TE and Greene RW (2019) Gating and the Need for Sleep: Dissociable Effects of Adenosine A<sup>1</sup> and A2A Receptors. Front. Neurosci. 13:740. doi: 10.3389/fnins.2019.00740

**176**

as slow, oscillatory neocortical activity (0.5–4.5 Hz) that intensifies in correlation with wake duration and declines during sleep, but is expressed only during slow wave sleep (SWS). Because SWS-SWA increases as sleep loss is prolonged and decreases as sleep progresses, it is widely used as a marker of mammalian sleep homeostasis. The rates of SWA buildup and decay can be altered by extreme sleep loss or by pharmacologic or genetic manipulations in mammals, especially those affecting adenosine systems of the central nervous system (CNS). The adenosine system can also affect the gating of SWS-SWA expression by modulating the arousal level, thereby altering the duration of time during which sleep homeostasis and function can occur.

Adenosine is the key building block of every cell's energy source, i.e., adenosine triphosphate (ATP), and the related adenosine mono- and di-phosphates (AMP and ADP, respectively). Adenosine fulfills a wide range of physiologic and pathophysiologic functions (Fredholm, 2014). In the nervous system, adenosine acts as a neuromodulator through metabotropic receptors. Although adenosine acts on four evolutionarily well-conserved receptors present on most cells, it is believed to modulate sleep need and arousal by acting through A<sup>1</sup> and A2A receptors (A1Rs and A2ARs), respectively.

In light of the emerging roles of adenosine and its receptors in regulating different aspects of sleep, we propose a model for the gating and function of sleep. In our model, A2ARs allow the brain to sleep, i.e., these receptors provide sleep gating, whereas A1Rs modulate the function of sleep, i.e., these receptors are essential for the expression and resolution of the sleep need.

### ASPECTS OF SLEEP/WAKE REGULATION

### Humoral Theory of Sleep-Wake Regulation

The humoral theory of sleep-wake regulation posits that during wakefulness, one or more endogenous somnogenic factors is produced and accumulated. Brain activity decreases when the concentration of somnogenic substances increases to a certain threshold. These substances are gradually metabolized during sleep, which leads to a return to the waking state. Rosenbaum (1892) hypothesized that sleep is regulated by humoral factors; i.e., excess water accumulation due to oxidative processes in nerve cells during wakefulness depresses neuronal excitability and removal of the excess water during sleep restores full brain activity, resulting in wakefulness. A few years later, Ishimori (1909) and Legendre and Pieron (1913) independently demonstrated the existence of sleep-promoting hypnogenic substances, also known as "hypnotoxins," in the cerebrospinal fluid of sleep-deprived dogs (Kubota, 1989; Inoué et al., 1995).

The hypnotic effect of adenosine in the mammalian brain was discovered in 1954 (Feldberg and Sherwood, 1954). Adenosine as a neuromodulator with somnogenic properties should thus be classified as a sleep substance. Extensive evidence also suggests that components of the immune system, such as proinflammatory cytokines (Krueger et al., 1984, 2001; Mullington et al., 2000, 2001; Krueger and Majde, 2003) [for review, see (Krueger et al., 2011)] and prostaglandins (Ushikubi et al., 1998; Lazarus et al., 2007; Urade and Lazarus, 2013; Oishi et al., 2015) [for review, see (Urade and Lazarus, 2013)], are interrelated with the regulation of sleep. The involvement of other putative hypnogenic substances, including anandamide (Garcia-Garcia et al., 2009), urotensin-II peptide (Huitron-Resendiz et al., 2005), and the Drosophila peptide NEMURI (Toda et al., 2019), is also implicated in the sleep process.

### Circuit-Based Theories of Sleep-Wake Regulation

A slow humoral process, however, cannot sufficiently explain the reversibility of sleep, especially rapid transitions from sleep to wake in response to external stimuli. Experimental work by Constantin von Economo in the early 20th century produced findings that inspired circuit-based theories of sleep/wake regulation. In 1916, von Economo began to see patients with a new type of encephalitis eventually referred to as encephalitis lethargica or von Economo's sleeping sickness. The disorder was characterized by lesions in the anterior hypothalamus leading to prolonged insomnia or lesions at the junction of the brainstem and forebrain leading to prolonged sleepiness (von Economo, 1917; Economo, 1930). Von Economo concluded that these brain areas must play a role in sleep/wake regulation. The "passive theory," which dominated in the 1940/1950s, suggested that sleep occurs passively due to decreased activity of the brainstem reticular formation (Bremer, 1938; Moruzzi and Magoun, 1949). Importantly, this "passive theory" implicates a necessary active neuronal modulation to maintain a behavioral state of wake via the ascending reticular activating system. Although overly restrictive to the reticular activating system with regard to the wakemodulatory components, the principle of a necessary activation for wake cannot be ruled out; nor can an active sleep-promoting modulation be ruled out, as these are not mutually exclusive types of modulation.

Many decades later, neurons that are active when animals sleep were identified in the ventrolateral preoptic area (VLPO) near the third ventricle in the anterior part of the hypothalamus (Sherin et al., 1996; Chung et al., 2017). Studies demonstrated that sleep is promoted by projections from the GABAergic preoptic area (POA), including the VLPO, to the tuberomammillary nucleus (TMN) in the posterior hypothalamus (Sherin et al., 1998; Chung et al., 2017), which contains neurons that produce histamine, a neurotransmitter having an important role in arousal (Huang et al., 2006; Haas et al., 2008; Oishi et al., 2008). These findings provided strong evidence of sleep control by the POA-TMN neural pathway.

Neural circuits in the brainstem and basal ganglia also regulate sleep/wake behavior. The parafacial zone (PZ) in the medulla contains sleep-promoting GABAergic neurons (Anaclet et al., 2012, 2014) that project to the parabrachial nucleus (PB), a critical nucleus for cortical activation as lesions of the PB result in a comatose state (Fuller et al., 2011).

More recently, the involvement of dopaminergic neurons in the ventral tegmental area (VTA) was strongly implicated in the arousal effect (Eban-Rothschild et al., 2016; Oishi et al., 2017a). A role for dopamine in arousal is also supported by evidence that amphetamine, which induces the release of monoamines (including dopamine), increases alertness and psychomotor performance in sleep-deprived individuals (Bonnet et al., 2005). Dopamine transporters are necessary for the wake-promoting effects of amphetamine (Wisor et al., 2001). Ablating or suppressing GABAergic neurons in the ventral medial midbrain/pons (VMP), including the VTA, produces wakefulness and prevents sleep mainly through dopaminergic systems (Takata et al., 2018; Yu et al., 2019). Furthermore, the ability of dopamine neurons in the VTA to promote wakefulness is at least in part mediated by projections to the nucleus accumbens (NAc; Eban-Rothschild et al., 2016). Medium spiny GABAergic neurons in the NAc can be divided into two groups that respond differentially to stimulation by dopamine or adenosine. Direct pathway neurons express excitatory dopamine D<sup>1</sup> receptors and inhibitory adenosine A1Rs, whereas neurons of the indirect pathway express inhibitory dopamine D<sup>2</sup> receptors and excitatory A2ARs. In fact, recent studies showed that NAc direct pathway neurons induce wakefulness (Luo et al., 2018) and A2AR-expressing indirect pathway neurons strongly induce SWS (Oishi et al., 2017b). The indirect pathway neurons in the NAc produce sleep by inhibiting the ventral pallidum (VP) in the basal forebrain (BF), although these neurons also have sparse to moderate projections to other well-known arousal-promoting areas, such as the lateral hypothalamus, which produces orexin, the TMN, and the VTA. Interestingly, chemogenetic activation of the BF, including the VP, largely reduces sleep (Anaclet et al., 2015).

Altogether, it is impossible to abolish sleep completely by lesioning the afore-mentioned inhibitory circuits, including the POA-TMN, PZ-PB, and NAc-VP pathways, making it unlikely that the regulation of sleep time depends on a single center (i.e., a master switch for sleep in the brain may not exist). On the contrary, the existence of various neural circuits controlling sleep suggests that sleep is gated by different processes. All of these sleep/wake circuits clearly modulate an animal's level of arousal (i.e., vigilance) to determine the behavioral state and GABA is the key neurotransmitter for promoting the transition from wake to sleep and the duration of sleep. For example, the observation that the level of wakefulness is regulated by VMP GABAergic neurons (Takata et al., 2018) indicates the ability of the brain to adapt an animal's sleep/wake time to its behavior.

The transition from waking to sleep may be essential, at least under physiologic conditions, for the facilitation of sleep function to occur. A sufficiently increased level of arousal, as may be experimentally or environmentally induced, prevents sleep occurrence and, accordingly, sleep function. The resulting increase in sleep need is normally reflected by an increase in SWA in the ensuing sleep episode. During this ensuing episode, SWS-SWA resolves toward a non-sleep deprived baseline and the threshold for arousal to waking decreases. As a matter of fact, local sleep, i.e., a phenomenon in which discrete regions of cortical neurons go "offline" similar to during sleep, but other regions do not, is insufficient for sleep function to occur (Vyazovskiy et al., 2011), most likely as a result of the brain's massive interconnectivity. Thus the integration of local sleep events into a global sleep state is necessary for effective sleep function, even at a local level.

### Homeostatic Regulation of Sleep (Two-Process Model)

In 1982, Alexander Borbély at the University of Zürich in Switzerland proposed a two-process model of sleep regulation (Borbély, 1982) that currently prevails as a major conceptual framework in sleep research. In a simplified version of the two-process model of sleep regulation, sleep propensities in homeostatic and circadian processes are commonly plotted against the time of day and interactions of the two processes determine the cardinal aspects of sleep regulation. The "homeostatic" process is controlled by the sleep pressure or need that builds up during the waking period and dissipates during sleep. In contrast, the "circadian" process, i.e., the sleep/wake cycle during the day and night, is controlled by a circadian pacemaker or biologic clock. Although it was originally hypothesized that the circadian process is independent of prior sleep and waking, experiments in mice lacking clock genes revealed that clock gene knockout (KO) disrupts not only circadian processes, but also sleep homeostatic processes (Franken, 2013). This suggests overlapping functions for the circadian genes in sleep homeostatic control.

Consistent with the two-process model, a homeostatic response to sleep loss, namely sleep rebound in an animal after sleep deprivation, is considered an essential criterion of sleep. Rebound can reflect an increase in SWS-SWA power and/or an increase in SWS duration along with an increase in consolidation. Of these two rebound parameters, SWS-SWA is better correlated with prior waking time. Although a rebound increase in SWA during SWS is often associated with an increase in SWS time or consolidation, its occurrence may be dissociated from an effect on sleep time (Douglas et al., 2007; Bjorness et al., 2009; Suzuki et al., 2013). Importantly, SWA during SWS is considered to be an indicator of sleep intensity (Borbély and Neuhaus, 1979), providing a dimension beyond time in the recovery from prolonged waking. Interestingly, although sleep rebound is widely observed after sleep loss, some species skip sleep in favor of migration, mating, or other social interactions and do not catch up on lost sleep (Berger and Phillips, 1994; Rattenborg et al., 2004; Lyamin et al., 2005; Fuchs et al., 2009; Thimgan et al., 2010; Lesku et al., 2012). Recently, scientists at the Imperial College London demonstrated that male flies in the presence of another male fly undergo sleep loss that results in a sleep rebound once the male intruder is removed, whereas a resident fly also loses sleep in the presence of a female fly, but shows no sleep rebound when the female fly is removed (Beckwith et al., 2017), suggesting that sexual arousal in flies prevents a homeostatic response to sleep loss. Altogether, there is ample evidence in nature challenging the view that sleep rebound is an inescapable outcome of sleep

loss. Nevertheless, the "rebound" in these cases refers only to sleep duration and not to SWS-SWA intensity. This potential dissociation of SWS duration from SWS-SWA expression (i.e., SWS duration may not reflect the SWA changes shaping sleep homeostasis) suggests that it may not be possible to fully interpret rebound sleep or the lack thereof in the absence of SWS-SWA assessment.

Another limitation of the two-process model is that it defines circadian input as the only allostatic component that drives the balance between waking and sleep. Sleep/wake behavior is also influenced by cognitive and emotional factors (Saper et al., 2005; Fernandez-Mendoza et al., 2014; Mullins et al., 2014) or other basic drives, such as a lack of food, predator confrontation, mating pressure, and seasonal migration (Yamanaka et al., 2003; Cano et al., 2008). The mechanisms by which motivational stimuli or stressors interact with sleep/wake behavior are not easily accounted for by the two-process model. On the other hand, if the exceptional conditions mentioned above primarily affect arousal level and thus gating of sleep, then sleep homeostasis, conceptualized as "process S" in the two-process model, may still occur. Sleep homeostasis, although related to arousal (sleep need can dissipate to the largest extent only during sleep) appears to follow an exponential rate of decay (Franken et al., 2001; Bjorness et al., 2016). With greater sleep need, there is greater rebound SWS-SWA, but the rate of decay is slowed, further enhancing the amount of SWS-SWA expressed (Bjorness et al., 2016).

The increased sleep duration and consolidation associated with increased sleep need may reflect a decreased level of arousal needed for waking although both external sensory input as well as the internal state (likely to include circadian drive, need for food, predator threat, sex drive, etc.) remain as effective determinants. Accordingly, level of arousal and sleep duration are dynamic, relying on the integration of multiple factors in addition to previous waking time.

### Sleep as a State of Adaptive Inactivity

An alternative view proposes that sleep enforces adaptive inactivity to conserve energy when activity is not beneficial (Siegel, 2009). The wide variability in sleep duration across the animal kingdom (Preston et al., 2009) suggests the sleep amount of an animal may be adapted to the species' behavior that is critical for survival. Consequently, animals may have the ability to dispense with sleep when varying ecologic demands favor wakefulness; e.g., the ability of male pectoral sandpipers to maintain high neurobehavioral performance despite greatly reduced sleeping time when competing for mating opportunities in a short annual window of female fertility (Lesku et al., 2012) may contradict the notion that decreased performance is an inescapable outcome of sleep loss. A model of sleep as a state of adaptive inactivity challenges the hypothesis that the sleep state persists because it has a vital physiologic function and proposes that sleep has not evolved for what happens when we are asleep, but rather for the energy-saving absence of activity during sleep. The magnitude of energy savings gained through sleep is still unknown, although a new framework for determining

relative energy savings during sleep was recently described (Schmidt et al., 2017).

The teleological problem of sleep function may arise from the presumption of sleep's evolution from a default state of waking. Humans are likely biased toward this presumption by the egocentricity of waking consciousness. The adaptive inactivity model could be modified by reorientation of the question "why do we sleep" to "why do we wake?" In this model, sleep is considered the state that facilitates vegetative functions like anabolism and replacement of proteins, complex carbohydrates, and complex lipids and organelles. The vegetative functions are clearly not inactivity or passive in terms of energy conservation. On the contrary, there is evidence for increased energy utilization in sleep, such as ATP mobilization and AMP dephosphorylation (Dworak et al., 2010). Moreover, the cellular metabolism of brain tissue does not coincide with the systemic eating and digestion of food. From this perspective, an organism is driven to waking primarily by non-vegetative, life-essential pursuits, such as foraging for food, avoiding predators, and, occasionally, sex, along with an integrated circadian timer (also controlling arousal). This model is thus consistent with an active drive or activating system needed to maintain wake.

### Dissociation of the Arousal State and Sleep Homeostasis

An increase in the response threshold to external stimuli is a core feature of sleep (Zepelin, 1994) and is critical for defining sleep in animals that lack a cortex (for review, see Ho and Sehgal, 2005). Prolonged waking by sleep deprivation increases the arousal threshold during subsequent sleep (Williams et al., 1964; Bonnet, 1985), and stronger stimuli are necessary to prevent sleeping/prolong waking (Blumberg et al., 2004). As with spontaneous waking, arousal during sleep deprivation is modulated by internal and external stimuli, but within the context of sleep deprivation the arousal threshold is typically increased. Arousal-related brain regions show greater activity, as measured by c-fos, under sleep deprivation mediated by exposure to novel environments or social interaction compared with gentle handling alone (Deurveilher et al., 2013). Furthermore, the nature of the homeostatic response to prolonged waking varies across development with increases in sleep time preceding the appearance of increased SWA power (Frank et al., 1998). Finally, SWA power, commonly used as an indicator of homeostatic sleep need, is increased within waking during prolonged sleep deprivation (Franken et al., 2001; Vyazovskiy et al., 2011), but the relationship between SWA and sleep homeostasis can be dissociated in pathologic disorders. SWA power is increased during waking and rapid eye movement (REM) sleep in Alzheimer's disease (Hassainia et al., 1997; Jeong, 2004); whereas SWA is increased during waking but decreased during sleep in schizophrenia (Keshavan et al., 1998; Hoffmann et al., 2000; Fehr et al., 2003). Conversely, faster EEG activity, such as beta and gamma rhythms commonly used as an indicator of arousal, is high during sleep in a subset of insomniacs (Perlis et al., 2001; Dolsen et al., 2017).

# ADENOSINE AND SLEEP

fnins-13-00740 July 15, 2019 Time: 15:26 # 5

# Adenosine Metabolism and Levels During Sleep and Wakefulness

Hydrolysis of AMP and S-adenosylhomocysteine (SAH) produces adenosine (Schrader, 1983; Fredholm, 2007). Adenosine is generated from SAH by SAH hydrolase, which also acts to trap adenosine in the presence of excess L-homocysteine. This takes place intracellularly and the bidirectional actions of the enzyme ensure the constant presence of a particular concentration of adenosine in the cell. Whether SAH hydrolase is involved in generating adenosine in the brain, however, is controversial (Latini and Pedata, 2001). Adenosine is formed intracellularly and extracellularly from 5<sup>0</sup> -AMP by different 5 0 -nucleotidase (5<sup>0</sup> -NT) (Zimmermann, 2000). A cascade of actions by an ecto-5<sup>0</sup> -NT, together with ecto-ATPases, terminates the action of ATP as extracellular signaling molecules (Zimmermann, 2000, 2006; Yegutkin, 2008; Kovacs et al., 2013).

High adenosine levels are reduced by the actions of adenosine deaminase (ADA), or are taken up by cells where adenosine is rapidly phosphorylated to AMP by adenosine kinase (AdK), an enzyme that effectively controls the intracellular adenosine concentration (**Figure 1**; Fredholm et al., 2005; Oishi et al., 2008; Parkinson et al., 2011). Importantly, AdK binds a molecule of ATP and adenosine, catalyzes the transfer of a phosphate group from ATP to adenosine and produces ADP and AMP. As a result, the rate of adenosine metabolism is reflected by the [ATP]/[ADP][AMP] ratio, linking the rate of adenosine metabolism to the metabolic state of the cell. In the adult CNS, AdK expression occurs predominately in the glia (Studer et al., 2006), and thus the concentration of adenosine is controlled by the metabolic state of the glia.

Bi-directional equilibrative nucleoside transporters regulate the concentration of adenosine available to cell surface adenosine receptors (Parkinson et al., 2011; Dos Santos-Rodrigues et al., 2014). Therefore, adenosine levels are dependent on the formation and removal of extracellular adenosine. Extracellular adenosine levels are low under basal conditions – approximately 30 to 300 nM (Ballarin et al., 1991), but may exceed 1 µM under more extreme conditions, such as mild hypoxia or strenuous exercise, and can reach up to several tens of micromolar concentration in severely traumatic situations, including local ischemia (Fredholm, 2007).

Adenosine triphosphate depletion and an increase of extracellular adenosine levels are positively correlated (Kalinchuk et al., 2003) and positively associated with sleep (Rainnie et al., 1994; Porkka-Heiskanen et al., 1997). Thus, adenosine may represent a state of relative energy deficiency. During spontaneous sleep/wake behavior in cats, adenosine levels in several brain regions are higher during SWS than wakefulness (Porkka-Heiskanen et al., 1997, 2000). Moreover, in vivo microdialysis studies in cats revealed that adenosine concentrations increase 2-fold in the BF during a prolonged 6-h period of wakefulness compared with that at the beginning of sleep deprivation. Under more chronic sleep deprivation protocols, however, increases in adenosine concentrations during prolonged waking are no longer observed (Clasadonte et al., 2014), suggesting that loss of the adenosine response may mark a shift from a homeostatic response to an allostatic response following reduced sleep.

Six decades after the discovery of adenosine's role in sleep, the mammalian brain cell types involved in the sleeppromoting effects of adenosine remain unclear (Feldberg and Sherwood, 1954). ATP, which is rapidly degraded to adenosine, and adenosine are released from glial cells and neurons. In genetically engineered mice in which the release of ATP is nonspecifically blocked in astrocytes by selective expression of a dominant negative SNARE domain, decreased concentrations of extracellular adenosine are observed (Pascual et al., 2005). While the amounts of wakefulness, SWS, and REM sleep in these mice are indistinguishable from those in wild-type mice, these mice exhibit reduced SWA and recovery sleep after sleep deprivation (Halassa et al., 2009). Furthermore, reducing AdK in astrocytes, thereby increasing the adenosine tone, is sufficient to increase SWS-SWA and sleep consolidation, reduce the decrease in SWA across the light phase, and slow the decay of SWS-SWA within an average SWS episode, whereas selectively reducing AdK in neurons has no effect (Bjorness et al., 2016). These observations suggest that adenosine mediates the sleep deprivation-induced homeostatic sleep response. The source of the released adenosine, however, remains controversial. Some of the adenosine may originate from astrocytes and the majority may originate from neurons, but direct proof is lacking and thus the exact source of adenosine remains unknown. On the other hand, control of extracellular adenosine modulating sleep need clearly involves glial metabolism mediated by AdK (Bjorness et al., 2016).

Radulovacki et al. (1983) extensively investigated the effects of adenosine on wakefulness. They found that increasing the levels of adenosine in the central nervous system of rats by systemic administration of the ADA inhibitor deoxycoformycin led to increases in REM and SWS. In addition, Oishi et al. (2008) reported that focal administration of the ADA inhibitor coformycin into the rat TMN, where ADA is dominantly expressed, increases SWS, further supporting a hypnotic role for adenosine.

### Effects of A<sup>1</sup> Receptors and Sleep Homeostasis

Adenosine acting through A1Rs facilitates sleep as non-selective and selective A1R agonists increase sleep and SWA (Radulovacki et al., 1984; Benington et al., 1995), whereas A1R antagonists decrease sleep and SWA (Virus et al., 1990; Methippara et al., 2005; Thakkar et al., 2008). Furthermore, A1R antagonism within the BF reduces the homeostatic sleep and SWA response following acute sleep deprivation (Gass et al., 2009). Conditional KO of A1Rs predominantly affecting forebrain glutamatergic neurons prevents sleep deprivation-induced increases in SWA, indicating that A1Rs are necessary for normal sleep homeostasis (Bjorness et al., 2009). In mixed background mice with constitutive KO of A1Rs, the normal sleep homeostatic response is maintained as measured by slow wave energy [SWE; SWA (0.5–4.5 Hz) × time] in SWS (Stenberg et al., 2003).

Further, acute application of a selective A1R antagonist blocks the homeostatic response of increased SWS-SWA in sleep-deprived wild-type mice, but is ineffective in the constitutive KO mice under the same conditions (Stenberg et al., 2003). This finding suggests the presence of compensatory mechanisms in mice with constitutive KO that were not present in mice with conditional KO. Sleep facilitation via A1Rs occurs through inhibition of wake-active neurons in several brain areas, including both the brainstem and forebrain regions of the cholinergic arousal system [mesopontine tegmentum (Rainnie et al., 1994) and BF (Alam et al., 1999; Thakkar et al., 2003)], and the lateral hypothalamus containing hypocretin/orexin neurons (Liu and Gao, 2007). Additionally, administration of a selective A1R agonist into the TMN decreases histamine in the frontal cortex while increasing sleep and SWA (Oishi et al., 2008), suggesting that adenosine also inhibits activity of this wake-promoting neurotransmitter system. An additional mechanism by which adenosine facilitates sleep through A1Rs is by disinhibiting sleep-active neurons in the VLPO and anterior hypothalamic area (Chamberlin et al., 2003; Morairty et al., 2004). Finally, A1Rs mediate homeostatic sleep pressure based on astrocytic gliotransmission (Halassa et al., 2009) and as part of a glial-neuronal circuit (Bjorness et al., 2016).

Prolonged waking through sleep deprivation increases the expression of A1Rs in both humans and rodents (Basheer et al., 2007; Elmenhorst et al., 2007), with expression levels normalizing after recovery sleep in humans (Elmenhorst et al., 2017).

As mentioned above, SWA power is the primary indicator of homeostatic sleep need. SWA power reflects both the number of cells firing at SWA frequencies, which is an intrinsic feature of thalamocortical neurons (McCormick and Pape, 1990; Dossi et al., 1992), and the synchronicity of firing across neurons, which is a circuit effect involving cortical neurons, thalamocortical neurons, and neurons of the reticular nucleus of the thalamus (Steriade et al., 1993). Activation of A1Rs influences SWA by both direct and indirect mechanisms; the direct mechanism is based on presynaptic inhibition of cortical and thalamic neurons, which results in relative functional deafferentation along with an A1R-induced increase in whole cell, GIRK channel conductance and decreased hyperpolarization activated currents (Ih), such that adenosine enhances slow oscillations in thalamocortical neurons (Pape, 1992). The indirect mechanism is a reduction of cholinergic tone by A1Rmediated inhibition of cholinergic arousal neurons (Rainnie et al., 1994; Porkka-Heiskanen et al., 1997). Acetylcholine inhibits slow oscillation in thalamocortical neurons (Curro Dossi et al., 1991; Steriade et al., 1991; McCormick, 1993); thus reduction of cholinergic tone is permissive for the expression of SWA.

Infusion of the selective A2AR agonist CGS21680 into the subarachnoid space below the ventral surface region of the rostral BF in rats or into the lateral ventricle of mice produces robust increases in SWS and REM sleep (Satoh et al., 1996; Urade et al., 2003). In vivo microdialysis experiments, infusing CGS21680 into the BF dose-dependently decreases histamine release in the frontal cortex and medial preoptic area, and increases the release of GABA in the TMN, but not in the frontal cortex (Hong et al., 2005). Infusion of the GABA antagonist picrotoxin into the TMN attenuates the CGS21680 induced inhibition of histamine release, suggesting that the A2AR agonist induces sleep by inhibiting the histaminergic system through increasing the release of GABA in the TMN. Intracellular recordings of VLPO neurons in rat brain slices demonstrated that two distinct types of VLPO neurons exist in terms of their responses to serotonin and adenosine. VLPO neurons are inhibited by noradrenaline, acetylcholine, and an A1R agonist, whereas serotonin inhibits type-1 neurons, but excites type-2 neurons. An A2AR agonist post-synaptically excites type-2, but not type-1, neurons. These findings suggest that type-2 neurons are involved in initiating sleep, whereas type-1 neurons may contribute to sleep consolidation, because they are only activated in the absence of inhibitory effects from wake-inducing systems (Gallopin et al., 2005).

Administration of CGS21680 into the rostral BF, however, produces c-fos expression not only in the VLPO, but also within the NAc shell and the medial portion of the olfactory tubercle (Satoh et al., 1999; Scammell et al., 2001). Direct infusion of the A2AR agonist into the NAc induces SWS that corresponds to approximately 75% of the sleep amount measured when the A2AR agonist is infused into the subarachnoid space (Satoh et al., 1999). This observation may indicate that activating A2ARs within or close to the NAc induces sleep. Acting opposite to adenosine, caffeine, which is the most widely consumed psychostimulant in the world, enhances wakefulness because it acts to antagonize both A1R and A2AR subtypes. At doses commonly consumed by humans, caffeine partially (estimated as 25–50%) and non-selectively (similar affinity for both A1Rs and A2ARs) blocks adenosine receptors (Fredholm et al., 1999). Experiments using mice with global genetic A1R and A2AR KO revealed that A2ARs, but not A1Rs, mediate the wakefulnessinducing effect of caffeine (Huang et al., 2005), while single nucleotide mutations of the A2AR gene confer sensitivity to caffeine and sleep deprivation (Bodenmann et al., 2012). The specific role of A2ARs in the striatum was investigated in conditional A2AR KO mice based on the Cre/lox technology and local infection with AAV carrying short-hairpin RNA of the A2AR to silence the expression of the receptor. Selective deletion of the A2ARs in the NAc shell blocked caffeine-induced wakefulness (Lazarus et al., 2011).

For caffeine to be effective as an A2AR antagonist, adenosine must tonically activate excitatory A2ARs within the NAc shell. This activation likely occurs in the NAc shell because A2ARs are abundantly expressed throughout the striatum,

FIGURE 2 | Adenosine receptors influence sleep/wake behavior by modulating the arousal level through A2ARs or A1Rs and the sleep need through A1Rs. Increased activity of the arousal centers promotes wakefulness. For example, activation of A2ARs in the nucleus accumbens (NAc) and hypothalamus facilitates sleep through the inhibition of arousal-promoting neurons. The duration of wake time positively correlates with sleep need and the buildup of extracellular adenosine. The buildup of adenosine in the cortex and thalamus increases SWS-SWA through the activation of A1Rs. Sleep need also increases the probability of a state change from wake to sleep, primarily by decreasing arousal center activity (in part by activating A1Rs in arousal centers and A2ARs in the NAc). The sleep state is permissive for sleep function that resolves the sleep need (as sleep function is accomplished), as reflected by the resolution of rebound SWS-SWA.

including the NAc shell and sufficient levels of adenosine are available under basal conditions (Rosin et al., 1998; Svenningsson et al., 1999). A recent study showed that chemogenetic or optogenetic activation of NAc A2AR core neurons projecting to the VP in the BF strongly induces SWS, whereas chemogenetic inhibition of these neurons prevents sleep induction, but does not affect homoeostatic sleep rebound (Oishi et al., 2017b). Interestingly, motivational stimuli suppress sleep and inhibit the activity of VP-projecting NAc A2AR neurons. In addition, another recent study revealed that adenosine is a plausible candidate molecule for activating NAc core A2AR neurons to induce SWS because elevated adenosine levels in the NAc core promote SWS via A2ARs (Zhou et al., 2019).

The sleep-gating ability of the NAc indirect pathway may explain the tendency toward falling asleep in boring situations. Interestingly, excessive daytime sleepiness is common in children with attention-deficit/hyperactivity disorder, who frequently start napping or daydream when they are bored (Weinberg and Brumback, 1990). Dopamine produced by VTA neurons has a key role in processing reward, aversive, or cognitive signals (Wise, 2004; Bromberg-Martin et al., 2010; Schultz, 2015), and projections from VTA dopaminergic neurons to the NAc, commonly known as the mesolimbic pathway, constitute a well-characterized reward circuit in the brain (Russo and Nestler, 2013; Volkow and Morales, 2015). Two independent studies recently examined the contribution of VTA

dopaminergic neurons to wakefulness under baseline conditions by chemogenetic inhibition. One study found that chemogenetic inhibition of VTA dopamine neurons decreases the amount of wakefulness, thus suggesting that these neurons are necessary for baseline wakefulness in mice (Eban-Rothschild et al., 2016). The other study showed that chemogenetic inhibition of VTA dopamine neurons does not significantly affect wakefulness at baseline in mice (Oishi et al., 2017a). A plausible explanation for the differences in the observations in these studies is different ectopic Cre expression (Lammel et al., 2015) in the midbrain of the tyrosine hydroxylase-Cre mice used by Eban-Rothschild et al. (2016) or the dopamine transporter-Cre mice used by Oishi et al. (2017a).

### UNIFIED MODEL OF SLEEP-WAKE REGULATION: GATING OF SLEEP HOMEOSTASIS BY AROUSAL

As knowledge of the molecular and circuit bases of sleep/wake regulation expands, new roles of adenosine receptors in modulating different aspects of sleep emerge. For example, A2ARs appear to promote sleep by suppressing arousal, whereas sleep need and the response to sleep deprivation are mediated by A1Rs, and these receptors may thus play a crucial role in the function of sleep. In light of the dissociable effects of adenosine for gating sleep and mediating sleep need at the receptor level, we propose a model of sleep-wake regulation in which the sleep state is regulated by arousal when an organism must consolidate wakefulness in response to environmental changes (**Figure 2**). A typical example is motivated behavior that efficiently suppresses sleep of all stages and produces arousal by utilizing mesolimbic dopaminergic systems, whereas the wake state is suppressed in the absence of motivating stimuli by activation of A2ARs in the NAc (Oishi and Lazarus, 2017; Oishi et al., 2017b). The circadian and hypothalamic feeding systems have indirect influences by driving internally generated arousal, e.g., increasing motivation to forage according to the circadian phase. Thus in the absence of motivating/external arousing stimuli, the loss of the arousing influence of the circadian system (the sleep phase) may be sufficient to allow transition to sleep. On the other hand, sleep is necessary for SWS-SWA to facilitate the expression of sleep need and for the resolution of sleep need, a process in which A1Rs play a crucial role.

# CONCLUSION

Adenosine is a well-known somnogenic substance that affects normal sleep-wake patterns. While the source of the adenosine involved in sleep remains poorly understood, the metabolism of adenosine in the CNS is significantly mediated by adenosine kinase, which modulates the concentration of adenosine at neuronal A1R sites. Similarly, adenosine promotes sleep by several mechanisms in various locations via A1Rs or A2ARs.

Adenosine receptor stimulation should be considered as a potential treatment for insomnia. Insomnia is a sleep disorder affecting millions of people around the world and frequently co-occurs with a wide range of psychiatric disorders (Roth, 2007; de Zambotti et al., 2018; Seow et al., 2018). Although A2AR agonists strongly induce sleep, classical A2AR agonists have adverse cardiovascular effects and cannot be used clinically to treat sleep disorders. Moreover, the development of adenosine analogs for treating central nervous system disorders, including insomnia, is hampered by the poor transport of these drugs across the blood-brain barrier. A small blood brain barrier-permeable monocarboxylate was recently demonstrated to induce sleep by enhancing A2AR signaling in the brain, and surprisingly did not exhibit the typical cardiovascular effects of A2AR agonists (Korkutata et al., 2017). Therefore, molecules that allosterically enhance A2AR signaling could help people with insomnia fall asleep and may also be a potential treatment for psychiatric illness. Similarly, molecules that enhance A1R signaling might enhance sleep efficiency.

### AUTHOR CONTRIBUTIONS

All authors wrote the review, and approved it for publication.

# FUNDING

Our work was generously supported by the Japan Society for the Promotion of Science [Grants-in-Aid for Scientific Research B (grant number 17H02215) to ML]; the Japan Science and Technology Agency [CREST (grant number JPMJCR1655) to ML]; the Ministry of Education, Culture, Sports, Science and Technology (MEXT) of Japan [Grantsin-Aid for Scientific Research on Innovative Areas "Living in Space" (grant numbers 15H05935, 15K21745, and 18H04966) and "WillDynamics" (grant number 19H05004) to ML]; the World Premier International Research Center Initiative (WPI) from MEXT (to ML, YO, and RG.); a research grant from the Astellas Foundation for Research on Metabolic Disorders (to ML); the National Institutes of Health (grant numbers MH 06777 and NS075545 to RG); and the Department of Veterans Affairs through Dallas VA Medical Center (grant numbers MH79710 and MH083711 to RG). The contents of this review do not represent the views of the United States Department of Veterans Affairs or the United States Government.

# ACKNOWLEDGMENTS

We appreciate the research spirit and diligence of all the colleagues who contributed to this fascinating and lively research area.

### REFERENCES




Frontiers in Neuroscience | www.frontiersin.org

postnatal development suggests dual functionality of the enzyme. Neuroscience 142, 125–137. doi: 10.1016/j.neuroscience.2006.06.016


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Lazarus, Oishi, Bjorness and Greene. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Transition Between Slow-Wave Sleep and REM Sleep Constitutes an Independent Sleep Stage Organized by Cholinergic Mechanisms in the Rostrodorsal Pontine Tegmentum

Carlos Carrera-Cañas, Miguel Garzón and Isabel de Andrés\*

Departamento de Anatomía, Histología y Neurociencia, Facultad de Medicina, Universidad Autónoma de Madrid, Madrid, Spain

There is little information on either the transition state occurring between slow-wave sleep (SWS) and rapid eye movement (REM) sleep, as well as about its neurobiological bases. This transition state, which is known as the intermediate state (IS), is welldefined in rats but poorly characterized in cats. Previous studies in our laboratory demonstrated that cholinergic stimulation of the perilocus coeruleus α nucleus (PLCα) in the pontine tegmentum of cats induced two states: wakefulness with muscle atonia and a state of dissociated sleep we have called the SPGO state. The SPGO state has characteristics in common with the IS, such including the presence of ponto-geniculooccipital waves (PGO) and EEG synchronization with δ wave reduction. Therefore, the aims of the present study were (1) to characterize the IS in the cat and, (2), to study the analogy between the SPGO and the different sleep stages showing PGO activity, including the IS. Polygraphic recordings of 10 cats were used. In seven cats carbachol microinjections (20–30 nL, 0.01–0.1 M) were delivered in the PLCα. In the different states, PGO waves were analyzed and power spectra obtained for the δ, θ, α, and β bands of the EEG from the frontal and occipital cortices, and for the θ hippocampal band. Statistical comparisons were made between the values obtained from the different states. The results indicate that the IS constitutes a state with characteristics that are distinct from both the preceding SWS and the following REM sleep, and that SPGO presents a high analogy with the IS. Therefore, the SPGO state induced by administering carbachol in the PLCα nucleus seems to be an expression of the physiological IS of the cat. Consequently, we propose that the PLCα region, besides being involved in the mechanisms of muscle atonia, may also be responsible for organizing the transition from SWS to REM sleep.

Keywords: intermediate state, transition state, perilocus coeruleus α, REM sleep, slow-wave sleep, carbachol, polygraphic recordings, cat

# INTRODUCTION

Multiple transitions occur between the different stages of the sleep wakefulness cycle (SWC) but these transitions are not so well-characterized as the SWC stages themselves. Transition states are difficult to analyze because of their short duration and complexity. However, these periods are critical since they involve changes in multiple brain areas that are essential for

### Edited by:

Michael Lazarus, University of Tsukuba, Japan

### Reviewed by:

Ramalingam Vetrivelan, Beth Israel Deaconess Medical Center, Harvard Medical School, United States Pablo Torterolo, Universidad de la República, Uruguay

> \*Correspondence: Isabel de Andrés isabel.deandres@uam.es

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 18 March 2019 Accepted: 05 July 2019 Published: 23 July 2019

### Citation:

Carrera-Cañas C, Garzón M and de Andrés I (2019) The Transition Between Slow-Wave Sleep and REM Sleep Constitutes an Independent Sleep Stage Organized by Cholinergic Mechanisms in the Rostrodorsal Pontine Tegmentum. Front. Neurosci. 13:748. doi: 10.3389/fnins.2019.00748

**188**

the proper establishment of each sleep stage (Durán et al., 2018; Sánchez-López et al., 2018). One of these critical transition states is the slow-wave sleep (SWS) to rapid eye movement (REM) sleep transition.

In cats, pontogeniculooccipital (PGO) activity is not only associated to REM sleep, it also appears about 1 min before REM sleep onset during SWS; therefore, the SWS with PGOs was considered the transitional period in this species (Ursin and Sterman, 1981). Pioneer studies in the rat had defined a transition state of shorter duration (1 to 5 s) that preceded REM sleep called the intermediate state (IS) (Gottesmann, 1973). The IS had large-amplitude sleep spindles in the frontal cortex electroencephalogram (EEG) and low frequency θ activity in the dorsal hippocampus (Gottesmann, 1973). The IS was also studied in the cat where it was found to share the same EEG characteristics of the rat IS, with the addition that PGO waves could also be observed (Gottesmann et al., 1984). The latter authors also reported that the IS was shorter in cats than in rats (1 to 3 s), only present in 30% of the animals, and one or more IS episodes could be observed just before and occasionally even after REM sleep. Therefore, for those authors the cat IS was not the whole SWS interval preceding REM sleep in which PGO activity was already present. In relation with IS mechanisms, it was reported that the forebrain would be transiently disconnected from the brainstem during IS in cats and rats (User et al., 1980; Gottesmann and Gandolfo, 1986; Piallat and Gottesmann, 1995; Gottesmann, 1996).

Recently the IS has again attracted interest, particularly the temporal sequence of the events that occur within this stage. On the one hand, rat studies on the dynamics of sleep stages in different brain structures, such as the neocortex and the hippocampus, have indicated that there may be functional dissociations between these structures and it seems that the sleep state showing the least concordance among the different brain structures is the IS (Durán et al., 2018). According to the latter authors, IS, which would have an average duration of 18 s, is identified by a decrease in δ activity accompanied by increased θ activity and the presence of sleep spindles, although these activities may not be simultaneous in the different cortical regions. On the other hand, high frequency oscillations (HFO, 110–160 Hz), which are modulated by θ rhythm and modified by behavioral state (Scheffzük et al., 2011; Cavelli et al., 2018), have been reported to completely delimit the IS in the rat since they are minimal during SWS and maximum during REM sleep (Sánchez-López et al., 2018). Therefore, a new definition for the IS has been proposed in the rat. It would be a period of approximately 20 s during which electroencephalographic variations occur before motor changes. In relation with IS mechanisms, structures located in the dorsal oral pontine tegmentum, like the sublaterodorsal nucleus (SLD) of the rat, have been proposed as being responsible for the coordination of all these events taking place during IS (Sánchez-López et al., 2018). In fact, REM sleep is enhanced after SLD glutamatergic stimulation or GABAergic disinhibition of this region, however, SLD cholinergic stimulation of this region produces wakefulness (W) with muscle atonia (Boissard et al., 2002).

The perilocus coeruleus α nucleus (PLCα) in the rostrodorsal pontine tegmentum of humans and cats or its homolog in rats, the SLD, is a part of the locus coeruleus complex (LCC) together with the locus coeruleus α (LCα) and the subcoeruleus (SCoe) nuclei (**Figure 5**). These regions contain cholinergic, noradrenergic and serotonergic neurons (Sakai et al., 2001; Reinoso-Suárez et al., 2011), as well as glutamatergic and GABAergic neurons (Boissard et al., 2002; Lu et al., 2006). The PLCα shows active neurons during REM sleep (PS-on neurons, Sakai et al., 2001) and injury or inhibition of that region can decrease or abolish REM sleep (Crochet and Sakai, 1999). Besides, the PLCα was initially considered as the region responsible for cholinergic REM sleep generation (Sakai, 1988; Vanni-Mercier et al., 1989). However, pioneer studies in the cat had indicated that cholinergic stimulation of the dorsal pontine tegmentum produced W without muscle tone (Van Dongen et al., 1978). Later studies with carbachol microinjections in the dorsal oral pontine tegmentum specifically located in the PLCα confirmed these results (Reinoso-Suárez et al., 1994), in contrast to the clear triggering of REM sleep produced by cholinergic stimulation of the near ventral part of the oral pontine reticular nucleus (vRPO, Reinoso-Suárez et al., 1994; Garzón et al., 1997, 1998). A more recent study (Moreno-Balandrán et al., 2008) has shown that small-volume low doses of carbachol (0.01–0.1 M) in the PLCα produce W with atonia alternating with another state in which the cat shows behavioral sleep, but shows a polygraphic pattern that does not conform to the typical characteristics of the spontaneous stages of SWS or REM sleep, although it has features common to both. Like physiological REM sleep, the sleep induced by cholinergic stimulation of the PLCα nucleus presents muscle atonia and PGO activity. Nevertheless, the EEG presents a striking synchronization but the typical δ activity of SWS is reduced while θ and α activity is increased. In addition, the hippocampal EEG also shows θ activity that is less rhythmic than that during spontaneous REM sleep (Moreno-Balandrán et al., 2008). These features of carbachol PLCα-induced sleep are reminiscent of the characteristics of IS in the transition between SWS and REM sleep (User et al., 1980; Gottesmann et al., 1984; Gottesmann and Gandolfo, 1986; Piallat and Gottesmann, 1995; Gottesmann, 1996; Durán et al., 2018; Sánchez-López et al., 2018). In order to establish the functional significance of PLCα in relation to the SWC mechanisms, these observations led us to further analyze the effects of low carbachol doses in this region, investigating their relationship with physiological SWS and REM sleep but also with the IS. Additionally, in light of the scant existing knowledge on cat IS (Gottesmann et al., 1984), we have also tried to further characterize the bioelectric features of IS in this species so as to compare it with carbachol-PLCα induced sleep.

### MATERIALS AND METHODS

### Subjects and Surgery

We used 10 adult cats (nine males and one female). All experiments were carried out in accordance with the European

Community Council Directive (2010/63/UE) and approved by the Institutional Animal Care and Use Committee of the Universidad Autónoma of Madrid (Spain) and the competent regional government agency (PROEX 004/15). Surgery was performed under general anesthesia [Medetomidina (Domtor) 0.1 mg/kg i.m. and Pentobarbital (Dolethal) 14 mg/kg i.p. using aseptic techniques. Using the coordinates of the Reinoso-Suarez (1961) cat brain atlas, standard steel screws welded to a conductor cable for EEG recording were placed bilaterally on the skull over the frontal somato-motor cortex (2 mm rostral to Bregma and 10 mm from the midline) and the occipital primary visual cortex (21 mm caudal to Bregma and 2 mm from the midline). A reference screw electrode was implanted in the midline over the frontal sinus. Silver wires (380 µm diameter) were attached subcutaneously to both supraorbital ridges to record the electrooculogram (EOG). Stainless steel wires were implanted in the neck muscles for electromyogram (EMG) recordings. Twisted stainless steel wires were placed subcortically with stereotaxic methods to record PGO activity in the lateral geniculate nucleus (LGN, AP, +7; V, +14; L, ±10) and hippocampal EEG in the CA1 region (AP, +3; V, +16; L, ±6). We also implanted a stainless steel guide tube or cannula (20 gauge) provided with a stylet. The cannula was inserted at a 32◦ angle to the coronal plane through the posterior fossa and, its tip, aimed at the PLCα nucleus (AP, −2/−3; V, 7; L, ±2), was left 4 mm above the target. All wires were affixed to an Amphenol strip connector, which, together with the cannula, was anchored to the skull with acrylic cement. After surgery, animals were given antibiotic and analgesic treatment for 5 days.

### Polygraphic Recordings and Drug Administration

After 8–10 days of recovery, each cat was placed in a soundproof ventilated recording chamber with constant temperature (22 ± 2 ◦C) and dark/light cycle (12/12 h lights on at 7:00 h) for a 3–5 day habituation period. Food and water were supplied ad libitum. The animals were video recorded while they were in the soundproof chamber. Thereafter, 6 h polygraphic-video recordings starting at 10:30–11:00 h were made at 1 week intervals on a randomized schedule after PLCα unilateral microinjections of sterile saline (for baseline o control recordings) and after two doses of carbachol (0.01 and 0.1 M, Carbamylcholine chloride; Sigma), a long lasting cholinergic agonist.

All of the microinjections were performed with the cats awake and gently restrained, introducing the needle of a 0.5 µL Hamilton syringe through the cannula. The tip of the syringe protruded 4 mm from the end of the cannula and 20–30 nL of saline/carbachol were delivered. The syringe was left in place for 1 min before its removal in order to avoid leakage of the microinjected solutions along the needle track. Afterward, cats were placed in the soundproof chamber to immediately begin the polygraphic-video recordings.

The first 3 h of the polygraphic recordings were simultaneously digitized, filtered at 0.3–30 Hz and fed into a computer at a sampling frequency of 200 Hz for offline analysis. Spike2-CED software (Cambridge Electronic Design, Cambridge, United Kingdom) was used.

### Histology

At the end of the experiments the animals were given an overdose of Pentobarbital (100 mg/kg intraperitoneal) and perfused transcardially with saline, 10% formalin and increasing concentrations of sucrose (5, 10, and 20%). Frozen coronal sections of the brain were serially cut at 40–50 µm thicknesses, stained (Nissl) and examined. The site of the microinjections as well as the position of the subcortical electrodes were respectively identified in all the cats by locating the tip of the needle tract or the tip of the electrodes. The cat brain atlas of Reinoso-Suarez (1961) was used for the analysis.

### Analysis of the Polygraphic Recordings

Using polygraphic patterns for the physiological SWC stages in the cat (Ursin and Sterman, 1981), one episode containing the last minute of SWS with PGO activity together with the first minute of REM sleep was selected from baseline-saline recordings to detect and characterize IS (n = 10 cats). In these samples the occipital cortex δ band was isolated using the Spike2-CED software. For that, we used a 4th order bandpass digital filter of the "Infinity Impulse Response" type (IIR) and "Butterworth" model, which allowed us to generate a new signal that only contained frequencies between 1.5 and 3.5 Hz. From the generated signal, its root mean square was calculated using a time constant of 0.05 s, and it was smoothed with the "smooth" filter using a time constant of 20 s. This made it possible to obtain a practically linear signal that represented the amplitude of the occipital cortex δ band over time (see section "Characterization of the Intermediate State" and **Figure 1**). This signal showed a progressive decline in the last seconds of SWS that reached minimum amplitude once EEG desynchronization of REM sleep was established. The interval with the highest slope values (see **Figure 1**) was taken into account in each cat to assess it as the IS. To determine whether or not these intervals actually corresponded to the IS, their bioelectric characteristics were compared with those of the SWS and REM sleep intervals of the same duration immediately before and after them.

After carbachol delivery in the PLCα nucleus, wakefulness with muscle atonia (Wa) and EEG synchronization with PGO waves (SPGO) were the main states observed. Latency to onset and time spent in these or the other stages of the SWC were quantified for the two first hours of recording. When REM sleep did not appear in those 2 h, the analysis continued until finding the first episode of REM sleep after PLCα carbachol microinjection. Additionally, for further off line analyses, the selected episodes of SWS and REM sleep on the baseline recordings and three 1-min SPGO epochs from the first oneand-a-half hour of experimental carbachol recordings (taken at 20 min intervals) were taken into account.

In the different polygraphic recording samples, the Spike2- CED software was used to obtain power spectra (frequency range: 0–20 Hz; resolution: 0.39 Hz) for the neocortical and hippocampal EEG from the selected intervals. The power values

of EEG bands δ (0–3.5 Hz), θ (3.5–7.8 Hz), α (7.8–14 Hz), and β (14–20 Hz) were normalized in percentages of total power. In order to avoid signal contamination produced by ocular movements, the 0–1.5 Hz interval on the δ band was discarded in all cases. Moreover, PGO waves were quantified and the EMG signal was visually analyzed.

### Statistical Analysis

The Statview statistical package was used for the analysis. Oneway ANOVAs for repeated measures were carried out to compare the relative values of the δ, θ, α, and β bands between: (1), IS and same length intervals of the previous SWS and the following REM sleep in control recordings; (2), 1-min periods of the three SPGO samples obtained after the low dose of carbachol in PLCα and both SWS and REM sleep of control experiments; and (3), the three SPGO samples obtained after low-dose carbachol and IS of control recordings. PGO waves were quantified and compared between the different states. In the cases in which the ANOVAs showed significant values, post hoc comparisons between value pairs were carried out applying Fisher's least difference test. Statistical significance was set at p ≤ 0.05.

### RESULTS

### Characterization of the Intermediate State

A close examination of the 1-min SWS with PGO activity that preceded REM sleep indicated that the EEG of the frontal and occipital cortices did not abruptly change from a synchronized SWS pattern to a fully desynchronized one in the REM sleep episodes. The EEG in both the frontal and occipital cortices showed a noticeable tendency to desynchronization in the last few seconds of SWS (see **Figure 1**). This fact was welldocumented when a root mean square analysis was applied to determine the voltage variation in the occipital cortex δ band in the 1-min SWS with PGOs and in the succeeding first minute of REM sleep samples. The signal showed a high and quite stable amplitude during SWS except for the last seconds, when a progressive decline took place before reaching the minimum amplitude once desynchronized EEG of REM sleep was well-established (**Figure 1**). The duration of these intervals showed some variability in the different cats, presenting a mean length of 14.6 ± 1.3 s (n = 10 cats). Also, as reflected by the frontal and occipital cortical power spectra (1.5–20 Hz) (**Figure 2**), the interval presenting the progressive decline in δ band power (**Figure 2B**) had an intermediate level of cortical synchronization (for the frontal and the occipital cortices) between the maximal EEG synchronization present in the preceding seconds of SWS (**Figure 2A**) and the minimal EEG synchronization in the following seconds that showed all the REM sleep characteristics (**Figure 2C**). Quantitative analysis comparing the total power values (1.5–20 Hz) of the frontal and occipital power spectra in these three intervals in the 10 cats (that is, between the interval showing the highest δ voltage variation and the two other accompanying intervals before and after it) confirmed the intermediate value of EEG synchronization in the former interval in comparison with the other two (**Figure 3**). Therefore, in terms of general cortical EEG synchronization/desynchronization, the interval showing the highest voltage variation for the δ band in the occipital cortex

represented an intermediate state (IS) between the previous seconds, which had the typical EEG characteristics of SWS, and the subsequent ones, which had unquestionable EEG REM sleep features. Hippocampal θ in the three intervals also confirmed the intermediate character of these IS intervals in the 10 cats (**Figures 2**, **4C**).

After having determined the duration of the IS in each of the 10 cats, further quantitative analyses were carried out. The δ, θ, α, and β bands normalized power values (as percentages of the total 1.5–20.0 Hz power) of the cortical EEG and hippocampal θ were compared between IS and the intervals of the same duration of the stages that preceded (SWS) and followed (REM sleep) it. Frontal cortex one-way ANOVAs for repeated measures showed significant differences for the δ (F2,<sup>29</sup> = 20.433, p = 0.0001), α (F2.<sup>29</sup> = 4.589, p = 0.0245), and β bands (F2,<sup>29</sup> = 4.985, p = 0.0189). However, the differences for the θ band (F2,<sup>29</sup> = 0.234, p = 0.794) were not significant. Post hoc comparisons demonstrated that in frontal cortex (**Figure 4A**, frontal cortex): (1), δ band power was significantly lower during IS than in SWS, but IS had higher values compared with REM sleep; (2), the power of the α band was significantly increased in IS versus SWS; and (3), the β band power was significantly increased in REM sleep with respect to IS. Concerning the occipital cortex, significant changes occurred in the δ (F2,<sup>29</sup> = 55.448, p = 0.0001), θ (F2,<sup>29</sup> = 3.716, p = 0.0446) and β bands (F2,<sup>29</sup> = 25.08, p = 0.0001), but the differences for the α band did not reach significant values (F2,<sup>29</sup> = 2.452, p = 0.1144). Pairwise comparisons (**Figure 4B**, occipital cortex) showed that: (1), the intermediate power values for the δ and β bands during IS reached statistically significant differences compared to both SWS and REM sleep; and (2), the θ band power was significantly higher during REM sleep than during IS. Regarding the hippocampal θ, there were significant differences (F2,<sup>29</sup> = 17.418, p = 0.0001) between IS and both SWS and REM sleep (**Figure 4C**, hippocampus). The intermediate value of the hippocampal θ power was significantly different in comparison with the other two states. All these results further support the individuality of the IS with respect to SWS and REM sleep.

Regarding the PGO wave analyses in the 10 cats, isolated and double PGO waves constituted an average of 95 and 97.5% of the total PGO waves in the last seconds of SWS prior to REM

sleep and IS respectively, but were only 67.5% of the PGO waves in the first seconds of REM sleep. The remaining percentage in each case corresponded to PGO waves firing in clusters (≥3 PGO waves), something which occurred only occasionally during SWS and IS. The isolated PGO waves remained unchanged during SWS (60.5 ± 7.6%) and IS (62.4 ± 7.9%), but were significantly less frequent in REM sleep (F2,<sup>29</sup> = 3.585, p = 0.04). However, the proportions of double PGO waves did not show significant differences (F2,<sup>29</sup> = 0.264, p = 0.7709) between SWS (34.5 ± 8.4%), IS (35.1 ± 7.6%), and REM sleep (26.4 ± 6.1%). Finally, although the EMG signal was not quantified, examining the polygraphic recordings of the animals indicated that the complete loss of muscle tone occurred either at the end of the IS or at the beginning of REM sleep.

### Carbachol Experiments

Histological analyses indicated that seven of the 10 cats used in this study received microinjections situated in the PLCα; the remaining three cats had received the injections outside the target region, so they were excluded from the carbachol studies. **Figure 5** shows the microphotograph illustrating the location of the microinjections in one animal, and the microinjection sites of the seven animals in coronal drawings using the cat brainstem from the Reinoso-Suarez (1961) atlas. Note that in all cases the microinjections in the rostrodorsal pontine tegmentum were located at the level of the PLCα.

After microinjecting small volumes of the long lasting cholinergic agonist carbachol in the PLCα, all the cats showed, at short latency, two states that did not polygraphically fit the physiological patterns of the cat SWC (**Table 1**). These two states were: (1) Wa, a state of wakefulness without muscle tone (that is, with muscle atonia), and (2) the state that we have called SPGO, in which muscle atonia and PGO activity were associated with EEG synchronization. There were dose-response effects in terms of the latency to onset and time spent in these states; the high carbachol dose (0.1 M) promoted a shorter latency and greater proportion of Wa, while the low dose (0.01 M) promoted the SPGO state (**Table 1**). Also, the administration of either carbachol dose in the PLCα practically abolished the

physiological stage of SWS and caused wakefulness with muscle tone to appear with long latencies and in low proportions (**Table 1**). Concerning REM sleep, except in one animal in which

cortex, and (C) hippocampus (n = 10 cats). <sup>∗</sup>Statistically significant differences between IS and SWS. #Statistically significant differences between IS and

REM sleep. Post hoc analyses (Fisher's test, p = 0.05).

PLCα, perilocus coeruleus alpha; PaG, periaqueductal gray; vRPO, ventral part of the oral pontine reticular nucleus; SLC, locus subcoeruleus; SO, superior olivary nucleus; Tp, tegmental pontine nucleus; Tr, trapezoid body; TV, ventral tegmental nucleus.

### a REM sleep episode was observed with a latency of 18 min, REM appeared after a long latency, more than an hour after carbachol delivery (**Table 1**).

TABLE 1 | Latency to onset and time spent (mean ± standard error) in the different sleep-wakefulness states observed in the first 2 h after unilateral carbachol miroinjections in the rostrodorsal pontine tegmentum at the level of the PLCα (n = 7 cats).


SPGO, EEG synchronization with pontogeniculooccipital activity; REM, REM sleep; Wa, wakefulness with muscle atonia; Wt, wakefulness with muscle tone.

# Matching Carbachol-Induced States With the Physiological Stages of the Sleep-Wakefulness Cycle

### Comparisons With SWS and REM Sleep

The Wa state is comparable to the state of cataplexy (Reinoso-Suárez et al., 1994; Moreno-Balandrán et al., 2008; Torterolo et al., 2015, 2016) in which the electroencephalographic characteristics are common to those of the W stage, but there is a loss of muscle tone. However, the SPGO state is more complex since it constitutes a behavioral sleep state with a dissociated polygraphic pattern in which characteristic manifestations of SWS (EEG synchronization) and REM sleep (θ rhythm in the hippocampus, PGO waves in the LGN and muscle tone absence) coexist (**Figure 6**). Therefore, the three 1-min samples of the SPGO state taken sequentially from the first hour-anda-half of experimental recordings were compared with the last minute of SWS with PGOs (which included the few seconds of IS intervals) prior to REM sleep, and with the first minute of REM sleep. One-way ANOVAs for repeated measures for the normalized power for the different frequency bands of the cortical EEG and for the hippocampal θ were carried out comparing the values of the SPGO state with those of

electrooculogram; EMG, electromyogram; LGN, lateral geniculate nucleus; HPC, hippocampus; FC-R, frontal cortex-reference; OC-R, occipital cortex-reference.

OC-δ, isolated delta band (1.5–3.5 Hz) from the occipital cortex.

for the δ band in frontal and occipital cortices with respect to SWS; (2), higher power values for α activity in both cortices than in SWS and REM sleep; and (3), intermediate values for the hippocampal θ rhythm, higher than in SWS but lower than in REM sleep. Therefore, these results demonstrate the lack of an analogy between this state and SWS and REM sleep, as well as suggesting that the SPGO state could be equivalent to the IS.

### Comparisons With IS

One-way ANOVAs for repeated measures comparing the values of the SPGO state and the IS for the frontal cortex indicated that there were significant differences for the δ (F3,<sup>27</sup> = 7.765, p = 0.0016) and θ bands (F3,<sup>27</sup> = 3.867, p = 0.0269), but not for the α (F3,<sup>27</sup> = 0.711, p = 0.5579) and β bands (F3,<sup>27</sup> = 1.211, p = 0.3343). Post hoc comparisons (**Figure 8A**, frontal cortex) showed that the power of the δ band was significantly lower in the SPGO state than in IS, and that the power of the θ band was significantly higher during the SPGO state than in IS. In relation to the occipital cortex, as well as the frontal cortex, there were significant differences between the SPGO state and the IS for the δ (F3,<sup>27</sup> = 8.391, p = 0.0011) and θ bands (F3,<sup>27</sup> = 10.777, p = 0.0003), but there were no significant differences with the α (F3,<sup>27</sup> = 0.83, p = 0.4947) and β bands (F3,<sup>27</sup> = 0.901, p = 0.4601). Post hoc comparisons (**Figure 8B**, occipital cortex) indicated that δ band power was significantly lower in the SPGO state than in the IS, whilst θ band power was significantly higher during the SPGO state than in the IS, again like occurring in the frontal cortex. Finally, comparing the hippocampal θ band powers between the two states (**Figure 8C**) indicated that there were no significant differences between them. Since differences between the three samples of the SPGO state were not found in any case (see **Supplementary Table S1**), **Figure 8** groups the values of the different SPGO samples for the purpose of comparing this state with the IS, as was done in all above comparisons.

In relation with PGO waves, the proportions of isolated PGOs in the seven animals were similar in the three SPGO state samples (mean percentages: 58.6 ± 5.4, 60.7 ± 4.1, and 59.9 ± 5.8%) and

FIGURE 7 | Comparisons of normalized power values for the different EEG frequency bands between the SPGO state, SWS and REM sleep. Each bar represents the average percentage value and the standard error of each EEG band relative to the total power (1.5–20 Hz) from 1-min epochs in the different states. (A) Frontal cortex, (B) occipital cortex, and (C) hippocampus (n = 7 cats). Values for the SPGO state are the mean ± standard error from the three successive 1-min epochs (20 min apart) obtained from each animal after 20–30 nl 0.01 M carbachol in the PLCα. <sup>∗</sup>Statistically significant differences between the SPGO state and SWS. #Statistically significant differences between SPGO state and REM sleep. Post hoc analyses (Fisher's test, p = 0.05).

IS (59.1 ± 9.7%) without significant differences between them (F3,<sup>27</sup> = 0.022, p = 0.9953). The proportions of double PGO waves were also unchanged (F3,<sup>27</sup> = 0.39, p = 0.7617) in the three SPGO samples (mean percentages: 34.7 ± 4.6, 28.8 ± 4.6, and 33.6 ± 5.6%) as well as being similar to IS (37.3 ± 9.3%). Thus, PGO data gives further indications of strong similarities between the SPGO state induced by cholinergic stimulation of the PLCα region and physiological IS.

# DISCUSSION

Currently there is great interest in studying how the transition to REM sleep occurs, specifically an interest in characterizing the IS and understanding its neural mechanisms, since the key to some pathologies like narcolepsy may lie in this transition (Sorensen et al., 2013). The cat is a good experimental model for studying the neurobiological bases of the transition to REM sleep since a large amount of our knowledge about the mechanisms of the SWC has been accumulated in this species and, additionally, the electrographic manifestations of the sleep phases in the cat, like PGO activity a landmark of REM sleep but also of the IS as reported in the present work- is both easier to record and better characterized in cats than in rodents. As far as we know, only one work has described the characteristics of IS in the cat (Gottesmann et al., 1984), and those characteristics, especially in regards to duration, are quite different from the IS characteristics described in recent studies in the rat (Sánchez-López et al., 2018). To reevaluate IS in cats we have first tried to identify its location and duration based on two widely accepted criteria from the literature associated to REM sleep onset: EEG desynchronization; and the decrease of δ band amplitude in the occipital cortex, where this activity is particularly patent during the deep SWS with PGOs that precedes REM sleep in the cat (Ursin and Sterman, 1981). These criteria make it easy to detect, as shown in **Figure 1**, the existence of a short yet very dynamic period, although with slight differences in each animal, that lies between SWS and REM sleep. This period is not abrupt, but has a progressive presentation as indicated by the slope reflecting the amplitude decrease in the occipital cortex δ band. Accordingly, the power spectra for this period (**Figures 2**, **3**) had values that were intermediate between those of the preceding and following intervals with the same duration. Therefore, these observations would ensure that the chosen period was a transition state corresponding to the IS, since its cortical and hippocampal EEG was qualitatively and quantitatively different from that of both the preceding SWS and the subsequent REM sleep.

According to the above criteria, our results show that, in contrast to previous work done in cats in relation with this topic (Gottesmann et al., 1984), the IS in the cat has an average duration of 14.6 ± 1.3 s. The analyses of the relative powers of the different cortical EEG bands indicate that the IS in cats is characterized by: (1), a significant decrease in amplitude of the δ band in the cortices with respect to the previous SWS, to a value that is even lower in the succeeding REM sleep; (2), a significant increase in amplitude of the α band in the frontal cortex compared with the

previous SWS, which would correspond to the large amplitude sleep spindles previously described during the IS (Gottesmann et al., 1984), an activity that decreases once REM sleep has started; (3), a significant increase in hippocampal θ in relation with the previous SWS, although the hippocampal θ does not become as powerful and rhythmic during IS as during REM sleep. In contrast, PGOs during IS did not differ significantly from those in the preceding SWS; isolated and double PGO waves occurred in both cases and in the same proportions. PGOs are considered to be the physiological signals that trigger REM sleep (Callaway et al., 1987). Our results support this point of view, and they add the probability that PGO activity is also necessary to trigger IS since PGOs are already present with the full EEG characteristics of SWS before IS appears. Concerning muscle tone, the IS presented a decrease of muscle tone with respect to the previous SWS, and that tone was completely lost either by the end of the IS or at the beginning of REM sleep. Finally, it is particularly noteworthy that our results indicate that, as in rats (Sánchez-López et al., 2018), the transition from SWS to REM sleep is not an abrupt phenomenon as occurs with the transition from SWS to wakefulness when the reticular formation is stimulated (Moruzzi and Magoun, 1949). Entrance into REM sleep occurs gradually over an intermediate stage, the IS. This implies the need to reconsider the transition models from SWS to REM sleep with flip-flop characteristics, that is, with sharp transitions from one state to another (Lu et al., 2006).

In relation with the carbachol experiments, as our results show, the cholinergic stimulation of the PLCα nucleus produced complete disturbance of the SWC, totally abolishing some stages of the cycle such as SWS, while considerably increasing the latency of others, like REM sleep. Instead of those, two new states appeared: W with atonia (Wa), mainly after high carbachol doses, and a state called SPGO, which was more frequent after low doses. During Wa the cats had increased respiratory and heart rates and they also had intermittent recoveries of muscle tone, which confirmed that the animals were paralyzed but awake (Reinoso-Suárez et al., 1994; Moreno-Balandrán et al., 2008; Torterolo et al., 2015). Therefore, during Wa the animals showed a clear analogy with the state of cataplexy. In contrast, during SPGO the cats were behaviorally asleep but showed a dissociated polygraphic pattern (Baghdoyan et al., 1982) that did not present a complete analogy with any of the physiological stages of SWS or REM sleep.

Both carbachol-induced states presented a common characteristic, the absence of muscle tone typical of REM sleep, confirming that the PLCα is a fundamental region involved in the triggering of partial signs of REM sleep such as muscle atonia. This observation concurs with results from experiments with bilateral PLCα lesions in cats, in which REM sleep without muscle atonia was observed (Jouvet, 1979; Sastre and Jouvet, 1979). Similarly, SLD nucleus lesions in the rat produced REM sleep with muscle tone (Lu et al., 2006). These conclusions would support the role of the PLCα/SLD region for the triggering of muscle atonia during REM sleep through its descending projections to the inhibitory interneurons of the spinal cord that would inhibit motoneurons. The PLCα/SLD projections would have an intermediate station in the ventromedial medulla (Sakai, 1985) or reach the spinal cord interneurons directly (Lu et al., 2006). Another feature of REM sleep observed after cholinergic stimulation of the PLCα nucleus is the presence of PGO waves. It is well-known that the rostrodorsal pontine tegmentum contains structures that mediate PGO activity after carbachol

administration (Datta et al., 1993, 1998). Our work demonstrates that the type of PGO waves that appear when the PLCα nucleus is cholinergically stimulated are isolated or double waves, just like the PGO patterns that precede REM sleep episodes and that are present in both the SWS and IS.

In our experiments, cholinergic stimulation of PLCα always produced muscle atonia, but this was not the case for PGO activity. After administering the high dose of carbachol we did not immediately observe PGO waves, contrary to the PGO activity observed in response to the low dose. The SLD nucleus in the rat holds several neuronal groups with differential responses to carbachol administration and some show a low bursting threshold similar to the firing of other pontine neurons known to be involved in PGO wave generation (Brown et al., 2006). The use of M2 acetylcholine muscarinic receptor antagonists blocks carbachol-induced PGOs (Datta et al., 1993). On the other hand, in cats, the inhibition of the M3 type acetylcholine muscarinic receptors in the PLCα nucleus had, as a consequence, the absence of muscle atonia during REM sleep (Sakai and Onoe, 1997). M2 receptor carbachol affinity is almost five times higher than that of the M3 receptor (Vanderheyden et al., 1990). Consequently, the absence of PGO activity after administering the high carbachol dose is probably mediated by some M2 receptor desensitization mechanism, such as receptor internalization, in order to maintain homeostasis in the neurons generating that activity. This would explain why PGO waves are not readily observed after the administration of the high carbachol dose, since they would not appear until sufficient time had elapsed for the extracellular concentration of the cholinergic agonist to be reduced.

Some authors have proposed that the PLCα is the responsible region for triggering complete REM sleep (Sakai, 1988; Vanni-Mercier et al., 1989; Boissard et al., 2002; Lu et al., 2006). However, according to our results, cholinergic stimulation of the PLCα nucleus does not produce complete REM sleep, only the appearance of some signs of REM sleep mixed with characteristics of other stages of the SWC. This agrees with previous works by our group, which found that the vRPO region was the only region in which the administration of low-volume microinjections with a wide range of carbachol doses produced REM sleep with all its manifestations (Reinoso-Suárez et al., 1994; Garzón et al., 1997, 1998; Moreno-Balandrán et al., 2008).

Returning to the SPGO state, it is not truly comparable to either SWS or REM sleep. However, it has a much better match with IS, since both states show, in comparison to SWS, increased α band amplitude in the frontal cortex, decreased δ band amplitude in both cortices, increased θ rhythm amplitude in the hippocampus, and decreased muscle tone, as well as presenting similar proportions of single and double PGO waves. The only significant differences observed between the SPGO state and IS were the lower δ band and higher θ band power in both cortices during the SPGO state. The decrease in δ power is one of the main EEG characteristics of IS in rats (Sánchez-López et al., 2018) and in cats as reported here. We think that the more pronounced δ decrease and its associated increase in θ power after carbachol in the PLCα could be explained by an excessive carbachol effect, one that would probably disappear at lower carbachol doses. Therefore, we believe that these differences do not suppose the rejection of an analogy between the IS and SPGO states.

Since stimulation of the PLCα nucleus with low carbachol doses induces SPGO, which constitutes a good expression of physiological IS, it seems that the PLCα region could be the organizing structure behind the transition from SWS (or NREM) sleep to REM sleep, therefore generating IS. It is striking that, in both the cat and the rat, α band voltage rises during IS. This could correlate with the fact that the entrance into REM sleep in humans does not occur from the N3 phase, an NREM sleep stage with predominance of δ waves, but always previously passes through the N2 phase that has predominant spindle α activity (Achermann and Borbély, 2011). Therefore, it seems that for REM sleep to occur, hyperpolarization levels in thalamo-cortical cells must vary beforehand since these cells are responsible for the EEG bioelectric manifestations during NREM sleep (De Andrés et al., 2011). Depending on the degree of hyperpolarization in thalamo-cortical neurons, typical SWS δ waves or sleep spindle α activity will be observed on the EEG (Nuñez et al., 1992). Accordingly, the electroencephalographic changes observed during IS, such as the voltage increase in the α band, will be mediated by the firing of reticular thalamic neurons transferred to the cortex through thalamic projection cells (Steriade et al., 1985, 1991). This effect could be carried out by neurons from the PLCα region, which has multiple direct connections with the various nuclei of the thalamus, including the reticular thalamic nucleus (Datta et al., 1998). Moreover, the hippocampal θ rhythm that characterizes REM sleep and that is enhanced during IS and SPGO compared to levels during SWS, as shown in the present work, is generated by GABAergic and cholinergic projections reaching the hippocampus from θ pacemaker neurons in the medial septum (Petsche et al., 1962; Gerashchenko et al., 2001; Fuller et al., 2007). However, brainstem structures are also involved in triggering this event, such as the excitatory projections reaching the medial septum from the precoeruleus region (Fuller et al., 2007) as well as from the nucleus incertus (Nuñez et al., 2006), which constitutes a relay station between the oral pontine reticular nucleus and the medial septum for hippocampal θ rhythm generation (Vertes, 1980; Nuñez et al., 1991). Therefore, it is likely that PLCα activation is also related to the increase of θ rhythmicity observed in the hippocampus during both IS and the carbacholinduced SPGO state. Finally, it is important to remember the reciprocal connections between the locus coeruleus complex and the vRPO region, which is responsible for the complete generation of REM sleep. The vRPO has strong connections with the thalamic nuclei, including reciprocal connections with the reticular nucleus (Reinoso-Suárez et al., 1994; Rodrigo-Angulo et al., 2008); thus the high α activity observed during IS could be blocked by the vRPO region thus giving way to the complete EEG desynchronization observed during REM sleep.

### CONCLUSION

The present results strongly indicate that the IS is an independent sleep stage, and confirm that the PLCα region, rather than a REM sleep triggering region, is a region that generates some partial signs of REM sleep, such as muscle atonia or PGO waves. Also, we present results that allow us to propose the PLCα nucleus as the structure organizing IS, which occurs in the transition from SWS to REM sleep.

### DATA AVAILABILITY

fnins-13-00748 July 19, 2019 Time: 15:31 # 12

All datasets generated for this study are included in the manuscript and/or the **Supplementary Files**.

### ETHICS STATEMENT

All experiments were carried out in accordance with the European Community Council Directive (2010/63/UE) and approved by the Institutional Animal Care and Use Committee of the Universidad Autónoma of Madrid (Spain) and the competent regional government agency (PROEX 004/15).

### AUTHOR CONTRIBUTIONS

CC-C scored and analyzed the polygraphic recordings, performed statistical analyses, and wrote the manuscript. MG performed the experiments and wrote the manuscript. IdA

### REFERENCES


conceived and designed the experiments, supervised all aspects of the study, and wrote the manuscript.

### FUNDING

Supported by the Spanish Ministerio de Economía y Competitividad (BFU2013-43741P) and Autonomous University of Madrid (UAM/92).

### ACKNOWLEDGMENTS

We would like to acknowledge Ms. Marta Callejo for technical assistance and Ms. Carol Fox Warren for revising English language use.

We want to express our appreciation of Dr. Fernando Reinoso-Suárez, who founded our Department, introduced us to sleep studies, and has recently passed away.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2019.00748/full#supplementary-material



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Carrera-Cañas, Garzón and de Andrés. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Differential Role of Pontomedullary Glutamatergic Neuronal Populations in Sleep-Wake Control

Evelyn T. M. Erickson, Loris L. Ferrari, Heinrich S. Gompf and Christelle Anaclet\*

Department of Neurobiology, University of Massachusetts Medical School, Worcester, MA, United States

Parafacial zone (PZ) GABAergic neurons play a major role in slow-wave-sleep (SWS), also called non-rapid eye movement (NREM) sleep. The PZ also contains glutamatergic neurons expressing the vesicular transporter for glutamate, isoform 2 (Vglut2). We hypothesized that PZ Vglut2-expressing (PZVglut2) neurons are also involved in sleep control, playing a synergistic role with PZ GABAergic neurons. To test this hypothesis, we specifically activated PZVglut2 neurons using the excitatory chemogenetic receptor hM3Dq. Anatomical inspection of the injection sites revealed hM3Dq transfection in PZ, parabrachial nucleus (PB), sublaterodorsal nucleus (SLD) or various combinations of these three brain areas. Consistent with the known wake- and REM sleep-promoting role of PB and SLD, respectively, chemogenetic activation of PBVglut2 or SLDVglut2 resulted in wake or REM sleep enhancement. Chemogenetic activation of PZVglut2 neurons did not affect sleep-wake phenotype during the mouse active period but increased wakefulness and REM sleep, similar to PBVglut2 and SLDVglut2 activation, during the rest period. To definitively confirm the role of PZVglut2 neurons, we used a specific marker for PZVglut2 neurons, Phox2B. Chemogenetic activation of PZPhox2B neurons did not affect sleep-wake phenotype, indicating that PZ glutamatergic neurons are not sufficient to affect sleep-wake cycle. These results indicate that PZ glutamatergic neurons are not involved in sleep-wake control.

Keywords: neuronal circuitry, DREADDs, brainstem, parafacial zone, sleep-wake control, sublaterodorsal nucleus, parabrachial nucleus

### INTRODUCTION

Over the past few years, the medullary parafacial zone (PZ) has been identified as a strong sleeppromoting brain area (Anaclet and Fuller, 2017). Both disruption of PZ GABAergic transmission (Anaclet et al., 2012) and chemogenetic inhibition of PZ GABAergic (PZGABA) neurons (Anaclet et al., 2014) result in insomnia. More importantly, chemogenetic activation of PZGABA neurons strongly increases SWS amount and consolidation and enhances cortical EEG slow-wave activity (SWA), a marker of SWS depth (Anaclet et al., 2014). Finally, chemogenetic activation of PZGABA neurons counteracts the wake-promoting action of psychostimulants (Anaclet et al., 2018). The PZ is generally located dorsal and lateral from the facial nerve but its exact boundaries are not precisely defined. A recent study has shown that, in mouse, the parvicellular reticular nucleus part alpha (PCRtA), ventral from PZ, does not contain sleep-active neurons (Sakai, 2017), indicating that the PZ sleep promoting neuronal population does not include the PCRtA.

Edited by:

Takeshi Sakurai, University of Tsukuba, Japan

### Reviewed by:

Radhika Basheer, Harvard Medical School, United States Hiromasa Funato, Toho University, Japan

\*Correspondence: Christelle Anaclet Christelle.Anaclet@umassmed.edu

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

> Received: 30 March 2019 Accepted: 08 July 2019 Published: 30 July 2019

### Citation:

Erickson ETM, Ferrari LL, Gompf HS and Anaclet C (2019) Differential Role of Pontomedullary Glutamatergic Neuronal Populations in Sleep-Wake Control. Front. Neurosci. 13:755. doi: 10.3389/fnins.2019.00755

In rats, about 40% of PZ neurons are sleep-active (Alam et al., 2018) and cell body specific PZ lesions result in insomnia (Anaclet et al., 2012). More specifically, PZGABA are involved in slow-wave-sleep (SWS) control. However, cFos expression, a marker of neuronal activity, showed that about half of sleep-active neurons are GABAergic (Anaclet et al., 2012), indicating that in the PZ, non-GABAergic neurons are also involved in sleep control. Within the PZ, the only other known neuronal population is glutamatergic, expressing the vesicular glutamate transporter isoform 2 (Vglut2; in situ hybridization data are available in Allen Mouse Brain Atlas, Allen Institute for Brain Science<sup>1</sup> ). We hypothesized that PZ glutamatergic neurons (PZVglut2) are also involved in sleep control and act synergistically with sleep-active PZGABA neurons to promote SWS. To start testing this hypothesis, we chemogenetically activated PZVglut2 neurons. Specific targeting of PZ was challenging and sleep phenotypes were difficult to interpret due to the possible transfection of the neighboring parabrachial (PB) and sublaterodorsal (SLD) nuclei that are involved in wakefulness and rapid eye movement (REM) sleep, respectively (Clement et al., 2011; Fuller et al., 2011). To get around these obstacles and specifically test the involvement of PZ glutamatergic neurons in sleep-wake control, we used Phox2B, a transcription factor expressed in PZ but not PB or SLD glutamatergic neurons. Data indicate that PZ glutamatergic neurons are not involved in sleepwake control. Additionally, we found that excitation of PB or SLD glutamatergic neurons promotes wakefulness or REM sleep, respectively, results that are complimentary to the reduction in wakefulness or REM sleep previously observed following lesion of PB or SLD, respectively (Clement et al., 2011; Fuller et al., 2011).

### MATERIALS AND METHODS

### Animals

In order to visualize Vglut2-, Phox2B- and Vgat-expressing neurons, Vglut2-IRES-cre [Jackson Laboratory #016963 (Vong et al., 2011)], Phox2B-IRES-cre [Jackson Laboratory #016223 (Rossi et al., 2011)] and Vgat-IRES-cre [Jackson Laboratory #016962 (Vong et al., 2011)] mice were crossed with a credependent reporter mouse Flox-L10-GFP [Jackson Laboratory #24750 (Liu et al., 2014)], producing Vglut2-GFP, Phox2B-GFP and Vgat-GFP mouse lines. Thirty two adult male Vglut2-GFP mice, seven adult male Phox2B-GFP and one adult male Vgat-GFP (8–12 weeks, 20–25 g) mice were used in this study. Mice were bred at our animal facility and underwent genotyping both before and after experiments. All procedures were approved by the Institutional Animal Care and Use Committee of Beth Israel Deaconess Medical Center and of University of Massachusetts Medical School.

### Surgery

Naïve mice were anesthetized with ketamine/xylazine [100 and 10 mg/kg, respectively, intraperitoneal (IP)] and then placed in a stereotaxic apparatus. To selectively express the hM3Dq receptors in glutamatergic (Vglut2+) or Phox2B-expressing neurons of the PZ, we performed bilateral injections of an adenoassociated viral (AAV; serotype 10) vector expressing the hM3Dq receptor in a cre-dependent configuration [hSyn-DIO-hM3DqmCherry-AAV; (Anaclet et al., 2014)] into the PZ [coordinates from Bregma: Antero-posterior = −5.6 mm, Lateral = ± 1.0 mm, Dorso-ventral = −4.2 mm, as per the mouse atlas of Paxinos and Watson (Paxinos and Franklin, 2001)] of Vglut2-IRES-cre (PZVglut2−hM3Dq) mice, Phox2B-IRES-cre (PZPhox2B−hM3Dq) mice or non-cre expressing littermate control mice. Injections of the viral vector (60 nl) into the PZ of these mice were performed using a compressed air delivery system as previously described (Anaclet et al., 2010). After injections, mice were implanted with four EEG screw electrodes (Pinnacle Technology Inc., Catalog #8403) and two flexible electromyogram (EMG) wire electrodes (Plastics One, catalog #E363/76/SPC), previously soldered to a 6-pin connector (Heilind Electronics, catalog #853-43-006-10- 001000) and the assembly was secured with dental cement. The scalp wound was closed with surgical sutures and the mouse was kept in a warm environment until resuming normal activity as previously described (Anaclet et al., 2015).

### Sleep-Wake Monitoring

Three weeks after surgery, the mice were housed individually in transparent barrels in an insulated sound-proofed recording chamber maintained at an ambient temperature of 22 ± 1 ◦C and on a 12 h light/dark cycle (lights-on at 7 A.M., Zeitgeber time: ZT0) with food and water available ad libitum. Mice were habituated to the recording cable for 5–7 days before starting polygraphic recording. Cortical EEG (ipsilateral frontoparietal leads) and EMG signals were amplified (A-M System 3500, United States) and digitalized with a resolution of 500 Hz using Vital Recorder (Kissei, Japan). Mice were recorded for a 24 h baseline period followed by IP injections of saline (control injection) or Clozapine-N-oxide (CNO, NIMH Chemical Synthesis and Drug Supply Program; 0.3 mg/kg in saline). Injections were performed at 10 A.M. (10:00, ZT3, light period, time of high sleep-drive) and 7 P.M. (19:00, ZT12, beginning of the dark period, time of high wake-drive), in a randomized cross-over design, with each injection separated by a 2–3 day washout period. In each experiment, recordings were simultaneously made from an equal number (batches of 2–4) of PZVglut2−hM3Dq and PZVglut2−wt mice.

### Sleep Scoring and Analysis

Using SleepSign for Animal (Kissei, Japan) assisted by spectral analysis using fast Fourier transform, polygraphic records were visually scored in 10 s epochs for wakefulness (W), SWS, and REM sleep. The percentage of time spent in wake, SWS and REM sleep were summarized for each group and each condition. The SWS to REM sleep latency is defined as the time between the onset of the first SWS episode, lasting >20 s, after injection and the onset of the first REM sleep episode, lasting >10 s.

Sleep-wake fragmentation was assessed by analyzing the distribution of each vigilance stage in different bout lengths. Vigilance stages were separated into eight bout lengths (<30, 40–70, 80–150, 160–310, 320–630, 640–1270, 1280–2550, and

<sup>1</sup>http://mouse.brain-map.org/experiment/show/73818754

>2550 s) (Mochizuki et al., 2004; Kantor et al., 2013). For each vigilance stage, the number of episodes and the percentage of the vigilance stages occurring in each bout length were used to produce a time-weighted frequency histogram.

Recordings were scored again in 5 s epochs to allow for performance of an EEG power spectrum analysis. On the basis of visual and spectral analysis, epochs containing artifacts occurring during active wake (with large movements) or containing two vigilance states were visually identified and omitted from the spectral analysis. Recordings containing wake artifacts during more than 20% of the time were removed from the spectral analysis. EEG power spectra were computed for consecutive 5 s epochs within the frequency range of 0.5–120 Hz using a fast Fourier transform routine (FFT). The data were collapsed into 0.5 Hz bins. To determine the effect of injection on sleepwake power spectra, EEG power spectra were analyzed during the 3 h period of time post-injection, starting 10 min after injection as a previous study had shown that CNO injection significantly affected SWS amount during 3 h post-injection and SWS latency was no more than 10 min (Anaclet et al., 2014). The data were standardized by expressing each frequency bin as a percentage relative to the same bin under baseline conditions from the same mouse and from the same time of the day (same Zeitgeber time). To analyze the EEG frequency bands, power bins were summed in δ 0.5–5 Hz, θ 5–9 Hz, α 9–15 Hz, β 15–30 Hz, low γ 30–60 Hz and high γ 60–120 Hz, and expressed in percentage of baseline power band, from the same circadian time.

Statistical analysis was performed using Prism v6 (GraphPad Software, San Diego, CA, United States). Following confirmation that the data met the assumptions of the ANOVA model, twoway ANOVA followed by a post hoc Bonferroni test were used to compare the effects of the drug injection and time period on sleep-wake parameters, the effect of the drug injection and the distribution of vigilance episodes, or the effect of drug injection and power band on cortical EEG power density. Paired Student's t-test was used to compare the effects of the drug injection on SWS to REM sleep latency. Sample size and power calculations were performed post hoc at http://www. biomath.info, using means and standard deviations derived from our analysis. The present study was sufficiently powered to detect effect sizes.

### Immunostaining and RNAscope

At the end of the behavioral experiments, mice were deeply anesthetized with ketamine/xylazine (200 and 20 mg/kg, respectively) and perfused transcardially with 20 ml of saline, followed by 100 ml of neutral phosphate-buffered formalin (4%; Thermo Fisher Scientific). Brains were removed from the skull and incubated in neutral phosphate-buffered formalin (4%; Thermo Fisher Scientific) for 2 h, followed by 20% sucrose until they sank.

For immunostaining, using a freezing microtome, brains were sectioned at 40 µm into 3 series. One series was used to label mCherry to visualize neurons transfected by hSyn-DIO-hM3DqmCherry-AAV. Floating brain sections were incubated overnight with the primary antiserum (1:10,000; rabbit polyclonal antibody against mCherry was raised against DsRed, catalog #632496, Clontech). The next day, sections were incubated in goat antirabbit biotinylated secondary antiserum (1:1,000; catalog # BA-1000, Vector Laboratories), followed by incubation in ABC reagents (1:1000; Vector Laboratories) for 90 min. Visualization reaction was in a 0.06% solution of 3,3-diaminobenzidine tetrahydrochloride (Sigma-Aldrich) in PBS plus 0.02% H2O<sup>2</sup> for 2–15 min. Finally, the sections were mounted on slides, dehydrated, cleared, and coverslipped. To map the extent of hSyn-DIO-hM3Dq-mCherry-AAV transfection, immunostained neurons were visualized with a brightfield microscope (Keyence BZ-X710, Japan) and mapped (**Figures 1B**, **3A**, **4A**, **7A**).

For RNAscope, using a cryostat (Thermo Scientific, Cryostar NX70), brains were sectioned at 10 µm and mounted onto Surgipath (Leica) adhesive microscope slides, 3 slices per slide. Slides were kept at −80◦C until shortly before in situ hybridization. Slides were first warmed to room temperature and then we performed the RNAscope hybridization using a RNAscope Multiplex Fluorescent Reagent Kit (Advanced Cell Diagnostics, Inc., Newark, CA, United States). Briefly, according to the manufacturer's instructions, target retrieval was performed at 99◦C after which slices were dehydrated in 100% ethanol and air-dried. Next, sections were treated with protease inhibitor (Protease III, RNAscope) for 30 min at 40◦C. After rinsing in RNAscope wash buffer, we incubated the sections in the RNAscope probes for Vglut2 (Mm-Slc17a6, catalog # 319171) and Phox2B (Mm-Phox2b-C2, catalog # 407861-C2). Additional sections were incubated in the manufacturer-supplied 2-plex positive control (catalog # 320761) and negative control (catalog # 320751) probes. Following the first 3 signal amplification steps, the fourth amplification was performed using Amp 4 Alt C-FL, such that channel 1 (Vglut2) was fluorescently labeled with Alto 550 and channel 2 (Phox2B) was labeled with Alto 647. Fluorescent images were collected with a confocal microscope (Zeiss LSM 700; **Figures 6D–P**).

**Figures 6A–C**, Phox2B-GFP Native GFP fluorescence images were collected with a fluorescence microscope (Keyence BZ-X710, Japan).

# Whole-Cell in vitro Experiments

For in vitro electrophysiological recordings, 10–13 days old Vglut2-ires-cre (N = 17) and Phox2B-ires-cre (N = 5) mice were injected bilaterally in the PZ, SLD or PB area with hSyn-DIO-hM3Dq-mCherry-AAV (100 nl/side). At about 3 weeks of age, 250 µm thick coronal brain slices of the PZ, SLD or PB area were prepared.

Mice were deeply anesthetized (200 mg/Kg Ketamine, 20 mg/Kg Xylazine) and transcardially perfused with icecold N-methyl-D-glucamine based artificial cerebrospinal fluid (NMDG-ACSF) containing (in mM): NMDG 98, HEPES 20, Glucose 25, NaHCO<sup>3</sup> 30, Na-ascorbate 5, Na-pyruvate 3, Thiourea 2, MgSO<sup>4</sup> 10, NaH2PO<sup>4</sup> 1.24, KCl 2.5, CaCl<sup>2</sup> 0.5; pH adjusted to ≈7.3 with HCl 37%. The brains were quickly extracted from the skull and sliced in carbogeneted (95% O2 5% CO2) ice-cold NMDG-ACSF using a vibrating microtome (7000-SMZ2, Campden Instruments). Slices containing the area of interest were immediately transferred to a chamber with carbogented NMDG-ACSF kept at 35◦C for 8 min, then moved to carbogeneted normal ACSF at room temperature containing

FIGURE 1 | Activation of PZ glutamatergic neurons during the active period (19:00 or ZT12). (A) In vitro confirmation. (A1) PZVglut2−hM3Dq whole-cell recording showing an increase in firing frequency in response to bath application of CNO (0.5 µM). (A2) Average firing frequency (±S.E.M.) during the last 2 min of the CNO (0.5 µM) application as compared with the 2 min period preceding CNO application (control; N = 5 PZVglut2−hM3Dq neurons). <sup>∗</sup>p < 0.05 Paired Student's t-tests. (B) Extent of transduced neurons (mCherry-positive somas) is shown for individual Vglut2-IRES-cre mice that received bilateral injections of hM3Dq-mCherry-AAV into the PZ (PZVglut2−hM3Dq). (C) Hourly amount of wakefulness (C1), SWS (C2) and REM sleep (C3) following CNO (0.3 mg/kg, N = 6 mice) as compared with control injection. (D1–D3) Percentage of sleep-wake states (±S.E.M.) during the 3 h post-injection period (19:00–22:00), the remainder (9 h) of the dark/active period (22:00–07:00) and the subsequent 12 h light period (07:00–19:00; N = 6 mice). (E1–E3) Number of episodes (±S.E.M.) of wakefulness (W), SWS or REM sleep (RS) in each bout length and (E1'–E3') time-weighted frequency histograms showing the proportion (±S.E.M.) of W, SWS or RS amounts in each bout length as a percentage of the total amount of W, SWS or RS during the 3 h post-injection period (19:00–22:00; N = 6). (F1–F3) Sleep-wake power spectrum changes over baseline during the 3 h (19:00–22:00) post CNO (0.3 mg/kg, N = 4 mice) injection as compared with control injection; and the quantitative changes (±S.E.M.) in power for the δ (0.4–5 Hz), θ (5–9 Hz), α (9–15 Hz), β (15–30 Hz), low γ (30–60 Hz) and high γ (60–120 Hz) frequency bands (±S.E.M.) following vehicle or CNO (0.3 mg/kg, N = 4 mice) administrations. (C–F) Control injection in Black, CNO injection in red; <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, two-way ANOVA followed by a post hoc Bonferroni test.

(in mM): NaCl 126, NaHCO<sup>3</sup> 26, Glucose 10, Na-ascorbate 1, Thiourea 2, Na-Pyruvate 3, NaH2PO<sup>4</sup> 1.24, KCl 2.5, CaCl<sup>2</sup> 2, MgCl<sup>2</sup> 1.3.

Recordings were guided using a combination of fluorescence and infrared differential interference contrast (IR-DIC) video microscopy using a fixed stage upright microscope (Axio Examiner.D1, Zeiss) equipped with a Nomarski immersion lens (40×/1.0) and an infrared-sensitive camera (Orca flash 4.0, Hamamatsu). Images were displayed in real time using Zen2 software (Carl Zeiss). Recordings were conducted in wholecell mode using an EPC-10 USB amplifier and Patchmaster software (Heka).

Recordings were performed in current clamp mode using a K-gluconate based pipette solution containing (in mM): Kgluconate 120, KCl 10, MgCl<sup>2</sup> 3, HEPES 10, K-ATP, Na-GTP 0.5. After at least 10 min of stable recording, ACSF containing CNO (500 nM) was perfused into the chamber for 3–5 min before washout. Recordings were analyzed, using Patchmaster software, by comparing the last 2 min before the application of CNO to the last 2 min of the CNO application. Paired Student's t-tests were used to calculate statistical significance.

### RESULTS

# PZVglut2 Neurons Are Not Sleep-Promoting

To test whether activation of PZVglut2 neurons affects sleepwake phenotype, Vglut2-IRES-Cre mice were injected into the PZ with a virus vector containing the excitatory hM3Dq receptor (AAV-mCherry-hM3Dq) to specifically express hM3Dq receptors in PZ glutamatergic neurons (PZVglut2−hM3Dq mice). First, responses of PZVglut2−hM3Dq neurons to the hM3Dq ligand, clozapine-N-oxide (CNO), were tested using whole-cell in vitro recordings (**Figure 1A1**). Bath application of CNO (500 nM) significantly increased firing rates in PZVglut2−hM3Dq neurons (1.2 ± 0.5 vs. 0.3 ± 0.2 Hz in control condition, p = 0.042; **Figure 1A2**), confirming that CNO activates PZ glutamatergic neurons. We then tested, in vivo, the sleep-wake phenotypes upon activation of PZ glutamatergic neurons. At the end of the behavioral studies, the injection sites were mapped using mCherry immunostaining. Of the sixteen Vglut2-IRES-Cre mice injected with AAV-mCherry-hM3Dq, six mice displayed more specific expression of mCherry bilaterally in the PZ (**Figure 1B**) and were used for the following sleep-wake analysis.

To test the effect of PZVglut2 neurons in sleep-wake control, mice were injected in a randomized cross-over design with saline or CNO (0.3 mg/kg) at the beginning of the dark/active period (19:00, ZT12) or during the light/rest period (10:00, ZT3). When injected at 19:00, CNO treatment did not significantly affect the hourly amount of wakefulness (two-way ANOVA, F(23,115) = 1.06, p = 0.40; **Figure 1C1**), SWS (two-way ANOVA, F(23,115) = 1.02, p = 0.45; **Figure 1C2**), or REM sleep (two-way ANOVA, F(23,115) = 1.38, p = 0.14; **Figure 1C3**). Because in previous studies (Anaclet et al., 2014, 2015, 2018) the effect of CNO-mediated neuronal activation or inhibition on sleep-wake cycles was most pronounced during the 3 h post-injection period, we performed a more refined analysis of this period. Neither wakefulness nor REM sleep amount and consolidation were affected by CNO injection during the 3 h post CNO injection as compared with control injection (**Figures 1D1,D3,E1,E1',E3,E3'**). However, SWS amounts were significantly decreased during the 3 h post CNO injection period as compared with control injections (15.0 ± 5.0 vs. 33.5 ± 11.9% of SWS in control condition, p < 0.01; **Figure 1D2**) but without change in bout length distribution (**Figure 1E2'**). This SWS decrease is associated with a significant increase in the number of very short SWS bouts (7.0 ± 3.1 vs. 1.8 ± 0.6 bouts lasting between 10 and 30 s, p = 0.0006; **Figure 1E2**). Wakefulness bout duration was also increased with a significant increase of the proportion of wakefulness from long bout lengths (>40 min long bouts: 72.9 ± 10.1 vs. 33.3 ± 21.1% of total wakefulness after control injection, p = 0.003; **Figure 1E1**'). These results indicate more labile switching between the two vigilance stages. Cortical EEG power spectral distribution was affected by the treatment in wakefulness (two-way ANOVA, F(243,729) = 1.48, p < 0.0001; **Figure 1F1**) and SWS (two-way ANOVA, F(243,729) = 3.88, p < 0.0001; **Figure 1F2**) but not in REM sleep (two-way ANOVA, F(243,729) = 0.82, p = 0.97; **Figure 1F3**). However, none of the frequency bands displayed any significant difference between CNO and control injection, in any vigilance stage (**Figures 1F1– F3**). Altogether, activation of PZ glutamatergic neurons at a time when the wake-promoting systems are active, during the active phase, did not induce SWS and showed only minimal effects on sleep-wake phenotype, indicating that PZVglut2 neurons are not sleep-promoting.

In order to test if activation of PZVglut2 neurons affects sleepwake phenotypes differently when the sleep-promoting system is driving sleep, during the light period, CNO was injected at 10:00. CNO treatment significantly affected wakefulness (twoway ANOVA, F(23,115) = 3.14, p < 0.0001; **Figure 2A1**), SWS (two-way ANOVA, F(23,115) = 3.12, p < 0.0001; **Figure 2A2**) and REM sleep (two-way ANOVA, F(23,115) = 2.16, p = 0.004; **Figure 2A3**) hourly distribution. Wakefulness amount was significantly increased during the 4 h post CNO injection period (74.6 ± 7.5 vs. 40.3 ± 3.9% of time in control condition, p < 0.001; **Figure 2B1**). This wakefulness increase was at the expense of SWS (21.9 ± 6.1 vs. 53.5 ± 4.2% of time in control condition, p < 0.001; **Figure 2B2**). The increase in wakefulness was due to bout elongation, after CNO injection, as the mice were spending most of their wake time in bouts longer than 40 min (44.5 ± 11.2 vs. 7.2 ± 7.2% of total wakefulness after control injection, p < 0.001; **Figure 2C1**'), while in control condition, they were spending most of their wake time in 20–40 min long bouts (15.5 ± 8.2 vs. 44.3 ± 4.5% of total wakefulness in control condition, p = 0.01; **Figure 2C1**'). The number of wakefulness episodes, however, remained unchanged (**Figure 2C1**). The decrease of SWS was due to fragmentation characterized by a significant increase in the number of very short SWS bouts (13.0 ± 4.0 vs. 3.2 ± 0.6 episodes 30 s long or shorter in control condition, p < 0.0001; **Figure 2C2**) and a significant decrease of long SWS bouts (0.0 ± 0.0 vs. 27.8 ± 6.1% of total SWS in bouts 10–20 min long in control condition, p < 0.0001; **Figure 2C2**'). Interestingly, REM sleep amount displayed a trend to increase during the second part

frequency histograms showing the proportion (±S.E.M.) of W, SWS or RS amounts in each bout length as a percentage of the total amount of W, SWS or RS during the 4 h post-injection period (10:00–14:00; N = 6). (D1–D3) Sleep-wake power spectrum changes over baseline during the 3 h (10:00–13:00) post CNO (0.3 mg/kg, N = 5 mice) injection time period as compared with control injection; and the quantitative changes (±S.E.M.) in power for the δ (0.4–5 Hz), θ (5–9 Hz), α (9–15 Hz), β (15–30 Hz), low γ (30–60 Hz) and high γ (60–120 Hz) frequency bands (±S.E.M.) following vehicle or CNO (0.3 mg/kg, N = 5 mice) administrations. Control injection in Black, CNO injection in red; <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001, two-way ANOVA followed by a post hoc Bonferroni test.

of the light period, 4 h following CNO injection (14:00–19:00; 10.3 ± 0.2 vs. 8.5 ± 0.7% of time after control injection, p > 0.05; **Figure 2B3**). This was associated with a significant increase in the number of episodes (10–30 s long bouts: 7.0 ± 1.3 vs. 3.3 ± 0.6 bouts after control injection, p < 0.0001; 10–30 s long bouts: 12.5 ± 0.6 vs. 6.8 ± 1.2 bouts after control injection, p < 0.0001; **Figure 2C3**). REM sleep bout length distribution, however, remained unchanged (**Figure 2C3'**). Cortical EEG power spectral distribution was not affected by the treatment in wakefulness (two-way ANOVA, F(243,972) = 0.76, p = 0.996; **Figure 2D1**). In contrast, both SWS (two-way ANOVA, F(243,972) = 2.79, p < 0.0001; **Figure 2D2**) and REM sleep (two-way ANOVA, F(243,972) = 1.43, p = 0.0001; **Figure 2D3**) cortical EEG power spectral distribution was affected by the treatment. Interestingly, during REM sleep, the theta band was significantly increased (128.1 ± 6.3 vs. 106.9 ± 3.2% of baseline theta power in control condition, p < 0.001; **Figure 2D3**).

# Activation of SLDVglut2 Neurons During the Inactive Period Enhances REM Sleep

The excitatory receptor, hM3Dq, was mostly expressed in the SLD in six Vglut2-hM3Dq mice (SLDVglut2−hM3Dq; **Figure 3A**). Whole-cell recording confirmed the expression of functional

FIGURE 3 | Activation of SLD glutamatergic neurons during the inactive period (10:00 or ZT3). (A) Extent of transduced neurons (mCherry-positive somas) is shown for individual Vglut2-IRES-cre mice that received bilateral injections of hM3Dq-mCherry-AAV into the SLD (SLDVglut2−hM3Dq). (B1) SLDVglut2−hM3Dq whole-cell recording showing an increase in firing frequency in response to bath application of CNO (0.5 µM). (B2) Average firing frequency (±S.E.M.) during the last 2 min of the CNO (0.5 µM) application as compared with the 2 min period preceding CNO application (control; N = 6 PBrmVglut2−hM3Dq neurons). <sup>∗</sup>p < 0.05 Paired Student's t-tests. (C) Hourly amount of wakefulness (C1), SWS (C2) and REM sleep (C3) following CNO (0.3 mg/kg, N = 6 mice) as compared with control injection. (D1,D2) Percentage of wakefulness and SWS (±S.E.M.), respectively, during the 3 h post-injection period (10:00–13:00), the following 3 h (13:00–16:00), the remainder (3 h) of the light/sleep period (16:00–19:00), the subsequent 12 h dark period (19:00–07:00) and first 3 h of the light period of the next day (07:00–10:00, N = 6 mice). (D3) Percentage of REM sleep (±S.E.M.) during the 2 h post-injection period (10:00–12:00), the following 4 h (12:00–16:00), the remainder (3 h) of the light/sleep period (16:00–19:00), the subsequent 12 h dark period (19:00–07:00) and first 3 h of the light period of the next day (07:00–10:00, N = 6 mice). (E) SWS to REM sleep latency defined as the time (min) between the onset of the first SWS episode (>20 s) and the onset of the first REM sleep episode (>10 s). (F1–F3) Number of episodes (±S.E.M.) of wakefulness (W), SWS or REM sleep (RS) in each bout length and (F1'–F3') time-weighted frequency histograms showing the proportion (±S.E.M.) of W, SWS or RS amounts in each bout length as a percentage of the total amount of W, SWS during the 3 h post-injection period (10:00–13:00) or RS during the 12:00–16:00 period (N = 6). (G1–G3) Sleep-wake power spectrum changes over baseline during the 3 h (10:00–13:00) post CNO (0.3 mg/kg, N = 6 mice) injection as compared with control injection; and the quantitative changes (±S.E.M.) in power for the δ (0.4–5 Hz), θ (5–9 Hz), α (9–15 Hz), β (15–30 Hz), low γ (30–60 Hz) and high γ (60–120 Hz) frequency bands (±S.E.M.) following vehicle or CNO (0.3 mg/kg, N = 6 mice) administrations. (C–G) Control injection in Black, CNO injection in blue; <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001, two-way ANOVA followed by a post hoc Bonferroni test.

hM3Dq receptors (**Figure 3B1**). CNO (500 nM) application significantly increased the firing rate of SLD neurons (3.15 ± 0.96 vs. 0.68 ± 0.23 Hz in control condition, p = 0.032, **Figure 3B2**). Sleep-wake analysis during the inactive phase (10:00) revealed that CNO (0.3 mg/kg, n = 6, 10:00) injection significantly affected wakefulness (two-way ANOVA, F(23,115) = 2.38, p = 0.0014; **Figure 3C1**), SWS (two-way ANOVA, F(23,115) = 2.82, p = 0.0001; **Figure 3C2**) and REM sleep (two-way ANOVA, F(23,115) = 3.30, p < 0.0001; **Figure 3C3**) in SLDVglut2−hM3Dq mice, as compared with control injection. Wakefulness was significantly increased during the 3 h post CNO injection period (65.0 ± 8.8 vs. 34.8 ± 2.1% of time after control injection, p < 0.0001; **Figure 3D1**). At the same time, SWS amount was significantly decreased (25.9 ± 7.0 vs. 57.3 ± 1.8% of time after control injection, p < 0.0001; **Figure 3D2**). REM sleep amount remained unchanged during the 2 h post CNO injection period (6.2 ± 2.2 vs. 6.5 ± 0.7% of time after control injection, p > 0.05; **Figure 3D3**). However, REM sleep amount was significantly increased during the 2–6 h post-injection period (15.5 ± 1.1 vs. 9.8 ± 1.2% of time after control injection, p < 0.0001; **Figure 3D3**). Interestingly, the SWS to REM sleep latency was significantly decreased after CNO injection (2.0 ± 0.7 vs. 17.8 ± 4.6 min between the beginning of the first SWS episode and the beginning of the first REM sleep episode in control condition, p = 0.018; **Figure 3E**). The observed wakefulness increases during the 3 h post CNO injection resulted from a non-significant increase in both the number of long bouts (>40 min; **Figure 3F1**) and in the proportion of wakefulness from long bouts (>40 min; **Figure 3F1**'). SWS decrease was due to a significant decrease of the proportion of SWS from long SWS bouts (0.0 ± 0.0 vs. 26.0 ± 7.2% of total SWS from 10– 20 min long bouts in control condition, p = 0.0038, **Figure 3F2**'), associated with a significant increase in the proportion of SWS from short SWS bouts (33.7 ± 9.2 vs. 10.8 ± 3.9% of total SWS from 1–2.5 min long bouts in control condition, p = 0.014, **Figure 3F2**'). At the same time, the number of very short SWS bouts (10–30 s long) were significantly increased (9.8 ± 3.0 vs. 3.5 ± 1.8 bouts in control condition, p = 0.015, **Figure 3F2**). The REM sleep increase 2–6 h post CNO injection was due to a significant increase in the number of medium-duration bouts (6.6 ± 1.6 vs. 3.6 ± 1.1 0.5–1 min long bouts in control condition, p < 0.0001; and 12.4 ± 3.0 vs. 6.8 ± 1.2 1–2.5 min long bouts in control condition, p < 0.0001, **Figure 3F3**) while REM sleep bout length is moderately affected (**Figure 3F3'**).

Cortical EEG power distribution was affected by CNO administration during wakefulness (two-way ANOVA, F(243,972) = 3.16, p < 0.0001; **Figure 3G1**), SWS (two-way ANOVA, F(243,972) = 2.27, p < 0.0001; **Figure 3G2**) and REM sleep (two-way ANOVA, F(243,972) = 4.89, p < 0.0001; **Figure 3G3**). During SWS, delta (101.1 ± 12.5 vs. 113.3 ± 7.4% of baseline power in control condition, p < 0.01), sigma (83.8 ± 4.7 vs. 98.0 ± 3.2% of baseline power in control condition, p < 0.001) and beta (91.9 ± 5.3 vs. 98.0 ± 3.2% of baseline power in control condition, p < 0.05) power bands were decreased (**Figure 3G2**). During REM sleep, both delta (86.5 ± 3.2 vs. 109.0 ± 5.9% of baseline power in control condition, p < 0.01) and sigma (78.1 ± 10.1 vs. 96.8 ± 3.0% of baseline power in control condition, p < 0.01) frequency bands were significantly decreased whereas theta (125.0 ± 3.9 vs. 99.1 ± 3.9% of baseline power in control condition, p < 0.001) was significantly increased (**Figure 3G3**). Similar to PZVglut2−hM3Dq mice, SLDVglut2−hM3Dq did not show any sleep-wake phenotypes when CNO was injected at the beginning of the dark/active period (data not shown), indicating a time of the day difference.

# Activation of PBVglut2 Neurons Induces Wakefulness

The excitatory receptor, hM3Dq, was mostly expressed in the PB in five Vglut2-hM3Dq mice (PBVglut2−hM3Dq; **Figure 4A**). Slice electrophysiology showed that firing rates of PBVglut2−hM3Dq neurons were significantly increased (1.48 ± 0.40 vs. 0.32 ± 0.04 Hz in control condition, p = 0.042) by bath application of CNO (500 nM; **Figures 4B1,B2**). Injection of CNO (0.3 mg/kg, 10:00) in PBVglut2−hM3Dq mice significantly affected wakefulness (two-way ANOVA, F(23,92) = 6.12, p < 0.0001; **Figure 4C1**), SWS (two-way ANOVA, F(23,92) = 6.24, p < 0.0001; **Figure 4C2**) and REM sleep (two-way ANOVA, F(23,92) = 3.44, p < 0.0001; **Figure 4C3**). Wakefulness amount was significantly increased during the remaining 9 h of the light period post-injection (79.3 ± 9.5 vs. 32.4 ± 1.5% of time, p < 0.001; **Figure 4D1**). At the same time, both SWS (17.8 ± 7.8 vs. 59.1 ± 1.4% of time, p < 0.001; **Figures 4C2–D3**) and REM sleep (2.9 ± 1.7 vs. 8.5 ± 0.4% of time, p < 0.01; **Figures 4C3–D3**) amount were significantly decreased. No sleep rebound followed the long-lasting wakefulness increase (47.5 ± 1.1 vs. 47.7 ± 2.7% of time spent in SWS during the following dark period, 19:00–07:00, p > 0.05; **Figures 4C2,D2**). Wakefulness enhancement was due to a significant increase in bout length (76.7 ± 7.3 vs. 0.0 ± 0.05 of wakefulness from >40 min long bouts, p < 0.0001, **Figure 4E1**'), associated with a significant decrease in the number of short episodes (**Figure 4E1**). Both SWS bout number (5.6 ± 2.5 vs. 20.8 ± 2.1 5–10 min long bouts, p = 0.0002; **Figure 4E2**) and bout duration (3.4 ± 2.4 vs. 23.9 ± 6.6% of SWS in 10–20 min long bouts, p = 0.025; **Figure 4E2**') were significantly decreased. Similarly, both REM sleep bout number (**Figure 4E3**) and bout duration (**Figure 4E3**') were significantly decreased during the 5 h period following injection. Cortical EEG power distribution was affected by CNO injection during wakefulness (two-way ANOVA, F(543,972) = 1.66, p < 0.0001; **Figure 4F1**). PBVglut2 induced wakefulness was characterized by a significant decrease in cortical EEG delta power (49.9 ± 5.2 vs. 102.1 ± 7.4% of baseline power in control condition, p < 0.01; **Figure 4F2**). Similar results were obtained when CNO was injected at the beginning of the active period (19:00; not shown).

# CNO Does Not Affect Sleep-Wake Cycle in Control Mice

To control for non-specific actions of CNO, non-cre expressing littermate mice were used. No hM3Dq receptor transfection was seen in these control mice. Treatment did not affect the hourly distribution of wakefulness (two-way ANOVA, F(23,92) = 1.53, p = 0.082; **Figure 5A1**), SWS (two-way ANOVA, F(23,92) = 1.56,

FIGURE 4 | Activation of PB glutamatergic neurons during the inactive period (10:00 or ZT3). (A) Extent of transduced neurons (mCherry-positive somas) is shown for individual Vglut2-IRES-cre mice that received bilateral injections of hM3Dq-mCherry-AAV into the PB (PBVglut2−hM3Dq). (B1) PBVglut2−hM3Dq whole-cell recording showing an increase in firing frequency in response to bath application of CNO (0.5 µM). (B2) Average firing frequency (±S.E.M.) during the last 2 min of CNO (0.5 µM) application as compared with the 2 min period preceding CNO application (control; N = 6 PBVglut2−hM3Dq neurons). <sup>∗</sup>p < 0.05 Paired Student's t-tests. (C) Hourly amount of wakefulness (C1), SWS (C2) and REM (C3) sleep following CNO (0.3 mg/kg, N = 5 mice) as compared with control injection. (D1–D3) Percentage of sleep-wake states (±S.E.M.) during the 5 h post-injection period (10:00–15:00), the following 4 h of the light period (15:00–19:00), the subsequent 12 h dark/wake period (19:00–07:00) and first 3 h of the light period of the next day (07:00–10:00, N = 5 mice). (E1,E2) Number of episodes (±S.E.M.) of wakefulness (W) or SWS in each bout length and (E1',E2') time-weighted frequency histograms showing the proportion (±S.E.M.) of W or SWS amounts in each bout length as a percentage of the total amount of W or SWS during the 9 h post-injection period (10:00–19:00; N = 6). (E3) Number of episodes (±S.E.M.) of REM sleep (RS) in each bout length and (E3') time-weighted frequency histograms showing the proportion (±S.E.M.) of RS amounts in each bout length as a percentage of the total amount of W or SWS during the 5 h post-injection period (10:00–15:00; N = 6). (F1)Wake power spectrum changes over baseline during the 3 h (10:00–13:00) post CNO (0.3 mg/kg, N = 5 mice) injection period as compared with control injection. (F2) Quantitative changes (±S.E.M.) in power for the δ (0.4–5 Hz), θ (5–9 Hz), α (9–15 Hz), β (15–30 Hz), low γ (30–60 Hz) and high γ (60–120 Hz) frequency bands (±S.E.M.) following vehicle or CNO (0.3 mg/kg, N = 5 mice) administrations. Power spectral analysis was not performed for slow-wave-sleep and REM sleep due to the low amount of these vigilance stages. (C–F) Control injection in Black, CNO injection in green; <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001, two-way ANOVA followed by a post hoc Bonferroni test.

p = 0.071; **Figure 5A2**) or REM sleep (two-way ANOVA, F(23,92) = 1.06, p = 0.4; **Figure 5A3**). Moreover, CNO treatment did not affect the number of episodes or the episode length distribution as compared with control injection (**Figures 5B1– B3**') in any vigilance state. Finally, the cortical EEG power distribution during wakefulness, SWS and REM sleep was similar after CNO injection, as compared with both control injection and baseline recording (**Figures 5C1–C3**). These results confirm that the sleep-wake effects seen in PZVglut2−hM3Dq, PBVglut2−hM3Dq and SLDVglut2−hM3Dq mice is due to the specific activation of glutamatergic neurons.

### Phox2B Is a Specific Marker for PZ Glutamatergic Neurons

Because chemogenetic activation of PZVglut2 neurons resulted in phenotypes resembling chemogenetic activation of PBVglut2 and SLDVglut2 neurons, i.e., wakefulness and REM sleep increase, respectively, we hypothesized that in the PZVglut2−hM3Dq mouse group some PBVglut2 and SLDVglut2 neurons were transfected and therefore responsible for the phenotypes. To specifically target PZ glutamatergic neurons we took advantage of a specific marker for PZ glutamatergic neurons, Phox2B. In the adult rat medullary and pontine regions in proximity to the PZ, Phox2B expression is restricted to the PZ, with a notable lack of expression in either the PB or the SLD (Kang et al., 2007). Moreover, Phox2B is co-localized with Vglut2 but not with Vgat or GAD67 [**Figures 6D–G**; (Stornetta et al., 2006)], suggesting that Phox2B is a specific marker for PZ glutamatergic neurons. We first confirmed the presence of Phox2B expression in PZ (**Figure 6A**) of mouse using Phox2B-GFP mice. No GFP positive neurons were seen in either the PB or the SLD (**Figures 6B,C**), indicating that Phox2B is specific for PZ glutamatergic neurons. Neurons of the locus coeruleus were GFP positive (LC; **Figures 6A,B**), which is consistent with previous studies showing that Phox2B is necessary for the differentiation of central noradrenergic and adrenergic neurons (Pattyn et al., 2000; Huber et al., 2005). We then assessed the extent of co-localization between Vglut2 and Phox2B in PZ. In each of the slices containing the PZ (n = 12 from 4 mice), Vglut2 co-localized exclusively with Phox2B and Phox2B was found primarily lateral to the facial nerve, in the entire PZ area (**Figures 6H–K**). Higher magnification photomicrographs show the cellular details of Vglut2/Phox2B co-localization

**Figures 6L–O**). Therefore, Phox2B is a specific marker for PZ glutamatergic neurons and Phox2B-IRES-cre mice can be used to specifically activate PZ glutamatergic neurons and study their role in sleep-wake control.

# Activation of PZPhox2B Neurons Does Not Affect Sleep-Wake Cycle

To assess the involvement of PZ glutamatergic neurons, five Phox2B-IRES-Cre mice were injected into the PZ with AAVhM3Dq-mCherry (**Figure 7A**). Three of the five cases also showed partial expression in the LC. Whole-cell in vitro recording (**Figure 7B1**) confirmed that bath application of CNO (500 nM) significantly increased firing rate in PZPhox2B−hM3Dq neurons (3.9 ± 1.8 vs. 1.4 ± 1.2 Hz in control condition, p = 0.028; **Figure 7B2**). CNO injection was successful in four of the five PZPhox2B−hM3Dq mice (one mouse displayed an atypical adverse reaction to the injection). CNO treatment during the light period (10:00) did not affect the hourly amounts of wakefulness (twoway ANOVA, F(23,69) = 1.26, p = 0.23; **Figure 7C1**), SWS (two-way ANOVA, F(23,69) = 1.21, p = 0.27; **Figure 7C2**) or REM sleep (two-way ANOVA, F(23,69) = 0.92, p = 0.57; **Figure 7C3**). In order to study the qualitative aspects of the sleep-wake cycle following activation of PZPhox2B neurons, we studied fragmentation (**Figures 7D1–D3**') and cortical EEG

FIGURE 7 | Activation of PZ Phox2B-expressing neurons during the inactive period (10:00 or ZT3). (A) Extent of transduced neurons (mCherry-positive somas) is shown for individual Phox2B-IRES-cre mice that received bilateral injections of hM3Dq-mCherry-AAV into the PZ (PZPhox2B−hM3Dq). (B1) PZPhox2B−hM3Dq whole-cell recording showing an increase in firing frequency in response to bath application of CNO (0.5 µM). (B2) Average firing frequency (±S.E.M.) during the last 2 min of CNO (0.5 µM) application as compared with the 2 min period preceding CNO application (control; N = 5 PZPhox2B−hM3Dq neurons). <sup>∗</sup>p < 0.05 Paired Student's t-tests. (C) Hourly amount of wakefulness (C1), SWS (C2) and REM sleep (C3) following CNO (0.3 mg/kg, 10 A.M., N = 4 mice) as compared with control injection. (D1–D3) Number of episodes (±S.E.M.) of wakefulness (W), SWS, or REM sleep (RS) in each bout length and (D1'–D3') time-weighted frequency histograms showing the proportion (±S.E.M.) of W, SWS or RS amounts in each bout length as a percentage of the total amount of W, SWS or RS during the 3 h post-injection period (10:00–19:00; N = 4). (E1–E3) Sleep-wake power spectrum changes over baseline during the 3 h (10:00–13:00) post CNO (0.3 mg/kg, N = 4 mice) injection as compared with control injection; and the quantitative changes (±S.E.M.) in power for the δ (0.4–5 Hz), θ (5–9 Hz), α (9–15 Hz), β (15–30 Hz), low γ (30–60 Hz) and high γ (60–120 Hz) frequency bands (±S.E.M.) following vehicle or CNO (0.3 mg/kg, N = 4 mice) administrations. (C–E) Control injection in Black, CNO injection in orange; ∗∗p < 0.01, two-way ANOVA followed by a post hoc Bonferroni test.

power distribution (**Figures 7E1–E3**) of the three vigilance stages during the 3 h post-injection time period. The number of sleep-wake episodes and episode length distribution were similar between CNO and control injections (**Figures 7D1–D3**). Treatment did not affect the cortical EEG power distribution during wakefulness and SWS. However, during REM sleep, the theta frequency band was significantly increased (118.4 ± 6.8 vs. 96.5 ± 2.5% of baseline power in control condition, p < 0.01; **Figure 7D3**) while the sigma frequency band was significantly decreased (78.4 ± 3.8 vs. 101.6 2.5% of baseline power in control condition, p < 0.01; **Figure 7D3**). These results indicate that activation of PZPhox2B−hM3Dq during the light, inactive, period does not affect the sleep-wake cycle but could be involved in cortical EEG activation during REM sleep. Similar results were

obtained when CNO was administrated at the beginning of the active period (19:00; **Figure 8**).

### DISCUSSION

To test the contribution of PZ glutamatergic neurons in sleepwake control, we chemogenetically activated Vglut2-expressing neurons. Based on the sleep-wake phenotype and anatomical confirmation of the injection sites, the mice were separated in three groups: (1) one group, with targeted neuronal transfection mainly in the PZ, displayed increased wakefulness followed by a trend to increased REM sleep during the rest period but not during the active period; (2) a second group, with transfections that included the SLD, displayed a shorter wake increase followed by a significant increase in REM sleep amount; and (3) the third group, which had significant PB transfection, displayed a prominent and long lasting increase in wakefulness, independent of the time of day the injections were performed. Thus, due to the difficulty in targeting only glutamatergic PZ neurons while avoiding glutamatergic neurons in surrounding areas, the role of PZVglut2 neurons in sleep-wake control was still unclear. CNO did not affect the sleep-wake cycle in control mice not expressing the hM3Dq receptor, confirming that the phenotypes seen in Vglut2 transfected mice were specifically due to the activation of glutamatergic neurons. To test the specific role of PZ glutamatergic neurons in sleep-wake control, we took advantage of Phox2B, a transcription factor expressed by a subset of brainstem glutamatergic neurons. We first confirmed that Phox2B is a specific marker for PZ glutamatergic neurons in mice. Because chemogenetic activation of PZPhox2B neurons did not affect sleep-wake phenotypes, we can conclude that PZ glutamatergic neurons are not sufficient to influence the sleepwake cycle.

The absence of a sleep-wake phenotype in control and PZPhox2B−hM3Dq mice after CNO injection provides additional

evidence that CNO, at the dose used in our studies, does not affect the baseline sleep-wake cycle. A recent study had shown that clozapine, a metabolite of CNO, but not CNO, crosses the blood brain barrier and binds chemogenetic receptors with high affinity in rats (Gomez et al., 2017). This finding was subsequently challenged by the observation that both clozapine and CNO cross the blood brain barrier in mice, and that unbound CNO is present in the brain at concentrations sufficient to activate DREADDs, albeit at a higher initial dose than we typically use (Jendryka et al., 2019). We had previously shown that, at a dose of 0.3 mg/kg, CNO does not affect sleep-wake quantity and quality in Vgat-IRES-cre mice (Anaclet et al., 2014), nor does it interfere with the wake-promoting actions of armodafinil and caffeine (Anaclet et al., 2018). In the present study, we confirm the absence of non-specific actions of CNO on the sleep-wake cycle, using two different mouse strains, Vglut2-IRES-cre and Phox2B-IRES-cre mice. Additionally, we confirmed that CNO is able to directly activate PZVglut2−hM3Dq, PBVglut2−hM3Dq, and SLDVglut2−hM3Dq neurons in vitro, where the short application duration (few minutes) and the absence of hepatic metabolism make back-conversion to clozapine highly unlikely. In summary, CNO was able to activate glutamatergic neurons expressing hM3Dq chemogenetic receptor and did not result in non-specific sleep-wake phenotypes.

### Phox2B Is a Specific Marker for PZ Glutamatergic Neurons

The transcription factor Phox2B has been studied for its involvement in the control of breathing and autonomic regulation. Phox2B mutations have been implicated in congenital central hypoventilation syndrome (Moreira et al., 2016). Phox2Bexpressing neurons located in the medullary retrotrapezoid nucleus (RTN), ventral from the facial nucleus, are sensitive to hypoxia (Onimaru et al., 2008), hypercapnic acidosis and serotonin (Wu et al., 2019). Phox2B is necessary for the differentiation of central noradrenergic and adrenergic neurons (Pattyn et al., 2000; Huber et al., 2005). Phox2B is expressed in PZ in adult rats (Kang et al., 2007) but these neurons have no known physiological function. In the present study, we showed that, in PZ, Phox2B is highly co-localized with Vglut2 and therefore, is a specific marker for PZ glutamatergic neurons. Phox2B is also highly co-localized with LC noradrenergic neurons, known to be wake-promoting. However, in this study, the three mice showing partial expression of hM3Dq receptors in LC, did not display an increase in wake amount following CNO injection. It is possible that either the partial coverage of LC was not enough to promote wakefulness or LC noradrenergic neurons were not activated by the chemogenetic ligand. In vitro recordings of LCPhox2B−hM3Dq neurons would be necessary to answer this question.

# The Role of PZ Glutamatergic Neurons in Sleep-Wake Control

A previous study has suggested that some PZ non-GABAergic neurons are sleep-active (Anaclet et al., 2012). Because glutamatergic neurons are the only other neuronal population identified in PZ thus far, we tested if chemogenetic activation of PZ glutamatergic neurons affects sleep-wake phenotypes. Specific targeting of PZ glutamatergic neurons using Vglut2-cre mice was challenging. Of the over 29 injected mice, only six displayed hM3Dq expression mainly in PZ (PZVglut2−hM3Dq). Five mice displayed hM3Dq expression mainly in PB (PBVglut2−hM3Dq) and six in SLD (SLDVglut2−hM3Dq). The remaining mice included seven showing expression of hM3Dq at multiple sites, and five died after surgery or during the sleep recordings. These last two mouse groups were excluded from the study.

Chemogenetic activation of PZVglut2−hM3Dq neurons at the beginning of the mouse active phase (19:00) had limited impact on the sleep-wake cycle. On the other hand, chemogenetic activation of PZVglut2−hM3Dq neurons during the mouse rest phase (10:00) resulted in an early wake enhancement followed by an increase in REM sleep amount. Because these phenotypes are reminiscent of the phenotypes observed in PBVglut2−hM3Dq and SLDVglut2−hM3Dq mice, we hypothesized that they were due to the inadvertent transfection of PB and SLD neurons. In other words, in the PZVglut2−hM3Dq group, transfection would not be restricted to PZ. To test this hypothesis and definitively confirm the role of PZVglut2 neurons in sleep-wake control, we took advantage of Phox2B, a specific marker for PZVglut2 neurons. Using Phox2B-cre mice to specifically target PZVglut2 neurons and not neighboring PB and SLD, we showed that PZVglut2 neurons are not sufficient to affect the sleep-wake cycle at any time of the day. These results indicate that PZ glutamatergic neurons have no role in sleep or wake induction and/or maintenance. It remains, however, to be tested whether PZVglut2 neurons are necessary for normal sleep-wake cycle control, using inhibitory chemogenetic receptors and/or cell body specific lesion.

### A New Mouse Model for REM Sleep Enhancement

Rostral to the PZ and PB, the SLD contains Vglut2-expressing neurons that are specifically active during REM sleep recovery (Clement et al., 2011). The SLD contains a large proportion of neurons with tonic discharge patterns immediately prior to and during REM sleep (Sakai, 2015). Cell body specific SLD lesions, knockout of glutamatergic transmission and genetic inactivation significantly reduce REM sleep amount and result in REM sleep without muscle atonia (Lu et al., 2006; Krenzer et al., 2011; Valencia Garcia et al., 2017). In the present study, we show for the first time that chemogenetic activation of SLDVglut2 neurons results in increased REM sleep amount and reduced SWS to REM sleep latency. Moreover, cortical EEG theta power is significantly enhanced during REM sleep. These data provide a new and unique model of REM sleep enhancement. Such a model will permit probing of the specific role of REM sleep in other neurophysiological functions, such as memory consolidation. However, specific targeting of SLD glutamatergic neurons is challenging due to the close proximity of PB wake-promoting glutamatergic neurons (Fuller et al., 2011). A specific marker for SLD glutamatergic neurons would be very useful.

# Additional Evidence for the Importance of PB in Wakefulness

In close proximity to PZ, just dorsal, lateral and rostral, the PB is a critical brainstem wake-promoting system. Following lesions of both PB and precoeruleus (PC), rats can no longer sustain cortical activation and become comatose (Fuller et al., 2011). Since this seminal study, the role of PB glutamatergic neurons in wakefulness has been refined. Specific lesions of medial PB result in hypersomnolence (Kaur et al., 2013). Glutamatergic neurons located in the external lateral PB are activated by hypoxia and are a key component of the vitally important circuitry regulating arousal from sleep apnea episodes (Kaur et al., 2017). In the present study we show that chemogenetic activation of medial PB results in long lasting wake enhancement. Moreover, CNO induced wakefulness was characterized by a decreased delta frequency band power. Because the delta band is considered a marker of EEG synchronization and is more prominent during quiet wakefulness, this result indicates a more active wake state induced by activation of PBVglut2 neurons. Finally, no sleep rebound was seen after the wake enhancement. This is in accordance with previous studies using chemogenetics to specifically activate wake-promoting neuronal populations (Anaclet et al., 2015; Venner et al., 2016; Pedersen et al., 2017) and indicates that chemogenetic activation of wake-promoting neuronal populations does not enhance the homeostatic drive for sleep. All together, these results confirm the strong wake-promoting action of PB glutamatergic neurons.

### CONCLUSION

This study shows, for the first time, that PZ glutamatergic neurons are not sufficient to affect the sleep-wake cycle in mouse. However, chemogenetic activation of PB or SLD glutamatergic neurons results in wake or REM sleep enhancement, respectively. Finally, Phox2B is a specific marker for PZ glutamatergic neurons. All together, these results provide a better understanding on how the brain regulates sleep-wake

### REFERENCES


cycles, forming a framework for future studies characterizing the sleep-promoting subpopulation of the PZ.

### DATA AVAILABILITY

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

# ETHICS STATEMENT

All procedures were approved by the Institutional Animal Care and Use Committee of Beth Israel Deaconess Medical Center and of University of Massachusetts Medical School.

### AUTHOR CONTRIBUTIONS

EE performed the immunostaining and analyzed the sleep data. LF performed and analyzed the in vitro experiments. HG performed and analyzed the in situ hybridization experiments and wrote the manuscript. CA performed the surgeries and the in vivo experiments, and wrote the manuscript.

# FUNDING

This research was funded by the National Institutes of Health grants K99MH103399 (CA) and R00MH103399 (CA), Coins for Alzheimer's Research Trust (CART) fund and by the University of Massachusetts Medical School startup funds. The latter were used to fund the open access publication fees.

# ACKNOWLEDGMENTS

We are grateful to Quan Hue Ha, Minh Ha, Myriam Debryune, Rebecca Broadhurst, and Tilar Martin for superb technical assistance. We thank Dr. Patrick Fuller for his support.

the rostral medullary brainstem. J. Neurosci. 32, 17970–17976. doi: 10.1523/ JNEUROSCI.0620-12.2012



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Erickson, Ferrari, Gompf and Anaclet. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Effects of 3 Weeks of Water Immersion and Restraint Stress on Sleep in Mice

Shinnosuke Yasugaki1,2, Chih-Yao Liu1,3, Mitsuaki Kashiwagi1,4, Mika Kanuka<sup>1</sup> , Takato Honda1,3, Shingo Miyata<sup>5</sup> , Masashi Yanagisawa1,6,7 and Yu Hayashi<sup>1</sup> \*

1 International Institute for Integrative Sleep Medicine (WPI-IIIS), University of Tsukuba, Tsukuba, Japan, <sup>2</sup> Doctoral Program in Biomedical Sciences, Graduate School of Comprehensive Human Sciences, University of Tsukuba, Tsukuba, Japan, <sup>3</sup> Ph.D. Program in Human Biology, School of Integrative and Global Majors, University of Tsukuba, Tsukuba, Japan, <sup>4</sup> Doctoral Program in Kansei, Behavioral and Brain Sciences, Graduate School of Comprehensive Human Sciences, University of Tsukuba, Tsukuba, Japan, <sup>5</sup> Division of Molecular Brain Science, Research Institute of Traditional Asian Medicine, Kindai University, Osaka, Japan, <sup>6</sup> Department of Molecular Genetics, University of Texas Southwestern Medical Center, Dallas, TX, United States, <sup>7</sup> Life Science Center for Survival Dynamics, Tsukuba Advanced Research Alliance, University of Tsukuba, Tsukuba, Japan

### Edited by:

Patrick Fuller, Beth Israel Deaconess Medical Center and Harvard Medical School, United States

### Reviewed by:

Roberto Amici, University of Bologna, Italy Ashley Miranda Ingiosi, Washington State University Health Sciences Spokane, United States

> \*Correspondence: Yu Hayashi hayashi.yu.fp@u.tsukuba.ac.jp

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 22 May 2019 Accepted: 24 September 2019 Published: 14 October 2019

### Citation:

Yasugaki S, Liu C-Y, Kashiwagi M, Kanuka M, Honda T, Miyata S, Yanagisawa M and Hayashi Y (2019) Effects of 3 Weeks of Water Immersion and Restraint Stress on Sleep in Mice. Front. Neurosci. 13:1072. doi: 10.3389/fnins.2019.01072 Repeated stress is a risk factor for mental disorders and can also lead to sleep disturbances. Although the effects of stress on sleep architecture have been investigated in rodents, the length of the stress exposure period in most studies has been limited to about 10 days, and few studies have analyzed the effects of chronic stress over a longer period. Here we investigated how sleep is affected in a mouse model of depression induced by 3 weeks of daily water immersion and restraint stress (WIRS). Sleep was recorded after 1, 2, and 3 weeks of stress exposure. Some stress-induced changes in several sleep measures were maintained across the 3 weeks, whereas other changes were most prominent during the 1st week. The total amount of non-rapid eye movement sleep (NREMS) was increased and the total amount of time spent awake was decreased across all 3 weeks. On the other hand, the amount of REMS during the dark phase was significantly increased in the 1st week compared with that at baseline or the 2nd and 3rd weeks. Electroencephalogram (EEG) power in the delta range was decreased during NREMS, although the total amount of NREMS was increased. These findings indicate that repeated WIRS, which eventually leads to a depression-like phenotype, differentially affects sleep between the early and subsequent periods. The increase in the amount of REMS during the dark phase in the 1st week significantly correlated with changes in body weight. Our results show how sleep changes throughout a long period of chronic stress in a mouse model of depression.

Keywords: stress, sleep, depression, mouse, REM sleep

# INTRODUCTION

Sleep disturbances are major symptoms of various psychiatric disorders. More than 90% of patients with depression experience sleep disturbances (for review see Reynolds and Kupfer, 1987). Sleep architecture refers to the cycles of non-rapid eye movement sleep (NREMS) and REMS. Abnormalities in the sleep architecture, especially the REMS cycle, are frequently observed in depressed individuals (for reviews see Armitage, 2007; Nutt et al., 2008; Steiger and Kimura, 2010;

Palagini et al., 2013; Medina et al., 2014). For example, several studies have demonstrated an increased density (Reynolds et al., 1985; Rotenberg et al., 2000; for reviews see Nutt et al., 2008; Steiger and Kimura, 2010; Palagini et al., 2013; Medina et al., 2014) and decreased latency of the first REMS episode (Reynolds et al., 1985; Papadimitriou et al., 1988; Hubain et al., 2006; for reviews see Armitage, 2007; Nutt et al., 2008; Steiger and Kimura, 2010; Palagini et al., 2013; Medina et al., 2014). A decrease in slow-wave sleep is also reported in patients with depression (Hubain et al., 2006; Lopes et al., 2007; for reviews see Armitage, 2007; Nutt et al., 2008; Steiger and Kimura, 2010; Palagini et al., 2013; Medina et al., 2014). To understand and overcome the sleep disturbances associated with psychiatric disorders such as depression, it is crucial to elucidate how changes in the sleep architecture emerge during the development of the disease.

Stress is a major environmental risk factor for psychiatric disorders. Especially, chronic stress and chronic exposure to high levels of stress hormones are associated with the development of depression in humans (for reviews see Lupien et al., 2009; Tafet and Nemeroff, 2016). Rodents are useful animal models for studying the effects of stress. Although careful interpretation is required, rodents exposed to chronic stress exhibit various phenotypes that resemble the features of depression, including despair-like and anhedonia-like behaviors (for reviews see Czéh et al., 2016; Slattery and Cryan, 2017; Wang et al., 2017). These phenotypes are alleviated by the administration of antidepressants that are effective in humans (for reviews see Willner, 2005; Mahar et al., 2014; Hare et al., 2017; Ramaker and Dulawa, 2017). Thus, the use of rodent models is expected to provide clues to the mechanisms underlying changes in the sleep architecture due to stress that potentially contribute to the development of psychiatric diseases. Various studies have examined the effects of stress on sleep in rodents. For example, exposure to acute restraint stress for 0.5–2 h increases the REMS (Rampin et al., 1991; Gonzalez et al., 1995; Bonnet et al., 1997; Marinesco et al., 1999; Meerlo et al., 2001; Koehl et al., 2002; Dewasmes et al., 2004; Paul et al., 2009; Rachalski et al., 2009). Social defeat stress has a similar effect (Meerlo and Turek, 2001; Henderson et al., 2017). REMS increases during the dark phase (Meerlo and Turek, 2001; Meerlo et al., 2001; Paul et al., 2009; Henderson et al., 2017) regardless of the timing of the exposure to stress (Koehl et al., 2002). This increase in REMS after daily stress is maintained throughout a 10-day period of chronic exposure (Henderson et al., 2017). One study, however, reported that the REMS amount increased significantly only after several days of daily stress exposure (Wells et al., 2017). In previous studies using restraint stress or social defeat stress, the exposure to stress was limited to a relatively short period ranging from a single day to approximately 10 days. Thus, the aim of this study was to analyze the effects of exposure to daily stress over a longer period.

In this study, we focused on water immersion and restraint stress (WIRS), which is a combination of restraint stress and water-immersion stress that leads to the robust emergence of depression-like behaviors (Miyata et al., 2011). Moreover, 3 weeks of WIRS alters white matter integrity in the corpus callosum, another feature in common with patients diagnosed with depression (Miyata et al., 2011, 2016). While social defeat stress was widely applied in previous studies, prolonged cycles could lead to injuries as well as damage to the electroencephalogram (EEG) and electromyogram (EMG) electrodes due to attacks by an aggressive mouse during the stress session. Thus, in the present study, we applied WIRS daily for 6 consecutive days per week for up to 3 weeks and analyzed the effect on the sleep architecture weekly.

### MATERIALS AND METHODS

### Animals

All animal experiments were approved by the Institutional Animal Care and Use Committee of the University of Tsukuba, and all procedures were conducted in accordance with the Guidelines for Animal Experiments of the University of Tsukuba. Adult male C57BL/6J mice (8–19 weeks old) were used in this study. The mice were housed in a standard cage and maintained in controlled environment (23.5 ± 2.0◦C, 51.0 ± 10.0% humidity, 12-h light/dark cycle). Food and water were available ad libitum.

### Surgery

The mice were anesthetized with isoflurane and placed in a stereotaxic frame (David Kopf Instruments, CA, United States). Core body temperature was maintained using a heating pad. EEG electrodes were stainless steel recording screws implanted epidurally over the parietal cortex (ML + 1.5 mm, AP + 1.0 mm from lambda) and the cerebellum (ML 0.0 mm, AP −6.5 mm from bregma). EMG electrodes were stainless steel Tefloncoated wires placed bilaterally into the trapezius muscles. The electrodes were fixed to the skull with dental cement (Super-Bond C&B set; Sun Medical, Shiga, Japan and Unifast II; GC Corporation, Tokyo, Japan). The mice were allowed to recover in their home cage for at least 3 weeks before transfer to the sleep recording chamber.

### EEG/EMG Recording and Analysis

Mice were attached to a recording cable and acclimatized to a sleep recording chamber for at least 3 days. EEG/EMG recordings were performed four times: baseline, 1st week (1W), 2nd week (2W), and 3rd week (3W) during the stress exposure period. For baseline sleep, the EEG/EMG was recorded for 24 h. For sleep following stress exposure, the EEG/EMG was recorded for 45 h after the stress session [Zeitgeber time (ZT) 3.0-] on Day 6 of each week. For the naïve mice, EEG/EMG recordings were performed two times: 3 and 6 weeks after surgery, corresponding to baseline and 3W of the stressed mouse group in terms of the period after surgery. The EEG/EMG data were filtered (band pass 0.5–64 Hz), and collected and digitized at a sampling rate of 128 Hz, and further filtered post hoc by software (EEG: high pass 0.5 Hz). EEG signals were subjected to fast Fourier transform and further analysis using SleepSign (Kissei Comtec, Nagano, Japan). The vigilance state in each epoch was manually classified as wake, NREMS, or REMS by every 4-s epoch based on EEG patterns of delta power (0.5–4.0 Hz), the theta power (6.0– 10.0 Hz) to delta power ratio, and the integral of EMG signals. Epochs with high EMG and low delta power were classified as wakefulness. Epochs with high delta power and low EMG were classified as NREMS. Epochs with even lower EMG (suggestive of muscle atonia) and high theta power to delta power ratio were classified as REMS (**Figure 2A**). If a single epoch contained multiple states, the state with the longest duration was assigned. For EEG spectrum calculation, to avoid the effect of mixed states, any epochs which contained multiple states were excluded. For each individual, the average EEG power spectrum of each state was calculated and normalized using the average absolute value of the total EEG power across all frequencies and across all 24 h.

### Behavioral Analysis

fnins-13-01072 October 17, 2019 Time: 16:57 # 3

All behavioral assays were performed during the light phase, between ZT 0.0 and 1.0. Prior to the behavioral experimental procedure, the mice were handled for 2 min twice daily for 5 days. The forced swim test (FST) and sucrose preference test (SPT) were performed on the 4th week, following 3 weeks of WIRS. Before each individual test session, the apparatus was sterilized with weak acidic water.

### Forced Swim Test

The FST was performed as described previously with some modifications (Hashikawa-Hobara et al., 2015). Briefly, mice were transferred to the behavioral testing room for acclimatization at least 10 min prior to the test session. The mice were placed individually in a plastic chamber (25 cm high and 20 cm in diameter) filled with water (23.5 ± 1.0◦C) to a depth of 18 cm and forced to swim for 10 min. The light intensity was 60 lx. Mouse activity was recorded using a video camera and analyzed by visual observation. Immobility was defined as the state where motion seemed to stop or was minimized to only that required to keep the head above water. The data between 1 and 6 min from the start point were used for analysis. The analysis was conducted twice for each mouse, and the mean value was taken as the score of each mouse.

### Sucrose Preference Test

The SPT was performed as described previously with some modifications (Ramirez et al., 2015). Briefly, on Day 1, mice were each habituated in a standard cage with two bottles containing drinking water. On Day 2, one bottle of drinking water was replaced with 2% (w/v) sucrose (Nacalai Tesque, Inc., Kyoto, Japan) water. The bottles of drinking water and sucrose water were placed randomly in one of the two water bottle holders in the cage to prevent the mice from developing a side preference. The weight of the bottles was measured at the starting time and 24 h later. The positions of the two bottles were changed at 12 h to avoid the development of a side preference. Sucrose preference was calculated by the following index: (weight of consumed sucrose water)/[weight of consumed (sucrose water + drinking water)] × 100.

### Water Immersion and Restraint Stress

Water immersion and restraint stress was applied as described previously with some modifications (Mizoguchi et al., 2000; Miyata et al., 2011). Briefly, mice were restrained by a 50-ml conical polypropylene centrifuge tube containing multiple air holes. To prevent the mice from escaping, each tube was placed in a stretched net, and the net and tail were fixed in place with a rubber band. The tubes were immersed in water vertically to the level of the xiphoid process for 2 h between ZT 0.0 and ZT 2.5. Water temperature was adjusted to 22.0 ± 1.0◦C. Mice were subjected to this stress session once a day for 6 consecutive days per week for 3 weeks. Body weight was measured before the stress session each day. For control mice, instead of exposure to 2 h of WIRS, the mice were placed in a novel cage for 2 h with no food or water. In an independent group of mice, core body temperature was measured using a heating pad system (Muromachi Kikai, Tokyo, Japan) before the stress session, immediately after (0), 0.5, 1, 1.5, or 2 h after the stress session on Day 1 and Day 5. While daily WIRS for 6–7 h induces gastric ulcers (Konturek et al., 1991; Brzozowski et al., 1993), we previously confirmed that the milder protocol used in the present study does not cause gastric ulcers (Mizoguchi et al., 2000; Miyata et al., 2011).

### Experimental Protocols

The number of animals that were used for each study is as follows. For behavioral analyses shown in **Figures 1A–D**, control group:

N = 4, stressed group: N = 8. For EEG/EMG recordings shown in **Figures 2B–D**, **3A–D**, **4A–E**, **6A–C**: N = 7; **Figures 3E**, **7D**: N = 6; and **Figures 5A–F**, **7C**: N = 5. For core body temperature measurement shown in **Figures 7A,B**, control group: N = 4, WIRS group: N = 4, WIRS with small water bath: N = 6. Throughout the 3 weeks of experiment, mice were individually housed in the sleep recording cages except during the WIRS.

### Statistics

All data were analyzed using Prism 7 (Graph Pad, San Diego, CA, United States) or Excel (Microsoft, Redmond, WA, United States). For behavioral analyses, a Mann–Whitney U-test was performed. Changes in body weight were analyzed using an unpaired t-test. For analysis of daily variations of wake, NREMS, and REMS, a two-way repeated measures ANOVA was performed followed by Bonferroni's multiple comparisons test. Time spent in each stage of wake, NREMS, and REMS, and the ratio of REMS to the total sleep amount was analyzed using a one-way repeated measures ANOVA followed by Bonferroni's multiple comparisons test. The effects of water bath type or period from surgery on time spent in each stage of wake, NREMS, and REMS were analyzed using paired t-test. Both number of episodes and mean episode duration in each stage of wake, NREMS, and REMS, and REMS latency were also analyzed using a one-way repeated measures ANOVA followed by Bonferroni's multiple comparisons test. The EEG power spectra were analyzed using a two-way repeated measures ANOVA followed by Bonferroni's multiple comparisons test. For analysis of the correlation between the REMS increase during the dark phase and body weight change, Pearson correlation coefficients were evaluated. Changes in core body temperature were analyzed using a two-way repeated measures ANOVA followed by Bonferroni's multiple comparisons test. Where applicable, all tests were two-tailed.

# RESULTS

### Assessment of Depression-Like Phenotypes Following WIRS

Prior to analyzing the effects of WIRS on sleep architecture, we confirmed the emergence of a depression-like phenotype by exposure to the stress for 3 weeks. In assessing depression-like behavior, the FST was applied as a measure for despair-like phenotypes and the SPT was applied as a measure for anhedonialike phenotypes. Stressed mice showed an increase in immobility time in the FST (**Figure 1A**) and a decrease in sucrose preference in the SPT (**Figure 1B**) compared with control mice. In addition, body weight gain during the 3 weeks was lower in stressed mice than in control mice (**Figures 1C,D**). These results are consistent with previous reports on depression-like phenotypes caused by WIRS (Miyata et al., 2011; Hashikawa-Hobara et al., 2015; Dai et al., 2018).

### Effect of WIRS on the Sleep/Wake Cycle

We next investigated the effects of 3 weeks of WIRS on sleep/wake (**Figure 2**). **Figure 2A** shows the experimental timeline of the stress exposure and EEG/EMG recordings. EEG/EMG was recorded each week for 45 h beginning on Day 6 (ZT3.0-). In the baseline sleep recording, EEG/EMG was recorded for 24 h.

Stress induced various changes in the amount of sleep/wake. At some times of day, sleep/wake was similarly affected at 1W, 2W, and 3W. By contrast, at some other times of day, sleep/wake was affected most strongly at 1W whereas the effect was milder at 2W and 3W.

Across all weeks during the stress session, at ZT 21.0–24.0 on Day 6 (∼18 h after the stress), there was a decrease in wake and concomitant increase in NREMS compared with that at baseline (**Figures 2B,C**). In contrast, across all weeks during the stress session, ZT 12.0–15.0 at Day 7 (∼33 h after the stress), there was an increase in wake and concomitant decrease in NREMS (**Figures 2B,C**).

Although the changes in the amount of wake and NREMS appeared to mirror each other, the change in the amount of REMS differed from that in the other stages. Across all weeks, at ZT 3.0– 6.0 on Day 6 (immediately after the stress), REMS was decreased, and at ZT 9.0–12.0 on Day 6 (6 h after the stress), REMS was increased, while the amounts of wake and NREMS were not significantly affected in either period (**Figure 2D**). Another feature of REMS was that the effect of stress changed dynamically across the whole stress exposure period. This was most obvious during the dark phase of Day 6, especially at ZT 15.0–21.0, where the REMS amount was largely increased at 1W, whereas the increase appeared more blunted at 2W and 3W (**Figure 2D**).

### Effect of WIRS on Various Sleep/Wake Parameters

We next compared the total amount in each stage of wake, NREMS, and REMS during the light phase, dark phase, or 24 h after the stress exposure (**Figure 3**). Again, the changes in wake and NREMS appeared to mirror each other (**Figures 3A,B**). Total amount of NREMS in the 24 h after the stress exposure was increased at all weeks compared with that at baseline, whereas the amount of wake was decreased. For REMS, the total amount at 24 h was largely increased at 1W, but not at 2W (**Figure 3C**).

When 24 h were further divided into the light and dark phases, no significant change in the amount of sleep or wake was detected in the light phase. In contrast, in the dark phase, the total amounts of both NREM and REMS were increased at 1W. Importantly, the ratio of REMS to total sleep was also increased during the dark phase at 1W, suggesting that the increase in REMS was not simply a result of an overall increase in sleep (**Figure 3D**).

To confirm that these changes in sleep parameters were not due simply to the gradual deterioration of implanted EEG/EMG electrodes, EEG/EMG recordings were also performed in naïve mice at 3 and 6 weeks after surgery, corresponding to baseline and 3W of the stressed mice in terms of the period after surgery. As a result, the total amount in each stage of wake, NREMS, and REMS did not significantly change in these mice across the period (**Figure 3E**).

Analyses of the number and duration of each episode of sleep/wake (**Figure 4**) revealed that the increase in the REMS amount could be largely attributed to the increase in the number,

FIGURE 3 | The total amount of each sleep/wake stage was affected by WIRS. (A–C) Total amount of wake (A), NREMS (B), and REMS (C) (24 h, light phase, dark phase). (D) Ratio of REMS amount to the total sleep amount (24 h, light phase, dark phase). (E) Total amount of each sleep/wake stage in naïve mice 3 and 6 weeks after surgery. (A–D) N = 7, (E) N = 6. (A–D) <sup>∗</sup>p < 0.05, ∗∗p < 0.01, one-way repeated measures ANOVA followed by Bonferroni's multiple comparisons test. (E) Paired t-test. (A–D) Data are a summary of the first 24 h of the 45 h recording periods and are presented as means ± SEM.

not duration, of REMS episodes (**Figures 4C,D**). For NREMS and wake, we detected no significant change in either the episode number or duration, and thus we could not determine which factor contributed more to the change in the total amount of each stage.

Additionally, we analyzed the latency to REMS, which is defined as the duration of the NREMS episode immediately preceding the first REMS episode. We detected no significant effect of the stress on REMS latency (**Figure 4E**).

# Effect of WIRS on the EEG Power Spectrum

We next analyzed the effect of stress exposure on the EEG power spectrum (**Figure 5**). For the wake stage, during the light phase, there was a decrease in delta power and an increase in theta power across all weeks, perhaps reflecting a more attentive state (**Figure 5A**). By contrast, during the dark phase, theta power was decreased across all weeks. In addition, delta power was decreased at 1W.

For the NREMS stage, both during the light phase and dark phase, a decrease in power in the delta range (0.5–4.0 Hz) was detected across all weeks compared with that at baseline (**Figure 5B**). This effect was strongest at 1W.

For the REMS phase, in contrast to the change in total amount detected during the dark phase, the change in the EEG power spectrum was detected at the light phase (**Figure 5C**). During the light phase, theta power was decreased across all weeks compared with that at baseline. Notably, the reduction in theta power was more pronounced at 2W and 3W compared with that at 1W. This is in large contrast to most other changes in the sleep/wake, which were either most drastic at 1W or comparable across all weeks. During the dark phase, in contrast to the dynamic change in the amount of REMS (**Figure 3C**), no significant change in the EEG spectrum was detected (**Figure 5C**).

We confirmed that the EEG power spectrum did not significantly differ in the naïve mice between 3 and 6 weeks after surgery (**Figures 5D–F**).

### Correlation Between Change in REMS Amount During the Dark Phase and Body Weight Change

To investigate whether changes in sleep are correlated with the development of depression-like phenotypes, we next focused on body weight. A strong effect of stress exposure on sleep/wake was observed in the amount of REMS during the dark phase (**Figures 2**, **3**). Thus, we tested whether the change in the REMS amount during the dark phase was correlated with the change in body weight. At 1W, there was a significant negative correlation (**Figure 6A**). By contrast, there was no significant correlation at 2W or 3W (**Figures 6B,C**).

### Effects of WIRS on Core Body Temperature

Besides stress, WIRS might have other effects that also affect subsequent sleep. Here, we assessed the effect on core body temperature. Immediately after the WIRS session, core

### FIGURE 5 | Continued

fnins-13-01072 October 17, 2019 Time: 16:57 # 9

following WIRS). (D–F) EEG power spectra during wake, NREMS, and REMS (light phase, dark phase) 3 and 6 weeks after surgery in naïve mice. (A–C) N = 5 (two of seven mice were excluded due to contamination of large noise during wake), (D–F) N = 5. <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001, two-way repeated measures ANOVA followed by Bonferroni's multiple comparisons test. Two-way repeated measures ANOVA was applied to the entire 0.5–30 Hz data. The post hoc test results are shown only for the 0.5–10 Hz region as there were no significant points outside of this region. Data are presented as means ± SEM.

body temperature significantly declined, which recovered to baseline level within 1 h (**Figure 7A**). Applying WIRS using a smaller water container partly suppressed the core body temperature decline (**Figure 7B**). Nonetheless, 6 consecutive days of WIRS under this condition resulted in a high amount of subsequent REM sleep comparable to the original WIRS condition (**Figure 7C**).

### Assessment of Sleep/Wake During WIRS

We tested whether mice slept during the WIRS. EEG/EMG recording from mouse undergoing WIRS indicated that mice were completely awake throughout the WIRS (**Figure 7D**).

### DISCUSSION

To our knowledge, this is the first rodent study to examine the effect of chronic stress on the sleep architecture by applying WIRS for as long as 3 weeks. While many rodent studies have applied social defeat stress to address the effect of stress on sleep, the longest period tested was 10 days because social stress may cause severe injuries to the animal and damage the EEG/EMG electrodes. The WIRS protocol is advantageous for circumventing such potential problems.

Comparisons of the sleep architecture between the 1W, 2W, and 3W, and baseline conditions revealed that stress differentially affected sleep at 1W compared with that at 2W and 3W. For example, while the stress induced an overall increase in REMS, this effect was especially strong at 1W. The correlation of the change in body weight with the change in the amount of REMS at 1W further supports the strong association between stress and sleep, especially during the early period. It is possible that, after 1W, animals gradually adapted to the repeated WIRS sessions and experienced less stress. The fact that the body weight continued to decrease to a similar extent at 2W and 3W as at 1W, however, does not support this possibility. It is also possible that prolonged exposure to a stressful environment attenuates the adaptive changes in sleep in response to a stressful environment. Further studies should be performed to examine whether the observed attenuation in the sleep alterations at later periods contributes to the emergence of depression-like behaviors.

Compared with wake and NREMS, changes in REMS occurred more rapidly. In the light phase immediately following stress exposure, REMS was initially strongly decreased and subsequently increased, and these responses occurred at a timing when no significant change was observed in the amount of NREMS or wake. The initial suppression of REMS might be a result of increased corticosterone, as a systemic increase in cortisol is suggested to inhibit REMS (Born et al., 1991). The subsequent strong increase in REMS might in part be a result of a homeostatic response to the prior reduction of REMS, although other mechanisms could also be involved. In the following dark phase, both REMS and NREMS amounts were increased, but the increased REMS amount was more pronounced as the ratio of REMS to the total sleep amount was also increased. This increase in the amount of REMS and NREMS during the dark phase was attenuated at 2W and 3W. The overall increase in the REMS amount is consistent with findings in mice exposed to 10 days of social defeat stress (Wells et al., 2017). Moreover, an increase in the REMS amount is reported in patients with depression (Papadimitriou et al., 1988; for reviews see Armitage, 2007; Palagini et al., 2013). The increased REMS amount might be explained by the REMS-promoting effect of corticotropinreleasing hormone in the forebrain (Kimura et al., 2010). We must note the possibility that effects other than stress that were caused by WIRS might have also contributed to changes in the sleep parameters. For example, effects on metabolism or thermoregulation might have contributed to changes in REMS (for review see Parmeggiani, 2003). In rodents, REMS amount changes dynamically according to environmental temperature, with a peak around 29–30◦C (Szymusiak and Satinoff, 1981; Kumar et al., 2009; for review see Cerri et al., 2017). The initial suppression and subsequent increase of REMS amount observed in our study might be in part due to the rapid decline and subsequent recovery of core body temperature caused by WIRS. However, under conditions of WIRS in which the decline in core body temperature was partly suppressed, REMS amount was still significantly increased to a comparable level, suggesting that such changes in REMS was not solely due to the changes in the core body temperature. During the WIRS session, the mice were sleep-deprived, which might have also contributed to changes in subsequent sleep. However, effects such as the decrease in delta power during NREMS and increase in the ratio of REMS amount to total sleep were not observed after total sleep deprivation (Mochizuki et al., 2004).

Due to uncertainties about the function of REMS, it is difficult to speculate whether the increase in the REMS amount at 1W and its dampening in the subsequent weeks are beneficial or non-beneficial for an animal coping with stress. Antidepressants such as monoamine oxidase inhibitors and selective serotonin reuptake inhibitors reduce REMS (Mayers and Baldwin, 2005). Moreover, REMS deprivation has an antidepressant effect (Vogel et al., 1975). Thus, an increase in REMS might negatively affect recovery from psychiatric diseases. However, because antidepressants act on many brain regions and REMS deprivation is usually followed by a large REMS rebound, it is difficult to differentiate the direct effects of REMS reduction from these other effects. When sleep was recorded from subjects that underwent life-threatening experiences, subjects who had a longer duration of REMS episodes during the acute period had a lower tendency to develop subsequent posttraumatic stress disorder, perhaps supporting the notion that REMS plays a beneficial role (Mellman et al., 2002). Future studies

calculated as follows: (the amount of REMS during the 12 h of dark phase after WIRS at each week) – (the amount of REMS during the 12 h of dark phase in the baseline recording). N = 7. (A–C) Pearson correlation coefficients.

using optogenetics or chemogenetics to manipulate REMS might provide more clues as to how REMS is involved in coping with stress.

Chronically stressed mice also exhibited changes in the EEG power spectrum. In NREMS, especially during the dark phase, the delta power was decreased across all weeks. This is consistent with the observations of decreased slow-wave sleep in patients with depression (Hubain et al., 2006; Lopes et al., 2007; for reviews see Armitage, 2007; Nutt et al., 2008; Steiger and Kimura, 2010; Palagini et al., 2013; Medina et al., 2014). Slow-wave sleep is characterized by a decrease in the blood cortisol levels (Gronfier et al., 1997). Moreover, slow-wave activity itself may contribute to reducing cortisol levels (Besedovsky et al., 2017). Thus, it is possible that the decrease in delta power contributes to increase the cortisol levels and the emergence of depression-like phenotypes. In REMS, theta power was decreased, especially during the light phase. In contrast to most other changes in sleep/wake, which were either strongest at 1W or comparable across all weeks, the reduction in theta power was more pronounced at 2W and 3W. In mice, the surface

ANOVA followed by Bonferroni's multiple comparisons test. (C) <sup>∗</sup>p < 0.05, paired t-test.

the 24 h following 6 consecutive days of WIRS with the smaller water bath. (D) Total amount of each sleep/wake stage during the WIRS session. (A,B) Control: N = 4, WIRS: N = 4, WIRS using smaller water bath: N = 6. (C) N = 5. (D) N = 6. (A,B) ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001, two-way repeated measures

EEG theta power mainly originates from the hippocampus. Thus, a progressive reduction of the theta power might reflect some accumulating damage to the hippocampus. Shrinkage of the hippocampal volume is well known to occur in patients with depression (Schmaal et al., 2016), and chronic stress in mice may have a similar effect.

While the increase in the REMS amount observed in the present study in mice is consistent with findings in human patients with depression (Papadimitriou et al., 1988; for reviews see Armitage, 2007; Palagini et al., 2013), some other characteristics of sleep in depressed patients, including shortening of the REMS latency (Reynolds et al., 1985; Papadimitriou et al., 1988; Hubain et al., 2006; for reviews see Armitage, 2007; Nutt et al., 2008; Steiger and Kimura, 2010; Palagini et al., 2013; Medina et al., 2014) and sleep fragmentation (Rotenberg et al., 2000; Lopes et al., 2007; for reviews see Nutt et al., 2008; Steiger and Kimura, 2010; Palagini et al., 2013; Medina et al., 2014) were not observed in the present study. This might be a limitation of using WIRS to model human stress disorders and depression. In mice, 10 days of social defeat stress results in shortened REMS latency (Wells et al., 2017) and long-term exposure to various types of unpredictable mild stressors results in fragmented sleep (Nollet et al., 2019). Thus, it appears that each model has advantages for recapitulating human diseases. Some studies, however, report no significant difference in the REMS latency in patients with depression (Rotenberg et al., 2000; Lopes et al., 2007). Thus, it is possible that depression itself is a disorder with a spectrum of diverse sleep characteristics.

### CONCLUSION

In summary, this study shows how 3 weeks of WIRS, which is a robust protocol for inducing depression-like behaviors in mice, affects sleep architecture. Several changes observed in the present study recapitulated the sleep disturbances that occur in patients with depression. Moreover, the results demonstrated that mouse sleep exhibits a graded change in response to stress applied for 3 weeks. The findings of this study provide a platform for future studies to address how sleep is affected by chronic stress and how the changes in sleep induce other behavioral or physiologic alterations.

### REFERENCES


### DATA AVAILABILITY STATEMENT

Data sets are available from the corresponding author upon reasonable request.

### ETHICS STATEMENT

The animal study was reviewed and approved by the Institutional Animal Care and Use Committee of the University of Tsukuba.

### AUTHOR CONTRIBUTIONS

SY, C-YL, and YH conceived and designed the experiments with advice from SM. SY, C-YL, and MKas performed the experiments with advice from MKan, TH, and SM. SY and YH analyzed the data and wrote the manuscript with advice from SM and MY.

# FUNDING

This work was supported by the Japan Agency for Medical Research and Development (AMED) under Grant Numbers JP19dm0107138 and 19gm1110008, the JSPS KAKENHI under Grant Numbers JP16H06141 and JP16H01264, the Japan Science and Technology Agency (JST) under Grant Number JPMJPR13AC, the MEXT WPI program, the Cell Science Research Foundation, the Asahi Glass Foundation, the Nakajima Foundation, the Senri Life Science Foundation, the Kanae Foundation for the Promotion of Medical Science, the Takeda Science Foundation, the Mitsubishi Foundation, the Mochida Memorial Foundation for Medical and Pharmaceutical Research, the Astellas Foundation for Research on Metabolic Disorders, the Sumitomo Foundation, the SENSHIN Medical Research Foundation, and the Japan Foundation for Applied Enzymology (TMFC) (to YH).

### ACKNOWLEDGMENTS

We thank all of the Hayashi Lab members for technical assistance, discussion, and comments.



Parmeggiani, P. L. (2003). Thermoregulation and sleep. Front. Biosci. 8:s557–s567.



**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Yasugaki, Liu, Kashiwagi, Kanuka, Honda, Miyata, Yanagisawa and Hayashi. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Dynamic Metabolic Changes in the Human Thalamus at the Transition From Waking to Sleep - Insights From Simultaneous Functional MR Spectroscopy and Polysomnography

### Edited by:

Michael Lazarus, University of Tsukuba, Japan

### Reviewed by:

Luigi De Gennaro, Sapienza University of Rome, Italy Arcady A. Putilov, Independent Researcher, Berlin, Germany

### \*Correspondence:

Hans-Peter Landolt landolt@pharma.uzh.ch

†These authors have contributed equally to this work ‡These authors share senior authorship

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 14 August 2019 Accepted: 14 October 2019 Published: 30 October 2019

### Citation:

Lehmann M, Hock A, Zoelch N, Landolt H-P and Seifritz E (2019) Dynamic Metabolic Changes in the Human Thalamus at the Transition From Waking to Sleep - Insights From Simultaneous Functional MR Spectroscopy and Polysomnography. Front. Neurosci. 13:1158. doi: 10.3389/fnins.2019.01158 Mick Lehmann1,2,3† , Andreas Hock3,4† , Niklaus Zoelch3,4,5† , Hans-Peter Landolt1,2, \* ‡ and Erich Seifritz2,3‡

1 Institute of Pharmacology and Toxicology, University of Zurich, Zurich, Switzerland, <sup>2</sup> Sleep & Health Zurich, University of Zurich, Zurich, Switzerland, <sup>3</sup> Department of Psychiatry, Psychotherapy and Psychosomatics, Hospital of Psychiatry, University of Zurich, Zurich, Switzerland, <sup>4</sup> Institute for Biomedical Engineering, ETH Zurich and University of Zurich, Zurich, Switzerland, <sup>5</sup> Department of Forensic Medicine and Imaging, Institute of Forensic Medicine, University of Zurich, Zurich, Switzerland

An important contribution of the thalamus to the transition from wakefulness to sleep is a consistent finding in animal studies. In humans, only little is currently known about the specific role of the thalamus in regulating wake-sleep transitions. Although changes in thalamic blood flow and activity have been reported, the underlying molecular mechanisms have not been investigated. Knowledge about neurotransmitter changes at the wake-to-sleep transition would be indispensable for a better translation of basic animal research findings to humans. Here, we start to fill this important scientific gap. More specifically, we benefit from recent advances in magnetic resonance (MR) spectroscopy, which allow for the non-invasive, local-specific and high-quality detection of naturally occurring metabolite changes in the human brain. We demonstrate in nine young adults able to produce consolidated sleep in the MR spectroscopy scanner, a specific decrease in thalamic glutamate concentration from wakefulness to stage N2 sleep. The magnitude of this decrease was highly correlated with individual N2 sleep duration. When five participants of the original experiment were kept awake in a separate control condition, no decrease in thalamic glutamate levels occurred. The study highlights for the first time in humans that dynamic changes in distinct brain metabolites can be reliably detected at the transition from waking to sleep. The reported methodology to simultaneously acquire functional MR spectroscopy data and neurophysiological signals offers great potential for investigating the molecular mechanisms underlying the transition between and the maintenance of sleep and wake states in humans.

Keywords: glutamate, electroencephalography (EEG), metabolite cycling, excitability, thalamic reticular nucleus

# INTRODUCTION

fnins-13-01158 October 26, 2019 Time: 15:17 # 2

Brain neuronal activity and metabolism are fundamentally different in wakefulness and sleep, particularly in thalamic and cortical networks (for reviews, see e.g., McCormick and Bal, 1997; Vantomme et al., 2019). The electroencephalogram (EEG) in deep non-rapid-eye-movement (NREM) sleep shows a high prevalence of slow waves, which can be quantified by power spectral analysis as slow-wave activity. EEG slow-wave activity (∼ 0.75–4.5 Hz) includes the slow oscillation (<∼1.5 Hz), reflecting widespread cortical alternations between neuronal UP (ON) and DOWN (OFF) periods (Vyazovskiy et al., 2009; Nir et al., 2011), and delta waves (>∼1.5 Hz) that are widely accepted as a homeostatically regulated physiological marker of sleep depth (Achermann and Borbély, 2017). The thalamus and the surrounding thalamic reticular nucleus are thought to make strong contributions to the generation mechanisms of the slow oscillation, delta waves and sleep spindles, which are the EEG hallmarks of NREM sleep in animals and humans (Crunelli et al., 2018; Vantomme et al., 2019).

Consistent with this notion, early positron emission tomography studies in humans revealed that the strongest reduction in regional cerebral blood flow (rCBF) occurred in the thalamus when slow wave sleep was compared to wakefulness (Maquet et al., 1997). Moreover, the changes in rCBF were correlated with EEG delta and spindle (12–15 Hz) frequency activity in NREM sleep (Hofle et al., 1997). More recent functional magnetic resonance imaging (MRI) studies confirmed decreased thalamic activity inferred from arterial spin labeling in NREM sleep when compared to waking (Tüshaus et al., 2017). Intriguingly, simultaneous intracortical and intrathalamic EEG recordings in human epileptic patients demonstrated that the thalamic deactivation occurring at sleep onset typically precedes that of the cortex by several minutes (Magnin et al., 2010). Together with preclinical insights from correlational, lesion/ablation, pharmacological and genetic studies, the available evidence is consistent with an important role of the thalamus for the occurrence of NREM sleep oscillations in the cortex, but also for the transition from wakefulness to sleep (David et al., 2013; Hughes and Crunelli, 2013; Lemieux et al., 2014; Gent et al., 2018a). Nevertheless, the molecular mechanisms controlling the thalamic neuronal firing patterns in wake and sleep states are virtually unknown.

Earlier neurochemical models of brain circuits controlling wakefulness and sleep primarily proposed monoaminergic and cholinergic mechanisms originating in hypothalamus and basal forebrain (Saper et al., 2010; Brown et al., 2012). More recently, it has been argued that these classic neuromodulators merely play a modulatory role, whereas fast acting neurotransmitters such as glutamate (GLU) and γ-aminobutyric acid (GABA) provide the main grid of the wake-sleep regulatory systems (Saper and Fuller, 2017). Indeed, it has long been proposed that a decrease in GLU and an increase in GABA function are fundamental for sleep-related behaviors (for an overview, see Jones, 2011). Interestingly, modafinil, a wake-promoting agent, increases GLU in thalamic areas (Ferraro et al., 1997). While subregions of the thalamus have emerged as essential hubs for controlling cortical activity and behavioral state (Giber et al., 2015; Liu et al., 2015; Gent et al., 2018b), the exact roles of GLU and GABA, the main excitatory and inhibitory neurotransmitters, in sleep-wake regulation remain unclear.

Knowledge about dynamic changes in these neurotransmitter systems in humans at the wake-to-sleep transition is currently lacking. To start filling this scientific gap, we used recent methodological advances that allow for the non-invasive detection of naturally occurring metabolite concentrations with high data quality in circumscribed areas of the human brain. We acquired functional MR spectroscopy data with high temporal resolution simultaneously with polysomnography, to objectively define the wake-to-sleep transition in healthy adults. Based on previous data in rats (Dash et al., 2009), we hypothesized that the GLU level in the thalamus would decrease from wakefulness to sleep and that the magnitude of this decrease may be associated with the depth of sleep.

# MATERIALS AND METHODS

### Participants

Fifteen adults between 19 and 24 years of age were recruited via advertisements at the University of Zurich. All participants had previous experience with MR studies, were familiar with staying in the scanner for more than 45 min and sleeping on their back in the constrained scanner environment. None fulfilled any of the following exclusion criteria: history of psychiatric/neurological diseases, sleep-wake abnormalities, drug abuse, concurrent medication use, cardiovascular disease, MR exclusion criteria, and pregnancy. No traveling across time zones was allowed for at least 6 weeks before the study. Subjects had to refrain from alcohol, caffeine and sports activities during 24 h prior to the experiment. The Ethics Committee of the Canton of Zurich approved the study. All subjects gave written informed consent before screening.

### Experimental Procedure

One week prior to the experimental night, participants were instructed to keep a regular sleep-wake schedule according to their individual habitual sleep-wake times. Compliance was verified with self-reported sleep logs and wrist motor actigraphy and light sensors (MotionWatch 8, CamNtech Ltd., Cambridge, United Kingdom). In the night prior to the assessment, the subjects' average sleep duration across the pre-study week was reduced by 2 h to slightly elevate sleep pressure. Actigraphic data were controlled at the beginning of each experimental session, to ensure compliance with the study instructions.

On the night of the MR spectroscopy and sleep recordings, subjects arrived ∼ 3 h before their usual bedtime at the MR center of the Psychiatric Hospital Zürich for the application of the electrodes for polysomnography. The experimental setup included 32 EEG channels according to an extended 10–20 system (Brain Products GmbH, Gilching, Germany), electrooculogram (EOG), submental electromyogram (EMG), and electrocardiogram (ECG). The MR protocol started with a structural MRI scan around 2 h before the subjects' habitual

bedtime, to avoid sleepiness in the scanner. Participants were asked to stay awake for at least 20 min during initial MR spectroscopy scanning, to collect sufficient data during wakefulness. Afterward, they were permitted to sleep. The MR spectroscopy and polysomnographic measurements were continued for as long as the volunteers were able to remain in the MR scanner without gross movements. Six study participants were unable to sleep (n = 2) or slept for less than 15 min (n = 4). They were excluded from data analyses, resulting in a final study sample of five females and four males (mean age: 22.3 ± 2.2 [SEM] years).

To control for diurnal or non-sleep-related temporal changes in metabolite ratios during the prolonged measurement period, all participants who were able to sleep in the scanner for longer than 15 min in sleep stages N2 and N3 were invited for a second recording. This additional experimental condition followed the same procedures as described above, yet the volunteers were required to stay awake as controlled by polysomnography for the entire duration of MR spectroscopy scanning.

### MR Spectroscopy Data Acquisition

<sup>1</sup>H-MR spectroscopy measurements were performed on a Philips Achieva 3T whole-body scanner equipped with a 32-channel receive-only phased-array head coil (Philips Healthcare, Best, The Netherlands). Non-water suppressed, PRESS-localized <sup>1</sup>H-MR spectroscopy spectra in the thalamus (**Figure 1A**) were obtained using the metabolite cycling (MC) technique (Hock et al., 2013). This technique allows for the correction of frequency and phase changes caused by system instabilities (e.g., B0-field drifts, blood or cerebrospinal fluid flow, patient movements) between each single acquisition. In addition, this method enables a flexible selection of acquisitions from different time points for a retrospective averaging within the scan session based on the sleep states defined by the polysomnographic analyses. The PRESS localization sequence parameters were chosen as follows: 2500 ms repetition time, 32 ms echo time and 2000 Hz spectral bandwidth and was combined with inner-volume saturation (Edden et al., 2006).

### Polysomnographic Recording During MR Spectroscopy Scanning and Artifact Correction

Wakefulness and sleep during scanning were confirmed by polysomnography as described above using an MR-compatible polysomnographic recording system (BrainAmp, BrainCap and BrainAmp ExG MR devices and electrodes; Brain Products GmbH, Gilching, Germany). Impedances of all electrodes were kept below 10 k. The bioelectric data were synchronized with the scanner clock, sampled at 5 kHz (filtered between 0.01 and 200 Hz) and referenced to the vertex (Cz).

### Sleep EEG Correction and Sleep-State Scoring

Offline EEG data correction was performed using BrainVision Analyzer 2.0 (Brain Products, Gilching, Germany). Removal of MR spectroscopy artifacts was based on the average artifact subtraction method, whereas 21 MR spectroscopy artifact intervals were used for sliding average (Allen et al., 2000). Subsequently, data were corrected with a standard cardioballistogram (CB) artifact template correction (Allen et al., 1998). The cardioballistic artifact template was determined from the cardiac waveform recorded by the ECG channel. The CB artifact to be subtracted was defined by a moving average over 21 cardiac periods. Both, CB and MR artifact correction routines are pre-implemented in the software. The EEG data were high-pass filtered at 0.3 Hz and low-pass filtered at 35 Hz (48 dB/octave) and subsequently down-sampled to 512 Hz for the scoring of waking and sleep states. After MR spectroscopy and CB artifact correction, the polysomnographic recordings were scored offline in 30-s epochs by two independent raters according to standard criteria (Iber et al., 2007).

### MR Spectroscopy Analyses

Each single MR spectroscopy acquisitions was phase corrected and frequency aligned exploiting the advantages of the nonwater suppressed acquisition scheme before acquisitions scored as the same sleep state were averaged. After averaging, eddy-current correction (Klose, 1990) was applied using the unsuppressed water signal to reduce induced line-shape distortions. Additionally, a Hankel singular value decomposition water filter was applied to remove the remaining water peak in the resulting metabolite spectrum (Cabanes et al., 2001).

For quantification of spectral data, LCModel (Version 6.3) with a set of simulated basis spectra, consisting of 18 metabolites, was used (Provencher, 1993). For each metabolite, the ratio to the unsuppressed water signal was calculated.

Based on the polysomnographic recordings, single MR acquisitions were classified into wakefulness and NREM stages, N1, N2, and N3. Signal-to-noise-ratio (SNR) and full-widthat-half-maximum of the spectra according to the LCModel output were compared between categories (i.e., wakefulness and sleep states).

### Statistical Analyses

Metabolic ratios between wakefulness and sleep were compared using paired t-tests. The relationship between sleep variables and changes in metabolic ratios were analyzed using Pearson's product-moment correlation coefficients. A Bonferronicorrected probability of p = 0.05 was set as the significance threshold, yet the uncorrected p-values are reported.

# RESULTS

Actigraphic data to control the pre-experimental study instructions were verified at the beginning of each experimental night.

The "sleep" experiment started at 9.56 pm ± 56 min. Following the pre-sleep wake scan and once permitted to fall asleep, the participants took on average 10.3 min (SD = 7.4 min) to produce N2 sleep and maintained 79.7 min (SD = 20.8 min) of sleep in the scanner (range: 37 - 115 min). Sleep was composed of roughly 20%

decreased from wakefulness to sleep in 8 of 9 study participants (arbitrary units; p < 0.01, two-tailed, paired t-tests) (C) The change from wakefulness to sleep was negatively correlated with N2 sleep duration. Individual participant's data have the same color code in panels (B,C). (D) When a subset of participants stayed awake during an equivalent recording period at the same time of day, no changes in GLU were observed. No changes were observed in GABA ratios, neither from wakefulness to sleep (E) nor during the wake control condition (F). Each individual participant's data have the same color code in panels (B–F).

N3, 57% N2 and 13% N1 sleep, while wakefulness after sleep onset equaled approximately 10% (**Table 1**). No REM sleep occurred.

Analyses of the metabolite changes indicated that the SNR differed significantly between waking and sleep, when "N3" sleep was included in the category "sleep" (t<sup>8</sup> = 7.20, p < 0.001,



Mean values ± SD are reported (n = 9). Sleep latency: time after permission to fall asleep until first occurrence of N2 sleep, defining sleep onset. REM sleep, rapideye-movement sleep; WASO, wakefulness after sleep onset.

TABLE 2 | Water-scaled metabolite concentrations in wakefulness before sleep and during sleep.


Mean values ± SD in arbitrary units are reported (n = 9). t- and p-values refer to two-sided, paired t-tests.

d = 1.01). This rendered the metabolite ratios non-comparable between wakefulness and sleep. The SNR did not differ when ratios were compared between wakefulness and N2 sleep (t<sup>8</sup> = 0.68, p = 0.541).

In all participants but one, the thalamic GLU concentration decreased from wakefulness to N2 sleep (t<sup>8</sup> = −3.74, p = 0.006, d = 1.25; **Figure 1B**). During the 2-h recording period, the decrease in GLU was tightly inversely correlated with the duration of N2 sleep (r = −0.84; p = 0.005; **Figure 1C**). Interestingly, the single subject not showing a reduction in thalamic GLU levels, only produced 17 min of N2 sleep in the MR scanner (**Figure 1C**; "red" participant). Apart from GLU, no other metabolite, including GABA (**Figure 1E**), showed a difference between wakefulness and sleep (p > 0.117; **Table 2**).

To determine whether the observed changes in thalamic GLU concentration were sleep related or merely reflected diurnal or temporal changes in metabolite ratios, seven participants of the "sleep" experiment completed a control recording at the same time of day, but in the absence of sleep. The presence of wakefulness during the entire duration of functional MR spectroscopy scanning was controlled by polysomnography. Five volunteers were able to stay awake during the entire control measurement (89.3 ± 21.5 min). Confirming that sleep is required for the observed decrease in GLU at the sleep-to-wake transition, no reliable changes in GLU (**Figure 1D**) nor in any other metabolite ratio [including GABA (**Figure 1F**); other data not shown] were observed when subjects stayed awake.

### DISCUSSION

The present study provides first experimental evidence for dynamic changes in thalamic GLU concentration at the transition from wakefulness to sleep in humans. The unique data set required the ability to average individual functional MR spectroscopy acquisitions based on the polysomnographydefined occurrence of wakefulness and sleep states. These methodological innovation enabled the reliable quantification of minute metabolic changes in the glutamatergic system between wakefulness and sleep.

At the transition from waking to sleep, a specific decrease in GLU was observed whereas other metabolites did not change. This suggests that our finding is not biased by a general reduction in the water signal, which was used for the scaling of the metabolite values. Moreover, the decrease in GLU was stronger in participants reaching more stable NREM sleep. This observation is in line with previous observations in rat cerebral cortex showing that the rate of decrease in GLU varies with sleep depth and increases when animals are awake (Dash et al., 2009). The latter indicates that changes in glutamatergic metabolism are not mainly driven by circadian mechanisms. Of note, Dash and colleagues studied sleep-wake related and diurnal changes in GLU in rats and their findings may not be transferred directly to humans. In the present study, a circadian contribution to the level of GLU in the human thalamus was not assessed. Moreover, while the removal of scanner artifacts from the EEG allowed the reliable scoring of wakefulness and sleep states, a reliable quantitative analysis of the EEG signals was not possible. Nevertheless, the wake control experiment corroborated the observation in rats that the cerebral GLU levels do not decrease in reliable manner in the absence of sleep. Substantial preclinical evidence supports important roles of thalamic neurons and thalamo-cortical networks in EEG NREM sleep oscillations and the control of behavioral states (wake vs. sleep) (Crunelli et al., 2018; Gent et al., 2018b; Vantomme et al., 2019). The present human data are consistent with the hypothesis that an elevated thalamic GLU concentration promotes the tonic, single-spike activity of thalamo-cortical neurons and the thalamic reticular nucleus typical for wakefulness, whereas reduced thalamic GLU levels facilitate the occurrence of EEG NREM sleep oscillations and consolidated sleep. This conclusion may also be consistent with computational modeling (Destexhe et al., 1994) and in vitro slice electrophysiology data (Cox and Sherman, 1999).

Even though 6 out of 9 participants showed an increase in GABA concentration from wakefulness to sleep, the change was not significant. The reliable detection of GABA with MR spectroscopy in small voxels is generally difficult because of the relatively low abundance of GABA and the spectral overlap with signals from more abundant metabolites. We found no correlation between the wake-sleep related changes in GLU and GABA ratios (r = −0.405, p = 0.280; data not shown). In addition, similar to our study, administration of modafinil to

promote wakefulness in rats specifically increased GLU release in ventromedial and ventrolateral thalamic nuclei while the release of GABA remained unaffected (Ferraro et al., 1997). Although the physiological significance of pharmacologically induced alterations in wake and sleep states has to be interpreted with caution, the present and previous data indicate that merely altering the balance between GLU and GABA may be sufficient to switch the system between wakefulness and sleep. It also has to be kept in mind that the participants in the present study were slightly sleep deprived. Future work may address the question whether differences in homeostatic sleep pressure affect the rate at which brain GLU levels change at the transition from waking to sleep.

Similar to GABA, no changes in thalamic GLX were found at the wake-to-sleep transition. The GLX signal captures the combination of GLU and glutamine which is the biosynthetic precursor of both, GLU and GABA. Virtually opposite changes in GABA and GLU concentrations from waking to sleep may, thus, obscure potential changes in GLX resulting from a decreased GLU level. Interestingly, a strong relationship between right thalamic GLX levels and disturbed sleep was reported in patients with restless legs syndrome (Allen et al., 2013). In that study, however, MR spectroscopy and polysomnographic recordings were not conducted simultaneously such as in the current work.

### CONCLUSION

In conclusion, this study provides first experimental evidence in humans that dynamic metabolic changes can be measured at the transition from wakefulness to sleep. Because of the relatively low sample size, the findings should be qualified as preliminary. Nevertheless, apart from their importance for informing the search for molecular mechanisms underlying human sleepwake regulation, these insights also have implications for the interpretation of studies using resting state MR spectroscopy as a diagnostic tool for sleep-wake and neuropsychiatric disorders (Allen et al., 2013; Moriguchi et al., 2019). Drowsiness and (N2) sleep likely occur in many individuals undergoing MR scanning (Tagliazucchi and Laufs, 2014). Given the evidence provided here, that GLU concentrations differ between wakefulness and N2 sleep, control for changes in vigilance state will be required for the careful interpretation of future MR

### REFERENCES


spectroscopy findings. Moreover, it will be of great interest to investigate whether levels of GLU are globally decreased when the awake brain enters sleep or whether this decrease is restricted to central hubs subserving sleep-wake control such as the thalamus.

### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/supplementary material.

### ETHICS STATEMENT

The studies involving human participants were reviewed and approved by the Ethics Committee of the Canton of Zurich. The patients/participants provided their written informed consent to participate in this study.

### AUTHOR CONTRIBUTIONS

NZ and AH developed the functional MR spectroscopy sequence and the MR spectroscopy data post-processing. ML and AH designed the experiment. ML, AH, and NZ carried out the data collection. ML analyzed the EEG data for sleep scoring. ML and NZ analyzed the MR spectroscopy data. All authors wrote, discussed, and approved the manuscript.

### FUNDING

This work was supported by the Clinical Research Priority Program "Sleep and Health" of the University of Zurich and Swiss National Science Foundation grant # 320030\_163439. ML was supported by the "Forschungskredit" of the University of Zurich (Grant # FK-16-044).

### ACKNOWLEDGMENTS

We thank Linda Hamida for assistance in data collection.

to disturbed sleep. Neurology 80, 2028–2034. doi: 10.1212/WNL. 0b013e318294b3f6



**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Lehmann, Hock, Zoelch, Landolt and Seifritz. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Transcriptome Analysis of Pineal Glands in the Mouse Model of Alzheimer's Disease

Kwang Il Nam<sup>1</sup> , Gwangho Yoon1,2 , Young-Kook Kim2,3\* and Juhyun Song1,3 \*

<sup>1</sup>Department of Anatomy, Chonnam National University Medical School, Jeollanam-do, South Korea, <sup>2</sup>Department of Biochemistry, Chonnam National University Medical School, Jeollanam-do, South Korea, <sup>3</sup>Department of Biomedical Sciences, Center for Creative Biomedical Scientists at Chonnam National University, Jeollanam-do, South Korea

The pineal gland maintains the circadian rhythm in the body by secreting the hormone melatonin. Alzheimer's disease (AD) is the most common neurodegenerative disease. Pineal gland impairment in AD is widely observed, but no study to date has analyzed the transcriptome in the pineal glands of AD. To establish resources for the study on pineal gland dysfunction in AD, we performed a transcriptome analysis of the pineal glands of AD model mice and compared them to those of wild type mice. We identified the global change of diverse protein-coding RNAs, which are implicated in the alteration in cellular transport, protein transport, protein folding, collagen expression, histone dosage, and the electron transfer system. We also discovered various dysregulated long noncoding RNAs and circular RNAs in the pineal glands of mice with AD. This study showed that the expression of diverse RNAs with important functional implications in AD was changed in the pineal gland of the AD mouse model. The analyzed data reported in this study will be an important resource for future studies to elucidate the altered physiology of the pineal gland in AD.

### Edited by:

Takeshi Sakurai, University of Tsukuba, Japan

### Reviewed by:

Christophe P. Ribelayga, University of Texas Health Science Center at Houston, United States Jinju Han, Korea Advanced Institute of Science & Technology (KAIST), South Korea

### \*Correspondence:

Young-Kook Kim ykk@jnu.ac.kr Juhyun Song juhyunsong@chonnam.ac.kr

Received: 18 September 2019 Accepted: 13 December 2019 Published: 09 January 2020

### Citation:

Nam KI, Yoon G, Kim Y-K and Song J (2020) Transcriptome Analysis of Pineal Glands in the Mouse Model of Alzheimer's Disease. Front. Mol. Neurosci. 12:318. doi: 10.3389/fnmol.2019.00318 Keywords: pineal gland, Alzheimer's disease, long noncoding RNA, circular RNA, RNA sequencing

# INTRODUCTION

Alzheimer's disease (AD), the most common neurodegenerative disease worldwide, is characterized by progressively impaired cognition, an excessive accumulation of amyloid-β (Aβ), and abnormally hyperphosphorylated tau in the brain (Jack et al., 2013; Song, 2019). It was reported that AD patients showed calcification of the pineal gland and reduced melatonin levels in the serum and cerebrospinal fluid (Skene and Swaab, 2003). Reduced pineal gland volume was also observed in AD patients (Mahlberg et al., 2008). Moreover, a decreased melatonin level is correlated with cognitive impairment (Srinivasan et al., 2010; Rosales-Corral et al., 2012). Although these studies showed causality, however, there is still controversy on the role of pineal gland and melatonin in AD because no clear mechanism has yet been identified.

The pineal gland, which contains neuroglial cells and predominantly pinealocytes, is one of the central organs that regulate the circadian system and is innervated by a neural synaptic pathway originating in the suprachiasmatic nucleus within the hypothalamus (Simonneaux and Ribelayga, 2003). The pineal gland also acts as a modulator in the sexual maturation and aging process as well as sleep disturbance (Bumb et al., 2014). Melatonin is synthesized and secreted in the pineal gland, and its secretory capacity is significantly proportional to pineal parenchymal volume (Nölte et al., 2009). Melatonin contributes to improved hippocampal neurogenesis in AD (Sarlak et al., 2013) and protects neurons against death induced by oxidative stress (Shukla et al., 2017) and Aβ toxicity (He et al., 2010). Pineal dysfunction and reduced melatonin levels are directly related to the pathological progression of AD (Wu et al., 2003). The reduced volume and calcification of the pineal gland influences its function; ultimately, it is strongly associated with the diverse neuropathology of AD patients. However, the mechanisms connecting pineal gland dysfunction and AD pathologies are not fully understood. Thus, more detailed analyses on the molecular level are required to identify the relationship between the pineal gland and AD.

Current annotation of the human genome shows that approximately 90% of the human genome is transcribed, 3% of the genome comprises protein-coding genes, and the rest is noncoding RNA (Harrow et al., 2006). Noncoding RNAs modulate gene expression and are divided into two subclasses according to their length. Long noncoding RNAs (lncRNAs; >200 nt in general) are of special issue owing to their large numbers and the possibility that they are functionally crucial components of the genome (Tsai et al., 2010). Many lncRNAs are associated with epigenetic processes affecting gene expression. The expression of lncRNAs could be regulated by the transsynaptic/cAMP system that controls the expression of hundreds of protein-coding genes in the pineal gland (Coon et al., 2012). CircRNA as a different class from other noncoding RNAs is produced by the back-splicing of a single-stranded linear transcript. In the central nervous system, some circRNAs have a regulatory role in synaptic plasticity induction in neurons (You et al., 2015). Although many noncoding RNAs are linked with the neurological function and identified in the pineal gland, the roles of noncoding RNAs in the AD pineal gland has not been investigated until now.

Here we analyzed diverse RNAs altered in the pineal gland of AD and investigated the function of those RNAs related to the pineal gland in the AD brain. We expect that our analysis of the identification and functional analysis of the transcriptome in the AD pineal gland may be an important resource to solve the neurological interaction between the pineal gland and AD pathogenesis.

# MATERIALS AND METHODS

### Sample Preparation

We obtained male 5xFAD transgenic mice [B6.Cg-Tg (APPSwFlLon, PS1∗M146L∗L286V) 6799Vas/Mmjax; JAX MMRRC stock number: 34848] from The Jackson Laboratory (Bar Harbor, ME, USA). Aβ<sup>42</sup> production was identified in the entire brain at 2 months. Wild-type male mice (C57BL/6) were purchased from Koatech (Pyeongtaek, South Korea). For pineal gland harvesting, the mice with 5 months of age were sacrificed under ether anesthesia. We took samples from all mice at the same time of the day [Zeitgeber time (ZT) 0.5] under conditions of 12 h in light and 12 h in dark. There was no noticeable difference in the phenotype between the pineal glands of wild type and 5xFAD mice. The pineal glands from five to seven mice were pooled in each sample, and four samples in each of wild type and 5xFAD group were prepared for the analysis. The experiment was performed in accordance with the recommendations of ''96 Guidance for Animal Experiments'' established by the Animal Ethics Committee at Chonnam National University. The experimental protocols were approved by the Animal Ethics Committee at Chonnam National University.

### RNA Sequencing

Total RNA from the pineal gland was extracted using TRIzol reagent (Thermo Fisher), and their integrity was checked using the Agilent 2100 BioAnalyzer (Agilent). Total RNA was treated with a Ribo-Zero Gold rRNA Removal Kit (Illumina) to remove ribosomal RNAs, and RNA sequencing libraries were constructed using a TruSeq Stranded Total RNA Kit (Illumina). The libraries were paired-end sequenced with 100 sequencing cycles on a HiSeq 2500 system (Illumina).

### Analysis of RNA Sequencing Data

By using FastQC, we assessed the quality of our sequencing data. When the median value of per base sequence quality was calculated from all sequencing libraries, the quality was shown to be very high across all nucleotides. The overall quality was satisfactory in all libraries as calculated from the high percentage of nucleotides with a quality score above 30 (**Table 1**).

The sequencing reads with low quality were trimmed using the Trimmomatic (Bolger et al., 2014). To quantify the expression levels of mRNAs and lncRNAs, we used two independent analysis pipelines and intersected the results as shown in our previous article (Yoon et al., 2019). In the first approach, the trimmed sequences were aligned into the mouse genome (mm10) using the STAR aligner (Dobin et al., 2013), while the Cuffnorm was used to calculate normalized values of fragments per kilobase of transcript per million mapped reads (FPKM) based on GENCODE annotation (Release M17, GRCm38.p6; Harrow et al., 2006; Trapnell et al., 2012). Those transcripts with average FPKM values less than 1 or those not detected in any sample were removed from the further analyses. To select the transcripts with a significantly different expression between the wild type and 5xFAD groups, the t-test was used. In the second approach, the Salmon tool was used to quantify expression levels of transcripts (Patro et al., 2017), while the edgeR package (Robinson et al., 2010) was used to select the transcripts with significantly differing expressions between wild type and 5xFAD groups. The results from these two pipelines were intersected, and only those transcripts with significantly different expressions in both approaches were selected for further analyses.

The unsupervised hierarchical clustering was performed with Cluster 3.0 (de Hoon et al., 2004) and the results were visualized using Java Treeview (Saldanha, 2004). For the clustering, FPKM values were log-transformed, and the genes and arrays were median-centered and normalized. Complete linkage analysis was used to examine hierarchical clustering using the centered-correlation method. Those mRNAs and lncRNAs with differential expressions between the wild type and 5xFAD groups as selected from each volcano plot were used for the clustering.

### Functional Analysis of mRNAs

For gene ontology (GO) enrichment analysis, we sorted the genes based on their expression differences between wild type and 5xFAD groups. Among them, the top 10% genes with higher expressions in the pineal gland from the 5xFAD group compared to the wild type group were selected for the GO analysis of the Molecular Signatures Database (Liberzon et al., 2011; The Gene Ontology Consortium, 2017). For the same group of genes, functional annotation clustering was performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID) tool (Huang da et al., 2009).

### Analysis of circRNA Expression

To detect the sequencing reads containing the back-splicing junction, the DCC algorithm was used (Cheng et al., 2016). Among the results, the exonic circRNAs, which are composed of only exonic sequences of host genes, were selected. The expression count for each circRNA was normalized by the count of total circRNA reads of each sample. To select circRNAs with differential expression between the wild type and 5xFAD groups, only circRNAs with average read numbers greater than 5 were used. To construct the regulatory network between circRNAs and miRNAs, we further selected the circRNAs with fold changes between the wild type and 5xFAD group of more than 2 and those with normalized expression counts higher than 10 among the differentially expressed circRNAs selected above.

### Prediction of miRNAs Targets

The TargetScan algorithm was used to predict the miRNA binding sites in the circRNA sequences (Agarwal et al., 2015). Among the miRNAs discovered in mice, we only used those that were ''highly conserved'' or ''conserved'' according to the annotation in the TargetScan database. After running the TargetScan to predict the miRNA binding sites in circRNAs, we only selected the targets with an ''8mer site'' or a ''7mer-m8 site.'' To draw the miRNA-circRNA interaction network, only the miRNAs with more than two binding sites in the circRNAs were used.

# Verification of the Circular Structure of circRNAs

Total RNA was treated with RNase R (Epicentre) for 20 min at 37◦C to remove linear RNAs, and the enzyme was inactivated at 95◦C for 3 min. The remaining RNAs were reverse-transcribed and amplified by PCR. The sequences of the PCR product were identified by Sanger sequencing to confirm the expected back-splice junctions of circRNAs. The list of PCR primers is included in **Supplementary Table S4**.

# RESULTS

### Transcriptome Analysis of Pineal Gland From Mouse AD Model

The pineal gland produces melatonin, which has an important role in maintaining circadian rhythm. In addition, melatonin protects the neurons in AD by working as an antioxidant. However, no study to date has analyzed the transcriptome profile of the pineal gland from AD in part due to the difficulty in obtaining a sufficient amount of sample. To identify the proteincoding or noncoding RNAs associated with the physiology of the pineal gland of AD, we prepared the pineal gland from wild type and 5xFAD mice, a model of AD that accumulates high levels of Aβ42. We performed RNA sequencing for the total RNA and selected only the high-quality sequencing reads using Trimmomatic (**Figure 1A**; see the ''Materials and Methods'' section; Bolger et al., 2014).

In our previous study, we combined the results from two different pipelines for RNA sequencing analysis to increase the reliability of the expression measurement (**Figure 1A**; see the ''Materials and Methods'' section; Yoon et al., 2019). In the first pipeline, the sequences from each sample were aligned to the mouse genome by STAR, while the expression level of each gene was measured by Cuffnorm (Trapnell et al., 2012; Dobin et al., 2013). Then, the expression difference between the wild type and 5xFAD groups was calculated. In the second method, the expression of each transcript was quantified by Salmon, while the difference in the expression level was calculated using edgeR (Robinson et al., 2010; Patro et al., 2017). The results from the two approaches were compared (**Supplementary Table S1**) and only the genes that were changed in both analyses were used in further studies (**Figure 1**).


Alzheimer's disease (AD). (A) Analysis scheme. The pineal glands were separated from the wild type and 5xFAD mice, and total RNAs were extracted for RNA sequencing analysis. The sequencing data were analyzed by two different pipelines, STAR-Cuffnorm and Salmon-edgeR, to identify differentially expressed mRNAs and lncRNAs. To quantify the circRNAs, chimeric reads were detected by STAR and circRNAs were measured by DCC. See the "Materials and Methods" section for details. (B) Volcano plot for protein-coding mRNAs. Those genes with P-value < 0.05 and those with a change greater than 2-fold are indicated by colored dots. (C) Clustering for protein-encoding mRNAs. For the selected genes from (B), unsupervised hierarchical clustering was performed. The color bar represents the expression differences between the wild type and 5xFAD samples. Note that each group clustered together properly. (D) The expression level change of genes related to the circadian rhythm. P-value was calculated using a one-tailed t-test (<sup>∗</sup> < 0.05).

# Analysis of mRNA Changes in the Pineal Gland With AD

For the analysis of protein-encoding mRNAs, we first selected the mRNAs whose expressions differed between the wild type and 5xFAD groups. There were 81 genes with significantly increased expression and 31 genes with significantly decreased expression (**Figure 1B**). The unsupervised hierarchical clustering for those genes showed clear separation of the sample into the wild type and 5xFAD groups, respectively (**Figure 1C**).

Because the pineal gland is the central organ governing circadian rhythm, we investigated the expressions of the genes related to this process. Among the genes, we noticed a lower expression of period circadian clock 3 (Per3) and a higher expression of cryptochrome 1 (Cry1) in the pineal glands of 5xFAD mice compared to those of wild type mice (**Figure 1D**). It was reported that Per3 and Cry1 regulated circadian feedback loops to generate 24-h oscillations (Reppert and Weaver, 2002). These clock genes are expressed in the pineal gland and provide information to other central and peripheral structures (Fukuhara et al., 2000; Takekida et al., 2000; Yamazaki et al., 2000). Therefore, there might be an alteration in the circadian regulation in the pineal gland of AD.

To identify the cellular pathways affected in the pineal gland of AD, we first performed the GO analysis for increased genes in the 5xFAD vs. wild type group (The Gene Ontology Consortium, 2017). The most significantly enriched terms included those related to cellular transport and protein localization (**Figure 2A**). Many studies suggested that the process related to protein sorting is defective in the neurons of AD (Small and Gandy, 2006). Interestingly, the expression of myosin genes was globally decreased in the 5xFAD group compared to the wild type group (**Supplementary Figure S1A**). Thus, the processes related to the movement of cellular molecules including proteins might also be dysregulated in the pineal gland of AD. We also performed a functional analysis of increased genes using the DAVID functional annotation tool (Huang da et al., 2009). Interestingly, all three of the most highly enriched clusters were related to the cellular pathway of protein folding (**Figure 2B**). In diverse human diseases of the brain including AD and Parkinson's disease, misfolded proteins are commonly observed (Chiti and Dobson, 2017). In the brain of AD, Aβ is highly accumulated, leading to a problem in the protein folding process within the neurons. Therefore, a similar defect in protein folding occurs in the pineal gland of AD. We also noticed the global decrease in collagen genes (**Figure 2C**). Because collagen has a protective role for neurons, its decreased expression might affect the impaired function of the pineal gland in patients with AD (Cheng et al., 2009).

Analysis of histone gene expression revealed that the expression of overall histone genes is decreased in the pineal gland of 5xFAD mice compared to wild type mice (**Supplementary Figure S1B**). Interestingly, the FPKM values of most decreased genes were low, while those of some increased genes were relatively higher. This expression change resulted in a non-meaningful difference overall when we considered the expression level of each histone group by combining the FPKM for each histone class. Thus, there was no difference in the total amount of most histone classes, although there was a redistribution on histone expression in each gene class. However, the overall transcripts dosage of histone H3 class decreased significantly in the 5xFAD group compared to the wild type group (**Figure 2D**). Histone influences overall DNA metabolism, such as DNA repair (Singh et al., 2009), and histone dosage controls DNA damage sensitivity in a checkpoint-independent

gland from 5xFAD vs. wild type mice. The top 15 GO terms based on the false discovery rate (FDR) q-value are shown. (B) Database for Annotation, Visualization and Integrated Discovery analysis for increased protein-coding mRNAs. The top three enriched clusters are visualized. (C) Expression changes in the collagen genes are shown. (D) Expression of histone gene groups. The sum of fragments per kilobase of transcript per million mapped reads value for the members of each histone group was calculated. A one-tailed t-test was applied to calculate the P-value. (E) Expression changes for the members of the electron transfer chain in mitochondria. The expression changes for members from each mitochondria complex between the pineal glands from the wild type and 5xFAD groups are shown.

manner in cells (Liang et al., 2012). Therefore, the dysregulation of histone dosage may have a role in the cellular physiology of the pineal gland of AD.

Strikingly, the expression of nearly all components of the electron transfer chain increased in the 5xFAD group compared to the wild type group (**Figure 2E**). The electron transfer chain creates a proton gradient across the inner membranes of mitochondria, driving the synthesis of adenosine triphosphate (ATP). One study showed that ATP is involved in the inhibition of melatonin synthesis in the rat pineal gland (Souza-Teodoro et al., 2016). Thus, it would be interesting to study whether the increased expression of electron transfer chain complexes might affect the physiology of AD pineal glands.

### Analysis of lncRNAs Change in the Mouse Pineal Gland With AD

Although lncRNAs are involved in diverse aspects of cellular processes, no study to date has profiled the lncRNAs in the pineal glands of AD. To identify lncRNAs that are dysregulated in the pineal gland of AD, we analyzed the expression level of lncRNAs from the RNA sequencing data produced above. Diverse types of lncRNA genes exist in the genome as annotated in GENCODE (Harrow et al., 2006). Among those classes, we selected long intergenic noncoding RNA (lincRNA), antisense RNA, the processed transcript, and the bidirectional promoter lncRNA for further analysis (**Figure 3A**) because these classes are more likely to be transcribed as RNA transcripts than other classes such as pseudogenes. Among them, we selected the lncRNAs with a significant difference in their expression between the pineal glands of the wild type and 5xFAD groups (**Figure 3B**, **Supplementary Table S2**). The unsupervised clustering analysis for those selected lncRNAs showed a clear separation of samples into the wild type and 5xFAD groups (**Figure 3C**).

Among the differentially expressed lncRNAs, we selected the lncRNAs with highly significant expression changes (**Figure 3D**). Because the genes closely located in the genomic location have a higher tendency to be influenced by a common signaling pathway (Lee and Sonnhammer, 2003; Michalak, 2008), we investigated the protein-coding gene with a known function related to neuronal processes near each lncRNA. For example, the most highly increased lncRNA, RP23-47P3.3, and the most highly decreased lncRNA, RP23-81C12.3, have no neighboring protein-encoding genes in their vicinity. However, RP23-479J7.2, which we confirmed its expression change (**Supplementary Figure S2A**) had its neighboring gene, Cops3, which resided in the genomic locus commonly deleted in patients with Smith-Magenis syndrome (**Figure 3E**; Potocki et al., 2000). Thus, it is possible that they are regulated under the common signaling pathway.

### Analysis of circRNAs Change in the Mouse Pineal Gland With AD

CircRNAs are emerging as important regulators of cellular processes, mainly through the suppression of miRNA function. No study examined the circRNAs in any aspect of the pineal gland. By analyzing the back-splicing junction from the RNA sequencing reads based on the DCC algorithm, we calculated the expression level of circRNAs in the pineal glands of the wild type vs. 5xFAD mice (**Figure 1A**; see the ''Materials and Methods'' section; Cheng et al., 2016). We identified a total of 744 circRNAs expressed in our samples (**Supplementary Table S3**), and some circRNAs showed high expression levels (**Figure 4A**). The circRNA produced from the locus of the Nfkb1 gene showed the highest expression level. When we analyzed the number of circRNAs produced from each gene locus, we found that most host genes produced only a single type of circRNA, although some produced diverse circRNAs with a different combination of host gene exons (**Figure 4B**). Moreover, one to six exons were combined to compose circRNAs with three exons as the highest frequency (**Figure 4C**). In our previous analysis of the circRNAs in the brain cortex, the combination of three exons was also the most frequent choice for the production of circRNA (Yoon et al., 2019). Therefore, there may be a similar principle to produce circRNA by combining exons between the brain cortex and the pineal gland.

We selected the circRNAs with significant expression differences between the pineal glands of wild type and 5xFAD mice (**Figure 4D**). We also confirmed that the samples that we analyzed were clustered properly from unsupervised hierarchical clustering (**Figure 4E**). Among the differentially expressed circRNAs, we further selected 11 circRNAs whose expression differences between the wild type and 5xFAD group were more significant or whose expression levels were high in the pineal gland (see the ''Materials and Methods'' section; **Figure 4F**). To predict the miRNAs that could be under the control of these circRNAs, we utilized the TargetScan algorithm to predict the miRNA binding sites in the circRNA sequences (Agarwal et al., 2015). Indeed, there were many miRNA binding sites predicted in the circ-RNAs sequence (**Figure 4G**). Therefore, the altered expression of these circRNAs in the pineal gland of AD may exert their roles by suppressing the predicted miRNAs. We confirmed the circular structure and expression changes of five randomly selected circRNAs (**Figure 4H**, **Supplementary Figure S2**). All of the selected circRNAs had a circular structure in the cells as confirmed by their resistance to RNase R, an enzyme that degrades linear RNAs, and their back-splicing junction prediction was verified by Sanger sequencing.

# DISCUSSION

In mammals, the sympathetic innervation of the pineal gland contributes to the transduction of environmental light information into a hormonal signal such as melatonin. A previous study analyzed the light-induced transcriptome of the pineal gland of zebrafish and showed that miR-183 affects the mRNA level of aanat2, the key enzyme in melatonin synthesis (Ben-Moshe et al., 2014). It was also reported that miR-483 regulated the synthesis of melatonin by inhibiting Aanat in the pineal glands of rats (Clokie et al., 2012). Another study identified lncRNAs that are differentially expressed between day and night in the pineal glands of rats (Coon et al., 2012). However, no study has analyzed the transcriptome in the pineal gland of AD. Therefore, here we scrutinized

GENCODE annotation. The number of identified lncRNAs for each lncRNA class is shown in parentheses. (B) Volcano plot for lncRNAs. Those genes with P-value < 0.1 and those with expression changes >50% in the 5xFAD group compared to the wild type group are indicated with colored dots. (C) Clustering for lncRNAs. For the selected lncRNAs from (B), unsupervised hierarchical clustering was performed. The color bar representing the expression difference between the wild type and 5xFAD groups is shown. (D) List of differentially expressed lncRNAs between the pineal gland from the wild type and 5xFAD groups. (E) Genomic information near the locus of lncRNA RP23-479J7.2 and its near protein-coding genes, Flcn and Cops3. The genomic information data were downloaded from the Genome Browser (Kent et al., 2002).

the differential expression of diverse RNA types including protein-encoding RNAs, lncRNAs, and circRNAs in the pineal gland of AD.

The circadian rhythm is regulated by the oscillations of the transcriptional-translational negative feedback system. During sleep, brain and muscle ARNT-like-1 (BMAL1)

FIGURE 4 | Analysis of circRNA changes in the mouse pineal gland with AD. (A) Expression counts of 15 highly expressed circRNAs in the pineal glands. The average read counts among the wild type and 5xFAD samples were calculated. (B) Distribution of the number of circRNAs produced from each host gene. (C) Distribution of the number of exons used to produce each circRNA. (D) Volcano plot for circRNAs. Those genes with values of P < 0.1 and those with expression changes >50% in the 5xFAD group compared to the wild type group are indicated. (E) Clustering of circRNAs. For the selected circRNAs from (D), unsupervised hierarchical clustering was performed. The color bar representing the expression difference between wild type and 5xFAD is shown. (F) Differentially expressed circRNAs between the pineal gland from the wild type and 5xFAD groups. A one-tailed t-test was applied to calculate the P-value (<sup>∗</sup> < 0.05, ∗∗ < 0.01). (G) The circRNAs-miRNAs regulatory network. The regulatory relationship was predicted by the existence of a miRNA binding site in the circRNA sequence. (H) Confirmation of the circular structure of circRNAs by RNase R treatment. The data in Supplementary Figure S2B were used for the quantitation. The white bars indicate circRNAs while the filled bars are linear RNAs. Expression change of each circRNA between untreated and RNase R-treated samples was calculated. P-value was calculated with a two-tailed t-test.

heterodimerizes with circadian locomotor output cycles gone kaput (CLOCK), and the CLOCK/BMAL1 heterodimer regulates the transcription of period (Per) and cryptochrome (Cry; Ko and Takahashi, 2006). Per loss has been known to regulate the transcriptional activity of CLOCK (Hardin and Panda, 2013) and accelerates cell death and motor dysfunction (Krishnan et al., 2012). The loss of Per3 protein in particular among the period genes inhibits the Per1/2 stabilization and induces circadian rhythm disruptions (Zhang et al., 2016). In addition, cryptochrome (Cry1/2) suppresses CLOCK and BMAL1 expressions (Reppert and Weaver, 2002). The rhythm of clock gene expression in the pineal gland is aberrantly regulated in AD patients; subsequently, the aberrant circadian regulation leads to memory deficits in AD in relation to the AD development (Coogan et al., 2013). In our data, the lower Per3 expression and higher Cry1 expression might lead to the aberrant circadian rhythm based on this previous evidence.

In our analysis, the expression of most components of the electron transfer chain increased in the pineal gland of AD. These proteins comprise the main complexes in the mitochondria to produce the cellular energy source, ATP, a known co-transmitter of noradrenaline in the pineal gland (Mortani Barbosa et al., 2000) and boost β1-adrenergic-induced N-acetylserotonin synthesis by triggering P2Y1 receptors (Ferreira and Markus, 2001). Recent studies demonstrated that ATP suppressed melatonin synthesis by the pineal gland (Souza-Teodoro et al., 2016). Furthermore, ATP is one of the damage-associated molecular pattern molecules (Di Virgilio, 2007), and a high amount of it is released under stress conditions (Carta et al., 2015). Thus, the increased expression of the members of the electron transfer system in 5xFAD mice may increase ATP production and can result in pineal gland impairment.

Collagens comprise a large family of proteins involved in a variety of functions ranging from the formation of fibrillary networks of the extracellular matrix to synapse formation. One study revealed that the increase in collagen VI expression in the brains of hAPP mice is a neuroprotective response (Cheng et al., 2009). Other collagen genes provide guidance for axon outgrowth (Taylor et al., 2018), help motor neurons to innervate (Kowa et al., 2004), and regulate Aβ peptide formation during the progression of AD (Osada et al., 2005). Because of these important roles of collagen in neurons, we assume that the loss of collagen in the AD pineal gland is linked to pineal gland dysfunction.

No study to date has analyzed the transcriptome of the pineal gland in an AD model. In this study, we identified diverse lncRNAs and circRNAs that were differentially expressed in the pineal gland in the 5xFAD group vs. the wild type group (**Figures 3**, **4**). Because many lncRNAs work by regulating their near protein-encoding genes, we investigated the genomic locus of lncRNAs and searched the genes near them. Among the lncRNAs, RP23-479J7.2 had a near proteincoding gene Cops3, whose genomic locus is commonly deleted in patients with Smith-Magenis syndrome (Potocki et al., 2000). Smith-Magenis syndrome is a syndrome that features mental retardation and sleep disturbances as well as problematic daytime behavior (De Leersnyder, 2006). This problem was linked to abnormal circadian secretion of melatonin. Thus, a future study must elucidate the detailed role of RP23-479J7.2 in this disease.

We also identified many altered circRNAs in the pineal gland of an AD model. Because the circRNAs mainly work by binding miRNAs and suppressing their function, we predicted the miRNAs with binding sites in circRNA sequences (**Figure 4G**). Among the identified miRNA-circRNA pairs, circMboat2 and circNlrp5-ps were predicted to be regulated by miR-483, which was shown to suppress Aanat, the enzyme critical for melatonin synthesis (Clokie et al., 2012). Therefore, these circRNAs may be involved in the regulatory network of pineal gland function such as melatonin synthesis. Further studies including the profiling of miRNAs are essential to identify the role and working mechanism of noncoding RNAs in the pineal gland of AD.

This is the first study to profile the transcriptome of the pineal gland from the AD model. We expect that our study will be helpful for the researchers interested in the gene expression change of the pineal gland with AD. Although we identified diverse RNAs differentially expressed between the pineal glands of wild type and 5xFAD mice, it needs to be noted that the experiment was performed at a 1-time point (ZT0.5). Therefore, there is a possibility that the gene expression changes could be the result of a shift in the circadian rhythm of the AD model. In a future study, large-scale profiling of the transcriptomes with various time points of the day will be required for a more comprehensive understanding of the change in the AD pineal gland.

# DATA AVAILABILITY STATEMENT

The raw sequencing data and processed data with FPKM values are available through the Gene Expression Omnibus database under accession number GSE129586 (Barrett et al., 2013).

# ETHICS STATEMENT

The animal study was reviewed and approved by Animal Ethics Committee at Chonnam National University.

# AUTHOR CONTRIBUTIONS

KN and JS prepared the samples for the analysis. Y-KK performed the bioinformatics analyses. GY confirmed the expression of the selected genes by biochemical analysis. JS and Y-KK wrote the manuscript.

# FUNDING

This study was supported by grants from the Basic Science Research Program through the National Research Foundation of Korea (NRF; NRF-2019R1F1A1054111 to JS, NRF-2018R1A2B6001104 and NRF-2019R1A4A1028534 to Y-KK).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol.2019.003 18/full#supplementary-material.

FIGURE S1 | Expression of protein-coding genes of the selected gene groups. (A) Expression changes of the myosin genes. (B) Expression changes of the histone genes. A one-tailed t-test was applied to calculate the P-value (<sup>∗</sup> < 0.05, ∗∗ < 0.01, ∗∗∗ < 0.001).

FIGURE S2 | Measurement of lncRNA and circRNAs level and confirmation of the circular RNA structure. (A) Measurement of lncRNA and circRNAs by PCR.

### REFERENCES


The expression change of the lncRNA RP23-479J7.2 in Figure 3E and five randomly selected circRNAs in Figure 4F were measured. P-values were calculated by a two-tailed t-test. (B) Confirmation of the circular structure of circRNAs by RNase R treatment. This data was used for the quantitation in Figure 4H.


melatonin synthesis in the rat pineal gland. J. Pineal Res. 60, 242–249. doi: 10.1111/jpi.12309


**Conflict of Interest**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Nam, Yoon, Kim and Song. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Molecular Mechanisms of REM Sleep

Rikuhiro G. Yamada<sup>1</sup> and Hiroki R. Ueda1,2 \*

<sup>1</sup> Laboratory for Synthetic Biology, RIKEN Center for Biosystems Dynamics Research, Osaka, Japan, <sup>2</sup> Department of Systems Pharmacology, Graduate School of Medicine, The University of Tokyo, Tokyo, Japan

Rapid-eye movement (REM) sleep is a paradoxical sleep state characterized by brain activity similar to wakefulness, rapid-eye-movement, and lack of muscle tone. REM sleep is a fundamental brain function, evolutionary conserved across species, including human, mouse, bird, and even reptiles. The physiological importance of REM sleep is highlighted by severe sleep disorders incurred by a failure in REM sleep regulation. Despite the intense interest in the mechanism of REM sleep regulation, the molecular machinery is largely left to be investigated. In models of REM sleep regulation, acetylcholine has been a pivotal component. However, even newly emerged techniques such as pharmacogenetics and optogenetics have not fully clarified the function of acetylcholine either at the cellular level or neural-circuit level. Recently, we discovered that the G<sup>q</sup> type muscarinic acetylcholine receptor genes, Chrm1 and Chrm3, are essential for REM sleep. In this review, we develop the perspective of current knowledge on REM sleep from a molecular viewpoint. This should be a starting point to clarify the molecular and cellular machinery underlying REM sleep regulation and will provide insights to explore physiological functions of REM sleep and its pathological roles in REM-sleep-related disorders such as depression, PTSD, and neurodegenerative diseases.

### Edited by:

Michael Lazarus, University of Tsukuba, Japan

### Reviewed by:

Akihiro Yamanaka, Nagoya University, Japan Hiroaki Norimoto, Max-Planck-Institut für Hirnforschung, Germany

> \*Correspondence: Hiroki R. Ueda uedah-tky@umin.ac.jp

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 03 September 2019 Accepted: 12 December 2019 Published: 14 January 2020

### Citation:

Yamada RG and Ueda HR (2020) Molecular Mechanisms of REM Sleep. Front. Neurosci. 13:1402. doi: 10.3389/fnins.2019.01402 Keywords: REM sleep, theta oscillation, hippocampus, bursting, muscarinic acetylcholine receptors

# INTRODUCTION

Rapid-eye movement (REM) sleep is a prominent brain state which is accompanied with multiple features such as random movements of eyes, vivid dreaming, quiet muscle tone, lessened homeostatic regulation of body (e.g., temperature, heart rate, etc.), and brain activity marked by the enhancement of specific brain oscillation. As indicated by the multiple features of REM sleep, the mechanisms of sleep, in general, involve multiple regulatory components at different layers of scales ranging from molecule level to organism level. The brain oscillation is an electrophysiological activity widely used to define stages of sleep. The identification of the brain oscillation associated with specific sleep stages has enabled researchers to untangle such intermingled systems in consideration of neural activity of the brain. There are largely three classes of oscillations in the brain of sleeping mice; slow-wave oscillation (SWO, <1 Hz), delta oscillation (1–4 Hz), and theta oscillation (4–10 or 12 Hz; this range slightly varies depending on literature). The SWO and delta oscillation are characteristic to non-REM (NREM) sleep, and the theta oscillation is characteristic to REM sleep. The early investigations into the SWO and delta oscillations provided the basis for current understanding of molecular and cellular mechanisms of NREM sleep. In contrast to the relatively well-described mechanisms of NREM sleep, that of REM sleep has been left unclear. Looking back some historical milestones of sleep research is helpful to grasp how the

brain oscillations became established as definitive features of sleep stages and contributed to the current understanding of the mechanisms of NREM sleep. This retrospective detour into NREM sleep studies should be beneficial to make extrapolations on how the brain oscillations provide insights into the mechanisms of REM sleep.

The first measurement of the electrical activity of brain dates back to 1875 when the pioneer electrophysiologist; Caton (1875) reported that the electrical current in a cortical region varies depending on the functional activity of the region using dog, rabbit, and monkeys. Notably, he already described that the suspension of functional activity increased the electrical current, and the current diminished when the cortical area was functionally active (Cohen of Birkenhead, 1959). His notion is consistent with today's knowledge that the sleeping cortex shows the relatively high amplitude of SWO and delta oscillation while the waking brain shows low amplitude fast oscillation. Without knowing Caton's work, in 1890, Beck (1890) observed spontaneous rhythms of electrical brain activity and that the rhythm changes upon sensory stimulation using frogs, dogs, and rabbits. In 1910s, W. Práwdicz-Neminski identified the distinguishable patterns of the spontaneous rhythms and referred them as A-waves and B-waves using dogs (Coenen and Zayachkivska, 2013). In 1920s, the first electrical recording of a human brain was made by Berger (1929), who called his method "electro encephalogram (EEG)." He also observed distinct patterns of spontaneous rhythms which consisted of slow and fast oscillations. The slow and fast oscillations are around 10 and 30 Hz, respectively. Berger (1929) referred those oscillations as alpha and beta waves and reported that the alpha wave was replaced with the beta wave in response to physiological stimuli. Importantly, his extensive investigation of brain oscillations in various pathological and pharmacological conditions revealed that the slower alpha wave, which is referred as delta oscillation in today's term, appeared when the subject was unconscious under anesthesia or epilepsy (Walter, 1938). The term of "delta wave" was first introduced by Walter in 1936 to describe slow oscillations produced in a cortical region adjacent to cerebral tumors when he studied the location of tumors by using EEG signal. Later, Loomis et al. (1937) used the term (delta wave) to describe the slow oscillations discovered in natural sleep of human and many other conditions (Walter, 1938). Loomis et al. (1937) built a large recording drum of EEG to observe a human subject continuously throughout a night, and they discovered several distinct stages of sleep and characteristic brain oscillations (Davis et al., 1937; Loomis et al., 1937). Later, these findings led to the identification of the sleep stage associated with REM and frequent dreaming (Aserinsky and Kleitman, 1953), and to the formulation of an objective method for studying sleep (Dement and Kleitman, 1957a,b). Since those seminal studies, EEG signal has been the definitive information for staging sleep. However, the understanding of the neuronal and molecular basis that underlie the characteristic EEG signals had to wait for the works of Steriade that attributed the high amplitude and lowfrequency EEG oscillations, that is SWO and delta oscillation, to the synchronized activity of the population of neocortical and thalamocortical neurons (Steriade et al., 1993a,c,d).

Importantly, all of the three brain oscillations, SWO, delta oscillation, and theta oscillation, are originated from synchronized activity of a population of neurons. The cellular property shared among synchronous neurons is the bimodality of membrane potentials. The alternating sequence of the active state (UP state) and the silent state (DOWN state) rises to the brain oscillations. The UP state is associated with vigorous firings due to the depolarized membrane potential and the DOWN state associated with the ceasing of firing due to the hyperpolarized membrane potential (McCormick et al., 2015). Some early studies suggested that the essence of the alternating sequence is the interaction between the relatively slow Ca2<sup>+</sup> oscillation and the fast action potentials, and suggested ion channels involved in the slow oscillation and action potentials (Jahnsen and Llinas, 1984; Steriade et al., 1993b; McCormick and Bal, 1997). Following the description of the SWO and the delta oscillation in the neocortex and the thalamus which underly the NREM sleep, the identification of involved brain regions and their brain-wide neural circuits have rapidly advanced, and we see further advancement in that direction thanks to the recent advent of innovative techniques such as opto- and pharmacogenetics (Weber and Dan, 2016; Saper and Fuller, 2017; Scammell et al., 2017).

While the neural circuits are relatively well described, the molecular and cellular properties essential to sleep regulation have been less investigated in the last decades. Moreover, the function of even basic neurotransmitters in regulating the cellular properties for sleep, especially the necessity of acetylcholine for REM sleep has been controversial for decades. Early studies implied the importance of acetylcholine for REM sleep by demonstrating that the injection of cholinergic agonists into the brainstem induced REM sleep-like state (Cordeau et al., 1963; George et al., 1964). Also, acetylcholine release was found to be abundant in the brainstem during REM sleep (Kodama et al., 1990; Leonard and Lydic, 1997). Therefore, a long-standing hypothetical model of the transition between NREM sleep and REM sleep incorporated acetylcholine as the key factor (Hobson et al., 1975; McCarley and Hobson, 1975; Sakai et al., 2001). However, the necessity of acetylcholine for REM sleep has been elusive, because lesioning of brain regions such as cholinergic neurons in the basal forebrain (BF), the laterodorsal tegmentum (LDT), and the pedunculopontine tegmentum (PPT) in the brainstem results in relatively minor effects (Lu et al., 2006; Blanco-Centurion et al., 2007). Hence, a proposed model for regulating REM sleep incorporates GABAergic and glutamatergic neurons as its core components (Luppi et al., 2013). Although recent opto- and pharmacogenetic approaches consolidated the role of acetylcholine in sleep regulation at neural-circuit level, the necessity of acetylcholine in REM sleep regulation remained unclear (Shi et al., 2015; Xu et al., 2015; Chen et al., 2016; Zant et al., 2016). Despite the controversy over the importance of cholinergic regulation on REM sleep, multiple lines of in vivo pharmacological evidence consistently indicated muscarinic acetylcholine receptors are important for REM sleep regulation. Muscarinic receptor agonists and acetylcholinesterase inhibitors increase REM sleep

and shorten the REM latency (the time-delay of REM start after the NREM start) (Sitaram et al., 1976; Hohagen et al., 1993; Lauriello et al., 1993; Riemann et al., 1994). On the other hand, muscarinic receptor antagonist decreased REM sleep and lengthened the REM latency (Gillin et al., 1991; Rauniar et al., 1998; Kim and Jeong, 1999). Nonetheless, genetic approaches to assess the contribution of muscarinic receptors to sleep-regulation have been limited (Goutagny et al., 2005), leaving the molecular component in the regulation of REM sleep unidentified.

To obtain deeper insights into the molecular mechanism of REM sleep, we need to address two issues: (1) identifying specific molecular components among the family members of acetylcholine receptors and (2) understanding the molecular function in regulating cellular properties of the identified receptors. A recent comprehensive reverse genetic study revealed that the G<sup>q</sup> protein-coupled muscarinic acetylcholine receptors, Chrm1 and Chrm3, are essential for REM sleep, as REM sleep and its associated enrichment of EEG theta oscillation could be hardly detected in Chrm1 and Chrm3 double-knockout (DKO) mice during sleep (Niwa et al., 2018). Also, a series of our studies suggested that the Ca2+-hyperpolarization pathway plays an important role in regulating cellular properties for the synchronized activity for NREM sleep, i.e., for the SWO and the delta oscillation. Because the synchronized activity of population of neurons is also a mechanism that underlies the theta oscillation; a definitive feature of REM sleep, the investigation into the molecular mechanisms involved in the synchronized activity should be a future direction of REM sleep research. In this review, we intend to give a perspective on molecular mechanisms of REM sleep by focusing on the EEG theta oscillation in sleeping mice. First, we summarize the cellular basis of synchronized neurons underling the SWO and the delta oscillations, then we shift our viewpoint to the theta oscillation and look neural circuits involved in generation and regulation of the theta oscillation. We also discuss the current views about the function of Chrm1 and Chrm3 in the theta oscillation, and a potential molecular basis for sleep homeostasis. Previous excellent reviews have extensively described the regulatory neural circuits of REM sleep, the characteristic muscle activities associated with REM sleep such as REM and muscle atonia, the evolutional perspective of REM sleep, or the mechanism and function of SWO in the neural network (Brown et al., 2012; Luppi et al., 2013; Neske, 2015; Miyazaki et al., 2017). Here, we put our focus on the molecular mechanisms of EEG brain oscillations.

### THE CELLULAR MECHANISM OF THE BRAIN OSCILLATIONS

Although any cell types in the brain may contribute to the EEG signal as ionic flows are generated, the primary contributor is the pyramidal neuron in the neocortex. The neocortical pyramidal neurons reside under the skull, aligned in parallel to each other, and have thick dendrite that can form strong dipoles along the somatodendritic axis. The synchronized activation of those layered pyramidal neurons generates the strong extracellular electrical field readily measurable on the skull (Buzsaki et al., 2012). In contrast, the cerebellum which also has the layer structure of giant Purkinje neurons generate very small extracellular fields because the cerebellar activity is mainly local, and the Purkinje cells are not synchronized. Thalamocortical cells, which have more circular morphology emanating dendrites in all directions with relatively equal size compared to pyramidal neurons, can form limited dipoles and their contribution to extracellular fields is small even when their action is highly synchronized. Besides the neocortex, the hippocampus is a major influencer on the EEG signal. The hippocampus has a layered structure, hippocampal pyramidal neurons are densely aligned in parallel, and they act in a synchronized manner to generate strong electrical fields (Colgin, 2013, 2016). Because the hippocampus is located deeper in the brain compared to the neocortex, the detailed recording of the extracellular electrical field requires deep electrodes placed close to the tissue (**Figures 1A,B**). However, the characteristic oscillation at the theta frequency band (4–10 Hz) recorded in EEG is thought to originate from the hippocampus. Supporting evidence includes that the average magnitude of theta power measured by multisite recordings along the hippocampus–neocortex axis monotonically decreased with distance from the hippocampus and that the distribution of theta power on the neocortical surface reflects the physical layout of the underlying hippocampus (Bland and Whishaw, 1976; Sirota et al., 2008).

### Molecular Mechanisms of the Bimodality: UP and DOWN States of the Burst Firing

The synchronized activity of neurons emerges from the bursting of individual neurons. The bursting consists of repeating cycles of the vigorous-firing state (UP state) and the silent state (DOWN state). The cycle can be observed in the neocortex and thalamus where a population of neurons synchronously generate the SWO and delta oscillation, respectively. The transition between UP and DOWN states is marked by a clear change of membrane potential, which traces the bimodality of neurons (**Figure 2**). The depolarized membrane potential underlies the UP state, and the hyperpolarized membrane potential underlies the DOWN state (Crunelli and Hughes, 2010). Some early studies focused on thalamic cells and depicted the molecular mechanism underlying the bimodality. In the classical explanation, the burst firing occurs from a hyperpolarized membrane potential. A burst firing consists of a series of ionic flows: (1) the hyperpolarization activated-cation channels (HCN) depolarizes the membrane (Ih) to activate the transient slow low-threshold Ca2<sup>+</sup> spike (IT), (2) the low-threshold Ca2<sup>+</sup> spike triggers action potentials consisting of the fast in- and out-flow of sodium (INa) and potassium (IK). In addition, the fast Na<sup>+</sup> spikes also activate high-threshold Ca2<sup>+</sup> current (ICa), (3) after the burst of action potentials, the membrane repolarizes as the low-threshold Ca2<sup>+</sup> spike (IT) ceases, (4) and the reduced depolarizing effect of I<sup>T</sup> is followed by the overshooting after-hyperpolarization which is caused by the outflows of K<sup>+</sup> due to the activation of Ca2<sup>+</sup> dependent potassium channels (IK[Ca]) (Jahnsen and Llinas, 1984;

FIGURE 1 | The brain region and the neural connections in the MS-DBB–hippocampus system. (A) The MS-DBB–hippocampus system depicted in 3D space. The blue region is the hippocampal formation; the orange region is the medial septal complex (including MS-DBB). The purple region is the somatomotor areas in the neocortex (isocortex) presented for visual aid to highlight the deep locations of the hippocampus and the medial septal complex. (B) The thalamocortical system depicted in 3D space. The transparent blue region represents the neocortex. The orange region represents thalamus. The purple region is the somatomotor areas in the neocortex presented for visual aid to highlight the deep location of the thalamus. (C) Schematic diagram of the MS-DBB–hippocampus system. The hippocampal pyramidal neurons (blue triangles) are aligned in parallel so that they produce the strong alteration in the electrical field measurable by EEG. The pyramidal neurons receive excitatory cholinergic and glutamatergic inputs from the MS-DBB and inhibitory inputs from hippocampal interneurons. The pyramidal neurons innervate cholinergic and GABAergic neurons in MS-DBB. Dark brown ellipsoids represent GABAergic neurons; light brown ellipsoids represent cholinergic neurons; blue ellipsoids and triangles represent glutaminergic neurons. Red connection and blue connections represent excitatory and inhibitory connections, respectively. The 3D plots were drawn with cocoframer available at a public mouse brain atlas for parts (A) and (B) (Allen Institute for Brain Science, 2018).

McCormick and Bal, 1997). The essence of this ionic-flow model is the interaction between the relatively slow Ca2<sup>+</sup> oscillation and the fast action potentials (Steriade et al., 1993b). Synaptic input is not critical part of the model. Indeed, the isolated thalamic neuron displays firing patterns similar to those of intact neurons in slice or in vivo (Hernandez-Cruz and Pape, 1989; Suzuki and Rogawski, 1989), and simple theoretical models generate bursting without explicit synaptic connections (Izhikevich, 2007).

In contrast to the established model of the cellular mechanism for the thalamic delta oscillation, cellular mechanisms for the neocortical SWO are less understood. This disparity is presumably because the neocortical SWO has been widely accepted to emerge from a finely tuned neural circuit consisting of balanced excitatory and inhibitory synaptic connections (Shu et al., 2003; Crunelli et al., 2015). One piece of supporting evidence of the view comes from the observation that the application of antagonist of non-NMDA glutamate receptor diminishes the SWO in the neocortical slices (Sanchez-Vives and McCormick, 2000). However, this view does not exclude the possibility that the ionic-flow mechanism similar to thalamic neurons also works in the neocortex. Indeed, the bursting persists in at least two groups of neocortical neurons even without synaptic connections at the frequency range of SWO (Le Bon-Jego and Yuste, 2007). Moreover, the neocortical bursting neurons have the electrophysiological properties characteristic to specific neurons that exhibit the slow Ca2<sup>+</sup> oscillation (lowthreshold Ca2<sup>+</sup> spike), such as the rebound burst of action potentials following negative current injection (Huguenard, 1996; Le Bon-Jego and Yuste, 2007). Although it is to be confirmed that the intrinsic oscillatory property of neocortical cells is relevant to natural sleep, the property as ubiquitous as the low-threshold spike observed throughout brain regions with synchronized bursting for functions seems to play an important role in the neocortex (Huguenard, 1996; Li et al., 2009). Notably, it is recently shown that the changes in the composition of cortical interstitial Ca2<sup>+</sup> and K<sup>+</sup> ions influence the sleep–wake cycle (Ding et al., 2016). This study suggests that the intrinsic properties of neural oscillation may depend on the intracellular concentration of Ca2<sup>+</sup> and K<sup>+</sup> ions, which is in line with the observations that the loss of Ca2<sup>+</sup> and K<sup>+</sup> channels, such as SK2 (Kcnn2) and SK3 (Kcnn3), Cav3.1 (Cacna1g), Cav3.2 (Cacna1h), and TASK3 (Kcnk9), affected the sleep duration in vivo (Tatsuki et al., 2016; Yoshida et al., 2018).

# Ca2+-Dependent Hyperpolarization Pathway for SWO

The investigations with emphasis on the circuit of SWO have provided the detailed view of intra-neocortical network of glutamatergic excitatory neurons and GABAergic inhibitory neurons and the inter-regional interaction between the thalamus and the neocortex (Crunelli et al., 2015; Neske, 2015). The circuit-based investigation has also been successful in describing the brain-wide regulatory neural circuits of sleep and wake cycle (Weber and Dan, 2016; Saper and Fuller, 2017; Scammell et al., 2017). On the other hand, the essential properties of oscillatory neurons remain elusive due to the complex nature of the interaction within the circuit. One approach is to simplify the neural circuit by constructing a computational model of "averaged" homogenous population of neurons (Tatsuki et al., 2016). The averaged-neuron (AN) model includes the excitatory glutamatergic AMPA and NMDA receptors that mediate Na<sup>+</sup> and Ca2<sup>+</sup> currents (IAMPA, INMDA), respectively, and the inhibitory GABA<sup>A</sup> receptors that regulate Cl<sup>−</sup> currents (IGABA). The model also contains voltage-gated Ca2<sup>+</sup> current (ICa), voltage-gated and persistent sodium current (INa, INaP), several types of K<sup>+</sup> currents including voltagegated (IK), leak (IL), fast A-type (IA), inwardly rectifying (IAR), slowly inactivating (IKS), and Ca2+-dependent potassium currents (IK[Ca]). The unbiased search over the almost 20 million sets of parameters demonstrated that the AN model could generate bursting in a homogenous population of neurons. Moreover, the model predicted that the cellular Ca2<sup>+</sup> plays a pivotal role in the alternation between UP and DOWN states.

To validate the prediction of the model, a series of knockout (KO) mice were generated. The KOs covered following genes: Ca2<sup>+</sup> dependent K<sup>+</sup> channels (eight genes) including SK2 (Kcnn2) and SK3 (Kcnn3), the voltage-gated Ca2<sup>+</sup> channels (10 genes) including Cav3.1 (Cacna1g) and Cav3.2 (Cacna1h), the plasma membrane calcium pumps (four genes) including Atb2b3 (PMCA3), and NMDA receptors (seven genes) including Nr3a (GRIN3A) (Sunagawa et al., 2016; Tatsuki et al., 2016). The results demonstrated that the changes of sleep duration observed in the mutant mice were consistent with the predictions. In addition, the acute or chronic pharmacological inhibition of NMDA receptors (possibly Nr1/Nr2b) in WT mice reduced the duration of sleep, suggesting the contribution of Nr1 or Nr2b, the KO of which is embryonically lethal, in sleep regulation (Tatsuki et al., 2016). Building upon these results, they further tested the role of the major calcium-dependent protein kinase, calcium-/calmodulin-dependent protein kinase type II (CaMKII) in sleep regulation. Among the four different subunits of the CaMKII tested, CaMK2a KO and CaMK2b were found to be involved in the regulation of sleep/wake cycle (Tatsuki et al., 2016). Taken together, these results provided a hypothesis that the Ca2+-dependent hyperpolarization pathway plays an important role in regulating sleep duration through modulating the neural bimodality.

The AN model demonstrated that theoretical models could provide fundamental insight into the complex nature of bursting neurons. However, the AN model, which contains 13 components, is too complicated to interpret its detailed mathematical structure. Because it was important to elucidate how the transition between UP and DOWN states occurs, mathematical analyses, for example, to reveal the currents responsible for the transition were demanded. This point was addressed by constructing a simplified AN (SAN) model (Yoshida et al., 2018). Bifurcation and detailed mathematical analyses of the SAN model predicted that leak K<sup>+</sup> channels play a role in generating bursting. Furthermore, the following comprehensive phenotype assays with 14 KO mice of leak K <sup>+</sup> channels family identified that potassium two pore domain channel subfamily K member 9 (Kcnk9) gene is involved in sleep regulation, validating the prediction of the mathematical analysis (Yoshida et al., 2018). It is interesting to note that this data of the involvement of constant K<sup>+</sup> currents suggested that the conductance of leak K<sup>+</sup> channels may alter the threshold for transition from UP to DOWN state mediated by the Ca2+-dependent K<sup>+</sup> channels. Collectively, the Ca2+-dependent hyperpolarization pathway and leak K<sup>+</sup> channels are involved in regulating SWO. The mathematical insight given by the potential role for constant currents such as leak K<sup>+</sup> currents in affecting

the bifurcation explains that the other constant synaptic currents mediated by, for example, AMPA receptors and GABA receptors can contribute to wakefulness and sleep.

# THE NEURAL CIRCUITS OF THE EEG THETA OSCILLATION

The extensive multisite measurements of local field potentials in rodents found the hippocampus as the cardinal source of the theta oscillation (Green and Arduini, 1954). The hippocampal theta oscillation was later to be found associated with REM sleep (Jouvet, 1969). Since then, the hippocampal theta oscillation in sleeping animal is recognized as a definitive feature of REM sleep. The EEG theta oscillation is a summation of multiple signals generated by spatially distributed oscillators in the hippocampal– entorhinal regions, and the oscillation amplitude and phase vary as a function of behavior (Sirota et al., 2008; Buzsaki et al., 2012). Notably, the isolated hippocampal neurons can exhibit oscillations at the theta frequency band in vitro when it is bathed in acetylcholine or kainate receptor agonist (Williams and Kauer, 1997; Garner et al., 2005; Manseau et al., 2005; Fuller et al., 2007). Moreover, the hippocampus neurons, such as CA3 pyramidal neurons, exhibit low-threshold Ca2<sup>+</sup> spike and bursting (Llinas, 1988; Huguenard, 1996). Therefore, it is plausible to assume that the neurons of hippocampus possess an intrinsic ability to generate the theta oscillation. Interestingly, a recent study showed that a majority of hippocampal neurons are self-oscillatory, and the properties of oscillation, including frequency, are affected by environmental ions and cellular Ca2<sup>+</sup> (Penn et al., 2016). This effect occurs without changes in synaptic connectivity or neural circuit, suggesting that the intrinsic neural properties directly affect circuit-level oscillation.

Together, the body of evidence suggests that the brain oscillations, including hippocampal theta oscillation, originate from intrinsic cellular properties. The intrinsic oscillation resonates and is amplified in neural circuits to operate physiological function implemented in each brain region. This view is consistent with the observed function of neural circuits. The intensive studies on neural circuits for regulating REM sleep have revealed multiple brain regions and extracellular neurotransmitters. The following sections briefly review the neural circuits focusing on the regulation of hippocampal theta oscillation.

### The Brain Regions Involved in the Regulation of the Hippocampal Theta Oscillation

Similar to the model of SWO or delta oscillation which consists of interacting intrinsic oscillators in the neocortex or the thalamus (Crunelli et al., 2015), the widely accepted model of hippocampal theta oscillation adopted the view of neural circuits as an oscillatory unit. In the model, the interaction between the medial septum (MS) and diagonal band of Broca (MS-DBB) and hippocampus mediates the generation of theta oscillation (**Figure 1**; Brown et al., 2012; Teles-Grilo Ruivo and Mellor, 2013). The loss of major afferent input from MS-DBB abolishes the theta oscillation in the hippocampal–entorhinal regions in urethane-anesthetized animals or awake animals indicating that the MS-DBB plays a critical role in hippocampal oscillation (Brazhnik and Fox, 1997; Yoder and Pang, 2005).

The MS-DBB is a region of the BF and composed of a heterogeneous population of neurons including GABAergic, cholinergic, and glutamatergic neurons. The large portion of the GABAergic neurons exhibits bursting activity at the theta frequency while the cholinergic neurons have a low firing rate not related to the theta oscillation (Simon et al., 2006). However, the targeted lesion of cholinergic neurons in MS-DBB by 192 IgG-saporin injections reduces the amplitude of the hippocampal theta oscillation indicating that the cholinergic projection also plays a role in the hippocampal theta (Lee et al., 1994; Yoder and Pang, 2005). The selective lesion of GABAergic neurons and potentially other non-cholinergic neurons by kainic acid reduces the hippocampal theta oscillation more than cholinergic lesion. The combined lesion of cholinergic and GABAergic neurons almost eliminates the hippocampal theta oscillation (Yoder and Pang, 2005). In addition, the isolated MS-DBB can exhibit oscillation at the theta frequency band in vitro when it is bathed in acetylcholine agonist (Konopacki et al., 1987a,b; Goutagny et al., 2009; Pignatelli et al., 2012). These observations supported the view that the MS-DBB, especially the GABAergic projection, provides the hippocampus with critical inputs to mediate the theta oscillation (Manseau et al., 2005; Teles-Grilo Ruivo and Mellor, 2013).

The MS-DBB-hippocampus system has afferent input from nuclei in the brainstem from which the major sleep/wake regulatory pathways arise (**Figure 3**; Petsche et al., 1962; Buzsaki, 2002). The pons of the brainstem contains a population of REMon neurons (i.e., neurons that are active during REM sleep) in the sublaterodosal nucleus (SLD), the REM-off neurons (i.e., neurons that are inactive during REM sleep) in ventrolateral periaqueductal gray matter (vlPAG), and the adjacent lateral pontine tegmentum (LPT) which is also known as the deep mesencephalic reticular nucleus (DpME) (Boissard et al., 2002; Lu et al., 2006; Scammell et al., 2017). The pathways ascend through the midbrain and then split into a dorsal pathway and ventral pathway. The dorsal pathway innervates the thalamus which projects to neocortical areas, while the ventral pathway innervates the BF including the MS-DBB, the hypothalamus, and the cortex (Brown et al., 2012).

### The Direct Projection From the Pons to MS-DBB

The cholinergic inputs from the brainstem are the major neuromodulators of MS-DBB (Mesulam et al., 1983). This cholinergic input consists of two different pathways (**Figure 3**). The one pathway is the direct cholinergic projection from the PPT and the LDT (Hallanger and Wainer, 1988). The activation of PPT and LDT by injection of glutamate results in neocortical desynchronization and hippocampal theta oscillation accompanied by wakefulness or REM sleep (Datta and Siwek, 1997). Additionally, the optogenetic activation of cholinergic

neurons in PPT and LDT increased the initiation of REM sleep (Van Dort et al., 2015). However, selective lesions of PPT and LDT do not much affect REM sleep (Lu et al., 2006).

Importantly, retrograde tracer injected in the MS found positive cells in the precoerulueus (PC) region, the periventricular gray matter, and a dorsal extension of the SLD which shows high levels of activity (c-Fos expression) in REM sleep (Lu et al., 2006; Fuller et al., 2007). The projected PC neurons are glutamatergic, and the selective lesion of the PC and SLD abolished the theta oscillations during sleep (Lu et al., 2006), providing support for the concept that glutamatergic neurons in the PC regions play a critical role in mediating the emergence of theta oscillation from the cellular intrinsic oscillations in the MS-DBB–hippocampus system during REM sleep (Fuller et al., 2007). These observations give rise to the hypothesis of the "flip-flop model" in which the bidirectional inhibition between REM-on regions (PC and SLD) and REM-off regions (LPT and vlPAG) works like a flip-flop switch via GABAergic projections. The state of the switch is regulated by the excitatory inputs from serotoninergic dorsal raphe nucleus and locus coeruleus (DRN-LC) to the REM-on neurons (Lu et al., 2006), glutaminergic inputs from neurons located rostrolateral of SLD (Atoh1-E10.5 medial neurons) to REM-off neurons (Hayashi et al., 2015), and by the inhibitory inputs from the GABAergic ventral medulla neurons to REM-off neurons (Weber et al., 2015; **Figure 3**). The glutamatergic neurons in the PC and the dorsal part of the SLD project to the MS. On the other hand, the glutaminergic neurons in the ventral part of the SLD project to the spinal cord and regulate muscle atonia (Lu et al., 2006; Weng et al., 2014). Thus, lesions of the ventral SLD and PC produce a specific loss of REM sleep components; that is, the muscle atonia and the EEG theta, respectively (Fuller et al., 2007).

# The Indirect Projection From the Pons to MS-DBB

The other pathway is the indirect projection mediated by nuclei in the hypothalamus (Woodnorth et al., 2003; **Figure 3**). The supramammillary nucleus (SuM) in the hypothalamus is one candidate which may relay the regulation from the nucleus pontis oralis (PnO). PnO is a region of the brainstem reticular formation which is projected by PPT. The PnO activity is associated with the presence of hippocampal theta (Kirk and McNaughton, 1991; Oddie et al., 1994; Vertes and Kocsis, 1997; Pignatelli et al., 2012). Since the neurons in the PnO did not show rhythmic firing, the SuM has been assumed to translate the tonic firing of PnO into rhythmic firing. However, the lesion of SuM fails to affect theta rhythm (Thinschmidt et al., 1995; Renouard et al., 2015) while the inactivation of SuM by procaine injection reduces both frequency and amplitude of theta rhythm in the hippocampus (Kirk and McNaughton, 1993). The recent finding of the role for SuM in the creation of oscillatory interference between the theta oscillation of itself and the ongoing oscillations in its target areas suggested that the SuM is a coordinator of phase coherence of theta oscillations among brain regions (Ito et al., 2018).

### THE ESSENTIAL GENES FOR REM SLEEP AND THE ASSOCIATED EEG THETA OSCILLATION: Chrm1 AND Chrm3

The accumulated evidence indicated that acetylcholine plays an important role in regulating REM sleep. However, it is demonstrated that the cholinergic function in a specific neural circuit can be limited (Grace et al., 2014). A possible function of acetylcholine is the regulation of cellular properties involved in the theta oscillation rather than of switching neural circuits. Indeed, isolated hippocampal and MS-DBB neurons can exhibit oscillations at the theta frequency band in vitro when it is bathed in acetylcholine receptor agonist (Williams and Kauer, 1997; Fellous and Sejnowski, 2000; Manseau et al., 2005). However, the molecular investigation into the necessity of acetylcholine for REM sleep has been hindered, because of the formidable redundancy resulting from the multitude of genes involved in the regulation: the 11 neuronal-type nicotinic acetylcholine receptors and 5 muscarinic acetylcholine receptors. The identification of critical cholinergic receptors has been unfeasible until the recent emergence of the efficient techniques in genetics such as CRISPR and ES-mice (Sunagawa et al., 2016; Ukai et al., 2017). The comprehensive study with these techniques on acetylcholine receptors revealed that DKO mice of G<sup>q</sup> proteincoupled muscarinic acetylcholine receptors: Chrm1 and Chrm3

abolish REM sleep and the associated enrichment of EEG theta oscillation during sleep, leaving the theta oscillation largely unaffected during wakefulness (Niwa et al., 2018; **Figure 4**). This work demonstrated the necessity of acetylcholine for REM sleep and the EEG theta oscillation during sleep, which is in line with the previous pharmacological studies that demonstrated that the muscarinic blockers, such as atropine or scopolamine, diminish the EEG theta oscillation in anesthetized animals (Kim and Jeong, 1999; Buzsaki, 2002).

In contrast to the drastic sleep phenotypes observed in the DKO mice of muscarinic receptors, the comprehensive KO study of nicotinic acetylcholine receptors did not show significant sleep phenotype (Niwa et al., 2018). Nicotinic acetylcholine receptors are ionotropic, and their response is fast compared to the metabotropic muscarinic acetylcholine receptors. In consistent with the slow property of muscarinic acetylcholine receptors relevant to sleep regulation, the majority (80–90%) of cholinergic axon terminals in hippocampus are diffusely organized (Descarries et al., 1997), and do not associate with distinct postsynaptic sites suggesting that the cholinergic signaling in the hippocampus is primarily mediated by volume transmission as an ambient cholinergic tone instead of synaptic transmission (Teles-Grilo Ruivo and Mellor, 2013). Indeed, the elevated level of acetylcholine in the hippocampus is found associated with REM sleep in vivo (Teles-Grilo Ruivo et al., 2017). Together, these observations imply that acetylcholine contributes to the EEG theta oscillation of hippocampal neurons mainly through the slow regulation and modulate the intrinsic neuronal properties to generate the bursting activity at theta frequency band.

### The Possible Molecular Mechanism of the Theta Oscillation With Chrm1 and Chrm3

The important insight given by the finding of the essential role for Chrm1 and Chrm3 in REM sleep is that, in contrast to the conventional neural-circuit view that the GABAergic input from MS-DBB neurons is primarily driving the EEG theta oscillation and the cholinergic regulation plays subsidiary role (Brown et al., 2012; Teles-Grilo Ruivo and Mellor, 2013), the cholinergic regulation in the MS-DBB–hippocampus system is comparably critical for the EEG theta oscillation at molecular and cellular level. The conventional view was based on the observations that the combined lesion of cholinergic and GABAergic neurons in MS-DBB almost eliminated the hippocampal theta oscillation (Yoder and Pang, 2005), theta activity survives in the hippocampus following the selective lesion of cholinergic neurons in MS-DBB (Lee et al., 1994), the selective lesion of GABAergic neurons and potentially other non-cholinergic neurons reduces the hippocampal theta oscillation more than cholinergic lesion (Yoder and Pang, 2005), and the large portion of the GABAergic neurons exhibits burst firing activity at the theta frequency while the cholinergic neurons have a low firing rate not related to the theta oscillation in MS-DBB (Simon et al., 2006). However, the almost complete absence of REM sleep in the Chrm1 and Chrm3 DKO mice re-emphasized the role for the cholinergic pathway in REM sleep. Indeed, cultured hippocampal slice bathed in the acetylcholine agonist, carbachol, shows oscillation at the range of frequencies, including theta band depending on the drug concentration (Fellous and Sejnowski, 2000). The hippocampus intrinsic oscillation was inhibited either by Chrm1 or Chrm3 inhibitors (Williams and Kauer, 1997).

Although the current knowledge of the essential role for Chrm1 and Chrm3 in REM sleep is based on the wholebody KO mice, investigations into the intrinsic cellular mechanisms involving Chrm1 and Chrm3 may contribute to understanding REM sleep regulation, much as the study of thalamocortical oscillations benefited from investigations into regulatory mechanisms of cellular excitability (Steriade et al., 1993b). In the thalamocortical cells, the membrane depolarization is mediated to a large part by the inhibition of a leak K<sup>+</sup> conductance (IKleak), the molecular instance of which is the two-pore domain K<sup>+</sup> channels TASK1 (Kcnk3) and TASK3 (Kcnk9) (Meuth et al., 2006; Yoshida et al., 2018). The metabotropic glutamate and muscarinic acetylcholine receptors competitively activate G<sup>q</sup> pathway, which in turn inhibit the leak K<sup>+</sup> conductance (Chen et al., 2006; Coulon et al., 2010). A relatively small shift of membrane potential (∼10 mV) is sufficient to mediate a switch of firing mode in vivo. Moreover, the depolarization induced by the activation of muscarinic acetylcholine receptors Chrm1 and Chrm3 mediated by G<sup>q</sup> proteins can mediate the switch in thalamocortical neurons (Broicher et al., 2008; Coulon et al., 2010). Because the level of acetylcholine is elevated in the hippocampus during REM sleep (Teles-Grilo Ruivo et al., 2017), it may be plausible to assume a similar mechanism works in the hippocampus to switch the hippocampal neurons between the oscillating modes. Notably, carbachol-induced depolarization of hippocampal CA1 neurons is eliminated from Chrm1/3 DKO mice while the depolarization remained in single KO mice, suggesting that Chrm1 and Chrm3 receptors are each redundantly capable of depolarizing hippocampus neurons (Dasari and Gulledge, 2011). However, further studies are necessary to dissect the functions of Chrm1 and Chrm3 in the MS-DBB–hippocampus system.

# The Distinct Roles of Chrm1 and Chrm3 for REM Sleep

Comprehensive investigation on the acetylcholine receptors found that the combinatorial function of Chrm1 and Chrm3 is essential for REM sleep and associated EEG theta oscillation during sleep, whereas the function of each gene has yet to be investigated. Especially, single KO of either gene showed different phenotypic responses. The Chrm1 KO mice had a reduced REM sleep duration, but NREM sleep duration was only moderately reduced. On the other hand, Chrm3 KO mice had a reduced NREM sleep duration, but REM sleep duration was similar to that of WT mice (Niwa et al., 2018). These observations raised a question of what molecular mechanisms account for the observed difference. Muscarinic acetylcholine receptors consist of five isoforms and coupled with G<sup>q</sup> proteins (M1, M3, and M5) or G<sup>i</sup> proteins (M2 and M4). Differences in the preference of G protein coupling come from the difference in an amino acid

sequence in the third intracellular loop between the M1, M3, and M5 sequences compared to the M2 and M4 sequences (Wess et al., 1997). However, several studies have shown that receptors coupling predominantly to one G protein family can also couple with other G proteins, though less efficiently. For example, Chrm3 receptor associates to both G<sup>q</sup> and G<sup>i</sup> in rat parotid glands (Dai et al., 1991), Chrm1 receptor also has G<sup>s</sup> activity when ectopically expressed in Chinese hamster ovary (CHO) cells (Burford and Nahorski, 1996), whereas both Chrm1 and Chrm3 predominantly link to Gq. In addition, a short variable sequence of the amphipathic helix (H8), typically three turns long and with palmitoylation sites at its C terminus, is present in several GPCRs including Chrm3 (Qin et al., 2011; Venkatakrishnan et al., 2013). The H8 forms the preassembly with G<sup>q</sup> proteins, which may contribute to the possible difference in the rate of receptor activation, compared with Chrm1 (Qin et al., 2011). Another possibility is the different spatial distribution of Chrm1 and Chrm3 in the brain (Levey et al., 1995). Indeed, about 60% of the total muscarinic acetylcholine receptors of the hippocampus is Chrm1 receptors, whereas Chrm3 receptor is up to 10% (Dasari and Gulledge, 2011). Future investigation of downstream pathways from the identified receptors: Chrm1 and Chrm3 may reveal the mechanism of REM sleep and their physiological roles. Especially, several functions related to the long-term potentiation (LTP) have been shown to be mediated particularly by Chrm1. For example, Chrm1-dependent inhibition of SK channels enhances NMDA receptor function to facilitate the induction of LTP (Buchanan et al., 2010), and the Chrm1-dependent inhibition of voltage-activated Kv7 potassium channels facilitate LTP (Petrovic et al., 2012; Teles-Grilo Ruivo and Mellor, 2013), while involvement of Chrm3 in this context is unknown.

### PHOSPHORYLATION HYPOTHESIS FOR THE HOMEOSTATIC REGULATION OF SLEEP

The amount of sleep is regulated to be in a physiologically feasible range. This regulation is called the homeostatic regulation of sleep. The homeostatic regulation can comprise two distinct hypothetical processes: "process C" and "process S" (Borbely, 1982). Process C is the circadian component that regulates

the propensity of sleep with the rhythm of 24 h. The process S is a sleep-dependent process that monitors accumulated amount of sleep and compensates the detected loss or excess of sleep. The mechanism of the homeostasis is under vigorous investigation at present.

### Homeostatic Regulation of NREM Sleep

The EEG power in the delta frequency band reflects the pressure for the NREM-sleep resulting from the loss of NREM sleep. The NREM-sleep need increases during wakefulness period, while it decreases during the sleep period. The changes of NREMsleep need are well described by exponential function (Borbely, 1982). An apparent but unresolved question is by what molecular mechanism NREM-sleep need is represented. One important criterion to be satisfied is that the molecular mechanism must work in the time scale of minutes to hours, which is slower in order of magnitude compared to the time scale of neural action potentials. Candidate components of the mechanism include ion concentration, gene expression, post-translational modification, and production/degradation of ion channels or pumps.

Interestingly, in the context of Process C, the phosphorylation plays an important role in keeping the circadian period about 24 h (Tomita et al., 2005; Isojima et al., 2009; Shinohara et al., 2017). Phosphorylation was indicated to play a role also in Process S, as the loss of calcium/CaMKII gene resulted in significant reduction of sleep (Tatsuki et al., 2016). This finding is further supported by the observation that the wakefulness induced phosphorylation in the extracellular signal-regulated kinase (ERK) proteins, which are upstream of a group of genes expressed in activity-dependent manner and involved in sleep regulation (Mikhail et al., 2017). Moreover, the following phosphoproteomics studies revealed a number of genes in the intracellular signaling pathways change their states of phosphorylation along with the sleep/wake cycles (Diering et al., 2017; Wang et al., 2018; Bruning et al., 2019). Because protein functions can be modulated by sitespecific phosphorylation or by cumulative phosphorylation of multiple sites, all of this evidence strongly suggests that the phosphorylation process is a component of the homeostatic regulation of sleep.

Because the Ca2+-dependent hyperpolarization pathway plays an important role in switching UP and DOWN states of neurons (Tatsuki et al., 2016, 2017), and the persistent UP state is associated with wakefulness, Ca2+-dependent phosphorylation is a promising regulatory component for the homeostasis in sleep and wake cycle (Ode et al., 2017; Shi and Ueda, 2018). A candidate gene family to be involved in Ca2+-dependent phosphorylation is calcium/CaMKII. Indeed, the KO mice of the CaMKII family revealed that either KO of CaMKIIα and CaMKIIβ results in significant reduction of sleep duration (Tatsuki et al., 2016), implying that CaMKII may be link between the actions of Ca2<sup>+</sup> in the time domain of second to the activity of the kinase in the time domain of hours.

### Homeostatic Regulation of REM Sleep

In contrast to the EEG delta power established as an indicator of NREM-sleep need, any single component of EEG spectral power has not been established to represent REM-sleep need. Nonetheless, REM sleep is also under the homeostatic regulation; that is, the loss of REM sleep is compensated for by the increase in the duration of REM sleep (Franken, 2002). Selective REM sleep deprivation induces a rebound increase in subsequent REM sleep. Because the selective REM sleep deprivation does not largely affect the amount of NREM sleep, the homeostatic regulation of REM sleep seems to be likely independent to that of NREM sleep (McCarthy et al., 2016). However, prolonged REM sleep increases the delta power in the subsequent NREM sleep indicating there is profound interaction between REM and NREM sleep homeostatic regulatory mechanisms (Hayashi et al., 2015). Although the molecular machinery of the REM sleep homeostasis has yet to be investigated, it is plausible to assume mechanisms similar to the NREM sleep homeostasis, such as phosphorylation process, also work in REM sleep homeostasis. Notably, while most antidepressants suppress REM sleep, the physiologically induced REM sleep deficits are compensated for regardless of the subsequent pharmacological suppression of REM sleep (McCarthy et al., 2016). This observation implies that the homeostatic regulation of REM sleep consists of molecular and cellular mechanisms distinct from the neural circuits mediated by neurotransmitters, which are targeted by the antidepressants, such as serotonin and acetylcholine pathways. This insight is in line with the hypothetical involvement of the phosphorylation process in the homeostatic regulation, suggesting that the homeostasis is implemented at the cellular level rather than at the neural circuits level.

# FUTURE PERSPECTIVE

The pioneering studies focused on the electrophysiological activity of brains and identified characteristic EEG signatures such as SWO, delta, and theta oscillation to define sleep stages. The investigations revealed the underlying cellular level machinery to generate bursting activity for explaining the population level EEG signals. Based on those findings, in the last couple of decades, the focused study on the neural circuits of sleep regulation has been successful in identifying brain regions and detailed interactions among the regions (Scammell et al., 2017). However, the conventional way to investigate the electrophysiological properties of neurons has been mostly pharmacological approaches, hence an identified molecular component in the machinery is a cluster of molecules responsive to the applied drug. This restriction has hindered the identification of specific genes involved in sleep regulation. The recent advent of so-called next-generation genetics such as CRISPR and ES-mouse methods has been easing the longstanding restriction by significantly reducing the time and cost to produce KO or knockin mice of specifically targeted genes (Sunagawa et al., 2016; Ukai et al., 2017). Using these methods, researchers can generate a variety of KO mice covering the cluster of genes involved in a sub-system of sleep regulation. The application of the methods revealed that the genes involved in the Ca2+-dependent hyperpolarization pathway are important in sleep regulation (Sunagawa et al., 2016; Tatsuki et al., 2016; Yoshida et al., 2018). Besides identifying genes involved

in NREM sleep duration, the next-generation genetics also identified genes essential for REM sleep: muscarinic acetylcholine receptors Chrm1 and Chrm3 (Niwa et al., 2018).

It is also notable that the reverse genetics approach demonstrated the role of the major calcium-dependent protein kinase, CaMKII in sleep regulation (Tatsuki et al., 2016), suggesting that the phosphorylation is involved in the sleep regulation. In addition to the reverse-genetic approach, the effort of forward-genetics also demonstrated that the mutation of Sik3 protein kinase gene causes a profound increase in sleep duration by a gain-of-function mutation (Funato et al., 2016). The phosphorylation process may occur in the time scale of hours, and be modulated by cumulative phosphorylation of multiple sites, e.g., casein kinase I (CKI) in the circadian regulation (Isojima et al., 2009; Shinohara et al., 2017). We note that the genes involved both in phosphorylation and sleep regulation are attractive candidates for the future studies on the homeostatic regulation of sleep (Ode et al., 2017; Tatsuki et al., 2017; Shi and Ueda, 2018).

The finding of the almost abolished REM sleep in the Chrm1 and Chrm3 DKO mice may provide a useful tool to clarify the function of REM sleep in learning and memory (Niwa et al., 2018). The dominance of theta oscillation in EEG signal during REM sleep indicates the synchronized activity of hippocampal neurons. The synchronization is believed to be critical for transferring information between neocortex and hippocampus and the sleep-related neural plasticity (Sirota et al., 2008; Grosmark et al., 2012). The optogenetic silencing of GABAergic neurons during REM sleep in the MS-DBB hindered the mice from properly consolidating what they learned prior to the sleep (Boyce et al., 2016), indicating the important roles for the theta oscillation in REM sleep and associated learning and memory. This function of theta oscillation is coinciding with that of delta or SWO in the thalamocortical system. SWO promotes learning and memory consolidation (Marshall et al., 2006; Miyamoto et al., 2016). On the other hand, a pharmacogenetic study revealed that reduction or induction of REM sleep attenuates or enhances SWO, respectively, in the subsequent NREM sleep (Hayashi et al., 2015). Thus, REM sleep might indirectly regulate memory formation in the neocortex through NREM sleep (Miyazaki et al., 2017).

Aside learning and memory in the hippocampus, other physiological functions of REM sleep remain obscure. Interestingly, the duration of REM sleep increases in some depression and the most antidepressants inhibit REM sleep

### REFERENCES


in animals and humans (McCarthy et al., 2016). This strong correlation between REM sleep and psychiatric disorders including post-traumatic stress disorder (PTSD) implies that controlling REM sleep may help PTSD patients to alleviate the symptoms. The further elucidation of the molecular mechanism of theta oscillation will provide significant insights on how to control the amount of REM sleep both in mice and humans and may facilitate, for example, to refine antidepressants and to reveal the physiological roles of REM sleep in its closely related higher cognitive functions such as dreaming or consciousness.

### AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

### FUNDING

This work was supported by grants from the Japan Agency for Medical Research and Development – Core Research for Evolutional Science and Technology (AMED-CREST) [AMED/Ministry of Education Culture Sports Science and Technology (MEXT) Grant JP17gm0610006 to HU], Brain Mapping by Integrated Neurotechnologies for Disease Studies (Brain/MINDS) (AMED/MEXT Grant JP17DM0207049 to HU), Basic Science and Platform Technology Program for Innovative Biological Medicine (AMED/MEXT Grant JP17AM0301025 to HU), World Premier International Research Center Initiative (MEXT; HU), a Grant-in-Aid for Scientific Research (S) (JSPS KAKENHI Grant Number 18H05270, HU), HFSP (Human Frontier Science Program) Research Grant Program (HFSP RGP0019/2018), and an intramural Grant-in-Aid from the RIKEN Center for Biosystems Dynamics Research (to HU).

### ACKNOWLEDGMENTS

We thank all the laboratory members at the RIKEN Center for Biosystems Dynamics Research and the University of Tokyo, particularly Dr. S. Shi and Dr. H. Ono, for kind help in preparing figures and discussions. We also thank Kylius Wilkins at the research promotion office of RIKEN for careful reading and editing.

Blanco-Centurion, C., Gerashchenko, D., and Shiromani, P. J. (2007). Effects of saporin-induced lesions of three arousal populations on daily levels of sleep and wake. J. Neurosci. 27, 14041–14048. doi: 10.1523/JNEUROSCI.3217-07.2007



Buzsaki, G. (2002). Theta oscillations in the hippocampus. Neuron 33, 325–340.





**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Yamada and Ueda. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Simultaneous Electrophysiology and Fiber Photometry in Freely Behaving Mice

Amisha A. Patel<sup>1</sup> , Niall McAlinden<sup>2</sup> , Keith Mathieson<sup>2</sup> and Shuzo Sakata<sup>1</sup> \*

<sup>1</sup> Strathclyde Institute of Pharmacy & Biomedical Sciences, University of Strathclyde, Glasgow, United Kingdom, <sup>2</sup> Department of Physics, Institute of Photonics, SUPA, University of Strathclyde, Glasgow, United Kingdom

In vivo electrophysiology is the gold standard technique used to investigate subsecond neural dynamics in freely behaving animals. However, monitoring celltype-specific population activity is not a trivial task. Over the last decade, fiber photometry based on genetically encoded calcium indicators (GECIs) has been widely adopted as a versatile tool to monitor cell-type-specific population activity in vivo. However, this approach suffers from low temporal resolution. Here, we combine these two approaches to monitor both sub-second field potentials and cell-type-specific population activity in freely behaving mice. By developing an economical custommade system and constructing a hybrid implant of an electrode and a fiber optic cannula, we simultaneously monitor artifact-free mesopontine field potentials and calcium transients in cholinergic neurons across the sleep-wake cycle. We find that mesopontine cholinergic activity co-occurs with sub-second pontine waves, called P-waves, during rapid eye movement sleep. Given the simplicity of our approach, simultaneous electrophysiological recording and cell-type-specific imaging provides a novel and valuable tool for interrogating state-dependent neural circuit dynamics in vivo.

### Edited by:

Kaspar Emanuel Vogt, University of Tsukuba, Japan

### Reviewed by:

Balázs Pál, University of Debrecen, Hungary Alessandro Silvani, University of Bologna, Italy

\*Correspondence:

Shuzo Sakata shuzo.sakata@strath.ac.uk

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 16 October 2019 Accepted: 07 February 2020 Published: 21 February 2020

### Citation:

Patel AA, McAlinden N, Mathieson K and Sakata S (2020) Simultaneous Electrophysiology and Fiber Photometry in Freely Behaving Mice. Front. Neurosci. 14:148. doi: 10.3389/fnins.2020.00148 Keywords: brain state, REM sleep, GCaMP, acetylcholine, pontine waves, brainstem

# INTRODUCTION

Intracranial electrophysiological recordings monitor neuronal activity at various spatial scales, from single cells to populations across brain regions, with high temporal resolution (Buzsaki, 2004; Buzsaki et al., 2012; Jun et al., 2017). However, one of limitations in this approach is identifying the source of the neural signal: because neuronal activity is typically monitored extracellularly in freely behaving condition, the identification/isolation of recorded neurons is challenging (Einevoll et al., 2012; Harris et al., 2016).

Genetically encoded indicators offer complementary advantages over in vivo electrophysiological approaches (Lin and Schnitzer, 2016; Deo and Lavis, 2018; Wang et al., 2019). Over the last two decades, genetically encoded calcium indicators (GECIs) have been widely used to interrogate not only neuronal ensemble dynamics, but also activity of non-neuronal cells, such as astrocytes in vivo (Nakai et al., 2001; Chen et al., 2013; Stobart et al., 2018; Dana et al., 2019; Inoue et al., 2019; Stringer et al., 2019). For example, GECIs enable cell-type-specific targeting and long-term monitoring of neuronal activity in vivo. However, because of the intrinsic nature of calcium signals, the low temporal resolution of GECIs are not ideal for monitoring sub-second

**264**

neural dynamics. It is also challenging to monitor individual neuronal activity in deep brain areas without causing significant tissue damage.

Here, we combine an electrophysiological approach with GECI-based fiber photometry to simultaneously monitor both field potentials and calcium transients in freely behaving mice. Fiber photometry is an imaging method used to monitor fluorescent signals via an implanted fiber optic cannula (Adelsberger et al., 2005; Lutcke et al., 2010; Kim et al., 2016; Sych et al., 2019). Although it is still invasive, the diameter of the cannula is thinner than an endoscope. Therefore, fiber photometry is well-suited to monitor neural population activity in deep tissue, such as the brainstem.

In the present study, we develop a versatile custom-made fiber photometry system with the capability to integrate in vivo electrophysiological recording in freely behaving mice. To validate our system, we focus on pontine waves (P-waves), which were reported in mice recently (Tsunematsu et al., 2020). Because mesopontine cholinergic neurons have been implicated in the induction of P-waves (Callaway et al., 1987; Datta, 1997), we monitor calcium transients from GCaMP6s-expressing mesopontine cholinergic neurons along with detecting P-waves electrophysiologically. We show that P-waves during REM sleep co-occurs with calcium transients in mesopontine cholinergic neurons. Thus, our system allows simultaneous electrophysiological recording and fiber photometry in freely behaving mice.

# MATERIALS AND METHODS

### Recording System Configuration

The recording system is shown in **Figure 1** and a parts list for the fiber photometry system is summarized in **Table 1**. A detailed construction manual is provided in **Supplementary Material**. Essential codes for data acquisition and data processing are also available<sup>1</sup> .

Briefly, the fiber photometry system consisted of two excitation channels. A 470 nm LED (M470L3, Thorlabs) was used to extract a Ca2+-dependent signal and a 405 nm LED (M405L3, Thorlabs) was used to obtain a Ca2+-independent isosbestic signal. Light from the LEDs was directed through excitation filters (FB470-10, FB405-10, Thorlabs) and a dichroic mirror to the fiber launch (DMLP425R and KT110/M, respectively). The fiber launch was connected to a multimode patch cable (M82L01, Thorlabs) which could be reversibly attached and detached to an implantable optic fiber on the mouse via a ceramic mating sleeve (CFM14L05 and ADAF1, respectively). Light emissions from GCaMP6s expressing neurons were then collected back through the optic fiber, and directed through a detection path, passing a dichroic mirror (MD498) to reach a photodetector (NewFocus 2151, Newport). A National Instruments DAQ (NI USB-6211) and custom-written LabVIEW software was used to control the LEDs and acquire fluorescence data at 1 kHz. LEDs were alternately turned on and off at 40 Hz in a square pulse pattern. Electrophysiology signals were recorded at 1 kHz using an interface board (RHD2000, Intan Technologies) and connected to the mouse via an amplifier (RHD2132 16-channel amplifier board, Intan Technologies).

### Implant Fabrication

Hybrid implants (also referred to as an optrode) consisted of a bipolar electrode (pair of electrodes) glued to the optic fiber and were fabricated though a multistep process. First two 0.1 mm diameter stainless steel wires (FE205850/2, Goodfellow) were cut to approximately 1.5 cm in length with fine scissors (14084- 08, Fine Science Tools). Both wires were glued together (offset by approximately 0.5–1 mm at the tip) and the other end of the bundle was cut so that the tips of the wires were aligned (**Figure 2A**, step 1). Insulation was scraped off from the flush end of the bundle using a scalpel blade and connected to a 2 piece connector (SS-132-T-2-N, Samtec) using conductive epoxy (186-3593, RS Pro) (**Figure 2A**, step 2). The conductive epoxy was left to dry for 10 min and then secured with dental cement. Impedances were checked by connecting the bipolar electrodes to the Intan system (RHD2132 16-channel amplifier board and RHD2000, Intan Technologies) with a custom-made connector, and placing the tips of the bipolar electrodes in saline. Electrodes with impedances between 200 k and 1 M at 1 kHz were folded (**Figure 2A**, step 3): the vertical shaft closest to the electrode tips was approximately 5 mm long and the horizontal section 3– 4 mm. The folded electrode was positioned alongside the optic fiber and fixed in place 500 µm below the tip of the optic fiber with superglue (473-455, RS Pro), taking care not to get glue on the tips of either the optic fiber or bipolar electrode (**Figure 2A**, step 4). Dental cement was then used to secure and stabilize the structure (**Figure 2B**). Impedances were checked again (the range was 276–452 k) and optrodes were ready for implantation.

### Animals

All animal experiments were performed in accordance with the United Kingdom Animals (Scientific Procedures) Act of 1986 Home Office regulations and approved by the Home Office (PPL 70/8883). Three ChAT-IRES-Cre (JAX006410) mice were used (female, 8–37 weeks-old) and housed individually in highroofed cages with a 12 h:12 h light/dark cycle (light on hours: 7:00–19:00). Mice had ad libitum access to food and water. All experiments were performed during the light period. No blind and randomized experimental design was adopted due to the nature of the technical development study.

### Surgery

The surgical procedures have been described previously (Tsunematsu et al., 2020). Briefly, mice were anesthetized with isoflurane (5% for induction, 1–2% for maintenance) and placed in a stereotaxic apparatus (SR-5M-HT, Narishige). Body temperature was maintained at 37◦C with a feedback temperature controller (40–90–8C, FHC). Lidocaine (2%, 0.1–0.3 mg) was administered subcutaneously at the site of incision. Two bone screws were implanted on the skull for monitoring cortical EEGs and twisted wires were inserted into

<sup>1</sup>https://github.com/Sakata-Lab

the neck muscle for obtaining EMG signals. An additional bone screw was implanted over the cerebellum to provide a ground/reference channel. These electrodes were connected to a two-by-three piece connector (SLD-112-T-12, Samtec). Two additional anchor screws were implanted bilaterally over the parietal bone to provide stability and a small portion (approximately 1 cm long) of a drinking straw was placed horizontally (opening facing medial/lateral axis) between the anchor screws and the connector. The purpose of the drinking straw was to create a hollow cavity within the head cap which allowed an Allen key to pass through and hold the mouse head still for connecting and disconnecting the mouse to the head-amp. The Allen key was securely clamped in place with a workbench vice. The viral vector (AAV5-CAG-flex-GCaMP6s-WPRE-SV40, Penn Vector Core; titer 8.3 × 10<sup>12</sup> GC/ml) was microinjected (500 nl at 30 ml/min) (Nanoliter2010, WPI) to target the pedunculopontine tegmental nucleus (PPT) and laterodorsal tegmental nucleus (LDT) (−4.5 mm posterior, 1 mm lateral from bregma, and 3.25 mm depth from brain surface). The micropipette was left in the brain for an additional 10 min and then slowly raised up. A hybrid implant (see above) was then implanted 3 mm deep from the surface of the brain and all components were secured to each other and the skull with dental cement.

### Recording Procedures

After a recovery period (3–4 weeks), mice were habituated to being handled and tethered to the freely behaving system over several consecutive days. Mice were scuffed and the straw on the head cap slotted into a custom-made clamp, to keep the head still and absorb any vertical forces when connecting the electrophysiology and fiber photometry tethers to the head cap. Once connected, mice were placed in an open top Perspex box (21.5 cm × 47 cm × 20 cm depth) lined with absorbent paper, bedding, and soft food (creamed porridge, Heinz). During the habituation

### TABLE 1 | Parts list for the fiber photometry system.

fnins-14-00148 February 20, 2020 Time: 20:13 # 4


is bent. (4) The bent bipolar electrode is attached to a fiber optic cannula. (B) A photograph of an assembled implant.

period, short recordings (20–30 min) were taken to test illumination parameters for the best signal to noise ratio. The illumination power was adjusted at the tip of optical fiber to 0.4–0.94 mW/mm<sup>2</sup> for the 405 nm LED and 0.7– 1.37 mW/mm<sup>2</sup> for the 470 nm LED. Following the habituation period, simultaneous electrophysiological recording and calcium imaging was performed for 4–5 h to allow for multiple sleep/wake transitions.

### Histology

After electrophysiological experiments, animals were deeply anesthetized with mixture of pentobarbital and lidocaine and perfused transcardially with 20 ml saline followed by 20 ml 4% paraformaldehyde/0.1 M phosphate buffer, pH 7.4. The brains were removed and immersed in the above fixative solution overnight at 4◦C and then immersed in a 30% sucrose in phosphate buffer saline (PBS) for at least 2 days. The brains were quickly frozen and were cut into coronal sections with a sliding microtome (SM2010R, Leica) with a thickness of 50 or 80 µm. To verify GCaMP6s expression in cholinergic neurons within the brainstem, sections were stained for choline acetyltransferase (ChAT) and green fluorescent protein (GFP). Brain sections were first washed (5 min, three times) at room temperature (RT) with PBS-Triton-X (PBST, 0.1 M PBS and 0.3% Triton-X) and then incubated in a blocking solution (10% normal donkey serum, NDS, Sigma-Aldrich, D9663 in 0.3% PBST) for an hour at RT. Next, sections were incubated with primary antibodies against GFP (mouse anti-GFP 1:2000, Abcam, ab1218) and ChAT (goat anti-ChAT 1:400, Millipore, AB144P) in PBST and 3% NDS overnight at 4◦C. Sections were washed with PBS (5 min, three times) and incubated in a secondary antibody solution (donkey anti-mouse IgG alexa fluor 488, 1:500, Thermo Fisher Scientific, SA5-10166; Donkey antigoat IgG alexa fluor 568, 1:500, Invitrogen, A11057) in PBST and 3% NDS for 2 h at RT. Sections were washed with PBS (20 min, two times) and then incubated with a 1/5000 solution of DAPI (Thermo Fisher Scientific) in PBS for 5 min. After washing (30 min, PBS at RT), sections were mounted on glass slides with gelatine solution. Slides were left to air dry and cover slipped. Sections were observed with an epifluorescence microscope (Nikon Eclipse E600, Grayscale).

### Signal Processing

All signal processing was performed offline using MATLAB (version 2018b, MathWorks).

### Fiber Photometry

Custom-written MATLAB scripts were used to compute fluorescent signals (**Figure 3C**). To extract 405 and 470 nm signals, illumination periods were determined by detecting synchronization ON/OFF pulses for each LED (see also **Figure 3A**). The median fluorescent signal was calculated during each illumination epoch (**Figure 3C**, step 1). Because each illumination epoch consisted of pulses at 40 Hz, the fluorescent signals originally sampled at 1 kHz were effectively downsampled to 40 Hz. Photobleaching was estimated by fitting a single exponential curve and the difference between the fluorescent signal trace and the estimate was further low-pass filtered at 4 Hz given the slow kinetics of GCaMP6s (**Figure 3C**, step 2). To estimate moving artifacts, the filtered 405 nm signals were normalized based on the filtered 470 nm signals using a linear regression (**Figure 3C**, step 2). To estimate fluorescent signals, the fitted 470 nm signals were subtracted from the scaled 405 nm signals (**Figure 3C**, step 3).

by 405 nm (left) and 470 nm (right) excitation pulses in a 20-s recording (∼800 pulses). No noticeable artifact was observed. (C) Signal processing. (top) Medians were computed from raw fluorescent signals during individual illumination periods for each LED. The profile of the median values shown is over a 4-h recording period. blue, signals with 405 nm illumination. light blue, 470 nm illumination. red, exponential fit to estimate photobleaching. (middle) The median values in the top panel were subtracted from their exponential curves fitted, respectively (blue, 405 nm; light blue, 470 nm). The subtracted 405 nm signals were then linearly scaled (red) to evaluate moving artifacts. (bottom) The subtracted 470 nm signals in the middle panel were corrected by subtracting signals from the scaled 405 nm signals to provide normalized fluorescent signals.

### Electrophysiology

Vigilance states were visually scored offline as described elsewhere (Tsunematsu et al., 2020). Wakefulness, NREM sleep, or REM sleep was determined over a 4-s resolution, based on cortical EEG and EMG signals using a custom-made MATLAB Graphical User Interface. The same individual scored all recordings for consistency.

To detect P-waves, the two EEG signals from the pons were subtracted and filtered (5–30 Hz band-pass filter). If the signals

crossed a threshold, the event was recognized as P-waves. To determine the detection threshold, a 1-min segment of the subtracted EEG signals was extracted from the longest NREM sleep episode to estimate stable noise level. The noise level was estimated by computing root-mean-square (RMS) values in every 10 ms time window. The threshold was defined as mean + 5 × the standard deviation of the RMS values. The timing of P-waves was defined as the timing of the negative peak. To generate surrogate P-wave timing during REM sleep (**Figure 4E**), the number of P-waves during each REM sleep episode was held, but P-wave timing was randomly allocated during the episode. This surrogate timing was used to extract GCaMP6s signals for comparisons. To assess the reproducibility of our observation in **Figure 4**, the activation index was defined as Freal−Fsurrogate |Fsurrogate| , where Freal and Fsurrogate were average real and surrogate fluorescent signals in 1-s window from the onset of P-wave, respectively.

### Statistical Analysis

Data was presented as mean ± SEM unless otherwise stated. Student's t-test was performed in **Figure 4E** (inset).

# RESULTS

fnins-14-00148 February 20, 2020 Time: 20:13 # 7

# Simultaneous Monitoring of Pontine EEGs and Calcium Transients in Cholinergic Neurons in Freely Behaving Mice

First, we evaluated whether our system is suitable for a longterm recording over several hours from mesopontine cholinergic neurons along with electrophysiological recording. To this end, we collected six datasets from three animals. The average recording duration was 253.5 ± 8.5 min (range, 212.7–267.1 min) (**Table 2**). **Figure 3A** shows representative raw traces of pontine EEGs, raw fluorescent signals, and LED illumination pulses. Our initial concern was that optical illumination might induce optical artifacts in pontine EEG signals as reported in optogenetic experiments (Kozai and Vazquez, 2015). However, no optical artifact was observed (**Figure 3B**).

We also evaluated the stability of calcium transient amplitudes during the recording. While overall fluorescent signals decreased exponentially (**Figure 3C**), calcium transients were robust over several hours (**Figure 3C**). Thus, our approach allows for the simultaneous monitoring of both electrophysiological signals and calcium transients in freely behaving animals across the sleepwake cycle.

# Pontine Waves and Calcium Transients in Mesopontine Cholinergic Neurons

We recently reported P-waves in mice (Tsunematsu et al., 2020). Because mesopontine cholinergic neurons have been implicated in P-wave genesis (Callaway et al., 1987; Datta, 1997), we examined whether P-waves co-appear with calcium transients in mesopontine cholinergic neurons during REM sleep. **Figures 4A,B** represent the position of the optrode and co-expression of GCaMP6s and ChAT, respectively. Example signals around a REM sleep episode is shown in **Figure 4C**. We observed frequent calcium transients during REM sleep as well as phasic, large fluctuations of pEEGs. These signals were qualitatively similar to those which were separately observed in our previous study (Tsunematsu et al., 2020).

We then quantified whether these two events co-occur across recordings (**Figures 4D,E**). We found that large calcium transients appeared around the timing of P-waves (**Figure 4D**).

### To quantify this trend, we computed the average fluorescent signals and compared this with surrogate signals across all detected P-wave events (**Figure 4E**). The calcium transient from the real data was larger than that from the surrogate data and this trend was consistent across all six recordings (**Figure 4E**, inset) (p < 0.01, t-test). In addition, we did not observe changes in EMGs associated with P-waves (**Figure 4F**), indicating that these transients and P-waves were not due to movement artifacts. Thus, we confirmed that P-waves co-occurs with calcium transients in mesopontine cholinergic neurons during REM sleep.

# DISCUSSION

A combination of electrophysiological recording and calcium imaging can be used to monitor cell-type-specific activity together with sub-second neuronal events in freely behaving animals. In this study, we utilized this approach to correlate calcium transients in mesopontine cholinergic neurons with P-waves during REM sleep in mice for the first time. The same approach can be applied in various experimental contexts. Thus, our method adds a novel tool to investigate state-dependent neural circuit dynamics in vivo.

# Comparisons With Existing Systems

Although there are a handful of commercially available fiber photometry systems, our system is easy-to-build and economical. All parts of our fiber photometry system can be purchased from well-known suppliers and cost approximately 6,800 USD in total. Our in vivo electrophysiological recording system can be built with an additional budget of up to 6,000 USD. Therefore, our system offers an affordable solution to integrate in vivo electrophysiology with calcium imaging in freely behaving rodents. One of the main considerations to be made when setting up a fiber photometry system is whether a lock-in amplifier is required. Due to our offline analysis pipeline (**Figure 3C**), our system does not require a lock-in amplifier and thus drastically reduces costs as they typically costs over 5,000 USD. Although direct comparisons with commercially available systems are not straightforward due to differences in their specifications, several commercial systems (e.g., Doric Lens) offer a photometry system virtually equivalent to our system without the added functionality of electrophysiological recording and cost around 10,000 USD.


AW, wakefulness; REM, rapid eye movement sleep; NREM, non-REM sleep. Data represents mean ± standard deviation.

Others with lock-in amplification are more expensive, but some of the more sophisticated commercial systems (e.g., RZ10x, Tucker-Davis Technologies) offer a multi-color, multi-channel options, which may be attractive to some of users.

A limitation of fiber photometry in general is that it provides only population-level activity. Although an alternative approach is the use of GRIN lenses (Ghosh et al., 2011; Skocek et al., 2018; Aharoni and Hoogland, 2019), this approach is more invasive due to larger lens diameters. Hence, simultaneous calcium imaging of individual neurons and electrophysiological monitoring may be challenging.

### Implications of Findings

fnins-14-00148 February 20, 2020 Time: 20:13 # 8

In our recent study (Tsunematsu et al., 2020), we performed in vivo electrophysiological recordings of P-waves and GCaMP6s-based fiber photometry in mesopontine cholinergic neurons, separately. In the present study, we investigated the temporal relationship between population activity in mesopontine cholinergic neurons and P-waves during REM sleep by simultaneously monitoring both signals. Although P-waves have been studied in several mammalian species, such as cats, monkey, and rats since the 1960s, few studies have investigated P-waves in mice (Tsunematsu et al., 2020). Previous studies suggest that cholinergic neurons play a role in the induction of P-waves (Callaway et al., 1987; Steriade et al., 1990; Datta et al., 1992; Datta, 1997). In line with this, our results directly demonstrated that indeed mesopontine cholinergic population activity co-occurs with P-waves for the first time. A limitation of GCaMP6s is that it provides only an approximate reflection of neuronal spiking activity. Therefore, the exact temporal relationship between the firing of cholinergic neurons and P-waves still need to be investigated with the use of genetically encoded voltage indicators or electrophysiological techniques with optogenetic tagging. In addition to cholinergic neurons, it would be also interesting to monitor calcium transients in different cell types across pontine nuclei to characterize neural ensemble dynamics underlying P-waves.

### Future Directions

A similar approach can be taken in different experimental settings. For example, field potentials can be monitored with cell type-specific calcium transients in task performing animals. Our system can be customized to add optogenetic stimulation by expressing red-shifted indicators and opsins sensitive to blue light (Chen et al., 2013). A bipolar electrode can be replaced by other types of electrodes to record broadband signals including spiking activity to correlate calcium transients with neuronal spiking

# REFERENCES


because various optrodes have been developed for optogenetic experiments (Ono et al., 2018; Sileo et al., 2018; Wang et al., 2018). The fiber photometry system can be updated to utilize a tapered optic fiber to monitor activity from a larger area (Pisanello et al., 2019) or to perform cell-type-specific voltage imaging (Marshall et al., 2016; Kannan et al., 2018). In conclusion, our combinatory approach with electrophysiological recording and fiber photometry offers an affordable, but powerful solution to interrogate state-dependent neural circuit dynamics across various brain regions and behavioral states.

# DATA AVAILABILITY STATEMENT

The data for this article can be found at https://doi.org/10.15129/ c7bb43e9-ffa5-490b-9fb2-41250c2ce449.

# ETHICS STATEMENT

The animal study was reviewed and approved by the United Kingdom Home Office (PPL 70/8883) and all protocols were performed in accordance with the Animals (Scientific Procedures) Act of 1986.

# AUTHOR CONTRIBUTIONS

AP and SS designed and conceived the project and analyzed the data. AP and NM developed the recording system. AP performed all experiments. NM created the construction manual of the photometry system. AP, NM, and SS wrote the manuscript. KM and SS supervised NM and AP, respectively.

# FUNDING

This work was supported by the BBSRC (BB/M00905X/1), Leverhulme Trust (RPG-2015-377), Alzheimer's Research UK (ARUK-3033bb-CRT), and Action on Hearing Loss (S45) to SS.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2020.00148/full#supplementary-material



**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Patel, McAlinden, Mathieson and Sakata. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Dieckol, a Major Marine Polyphenol, Enhances Non-Rapid Eye Movement Sleep in Mice via the GABAA-Benzodiazepine Receptor

Minseok Yoon<sup>1</sup> , Jin-Soo Kim<sup>2</sup> , Sangwoo Seo<sup>3</sup> , Kiwon Lee3,4, Min Young Um<sup>1</sup> , Jaekwang Lee<sup>1</sup> , Jonghoon Jung<sup>1</sup> and Suengmok Cho5\*

<sup>1</sup> Korea Food Research Institute, Wanju-gun, South Korea, <sup>2</sup> Department of Seafood Science and Technology, Institute of Marine Industry, Gyeongsang National University, Tongyeong, South Korea, <sup>3</sup> WCU Biomodulation Major, Department of Agricultural Biotechnology, Seoul National University, Seoul, South Korea, <sup>4</sup> Advanced Institutes of Convergence Technology, Seoul National University, Suwon, South Korea, <sup>5</sup> Department of Food Science and Technology/Institute of Food Science, Pukyong National University, Busan, South Korea

### Edited by:

Michael Lazarus, University of Tsukuba, Japan

### Reviewed by:

Zhi-Li Huang, Fudan University, China Yoan Chérasse, University of Tsukuba, Japan

> \*Correspondence: Suengmok Cho scho@pknu.ac.kr

### Specialty section:

This article was submitted to Neuropharmacology, a section of the journal Frontiers in Pharmacology

Received: 09 October 2019 Accepted: 30 March 2020 Published: 17 April 2020

### Citation:

Yoon M, Kim J-S, Seo S, Lee K, Um MY, Lee J, Jung J and Cho S (2020) Dieckol, a Major Marine Polyphenol, Enhances Non-Rapid Eye Movement Sleep in Mice via the GABAA-Benzodiazepine Receptor. Front. Pharmacol. 11:494. doi: 10.3389/fphar.2020.00494 We had previously demonstrated that phlorotannins, which are marine polyphenols, enhance sleep in mice via the GABAA-benzodiazepine (BZD) receptor. Among the constituents of phlorotannin, dieckol is a major marine polyphenol from the brown alga Ecklonia cava. Although phlorotannins are known to exert hypnotic effects, the sleepenhancing effect of dieckol has not yet been determined. We evaluated the effect of dieckol on sleep-wake state of mice by analyzing electroencephalograms (EEGs) and electromyograms. Flumazenil, a GABAA-BZD antagonist, was used to investigate the molecular mechanism underlying the effects of dieckol on sleep. The polygraphic recordings and corresponding hypnograms revealed that dieckol accelerated the initiation of non-rapid eye movement sleep (NREMS); it shortened sleep latency and increased NREMS duration. According to the change in time-course, dieckol showed sleep-enhancing effects by increasing the amount of NREMS and decreasing wakefulness during the same hours. Additionally, sleep quality was evaluated by analyzing the EEG power density, and dieckol was found to not affect sleep intensity while zolpidem was found to reduce it. Finally, we treated mice with zolpidem or dieckol in combination with flumazenil and found the latter to inhibit the sleep-enhancing effect of dieckol and zolpidem, thereby indicating that dieckol exerts sleep-enhancing effects by activating the GABAA-BZD receptor, similar to zolpidem. These results implied that dieckol can be used as a promising herbal sleep aid with minimal side effects, unlike the existing hypnotics.

Keywords: dieckol, phlorotannins, marine polyphenols, sleep, electroencephalogram, hypnotic

### INTRODUCTION

Insomnia is a highly prevalent complaint in modern society (Jespersen et al., 2015). Sleep deprivation exerts negative effects on physical and mental performance, mood, as well as the immune system; overall, it affects the quality of life (Alami et al., 2018). With insomnia becoming increasingly prevalent, herbal sleep aids are gaining popularity worldwide as alternatives to prescription drugs for improving sleep quality or treating insomnia (Meletis and Zabriskie, 2008). Therefore, the sleepenhancing effects of herbal plants or phytochemicals have been reported widely.

GABAA receptors have been considered to be important molecular targets for the development of sleep-enhancing drugs and herbal sleep aids. In fact, polyphenols, mainly flavonoids, exert their hypnotic effects through the positive allosteric modulation of GABAA receptors (Johnston, 2015). For example, honokiol and magnolol increase non-rapid eye movement sleep (NREMS) in mice by acting on the GABAAbenzodiazepine (BZD) receptor (Chen et al., 2012; Qu et al., 2012). Glabrol, which is licorice component, has been characterized as a GABAA-BZD receptor ligand that exhibits hypnotic effects (Cho et al., 2012a). Although many studies have been conducted to reveal the hypnotic effects of polyphenols, they have all been limited to terrestrial plants (Cho and Shimizu, 2015). Recently, we reported for the first time that phlorotannins, which are marine polyphenols, enhance sleep in mice via the GABAA-BZD receptor (Cho et al., 2014). Phlorotannins consist of oligomers and polymers of phloroglucinol as the basic unit (Koivikko et al., 2007). Currently, approximately 150 phlorotannin compounds, such as dieckol, bieckol, eckol, triphlorethol A, and trifucol, have been identified from various brown seaweeds (Kim et al., 2014). Among the constituents of phlorotannin, dieckol is regarded as an indicator of phlorotannin extracts from brown alga based on the amount of it present (Shibata et al., 2002; Shibata et al., 2004; Cho et al., 2014). Although hypnotic effects of phlorotannins have been reported, the effect of dieckol on sleep remains to be investigated.

In this study, we aimed to investigate whether dieckol indeed exhibits sleep-enhancing effects by analyzing its effects on the sleep-wake profiles of C57BL/6N mice using recorded electroencephalograms (EEGs) and electromyograms (EMGs). In addition, the underlying GABAergic mechanism of dieckol was delineated using flumazenil, an antagonist of the GABAA-BZD receptors.

### MATERIALS AND METHODS

### Isolation of Dieckol From Phlorotannin Preparation

To obtain highly pure phlorotannin preparation (PRT), the ethanol extract of Ecklonia cava was purified using a Diaion HP-20 resin (Mitsubishi Chemical Industries Ltd., Tokyo, Japan) (Cho et al., 2014). The total content of phlorotannin was 900 mg of phloroglucinol equivalents/g of dry extract, as determined by the Folin-Ciocalteu method, and standardized to 67 mg/g dry extract of dieckol. Next, PRT (250 g) was partitioned by ethyl acetate (EA, 110 g) and H2O (132 g). The EA fraction was subjected to SiO2 (Kiesel gel 60, Merck, Darmstadt, Germany) column chromatography (CC). The column was eluted using mixtures of CHCl3-methanol (MeOH) at a ratio of 10:1, 6:1, 3:1, and 1:1 in sequence, and the eluates were collected into five sub-fractions (E1 – E5) by thin-layer chromatography. The sub-fraction E4 (68 g) was further separated with Sephadex LH-20 CC (Ø 2.5 cm × 50 cm, 80% MeOH) to get four sub-fractions (E4-1–E4-4), from which the subfraction E4-3 (850 mg) was subjected to ODS-C18 reverse chromatography. Finally, highly pure dieckol (87.2%) (Figure 1B) was obtained by Sephadex LH-20 CC.

### Animals and Treatment

Eleven-week-old C57BL/6N male mice (weighing 25–28 g) were obtained from Koatech Animal Inc. (Pyeongtaek, Korea) and acclimatized for a week before being used in experiments. The animals were randomly assigned to different experimental

groups. Theywere housed in individual cages andfed sterilizedfood and water. The cages were placed in an insulated sound-proof recording room with automatically controlled light/dark cycle (12 h/12-h, respectively; lights on at 05:00, illumination intensity of approximately 300 lux). Ambient temperature and relative humidity were maintained at 23 ± 0.5°C and 55 ± 2%, respectively.

Dieckol was freshly dissolved in sterile saline including 2% dimethyl sulfoxide (DMSO) and 0.5% carboxymethyl cellulose (CMC) before use. Mice were divided into three groups (n = 6–7 per group), and were orally administered (p.o.) dieckol (50, 100, or 150 mg/kg) at 17:00 on the day of the experiment. Zolpidem (Ministry of Food and Drug Safety, Cheongwon-gun, Chungcheongbuk-do, Korea), a potent GABAA-BZD agonist, was selected as the positive control (Figure 1A). Flumazenil (Sigma-Aldrich Inc., St. Louis, MO, USA) was dissolved in sterile saline for intraperitoneal injection 15 min prior to oral administration of vehicle, dieckol or zolpidem. For baseline data, mice were treated with the vehicle at 17:00 (p.o.).

### Vigilance State Analysis Based on Polygraphic Recordings

Figure 2Ashows the sleep analysis with respect to the experimental procedure and timeline. To record polygraphic signals, a head mount (#8201; Pinnacle Technology, Inc., Lawrence, KS, USA) equipped with EEG and EMG electrodes was implanted in mice under pentobarbital anesthesia (50 mg/kg, i.p.) (Cho et al., 2014). The head and neck of anesthetized mouse were shaved and cleaned with 70% alcohol before surgery. The anesthetized mouse was incised and the front edge of head mount was inserted 3 mm anterior to the bregma on skull. Four stainless-steel screws for EEG recording were inserted into the skull. Two EMG wire electrodes were placed bilaterally into the nuchal muscles. Dental cement was used to fix the head mount on the skull. Once surgery was completed, the mice were moved to separate cages for recovery at least for a week. Three to four days before the experiments, animals were adapted to the recording conditions. EEG and EMG recordings were performed under freely moving condition using

EMG, electromyogram; FFT, fast Fourier transform; NREMS, non-rapid eye movement sleep; REMS, rapid eye movement sleep; Wake, wakefulness.

a slip ring. The EEG and EMG signals were collected using the PAL-8200 data acquisition system (Pinnacle Technology, Inc.). All signals were amplified (100×), filtered (high-pass filter: 0.5 Hz for EEG and antialiasing filter: 10 Hz for EMG), and stored at a sampling rate of 200 Hz. Sleep-wake cycle was monitored over 48 h, including baseline data acquisition and experimentation. Recording of baseline was conducted over 24 h in each mouse, beginning at 17:00. Obtained baseline from each mouse was used as the control. Mice were considered to be falling asleep when there was no detectable signal in EMG. The vigilance states were automatically classified by a 10-s epoch as wakefulness (Wake), rapid eye movement sleep (REMS), or non-REM sleep (NREMS) using SleepSign version 3.0 (Kissei Comtec, Nagano, Japan) according to the standard criteria (Kohtoh et al., 2008). At the end of step, visual examinationwas done to define sleep-wake stages and to correct, if necessary. Sleep latency was measured by the time taken for the first NREMS episode to appear (lasting for at least 120 s) from the time of drug administration. Delta activity (the range of 0.5–4 Hz) during NREMS was first averaged across the individual animal and then they were summed up, and subsequently normalized to obtain the percentage of corresponding mean delta activity during NREMS. Figure 2B shows the typical waveforms of EEG and EMG, and the fast Fourier transform spectra of delta and theta waves. Bouts of each stagewere defined as the periods of one or more consecutive 10-s epochs (Figure 2C).

### Statistical Analysis

Data plotted in figures are indicated as the mean ± SEM. Statistical analysis was performed with GraphPad Prism 5.0 software (GraphPad Software Inc., San Diego, CA, USA). Sleep latency, amount of NREMS and REMS, mean duration, stage transition number, and the number and duration of bouts were analyzed by the paired Student's t-test. For time-course of the hourly amounts of each stage and delta activity were assessed by two-way ANOVA with Bonferroni post-test. P-values less than 0.05 were considered significant for all statistical tests.

# RESULTS

### Effect of Dieckol on Sleep Latency and Amounts of REMS and NREMS in C57BL/6N Mice

To investigate the effect of orally administered dieckol on sleep structure, we analyzed the sleep architecture in C57BL/6N mice using EEG and EMG recordings. Figure 3A shows a

representative example of polygraphic recordings and corresponding hypnograms from a single mouse during the first 3 h after treatment with vehicle, dieckol, and zolpidem. Values of sleep latency were 19.2 ± 3.2 min in mice administrated 150 mg/kg dieckol and 14.8 ± 1.7 min in those administered 10 mg/kg zolpidem (Figure 3B). Sleep latency with dieckol and zolpidem was significantly shorter than that after vehicle treatment, specifically 44.5 ± 5.9 and 42.8 ± 5.8 min, respectively. Dieckol induced short sleep latency in mice, indicating that it accelerated the initiation of NREMS, similar to zolpidem.

To compare the effect of dieckol and zolpidem on sleep duration, we computed the total amounts of NREMS and REMS for 2 h after the administration of both chemicals (Figure 3C). As expected, the zolpidem-administered group, as a positive control, showed significantly (p < 0.01) increased total NREMS (by 1.6-fold) compared to that in the vehicle-treated group. Dieckol administration increased NREMS duration in a dose-dependent manner. Dieckol concentrations of 100 and 150 mg/kg significantly increased NREMS by 1.4-fold (p < 0.05) and 1.7-fold (p < 0.01), respectively, as compared to the vehicle. In particular, there was no significant differences in sleep latency and NREMS amount between 150 mg/kg dieckol and 10 mg/kg zolpidem group. However, dieckol at 50 mg/kg did not affect either sleep latency or NREMS amount. In addition, changes in REMS amount were not found after dieckol or zolpidem administration, as compared to vehicle.

### Effects of Dieckol on Time-Course Changes of Sleep-Wake Stage in C57BL/6N Mice

Figure 4 shows the time-course changes of the hourly amounts of NREMS, REMS, and Wake. Dieckol (150 mg/kg) showed a 2.1- and 1.30-fold increase in the amount of NREMS during the first and second hours, respectively, relative to the vehicle (Figure 4A). This improvement in NREMS was accompanied by reduction in Wake during the same hours. After an initial increase in NREMS, sleep architecture did not change significantly during the subsequent periods. Moreover, dieckol at 100 mg/kg exhibited similar time-course profiles, although the enhancing effect of sleep was minor, lasting over approximately 2 h after administration (data not shown). Unlike dieckol,

**277**

zolpidem at 10 mg/kg significantly enhanced NREMS by 7 h (Figure 4B). Neither of dieckol and zolpidem was significantly different compared to each vehicle over 24 h.

### Effects of Dieckol on Sleep-Wake Episode and Delta Activity

To evaluate the sleep-enhancing effect of dieckol, mean duration and total number of NREMS, REMS, and Wake episodes were analyzed. Dieckol (150 mg/kg) and zolpidem (10 mg/kg) significantly decreased the mean duration of Wake by 53.8% (p < 0.01) and 38.0% (p < 0.05), respectively, without affecting the mean duration of NREMS or REMS (Figure 5A). Moreover, both dieckol and zolpidem produced an increase in the number of NREMS bouts by 1.7- and 2.2-fold, and those of Wake bouts by 1.6- and 2.1-fold, respectively (Figure 5B). REMS bouts remained unchanged. In addition, both dieckol and zolpidem showed an increase in the number of state transitions from Wake to NREMS and NREMS to Wake (Figure 5C). However, we did not find any change in the number of transitions from NREMS to REMS or from REMS to Wake. Finally, dieckol induced an increase in bout number for NREMS, ranging from 10 to 60 s each. Zolpidem also raised the number of bouts of short-length NREMS; it resulted in significant differences in middle-length NREMS bouts, ranging from 60 to 480 s (Figure 6A).

In order to evaluate the sleep quality, delta activity was analyzed from the EEG power density during NREMS. As shown in Figure 6B, there was no significant difference in EEG power density (0–20 Hz), and delta activity (frequency range of 0.5–4 Hz), for NREMS between the dieckol- and vehicle-treated mice. However, zolpidem significantly (p < 0.05) decreased the

FIGURE 6 | (A) Changes in the number of NREMS and Wake bouts of different durations in C57BL/6N mice after oral administration of dieckol or zolpidem. Light and dark bars indicate the baseline day (vehicle administration) and experimental day (dieckol or zolpidem administration), respectively. Each column represents the mean ± SEM (n = 6–7) with data points. \*p < 0.05 and \*\*p < 0.01, significant difference compared to the vehicle (paired Student's t-test). (B) Electroencephalogram (EEG) power density curves of NREMS caused by dieckol and zolpidem. Delta activity, an index of sleep intensity, is shown in the inset histogram. The dash (▬) represents the range of the delta wave (0.5‒4 Hz). \*p < 0.05, significant difference compared to the vehicle (two-way ANOVA with Bonferroni post-test). NREMS, non-rapid eye movement sleep.

delta activity. These results suggested that, unlike zolpidem, dieckol increased sleep quantity without loss of sleep intensity.

### Possible Mechanism of Action of Dieckol

In our previous study, we had suggested that phlorotannin extracts, with dieckol as the major component, enhance NREMS through the activation of BZD binding site of GABAA receptor (Cho et al., 2014). This led us to investigate the sleep-enhancing effect of dieckol, if at all, through the GABAergic system. To confirm the action of dieckol on GABAA-BZD receptor, mice were pretreated with flumazenil, which is the specific GABAA-BZD receptor antagonist. There was no significant difference in sleep architecture of mice with the flumazenil (1 mg/kg) injection alone (Figure 7A). Effects of zolpidem, the GABAA-BZD agonist, were completely inhibited by flumazenil. Flumazenil also fully suppressed the sleep-enhancing effect of dieckol. The time-course changes in each stage over a period of 24 h revealed that dieckol did not affect sleep architecture in presence of flumazenil, in comparison with the vehicle (Figure 7B). These results implied that the sleep-enhancing effects of dieckol could be via the modulation of GABAA receptor at the BZD binding site.

# DISCUSSION

In this study, we found both dieckol and zolpidem to reduce sleep latency and increase NREMS without altering REMS. Moreover, we showed the amounts of NREMS between zolpidem at 10 mg/kg and dieckol at 150 mg/kg to not be significantly different (p > 0.05). Dieckol exerted a significant effect during the first 2 h after administration. During the subsequent period, we did not observe further significant disruption of sleep architecture. These results suggested that dieckol induces NREMS without causing harmful effects after sleep induction (Masaki et al., 2012). Considering the finding that dieckol reduced the mean duration of Wake episodes and increased the total number of NREMS bouts, we could strongly

in mice. Each column represents the mean ± SEM (n = 6-7) with data points. \*\*p < 0.01, significant difference compared to the vehicle (paired Student's t-test). (B) Time-course changes in NREMS, REMS, and Wake after administration of vehicle, flumazenil, and dieckol. The horizontal filled and open bars on the X-axis (time) indicate the 12 h dark and 12 h light periods, respectively.

suggest that maintenance of Wake was inhibited by dieckol (Masaki et al., 2012). Delta activity is a marker of the depth or intensity of NREMS (Winsky-Sommerer, 2009). The main effects of BZD agents have been reported as shortening of sleep latency and enhancement of sleep duration while suppressing delta activity (Lancel et al., 1997; Tobler et al., 2001; Kopp et al., 2004a). Zolpidem has been reported to produce a decrease in delta activity during NREMS (Kopp et al., 2004b; van Lier et al., 2004; Alexandre et al., 2008). Similarly, we also found delta activity to be significantly decreased by zolpidem; however, it remained unchanged by dieckol. Our results showed that dieckol produces NREMS similar to physiological sleep (Xu et al., 2014). Classical BZD-hypnotics have high affinity for a1, a2, a3, and a<sup>5</sup> subunits of GABAA receptors (Barnard et al., 1998; Möhler et al., 2002). Zolpidem has been reported to have a high affinity to a1-GABAA receptors, and an intermediate affinity to a2- and a3- GABAA receptors. However, it has a limitation in binding to a5- GABAA receptors (Crestani et al., 2000). Although zolpidem suppresses delta activity by binding to a<sup>2</sup> subunits of GABAA receptors, as does diazepam (Kopp et al., 2004a), the hypnotic action of zolpidem is mediated by GABAA receptors containing a<sup>1</sup> subtypes (Sanna et al., 2002; Sanger, 2004; Tsai et al., 2013). Therefore, we suggest that the action of dieckol and zolpidem is mediated by different GABAA receptor subunits.

We had previously reported that dieckol works as a ligand of GABAA-BZD receptor and its K<sup>i</sup> (binding affinity) value for [3 H] flumazenil binding is 3.072 µM (Cho et al., 2012b). These results suggested that the hypnotic effect of dieckol is attributed to GABAergic pathways. For this reason, we examined the effect of dieckol with co-application of flumazenil, which is a specific GABAA-BZD receptor antagonist, on the sleep-enhancing effect of dieckol in vivo. Flumazenil acts as an inhibitor of the hypnotic effect of GABAA-BZD receptor agonists (such as zolpidem) by interfering with their binding sites (Johnston, 2005). In the current study, we found the sleep-enhancing effect of dieckol, similar to that of zolpidem, to be completely blocked by flumazenil compared to that in the vehicle. Previously, honokiol and magnolol had been reported to promote NREMS by activating the BZD binding site of GABAA receptor, since their somnogenic effects and activation of ventrolateral preoptic area neurons were blocked by flumazenil (Chen et al., 2012; Qu et al., 2012). These findings supported our hypothesis that dieckol enhances sleep by acting as a positive allosteric modulator of GABAA receptors at the BZD binding site, similar to zolpidem.

# CONCLUSION

We demonstrated the sleep-enhancing effects of dieckol in mice and established that these effects are mediated via the GABAA-BZD receptor. Our results provide important insights that can contribute to the development of hypnotics with new structures, since dieckol, derived from brown seaweed, has a different structure compared to the polyphenols of terrestrial plants and other hypnotic drugs. However, for the development of novel polyphenol-based hypnotics, further electrophysiological studies on the in-vivo effects of chronic administration would be recommended.

# DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

# REFERENCES


# ETHICS STATEMENT

All procedures involving animals were conducted in accordance with the animal care and use guidelines of the Korea Food Research Institutional Animal Care and Use Committee (permission number: KFRI-M-12027).

# AUTHOR CONTRIBUTIONS

MY, J-SK, and SC conceived and designed the experiments. MY, MU, JL, and JJ performed the experiments. Data analysis was performed by MY and SC. The manuscript was prepared by MY, SS, and SC. KL contributed expert opinion to the manuscript correction.

# FUNDING

This research was supported by the Main Research Program (E0164501-04 and E0164503-02) of the Korea Food Research Institute (KFRI), funded by the Ministry of Science and ICT.


Conflict of Interest: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Yoon, Kim, Seo, Lee, Um, Lee, Jung and Cho. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Hyper-Activation of mPFC Underlies Specific Traumatic Stress-Induced Sleep–Wake EEG Disturbances

Tingting Lou<sup>1</sup>† , Jing Ma1,2† , Zhiqiang Wang1,2 \* † , Yuka Terakoshi<sup>1</sup> , Chia-Ying Lee<sup>1</sup> , Greg Asher<sup>1</sup> , Liqin Cao<sup>1</sup> , Zhiyu Chen3,4, Katsuyasu Sakurai<sup>1</sup> \* and Qinghua Liu1,3,4 \*

1 International Institute for Integrative Sleep Medicine (WPI-IIIS), University of Tsukuba, Tsukuba, Japan, <sup>2</sup> HIT Center for Life Sciences (HCLS), School of Life Sciences and Technology, Harbin Institute of Technology, Harbin, China, <sup>3</sup> National Institute of Biological Sciences (NIBS), Beijing, China, <sup>4</sup> Tsinghua Institute of Multidisciplinary Biomedical Research (TIMBR), Tsinghua University, Beijing, China

### Edited by:

Zhi-Li Huang, Fudan University, China

### Reviewed by:

Suengmok Cho, Pukyong National University, South Korea Christelle Anaclet, University of Massachusetts Medical School, United States

### \*Correspondence:

Zhiqiang Wang zhiqiang.wang@hit.edu.cn; zhiqiang.wangmj@foxmail.com Katsuyasu Sakurai sakurai.katsuyasu.gm@ u.tsukuba.ac.jp Qinghua Liu liuqinghua@nibs.ac.cn †These authors have contributed equally to this work

### Specialty section:

This article was submitted to Sleep and Circadian Rhythms, a section of the journal Frontiers in Neuroscience

Received: 23 November 2019 Accepted: 28 July 2020 Published: 18 August 2020

### Citation:

Lou T, Ma J, Wang Z, Terakoshi Y, Lee C-Y, Asher G, Cao L, Chen Z, Sakurai K and Liu Q (2020) Hyper-Activation of mPFC Underlies Specific Traumatic Stress-Induced Sleep–Wake EEG Disturbances. Front. Neurosci. 14:883. doi: 10.3389/fnins.2020.00883 Sleep disturbances have been recognized as a core symptom of post-traumatic stress disorders (PTSD). However, the neural basis of PTSD-related sleep disturbances remains unclear. It has been challenging to establish the causality link between a specific brain region and traumatic stress-induced sleep abnormalities. Here, we found that single prolonged stress (SPS) could induce acute changes in sleep/wake duration as well as short- and long-term electroencephalogram (EEG) alterations in the isogenic mouse model. Moreover, the medial prefrontal cortex (mPFC) showed persistent high number of c-fos expressing neurons, of which more than 95% are excitatory neurons, during and immediately after SPS. Chemogenetic inhibition of the prelimbic region of mPFC during SPS could specifically reverse the SPS-induced acute suppression of delta power (1–4 Hz EEG) of non-rapid-eye-movement sleep (NREMS) as well as most of long-term EEG abnormalities. These findings suggest a causality link between hyper-activation of mPFC neurons and traumatic stress-induced specific sleep–wake EEG disturbances.

Keywords: traumatic stress, single prolonged stress (SPS), sleep disturbances, electroencephalogram (EEG), medial prefrontal cortex (mPFC)

### INTRODUCTION

Exposure to catastrophic traumatic events could lead to severe mental and behavioral disorders, so called post-traumatic stress disorders (PTSD), which are characterized by symptoms of reexperiencing, numbing, avoidance, and hyperarousal (Germain, 2013; Khazaie et al., 2016). Sleep disturbances represent a core symptom of PTSD patients, including insomnia, nightly awakenings, nightmares, sleep paralysis, and restless sleep (Ross et al., 1989; Steine et al., 2012; Khazaie et al., 2016). Although polysomnographic studies in PTSD patients have reported abnormal sleep–wake architecture, previous studies have produced inconsistent results, such as changes in sleep amount, sleep latency, and frequency of nightly awakenings (Kobayashi et al., 2007; Yetkin et al., 2010).

There are also inconsistent results from quantitative analysis of the sleep–wake electroencephalogram (EEG) of PTSD patients (Germain, 2013; Khazaie et al., 2016). Typically, EEG signals can be decomposed into four distinct frequency bands, such as delta (1–4 Hz), theta (5–8 Hz), alpha (9–14 Hz), and beta (15–30 Hz), which may correspond to the underlying intraand inter-cellular signaling, neuronal activities of different brain regions, brain physiological states, cognitive and mental conditions (Campbell, 2009). For example, there have been reports

of increased (Woodward et al., 2000; Insana et al., 2012), decreased (Cohen et al., 2013), or no difference (Mellman et al., 2007) in the beta power of EEG during rapid-eye-movement sleep (REMS) in adult PTSD patients. Both reduced and increased delta power activity during non-rapid-eye-movement sleep (NREMS) and REMS have also been reported in PTSD patients (Woodward et al., 2000; Germain et al., 2006; Insana et al., 2012; de Boer et al., 2020; Wang et al., 2020). These conflicting findings may be attributed to the effects of many confounding variables in the experimental settings, the inter-individual differences and disease heterogeneity, such as differences in initial traumatic stimuli, analysis stages of the illness, comorbidities with other psychiatric conditions, and diversity of underlying neural mechanisms (Kobayashi et al., 2007; Yetkin et al., 2010; Germain, 2013; Baglioni et al., 2016; Khazaie et al., 2016; Deslauriers et al., 2018).

Because PTSD is a heterogeneous disorder affected by many physiological and environmental factors, the development of effective animal models to study traumatic stress-induced sleep abnormalities is urgently needed to fully understand PTSD pathogenesis and pre-clinically evaluate potential treatments (Deslauriers et al., 2018; Mysliwiec et al., 2018). Multiple traumatic stress protocols, such as single prolonged stress (SPS) (Liberzon et al., 1997; Liberzon et al., 1999; Perrine et al., 2016), inescapable foot shocks (Philbert et al., 2011; Yu et al., 2015), restraint stress (Meerlo et al., 2001; Hegde et al., 2008), predator scent stress (Sharma et al., 2018), acute and chronic social defeat stress (Meerlo et al., 1997; Kamphuis et al., 2015; Henderson et al., 2017; Olini et al., 2017; Fujii et al., 2019) have been used to develop PTSD models in rodents. Among these, SPS is a simple and well-established rodent model of traumatic stress that can reliably induce PTSD-like behavioral and physiological abnormalities (Liberzon et al., 1997, 1999; Perrine et al., 2016; Deslauriers et al., 2018).

Although sleep disturbances has long been recognized as a core symptom of PTSD, the PTSD-related sleep phenotypes remain an understudied area. Previously, the effects of SPS on the sleep–wake architecture have only been investigated in two rat studies that yielded inconsistent results (Nedelcovych et al., 2015; Vanderheyden et al., 2015). While one study showed that SPS caused an increase of REMS in the dark phase, but no change in NREMS (Vanderheyden et al., 2015), another reported that SPS reduced both NREMS and REMS in the light phase, followed by a strong rebound in NREMS and REMS in the dark phase (Nedelcovych et al., 2015). These inconsistent results across different laboratories need to be carefully re-examined (Deslauriers et al., 2018).

An important issue that has received little attention is the adopted method for quantitative analysis of the EEG power spectrum. Both absolute and relative EEG power analyses have been commonly used in the literature according to the experimental design (Campbell, 2009; Wang et al., 2018). Absolute EEG power analysis is appropriate for the longitudinal design to measure absolute changes in the EEG power spectrum before and after a traumatic event in the same subject. Relative EEG power analysis, which is calculated as the percentage of power density in a specific frequency bin in the total power of all frequency bins, is suitable for the cross-sectional design to compare the EEG alterations among different subjects. This is because the individual differences in bone thickness, skull resistance and impedance will cause variations in absolute EEG power values (Benninger et al., 1984). Most studies in PTSD patients use the cross-sectional design and, hence, relative EEG power analysis because it is impossible to measure the baseline sleep–wake architecture immediately before a traumatic event in the same individual (Vanderheyden et al., 2015). Some researchers argue that relative EEG power analysis has better test-retest reliability (Salinsky et al., 1991) and sensitivity to agedependent changes in the frequency composition of EEG signals (Clarke et al., 2001). Based on our previous studies (Funato et al., 2016; Wang et al., 2018), we recognize that relative EEG power analysis, due to the normalization process needed, is likely to miss critical changes of the EEG power spectrum that can be observed from absolute EEG power analysis.

Animal model and clinical studies of PTSD have revealed structural and functional alterations in multiple brain regions, however, the neurological correlates of traumatic stress-induced sleep abnormalities remain largely unexplored (Karl et al., 2006; Deslauriers et al., 2018; Mysliwiec et al., 2018). In particular, it has been challenging to establish the causality link between any specific brain region and traumatic stress-induced sleep abnormalities. In this study, we aim to characterize the SPS mouse model of PTSD, with an emphasis on the sleep–wake phenotypes. We found that SPS-treated mice exhibited specific changes in the sleep–wake architecture, including both short- and long-term EEG alterations. Moreover, our results suggest for the first time a causality link between the hyper-activation of medial prefrontal cortex (mPFC) neurons and the SPS-induced specific sleep–wake EEG abnormalities. This type of investigations should be important to understanding the neural mechanisms and facilitating development of effective therapies for at least a subset of PTSD patients.

# MATERIALS AND METHODS

### Animal Subjects

All mice were housed under humidity and temperature (22– 25±◦C) controlled conditions on a 12-h light–dark cycle with food and water provided ad libitum. We used 12–20 weeks old (26–33 g body weight) C57BL/6N male mice (CLEA Japan) in this study. All experimental animal procedures were approved by the Institutional Animal Care and Use Committee of University of Tsukuba. All mice were singly housed for one week before each experiment.

### Sleep Deprivation and Single Prolonged Stress

For sleep deprivation, mice were sleep deprived for 4 h from the onset of the light phase (ZT0–ZT4) by gently touching the cages when they started to recline and lower their heads in the home cage. The SPS was performed at the onset of light phase (ZT0) as previously described (Liberzon et al., 1997, 1999; Deslauriers et al., 2018). First, each mouse was restrained for 2 h in a 50 ml Falcon tube with the bottom removed. Second, the mouse was

forced to swim for 20 min in a plastic cylinder (height: 25 cm; diameter: 18.5 cm) filled with water (21–24◦C), such that the mouse's hind limbs could not touch the bottom. Third, after recuperating for 15 min in a new cage, the mouse was exposed to ether until general anesthesia (no more than 5 min). Finally, the mouse was returned to its home cage (around ZT3.5) and sleep deprived until ZT4 by gently touching the cages.

### EEG/EMG Electrode Implantation

Mice (8–10 weeks old) were implanted with the EEG/electromyogram (EMG) electrodes under anesthesia by isoflurane (3% for induction and 1% for maintenance). Briefly, four electrode pins were lowered to the dura under stereotaxic control. Two electrodes for EEG signals were positioned over the frontal and occipital cortices [anteroposterior (AP): 0.5 mm, mediolateral (ML): 1.3 mm, dorsoventral (DV): −1.3 mm; and AP: −4.5 mm, ML: 1.3 mm, DV: −1.3 mm]. Two electrodes with flexible wires for EMG recording were threaded through the dorsal neck muscle. Afterward, the EEG/EMG electrodes were glued to the skull with dental cement. Mice were individually housed following surgery, and allowed a minimum recovery period of 7 days.

### Sleep–Wake Behaviors Analysis

The sleep–wake behaviors were analyzed as previously described (Funato et al., 2016; Wang et al., 2018). Mice were tethered to a counterbalanced arm (Instech Laboratories) that allowed free movement and exerted minimal weight, and acclimatized to the recording chamber around 7 days before recording. EEG/EMG signals were recorded at the age of 12–20 weeks; age-matched animals were used in each experiment for control and treatment groups. EEG/EMG data were analyzed using a MatLab (MathWorks)-based semi-automated staging software followed by manual correction. EEG signals were decomposed by fast Fourier transform analysis for 1 to 30 Hz with 1 Hz bins. Sleep/wake states were scored in 20 s epoch as wake (low amplitude, fast EEG and high amplitude, and variable EMG), REMS [dominant theta (5–8 Hz) EEG and EMG atonia], or NREMS [high amplitude delta (1–4 Hz) EEG and low EMG tonus]. Absolute (arbitrary units) and relative EEG power density analysis was performed to examine the delta (1–4 Hz), theta (5–8 Hz), alpha (9–14 Hz), and beta (15–30 Hz) activities during NREMS, REMS, or wake state at indicated ZT period. To minimize the inter-individual differences for following statistical analysis, absolute EEG power data of each individual animal for the corresponding NREMS, REMS, or wake state was normalized to the mean power from ZT8 to ZT11 of baseline recording day of all animals used within each corresponding experiment, which is at the end of the major rest period (Franken et al., 2001; Mang et al., 2016). Relative EEG power density analysis (%) is defined by the ratio of a specific frequency bin to the total power over all frequency bins (1–30 Hz). In hourly analysis of sleep– wake architecture, each data point represents the mean value of either duration or EEG power density in the following 1 h during NREMS, REMS, and wake states. Researchers were blinded to genotype and/or treatment before data analysis, and only animals with unreadable EEG signals were excluded from final analysis.

# Behavioral Experiments

Two groups of mice were sleep deprived for 4 h (SD4) or exposed to SPS treatment, respectively. On the 7th day after the SD4/SPS procedure, the tail suspension test (TST) was performed as previously described (Can et al., 2012b). Each mouse was suspended in the hook of an open front TST box, approximately 50 cm above the surface of table with a small piece of adhesive tape placed 2 cm away from the tip of the tail. The duration of immobility was recorded for 10 min by a video camera positioned in front of the test box. Mice were considered immobile only when they hung passively and were completely motionless. Mice were returned to their home cages to rest for at least 1 h, and then the forced swim test (FST) was performed as previously described (Can et al., 2012a). The mice were placed individually for 10 min in a plastic cylinder (height: 25 cm; diameter: 18.5 cm) filled with water (21–24◦C) to a depth of 14 cm. The water depth was adjusted so that the animal's hind limbs cannot touch the bottom. Water was changed between subjects. All test sessions were recorded by a video camera positioned on the top of the plastic cylinder. Mice were considered to be immobile when floating motionless or making only those movements necessary to keep its head above the water. The duration of immobility was measured manually by an observer blind to group assignment.

### Immunohistochemistry

All mice were singly housed at least for 1 week before experiments. After experimental treatments, test mice were allowed to recover in the home cage. Specifically, at least one paired control and stressed mice brains were harvested at 30 min (ZT4.5) or 3.5 h (ZT7.5) after SPS or SD4 treatment at the same experimental day and processed at the same time in the following steps. At indicated ZT time, paired control and stressed mice were rapidly anesthetized with pentobarbital (50 mg/kg, i.p.), and then transcardially perfused with 0.1 M phosphate buffer saline, pH7.4 (PBS), followed by 4% paraformaldehyde in PBS (PFA). Whole brain was dissected and post-fixed for 24-h in 4% PFA at 4 ◦C, and then cryoprotected with 30% sucrose (wt/vol) in PBS for 48 h at 4◦C. The tissues were frozen in the Tissue-Tek O.C.T compound (Sakura Finetek), and 80-µm-thick coronal sections were cut on a cryostat (CM3050S, Leica). For c-Fos staining, the floating brain sections were washed three times with PBS for 5 min each, incubated with 1% Triton X-100 in PBS for 2 h. The sections were incubated in 10% Blocking One (nacalai tesque) in PBS with 0.1% Triton-X-100 (blocking solution) for 1-h at room temperature. The sections were incubated with rabbit anti-c-Fos antibody (1:2,500, EMD Millipore, ABE457) in blocking solution at 4◦C overnight. After washing three times with PBS, the sections were incubated with Donkey anti-rabbit Alex488 (1:500, Thermo Fisher R37118) and Fluorescent Nissl Stain (1:500, Thermo Fisher N21479) in Blocking solution at 4◦C overnight. After washing three times with PBS, the sections were mounted and covered with coverslip. All images were acquired using the Zeiss LSM700 confocal microscope with a 10× objective lens (NA = 0.45) under the Zen 2010 software (Carl Zeiss). The c-Fos positive neurons were counted in all sections from the same mouse brain (ImageJ). No normalization was performed for the c-fos expression at ZT4.5 and ZT7.5. Representative images shown in the figures were chosen from a similar region based on morphology.

### In situ Hybridization

fnins-14-00883 August 17, 2020 Time: 15:15 # 4

The cDNA fragments of mouse c-fos, vGlut1, and vGat were amplified by PCR with antisense primers containing T3 or T7 promoter sequence. In vitro transcription was performed with PCR-amplified template using T3 RNA polymerase (Promega) or T7 RNA polymerase (Roche) for the synthesis of antisense probes. Fluorescent two-color in situ hybridization was performed based on a basic method (Ishii et al., 2017). Briefly, mice were subjected to SPS treatment and, after 30 min, were anesthetized with pentobarbital (50 mg/kg, i.p.) followed by perfusion with 4% paraformaldehyde (PFA) in PBS. Brain slices (40 µm) were treated with protease K (Roche, cat#03115887001), followed by acetylation. The brain slices were incubated with hybridization buffer containing RNA probe mix at 60◦C for 16 h. After stringent washing, brain slices were incubated with horseradish peroxidase (HRP) conjugated anti-FITC antibody (PerkinElmer, 1:1,000) or HRP-conjugated anti-Dig antibody (Roche; 1:1,000) overnight at 4◦C. TSA system (TSA-FITC or TSA-Biotin; PerkinElmer) was applied to visualize the mRNA signal. All images were acquired using the Zeiss LSM700 confocal microscope with a 10× objective lens (NA = 0.45) under the Zen 2010 software (Carl Zeiss). The c-fos, vGlut1, and vGat positive neurons were counted in all sections from the same mouse brain (Image J).

### Stereotaxic AAV Injection and Drug Administration

For bilateral injection of adeno-associated viruses (AAV) (AAV2/9-CMV-mCherry; AAV2/9-hSyn-hM4Di–mCherry) into the mPFC, male mice (8–10 weeks old) were anesthetized with isoflurane (3% for induction and 1% for maintenance) and placed in a stereotaxic frame (David Kopf Instruments). An incision was made on the top of the skull, and the skin was retracted and connective tissue gently scraped away. After exposing the skull and cleaning the surface with 3% hydrogen peroxide, bilateral craniotomies (∼1 mm diameter each) were made to allow virus delivery (500 nl at 100 nl/min). Stereotaxic coordinates of virus injection were based on Paxinos and Franklin mouse brain atlas (AP: −1.94 mm, L: ±0.4 mm, DV: −2 mm). For EEG/EMG analysis of AAV-injected mice, the EEG/EMG electrode implantation was performed immediately following AAV injection. Clozapine N-oxide (CNO; Cayman Chemical, Item No. 12059) was dissolved in saline. Vehicle (0.9% saline) or CNO (3 mg/kg) was administered by intraperitoneal injection at ZT0 and ZT3.5 before the mouse returned to the home cage.

### Statistical Methods

GraphPad Prism 6 software was used for statistical tests. No statistical method was used to predetermine sample size. Randomization was not used. Following two-way ANOVA analysis of variance, Sidak's test was performed to compare a set of means, repeated measures was performed for matched subject comparisons. Paired t-test was performed for matched subject comparisons, whereas unpaired t-test for group comparisons. The complete sample size, statistical test method and results for each comparison are reported in the figure legends and described in detail in **Supplementary Table 1**. P < 0.05 was considered statistically significant. Unless otherwise noted, all experimental subjects are biological replicates and at least two independent experiments were performed.

# RESULTS

We adopted the standardized SPS paradigm to investigate the effects of traumatic stress on the sleep–wake architecture in wildtype C57BL/6N male mice. We used a longitudinal experimental design by sequentially comparing sleep/wake changes before and after 4-h sleep deprivation by gentle handling (SD4, ZT0–ZT4) or SPS (ZT0–ZT4) on the same subjects (**Figure 1**). For the SD4 segment, after continuous 24-h baseline (SD4-BL) recording, all test mice are subjected to SD4 (SD4-D1) and continuously monitored for EEG and EMG in the home cage until the seventh day (SD4-D7). After 1–3 days' rest, the same mice would be subjected to SPS (ZT0–ZT4) to study how traumatic stress caused sleep–wake disturbances. For the SPS segment, after continuous 24 h baseline (SPS-BL) recording, all test mice were subjected sequentially to 2-h restraining, 20-min forced swimming, and up to 5-min anesthesia by ether (SPS-D1), and followed by continuous EEG/EMG recording until the 7th day (SPS-D7). This longitude design gave us two important advantages over previous studies: (a) comparison of SPS and SD4 could distinguish the specific effects of SPS (as opposed to prolonged wakefulness) on the sleep–wake architecture; (b) the baseline and post-SD4 or post-SPS EEG/EMG recordings of the same mice allowed for both absolute and relative EEG power analysis to comprehensively evaluate the SPS-induced short-term (D1) and long-term (D7) EEG abnormalities, which is not possible in previous SPS rat studies (Nedelcovych et al., 2015; Vanderheyden et al., 2015).

### Traumatic Stress Induces Acute Changes in Sleep/Wake Duration

To examine the acute effect of traumatic stress on the sleep–wake architecture, we compared the EEG/EMG data of test mice on the day before (SD4-BL or SPS-BL) and after SD4/SPS (SD4-D1 or SPS-D1) (**Figure 2**). It is important to note that there were essentially no difference in the baseline sleep–wake pattern of the same mice before SD4 and SPS treatment (SD4-BL vs SPS-BL), making it possible to directly compare the effects of SPS and SD4 on the sleep–wake architecture (**Figure 2A**, **Supplementary Figure 1A** and **Table 1**). On the day after SD4, there was on average a 58.1% reduction in REMS duration (SD4-D1, 2.6 ± 1.9 min vs SD4-D0, 6.2 ± 1.9 min) at the first hour (ZT4) after sleep deprivation, and a 72.3 and 8.2% rebound of REMS (SD4-D1, 25.5 ± 7.8 min vs SD4-D0, 14.8 ± 6.2 min) and NREMS (SD4-D1, 254.2 ± 53.6 min vs SD4-D0, 234.9 ± 50.6 min) in the dark phase, respectively (**Figure 2A** and **Supplementary Figure 1B**). On the day after SPS, there was on average a 96.6 and 47.5% reduction in REMS duration (SPS-D1, 0.2 ± 0.6 min vs SPS-BL, 5.9 ± 1.5 min) and NREMS duration (SPS-D1,

21.5 ± 11.2 min vs SPS-BL, 41.0 ± 6.7 min) at the first hour (ZT4) after SPS, as well as a 183.9 and 33.9% rebound of REMS (SPS-D1, 44.0 ± 10.1 min vs SPS-BL, 15.5 ± 7.3 min) and NREMS (SPS-D1, 325.4 ± 49.5 min vs SPS-BL, 243.1 ± 37.6 min) in the dark phase, respectively (**Figure 2A** and **Supplementary Figure 1C**).

By direct comparison of the SPS-D1 vs SD4-D1 data, we found that SPS, relative to SD4, resulted in about 36.2% less NREMS at ZT4 (SPS-D1, 21.5 ± 11.2 min vs SD4-D1, 33.7 ± 10.9 min) and 56.3% less REMS during ZT4-6 (SPS-D1, 6.3 ± 3.1 min vs SD4-D1, 14.4 ± 4.0 min) (**Figure 2A**). In the dark phase, SPS mice spent 28 and 72.5% more time than SD4 mice in NREMS (SPS-D1, 325.4 ± 49.5 min vs SD4-D1, 254.2 ± 53.6 min) and REMS (SPS-D1, 44.0 ± 10.1 min vs SD4-D1, 25.5 ± 7.8 min), respectively (**Figure 2B**). Thus, our results indicate that traumatic stress by SPS can induce specific changes in the sleep–wake architecture that are distinct from sleep deprivation.

# Traumatic Stress Induces Short-Term Sleep/Wake EEG Abnormalities

Similarly, the baseline sleep/wake EEG power spectrum of the test mice was essentially the same before SD4 and SPS treatment (**Supplementary Figure 2**; SD4-BL vs SPS-BL). By absolute EEG power analysis, SD4 resulted in a broad increase over baseline in all frequency bands of EEG signals during NREMS in the light phase, particularly in the first hour (ZT4) after sleep deprivation (↑56.4% delta; ↑24.3% theta; ↑15.1% alpha; ↑19.0% beta) (**Supplementary Figure 3**; SD4-D1 vs SD4-BL). On the other hand, SPS resulted in a 13.2 and 9.5% increase over baseline, respectively, in the delta and theta power of EEG signals during NREMS at ZT4 (**Supplementary Figure 3A**; SPS-D1 vs SPS-BL). Comparison of the SPS and SD4 data reveals that SPS, relative to SD4, caused a broad suppression in all frequency bands of EEG signals during NREMS at ZT4 (↓26.1% delta; ↓12.1% theta; ↓18.0% alpha; ↓16.2% beta) and in the dark phase (↓9.0% delta; ↓11.0% theta; ↓7.2% alpha; ↓9.4% beta), as well as a specific suppression (↓9.4%) of NREMS delta power, a measurable index of sleep need, in the light phase (**Figures 3A–D**; SPS-D1 vs SD4-D1).

During REMS, SPS, relative to SD4, causes a significant increase in the absolute delta (↑8.8%), alpha (↑20.6%), and beta (↑10.6%) EEG power in the light phase, as well as a 10% reduction in theta EEG power in the dark phase (**Figures 3E– H**; SPS-D1 vs SD4-D1). During wakefulness, SPS, relative to SD4, caused a significant decrease in absolute delta (↓16.3%) and beta (↓7.9%) power in the light phase (**Figures 3I–L**; SPS-D1 vs SD4-D1). Additionally, SPS mice exhibited a broad reduction in all frequency bands of EEG signals (↓8.1% delta; ↓7.5% theta; ↓12.8% alpha; ↓12.6% beta) in the dark phase (**Figures 3I–L**; SPS-D1 vs SD4-D1). These observations indicate that SPS causes specific short-term sleep/wake EEG abnormalities.

# Traumatic Stress Induces Long-Term Sleep/Wake EEG Abnormalities

To examine the long-term effect of SPS on sleep–wake architecture, we compared the EEG/EMG data of the same mice on the seventh day (D7) after SD4 and SPS treatment (**Figure 4A**; SD4-D7 vs SPS-D7). Consistent with the previous study of SPS rats (Nedelcovych et al., 2015), there was no significant difference in the total duration, episode duration, or episode number of NREMS, REMS and wakefulness on D7 after SPS (**Supplementary Figures 4A–D**). By contrast, SPS, relative to SD4, caused a broad reduction in sleep/wake EEG power densities in the light phase, including the alpha (↓5.3%) and beta (↓6.4%) power during NREMS; the theta (↓4.9%) and alpha (↓4.8%) power during REMS; the alpha (↓3.5%) and beta (↓3.1%) power during wakefulness (**Figures 4B–D**; SPS-D7 vs

SD4-D7). In the dark phase, SPS caused a specific decrease in absolute theta (↓4.1%), alpha (↓6.7%), and beta (↓8.0%) power during NREMS; theta (↓7.8%) power during REMS; and alpha (↓5.7%) power during wakefulness (**Figures 4B–D**; SPS-D7 vs SD4-D7). Taken together, these observations indicate that unlike sleep deprivation, traumatic stress by SPS can lead to long-term sleep/wake EEG abnormalities.

### Absolute EEG Power Analysis Is Superior to Relative EEG Power Analysis

Our previous studies suggest that relative EEG power analysis is likely to miss critical changes of the EEG signals, such as a global reduction in EEG power densities, which can be detected by absolute EEG power analysis (Funato et al., 2016; Wang et al., 2018). Accordingly, we obtained very different outcomes by relative EEG power analysis (**Figure 4E**). During NREMS, SPS, relative to SD4, causes variable changes in the EEG power spectrum in the light phase (delta, ↓3.4%; theta, ↑3.8%; alpha,↑4.5%; beta, ↑5.5%) and in the dark phase (delta, ns; theta, ↓1.8%; alpha, ↑2.4%; beta, ns). Remarkably, only the modest reduction in NREMS delta (↓3.4%) power is verified by absolute EEG power analysis (**Figure 4E**; SPS-D1 vs SD4-D1). During REMS, relative EEG power analysis also reveals variable changes of the EEG power spectrum in the light phase (delta, ns; theta, ↓5.8%; alpha, ↑11.1%; beta, ns) and in the dark phase (delta, ns; theta, ↓4.2%; alpha, ns; beta, ns). Among these changes, only the 11.1% alpha power increase in the light phase and 4.2% theta power reduction in the dark phase are consistent with absolute EEG power analysis (**Figure 4E**; SPS-D1 vs SD4-D1).

For the long-term sleep/wake EEG abnormalities, relative EEG power analysis reveals that SPS, relative to SD4, causes a modest reduction in the alpha (↓2.5%) and beta (↓4.0%) power of NREMS in the light phase, beta (↓4.1%) power of NREMS in the dark phase, and alpha (↓3.4%) power of wake in the dark phase. Although these changes are largely consistent with those of absolute EEG power analysis, relative EEG power analysis failed to detect many critical changes of the EEG power spectrum in the dark phase (**Figure 4E**; SPS-D7 vs SD4-D7). Based on these observations, we conclude that absolute EEG power analysis is superior to relative EEG power analysis, which should be adopted especially in the longitude experimental setting.

# Persistent Activation of mPFC Neurons During and After SPS Treatment

Accumulating studies suggest that PTSD may be mediated by structural and functional alterations in multiple brain regions,


FIGURE 4 | Traumatic stress induces long-term alterations of sleep/wake EEG power spectrum. (A) A schematic of sleep–wake analysis at D7 after SD4/SPS treatment. (B–D) Analysis of mean absolute EEG power density in NREMS (B), REMS (C), and wake (D) states of the same test mice on day 7 after SD4/SPS treatment (SD4-D7 vs SPS-D7). (E) A table comparing the specific change ratios of the delta, theta, alpha, and beta power bands of EEG signals detected by absolute and relative EEG power analysis on D1 [("SPS-D1" – "SD4-D1")/"SD4-D1"] and on D7 [("SPS-D7" – "SD4-D7")/"SD4-D7"] after SD4/SPS treatment. Mean ± s.d., paired t-test, two-tailed (B–D). \*P < 0.05; †P < 0.01; nsP > 0.05.

including the prefrontal cortex, locus coeruleus, amygdala, hippocampus, and the hypothalamic–pituitary–adrenal (HPA) axis (Lindauer et al., 2004; Wignall et al., 2004; Lindauer et al., 2006; Chen et al., 2018; Deslauriers et al., 2018; Logue et al., 2018; Naegeli et al., 2018; van Rooij et al., 2018; Heyn et al., 2019). To explore the neurobiological correlates of traumatic stress-induced sleep abnormalities, we performed comparative analysis of the expression of immediate early gene c-Fos by immunostaining of mouse brain samples harvested at ZT4.5 and ZT7.5 after SD4/SPS treatment (ZT0–ZT4) (**Figure 5A**). Because c-Fos proteins exhibit a half-life of 45 min for fast decay and 1.5–2 h for slow decay (Shah and Tyagi, 2013), we wanted to identify specific brain regions showing persistent hyper-activity in response to SPS. At ZT4.5, SPS mice, relative to SD4 mice, showed significantly more c-Fos-expressing neurons in multiple subregions of the prefrontal cortex (**Figures 5B–D**), including the primary (M1) and secondary (M2) cortex of motor cortex (MC), the cingulate (Cg1), prelimbic (PrL), infralimbic (IL), and dorsal peduncular (DP) cortex within the mPFC, and the medial (MO), ventral (VO), and lateral (LO) part of the orbitofrontal cortex (OFC) (**Figure 5D**). By two-color fluoresence in situ hybridization, we showed that more than 95% of c-fos positive neurons in the

mPFC express the excitatory neuron marker vGlut1, but not the inhibitory neuron marker vGat (**Figures 5E,F**). Moreover, while SD4-induced c-Fos expression dissipated, SPS-induced c-Fos expression could still be observed in the mPFC, most notably in the PrL, IL and DP at ZT7.5 (**Figures 5C,D**). These results suggest SPS causes persistent hyper-activities of mPFC neurons during and immediately after SPS treatment.

# Chemogenetic Inhibition of mPFC Reverses SPS-Induced Sleep/Wake EEG Disturbances

We hypothesized that the persistent hyper-activities of mPFC could contribute to the SPS-induced short- and long-term alterations in the sleep–wake architecture and EEG power spectrum. The prelimbic (PrL) region of mPFC is a major subregion to control neuroendocrine outputs of the paraventricular hypothalamic nucleus (PVH) to restore homeostasis of the HPA axis-the central stress response system (Radley et al., 2006; Herman et al., 2012). Therefore, we used the inhibitory Designer Receptors Exclusively Activated by Designer Drugs (DREADD) system to investigate whether hyper-activation of PrL neurons play an important role in traumatic stress-induced sleep–wake disturbances. Specifically, we bilaterally injected AAV expressing mCherry (AAV2/9- CMV-mCherry) or hM4Di (AAV2/9-hSyn-hM4Di-mCherry) into the PrL of mPFC in C57BL/6N mice (**Figures 6A,B**). All AAV-injected mice were sequentially subjected to SD4 and SPS treatments as described above (**Figure 1**), except for intraperitoneal injection of vehicle during SD4 or CNO during SPS at ZT0 and ZT3.5, and followed by continuous EEG/EMG recording for seven days (**Figure 6A**).

We found that chemogenetic inhibition of PrL during SPS could not rescue the SPS-induced acute changes in sleep/wake duration on day 1 (**Supplementary Figure 5**; "SPS-D1" − "SD4-D1": mCherry vs hM4Di). However, inhibition of PrL activity could specifically reverse the SPS-induced acute suppression of NREMS delta power [**Figures 6C–E** and **Supplementary Figure 6A**; ("SPS-D1" − "SD4-D1")/"SD4-D1": mCherry vs hM4Di], particularly in the first hour (ZT4) after SPS (**Figures 6C,D**). By contrast, there were no statistically significant differences in other EEG power densities during NREMS, REMS or wake states between mCherry and hM4Di mice (**Supplementary Figure 6**). These results are consistent with the idea that hyper-activities of PrL neurons could result in specific suppression of the NREMS delta power, the best known measurable index of sleep need, immediately after traumatic stress.

To test whether the hyper-activities of PrL neurons during SPS might also result in the traumatic stress-induced long-term sleep/wake EEG disturbances, we analyzed EEG/EMG data of the mCherry and hM4Di mice on the seventh day after SD4/SPS treatment (**Figure 6A**). Remarkably, we found that chemogenetic inhibition of PrL neurons could abrogate the majority of SPS-induced long-term sleep/wake EEG abnormalities on day 7 [**Figures 6F**, **7A** and **Supplementary Figure 7**; ("SPS-D7" − "SD4-D7")/"SD4-D7": mCherry vs hM4Di]. Taken together, these results suggest that SPS-induced hyper-activation of mPFC neurons, particularly in the PrL region, may play a critical role in the development of both short- and long-term sleep–wake EEG disturbances (**Figure 7B**).

# DISCUSSION

The sleep–wake disturbances may be one of the most debilitating symptoms associated with PTSD (Pawlyk et al., 2008). In this study, we adopted the well-established SPS paradigm to investigate the effects of traumatic stress on the sleep–wake architecture in the isogenic mouse model. In accordance with what Liberzon and colleagues had originally observed in SPS rats (Yamamoto et al., 2009), we showed that SPS mice also exhibited higher immobility time than control mice in the FST, but similar immobility time in the TST on the seventh day after SD4/SPS treatment (**Supplementary Figure 4E**). This result, together with our findings that SPS mice exhibited robust short and long-term sleep disturbances – a core symptom of PTSD patients – further validated the cross-species utility of mouse SPS-PTSD model. Because of the isogenic background and many genetics tools available, the mouse SPS-PTSD model offers unique advantages than the rat SPS-PTSD model in future mechanistic studies of traumatic stress-induced sleep disturbances.

The acute effects of SPS on sleep–wake architecture have been reported in two previous studies in rats (Nedelcovych et al., 2015; Vanderheyden et al., 2015). Our results are mostly consistent with earlier findings of Nedelcovych et al. (2015), but not those of Vanderheyden et al. (2015). A common finding of all three studies is the significant increase in REMS in the dark phase on the day after SPS (**Figure 2**). The importance of REMS rebound after acute stress is highlighted by the sleep assessment of humans who experience a traumatic event: those who exhibit long episodes of REMS do not develop PTSD, whereas those who have very short episodes of REMS are likely to develop PTSD (Mellman et al., 2002). Taken together, these findings suggest that REMS rebound during the first dark phase, especially long REMS episodes, may represent an essential adaptive strategy for animals or humans to cope with traumatic stress and avoid the development of PTSD (Stickgold, 2007).

We also observed a specific increase in absolute alpha and beta power of EEG signals during REMS and a broad reduction in absolute EEG power densities during NREMS and wake states after SPS (**Figure 3**). These significant changes in sleep/wake EEG power spectra may be attributed to traumatic stress-induced dys-regulation of multiple neuronal networks mediated by distinct neuromodulators (Vakalopoulos, 2014). It has been shown that traumatic stress causes serotonin release and regional utilization changes in multiple brain regions (Germain et al., 2008; Pawlyk et al., 2008; Nedelcovych et al., 2015). Several studies have also reported that acute stress increases acetylcholine release in the hippocampus and frontal cortex (Mark et al., 1996) and impairs signaling in the prefrontal cortex (Picciotto et al., 2012). These brain region-specific changes of neuromodulator signaling may lead to acute changes

(Cg1), prelimbic (PrL), infralimbic (IL), and dorsal peduncular (DP) cortex within mPFC; medial (MO), ventral (VO), and lateral (LO) part of the orbitofrontal cortex (OFC).

in sleep/wake duration and/or short- and long-term statedependent EEG abnormalities.

\*P < 0.05; †P < 0.01; ‡P < 0.001; nsP > 0.05.

Consistent with our previous studies (Funato et al., 2016; Wang et al., 2018), we found that absolute EEG power analysis could consistently outperform relative EEG power analysis by revealing more critical changes in the EEG power spectrum. Moreover, relative EEG power analysis could sometimes distort the data and reach the wrong conclusion (**Figure 4E**). Thus, we recommend that both absolute and relative EEG power analysis should be performed to obtain comprehensive phenotypic analysis in future patho/physiological sleep studies, especially when using a longitude experimental design in the isogenic mouse models.

Both acute and chronic stress can cause structural and functional alterations of the mPFC, resulting in dys-regulation of the cognitive-emotional control and threat extinction (Holmes and Wellman, 2009; Herringa, 2017; Heyn et al., 2019). Our chemogenetic inhibition experiments strongly suggest that the hyper-activation of mPFC neurons during SPS may mediate specific suppression of NREMS delta power immediately after SPS treatment (**Figure 6**), and eventually lead to the long-term sleep/wake EEG abnormalities (**Figure 7**). To our best knowledge, our study represents the first attempt to establish such a causality link between dysfunction of a specific brain region and traumatic stress-induced sleep/wake EEG abnormalities.

Recent studies suggest that the mPFC contains a heterogeneous neural population, including the pyramidal neurons and interneurons that may exert opposite regulation on EEG activities. Whereas pyramidal neuronal activity results in cortical activation and desynchronization, inhibitory interneurons that express somatostatin (SOM) are involved in the generation and propagation of slow waves characteristic of NREM sleep (Funk et al., 2017). Although the detailed mechanism by which the mPFC responds to SPS is unclear, we found that more than 95% of c-fos-expressing neurons in the mPFC are excitatory neurons (**Figures 5E,F**), suggesting that hyper-activities of pyramidal neurons, rather than interneurons such as SOM+ or parvalbumin+ interneurons, in the mPFC

EEG/EMG recording after sleep deprivation (SD4, ZT0–ZT4), and subsequently subjected to seven day EEG/EMG recording after SPS (ZT0–ZT4). Intraperitoneal injection of vehicle (0.9% saline) during SD4 or CNO (3 mg/kg) during SPS was administered at ZT0 and ZT3.5. (B) Representative image showing correct AAV injection sites marked by mCherry positive cells. (C) Hourly analysis of mean absolute delta power density of NREMS in mCherry (n = 9) and hM4Di (n = 9) mice on the day after SD4/SPS treatment (SD4-D1 vs SPS-D1). (D,E) Comparison of the change ratio [("SPS-D1" – "SD4-D1")/"SD4-D1"]% in the mean absolute NREMS EEG power density of mCherry and hM4Di mice at ZT4 (D), and in the light or dark phase (E). (F) Comparison of the change ratio [("SPS-D7" – "SD4-D7")/"SD4-D7"]% in the mean absolute NREMS EEG power density of mCherry and hM4Di mice in the light or dark phase on day 7 after SD4/SPS treatment. Mean ± s.e.m., two-way ANOVA, Sidak's test (C); Mean ± s.e.m., unpaired t-test, two-tailed (D–F). \*P < 0.05; †P < 0.01; ‡P < 0.001; nsP > 0.05.

are probably involved in the acute suppression of NREMS delta power. However, future studies are needed to investigate the precise roles of different types of mPFC neurons in the SPS-induced sleep–wake EEG disturbances as the chemogenetic inhibition approach in our study result in the inhibition of all neuronal populations.

inhibition of mPFC. (B) A model showing that hyper-activation of mPFC contributes critically to the SPS-induced sleep–wake EEG disturbances.

Both reduced and increased delta power activity during NREMS have been reported in PTSD patients (Woodward et al., 2000; Germain et al., 2006; Insana et al., 2012; de Boer et al., 2020; Wang et al., 2020). Thus, the SPS mouse model may recapitulate the symptoms of the subset of PTSD patients showing reduced NREMS delta power (Woodward et al., 2000; de Boer et al., 2020; Wang et al., 2020). In our study, we found that chemogenetic inhibition of the mFPC activity could specifically reverse the SPSinduced acute suppression of delta power during NREMS and most of the long-term sleep/wake EEG abnormalities. Moreover, sleep deprivation immediately after trauma, which normally elevates NREMS delta power during recovery sleep, has been reported as an effective intervention for attenuating PTSDlike behavioral disruptions (Cohen et al., 2012, 2017). These observations underscore the importance of sleep-dependent processes of neural reactivation in the development of PTSD (Cohen et al., 2012, 2017). Our findings may suggest the mPFC as an attractive target for the development of effective therapeutics for traumatic stress-induced psychiatric disorders, such as PTSD.

### DATA AVAILABILITY STATEMENT

Source data and all other datasets generated and/or analyzed in the current study are available from the corresponding author on reasonable request.

# ETHICS STATEMENT

The animal study was reviewed and approved by the Institutional Animal Care and Use Committee of University of Tsukuba.

# AUTHOR CONTRIBUTIONS

ZW, JM, and TL designed the experiments with technical assistances from YT, LC, C-YL, GA, and ZC. KS made the AAV virus. JM, TL, and ZW collected and analyzed the data. JM and ZW made the figures. ZW, QL, and KS wrote the manuscript. All authors contributed to the article and approved the submitted version.

### FUNDING

This work was supported by the JSPS KAKENHI (16K16639 to ZW; 17K15592 to JM; 17K08133 to LC; and 19H03263 to QL) and the World Premier International Research Center Initiative (WPI) program from Japan's MEXT.

### ACKNOWLEDGMENTS

We thank all of Liu/Sakurai and Yanagisawa/Funato laboratory members for discussions and comments, especially Noriko Hotta-Hirashima, Miyo Kakizaki, and Chika Miyoshi for technical assistance.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins.2020. 00883/full#supplementary-material

FIGURE S1 | (A) Quantitative analysis of total NREMS, REMS, or wake time on the day before SD4/SPS treatment (SD4-BL vs SPS-BL). (B,C) Quantitative analysis of NREMS, REMS, or wake time on the day before and after 4 h sleep deprivation (SD4-D1 vs SD4-BL) (B), and on the day before and after SPS treatment (SPS-D1 vs SPS-BL) (C). n = 20, Mean ± s.d., paired t-test, two-tailed (A–C). <sup>∗</sup>P < 0.05; †P < 0.01; ‡P < 0.001; nsP > 0.05.

FIGURE S2 | (A–L) Analysis of mean absolute EEG power density in every hour (left, hourly) or in the light/dark phase (right) in NREMS (A–D), REMS (E–H) and wake (I–L) states of test mice (n = 20) on the day before SD4/SPS treatment (SD4-BL vs SPS-BL). Mean ± s.e.m., two-way ANOVA, Sidak's test (for hourly analysis); Mean ± s.d., paired t-test, two-tailed (for mean analysis). <sup>∗</sup>P < 0.05; †P < 0.01; nsP > 0.05.

FIGURE S3 | (A) Analysis of mean absolute NREMS ZT4 EEG power density on the day before and after SD4/SPS treatment (SD4-D1 vs SD4-BL; SPS-D1 vs SPS-BL), (n = 20). (B–D) Analysis of mean absolute EEG power density in the light/dark phase during NREMS (B), REMS (C) and wake (D) states of test mice (n = 20) on the day before and after SD4 treatment (SD4-D1 vs SD4-BL). Mean ± s.e.m., unpaired t-test, two-tailed (A); Mean ± s.d., paired t-test, two-tailed (B–D). <sup>∗</sup>P < 0.05; †P < 0.01; ‡P < 0.001; nsP > 0.05.

FIGURE S4 | (A, B) Hourly (A) and quantitative (B) analysis of NREMS, REMS and wake duration of test mice (n = 19) on day 7 after SD4 or SPS treatment (SD4-D7 vs SPS-D7). (C,D) Hourly analysis of episode duration (C) or episode number (D) of NREMS, REMS and wake states of test mice (n = 19) on D7 after SPS treatment (SD4-D7 vs SPS-D7). (E) Comparison of immobility time in the tail suspension test (TST) and forced swimming test (FST) on day 7 after SPS/SD4 treatment. SD4, n = 17; SPS, n = 12. Mean ± s.e.m., two-way RM ANOVA, Sidak's test (A,C,D); Mean ± s.d., paired t-test, two-tailed (B); Mean ± s.e.m., unpaired t-test, two-tailed (E). <sup>∗</sup>P < 0.05; †P < 0.01; ‡P < 0.001; nsP > 0.05.

FIGURE S5 | (A,B) Hourly (A) and quantitative (B) analysis of NREMS, REMS and wake duration of mCherry (n = 9) mice on the day after SD4/SPS treatment (SD4-D1 vs SPS-1). (C,D) Hourly (C) and quantitative (D) analysis of NREMS, REMS and wake duration of hM4Di (n = 9) mice on the day after SD4/SPS treatment (SD4-D1 vs SPS-D1). (E) Comparison of the difference ("SPS-D1" − "SD4-D1") in NREMS, REMS or wake time of mCherry (n = 9) and hM4Di mice (n = 9) on the day after SD4/SPS treatment. Mean ± s.e.m., two-way RM ANOVA, Sidak's test (A,C); Mean ± s.d., paired t-test, two-tailed (B,D); Mean ± s.e.m., unpaired t-test, two-tailed (E). <sup>∗</sup>P < 0.05; †P < 0.01; ‡P < 0.001; nsP > 0.05.

FIGURE S6 | (A–E) Analysis of mean absolute NREMS (A), REMS (B) or wake (D) EEG power density of mCherry (n = 9) and hM4Di (n = 9) mice on the day after SD4/SPS treatment (SD4-D1 vs SPS-D1). Comparison of the change ratio [("SPS-D1" − "SD4-D1")/"SD4-D1"]% in the mean absolute REMS (C) and wake (E) EEG power density of mCherry and hM4Di mice (n = 9) in the light or dark phase. Mean ± s.d., paired t-test, two-tailed (A, B, D); Mean ± s.e.m., unpaired t-test, two-tailed (C, E). <sup>∗</sup>P < 0.05; †P < 0.01; ‡P < 0.001; nsP > 0.05.

FIGURE S7 | (A–E) Analysis of mean absolute EEG power density during NREMS (A), REMS (B) and wake (D) states of mCherry (n = 9) and hM4Di (n = 9) mice on day 7 after SD4/SPS treatment (SD4-D7 vs SPS-D7). Comparison of the change ratio [("SPS-D7" − "SD4-D7")/"SD4-D7"]% in the mean absolute EEG power density during REMS (C) and wake (E) states of mCherry (n = 9) and hM4Di (n = 9) mice in the light or dark phase on day 7 after SD4/SPS treatment. Mean ± s.d., paired t-test, two-tailed (A,B,D); Mean ± s.e.m., unpaired t-test, two-tailed (C,E). <sup>∗</sup>P < 0.05; †P < 0.01; ‡P < 0.001; nsP > 0.05.

TABLE S1 | Statistical analysis. The complete sample size, statistical test method, and precise value results for each comparison were reported.

# REFERENCES

fnins-14-00883 August 17, 2020 Time: 15:15 # 14


of sexual abuse: the role of perceived social support and abuse characteristics. J. Interpers. Violence 27, 1827–1843. doi: 10.1177/0886260511430385


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Lou, Ma, Wang, Terakoshi, Lee, Asher, Cao, Chen, Sakurai and Liu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

fnins-14-00883 August 17, 2020 Time: 15:15 # 15

digital media

of impactful research

article's readership