# NEUROIMAGING FINDINGS IN SLEEP DISORDERS AND CIRCADIAN DISRUPTION

EDITED BY : Xi-Jian Dai, Hengyi Rao and Kai Spiegelhalder PUBLISHED IN : Frontiers in Neurology and Frontiers in Psychiatry

#### Frontiers Copyright Statement

© Copyright 2007-2019 Frontiers Media SA. All rights reserved. All content included on this site, such as text, graphics, logos, button icons, images, video/audio clips, downloads, data compilations and software, is the property of or is licensed to Frontiers Media SA ("Frontiers") or its licensees and/or subcontractors. The copyright in the text of individual articles is the property of their respective authors, subject to a license granted to Frontiers.

The compilation of articles constituting this e-book, wherever published, as well as the compilation of all other content on this site, is the exclusive property of Frontiers. For the conditions for downloading and copying of e-books from Frontiers' website, please see the Terms for Website Use. If purchasing Frontiers e-books from other websites or sources, the conditions of the website concerned apply.

Images and graphics not forming part of user-contributed materials may not be downloaded or copied without permission.

Individual articles may be downloaded and reproduced in accordance with the principles of the CC-BY licence subject to any copyright or other notices. They may not be re-sold as an e-book.

As author or other contributor you grant a CC-BY licence to others to reproduce your articles, including any graphics and third-party materials supplied by you, in accordance with the Conditions for Website Use and subject to any copyright notices which you include in connection with your articles and materials.

All copyright, and all rights therein, are protected by national and international copyright laws.

The above represents a summary only. For the full conditions see the Conditions for Authors and the Conditions for Website Use. ISSN 1664-8714 ISBN 978-2-88945-859-2 DOI 10.3389/978-2-88945-859-2

### 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

# NEUROIMAGING FINDINGS IN SLEEP DISORDERS AND CIRCADIAN DISRUPTION

Topic Editors: Xi-Jian Dai, Nanjing University, China Hengyi Rao, University of Pennsylvania,United States Kai Spiegelhalder, University of Freiburg, Germany

Each of us spends almost a third of our life asleep. Sleep is important for normal life processes including blood, metabolism, immune, endocrine, and brain activity. Neuroimaging studies of sleep disorders have not received as much attention as other psychiatric diseases. Here, we introduce some new findings in neuroimaging field of sleep disorders from five chapters in different aspects.

Citation: Dai, X.-J., Rao, H., Spiegelhalder, K., eds. (2019). Neuroimaging Findings in Sleep Disorders and Circadian Disruption. Lausanne: Frontiers Media. doi: 10.3389/978-2-88945-859-2

# Table of Contents

### CHAPTER 1

### INTRODUCTION

*05 Editorial: Neuroimaging Findings in Sleep Disorders and Circadian Disruption*

Xi-Jian Dai, Hengyi Rao and Kai Spiegelhalder

## CHAPTER 2

### CHRONIC INSOMNIA

*08 Aberrant Effective Connectivity of the Right Anterior Insula in Primary Insomnia*

Chao Li, Mengshi Dong, Yi Yin, Kelei Hua, Shishun Fu and Guihua Jiang

*17 Plasticity and Susceptibility of Brain Morphometry Alterations to Insufficient Sleep*

Xi-Jian Dai, Jian Jiang, Zhiqiang Zhang, Xiao Nie, Bi-Xia Liu, Li Pei, Honghan Gong, Jianping Hu, Guangming Lu and Yang Zhan

### CHAPTER 3

### SLEEP DEPRIVATION


### CHAPTER 4

### OBSTRUCTIVE SLEEP APNEA

*64 Topological Reorganization of the Default Mode Network in Severe Male Obstructive Sleep Apnea*

Liting Chen, Xiaole Fan, Haijun Li, Chenglong Ye, Honghui Yu, Honghan Gong, Xianjun Zeng, Dechang Peng and Liping Yan

### CHAPTER 5

## OTHER SLEEP-RELATED DISEASES SECTION 1

### NARCOLEPSY WITH CATAPLEXY

*75 Recursive Partitioning Analysis of Fractional Low-Frequency Fluctuations in Narcolepsy With Cataplexy*

Xiao Fulong, Lu Chao, Zhao Dianjiang, Zou Qihong, Zhang Wei, Zhang Jun and Han Fang

### SECTION 2 RESTLESS LEGS SYNDROME

### *83 Lack of Association Between Shape and Volume of Subcortical Brain Structures and Restless Legs Syndrome*

Marco Hermesdorf, Benedikt Sundermann, Rajesh Rawal, András Szentkirályi, Udo Dannlowski and Klaus Berger

## SECTION 3

CIRCADIAN DISRUPTION

*90 Imaging Individual Differences in the Response of the Human Suprachiasmatic Area to Light*

Elise M. McGlashan, Govinda R. Poudel, Parisa Vidafar, Sean P. A. Drummond and Sean W. Cain

# Editorial: Neuroimaging Findings in Sleep Disorders and Circadian Disruption

#### Xi-Jian Dai <sup>1</sup> \*, Hengyi Rao<sup>2</sup> and Kai Spiegelhalder <sup>3</sup>

*<sup>1</sup> Department of Medical Imaging, Jinling Hospital, Medical School of Nanjing University, Nanjing, China, <sup>2</sup> Division of Sleep, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, United States, <sup>3</sup> Department of Psychiatry and Psychotherapy, Faculty of Medicine, Medical Center – University of Freiburg, Freiburg, Germany*

Keywords: insomnia, obstructive sleep apnea, sleep deprivation, functional magnetic resonance imaging, sleep disorders, circadian rhythm

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

#### **Neuroimaging Findings in Sleep Disorders and Circadian Disruption**

#### Edited by:

*Jan Kassubek, University of Ulm, Germany*

#### Reviewed by:

*Menno Michiel Schoonheim, VU University Medical Center, Netherlands*

> \*Correspondence: *Xi-Jian Dai daixjdoctor@126.com*

#### Specialty section:

*This article was submitted to Applied Neuroimaging, a section of the journal Frontiers in Neurology*

Received: *02 February 2019* Accepted: *25 February 2019* Published: *21 March 2019*

#### Citation:

*Dai X-J, Rao H and Spiegelhalder K (2019) Editorial: Neuroimaging Findings in Sleep Disorders and Circadian Disruption. Front. Neurol. 10:249. doi: 10.3389/fneur.2019.00249* Each of us spends almost a third of our life asleep. Thus, obviously, sleep is a necessary physical need in human life. After sleep, the tired nerve cells and the biological characteristics of longdistance signal transmission recover to normal physiological function. In general, precise control of the sleep process is the basis of normal life processes including blood, metabolism, immune, endocrine, and brain activity, and is key to plasticity formation, information processing, and function implementation [(1–5); Dai et al.].

Sleep has played a minor role as object of research for a long time. Yet, recently there is a growing public interest in sleep. Sleep disorders are a major public health problem and widespread in today's society. In modern society, more and more people undergo an increased curtailment of daily sleep because of work overtime, exam preparation, shift working and long-term working or driving, resulting in an increased incidence of sleep disorders. The disturbed and/or interrupted sleep may be associated with a number of clinical conditions and has a detrimental effect on attention, working memory, executive functioning, emotion, or even metabolism. Nowadays, important challenges are posed to sleep disorders for which approved treatments are of limited efficacy.

Although there is surprising upsurge in neuroimaging findings in addressing the brain structural and functional changes associated with sleep disorders and circadian disruption, it is still difficult to glean a consistent story about its neuropathology of brain alterations. Therefore, a more comprehensive understanding of brain structural and functional changes associated with sleep disorders and circadian disruption are needed. The aim of this Research Topic is to contribute to a better understanding of the link between brain and sleep disorders, and offer an up-to-date view on how sleep affects our brain.

### PRIMARY INSOMNIA

This specific issue includes two studies focusing on insomnia. In one study by Li et al. the authors found decreased effective connectivity from right ventral and dorsal anterior insula to the precuneus, postcentral gyrus, and cerebellum posterior lobe, which negatively correlated with Pittsburgh Sleep Quality Index and Insomnia Severity Index scores.

**5**

In another study by Dai et al. the authors found that acute sleep deprivation (SD) and chronic insomnia showed widespread changes in gray matter volumes (GMVs) with shared but also distinct neurobiological representation in brain morphology. Acute SD may be associated with inhibition in sensory-informational processing with decreased GMVs in the somatosensory areas to compensate for the effects of sleep loss on advanced cognitive function, while primary insomnia may be associated with increased GMVs in several brain areas, which may be key a core predisposing or perpetuating factor of ultimately hampering the ability to initiate or maintain sleep.

### SD

This specific issue also includes four studies focusing on SD and one study on narcolepsy. In one study by Satterfield et al. the authors reported that increased baseline responsiveness within reward regions are more vulnerable to SD-induced overeating. Functional activation within the ventral striatum during the multi-source interference task (MSIT) and n-back task positively correlated with total caloric and carbohydrate intake during the final 6 h (06:00–12:00) of acute SD. Activation within the middle and superior temporal gyri during the MSIT also correlated with total carbohydrates consumed.

In the second study by Dai et al. these authors found prolonged acute SD hours (20, 24, 32, 36 h SD) exhibit accumulative brain atrophic effects and recovering plasticity (after one night sleep recovery) on brain morphology, in line with the behavioral changes on attentional and working memory tasks, which may provide the neurobiological basis for attention and memory impairments following sleep loss.

The last two studies focus on finding potential indicators. Chen et al. and Kong et al. found that the amplitude of lowfrequency fluctuation and short-range and long-range functional connectivity density may be potential biomarkers to describe the altered regional brain cortical activities and intrinsic brain functional organization disturbed by acute SD with high discriminating performances.

### OBSTRUCTIVE SLEEP APNEA

One study by Chen et al. examined topological changes in obstructive sleep apnea and found decreased functional connectivity within the default mode network, which may contribute to the observed topological reorganization of clustering coefficient, path length, global efficiency, and Montreal cognitive assessment score. These findings may provide evidence of cognitive deficits in obstructive sleep apnea.

### REFERENCES

1. Ohayon MM, Smolensky MH, Roth T. Consequences of shiftworking on sleep duration, sleepiness, and sleep attacks. Chronobiol Int. (2010) 27:575–89. doi: 10.3109/07420521003749956

### NARCOLEPSY WITH CATAPLEXY

One study by Fulong et al. found that both adult and juvenile narcolepsy had lower fractional low-frequency fluctuations (fALFF) values in bilateral medial superior frontal gyrus, bilateral inferior parietal lobule and supra-marginal gyrus, and higher fALFF values in bilateral sensorimotor cortex and middle temporal gyrus. The right medial superior frontal gyrus discriminated between narcolepsy and healthy controls with high degree of sensitivity (100%) and specificity (88.9%), which may suggest that the fALFF may be a helpful imaging biomarker.

### RESTLESS LEGS SYNDROME

One study by Hermesdorf et al. evaluated the relationship between the genetic risks and subcortical volumes for restless legs syndrome, but neither of them gave rise to the GMV changes in the hippocampal and subcortical shapes.

### CIRCADIAN DISRUPTION

The ninth study by McGlashan et al. investigated whether BOLDfMRI activation of human suprachiasmatic area in response to light in a 30 s block-paradigm of lights on (100 lux) and lights off (<1 lux) is related to a functional outcome. They found a positive correlation between this activation and melatonin suppression, which may help to better understand the clinical vulnerability influenced by circadian disruption.

### CONCLUSIONS

Together, this issue features articles that address the relationships between sleep-related disorders and the brain structure and function using neuroimaging methods. We hope this special issue will contribute to a better understanding of the link between brain and sleep disorders and offer an up-to-date view on how sleep affects our brain. We believe that this special issue will stimulate discussions in a wider public involving not only those working in the field, since both conditions cause an extreme impairment of quality of life, in particular in those patients suffering from both conditions.

### 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 National Natural Science Foundation of China (grant No. 81701678).

2. Dai XJ, Liu CL, Zhou RL, Gong HH, Wu B, Gao L, et al. Longterm total sleep deprivation decreases the default spontaneous activity and connectivity pattern in healthy male subjects: a resting-state fMRI study. Neuropsychiatr Dis Treat. (2015) 11:761–72, doi: 10.2147/NDT. S78335


**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 Dai, Rao and Spiegelhalder. 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.

# Aberrant Effective Connectivity of the Right Anterior Insula in Primary Insomnia

*Chao Li, Mengshi Dong, Yi Yin, Kelei Hua, Shishun Fu and Guihua Jiang\**

*Department of Medical Imaging, Guangdong Second Provincial General Hospital, Guangzhou, China*

Objective: Daytime cognitive impairment is an essential symptom of primary insomnia (PI). However, the underlying neural substrate remains largely unknown. Many studies have shown that the right anterior insula (rAI) as a key node of salience network (SN) plays a critical role in switching between the executive control network (ECN) and the default mode network (DMN) for better performance of cognitively demanding tasks. Aberrant effective connectivity (directional functional connectivity) of rAI with ECN or DMN may be one reason for daytime cognitive impairment in PI patients. Up to now, no effective connectivity study has been conducted on patients with PI during resting state. Our aim is to investigate the effective connectivity between the rAI and the other voxels in the whole brain in PI.

#### *Edited by:*

*Xi-jian Dai, Medical School of Nanjing University, China*

#### *Reviewed by:*

*Zhen Yuan, University of Macau, Macau Axel Steiger, Max-Planck-Institut für Psychiatrie, Germany*

*\*Correspondence:*

*Guihua Jiang jiangguihua1970@163.com*

#### *Specialty section:*

*This article was submitted to Sleep and Chronobiology, a section of the journal Frontiers in Neurology*

*Received: 28 January 2018 Accepted: 23 April 2018 Published: 08 May 2018*

#### *Citation:*

*Li C, Dong M, Yin Y, Hua K, Fu S and Jiang G (2018) Aberrant Effective Connectivity of the Right Anterior Insula in Primary Insomnia. Front. Neurol. 9:317. doi: 10.3389/fneur.2018.00317*

Materials and methods: Fifty drug-naive patients with PI and forty age- and sexmatched healthy controls were scanned using resting-state functional MRI. Seed-based Granger causality analysis was used to examine effective connectivity between the rAI, including ventral and dorsal part, and the whole brain. The effective connectivity was compared between the two groups and was correlated with clinical characteristics.

Results: Compared with controls, patients showed decreased effective connectivity from the rAI to the bilateral precuneus, the left postcentral gyrus (extending to bilateral precuneus) and the bilateral cerebellum posterior lobe, and decreased effective connectivity from the bilateral orbitofrontal cortex (OFC) to the rAI (single voxel *P* < 0.001, AlphaSim corrected with *P* < 0.01). In addition, effective connectivity from the ventral rAI to the left postcentral gyrus and from the left OFC to the ventral rAI were significantly negatively correlated with Insomnia Severity Index scores (*r* = −0.28/*P* = 0.046 and *r* = −0.29/*P* = 0.038, respectively).

Conclusion: The present study is the first to reveal aberrant effective connectivity between the SN hub (rAI) and the posterior DMN hub (precuneus) as well as decisionmaking region (OFC) and sensori-motor region in PI. These findings suggest an aberrant salience processing system of the rAI in PI patients.

Keywords: primary insomnia, functional magnetic resonance imaging, effective connectivity, insular cortex, executive function, cognitive impairment

## INTRODUCTION

Primary insomnia (PI) is one of the most common health problems. It is characterized by difficulties in falling asleep, maintaining sleep, or early awakening for at least 1 month (1). The worldwide prevalence of insomnia symptoms is approximately 30–35% and approximately 10% of people are diagnosed with PI (2, 3). Insomnia is associated with cognitive impairment, daytime fatigue, and

**8**

mood disruption (4, 5). Among a series of adverse consequences caused by insomnia, daytime cognitive impairment is an essential symptom with regard to working memory, episodic memory, and some aspects of executive functioning (4, 6). However, the underlying neural substrate is incompletely understood.

Neuroimaging techniques provided a new avenue to study the pathophysiological mechanisms underlying many psychiatric disorders (7–12). With regard to PI, PET, functional, and structural MR imaging have shown abnormal glucose metabolism (13), activation (14, 15), spontaneous activity (16–18), functional or structural connectivity (19–25), or atrophic structure (26–28) related to the cognitive system, especially the salience network (SN) (11, 21), executive control network (ECN), and default mode network (DMN) (22, 29).

It is worth noting that failed reducing activities of DMN is an important feature of PI during cognitively demanding tasks, such as during working memory task (15). During working memory task, with increasing task difficulty PI patients not only showed reduced activation in task-positive regions but also showed reduced deactivation in task-negative regions (DMN) (15). This study suggested that it was not simply the failure to recruit ECN that was associated with the reduced cognitively demanding task performance, but there was a conjoint failure to deactivate the DMN. However, very little is known about the neural mechanism behind this phenomenon.

The right anterior insula (rAI) as a key node of SN can modulate activity in the ECN and the DMN in healthy individuals for better performance of cognitively demanding tasks (30, 31). It is worth noting that the modulating function of anterior insula is right lateralized. So, we only choose the right side as the seed region but not left insula. Present theory holds that one fundamental mechanism underlying cognitive control is a transient signal from the right fronto-insular cortex, which engages the brain's attention, working memory, and higher-order control processes while disengaging other systems (such as DMN) that are not immediately task relevant (30, 32). Therefore, aberrant directed functional connectivity (FC) (effective connectivity) of rAI over ECN or DMN may be one reason of daytime cognitive impairment in PI patients. Previous studies have found altered FC in the right insula in PI patients (20–22, 33–35). However, there is no research to study the effective connectivity of the rAI.

In contrast to FC, which is zero time-lagged correlation between time series at spatially distinct regions of brain, the effective connectivity is the time-lagged correlation between time series. Effective connectivity from a region *X* to another region *Y* implies that the neuronal activity in region *X* precedes and predicts the neuronal activity that occurs in region *Y*. As mentioned above, some FC studies have already been performed to study the right insula (including rAI). However, there is still no research to study the directional FC (effective connectivity) in PI. Investigation of effective connectivity in PI patients may deepen our understanding of neurologic mechanism of PI.

We employed Granger causality analysis (36–38) in restingstate functional MR imaging to investigate the rAI-centered effective connectivity. Granger causal influence from a region *X* to another region *Y* implies that the neuronal activity in region *X* precedes and predicts the neuronal activity that occurs in region *Y* (38). Thus, the whole-brain Granger causal analysis is a useful approach to study the effective connectivity that may exist across networks. In contrast to undirected FC which does not support inferences about directed (causal) brain connections, effective connectivity refers to the influence that one neural system exerts over another and quantifies the directed coupling among brain regions. In addition to Granger causal analysis, there are also several ways to capture the directional brain dynamics, such as dynamic causal modeling and structural equation modeling. Granger causal analysis has been widely used to study effective connectivity in normal brains (30), schizophrenia (38), and major depressive disorders (37). To date, to the best of our knowledge, no studies have been published reporting effective connectivity in PI patients. The purpose of this study was to analyze the effective connectivity between the rAI and the whole brain in PI patients using first-order Granger causality analysis and its association with sleep and emotion scales of PI. We hypothesized that the effective connectivity between rAI and ECN or DMN was disrupted.

### MATERIALS AND METHODS

### Participants

This prospective study was approved by the ethics committee of the Guangdong Second Provincial General Hospital. All PI patients were recruited from the Department of Neurology at Guangdong Second Provincial General Hospital, Guangzhou, China from April 2010 to May 2016. Written informed consent was obtained from all patients. The inclusion criteria for PI patients were (a) all patients must meet Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) for diagnosis of PI; (b) patients had been complaining of difficulty falling asleep, maintaining sleep, or early awakening for at least 1 month; (c) patients had no other sleep disorders such as hypersomnia, parasomnia, sleep-related movement disorder, or other psychiatric disorders; (d) patients were younger than 60 years old; (e) free of any psychoactive medication at least 2 weeks prior to and during the study; and (f) patients were right-hand dominant as assessed with the Edinburgh Handedness Inventory. Exclusion criteria were as follows: (a) patients had an abnormal signal in any region of the brain verified by conventional T1-weighted or T2-fluid-attenuated inversion recovery MR imaging; (b) the insomnia disorder was caused by organic disease or severe mental disease such as secondary to depression or generalized anxiety; (c) other sleep disorder; (d) women who were pregnant, nursing, or menstruating; and (e) head motion more than or equal to 1.5 mm or 1.5° during MR imaging. Then three patients were discarded. Finally, 50 PI patients who met the requirements were included in the study.

A total of 40 age-, gender-, and education-matched healthy control (HC) subjects were recruited (17 men, 23 women; mean age, 39.38 ± 9.26 years) from the local community by using advertisements. Each HC subject gave written informed consent. HCs must met the following criterion: (a) Insomnia Severity Index (ISI) score was less than 7; (b) no history of swing shifts, shift work, or sleep complaints; (c) no medication or substance abuse such as caffeine, nicotine, or alcohol; (d) no brain lesions or prior substantial head trauma, which was verified by conventional T1-weighted or T2-fluid-attenuated inversion recovery MR imaging; (e) no history of psychiatric or neurological diseases; (f) head motion less than 1.5 mm or 1.5° during MR scan; and (g) right-hand dominant. Three controls were discarded due to head motion.

### Sleep and Emotion Scales

Several questionnaires were filled out by study participants. These questionnaires included the ISI (39), the Pittsburgh Sleep Quality Index (PSQI) (40), the Self-rating Anxiety Scale (SAS) (41), and the Self-rating Depression Scale (SDS) (42).

### MR Imaging

Resting-state functional MR imaging data were acquired using a 1.5 T MR scanner (Achieva Nova-Dual; Philips, Best, the Netherlands) in the Department of Medical Imaging, Guangdong Second Provincial General Hospital. To minimize head movements, a belt and foam pads were used. During the scanning, subjects were instructed to rest with their eyes closed and remain still but emphatically without falling asleep. The functional MR images were acquired in about 10 min using a gradient-echo planar imaging sequence as follows: interleaved scanning, repetition time = 2,500 ms, echo time = 50 ms, matrix = 64 × 64, field of view = 224 mm × 224 mm, flip angle = 90°, section thickness = 4 mm, gap = 0.8 mm, 27 axial slices, and 240 volumes.

### Data Preprocessing

The Data Processing Assistant for Resting-State Functional MR Imaging toolbox1 (version 2.3) was used to process the restingstate functional MR imaging data. Volumes at the first 10 time points were discarded so that magnetization reached a steady state and subjects had adapted to the MR scanning noise. The slice timing and realignment for head motion correction were conducted on the remaining images. Then, the realigned images were spatially normalized to the Montreal Neurological Institute template by applying the EPI template, and each voxel was resampled to 3 mm × 3 mm × 3 mm. We spatially smoothed the spatially normalized images with a 6-mm full-width halfmaximum isotropic Gaussian kernel. In order to reduce effects of low-frequency drift and high-frequency noise, we processed the data to remove linear trends and filtered temporally (band-pass, 0.01–0.08 Hz). Nine nuisance covariates, including cerebrospinal fluid signals, white matter signals, global brain signal, and six head motion parameters were regressed from the imaging data. The residuals of these regressions were used for the following analysis.

### Granger Causality Analysis

We calculated the effective connectivity of the time series of the dorsal and ventral rAI on every voxel in the whole brain (*X* to *Y*) and the effective connectivity of the time series of every voxel in the whole brain on the dorsal and ventral rAI (*Y* to *X*). The mean temporal-domain bold signals for the dorsal and ventral rAI are displayed in Figure S10 in Supplementary Material. Regions of interest (ROI) in the dorsal and ventral rAI were selected based on the brain atlas based on connectional architecture (43).2 Bivariate first-order coefficient-based voxel-wise Granger causality analysis was performed using REST-GCA (44). We followed Chen's (37) extended model as following:

$$\begin{aligned} Y\_l &= \sum\_{i=1}^p A\_i X\_{(l-i)} + \sum\_{i=1}^p B\_i Y\_{(l-i)} + C Z\_l + \varepsilon\_l \\ X\_l &= \sum\_{i=1}^p A\_i' Y\_{(l-i)} + \sum\_{i=1}^p B\_i' X\_{l-i} + C Z\_l + \varepsilon\_l' \end{aligned}$$

where *Yt* is the BOLD time series of one voxel in the brain at time *t*; *X* is the BOLD time series of seed region; *Zt* is a *q* × 1 vector containing exogenous variables (covariates or confounds) at time *t*; ε*t* is the error term; *p* and *q* are the number of lags and confounds, respectively; *Ai* is the signed path coefficient at time lag *i* (*i* = 1, …, *p*); *Bi* is the autoregression coefficient. In the present study, the number of lags *p* = 1 (1 TR = 2.5 s).

Explanation of the coefficient was the same as the previous study (38). The positive coefficient is referred as excitatory influence and *vice visa*.

### Statistical Analysis

Differences in age, education level, ISI, PSQI, SAS, and SDS scores between PI patients and HCs were compared by using two-sample *t* tests. Differences associated with gender between the two groups were assessed by using chi-squared tests.

First, the effective connectivity maps were analyzed using onesample *t*-test for the entire sample (both PI patients and HCs) with an uncorrected *P* < 0.001, cluster size = 50. Then, betweengroup differences in effective connectivity were compared by using two-sample *t* tests in a voxel-by-voxel fashion with age, sex, and education level imported as covariates. Multiple comparisons were corrected by an AlphaSim method implemented in the DPABI software [DPABI version 2.3, Data Processing & Analysis for (Resting-State) Brain Imaging] (45) and using significant corrected thresholds of *P* < 0.01 with combined with single voxel *P* < 0.001. The estimated FWHM (*x*–*y*–*z*) for the 4T maps (from ventral rAI, from dorsal rAI, to ventral rAI, and to dorsal rAI) were 6.9857–7.1202–7.9652, 5.8353–5.9840–6.7710, 9.1891–9.4015–9.4269, and 5.0546–5.1626–5.8403. The cluster size thresholds for the 4T maps were 22, 16, 48, and 26.

Besides, we used permutation threshold-free cluster enhancement (TFCE) correction method to perform statistical analysis (46, 47). The permutation TFCE correction method implements correction through a permutation testing approach which controls family-wise error rate by comparing voxel-wise statistics (TFCE) to the maximal statistics obtained from repeating the analysis with randomized data. The Matlab scripts for the permutation TFCE correction have been made available online: https://github.com/markallenthornton. ROI were defined as 6-mm-diameter spheres centered on voxels that exhibited the

<sup>1</sup>http://www.restfMRI.net/forum/DPARSF (Accessed: July, 2017).

<sup>2</sup>http://atlas.brainnetome.org/ (Accessed: July, 2017).

largest absolute *t* value in each of the significant clusters in the *t* map of between-group differences in effective connectivity. Then, effective connectivity was calculated for each subject by averaging the values of effective connectivity across all voxels within each of the ROI and correlated with the sleep and emotion scales using Pearson's correlation analysis.

### RESULTS

### Sleep and Emotion Scales

As **Table 1** shown, the PI patients and the controls showed no significant differences in age (*P*= 0.37), sex (*P*= 0.81), and education level (*P* = 0.28). PI patients had higher ISI, PSQI, SAS, and SDS scores than those of HCs (all *P* < 0.001).

### Effective Connectivity

One-sample *t*-test showed that the rAI exerted excitatory influence on the bilateral dorsolateral prefrontal cortex (DLPFC), the inferior parietal regions, the cingulate gyrus, and the left cerebellar crus. Inhibitory influence of the rAI was noted on the left precentral gyrus, the postcentral gyrus, and the bilateral occipital lobe. Furthermore, the bilateral DLPFC, the inferior parietal regions, and the cingulate gyrus, in turn, had inhibitory influence on the rAI, and in the same way, the bilateral occipital lobe had excitatory influence on the rAI. It is worth noting that the results of one-sample *t*-test were very similar to those of previous study (38). The results of the one-sample *t* tests are presented in Figures S4–S7 in Supplementary Material.

Compared with HCs, patients with PI showed negative effective connectivity (inhibitory influences) from the ventral rAI to the left precuneus, the left postcentral gyrus extending to the bilateral precuneus, and bilateral cerebellum posterior lobe including the bilateral cerebelum\_crus1 and left cerebelum\_6 (**Figure 1A**), and negative effective connectivity from the dorsal rAI to the bilateral precuneus and left postcentral gyrus extending to the left precuneus (**Figure 1B**). Also, patients with PI showed negative effective connectivity from bilateral orbitofrontal cortex (OFC) to ventral rAI (**Figure 1C**) (single voxel *P* < 0.001, corrected by AlphaSim correction with cluster *P* < 0.01). All above results of between-group differences in effective connectivity are shown in **Table 2**. Figures S1–S3 in


*Unless otherwise noted, data are mean* ± *SD.*

*PSQI, Pittsburgh Sleep Quality Index; ISI, Insomnia Severity Index; SAS, Self-rating* 

*Anxiety Scale; SDS, Self-rating Depression Scale; PI, primary insomnia. a The P value was obtained by using chi-square test.*

*bThe P value was obtained by using two-sample t tests.*

Supplementary Material showed the bar graphs demonstrating the mean effective connectivity values in the ROI defined as 6-mm-diameter spheres centered on voxels that exhibited the largest absolute *t* value in each of the significant clusters in the *t* map.

Results from permutation TFCE correction (5,000 times permutation, default parameters, FWE corrected, *P*< 0.05) were very similar to those derived from our parameter statistical method (two-sample *t* test with AlphaSim correction). Therefore, we only discussed these results. Figure S8 in Supplementary Material showed the *P* map of the permutation TFCE correction.

### Relationships Between Effective Connectivity and Sleep and Emotion Scales

As **Figure 2** shown, effective connectivity from the ventral rAI to the left postcentral gyrus extending to the bilateral precuneus and from the left OFC to the ventral rAI were significantly negatively correlated with ISI scores in PI group (*r* = −0.28/*P* = 0.046 and *r* = −0.29/*P* = 0.038, respectively).

Effective connectivity from the ventral rAI to the left precuneus was significantly negatively correlated with PSQI and ISI scores in HC group (*r* = −0.31/*P* = 0.047 and *r* = −0.32/*P* = 0.045, respectively). Figure S9 in Supplementary Material shows the results of the correlation analysis of the HC group.

### DISCUSSION

The present study investigated the effective connectivity between the rAI and the whole brain in PI patients. Our findings showed aberrant effective connectivity of rAI (a key node of SN) with the posterior DMN hub (precuneus) as well as regions involved in decision-making (OFC) and regions involved in sensori-motor function in PI. In addition, effective connectivity from the ventral rAI to the left postcentral gyrus extending to the bilateral precuneus and from the left OFC to the ventral rAI were significantly negatively correlated with ISI scores in PI group.

In contrast to FC which does not support inferences about directional brain connections, effective connectivity refers to the influence that one neural system exerts over another and quantifies the directional connectivity among brain regions (38). Consequently, effective connectivity may provide new insight into the neurological mechanism of insomnia.

The important findings in the current study were aberrant effective connectivity from the ventral rAI to the left precuneus and from the dorsal rAI to the bilateral precuneus at resting state. The rAI was a hub node of SN which is involved in detecting and orienting to both external and internal salient stimuli and events (31, 32). The precuneus was a hub node of DMN which is involved in self-referential/internally oriented processes (48). Previous study using chronometry and Granger causality analysis confirmed that rAI plays a critical and causal role in switching between the ECN and the DMN during visual attention tasks, oddball tasks, and even resting state (30). Furthermore, the present theory holds that one fundamental mechanism underlying cognitive control is a transient signal

from the right fronto-insular cortex, which engages the brain's attentional, working memory, and higher-order control processes while disengaging other systems (such as DMN) that are not immediately task relevant (30, 32). Interestingly, a recent study found that PI patients showed both reduced activation in task-related working memory regions and reduced deactivation in regions of the DMN with increasing task difficulty (15). This finding demonstrated a failed disengagement from DMN during working memory tasks in PI patients. It is a complement to previous studies that only found decreased metabolism or decreased activation in cognitive or task-related regions (13, 14). In our study, although we did not found any altered effective

#### Table 2 | Between-group differences in Granger causal influences.


*ROI, regions of interest, which was defined as 6-mm-diameter spheres centered on voxels that exhibited the largest absolute t value in each of the significant clusters in the t map of between-group differences in Granger causal influences.*

*PI, primary insomnia; HC, healthy control, rAI; right anterior insula.*

connectivity from rAI to ECN, we found aberrant effective connectivity from the rAI to regions of DMN at resting state or baseline condition. Our findings offer a parsimonious explanation for failed disengagement from DMN during cognitively demanding tasks (especially the working memory task) in PI patients.

Another finding was that PI patients showed aberrant effective connectivity from the bilateral OFC to the ventral rAI. Besides, the effective connectivity from left OFC to ventral rAI was significantly negatively correlated with ISI scores in PI group. The OFC is an important brain area responsible for emotion and decision-making (49). Previous study showed that PI patients' speed are slower than controls on a vigilance task which only need decision-making (50). Impaired decision-making may also lead to a lack of ability to solve problems in the insomnia patients (51). Indeed, PI patients showed reduced orbitofrontal gray matter volume or density (27, 28). Recent studies found that rAI also acted as a main outflow hub within SN for easier decision-making task (52). Together with previous studies, our findings suggest an abnormal OFC-rAI circuit in PI patients, which might be one of underlying substrate of impaired decision-making observed in PI patients.

We also found that PI patients showed aberrant effective connectivity from the rAI to the left postcentral gyrus (extending to bilateral precuneus) and the bilateral cerebellum posterior lobe. In addition, effective connectivity from ventral rAI to the left postcentral gyrus was significantly negatively correlated with ISI scores in the PI group. The postcentral gyrus and the cerebellum are locations of primary somatosensory cortex and motor control area, respectively. In recent years, several studies have also frequently reported abnormal spontaneous brain activity in the cerebellum posterior lobe as well as the postcentral gyrus in PI (16, 17, 53). On the other hand, existing evidence suggests the anterior insula and the anterior cingulate cortex serve as complementary limbic sensory and motor regions. They work together, similar to the somatosensory and motor cortices (54, 55). Relative to rAI, which is high on the level of the hierarchy due to its function of switching between other large-scale networks, the postcentral gyrus and the cerebellum is lower in the hierarchy (32, 56). Therefore, our findings that aberrant effective connectivity from the rAI to the left postcentral gyrus and the bilateral cerebellum posterior lobe may reflect aberrant top-down sensory and motor control of rAI in PI patients.

Our study had several limitations. First, it was a cross-sectional study, and we cannot directly identify the causal relation between PI and the abnormal effective connectivity. Longitudinal studies may help address this question. Second, we did not directly investigate the inter-network effective connectivity among SN, ECN, and DMN using independent component analysis, even though selected seed regions of rAI for Granger causality analysis was the widely recognized hub node of SN. Future researchers can use independent component analysis to study inter-network effective connectivity in PI. Third, analyses of combination of mental chronometry and Granger causality analysis will increase our understanding of PI. However, we did not do these analyses for technical reasons. Future study is suggested to do these analyses. Finally, the activity of the brain at resting state is not static but is a highly dynamic system. Therefore, static effective connectivity may not be enough to fully characterize the human brain. Future study is suggested to use dynamic FC to investigate the brain in PI.

In summary, we for the first time found aberrant effective connectivity of rAI (a key node of SN) with the posterior DMN hub (precuneus) as well as regions involved in decision-making (OFC) and regions involved in sensori-motor function in PI. These findings suggest an aberrant salience processing system of the rAI, which may be a candidate substrate for cognitive impairment, especially the impairment of working memory and decision-making in PI patients.

### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of "ethics committee of the Guangdong Second Provincial General Hospital" with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the "ethics committee of the Guangdong Second Provincial General Hospital."

### AUTHOR CONTRIBUTIONS

Study concepts/study design or data acquisition, manuscript drafting for important intellectual content, and approval of final version of submitted manuscript: all authors; literature research: CL, YY, KH, SF, and GJ; clinical studies: CL, MD, and GJ; experimental studies: CL and GJ; statistical analysis: CL and SF; and manuscript editing: CL, MD, and GJ.

### FUNDING

This study has received funding by the National Natural Science Foundation of China (grant no.: 81471639); the National Natural Science Foundation of China (grant no.: 81771807); the Natural Science Foundation of Guangdong Provincial (grant no.: 2015- A030313723); the Science and Technology Foundation of Guangdong Province (grant no.: 2016A020215125; 2017A020215077); and the Science and Technology Foundation of Guangzhou City (grant no.: 201607010056).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at https://www.frontiersin.org/articles/10.3389/fneur.2018.00317/ full#supplementary-material.

Figure S1 | The bar graphs demonstrating the mean effective connectivity values in the regions of interest defined as 6-mm-diameter spheres centered on voxels that exhibited the largest absolute *t* value in each of the significant clusters in the *t* map. Error bars indicate SD. Abbreviations: L\_CER\_crus1, left cerebelum\_crus1; R\_CER\_crus1, right cerebelum\_crus1; L\_CER\_6, left cerebelum\_6; L\_precuneus, left precuneus; L\_postcentral gyrus, left postcentral gyrus (extending to bilateral precuneus).

Figure S2 | The bar graphs demonstrating the mean effective connectivity values in the regions of interest defined as 6-mm-diameter spheres centered on voxels that exhibited the largest absolute *t* value in each of the significant clusters in the *t* map. Error bars indicate SD. Abbreviations: Bi\_precuneus, bilateral precuneus; L\_postcentral gyrus, left postcentral gyrus (extending to left precuneus).

Figure S3 | The bar graphs demonstrating the mean effective connectivity values in the regions of interest defined as 6-mm-diameter spheres centered on voxels that exhibited the largest absolute *t* value in each of the significant clusters in the *t* map. Error bars indicate SD. Abbreviations: R\_OFC, right orbitofrontal cortex; L\_OFC, left orbitofrontal cortex.

Figure S4 | The effective connectivity from ventral right anterior insula to the whole brain.

Figure S5 | The effective connectivity from dorsal right anterior insula to the whole brain.

### REFERENCES


Figure S6 | The effective connectivity from the whole brain to ventral right anterior insula.

Figure S7 | The effective connectivity from the whole brain to dorsal right anterior insula.

Figure S8 | The statistical results of the permutation threshold-free cluster enhancement correction (5,000 times permutation, default parameters, FWE corrected, *P* < 0.05). (A) Between-group differences in effective connectivity from ventral right anterior insula (rAI) to the whole brain. (B) Between-group differences in effective connectivity from dorsal rAI to the whole brain. (C) Between-group differences in effective connectivity from the whole brain to ventral right anterior insula.

Figure S9 | Relationships between effective connectivity and sleep and emotion scales in healthy control group.

Figure S10 | Temporal-domain bold signals for seed regions.


networks. *Proc Natl Acad Sci U S A* (2008) 105(34):12569–74. doi:10.1073/pnas. 0800005105


**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 Li, Dong, Yin, Hua, Fu and Jiang. 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 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.*

# Plasticity and Susceptibility of Brain Morphometry Alterations to Insufficient Sleep

Xi-Jian Dai 1,2, Jian Jiang<sup>2</sup> , Zhiqiang Zhang<sup>1</sup> , Xiao Nie2,3, Bi-Xia Liu<sup>4</sup> , Li Pei <sup>2</sup> , Honghan Gong<sup>2</sup> , Jianping Hu<sup>1</sup> , Guangming Lu<sup>1</sup> \* and Yang Zhan<sup>5</sup> \*

<sup>1</sup> Department of Medical Imaging, Medical School of Nanjing University, Jinling Hospital, Nanjing, China, <sup>2</sup> Department of Radiology, The First Affiliated Hospital of Nanchang University, Nangchang, China, <sup>3</sup> Department of Radiology, Yiyang Central Hospital, Yiyang, China, <sup>4</sup> Department of ICU, Jiangxi Cancer Hospital, Nanchang, China, <sup>5</sup> Brain Cognition and Brain Disease Institute, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China

Background: Insufficient sleep is common in daily life and can lead to cognitive impairment. Sleep disturbance also exists in neuropsychiatric diseases. However, whether and how acute and chronic sleep loss affect brain morphology remain largely unknown.

#### Edited by:

Thomas Pollmächer, Klinikum Ingolstadt, Germany

#### Reviewed by:

Axel Steiger, Max-Planck-Institut Für Psychiatrie, Germany Thomas Penzel, Charité Universitätsmedizin Berlin, Germany

#### \*Correspondence:

Guangming Lu cjr.luguangming@vip.163.com Yang Zhan yang.zhan@siat.ac.cn

#### Specialty section:

This article was submitted to Sleep and Chronobiology, a section of the journal Frontiers in Psychiatry

Received: 31 March 2018 Accepted: 31 May 2018 Published: 27 June 2018

#### Citation:

Dai X-J, Jiang J, Zhang Z, Nie X, Liu B-X, Pei L, Gong H, Hu J, Lu G and Zhan Y (2018) Plasticity and Susceptibility of Brain Morphometry Alterations to Insufficient Sleep. Front. Psychiatry 9:266. doi: 10.3389/fpsyt.2018.00266 Methods: We used voxel-based morphology method to study the brain structural changes during sleep deprivation (SD) at six time points of rested wakefulness, 20, 24, 32, 36 h SD, and after one night sleep in 22 healthy subjects, and in 39 patients with chronic primary insomnia relative to 39 status-matched good sleepers. Attention network and spatial memory tests were performed at each SD time point in the SD Procedure. The longitudinal data were analyzed using one-way repeated measures ANOVA, and post-hoc analysis was used to determine the between-group differences.

Results: Acute SD is associated with widespread gray matter volume (GMV) changes in the thalamus, cerebellum, insula and parietal cortex. Insomnia is associated with increased GMV in temporal cortex, insula and cerebellum. Acute SD is associated with brain atrophy and as SD hours prolong more areas show reduced GMV, and after one night sleep the brain atrophy is restored and replaced by increased GMV in brain areas. SD has accumulative negative effects on attention and working memory.

Conclusions: Acute SD and insomnia exhibit distinct morphological changes of GMV. SD has accumulative negative effects on brain morphology and advanced cognitive function. The altered GMV may provide neurobiological basis for attention and memory impairments following sleep loss.

### STATEMENT OF SIGNIFICANCE

Sleep is less frequently studied using imaging techniques than neurological and psychiatric disorders. Whether and how acute and chronic sleep loss affect brain morphology remain largely unknown. We used voxel-based morphology method to study brain structural changes in healthy subjects over multiple time points during sleep deprivation (SD) status and in patients with chronic insomnia. We found that prolonged acute SD together with one night sleep recovery exhibits accumulative atrophic effect and

**17**

recovering plasticity on brain morphology, in line with behavioral changes on attentional tasks. Furthermore, acute SD and chronic insomnia exhibit distinct morphological changes of gray matter volume (GMV) but they also share overlapping GMV changes. The altered GMV may provide structural basis for attention and memory impairments following sleep loss.

Keywords: insomnia, sleep deprivation, voxel-based morphometry, gray matter, attention network test, spatial working memory

### INTRODUCTION

We spend a third of our lives in sleep, yet sleep is less frequently studied using imaging techniques than many neurological and psychiatric disorders. Sleep is increasingly found to have far more health impact than what was previously thought. The precise control of sleep process is the basis of normal life process such as blood, metabolism, immune, endocrine, brain activity, and is the key of plasticity forming, information processing and function implementation (1–4). Sleep deprivation (SD) is associated with a series of maladaptive changes in alertness, judgment, emotion, memory, learning, immunity and central nervous system (5– 12). Short-time SD may influence the expression of certain genes (13) while long-term SD can result in genetic changes (14). Insomnia as a general sleep disorder affects nearly 10– 15% of the adult population (15). Insufficient sleep can lead to cognitive impairment, emotional change, brain dysfunction, psychomotor retardation and metabolic dysregulation (7, 8, 10, 12, 16–21). Despite the adverse socioeconomic impact of insufficient sleep, its neurobiological substrates are still elusive. Evidence suggests that chronic insomnia is accompanied by brain structural and functional changes (20–31). Elucidating brainmorphological changes of insufficient sleep can gain insights on the cognitive and emotional impacts by the sleep loss and bridges the gap between insufficient sleep and neurological or psychiatric disorders. Although SD is a frequently used protocol to investigate the functional consequences and behavioral changes associated with sleep loss (32, 33), what the brain structure changes temporally during the course of acute SD and what the brain structure changes after SD compare to those in patients with chronic primary insomnia remains unknown.

Sleep is associated with increased brain expression of genes involved in regulating macromolecule biosynthesis (34–37), and elevated transcription of genes involved in synthesis and maintenance of cell membrane lipids and myelin in the brain (34, 38, 39). Nevertheless, these structures might be particularly susceptible to insufficient sleep (39, 40). Recently a emerging view that structural brain gray matter and white matter changes can be observed within brief periods of time, from hours to days, following short-term learning (41) or neurotransmitter blockade (42). SD was associated with disturbed level of neurotransmitters (43), neuropeptides (44) and various kinds of cytokines (45) in the brain. In the longer term, rodent studies have shown that chronic sleep restriction and chronic stress are associated with brain structural changes (46, 47). The reported structural changes reflect the underlying pathology of the disease and may determine clinical phenomenology (48). Given the neurochemical changes by the SD and the link between the brain morphology and the neurochemical manipulation, we tested whether the brain structures exhibit changes as a result of insufficient sleep. First, we asked whether SD at different length of time could contribute to the changes in brain morphometry and its plasticity. Second, we examined the brain morphometry in patients with insomnia to understand whether short-term and chronic sleep loss may underlie shared structural basis.

Previous studies suggest that the reported structural changes reflected the underlying pathology of the disease and may determine clinical phenomenology (48). Voxel-based morphometry (VBM) method uses refined image registration and segmentation and provides sensitive measurements on the structural gray matter and white matter changes (49–51). In this study we applied VBM method to explore the dynamic evolution procedure of whole brain morphometry changes in the longitudinal data of 36-h acute SD and in a large sample of patients with insomnia. In the 36 h SD procedure, we performed repeated MRI sessions at 20, 24, 32, and 36 h after the SD started. We also performed one MRI session before the SD started and another one after one-night sleep recovery. Along with each MRI session, attentional network test (ANT) and spatial working memory task (SWM) were performed to evaluate the cognitive vulnerability to SD. In the insomnia study, we collected MRI data from patients with insomnia together with good sleepers (GSs).

### MATERIALS AND METHODS

### Subjects

This study was approved by the Medical Research Ethical Committee of Jinling Hospital and the First Affiliated Hospital of Nanchang University in accordance with the Declaration of Helsinki. The MRI and behavioral data were collected from two studies with a total of 100 subjects including an acute SD study and a chronic insomnia study. In the acute SD study, a total of 22 healthy university students (13 female, 9 male;

**Abbreviations:** SD, Sleep deprivation; GMV, Gray matter volume; VBM, Voxelbased morphometry; ANT, Attentional network test; SWM, Spatial working memory; ROIs, Regions of interest; GSs, Good sleepers; RW, Rested wakefulness; PSQI, Pittsburgh Sleep Quality Index; HAMD, Hamilton Depression Rating Scale; HAMA, Hamilton Anxiety Rating Scale; DSM-IV, Diagnostic and Statistical Manual of Mental Disorders, version 4; ISI, Insomnia Severity Index; SDS, Self-Rating Depression Scale; SAS, Self Rating Anxiety Scale; SRSS, Self-Rating Scale of Sleep; POMS, Profile of Mood States; SPM12, Statistical Parametric Mapping 12; DICOM, Digital Imaging and Communications in Medicine; CSF, Cerebrospinal fluid; DARTEL, Diffeomorphic Anatomical Registration Through Exponentiated Lie algebra; MNI, Montreal Neurological Institute; TIV, Total intracranial volume; FWE, Family-wise error; WMV, White matter volume.

mean age, 21.91 ± 1.38 years, mean ± standard deviation) participated an experiment of 36h SD design. In the chronic insomnia study, 39 patients with chronic primary insomnia (29 female, 10 male; mean age 48.92 ± 11.38 years, mean ± standard deviation) and 39 age-, sex-, and education-matched GSs (26 female, 13 male; mean age, 47.87 ± 9.15 years, mean ± standard deviation) were recruited. All volunteers participated voluntarily and were informed of the purposes, methods, and potential risks of this study, and signed an informed consent form. Twenty-one patients with insomnia (4 males, 17 females) were not the first-time visitors and previously had taken hypnotic or psychoactive medications. The other eighteen patients with insomnia (6 males, 12 females) were drug-naive and had never taken any medications before. The medication history duration was 1 month to 5 years. To avoid the possible effect of the medications, the patients with insomnia were kept medicationsfree for at least 2 weeks prior to the data collection and for the duration of this study, except that three patients with insomnia were medications-free for only 2–4 days. The mean duration of insomnia for patients with insomnia was 6.52 ± 5.65 (years, mean ± standard deviation).

Patients with insomnia met the relevant diagnostic criteria of the International Classification of Sleep Disorders, Second Edition(52), duration of insomnia > 1 year, Pittsburgh Sleep Quality Index (PSQI) score > 5, and sleep diary for >2 weeks duration. Furthermore, they had to report a total sleep time ≤ 6.5 h and (a) sleep onset latency > 45 min or (b) wake after sleep onset > 45 min or (c) total wake time during the sleep period (sleep latency + wake after sleep onset) > 60 min. To evaluate their sleep status, all subjects were asked to wear a Fitbit Flex tracker (http://help.fitbit.com) (20). These data were primarily used to verify sleep-wake diary information and not for independent assessment of inclusion and exclusion criteria.

All GSs and the 22 healthy university students met the following criteria: good sleeping habits, good sleep onset (<30 min) and/or maintenance (without easily wakened or morning awakening symptom) and regular dietary habits as measured by the Fitbit Flex tracker and sleep diary; no consumption of any stimulants, hypnotic or psychoactive medication, during or prior to the study for ≥3 months; PSQI score < 5, and Hamilton Depression Rating Scale (HAMD) and Hamilton Anxiety Rating Scale (HAMA) < 7. All subjects were right-handed. The exclusion criteria for all subjects comprised pathological brain magnetic resonance imaging (MRI) findings; inborn or other acquired diseases; any foreign implants in the body; BMI >32 or <19.8; present or past psychiatric or neurological disorders, substance dependency or substance abuse (including heroin, nicotine, or alcohol addiction); foreign implants in the body; any history of swing shift, night shift, or other shift work within the preceding year; any history of sleep complaints, or other sleep disorders, including hypersomnia, parasomnia, sleep related breathing disorder, sleep related movement disorder, or circadian rhythm sleep disorder, confirmed by overnight polysomnography; any history of significant head trauma or loss of consciousness >30 min; current smoking of more than 10 cigarettes per day; and consumption of >2 caffeinated beverages or potent tea per day.

### SD Procedure

In the acute SD study, the SD Procedure started from 20:00 in the first day and lasted until 8:00 in the fourth day (**Figure 1**). All subjects were asked to arrive the lab at 19:00 in the first day (the day before the SD process) and underwent an MRI session as a base-line. All subjects were asked to sleep in the laboratory at the same time as usual. During this process, the subjects who had poor sleep quality were excluded. The 36 h SD Procedure started at 8:00 in the morning in the second day and lasted until 20:00 in the third day. The participants were required to stay awake during the entire time of the SD procedure. All subjects were not allowed to lie down and do some vigorous exercise, and they were not allowed to continue to do one thing for a long time, such as read and talk about an exciting topic. The food and water were provided during the SD procedure. Specially, all subjects eat the same food at every meal during the SD procedure to control the food intake, but the water are not controlled. The temperature of the room was maintained between 23 and 27◦C. The staffs of the research team took charge of monitoring in turns through video monitors to make sure that the participants did not fall asleep during the SD procedure. If there were any signs of falling asleep, the participants were awakened by an alarm clock immediately.

Each subject underwent MRI sessions at the following time points: the start of the experiment during rested wakefulness (RW), 20, 24, 32, and 36 h after the experiment started (**Figure 1**). The subjects then spent one night of sleep for recovery and underwent another MRI session at 8:00 in the next morning in the fourth day. Furthermore, each subject underwent the longterm task of ANT and short-term simple task of SWM at each measurement time point before each of those MRI scans.

### Insomnia Procedure

An experienced psychiatrist evaluated the patients with insomnia with the Diagnostic and Statistical Manual of Mental Disorders, version 4 (DSM-IV) (53) for the life history of psychiatric disorders, as well as an unstructured clinical interview for the history of medicine and sleep disorders. The patients with insomnia and status-matched GSs were asked to complete a number of questionnaires, including PSQI (54), Insomnia Severity Index (ISI) (55), Self-Rating Depression Scale (SDS) (56), Self Rating Anxiety Scale (SAS) (57), HAMD (58), HAMA (59), Self-Rating Scale of Sleep (SRSS) and Profile of Mood States (POMS) (60). The POMS questionnaire contains of 7 indexes, including 5 negative emotion indexes (nervousness, wrath, fatigue, depression and confusion) and 2 positive emotion indexes (energy and self-esteem). Then the patients with insomnia and the status-matched GSs each underwent the MRI scan once between 19:00 and 20:00.

### Attention Network Test (ANT)

The ANT, adapted from Fan et al.'s study (61, 62), contains of three cue conditions (no cue, center cue, spatial cue) and two target conditions (congruent and incongruent). The visual stimuli consisted of a row of 5 horizontal black arrows pointing leftward or rightward with the target arrow in the center. The participants responded to the direction of central arrow by pressing the left or right buttons of the computer mouse. The task measures alerting, orienting and conflict effects by calculate time

difference between the response time and the presentation time under three different cue conditions.

The accuracy rate using corrected recognition, reaction time using only trials with correct responses, and lapse rate using missing recognition, were calculated. Finally, the intraindividual coefficient of variation was calculated for each participant by dividing the mean value of accuracy rate or correct reaction time by that of standard deviation.

### Spatial Working Memory Test (SWM)

The SWM was based on visual recognition of a series of 6 × 6 smaller squares filled in a large square with size of 7.2<sup>∗</sup> 7.2 mm<sup>2</sup> (**Figure 2**). All these 36 small squares were filled with white. First, there was only shown a single small square filled with black in one location among these 36 smaller square. Next, the second and third small square was filled with black in another location respectively. Then, the fourth square will be filled with black immediately once the first small black square was recovered from black to white. At this time, the subjects were asked to make a keypress response to determine whether the location of the fourth black square was in the same location with the first black square, or subsequent the fifth black square was in the same location with the second black square, and so on. If they are in the same location, the subjects were asked to press the right button, conversely, the left button was conducted. The accuracy rate using corrected recognition, reaction time using only trials with correct responses, and lapse rate using missing recognition, were calculated. Finally, the intraindividual coefficient of variation was calculated for each participant by dividing the mean value of accuracy rate or correct reaction time by that of standard deviation.

### MRI Parameter

The MRI scan was performed on a 3-Tesla MR scanner (Trio, Siemens, Erlangen, Germany). High-resolution T1-weighted anatomical images were acquired with a three-dimensional spoiled gradient-recalled sequence in sagittal orientation: 176 images (repetition time = 1,900 ms, echo time = 2.26 ms, thickness = 1.0 mm, gap = 0.5 mm, acquisition matrix = 256 × 256, field of view = 250 × 250 mm, flip angle = 9 0 ) were obtained. A simple questionnaire was administered immediately after the ∼3-min MRI scan to ask whether the subjects were awake during the scan. The data of subjects who fell asleep during the scan were excluded.

### Voxel-Based Morphometry (VBM)

MRIcro software (www.MRIcro.com) was used to ensure data quality. The data pre-processing was conducted using the available CAT12 toolbox (http://dbm.neuro.uni-jena.de/cat12/) which is based on Statistical Parametric Mapping 12 (SPM12, http://www.fil.ion.ucl.ac.uk/spm). First, the Digital Imaging and Communications in Medicine (DICOM) standard images were transformed into NIFIT format and were realigned into sagittal orientation. The images were corrected for bias field inhomogeneity by linear (12-parameter affine) and nonlinear transformations. Next, the structural images were segmented into gray matter, white matter, and cerebrospinal fluid (CSF). Diffeomorphic Anatomical Registration Through Exponentiated Lie algebra (DARTEL) segmentation procedure was performed in the present study. The 36 h SD Procedure were analyzed using segment procedures for the longitudinal data. The original unwrapped individual gray matter and white matter segmentations were warped to a newly constructed template with a combination of linear and nonlinear registration.

Then the data were spatially normalized using East Asian brain template to the Montreal Neurological Institute (MNI; http://www.mni.mcgill.ca/) space. The segmented data were modulated and smoothed using a Gaussian kernel of 8 × 8 × 8 mm<sup>3</sup> full-width at half- maximum.

### Multiple Linear Regression Analysis

Multiple linear regression analysis was performed to evaluate the relationships between the behavioral performance (dependent variable) in the ANT and SWM and the beta value of the main effect brain regions (independent variable) in each group of the 36 h SD study.

### Statistics

In the SD study, the behavioral data of the ANT and the SWM were analyzed using one-way repeated measures ANOVA. In the insomnia study, the demographic factors (age, education, and years of education) and the sleep questionnaire data were compared between the patients with insomnia and the GSs using two sample t-test. Chi-square (χ 2 ) test was used for categorical data. The statistical analysis was performed using SPSS version IBM 21.0.

For the VBM data of the SD study, one-way within-subject repeated measures ANOVA was used to analyze the longitudinal data across the six time points during the 36 h process. The different brain regions of the main effect were saved as a mask. For the post-hoc analysis between two time points, we either calculated the product between the mask and the T maps or analyzed the difference without applying the mask.

For the VBM data in the insomnia study, unpaired t-test was used to investigate the gray matter volume (GMV) difference between the patients with insomnia and the GSs with the age, sex, years of education and total intracranial volume (TIV) as nuisance covariates of no interest.

We analyzed group differences in two ways. First, we used a threshold of p < 0.05, corrected for multiple comparisons by family-wise error (FWE) method. Second, we used an uncorrected statistical threshold of p < 0.001 with a minimum cluster size (k) of 100 voxels if the correction for multiple comparison failed to detect any difference.

### RESULTS

### Main Effect in 36 h SD Study

One-way within-subject repeated measures ANOVA with the SD time points as main factor revealed significant GMV differences in the right cerebellum anterior lobe, bilateral striatum (caudate), bilateral thalamus, bilateral insula (BA13), bilateral somatosensory association cortex (paracentral lobule, BA5; precuneus, BA7), bilateral somatosensory cortex (BA2), left superior parietal lobule (BA40), bilateral inferior parietal lobule (BA40), right supplementary motor area (SMA; BA6), bilateral posterior cingulate cortex extending to corpus callosum (BA23), and right cingulate cortex (BA24) [F(5, 105) = 8.637, p < 0.05, k ≥ 100, corrected by FWE; **Supplemental Table 1**, **Figure 3**].

To account for the intra-individual differences, we then examined the beta values of the main effect areas from the ANOVA in each individual subject. Eighteen out of the 22 subjects (81.82%) exhibited smaller total and mean GMV at 36 h

SD compared to RW, and the other 4 subjects showed increased GMV (in total 1.02% mean decrease). After one night sleep recovery 20 out of the 22 subjects (90.91%) showed larger GMV and the other two subjects showed decreased GMV (in total 1.75% mean increase) compared to RW. In all subjects, from RW to 36 h SD and from 36 h SD to one night sleep recovery, the total and mean GMV in the main effect brain areas showed a tendency of reduction first and then increase (**Figure 4**).

### Post-hoc Tests in 36 h SD Study

To understand how SD at different length of time could contribute to the brain morphometry changes, we applied posthoc tests to assess the GMV differences between various SD time points and the time point of RW using main effect brain regions as mask with an uncorrected statistical threshold (p < 0.001, k ≥ 100, uncorrected; **Supplemental Table 2**, **Figure 5**). A number of brain areas showed increased GMV at 20, 24, 32, and 36 h after the SD started, including the left striatum, right middle cingulate cortex (BA24) and bilateral posterior cingulate cortex extending to corpus callosum (BA23) (**Figures 5A–D**). On the other hand, a number of brain areas began to show decreased GMV at 32 h after the SD started, including the right thalamus, right insula (BA13), bilateral somatosensory association cortex (paracentral lobule, BA5; precuneus, BA7) and right inferior parietal lobule (BA40) (**Figure 5C**). Interestingly, 4 h later at 36 h after the SD started, the number of brain areas with decreased GMV increased, expanding to bilateral somatosensory cortex (BA2, BA3) and right SMA (BA6) (**Figure 5D**). After one night sleep recovery, no areas with decreased GMV were found but the right cerebellum anterior lobe, right striatum (caudate body), bilateral thalamus, bilateral insula (BA13), bilateral somatosensory association cortex (paracentral lobule, BA5; precuneus, BA7), bilateral inferior parietal lobule (BA40), bilateral somatosensory cortex (BA2), left superior parietal lobule (BA40) and right SMA (BA6) showed increased GMV (**Figure 5E**). Many of these brain areas with increased GMV (**Figure 5E**) were in the similar location as the areas showing decreased GMV at the SD time points relative to the time point of RW (**Figures 5C,D**), but they also contained more brain areas.

Additionally we analyzed the GMV differences at each SD time point relative to the time point of RW using a corrected statistical threshold without applying the mask (p < 0.05, FWE corrected; **Supplemental Table 3**, **Figure 6**). This allowed to investigate the GMV changes on a broader scale. At 20 h after the SD started, VBM did not reveal any GMV difference relative to the RW. At 24 h after the SD started, bilateral striatum, bilateral cingulate gyrus (BA23), right posterior cingulate cortex (BA30) and right medial prefrontal cortex (BA10) showed increased GMV (**Figure 6A**). No decreased GMV was found. At 32 h after the SD started, only right cingulate gyrus (t = 5.0707; x = 13.5, y = −23.5, z = 36.5) showed increased GMV and no decreased GMV was found. At 36 h after the SD started, right cerebellum posterior lobe, left striatum and left somatosensory cortex (BA3) showed increased GMV (**Figure 6B**). The areas that showed decreased GMV included the left cingulate gyrus (BA24) and right temporal pole (BA38) (**Figure 6B**). After one night sleep recovery, bilateral thalamus, left orbitofrontal cortex (BA11), bilateral insula (BA13), right visual association cortex (BA18), bilateral somatosensory association cortex (BA7),

each SD time point against RW was conducted as the product between the GMV differences of each time point and the GMV differences of main effect brain areas. Brain areas that showed GMV differences at each time point during the 36 h SD procedure against RW from the post-hoc tests, including the time point of 20 h SD (A), 24 h SD (B), 32 h SD (C), 36 h SD (D), and after one night sleep recovery (E). Red areas denote increased GMV (A–E) and green areas denote decreased GMV (C–D) in brain areas.

bilateral parietal lobe (BA2, BA40) and left primary motor cortex (BA4) showed increased GMV (**Figure 6C**). No decreased GMV was found.

### Sample Characteristics of Patients With Insomnia

The demographic characteristics of the patients with insomnia sample are presented in **Table 1**. There were no significant differences in sex distribution (p = 0.456), mean age (p = 0.654), mean education (p = 0.694) and PSQI time in bed (p = 0.725). However, compared with GSs, patients with insomnia showed shorter PSQI total sleep time, lower PSQI sleep efficiency, and higher PSQI score, higher SRSS score, higher SAS score, higher SDS score, higher HAMA score, higher HAMD score, higher POMS score, higher score of five negative index in POMS and lower score of two positive index in POMS (p < 0.001).

### VBM Difference in Patients With Insomnia vs. GSs

There were no significant differences in the TIV, GMV, white matter volume (WMV), CSF volume, GMV/TIV, WMV/TIV, CSF volume (CSFM)/TIV and GMV/WMV between patients with insomnia and GSs (p > 0.05) (**Table 2**).

VBM did not reveal any GMV difference between patients with insomnia and GSs using a two sample t-test (p < 0.05, FWE corrected). When using an uncorrected statistical criterion (p < 0.001, k ≥ 100), we found GMV differences localized in the right hemisphere (**Supplemental Table 3**, **Figure 7**), with increased GMV in the fusiform gyrus (BA 37), cluster of cerebellum anterior lobe and visual association cortex (BA18), cluster of claustrum and insula (BA13), primary auditory area (superior temporal gyrus, BA22, BA42) and SMA (BA 6), and with decreased GMV in the visual association cortex (BA18).

### Behavioral Findings of 36 h SD Study

We examined the attention and working memory in the ANT and SWM tests at different time points during the SD Procedure. The accuracy rate, reaction time and lapse rate of the ANT were different across the six SD time points using one-way repeated measures ANOVA [Greenhouse-Geisser correction, accuracy rate: F(1.956, 41.072) = 8.299, p = 0.001, **Figure 8A**; reaction time F(3.268, 68.631) = 11.242, p < 0.001, **Figure 8B**; lapse rate F(1.975, 41.473) = 7.034, p = 0.002, **Figure 8C**]. The accuracy rate showed a tendency of gradual decrease (**Figure 8A**), and the reaction time and lapse rate showed a tendency of gradual increase as the SD hours prolonged (**Figures 8B,C**). The accuracy and reaction time restored after one night sleep recovery, but the accuracy rate and reaction time did not reach the level of RW completely. Interestingly, the subjects showed the lowest accuracy rate, longest reaction time and highest lapse rate at the time point of 24 h after the SD started. Furthermore, we measured the alertness, orienting and executive control from the ANT processes. The reaction time of spatial orientation and executive control were different across the SD time points [orienting: F(5, 105) = 2.683, p = 0.025; executive control: F(5, 105) = 6.003, p < 0.001; **Figure 8D**]. The reaction time of alertness was not different across the six time points [F(5, 105) = 0.277, p = 0.925;

#### TABLE 1 | Group characteristics of patients with insomnia and good sleepers.


#chi-square value; Data are mean ± standard deviation values; GSs, Good sleepers; N/A, Not applicable; Self-rating Anxiety Scale and Self-Rating Depression Scale showed the standard score. The five negative index comprised nervousness, wrath, fatigue, depression and confusion, and the two positive index comprised energy and self-esteem.


Data are mean ± standard deviation values; GSs, Good sleepers; TIV, Total intracranial volume; GMV, Gray matter volume; WMV, White matter volume; CSFM, Cerebrospinal fluid matter; Vol, Volume.

**Figure 8D**]. In the SWM test, the accuracy rate did not show an effect of SD time [F(5, 105) = 0.935, p = 0.461; **Figure 8E**], however there was a trend of gradual decrease as the SD hours prolonged and then a trend of increase after one night sleep recovery.

The intra-individual coefficient of variability for ANT accuracy rate, ANT reaction time, and SWM accuracy rate showed a tendency of increase as the SD hours prolonged and showed decrease after one night sleep recovery (**Figure 8F**). The accuracy rate of the ANT showed the highest intra-individual coefficient of variability at the time point of 24 h SD compared to other five time points.

### Intra-Individual Differences in Behavior for Each Subject

We conducted the intra-individual differences in behavior as the intra-individual GMV differences in brain areas. We calculated the accuracy rate and reaction time for each subject (**Supplemental Figure 1**). In the ANT test, eighteen of the subjects shower lower accuracy rate, and the other four subjects showed higher accuracy rate at the time point of 36 h SD compared to RW (**Supplemental Figure 1A**). Twelve of the subjects demonstrated lower accuracy rate, two subjects showed no differences, and the other eight subjects indicated higher accuracy rate after one night sleep recover compared with RW (**Supplemental Figure 1A**). Seventeen of the subjects demonstrated longer reaction time, and the other five subjects showed shorter reaction time at the time point of 36h SD compared to RW (**Supplemental Figure 1B**). Ten of the subjects showed longer reaction time and the other twelve of the subjects indicated shorter reaction time after one night sleep recover compared with RW (**Supplemental Figure 1B**). In individual subjects, from RW to 36 h SD and from 36 h SD to one night sleep recovery, the accuracy rate in ANT showed a tendency of reduction first (36 h SD vs. RW, 3.94% mean decrease) and

Frontiers in Psychiatry | www.frontiersin.org

then increase (recovery vs. RW, 1.39% mean decrease), and the reaction time showed an inverse tendency of increase first (36 h SD vs. RW, 9.00% mean increase) and then decrease (recovery vs. RW, 0.05% mean increase) (**Supplemental Figure 1**).

In the SWM, nine of the subjects demonstrated lower accuracy rate, seven subjects showed no differences, and the other six subjects indicated higher accuracy rate at the time point of 36 h SD compared to RW (36 h SD vs. RW, 2.7% mean decrease) (**Supplemental Figure 1C**). Five of the subjects demonstrated lower accuracy rate, nine subjects showed no differences, and the other eight subjects indicated higher accuracy rate after one night sleep recover compared with RW (recovery vs. RW, 1.46% mean increase) (**Supplemental Figure 1C**). Eleven of the subjects demonstrated longer reaction time, and the other eleven subjects showed shorter reaction time at the time point of 36 h SD compared to RW (36 h SD vs. RW, 0.26% mean decrease) (**Supplemental Figure 1D**). Five of the subjects showed longer reaction time and the other seventeen of the subjects indicated shorter reaction time after one night sleep recover compared with RW (recovery vs. RW, 5.39% mean decrease) (**Supplemental Figure 1D**).

### Regression Analysis Between VBM and Behavior

To investigate whether the structural changes during the SD status may have correlations with the behavior, we performed multiple linear regression between the behavioral parameters in the attention and working memory tasks and the beta values of the main effect brain areas at each SD time point (**Table 3**). Across the SD time points, the accuracy rate in the ANT and the SWM tests showed linear correlations with the beta value of many brain areas, including the somatosensory association cortex and insula. At the time points of RW and one night sleep recovery, linear relationships were found in the parietal lobe (somatosensory cortex and inferior parietal lobule).

The alertness in the ANT showed linear relationships with the beta value of the striatum, parietal lobe, insula and thalamus across the SD time points, and the executive control in the ANT showed linear relationships with the beta value of the insula, somatosensory association cortex, parietal lobe and SMA. However, no brain areas showed correlations with the alertness or executive control at the time point of RW and one night sleep recovery. Interestingly, at the time point of 24 h SD, the alertness displayed linear correlation with the beta value of the striatum and the executive control showed linear relationship with the beta value of the SMA, and no correlation was found at other time points.

## DISCUSSION

## Brain Morphological Changes During the Acute SD

We found in our SD study that insufficient sleep is associated with widespread brain morphological changes. Although molecular basis for the microstructure-level changes requires further investigation, acute SD is associated with altered gene expression involved in macromolecule biosynthesis (34–37) in human studies and altered gene expression involved in cell membrane lipids and myelin in the mouse brain (34, 38, 39). The susceptibility of these cellular substrates to the rapid changes following sleep loss might contribute to the brain microstructure changes as we observed in our VBM analysis. In the animal studies, SD could lead to neuronal marker changes for apoptosis and morphology, and these changes were restored after sleep recovery (63, 64). Consistent with these patterns, our VBM data showed progressive structural atrophy as SD hours prolonged and these changes were restored and replaced by extensive and larger morphologic brain inflation after one night sleep recovery. Previous diffusion tensor imaging study showed that SD is associated with widespread fractional anisotropy decreases in several brain areas and as the waking prolonged the decreases become larger (65), which may associated with the reduced interstitial space volumes and increased resistance to water flux in the brain after waking than during sleep (66) and particularly susceptible of cell membrane lipids and myelin to insufficient sleep (39, 40). In our longitudinal data of 36 h SD procedure, the brain atrophy began to appear at 32 h SD and aggravated at 36 h SD. In agreement with the brain morphology, the accumulative negative effects were found in attention and spatial memory tests, but after one night sleep recovery they were restored incompletely showing a delayed recovery. We hypothesized that the VBM changes observed in the present study are more likely to be related to the changes in tissue hydration or other phenomena.

## Circadian Rhythm Influences During SD

A recent study has shown that the brain responses during the day and prolonged wakefulness showed circadian rhythmic patterns (67). Subcortical areas including the striatum and thalamus showed strong correlations with the melatonin levels and these areas showed increased responses when later hours in the day start. We also found increased GMV in the striatum starting at the time point of 20 h SD, and these increased GMV remained into the later stage of the SD. Evidence also indicates that individuals with late chronotype who prefer to go to bed late in

the evening had structural differences in the cingulate cortex and corpus callosum (68). In our SD study, we also found increased GMV in the corpus callosum and cingulate cortex. Considering the similar brain areas found in our study and the others, the structural changes in our VBM analysis may reflect the influence of circadian rhythm.

ANT, accuracy rate of the ANT, and accuracy rate of the SWM. Data are presented as mean ± standard values.

At the time point of 24 h SD, the subjects exhibited the lowest accuracy rate, longest reaction time and highest lapse rate in the ANT compared to the other time points. Accordingly, at this time point a number of increased GMV areas in the bilateral striatum and bilateral cingulate cortex extending to corpus callosum were found, and these areas existed even after the FWE correction. Specifically, we found that at 24 h SD the alertness displayed linear correlation with the beta value of striatum and the executive control showed linear correlation with the beta value of SMA, and these relationships were not found at other SD time points. At this time point of 8:00 a.m. in the morning, the participants usually woke up in their daily schedule in the process of getting out of bed and they showed reduced alertness. The SMA area was implicated in sensory processing, working memory, executive control and spatial-bodily attention (69, 70). Therefore the structural changes we observed in our data may underlie the circadian rhythm and prolonged wakefulness to modulate the attentional performance.

### Acute SD vs. Chronic Insomnia

Chronic insomnia is thought to be maintained by excessive negatively toned cognitive activity with autonomic arousal and emotional distress (71). The exaggerated cortical and somatic



activation can lead to increased sensory information processing and inability to initiate or maintain sleep (20, 72, 73), and may be a result of increased brain activities. We found that both in the SD study and the insomnia study that the insula and cerebellum showed increased GMV. This demonstrates that acute and chronic sleep loss may also share similar neurobiological representation in brain morphology. In the SD study after one night sleep recovery only brain areas with increased GMV were found. Similarly, patients with insomnia also mainly found morphological differences in brain regions with increased GMV. This demonstrates that the status of subjects who underwent SD process and then received one night sleep recovery may exhibit analogous brain activation characteristics to the status of patients with insomnia who underwent subjective experience of chronically disturbed and non-refreshing sleep.

Although using an uncorrected statistical threshold we found a number of brain areas with GMV changes in patients with insomnia, using a strict criterion we did not observe differences between patients with insomnia and GSs. This probably demonstrates that the patients with insomnia were not prone to the brain microstructure changes already. Under the less stringent method we found altered GMV in the patients with insomnia in the temporal cortex, primary auditory area, insula, SMA and visual association cortex. For the acute SD the main effect areas with GMV changes were showed in the sensory cortex, motor cortex and subcortical thalamus. Therefore the two types of short-term and chronic sleep losses mainly exhibited non-overlapping altered brain areas.

In the insomnia study, the superior temporal cortex with increased GMV contains the primary auditory area (BA42). Previous study has shown that normal activation of the auditory cortex is decreased to help maintain sleep in response to external stimuli (74). Therefore, our observation of increased GMV in the auditory cortex may highlight the reduced capacity to disengage from external information processing of auditory stimuli, which was consistent with the clinical characteristics of insomnia patients with shallow sleep and increased sensitivity to surrounding environments. Our data therefore support the theory of hyperarousal, which may be a core predisposing or perpetuating factor of ultimately hampering the ability to initiate or maintain sleep.

Previous meta-analytical data demonstrated that the threat or anxiety hypothesis is associate with insula (75), and the craving hypothesis is associated with ventral striatum and cingulated cortex (76). Patients with insomnia underwent prolonged experience of chronically disturbed and non-refreshing sleep, and may display threat or anxiety in response to sleep-related cues. Subjects who underwent acute SD process may mainly display craving to sleep but not threat or anxiety (77). Our previous neuroimaging studies also found that insufficient sleep resulted in abnormal regional brain activity in the threat-related brain areas and craving-related brain areas (8, 20, 21, 78–80). Our observation of increased GMV in the insula or cingulated cortex in the insomnia study and increased GMV in the striatum in the SD study might support the theory of threat and craving hypothesis.

The paracentral lobule is considered to be negatively correlated with vigor activity (81). It has been found that the inferior parietal cortex area is impaired after SD and may represent an most reliable early biological marker of individual resistance to SD (7, 8, 82, 83). The postcentral gyrus is the main receptive region for external stimuli as the location of the primary somatosensory cortex. Recently the postcentral gyrus was shown to be implicated with the default mode network (84), which is a functional brain hub showing coupled slow signal fluctuations in the absence of external stimuli during restful waking and sleep (85). In the SD study, these areas with decreased GMV were found with several correlations with the ANT and SWM. Our observations of decreased GMV in theses somatosensory areas after acute SD in individuals who showed a possible insufficiency to enter into "resting state" status may reflect inhibition in sensory-informational processing and difficulties in cognitive function.

### CONCLUSIONS

In summary, acute SD and insomnia showed widespread changes in gray matter microstructure with shared but also distinct neurobiological representation in brain morphology. Acute SD may be associated with inhibition in sensory-informational processing with decreased GMV in the somatosensory areas to compensate for the effects of sleep loss on advanced cognitive function, while the insomnia may be associated with inability to disengage from external information processing of auditory stimuli with increased GMV in the primary auditory area. Prolonged acute SD together with one night sleep recovery exhibit accumulative atrophic effect and recovering plasticity on brain morphology, in line with the behavioral changes on attentional and working memory tasks. Taken together, clarifying the biological underpinnings of these microstructural alterations could advance our understanding of the neurobiological mechanism of waking and sleep. One of the strengths of the present study is the relatively large sample size in the insomnia study and longitudinal data in the SD study; however, there are several limitations that should be noted. First, our findings are limited by the use of the Fitbit Flex tracker to monitor the sleep quality in our experience (20, 86). Although we cannot provide direct evidence to prove whether the FITBIT tracker provides a valid and reliable measure of objective sleep, we compared some patients' data between the FITBIT and the PSG, and found the results were similar. In fact, our sample was screened to exclude individuals with medical or psychiatric disorders that may affect sleep, and the diagnosis of primary insomnia mainly depends on the experience of senior physicians who have been working for more than 20 years. Second, the subjects were not monitored by continuous EEG in the SD procedure, but a simple questionnaire was administered immediately after the MRI scan to ask whether the subjects were awake during the scan. The data of subjects who fell asleep during the scan were excluded.

## AUTHOR CONTRIBUTIONS

X-JD and YZ wrote the main manuscript text. X-JD, JJ, LP, HG, GL, and YZ conceived and designed the whole experiment. X-JD, XN, and B-XL collected the data. X-JD, JH, and ZZ analyzed the data.

### ACKNOWLEDGMENTS

This work was supported by grants from National Natural Science Foundation of China (grant No 81701678), Science and Technology Support Program of Jiangxi Province (grant No: 20141BBG70026), One Hundred Talents Program of CAS (YZ), Shenzhen city government KQCX2015033117354153 (YZ), Guangdong Innovative and Entrepreneurial Research Team Program No. 2014ZT05S020 (YZ) and Shenzhen Peacock Innovation Team Project KQTD20140630180249366 (YZ).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyt. 2018.00266/full#supplementary-material

Supplemental Figure 1 | Accuracy rate and reaction time in the ANT and spatial working memory (SWM) for each subject in the 36 h sleep deprivation (SD) study. (A) Accuracy rate of the ANT for each subject. (B) Reaction time of the ANT for each subject. (C) Accuracy rate of the SWM for each subject. (B) Reaction time of the SWM for each subject. In all subjects, from rested wakefulness (RW) to 36 h SD and from 36 h SD to one night sleep recovery, the accuracy rate showed a tendency of reduction first and then increase, and the reaction time showed a tendency of increase first and then decrease (A–D).

Supplemental Table 1 | The gray matter volume differences of main effect in the 36 h sleep deprivation study. R, right; L, left; BA, Brodmann's area; MNI, montreal neurological institute; N/A, Not applicable. The statistical threshold was set at corrected voxel threshold of p < 0.05 with a minimum cluster threshold of 100 voxels, corrected by family-wise error.

Supplemental Table 2 | The post-hoc differences of gray matter volume with the product with the mask of the different brain regions of main effect in the 36 h sleep deprivation (SD) study. RW, Rested wakefulness; R, right; L, left; BA, Brodmann's area; MNI, montreal neurological institute; N/A, Not applicable. The statistical threshold was set at uncorrected voxel threshold of p < 0.001 with a minimum cluster threshold of 100 voxels (t = 3.1697).

Supplemental Table 3 | The gray matter volume differences without applying mask method in the 36 h sleep deprivation (SD) study and in the chronic insomnia

### REFERENCES


study. RW, Rested wakefulness; R, right; L, left; BA, Brodmann's area; MNI, montreal neurological institute; N/A, Not applicable; GSs, good sleepers. The statistical threshold was set at family-wise error corrected voxel threshold of p < 0.05 of each time in the 36 h SD study without product with the mask image of main effect, and at uncorrected voxel threshold of p < 0.001 with a minimum cluster threshold of 100 voxels in patients with insomnia.


**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 Dai, Jiang, Zhang, Nie, Liu, Pei, Gong, Hu, Lu and Zhan. 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 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.

# Rested-Baseline Responsivity of the Ventral Striatum Is Associated With Caloric and Macronutrient Intake During One Night of Sleep Deprivation

Brieann C. Satterfield<sup>1</sup> , Adam C. Raikes <sup>1</sup> and William D. S. Killgore1,2 \*

<sup>1</sup> Social, Cognitive, and Affective Neuroscience Laboratory, Department of Psychiatry, College of Medicine, University of Arizona, Tucson, AZ, United States, <sup>2</sup> Department of Psychiatry, McLean Hospital, Harvard Medical School, Belmont, MA, United States

#### Edited by:

Hengyi Rao, University of Pennsylvania, United States

#### Reviewed by:

Angel Nunez, Universidad Autónoma de Madrid, Spain Qihong Zou, Peking University, China Tingyong Feng, Southwest University, China

\*Correspondence:

William D. S. Killgore killgore@psychiatry.arizona.edu

#### Specialty section:

This article was submitted to Sleep and Chronobiology, a section of the journal Frontiers in Psychiatry

Received: 07 August 2018 Accepted: 19 December 2018 Published: 17 January 2019

### Citation:

Satterfield BC, Raikes AC and Killgore WDS (2019) Rested-Baseline Responsivity of the Ventral Striatum Is Associated With Caloric and Macronutrient Intake During One Night of Sleep Deprivation. Front. Psychiatry 9:749. doi: 10.3389/fpsyt.2018.00749 Background: Sleep loss contributes to obesity through a variety of mechanisms, including neuroendocrine functioning, increased hunger, and increased food intake. Additionally, sleep loss alters functional activation within brain regions associated with reward and behavioral control. However, it remains unknown whether individual differences in baseline neural functioning can predict eating behaviors during total sleep deprivation (TSD). We used functional magnetic resonance imaging (fMRI) to test the hypothesis that individuals with increased baseline responsiveness within reward regions are more vulnerable to TSD-induced overeating.

Methods: N = 45 subjects completed several fMRI scans during a single pre-TSD session that included performance on the Multi-Source Interference Task (MSIT) and the n-back task. Subjects returned to the laboratory for an overnight TSD session, during which they were given ad libitum access to 10,900 kcal of food. Leftover food and packaging were collected every 6 h (00:00, 06:00, and 12:00) to measure total food consumption. Subjects reported sleepiness every hour and performed a food rating task every 3 h.

Results: Functional activation within the ventral striatum during the MSIT and n-back positively correlated with total caloric and carbohydrate intake during the final 6 h (06:00–12:00) of TSD. Activation within the middle and superior temporal gyri during the MSIT also correlated with total carbohydrates consumed. Food consumption did not correlate with subjective sleepiness, hunger, or food desire.

Conclusions: Individual differences in neural activity of reward processing areas (i.e., nucleus accumbens) prior to sleep deprivation are associated with an individual's propensity to overeat during subsequent sleep deprivation. This suggests that individual differences within reward processing pathways are potential key factors in sleep loss related overeating. Sleep loss and obesity are tightly linked. Both phenomena have been associated with increased neural activation in regions associated with reward, inhibitory control, and disrupted dopamine signaling. Elevated baseline reward sensitivity

**33**

in the ventral striatum appears to be further compounded by sleep deprivation induced dysfunction in the reward neurocircuitry, increasing the likelihood of overeating. Our findings suggest that large individual differences in baseline responsiveness of hedonic reward pathways may modulate the association between sleep loss and obesity.

Keywords: sleep deprivation, ventral striatum, nucleus accumbens, food consumption, reward

### INTRODUCTION

The social and economic demands of living in our modern 24/7 society have contributed to two pervasive problems: insufficient sleep and obesity (1). The National Sleep Foundation recommends that adults obtain ≥7 h of sleep per night (2). However, 35% of adults in the United States sleep <7 h per night (3), a nightly duration that has been on the decline for the last several decades. Simultaneously, obesity rates have dramatically increased, with over one third of the adult population being classified as obese (4). Epidemiological evidence suggests a strong association between the declining levels of sleep and increasing rates of obesity (1, 5–7).

Sleep loss contributes to weight gain through several physiological, behavioral, and neural mechanisms. From a physiological perspective, sleep loss disrupts the normal functioning of hormones that regulate appetite (ghrelin) and satiety (leptin). Studies have demonstrated that sleep restriction results in elevated levels of ghrelin and reduced levels of leptin, leading to increased feelings of hunger (8–11). Behavioral studies of sleep restriction and sleep deprivation have also demonstrated how sleep loss impacts eating behaviors. Sleep loss leads to increased energy intake, while energy expenditure does not change (12, 13), contributing to overall weight gain (14). Individuals tend to increase their overall total caloric intake (13–16), especially in the form of carbohydrates (16) and fat (13) during sleep loss. These extra calories come from snacks (17, 18) and increased meal frequency (14).

While studies have focused on how sleep loss disrupts neuroendocrine signaling and modifies eating behavior, few have investigated the brain's neural response to food stimuli. There are several interacting neural networks which drive food intake behavior, including connections between several key cortical [orbital frontal cortex (OFC), prefrontal cortex (PFC), insula, and anterior cingulate cortex (ACC)] and limbic (amygdala, hippocampus, and basal ganglia) regions (19–22). Volkow et al. (22) suggest that obesity may be the result of an imbalance between neural circuits that promote eating behaviors (rewardsaliency and motivation-drive circuits) and those that control and inhibit behavioral responses (learning-condition and inhibitory control-emotion regulation circuits) (22). Sleep loss also disrupts communication between cortical regulatory and subcortical reward systems. There is elevated neural activity in regions associated with reward and risky decision-making and attenuated activity in cortical regions associated with inhibitory control (23–25). The parallels in altered signaling patterns in these key systems for both obesity and sleep loss suggest that both conditions result in a loss of top-down inhibitory control over reward-processing regions.

Studies investigating the neural responses to food stimuli associated with daytime sleepiness, sleep restriction, and total sleep deprivation (TSD) have found evidence supporting a disruption of behavioral control and reward systems (21, 26–30). Excessive daytime sleepiness is associated with reduced neural activation in the ventromedial PFC (vmPFC), an area involved in inhibitory control, when viewing images of high calorie foods. Reduced activation in this region also correlates with subjective difficulty restricting food consumption (21). Lack of inhibition from the frontal control system may release a "brake" on subcortical pathways involved in modulating reward-based behaviors, such as eating.

Sleep restriction also impacts neural activity in regions associated with reward, including the nucleus accumbens (NAc), putamen, and vmPFC. Neural activation within these regions tends to increase when viewing food items, further supporting the notion that sleep loss alters normal reward processing and inhibitory control (27, 31). Similarly, viewing images of calorierich unhealthy food increases activity in regions associated with hedonic eating (i.e., eating for pleasure), including the frontal, temporal, and parietal cortices, as well as the OFC and insula (30). Activity in the inferior frontal gyrus, a region associated with inhibitory control, has also been found to increase in response to food cues (27). In addition to the regions described above, several avenues of research have found functional activation and connectivity changes in areas of the salience network [i.e., ACC, insula, etc. (32)]. This network is involved in identifying homeostatically relevant stimuli and evaluating and selecting valued options, such as food (26, 33, 34). The ACC also makes efferent and afferent connections with regions involved in reward pathways (33). Benedict et al. (26) were the first to investigate neural responses to food stimuli during TSD, finding that one night of sleep loss resulted in increased neural activation in the ACC (26). Further, one night of TSD enhanced functional connectivity between regions of the salience network, including the dorsal ACC and putamen, in relation to total fat consumption (28). Findings within the ACC suggest that sleep loss may modulate the desire for and rewarding properties of food, thus increasing the likelihood of overconsumption. Greer et al. (29)

**Abbreviations:** ACC, Anterior Cingulate Cortex; ART, Artifact Detection Tool; BMI, Body Mass Index; DLPFC, Dorsal Lateral Prefrontal Cortex; GLM, General Linear Model; fMRI, Functional Magnetic Resonance Imaging; FWE, Family-Wise Error; kcal, Kilocalories; KSS, Karolinksa Sleepiness Scale; MINI, Mini International Neuropsychiatric Interview; MNI, Montreal Neurological Institute; MSIT, Multi-Source Interference Task; NAc, Nucleus Accumbens; OFC, Orbital Frontal Cortex; PET, Positron Emission Tomography; PFC, Prefrontal Cortex; RDI, Recommended Daily Intake; SPM, Statistical Parametric Mapping; TE, Echo Time; TR, Repetition Time; TSD, Total Sleep Deprivation; vmPFC, Ventromedial Prefrontal Cortex.

found that one night of sleep loss resulted in decreased neural activity in appetite evaluation regions (ACC, PFC, insula) and increased activation in the amygdala, further supporting the notion of reduced inhibitory control and increased reward drive during sleep loss (29).

Strong evidence points to altered functional activation within behavioral control and reward systems as one mechanism underlying the propensity to overeat during sleep loss. However, little work has focused on how individual differences in baseline neural activity within these circuits are associated with food consumption throughout a period of TSD. Individual differences in reward sensitivity are associated with the brain's response to food stimuli under rested conditions (19). Reward drive, as assessed by questionnaire, correlates with increased neural activation in regions of the fronto-striatal-midbrain reward circuitry (19). This hyper-responsivity of the reward network leads to an increased vulnerability to overeat. Here, we used functional magnetic resonance imaging (fMRI) to investigate how differences in pre-TSD functional activation within rewardrelated neural circuits can predict an individual's propensity to increase caloric and macronutrient intake during sleep loss. Specifically, in light of the well-established deficits in prefrontal inhibitory control during sleep deprivation (24, 35), we hypothesized that baseline hyper-activation in regions related to reward drive (e.g., ventral striatum) would be associated with an increased tendency to overeat during a subsequent period of sleep deprivation.

## METHODS

### Subjects

Forty-five healthy adults (20–45 y; 22 females) participated in this three-part study conducted in the McLean Hospital Sleep Research Laboratory. Subjects eligible for study participation met the following criteria: free from sleep, psychological, neurological or other medical disorders; right-handed as assessed by the Edinburgh Handedness Inventory (36); primary English speakers; no vision impairment, unless corrected to normal with contact lenses; no drug or alcohol abuse in the past 6 months; no history of smoking or tobacco use in the past year; and no contraindications for neuroimaging, including pregnancy or metal in the body. In order to control for fluctuations in menstrual hormones which could directly impact performance and brain neurochemistry, females underwent the functional neuroimaging scan during the follicular phase of their menstrual cycle or were taking monophasic contraceptives. Female subjects were excluded if they used multiphasic birth control.

All study procedures were approved by the Institutional Review Board (IRB) of McLean Hospital and the United States Army Human Research Protection Office (HRPO). All subjects gave written informed consent in accordance with the Declaration of Helsinki and were financially compensated for their time.

### Experimental Design

Subjects visited the laboratory on three separate occasions: a screening session, a neuroimaging session, and a TSD session. See **Figure 1**.

### Screening Session

Subjects first visited the lab for a comprehensive screening session. All study procedures were explained and subjects completed a series of questionnaires, including a brief psychiatric evaluation using the Mini International Neuropsychiatric Interview [MINI; (37)] to further confirm eligibility. Subjects were fitted with a wrist activity monitor (Actiwatch-2, Philips Respironics, Bend, OR) to track at-home sleep and wake patterns for at least 1 week (10.8 ± 3.3 days) prior to the third visit (i.e., the TSD session). Subjects also completed a daily online sleep diary during this time.

### Neuroimaging Session (Pre-TSD)

Subjects returned to the lab ∼1 week (8.4 ± 3.2 days) following the initial screening session for a second visit. A 2 h afternoon neuroimaging session was conducted to collect structural and functional images while subjects performed a series of neurobehavioral tasks, including the Multi-Source Interference Task (MSIT) and the n-back task, which are described in detail below. Subjects were asked to refrain from alcohol consumption 48 h prior to the second visit and were not allowed to take any over-the-counter medications on the day of the neuroimaging scan. Additionally, subjects were asked to maintain their habitual caffeine usage on the day of the scan to minimize withdrawal effects on brain vasculature.

### Total Sleep Deprivation (TSD) Session

Subjects returned to the lab for their final visit, which was scheduled 1–4 days following the neuroimaging session (2.4 ± 1.5 days). On the two nights prior to the TSD session, subjects were instructed to go to bed between 22:00 and 23:00 and remain in bed for at least 8 h. Subjects were required to wake no later than 08:00 and received a wake-up call at 07:30 from the research staff on the morning of the TSD session. Compliance was verified by wrist actigraphy. The TSD session began when subjects woke

on the day of the visit, and ended at 12:00 the following day, after ∼29 h of continuous wakefulness. After awakening on the morning of the TSD session, naps were prohibited until the end of the study. Subjects were also not allowed to consume caffeine for 24 h prior to arriving at the lab. Subjects were also asked to fast, from 13:00 until arrival at the lab later that afternoon (∼5.5 h of fasting before arrival).

Subjects arrived at the lab by 18:30 and remained awake until they were released at 12:00 the following day (**Figure 2**). Upon arrival, height and weight measurements were collected and used to calculate body mass index (BMI) following the standard formula [(weight (lbs.)/height (in.)<sup>2</sup> ) x 703]. Subjects were allowed to read, play games, and watch TV or DVDs during their free time. Additionally, subjects could access the internet to read news, watch videos, or play games. However, contact with individuals outside of the lab via personal cellphones, social media, chatting, or email was prohibited. Subjects participated in the TSD session in pairs in our controlled laboratory environment, which included a common room and individual testing rooms used for cognitive performance assessments. Light levels were kept at a fixed level and the ambient temperature remained constant for the duration of the study. Further, subjects did not have exposure to natural sunlight while in the lab. A trained research assistant was on staff at all times to administer study procedures and observe behavior throughout the overnight session. Subjects were also monitored with a closed-circuit camera for the duration of the session.

### Neurobehavioral Tasks

Subjects performed a series of computerized neurobehavioral tasks during the baseline neuroimaging session and throughout the overnight TSD session. Here, we will focus on two pre-TSD (i.e., neuroimaging session) fMRI tasks (MSIT and n-back task), and one TSD task (Food Rating task). Subjects performed practice sessions of both the MSIT and n-back before entering the fMRI scanner.

### Multi-Source Interference Task

The MSIT (38) was the first of three neurobehavioral tasks performed in the MRI scanner at ∼14:20 (± 55 min), following scanner calibration and structural imaging. The MSIT commonly activates the cingulo-frontal parietal network, which is important for attention and monitoring cognitive interference (39). During the task, subjects were shown a series of three numbers (0, 1, 2, or 3). One number (target) was always different from the other two numbers (distractors). Numbers were presented during either control or interference trials. In the control trials, subjects pressed the button that corresponded to the spatial location of the target number (i.e., 100, 020, 003), where 0s served as distractors. In the interference trials, subjects pressed the button that corresponded to the identity of the target (i.e., different) number (i.e., 221, 233, 322), where identical numbers served as distractors. The MSIT is described in more detail elsewhere (40). Brain activation contrast maps were created for the Interference > Control condition for each individual.

### n-back Task

The n-back task was the second of three neurobehavioral tasks performed in the MRI scanner, immediately following the MSIT at ∼14:30 (± 55 min). The n-back task activates the dorsal lateral prefrontal cortex (DLPFC) and parts of the parietal cortex important for working memory (41). During the task, subjects were presented with a series of letters. Subjects pressed a button to indicate whether or not the current letter on the screen was the pre-specified target letter, the same as the letter displayed one letter earlier in the series (1-back), or the same as the letter displayed two letters back (2-back), depending on the cognitive load of the trial. The n-back is described in more detail elsewhere (42). Brain activation contrast maps were created for the 2-back > 0-back condition for each individual.

### Food Rating Task

During the overnight TSD session, subjects performed a food rating task once every 3 h, beginning at 23:35. During the task, subjects were first asked to rate their current level of hunger on a 7-point Likert scale from 1 ("not at all hungry") to 7 ("extremely hungry"). Subjects were then shown a total of 70 food and non-food images in a randomized order. Images were either of neutral objects (e.g., flowers, trees, rocks), high-calorie foods (e.g., cheeseburgers, French fries, milkshakes), or low-calorie foods (e.g., fruits, vegetables, whole-grains). Subjects were asked to rate how much they would like to eat each item at that moment on a 7-point Likert scale from 1 ("do not want to eat it") to 7 ("strongly desire to eat it"). Similar versions of this task have been reported and described elsewhere (20, 43, 44).

### Karolinska Sleepiness Scale (KSS)

The KSS is a 9-point Likert scale used to measure subjective sleepiness (45). Subjects rated their current level of sleepiness from 1 ("extremely alert") to 9 ("very sleepy, great effort to keep awake, fighting sleep"). The KSS was administered hourly beginning at 19:15 as part of a larger standardized test battery that is outside the scope of this paper.

## Food

Prior to the start of the TSD session, a trained research assistant prepared identical food baskets for each subject, labeled with the corresponding subject number. For a full list of food items, see the Supplemental Material (**Table S1**). Subjects recorded all food and drink items they consumed the morning of the TSD session and were required to fast from 13:00 to 18:30 (∼5.5 h) prior to entering the laboratory. Following study arrival and intake procedures at 18:30, subjects had ad libitum access to 10,900 kcal of food throughout the entire TSD period (see **Table 2** for nutritional breakdown). Subjects discarded food packaging and unwanted leftovers in individualized trash bins labeled with their corresponding subject number. Trained research assistants also observed and recorded food intake throughout the TSD session. Discarded packaging and leftover food items were collected at 6 h intervals (00:00, 06:00, and 12:00). Total calories, calories from fat, grams of fat, grams of carbohydrates, grams of protein, and grams of sugar were documented according to the per serving nutritional values on the food packaging. For items that did not have packaging (apples, bananas) nutritional information was obtained from the U.S. Department of Agriculture Food Composition Database (https://ndb.nal.usda.gov/ndb/). For items partially consumed, values were recorded to the nearest fraction (i.e., <sup>1</sup> /4, 1 /3, 1 /2, 2 /3) of the full nutritional value listed. For analytic purposes, food consumption was broken down into three 6 h periods (Baseline: 18:00–23:59; Nighttime TSD: 00:00–05:59; Daytime TSD: 06:00– 12:00) see **Figure 2**.

### Neuroimaging Methods

Subjects underwent an fMRI scan at the end of the second visit. Neuroimaging scans were collected on a 3.0 Tesla Siemens Tim Trio Scanner (Siemens, Erlangen, Germany) with a 32-channel head coil. First, structural T1-weighted 3D images were collected with a magnetization-prepared rapid gradient-echo (MPRAGE) sequence (repetition time [TR] = 2.1 s; echo time [TE] = 2.3 ms; flip angle (FA) = 12◦ ) over 176 sagittal slices (256 × 256 matrix) with a slice thickness of 1.0 mm (voxel size = 1.0 × 1.0 × 1.0 mm). T2<sup>∗</sup> -weighted functional scans were collected over 34 transverse slices (3.5 mm thickness, no gap) using an interleaved sequence (TR = 2.0 s; TE = 30 ms; FA = 90◦ ) with 198 images and 239 images collected per slice for the MSIT and n-back, respectively. Data were collected with a 22.4 cm field of view, 64 × 64 acquisition matrix, and a voxel size of 3.5 × 3.5 × 3.5 mm<sup>3</sup> .

### Image Processing

Functional neuroimaging scans were analyzed and processed using Statistical Parametric Mapping software (SPM12; Welcome Department of Cognitive Neurology, London, UK; http://www. fil.ion.ucl.ac.uk/spm/). The raw functional images were first realigned and unwarped. Realigned images were co-registered to each individual's T1-weighted structural image. Subject images were then normalized from the original native space to the 3D space of the Montreal Neurological Institute (MNI) using forward deformation fields. The images were then spatially smoothed with a 6 mm full-width half maximum isotropic Gaussian kernel and resliced to 2 × 2 × 2 mm<sup>3</sup> voxels using 4th degree B-spline interpolations. Low frequency confounds were removed using a high-pass filter with a 128 s cutoff period. The standard canonical hemodynamic response function for SPM12 was employed, and serial correlation was corrected using a firstorder autoregressive model (AR1). The Artifact Detection Tool (ART; http://www.nitrc.org/projects/artifact\_detect/) for SPM12 was used to remove motion and spiking artifacts. Scans exceeding 3 standard deviations in mean global intensity, scan-to-scan motion that exceeded 1 mm, and the first scan of each run were regressed out of each 1st level analysis.

### Statistical Analyses

First level analyses were conducted in SPM12 using a general linear model (GLM) to create subject-specific brain activation maps by contrasting the Interference and Control conditions (Interference > Control) on the MSIT and the 2-back and 0-back conditions (2-back > 0-back) on the n-back. Contrast images were then entered into separate second level multiple regression models to assess the relationship between well-rested functional activation during the MSIT and n-back tasks and caloric and macronutrient consumption during the three periods of the TSD session. Gender, BMI, and caloric/macronutrient intake during baseline (18:00–00:00) were included as covariates for nighttime TSD and daytime TSD analyses.

Whole brain-analyses were initially height thresholded at p < 0.001 (uncorrected). Cluster-level statistics were corrected for family-wise error (FWE) at p < 0.05. For post-hoc analyses, the first eigenvariates for the significant clusters were extracted from SPM12 for regression and plotting purposes in SAS (v9.4). Additionally, Pearson's partial correlations, controlling for BMI and gender, were used to compare caloric and macronutrient intake to baseline and TSD subjective sleepiness levels. Simple linear regression models were used to individually assess the relationship between individual caloric intake and macronutrient intake and subjective hunger food desire ratings and KSS score as a function of time awake. Additionally, linear regression was used to assess the relationship between BMI and caloric/macronutrient and gender and caloric/macronutrient intake during each TSD period.

### RESULTS

### Subject Characteristics

Subject characteristics are summarized in **Table 1**.

### Caloric and Macronutrient Intake

During the at-home portion of the study on the first day of sleep deprivation (∼11 h total), subjects consumed an average of 1135.2 ± 414.9 total calories and 379.9 ± 232.1 calories from fat prior to entering the laboratory. They consumed 158.2 ± 169.8 g of carbohydrates, 52.2 ± 27.0 g of sugar, 117.1 ± 1.0 g of fat, and 45.6 ± 23.3 g of protein. Throughout the in-laboratory portion of the TSD session (17 h total), subjects consumed an average of 2,503.4 ± 754.0 total calories of the 10,900 calories available. On average, they consumed 769.4 ± 302.6 calories from fat, 380.7 ± 124.6 g of carbohydrates, 187.9 ±75.5 g of sugar, 85.5 ± 33.6 g of fat, and 61.2 ± 18.9 g of protein. Total caloric and macronutrient intake for the overnight session exceeded the recommended daily intake (RDI) values for an entire day based on a 2,000-calorie diet (**Table 2**).

Body mass index (BMI) did not predict total caloric intake during any of the three TSD periods (baseline: [F(1, 43) = 2.80, p = 0.10, R <sup>2</sup> = 0.06]; nighttime TSD: [F(1, 43) = 2.56, p = 0.12, R <sup>2</sup> = 0.06]; daytime TSD: [F(1, 43) = 1.96, p = 0.17, R <sup>2</sup> = 0.04]). Likewise, BMI did not predict total calories from fat, grams of fat, or grams of carbohydrates for any of the TSD periods [F(1, 43) < 3.48, p > 0.07]. However, higher BMI was significantly associated with more grams of sugar consumed during baseline [F(1, 43) = 5.66, p = 0.02], but not during nighttime or daytime TSD. Higher BMI was also significantly associated with more grams of protein consumed during nighttime TSD [F(1, 43) = 4.63, p = 0.04], but


<sup>a</sup>BMI categories are defined as follows: Normal: 18.5–24.9 kg/m<sup>2</sup> ; Overweight: 25.0–29.9 kg/m<sup>2</sup> ; Obese: >30.0 kg/m<sup>2</sup> .

<sup>b</sup>Time in bed, sleep duration, and bed and wake times were derived from wrist actigraphy for the one night prior to the TSD period.

not during baseline or daytime TSD. Gender was not a significant predictor of total caloric intake during any of the three TSD periods (baseline: [F(1, 43) = 1.69, p = 0.20]; nighttime TSD: [F(1, 43) = 0.0, p = 0.97]; daytime TSD: [F(1, 43) = 0.09, p = 0.76]). Gender did not predict total calories from fat, grams of fat, grams of carbohydrates, or grams of protein consumed during any of the TSD periods [F(1, 43) < 3.13, p > 0.08]. However, males consumed significantly more sugar during baseline [F(1, 43 = 4.81, p = 0.03] compared to females. These differences were not apparent during nighttime or daytime TSD. Full statistical results can be found in the Supplemental Material (**Table S2**).

### Neural Correlates of Caloric and Macronutrient Intake

**Figure 3** shows the clusters (FWE corrected) with significant correlations between MSIT (Interference > Control) performance and total caloric (blue) or carbohydrate (green) intake. While no regions of the brain were associated with calorie consumption during the baseline and overnight periods, we observed a significant correlation between activation within the ventral striatum and calories consumed (p = 0.024) in the next-day period between 06:00 and 12:00 during TSD (**Figure 3A**). This pattern was accounted for primarily by a significant correlation in the ventral striatum for grams of carbohydrates consumed (p = 0.016) during the same timeframe (**Figure 3A**). In addition, we observed significant correlations in the right middle temporal gyrus (p < 0.001) and the left superior temporal gyrus (p = 0.038) for grams of carbohydrates consumed (**Figure 3B**). Activation in the reward circuitry, specifically the NAc, while performing the MSIT under well-rested conditions predicted both total calories consumed and total carbohydrates consumed during hours 23–29 of continuous wakefulness (i.e., the last 6 h of the TSD session; **Figure 3A**). Similarly, activation in areas associated with viewing unhealthy food (30) (middle and superior temporal gyri) predicted increased carbohydrate consumption (**Figure 3B**). Higher activation in each region was associated with increased caloric and carbohydrate consumption (**Figure 4**). However, activation in these regions did not predict consumption of total calories from fat, grams of sugar, grams of


TABLE 2 | Caloric and macronutrient intake across the in-laboratory sleep deprivation session.

<sup>a</sup>Values are rounded to the nearest 100. Available values refer to in-lab portion only.

<sup>b</sup>Self-reported food diaries were used to calculate nutritional information. Subjects were required to fast from 13:00–18:30 prior to arriving at the lab.

<sup>c</sup>Recommended Daily Intake (RDI) as set by the United States Food and Drug Administration.


TABLE 3 | Cluster maxima for whole-brain multiple regression analyses of the MSIT Interference > Control condition and n-back 2-back > 0-back condition as it pertains to caloric and carbohydrate consumption.

<sup>a</sup>Cluster level analyses were family-wise error corrected at p < 0.05, with whole-brain analyses thresholded at p < 0.001. MSIT, Multi-Source Interference Task.

fat, or grams of protein. Neuroimaging results are summarized in **Table 3**.

**Figure 5** shows the clusters (FWE corrected) with significant correlations between n-back performance (2-back > 0-back) and total caloric (blue) and carbohydrate (green) intake. Critically, there was no association between brain activation and calorie consumption at baseline or during the night. However, similar to the previous findings for the MSIT, we observed a significant correlation between activation of the ventral striatum during the n-back and subsequent calories consumed (p = 0.026) during the morning between 06:00 and 12:00 during TSD (**Figure 5**). Likewise, we observed a significant correlation in the ventral striatum for grams of carbohydrates consumed (p = 0.013) in the same timeframe (**Figure 5**). Similar to the MSIT, activation in the NAc while performing the n-back under well-rested conditions predicted both total caloric intake and grams of carbohydrates consumed during the last 6 h of TSD. Higher activation in this region was associated with higher caloric and carbohydrate intake (**Figure 6**). However, like the MSIT, activation did not predict consumption of total calories from fat, grams of sugar, grams of fat, or grams of protein. Neuroimaging results are summarized in **Table 3**.

### Subjective Sleepiness (KSS) Ratings

Subjective ratings of sleepiness were low during the baseline period (3.73 ± 0.73) and steadily increased across the nighttime (5.78 ± 2.01) and daytime TSD hours (6.41 ± 2.00) as expected with extended wakefulness. KSS scores during the baseline, nighttime, and daytime TSD periods were not significantly correlated with total caloric (r < 0.10, p > 0.53) or carbohydrate (r < 0.16, p > 0.30) intake (**Figure S1**). Full statistical results are reported in the Supplemental Material (**Table S3**).

### Subjective Hunger and Food Desire Ratings

Average subjective hunger ratings were low during both the nighttime (2.34 ± 0.89) and daytime TSD (2.26 ± 1.17) periods. **Figure 7** shows subjective hunger ratings and desire ratings for high- and low-calorie foods across the TSD period in relation to total caloric intake. Regression analyses indicated that average hunger ratings did not change across the sleep deprivation session [F(1, 223) = 0.12, p = 0.73, R <sup>2</sup> < 0.001]. Likewise, there was no change in desire for high-calorie [F(1, 223) = 2.32, p = 0.13, R <sup>2</sup> = 0.01] or low-calorie [F(1, 223) = 3.03, p = 0.08, R <sup>2</sup> = 0.01] foods across time. However, total caloric and carbohydrate intake did change across the sleep deprivation period (calories: [F(1, 133) = 59.49, p < 0.001, R <sup>2</sup> = 0.31]; carbohydrates: [F(1, 133) = 23.11, p < 0.001, R <sup>2</sup> = 0.15]), generally declining with longer time awake. Hunger ratings were stable for the duration of the overnight session.

### DISCUSSION

This study investigated whether pre-TSD neural activation in reward-related brain regions was associated with total caloric and macronutrient intake during one night of sleep deprivation. On average, subjects consumed 2,500 kcal throughout the 17 h inlab portion of the TSD session, exceeding the recommended total daily value of 2,000 kcal (46). The observed caloric intake during extended wakefulness was similar to other sleep restriction (18) and sleep deprivation studies (28). Moreover, we found that functional activation in the bilateral NAc (most prominently on the left) during two independent cognitively demanding tasks (i.e., MSIT and n-back) pre-TSD (**Figures 3A**, **5**) was significantly associated with total caloric (**Figures 4A**, **6A**) and carbohydrate (**Figures 4B**, **6B**) intake during the last 6 h of TSD (06:00–12:00). In addition, while not predicted, we found that activation within the middle and superior temporal gyri (**Figure 3B**) also correlated with total carbohydrate consumption during the same timeframe (**Figures 4C,D**). Baseline activation was not associated with calories from fat, grams of fat, grams of protein, or grams of sugar consumed during any portion of the overnight TSD session. To our knowledge, this is the first study to demonstrate that individual differences in baseline activation of the ventral striatum are potentially predictive of eating behaviors several days later during one night of TSD.

Our findings suggest that individuals with greater baseline responsiveness within the reward system (i.e., NAc) when wellrested may be most vulnerable to overeating during subsequent

FIGURE 5 | Whole-brain analyses (p < 0.001, cluster corrected; p < 0.05, FWE) of n-back activation during the 2-back > 0-back condition. Sagittal, axial, and coronal views of significant activation in the nucleus accumbens (MNI:−4, 18, 2) as it correlates with total calories (blue) and total carbohydrates (green) consumed during the last 6 h of sleep deprivation (i.e., 06:00–12:00). FWE, Family-wise error; MNI, Montreal Neurological Institute.

sleep deprivation. As outlined in greater detail in the sections that follow, we suggest that when considered in light of the well-known decreases in prefrontal inhibitory control that occur during sleep deprivation, those individuals with the greatest sustained NAc reward responses may be most prone to overconsuming calories when lacking sleep. Excess reward responsiveness in conjunction with sleep loss-induced deficits in prefrontal inhibitory control is likely to represent a problematic combination when one encounters attractive high-calorie foods.

Our findings complement evidence from previous studies assessing the neural responses to food cues and subsequent food intake (47) or weight gain (48) during rested conditions and studies assessing neural responses to food stimuli during sleep restriction or TSD (26, 27, 29, 31). Studies using fMRI have shown increased neural reactivity in reward-related brain regions, specifically in the NAc, in response to food images during both rested (47, 48) and sleep-restricted conditions (27, 31). St. Onge et al. (31) found that when sleep was restricted to 4 h per night for six nights, subjects showed increased neural activity in reward-related regions, including the NAc and the superior temporal gyri, when viewing images of food vs. images of non-food items (31). In a follow-up study, they found that viewing unhealthy foods during sleep restriction increased neural activation in several cortical regions, including the middle and superior temporal gyri (30). Further, the temporal gyri have been implicated in the perception and cognitive processing of emotional stimuli (49), including increased neural responses to food stimuli during a satiated state (50). Individuals diagnosed with anorexia nervosa, compared to controls, do not show such elevated responses, suggesting that altered activation patterns in these regions may be related to increased responsiveness to the pleasurable characteristics of food, and therefore may contribute to the control of food intake (50). We found that high pre-TSD activation in these cortical areas is strongly associated with carbohydrate consumption during TSD. Additionally, a recent study by Demos et al. (27) found that when sleep was restricted to 6 h per night for four nights, neural activity increased in the left and right NAc when viewing images of food (27). In addition to findings in studies of sleep restriction, studies in well-rested individuals have shown that increased reactivity of the NAc to food stimuli is associated with both weight gain (48) and increased snacking behavior (47). Taken together, these studies demonstrate that individual differences in responsivity of the fronto-striatal-midbrain reward circuitry contribute to increased hedonic food consumption during sleep loss. While we did not measure fMRI activation during sleep deprivation, prior evidence suggests that sleep loss does in fact alter activation patterns of the reward system in response to food stimuli (26, 27, 29, 31). Therefore, it is plausible that individuals with increased striatal activity prior to sleep loss, as shown here, may be more susceptible to TSD-induced perturbations to the underlying neurocircuitry associated with reward-driven behaviors, thus contributing to the tendency to overeat.

Imaging studies in obese individuals suggest that there is a discrepancy between the enhanced sensitivity of an expected reward (i.e., seeing food) and a decreased sensitivity to the gratifying effects of the reward (i.e., eating food). That is, there is an increased propensity to overeat because the reward expectation is never met (22). We propose that a similar phenomenon may be occurring during TSD, especially among those individuals with elevated baseline reward sensitivity. The data here suggest that individuals with higher baseline neural responsiveness within the NAc have a "reward anticipation reserve," meaning that they are more likely to expect a reward from their actions (25), but the reward expectations may not be fully met.

While not directly assessed in the present study, we speculate dopamine<sup>1</sup> may play a mechanistic role in the present findings. Dopamine is a key neurotransmitter in reward circuits and underlies the pleasurable properties of food. Overweight and obese individuals show signs of disrupted reward processing within the striatum, including altered neural activity and dopamine release, as well as decreased dopamine D<sup>2</sup> receptor availability (22, 52). Interestingly, sleep loss impairs dopamine signaling in a similar manner, such that dopamine D<sup>2</sup> receptors are downregulated, (53, 54) and neural activation in the ventral striatum is increased, (27, 29–31) both of which are similar to findings in overweight and obese individuals. Due to the downregulation of D<sup>2</sup> receptors in the striatum, it is probable that dopamine is unable to bind effectively to the limited number of D<sup>2</sup> receptors that are available. Therefore, dopamine signals that indicate a reward expectation have been met are potentially disrupted. The effect that sleep loss has on dopamine function within the striatum may be amplified in individuals with elevated baseline reward sensitivity, making these individuals more vulnerable to overeating during sleep loss. Alternatively, the reduction of dopamine D<sup>2</sup> receptor availability may push the system into a D<sup>1</sup> receptor dominated state, thereby promoting reward-driven behavior, such as eating (55).

Dopamine signaling within the striatum is also linked to the receptor activity of a well-characterized neural substrate, adenosine. Adenosine accumulates with extended wakefulness and has been implicated in homeostatic sleep regulation, as it inhibits neural activity in wake-promoting regions of the brain (56). Within the striatum, dopamine D<sup>2</sup> receptors are co-localized with adenosine A2A receptors and functionally interact in an antagonistic manner (57). That is, the binding of adenosine to the A2A receptors inhibits the actions of the dopamine D<sup>2</sup> receptors, thus impairing downstream dopaminerelated neurotransmission. The inhibition of D<sup>2</sup> receptor activity by adenosine, in combination with the downregulation of D<sup>2</sup> receptors during sleep loss, may also contribute to overeating by further disrupting normal reward signaling pathways.

In addition to increased reward sensitivity and disrupted dopamine signaling in the basal ganglia, there is evidence from positron emission topography (PET) imaging that one night of TSD reduces glucose metabolism within the PFC, including the vmPFC (24). The vmPFC is considered an inhibitory emotional control region, and dysfunction may contribute to a loss of inhibitory control over subcortical reward regions, such as the NAc. Loss of inhibitory control over emotional responses can lead to increased risky behavior and impulsivity, potentially increasing the tendency to overeat

<sup>1</sup>Dopamine exerts its effects on reward circuits by binding to either D<sup>1</sup> or D<sup>2</sup> receptor subtypes located throughout the brain, including the fronto-striatal reward pathways. It should be noted that these subtypes have different functional implications that are beyond the scope of this paper (51).

(58–61). Risky decisions during sleep loss are associated with increased NAc activation, (25) similar to how the brain responds when viewing images of food when sleep restricted (27, 31). This increase in neural activation during risky decision making presumably results in an elevated expectation of reward (25). Simultaneously, neural activation within the OFC is reduced, suggesting an attenuated ability to learn from any negative consequences from the risky behavior, (25) such as weight gain associated with increased caloric intake. Additionally, reduced PFC glucose metabolism is correlated with reduced D<sup>2</sup> receptor availability in obese individuals, further implicating impaired inhibitory control in overeating (52). Taken together, these earlier studies suggest that in combination with individual differences in baseline NAc activation, sleep loss may amplify the inability to inhibit impulsive eating behavior and increase the expectation of reward from eating, ultimately leading to over consumption of high calorie foods and carbohydrates.

Similar to findings in previous studies (27, 29, 47) we found no association between hunger ratings and food consumption during TSD (**Figure 7A** and **Figure S2A**), as hunger ratings remained stable throughout each period of the TSD session. Desire for high- and low-calorie foods also remained stable across the night (**Figure 7B** and **Figure S2B**). While both hunger and food desire ratings remained stable, it should be noted that overall caloric (**Figure 7**) and carbohydrate (**Figure S2**) intake decreased as a function of time awake. However, subjects consumed roughly 60% of the recommended RDI within the first 6 h of the study (**Table 2**), suggesting that the decline in consumption is a result of satiation. It is important to note that subjects still consumed a significant number of calories during the nighttime and daytime TSD periods, and the findings for the final morning period were statistically controlled for calorie consumption in the prior periods. Further, we did not find evidence that subjective sleepiness is correlated with increased caloric and carbohydrate intake (**Figure S1**). Taken together, our results suggest that: (1) hunger or desire for food was not the primary driving force for the excessive food consumption demonstrated here; and (2) subjects were not overeating due to increased sleepiness or as a means to stay awake. These findings offer further support of the mechanistic theory that impaired functioning within the reward circuitry contributes to heightened hedonic motivations for food during TSD.

The present study demonstrates a strong association between pre-TSD reward sensitivity and the tendency to overeat during sleep loss. However, several limitations should be addressed. First, neuroimaging scans were not collected during the TSD session, which limits our ability to understand the dynamic changes in brain activation patterns as they relate to caloric and macronutrient intake, hunger, desire for food, and sleepiness. Further, it remains unknown whether or not the findings presented are unique to TSD, or whether the neural activation and food consumption patterns are also observed under wellrested control conditions. However, it is important to note that we only found a significant relationship between pre-TSD activation and food consumption during hours 23–29 of wakefulness, and not during the first 12 h of the in-lab TSD session, suggesting that the effects are only present following a sufficient amount of sleep loss. Future studies should assess the relationship between brain activation patterns and food consumption in well-rested and TSD groups. Second, although subjects provided a self-report log of foods consumed prior to arriving at the laboratory, it is impossible to completely verify compliance with the 5 h fasting period before the inlab portion of the study. Third, it should be noted that our study population consisted of young, healthy adults, and we do not know how our findings are generalizable to other populations. Additionally, increased caloric and macronutrient intake during sleep loss, and its relation to obesity, is a multifaceted problem. Eating behaviors may be modified by a number of physiological and environmental factors including changes in appetite and satiety hormones, meal timing, gender, BMI, genetics, and lifestyle (6, 62). Our BMI range was not restricted, and included individuals classified as either normal, overweight, or obese (**Table 1**). While we showed no association between BMI and caloric/carbohydrate consumption, this factor should not be overlooked in future studies as some evidence suggests that brain responses to food images differs among lean and obese individuals (63, 64). In addition, we showed no gender differences in total caloric or carbohydrate intake, only differences in baseline sugar consumption. Due to the limited scope of the paper, we did not assess lifestyle, hormones, or genetic markers as possible additional factors that may modulate some of the hedonic pathways discussed here.

Overall, we demonstrated that pre-TSD activation within the ventral striatum, as well as the middle and superior temporal gyri, is associated with eating behaviors during a single night of sleep loss. Individuals with elevated neural activity in these regions consumed significantly more calories and carbohydrates after a night of sleep deprivation. These findings suggest that there are large individual differences in baseline functioning within hedonic reward pathways and sleep loss further disrupts functioning in these pathways. Elevated reward sensitivity appears to impact eating behaviors during sleep loss and may be a major contributor to the etiology of sleep loss related obesity.

### AUTHOR CONTRIBUTIONS

BS conducted the MRI data processing and statistical analyses, and drafted the initial manuscript. AR assisted with manuscript revisions. WK designed the study, secured funding, collected the data, assisted with data interpretation, and critique, as well as contributed to manuscript review and revisions.

## FUNDING

This work was supported by the Defense Advanced Research Projects Agency (DARPA) by the DARPA Young Faculty Award (D12AP00241) granted to WK. Opinions, interpretations, conclusions, and recommendations are those of the authors and are not necessarily endorsed by DARPA or the U.S. Department of Defense.

### ACKNOWLEDGMENTS

We gratefully acknowledge the contributions of Mareen Weber, Elisabeth Olson, Christian Webb, Maia Kipman, Sophie

### REFERENCES


DelDonno, Zack Schwab, Lily (Preer) Sonis, Hanna Gogel, Olga Tkachenko, and David Penetar to data collection.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyt. 2018.00749/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 Satterfield, Raikes and Killgore. 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.

# Altered Regional Cortical Brain Activity in Healthy Subjects After Sleep Deprivation: A Functional Magnetic Resonance Imaging Study

#### Lingling Chen<sup>1</sup> , Xueliang Qi <sup>1</sup> and Jiyong Zheng<sup>2</sup> \*

*<sup>1</sup> Department of Pediatric Internal Medicine, Linyi Central Hospital, Yishui, China, <sup>2</sup> Department of Medical Imaging, The Affiliated Huaian No. 1 People's Hospital of Nanjing Medical University, Huaian, China*

#### Edited by:

*Xi-jian Dai, Jinling Hospital and Medical School of Nanjing University, China*

#### Reviewed by:

*Angel Nunez, Universidad Autonoma de Madrid, Spain Weiguang Li, Beijing Normal University, China*

> \*Correspondence: *Jiyong Zheng jyzhengdoctor@126.com*

#### Specialty section:

*This article was submitted to Sleep and Chronobiology, a section of the journal Frontiers in Neurology*

Received: *18 April 2018* Accepted: *29 June 2018* Published: *02 August 2018*

#### Citation:

*Chen L, Qi X and Zheng J (2018) Altered Regional Cortical Brain Activity in Healthy Subjects After Sleep Deprivation: A Functional Magnetic Resonance Imaging Study. Front. Neurol. 9:588. doi: 10.3389/fneur.2018.00588* Objective: To investigate acute sleep deprivation (SD)-related regional brain activity changes and their relationships with behavioral performances.

Methods: Twenty-two female subjects underwent an MRI scan and an attention network test at rested wakefulness (RW) status and after 24 h SD. The amplitude of low-frequency fluctuations (ALFF) was used to investigate SD-related regional brain activity changes. We used the receiver operating characteristic (ROC) curve to evaluate the ability of the ALFF differences in regional brain areas to distinguish the SD status from the RW status. We used Pearson correlations to evaluate the relationships between the ALFF differences in brain areas and the behavioral performances during the SD status.

Results: Subjects at the SD status exhibited a lower accuracy rate and a longer reaction time relative to the RW status. Compared with RW, SD showed significant lower ALFF values in the right cerebellum anterior lobe, and higher ALFF areas in the bilateral inferior occipital gyrus, left thalamus, left insula, and bilateral postcentral gyrus. The area under the curve values of the specific ALFF differences in brain areas were (mean ± std, 0.851 ± 0.045; 0.805–0.93). Further, the ROC curve analysis demonstrated that the ALFF differences in those regional brain areas alone discriminated the SD status from the RW status with high degrees of sensitivities (82.16 ± 7.61%; 75–93.8%) and specificities (81.23 ± 11.39%; 62.5–93.7%). The accuracy rate showed negative correlations with the left inferior occipital gyrus, left thalamus, and left postcentral gyrus, and showed a positive correlation with the right cerebellum.

Conclusions: The ALFF analysis is a potential indicator for detecting the excitation– inhibition imbalance of regional cortical activations disturbed by acute SD with high performances.

Keywords: sleep deprivation, receiver operating characteristic, area under the curve, amplitude of low frequency fluctuations, functional magnetic resonance imaging

### INTRODUCTION

Sleep is a necessary physical need for normal life, and we spend nearly one-third of our life sleeping. Sleep deprivation (SD), widespread in the current society, is caused by environmental factors or personal reasons and generally has deleterious effects on emotional regulation, memory, attention, and executive control function (1–5). Long-term SD can lead to multiorgan and multisystem dysfunction and has been shown to have negative impacts on metabolic, physiological, psychological, and/or behavioral reactivity with a greater risk of being a serious disease (6–10). However, their mechanisms are still unclear.

Resting state functional MRI (rfMRI) does not need the use of radioactive tracers and can combine functional and structural images, making the imaging method suitable for exploring the mechanisms of and obtaining insights into the pathophysiology of diseases (3); furthermore, rfMRI can be used to find the location of altered neuronal spontaneous brain activity. Recently, numerous scholars have focused their attentions on whether short-term SD has detrimental effects on regional neuronal spontaneous brain activity and cognitive function. RfMRI studies have consistently found altered cognitive domains, and altered regional spontaneous brain activity and functional connectivity patterns in the sleep-deprived brain (3, 6, 11–19), suggesting that the internal brain activity and intra-/inter- connectivity patterns for the internal processing of information are disturbed by SD. Furthermore, recent studies have found that SD has accumulative negative effects on brain morphology and advanced cognitive function (attention and working memory), showing that as SD hours prolonged, more areas show reduced gray matter volume, and after one night's sleep the brain atrophy is restored and replaced by increased gray matter volume (10). However, few studies have considered the gender factor in the neuroimaging studies of sleep disorders, and both female and male subjects were combined in these studies. Thus, the neurological mechanism of the location of altered neuronal spontaneous brain activity based on gender has not been fully studied.

Amplitude of low-frequency fluctuations (ALFF) measurement has the ability to locate where (in which brain region) regional spontaneous brain activity was disturbed with less computation complexity and high test–retest reliability characterization (20–24). Theses characterizations may make the ALFF analysis a useful tool and potential indicator for rs-fMRI data to explore the various potential neurobiological mechanisms by locating the altered regional spontaneous brain activity and functional connectivity patterns (3). Recently, ALFF analysis has been successfully applied to the exploration of neural mechanism of primary insomnia (24), wakefulness and light sleep (25), and obstructive sleep apnea (22). In this framework, in the present study we hypothesized that the ALFF measurement has the ability to locate acute SD-induced regional brain activity with high sensitivity and specificity. To test this hypothesis, we used the ALFF analysis as a potential indicator to locate the underlying altered regional functional brain activity during the SD status relative to the rested wakefulness (RW) status, and further explored the potential neurobiological mechanisms of SD in female subjects with respect to the location of altered neuronal spontaneous brain activity. Specifically, the receiver operating characteristic (ROC) curve was used to investigate the abilities of the ALFF analysis in distinguishing the SD status from the RW status. Pearson correlations were used to evaluate the relationships between the ALFF differences in brain areas and the behavioral performances during the SD status.

### MATERIALS AND METHODS

### Subjects

The present study was approved by the Medical Research Ethical Committee. The Affiliated Huaian No.1 People's Hospital of Nanjing Medical University. Twenty-two healthy female subjects (age, 26.91 ± 6.05 years; education, 15.77 ± 1.15 years; mean ± std) were recruited. All subjects met the following criteria as in previous studies (3, 6): (1) right-handed; (2) good sleep habit without any symptoms of sleep disorders such as difficulties in initiating and/or maintaining sleep, with Pittsburgh sleep quality index score < 5; (3) never taken alcohol, stimulants, cigarette, hypnotic or psychoactive medications, diet pills, and caffeine for ≥3 months during and prior to the current study; (4) regular dietary habit with moderate weight and body shape; (5) without foreign implants, and inborn and acquired diseases.

Each of the subjects underwent the MRI scan twice, once during RW status and the other after 24 h acute SD. The acute SD session started from 19:00 p.m. on the first day and lasted until 7:00 p.m. on the second day. Before the MRI scan, all volunteers underwent an attention network test (26, 27). Food and water were provided during the SD procedure. The temperature of the room was maintained between 23 and 27◦C. The staff of our team used video monitors and worked in turns to make sure that the participants did not fall asleep. If the participants showed signs of falling asleep, they were immediately awakened using an alarm clock by staff. A simple questionnaire was used to evaluate whether the subjects were asleep during the MRI scan. All subjects provided their written informed consent voluntarily.

### MRI

The MRI examination was performed using an acquired clinical 3.0-Tesla MRI scanner (SIEMENS Trio Tim, Siemens Healthcare, Erlangen, Germany) with a standard eight-channel head coil. First, we acquired a high-resolution 3D anatomical image with 176 T1-weighted images in a sagittal orientation: repetition time = 1,950 ms, gap = 0 mm, echo time = 2.3 ms, thickness = 1 mm, acquisition matrix = 248 × 256, flip angle = 9 ◦ , field of view = 244 × 252 mm. Second, we also acquired 240 functional images using a single-shot gradient-recalled echo-planar imaging pulse sequence (repetition time = 3,000 ms, gap = 0.5 mm, echo time = 25 ms, thickness = 5.0 mm, flip angle = 90◦ , acquisition matrix = 32 × 32, field of view = 210 × 210 mm).

### Data Analysis

The first 10 time points of the functional images were discarded because of the possible instability of the initial MRI signal and to allow the participants to adapt to the scanning environment. Data preprocessing of the remaining resting-state images was performed using the Data Processing & Analysis for Brain Imaging (DPABI 2.1, http://rfmri.org/DPABI) toolbox, adopting the Digital Imaging and Communications in Medicine (DICOM) standard for form transformation, slice timing, head motion correction, spatial normalization, and spatial smoothing using a Gaussian kernel of 8 × 8 × 8 mm<sup>3</sup> full-width at half-maximum. Participants with more than 1.5 mm maximum translation in x, y, or z directions and 1.5◦ of motion rotation were removed. After the head motion correction, the rest of the functional images were spatially normalized and resampled to Montreal Neurological Institute (MNI) space at a resolution of 3 × 3 × 3 mm<sup>3</sup> . Linear regression was applied to remove several sources of possible spurious covariates, including 24 head motion parameters obtained in the realigning step, signal from a region in the cerebrospinal fluid or/and centered in the white matter, and global signal averaged over the whole brain. After preprocessing, the time series were further linearly detrended and temporally band-pass filtered (0.01–0.1 Hz). The details of the ALFF calculation have been reported in previous studies (3, 28). To reduce the global effects of variability across the participants, the mean ALFF value of each voxel was divided by the global mean ALFF value for each participant.

### Statistical Analysis

Data are presented as mean ± standard deviation (mean ± std). Pair t-tests were used for demographic factors (age, years of education, and clinical factors), and a chi-squared (χ 2 ) test was used for categorical data (gender). p < 0.05 was considered to be a significant difference.

A pair t-test was used to investigate the ALFF differences in regional brain areas of the subjects during the acute SD status relative to the RW status with the gender, age, and years of education as nuisance covariates of no interest. AlphaSim correction (threshold of individual voxel of p < 0.01 and cluster level of p < 0.05 with contiguous voxel size ≥20) was used to determine the statistical differences.

We used the ROC curve to evaluate the ability of the ALFF differences in regional brain areas to distinguish the SD status from the RW status, and we used Pearson correlations to evaluate the relationships between the ALFF differences in brain areas and the behavioral performances during the SD status. The statistical threshold was set at p < 0.05.

FIGURE 1 | ALFF differences between SD group and RW group. The differences covered cerebellum with lower ALFF (A), and inferior occipital gyrus, thalamus, insula, and postcentral gyrus with higher ALFF (B). The color in the map represents the differences. The red color signifies increase in ALFF areas, and the blue signifies decrease in ALFF areas. ALFF, Amplitude of low-frequency fluctuations; SD, Sleep deprivation; RW, Rested wakefulness; R, right; L, left.


*The statistical threshold was set at corrected significance level of individual voxel p* < *0.01 using an AlphaSim-corrected threshold of cluster p* < *0.05.*

*ALFF, Amplitude of low-frequency fluctuation; SD, Sleep deprivation; RW, Rested wakefulness; R, right; L, left; BA, Brodmann's area; MNI, Montreal neurological institute; N/A, Not applicable.*

TABLE 1 | ALFF differences in brain areas between SD and RW.

### RESULTS

### Behavioral Characteristics

Compared with the RW status, the acute SD status had a lower response in accuracy rate (mean ± std, 24 h SD = 96.07 ± 3.2%, RW = 97.85 ± 1.69%; t = −2.125, p = 0.046) and a longer response in reaction time (24 h SD = 633.99 ± 79.05 ms; RW = 537.97 ± 46.49 ms; t = 5.554, p < 0.001).

### ALFF Differences Between Groups

Compared with RW, SD had significant lower ALFF areas in the right cerebellum anterior lobe (**Figure 1A**), and higher ALFF areas in the bilateral inferior occipital gyrus (Brodmann's area, BA 18, 19), left thalamus, left insula (BA 13), and bilateral postcentral gyrus (BA 3, 6) (**Table 1**, **Figure 1B**).

### ROC Curve

The mean beta value of ALFF differences in the altered areas were extracted (**Figure 2**). These different ALFF differences in brain areas were further used for the ROC curve to evaluate their abilities to distinguish the acute SD status from the RW status. The area under the curve (AUC) values of those specific ALFF differences in brain areas were (0.851 ± 0.045; 0.805–0.93).

Further, the ROC curve demonstrated that the ALFF differences in those regional brain areas alone discriminated the acute SD status from the RW status with high degrees of sensitivities (82.16 ± 7.61%; 75–93.8%) and specificities (81.23 ± 11.39%; 62.5–93.7%) with cut-off points of −0.351, 0.206, 0.2065, 0.1155, −0.8015, −0.405, and −0.3095 (mean beta signal value), respectively (**Table 2**, **Figure 3**).

### Pearson Correlation Analysis

The accuracy rate demonstrated a positive correlation with the ALFF value in the right cerebellum anterior lobe (r = 0.496, p = 0.019; **Figure 4A**), and negative correlations with the ALFF

TABLE 2 | ROC curve for the ALFF differences in brain areas between SD and RW.


\**Cut off point of mean ALFF signal value.*

*ROC, Receiver operating characteristic; ALFF, Amplitude of low-frequency fluctuation; SD, Sleep deprivation; RW, Rested wakefulness; AUC, Area under the curve; R, Right; L, Left.*

values in the left inferior occipital gyrus (r = −0.602, p = 0.003; **Figure 4B**), left thalamus (r = −0.522, p = 0.013; **Figure 4C**) and left postcentral gyrus (r = 0.656, p = 0.001; **Figure 4D**) during the acute SD status, respectively. None of the other correlations between the ALFF values in those different areas and the behavioral performances during the acute SD status were found (p > 0.05).

### DISCUSSION

In the present study, we used ALFF analysis to demonstrate the differences in regional brain areas associated with acute SD, and their correlations with the clinical performances. Specifically, we found that SD was associated with widespread regional brain activities with lower ALFF values in the right cerebellum anterior lobe, and with higher ALFF values in the bilateral inferior occipital gyrus (BA 18, 19), left thalamus, left insula (BA 13), and bilateral postcentral gyrus. Furthermore, during the SD status, the accuracy rate showed correlations with the beta value of ALFF differences in those brain areas. Recently, the ROC curve is widely used to evaluate the reliability of a neuroimaging technique in distinguishing one group from another group (3, 22, 24). In general, the AUC value is considered as excellent between 0.9 and 1, considered as good between 0.8 and 0.9, considered as fair between 0.7 and 0.8, considered as poor between 0.6 and 0.7, and considered as failed between 0.5 and 0.6. In the present study, the ROC curve revealed that the ALFF differences in those brain areas had good discriminating abilities with a high AUC value (> 0.8). Further diagnostic analysis showed that these areas discriminated the SD status from the RW status with high degrees of sensitivities (mean, 82.16 ± 7.61%; 75–93.8%) and specificities (mean, 81.23 ± 11.39%; 62.5–93.7%).

In a previous study, a total of 16 healthy subjects (8 females, 8 males) were recruited, and SD was found to be associated with several ALFF differences in brain areas (29); however, the study did not take the gender differences into account. Previous studies have shown that there are wide gender differences in brain

activity in healthy subjects both at the SD status and the RW status, and in patients with chronic insomnia relative to good sleepers in sleep neuroimaging studies (6, 24). In this framework, the present study only recruited healthy female subjects to exclude the effect of the gender factor. Specifically, we found that SD was associated with widespread regional brain activities with lower ALFF values and higher ALFF values, and this finding is different from that of the previous study. Therefore, our findings support the standpoint that the gender factor should be taken into account in the neuroimaging studies of sleep disorders (6, 24).

The hyperarousal and increased glucose utilization in patients with chronic primary insomnia were found in neurocognitive, neuroimaging, and physiological studies (30–32). Hyperarousal refers to magniloquent cognitive, somatic, and/or cortical activation, further leading to increased sensory information processing (33, 34), which is a core predisposing factor of chronic primary insomnia (35). Previous neuroimaging studies also found hyperarousal reactivation in several brain areas in individuals after SD and patients with chronic primary insomnia (6, 24). The present study found that SD is associated with increased ALFF areas in widespread regional brain areas, and these increased ALFF areas show negative correlations with the accuracy rate. There are two prevalent speculations for the increased regional brain activities (36). One explanation of the hyperarousal model could be that this is a brain compensation mechanism. Previous diffusion tensor imaging study showed that 23 h SD is associated with widespread fractional anisotropy decreases in several brain areas and as the waking prolonged the decreases become larger (37). Another explanation of the increased ALFF areas in widespread regional brain areas may be that the hyperactivation in these widespread regional brain areas may be interpreted as an enhanced neural effort to offset these decreased brain structures associated with SD. A previous task study found that the parietal lobe was not activated after normal sleep but was activated after SD (38). The occipital lobe and postcentral gyrus were found with higher regional homogeneity and ALFF values (6, 17); the thalamus was also found with higher ALFF value after acute SD (17), and the thalamus and insula were activated by acupuncture after acute SD (39). These findings were consistent with our study, and may reflect dynamic, compensatory changes in cerebral activation after SD.

The lower ALFF values in brain areas may indicate a consistent decrease of regional neuronal activity with poor synchronization and without in order (6). Poor regulation of behaviors and emotions are core features of SD. The cerebellum is involved in coordinating movement, and emotional and cognitive functions (6, 24), and associated with the aberrant regional brain activity in sleep disorders, such as patients with primary insomnia (24, 40) and obstructive sleep apnoea (41). In the present study, SD compared with RW had a significant lower ALFF value in the right cerebellum anterior lobe, and the

### REFERENCES


mean ALFF value in this area had a positive correlation with accuracy rate (r = 0.496, p = 0.019). The decreased regional brain activity in the right cerebellum anterior lobe may reflect that the sleep-deprived brain needs to attempt to recruit more specific brain areas with advanced cognitive functions to accomplish the cognitive performance because of a continuing decline in the cerebellum activity. Interestingly, Wang et al. showed different findings of altered SD-related regional brain activities in several areas (29). Since the gender factor may influence the results in the neuroimaging studies of sleep disorders (6, 24), we therefore speculated that the differences between our study and Wang et al.'s study may be associated with the gender factor.

### CONCLUSIONS

In summary, the ALFF analysis is a useful index to locate the underlying altered regional brain activities in individuals during the SD status relative to the RW status with high degrees of sensitivities and specificities. SD is associated with the model of excitation–inhibition imbalance of cortical activations. These findings expand our knowledge and may help in deeper understanding of the neurobiological mechanisms underlying acute SD. Furthermore, the gender factor should be taken into account in the neuroimaging studies of sleep disorders. However, there are several potential limitations that should be noted. First, our study has a relatively small sample size and future studies on a larger number of sample sizes are necessary to corroborate our findings. Second, in our study the design of replication is not addressed. Third, the electronystagmogram has been used to dynamically monitor the sleep.

### AUTHOR CONTRIBUTIONS

LC wrote the main manuscript text. JZ conceived and designed the whole experiment. LC and XQ collected the data. JZ analyzed the data.


amplitude of low-frequency fluctuation: a fMRI study. Neuropsychiatr Dis Treat. (2015) 11:1877–83. doi: 10.2147/NDT.S87596


**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 Chen, Qi and Zheng. 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.

# Altered Long- and Short-Range Functional Connectivity Density in Healthy Subjects After Sleep Deprivations

Dan Kong<sup>1</sup> \*, Run Liu<sup>2</sup> , Lixiao Song<sup>3</sup> , Jiyong Zheng<sup>1</sup> \*, Jiandong Zhang<sup>1</sup> and Wei Chen<sup>4</sup>

*<sup>1</sup> Department of Medical Imaging, The Affiliated Huai'an No. 1 People's Hospital of Nanjing Medical University, Huai'an, China, <sup>2</sup> Department of Radiology, The Affiliated Xi'an Central Hospital of Xi'an Jiaotong University, Xi'an, China, <sup>3</sup> Department of Hematology, The Affiliated Huai'an No. 1 People's Hospital of Nanjing Medical University, Huai'an, China, <sup>4</sup> Department of Interventional Radiology, The Affiliated Huai'an No. 1 People's Hospital of Nanjing Medical University, Huai'an, China*

#### Edited by:

*Xi-jian Dai, Jinling Hospital and Medical School of Nanjing University, China*

#### Reviewed by:

*Xuming Liu, Wenzhou City People's Hospital, China Alexander Nikolaevich Savostyanov, State Scientific-Research Institute of Physiology & Basic Medicine, Russia*

#### \*Correspondence:

*Dan Kong 156017532@qq.com Jiyong Zheng jyzhengdoctor@126.com*

#### Specialty section:

*This article was submitted to Sleep and Chronobiology, a section of the journal Frontiers in Neurology*

Received: *22 May 2018* Accepted: *19 June 2018* Published: *16 July 2018*

#### Citation:

*Kong D, Liu R, Song L, Zheng J, Zhang J and Chen W (2018) Altered Long- and Short-Range Functional Connectivity Density in Healthy Subjects After Sleep Deprivations. Front. Neurol. 9:546. doi: 10.3389/fneur.2018.00546* Objective: To investigate the brain functional organization induced by sleep deprivation (SD) using functional connectivity density (FCD) analysis.

Methods: Twenty healthy subjects (12 female, 8 male; mean age, 20.6 ± 1.9 years) participated a 24 h sleep deprivation (SD) design. All subjects underwent the MRI scan and attention network test twice, once during rested wakefulness (RW) status, and the other was after 24 h acute SD. FCD was divided into the shortFCD and longFCD. Receiver operating characteristic (ROC) curve was used to evaluate the discriminating ability of those FCD differences in brain areas during the SD status from the RW status, while Pearson correlations was used to evaluate the relationships between those differences and behavioral performances.

Results: Subjects at SD status exhibited lower accuracy rate and longer reaction time relative to RW status. Compared with RW, SD had a significant decreased shortFCD in the left cerebellum posterior lobe, right cerebellum anterior lobe, and right orbitofrontal cortex, and increased shortFCD in the left occipital gyrus, bilateral thalamus, right paracentral lobule, bilateral precentral gyrus, and bilateral postcentral gyrus. Compared with RW, SD had a significant increased longFCD in the right precentral gyrus, bilateral postcentral gyrus, and right visuospatial network, and decreased longFCD in the default mode network. The area under the curve values of those specific FCD differences in brain areas were (mean ± std, 0.933 ± 0.035; 0.863∼0.977). Further ROC curve analysis demonstrated that the FCD differences in those brain areas alone discriminated the SD status from the RW status with high degree of sensitivities (89.19 ± 6%; 81.3∼100%) and specificities (89.15 ± 6.87%; 75∼100%). Reaction time showed a negative correlation with the right orbitofrontal cortex (*r* = −0.48, *p* = 0.032), and accuracy rate demonstrated a positive correlation with the right default mode network (*r* = 0.573, *p* = 0.008). Conclusions: The longFCD and shortFCD analysis might be potential indicator biomarkers to locate the underlying altered intrinsic brain functional organization disturbed by SD. SD sustains the cognitive performance by the decreased high-order cognition related areas and the arousal and sensorimotor related areas.

Keywords: sleep deprivation, functional connectivity density, receiver operating characteristic, sensorimotor, short-range, long-range

### INTRODUCTION

Sleep deprivation, widespread in current society, can be caused by environmental factors or personal reasons. It generally has a deleterious effect on emotional regulation, memory, attention, and executive control function (1–5), and even metabolic, physiological, psychological, and/or behavioral reactivity with a greater risk of being multiorgan and multisystem dysfunction (6–9). Recently, several studies have demonstrated structural and functional changes in the frontal cortex, parietal cortex, and temporal cortex in individuals after acute SD (1, 6, 10–21); however, the neurologic mechanism of acute SD has not been fully studied.

Resting-state functional MRI (rfMRI) can combine the functional images and structural images without exposure to radioactive tracers, which makes the rfMRI suitable for mechanism and pathophysiology exploration in several diseases (1). The advance of rfMRI can help us non-invasively explore the functional organization in the human brain thus better characterize the changes of regional neuronal spontaneous brain activity and intrinsic connectivity patterns to understand the underlying neural basis of neuropsychiatric disorders.

Seed-based functional connectivity studies have revealed abnormal connectivity patterns in individuals with insufficient sleep in brain regions related to emotion and cognition (13, 18, 21–26); however, the seed-based functional connectivity analysis provides limited information about the relationships between the time series of a given seed point area and the time series of other areas in a whole brain network (27, 28). Voxel-based functional connectivity density (FCD) was used to identify the distribution of hubs in the human brain (29). In contrast to the seed-based functional connectivity analysis, the FCD analysis, similar to the degree centrality analysis, provides an opportunity for unbiased searches abnormalities within the whole brain without the need for a prior definition of regions of interest (27). The FCD can be divided into the short-range FCD and long-range FCD on the basis of the neighboring relationships between brain voxels (30). Recently, the FCD analysis has been widely applied to the exploration of the neurophysiological basis of several diseases (31–34), and reveals extra information which cannot be provided by the seed-based functional connectivity analysis. In this framework, in the present study we utilized the potential indicators of shortFCD and longFCD approaches to characterize the changes of intrinsic functional connectivity strength after acute SD status relative to rested wakefulness (RW) status, and further explore the potential neurobiological mechanisms of SD.

### MATERIALS AND METHODS

### Subjects

Twenty healthy subjects (12 female, 8 male; mean age, 20.6 ± 1.9 years; mean education, 14.5 ± 1.19 years) participated in a 24 h SD design experiment. All subjects met the following criteria, as in previous studies (1, 6):


Each of the subjects underwent the MRI scan twice; once during RW status, and the other after 24 h' acute SD. The acute SD process started at 19:00 on the first day and lasted until 07:00 in the second day. The food and water were provided during the SD procedure. The temperature of the room was maintained between 23◦C and 27◦C. The team took turns to monitor and make sure that the participants did not fall asleep using video monitors. This study was approved by the Medical Research Ethical Committee of The Affiliated Huai'an No. 1 People's Hospital of Nanjing Medical University in accordance with the Declaration of Helsinki. All volunteers participated voluntarily and were informed of the purposes, methods, and potential risks of this study, and signed an informed consent form.

### MRI

The MRI examination was performed, via acquisition, on a clinical 3T MRI scanner (SIEMENS Trio, Erlangen, Siemens, Germany) with a standard eight-channel head coil using a 12 channel array coil. First, we acquired a high-resolution 3D anatomical images with 176 T1-weighted images in a sagittal orientation: repetition time = 1950 ms, gap = 0 mm, echo time = 2.3 ms, thickness = 1 mm, acquisition matrix = 248 × 256, flip angle = 9 ◦ , field of view = 244 × 252 mm. Second, we also acquired 240 functional images using a single-shot Gradient-Recalled Echo-Planar Imaging pulse sequence (repetition time = 3000 ms, gap = 0.5 mm, echo time = 25 ms, thickness = 5.0 mm, flip angle = 90◦ , acquisition matrix = 32 × 32, field of view = 210 × 210 mm).

wakefulness; shortFCD, short-range functional connectivity density; FDR, false discovery rate.

FIGURE 3 | One sample *t*-test differences of SD and RW in Binarized longFCD. The Binarized longFCD maps in the SD group (A) and the RW group (B), respectively. These maps are the results of the within-groups using one-sample *t*-tests, corrected by FDR. L, left; R, right; SD, sleep deprivation; RW, rested wakefulness; longFCD, long-range functional connectivity density; FDR, false discovery rate.

FIGURE 4 | Binarized shortFCD differences between SD and RW. The color in the map represents the differences. The blue signifies decreased binarized shortFCD in brain areas (A), and the red color signifies increased binarized shortFCD in brain areas (B). shortFCD, short-range functional connectivity density; SD, Sleep deprivation; RW, Rested wakefulness; L, left; R, right.

#### TABLE 1 | The binarized shortFCD differences between SD and RW.


*Between-group differences in binarized shortFCD thresholded at r* = *0.3. The statistical threshold was set at corrected significance level of individual two-tailed voxel-wise p* < *0.05 using an AlphaSim corrected threshold of cluster p* < *0.05.*

*shortFCD, short-range functional connectivity density; SD, sleep deprivation; RW, rested wakefulness; R, right; L, left; BA, Brodmann's area; MNI, Montreal Neurological Institute; N/A, Not applicable.*

### Attention Network Test (ANT)

Before the MRI scan, all volunteers underwent an attention network test (ANT) (1, 12, 35, 36). The ANT contained three cue conditions (no cue, center cue, spatial cue) and two target TABLE 2 | The binarized longFCD differences between SD and RW.


*Between-group differences in binarized shortFCD thresholded at r* = *0.3. The statistical threshold was set at corrected significance level of individual two-tailed voxel-wise p* < *0.05 using an AlphaSim corrected threshold of cluster p* < *0.05.*

*longFCD, long-range functional connectivity density; R, right; L, left; BA, Brodmann's area; MNI, Montreal Neurological Institute; N/A, Not applicable.*

conditions (congruent and incongruent). The visual stimuli consisted of a row of five horizontal black arrows pointing leftward or rightward with the target arrow in the center. The participants responded to the direction of the central arrow by pressing the left or right buttons of the computer mouse. The task measured alerting, orienting, and conflict effects by calculating the difference between the response time and the presentation time under three different cue conditions. The accuracy rate using corrected recognition, reaction time using only trials with correct responses, and lapse rate using missing recognition, were calculated.

### Data Analysis

First, the first 10 time points of the functional images were deleted, due to the possible instability of the initial MRI signal. The remaining data was analyzed by Data Processing & Analysis for Brain Imaging (DPABI 2.1, http://rfmri.org/DPABI) toolbox based on MATLAB2010a (Mathworks, Natick, MA, USA). The data preprocessing contained the following steps: including the format transformation, slice timing, head motion correction spatial normalization to the Montreal Neurological Institute (MNI) space, and smooth. The data of participants with > 1.5 mm maximum translation in x, y, or z directions and >1.5◦ degree of motion rotation were removed. Based on the recent work showing that higher-order models benefit from the removal of head motion effects (37, 38), after the head motion correction, The functional images were re-sampled at a resolution of 3 × 3 × 3 mm<sup>3</sup> during the spatial normalization. Linear regression was applied to remove the effects of spurious covariates, including the Friston 24 head motion parameters, global mean signal, white matter and cerebrospinal fluid signal. Next, the functional images were entered into temporally bandpass filtered (0.01–0.1 Hz) and linearly detrended.

## Calculation of Long FCD and ShortFCD Calculation Maps

The local and global FCD maps for each individual were calculated in a gray matter (GM) mask. The number of functional connections of a given voxel was considered as a degree of a node in a binary graph. First, we defined the functional connectivity between a given voxel with each of other voxels in the whole brain with a correlation threshold of r > 0.25 (39). In the present study, we adopted the threshold of r = 0.3 to calculate the FCD maps. Second, the longFCD and shortFCD were defined based on the neighborhood strategy. We defined the voxels with a correlation threshold of r > 0.25 inside their neighborhood (radius sphere ≤6 mm) as shortFCD, and defined the voxels with a correlation

FIGURE 5 | Binarized longFCD differences between SD and RW. The color in the map represents the differences. The red color signifies increased binarized longFCD in brain areas, and the blue signifies decreased binarized longFCD in brain areas. longFCD, long-range functional connectivity density; SD, Sleep deprivation; RW, Rested wakefulness; L, left; R, right.

threshold of r > 0.25 outside their neighborhood (radius sphere > 6 mm) as long FCD. Next, the shortFCD and longFCD maps of each subject were divided by the mean value so as to convert to Z scores to improve the normality. Finally, the shortFCD and longFCD maps underwent spatial smoothing with a Gaussian kernel of 6 × 6 × 6 mm<sup>3</sup> full-width at half-maximum using SPM8. The detailed procedure of the shortFCD and longFCD is given in a previous study (29).

### Statistical Analysis

Data was presented as mean ± standard deviation (mean ± std). Pair t-tests were used for demographic factors (age, years of education, and ANT findings). p < 0.05 was considered as significant.

Pair t-tests were used to investigate the FCD differences in regional brain areas of the subjects during the acute SD status relative to the RW status. AlphaSim correction (threshold of individual voxel of p < 0.05 and cluster level of p < 0.05) was used to determine the statistical differences.

We used the receiver operating characteristic (ROC) curve to investigate the ability of those binarized FCD differences in regional brain areas to distinguish the SD status from the RW status, and we used Pearson correlations to evaluate the relationships between those binarized FCD differences in brain areas and ANT during the SD status. The statistical threshold was set at P < 0.05.

## RESULTS

## Ant Findings

Individuals at acute SD status showed a lower accuracy rate (acute SD = 96.25 ± 2.32%, RW = 97.85 ± 1.77%; t = −2.482, p = 0.023; **Figure 1A**) and a longer reaction time (acute SD = 635.27 ms ± 82.68 ms; RW = 540.01 ± 48.37 ms; t = 5.013, p < 0.001; **Figure 1B**) during the ANT relative to the individuals at RW status.

## FCD Differences Between-Groups

First, we performed one-sample t-test to explore the FCD differences at within-group level for each group. **Figure 2** shows the shortFCD maps in the SD group (**Figure 2A**) and RW group (**Figure 2B**), respectively. **Figure 3** shows the longFCD maps in the SD group (**Figure 3A**) and RW group (**Figure 3B**), respectively. The covered differences in brain areas both in binarized shortFCD and in binarized longFCD were larger in the SD group than that of RW group.

Second, we performed pair t-tests to explore the FCD differences between-groups. Compared with RW, acute SD had significant decreased binarized shortFCD areas in the left cerebellum posterior lobe, right cerebellum anterior lobe (**Figure 4A**) and right inferior frontal gyrus (orbitofrontal cortex), and increased binarized shortFCD areas in the left occipital gyrus, bilateral thalamus, right paracentral lobule, bilateral precentral gyrus, and bilateral postcentral gyrus (**Table 1**, **Figure 4B**). Compared with RW, acute SD had significant increased binarized longFCD areas in the right precentral gyrus, bilateral postcentral gyrus, and right superior

Kong et al. Sleep Deprivation Disturbed Connectivity Density



\**Cut off point of mean shortFCD signal value.*

*ROC, Receiver operating characteristic; shortFCD, short-range functional connectivity density; SD, Sleep deprivation; RW, Rested wakefulness; AUC, Area under the curve; R, Right; L, Left.*

parietal lobule in the visuospatial network, and decreased binarized longFCD areas in the right supramarginal gyrus in the default mode network (**Table 2**, **Figure 5**).

### ROC Curve

The mean beta value of binarized shortFCD (**Figure 6A**) and binarized longFCD (**Figure 6B**) differences in those altered brain areas were extracted. These different binarized FCD differences in brain areas were further used for the ROC curve to evaluate their ability to distinguish the acute SD status from the RW status. The area under the curve (AUC) values of those specific binarized FCD differences in brain areas were (mean ± std, 0.933 ± 0.035; 0.863∼0.977). Further ROC curve demonstrated that the binarized FCD differences in those regional brain areas alone discriminated the acute SD status from the RW status with high degree of sensitivities (mean ± std, 89.19 ± 6%; 81.3∼100%) and specificities (mean ± std, 89.15 ± 6.87%; 75∼100%) (**Tables 3–4**, **Figure 7**).

### Pearson Correlation Analysis

The reaction time showed negative correlation with the mean beta value of binarized shortFCD in the right inferior frontal gyrus (r = −0.48, p = 0.032; **Figure 8A**), and the accuracy rate demonstrated a positive correlation with the mean beta value of binarized longFCD in the right supramarginal gyrus (r = 0.573, p = 0.008; **Figure 8B**). None of the other correlations between the mean beta value of binarized FCD in other different areas and the ANT during the acute SD status were found (p > 0.05).

TABLE 4 | ROC curve for the binarized longFCD differences in brain areas between SD and RW.


\**Cut off point of mean longFCD signal value.*

*ROC, Receiver operating characteristic; longFCD, long-range functional connectivity density; SD, Sleep deprivation; RW, Rested wakefulness; AUC, Area under the curve; R, Right; L, Left.*

### DISCUSSION

In the present study, we utilized shortFCD and longFCD analysis to characterize the differences of intrinsic functional connectivity induced by acute SD, and their correlations with the ANT. Specifically, we found that acute SD was associated with binarized shortFCD alterations in more regional brain areas than that of binarized longFCD. Acute SD was associated with a significant decrease in binarized shortFCD areas in the cerebellum posterior/anterior lobe and orbitofrontal cortex, and significant increase in the occipital gyrus, thalamus, paracentral lobule, and precentral/postcentral gyrus. Using the binarized longFCD method, only the supramarginal gyrus in the default mode network with decreased binarized longFCD were observed after acute SD relative to RW, and significantly increased binarized longFCD in the precentral/postcentral gyrus and visuospatial network were found. Furthermore, the ANT showed correlations with the beta value of FCD differences in those brain areas during the SD status. Recently, the ROC curve was widely used to applied into the exploration of the reliability of one neuroimaging approach as a potential indicator in distinguishing one group from the other group (1, 40, 41). In general, an AUC value between 0.9 and 1 is considered as excellent, while a value between 0.8 and 0.9 is considered as good. In the present study, the ROC curve demonstrated that the AUC values of the binarized FCD differences in those brain areas showed good discriminating abilities with extremely high AUC values (0.933 ± 0.035; 0.863∼0.977). Further diagnostic analysis revealed that the binarized FCD differences in those regional brain areas alone discriminated the acute SD status from the RW status with extremely high degree of sensitivity (89.19 ± 6%; 81.3∼100%) and specificities (89.15 ± 6.87%; 75∼100%).

The default-mode network is thought to be associated with self-referential mental activity (42), extraction of episodic memory (43), sleep and daydreaming (1, 44), and social cognitive processes related to decision making and self-regulation (45, 46). The orbitofrontal cortex, connected with prefrontal, and deep structures known to mediate sensorimotor processing,

motivation, and self-evaluation, is thought to be responsible for mediating the interactions between emotional processes and cognitive functions (47, 48), and play a significant role in fatigue, executive functions, attention, and motivation (49– 51). This area is particularly vulnerable to subjects with sleep loss (40, 41, 52, 53). The decreased gray matter volume in the orbitofrontal cortex has previously been reported in patients with daytime sleepiness and chronic insomnia (54, 55). In the present study, we found that acute SD was associated with a significant decreased binarized longFCD within the default mode network node and decreased binarized shortFCD in the right orbitofrontal cortex, which showed an extremely high degree of sensitivity and specificity in distinguishing the acute SD status from the RW status. In addition, the accuracy rate of the ANT demonstrated a positive correlation with the mean beta value of binarized longFCD in the default mode network node, and the reaction time of the ANT showed negative correlation with the mean beta value of binarized shortFCD in the orbitofrontal cortex. We speculated that the decreased long-/shortFCD in the default mode network and orbitofrontal cortex implicated the brain's exertion of voluntary control to remain awake and perform, which might be sensitive biomarkers for advanced cognitive function.

Higher level visual brain areas are divided into two distinct visual pathways: the object properties processing pathway and the spatial properties processing pathway (56–58). The spatial properties processing pathway runs from the occipital lobe up to the posterior parietal lobe and has been called the dorsal system. This system processes object localization and spatial attributes, and is also essential for guiding movements. Damage to the dorsal pathway disrupts the ability to visualize locations or perceive space. The postcentral gyrus is the main receptive region for external stimuli as the location of the primary somatosensory cortex. Recently the postcentral gyrus was implicated with the default mode network (59), which are functional brain hubs showing coupled slow signal fluctuations in the absence of external stimuli during restful waking and sleep (60). The thalamus is a vital region in integrating neural activity from widespread neocortical inputs and outputs, and is thought to play an important role in regulating state of sleep and wakefulness. Previous PET studies have revealed that SD could increase the metabolic rate of glucose in the visual cortex, somatosensory cortex, and fusiform gyrus, which were much higher after 48 h and 72 h than after 24 h SD (61, 62). Previous neuroimaging studies observed disturbed regional spontaneous neural activities in brain areas of the two visual pathways in insomnia patients and individuals after SD (6, 15, 25, 40, 63). In the present study we observed acute SD was associated with altered FCD areas in the thalamus and dorsal system, including significant increased binarized shortFCD areas in the occipital gyrus, thalamus and postcentral gyrus, and increased binarized longFCD areas in the postcentral gyrus and visuospatial network. The increased FCD in these regions in the visual pathway could be considered a compensatory effect to sustain the cognitive performance despite a continuing decline of activity in the higher cognition related areas. This may generate an excessive hyperarousal status, which leads to increased sensory information processing (64).

There are extensive round-trip nerve interactive fibers between the cerebellum posterior lobe(s) and the cerebral cortex. The cerebellum posterior lobe(s) has been widely used for adjusting nerve function, adjusting the start, and planning and coordinating movement. It also works together with the cerebrum to complete functions; such as cognition, language, and emotion; and to initiate, plan, and coordinate movement (65–67). In light of mounting evidence for cerebellar involvement in various neurologic and psychiatric conditions, including obstructive sleep apnea (53), depression (68), primary insomnia (40, 63), mood

binarized shortFCD (B), increased binarized longFCD (C), and decreased binarized longFCD (D). ROC, Receiver operating characteristic; R, right; L, left; CPL, Cerebellum posterior lobe; CAL, Cerebellum anterior lobe; IFG, Inferior frontal gyrus; LG, Lingual gyrus; PrG, precentral gyrus; PG, Postcentral gyrus; PL, Paracentral lobule; SPL, Superior parietal lobule; SG, Supramarginal gyrus; SD, Sleep deprivation; RW, Rested wakefulness; shortFCD, short-range functional connectivity density; longFCD, long-range functional connectivity density.

disorders (69) and sleep deprivation (6); this is crucial. In the present study we found acute SD showed decreased shortFCD in the cerebellum, which may indicate functional deficits associated with decreased ability in adjusting coordinate movement.

### CONCLUSIONS

In summary, the longFCD and shortFCD analysis might be sensitive biomarkers to locate the underlying altered intrinsic brain functional organization in individuals during SD status relative to RW status with high degree of sensitivities and specificities. Specifically, the shortFCD analysis is more sensitive to locating the functional organization with more alterations in regional brain areas than that of longFCD. In the present study, we found that the longFCD and short FCDs in the highorder cognition related areas decreased while the arousal and sensorimotor related areas increased to sustain the cognitive performance. These findings expand our

long-range functional connectivity density.

knowledge and may help give us insight into a deeper understanding of the neurobiological mechanisms of how the functional organization was altered in the sleep-deprived brain.

There are several limitations that should be noted. First, our study has a relatively small sample size and future studies with a larger sample size is necessary to corroborate our findings. Second, the electronystagmogram has not been used to dynamically monitor the sleep in the SD procedure.

### REFERENCES


### AUTHOR CONTRIBUTIONS

DK wrote the main manuscript text, DK, RL, JiyZ, and WC conceived and designed the whole experiment, DK, LS, and JiaZ collected the data, DK, RL, and JiyZ analyzed the data.

### ACKNOWLEDGMENTS

This work was supported by Key Research and Development Projects in Shaanxi province in the field of social development (2017SF-081).


a day of waking and sleep deprivation. PLoS ONE (2015) 10:e0127351. doi: 10.1371/journal.pone.0127351


insomnia: a voxel-based morphometric study. Biol Psychiatry (2010) 67:182–5. doi: 10.1016/j.biopsych.2009.08.003


**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 Kong, Liu, Song, Zheng, Zhang and Chen. 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.

# Topological Reorganization of the Default Mode Network in Severe Male Obstructive Sleep Apnea

*Liting Chen1†, Xiaole Fan2†, Haijun Li <sup>1</sup> , Chenglong Ye1 , Honghui Yu1 , Honghan Gong1 , Xianjun Zeng1 , Dechang Peng1 \* and Liping Yan3 \**

*1Department of Radiology, The First Affiliated Hospital of Nanchang University, Nanchang, Jiangxi, China, 2Department of General Surgery, The First Affiliated Hospital of Nanchang University, Nanchang, Jiangxi, China, 3Department of Cardiology, People's Hospital of Jiangxi Province, Nanchang, Jiangxi, China*

#### *Edited by:*

*Xi-jian Dai, Medical School of Nanjing University, China*

#### *Reviewed by:*

*Feng Liu, Tianjin Medical University General Hospital, China Federico Giove, Centro Fermi – Museo storico della fisica e Centro studi e ricerche Enrico Fermi, Italy*

#### *\*Correspondence:*

*Dechang Peng pengdcdoctor@163.com; Liping Yan 824852148@qq.com*

*† These authors have contributed equally to this work.*

#### *Specialty section:*

*This article was submitted to Sleep and Chronobiology, a section of the journal Frontiers in Neurology*

*Received: 10 December 2017 Accepted: 04 May 2018 Published: 13 June 2018*

#### *Citation:*

*Chen L, Fan X, Li H, Ye C, Yu H, Gong H, Zeng X, Peng D and Yan L (2018) Topological Reorganization of the Default Mode Network in Severe Male Obstructive Sleep Apnea. Front. Neurol. 9:363. doi: 10.3389/fneur.2018.00363*

Impaired spontaneous regional activity and altered topology of the brain network have been observed in obstructive sleep apnea (OSA). However, the mechanisms of disrupted functional connectivity (FC) and topological reorganization of the default mode network (DMN) in patients with OSA remain largely unknown. We explored whether the FC is altered within the DMN and examined topological changes occur in the DMN in patients with OSA using a graph theory analysis of resting-state functional magnetic resonance imaging data and evaluated the relationship between neuroimaging measures and clinical variables. Resting-state data were obtained from 46 male patients with untreated severe OSA and 46 male good sleepers (GSs). We specifically selected 20 DMN subregions to construct the DMN architecture. The disrupted FC and topological properties of the DMN in patients with OSA were characterized using graph theory. The OSA group showed significantly decreased FC of the anterior–posterior DMN and within the posterior DMN, and also showed increased FC within the DMN. The DMN exhibited small-world topology in both OSA and GS groups. Compared to GSs, patients with OSA showed a decreased clustering coefficient (*C*p) and local efficiency, and decreased nodal centralities in the left posterior cingulate cortex and dorsal medial prefrontal cortex, and increased nodal centralities in the ventral medial prefrontal cortex and the right parahippocampal cortex. Finally, the abnormal DMN FC was significantly related to *C*p, path length, global efficiency, and Montreal cognitive assessment score. OSA showed disrupted FC within the DMN, which may have contributed to the observed topological reorganization. These findings may provide further evidence of cognitive deficits in patients with OSA.

Keywords: obstructive sleep apnea, default mode network, cognitive function, resting-state functional magnetic resonance imaging, graph theory

### INTRODUCTION

Obstructive sleep apnea (OSA) is a common sleep-disordered breathing condition characterized by repetitive cessations of breathing and/or reduced airflow due to frequent episodes of complete (apnea) or partial (hypopneas) obstruction of the upper airway during sleep. These respiratory events lead to sleep fragmentation (1), chronic intermittent hypoxemia (2), repetitive arousals, oxygen desaturation, and hypercapnic hypoxia. Moderate to severe OSA is estimated to occur in 12% of women and

**64**

Chen et al. Topological Reorganization of the DMN in OSA

up to 30% of men aged between 30 and 70 years, and these estimated prevalence rates are increasing as the population ages and due to the ongoing obesity epidemic (3). OSA is associated with an increased risk of both traffic and occupational accidents (4), decreased quality of life, and long-term health problems resulting from a number of concomitant diseases, including hypertension, cardiovascular impairment, stroke, chronic kidney disease (5), depression (6), anxiety, metabolic syndrome, insomnia, cognitive dysfunction, and even Alzheimer's disease (7). OSA is also associated with cognitive dysfunction, which is an important independent predictor of mortality, even in the absence of dementia manifestations. Cognitive deficits, including deficits in attention, memory, psychomotor function, executive functions, visuospatial function, and language ability, have been observed in patients with OSA (8, 9). Unfortunately, the neurological basis of neurocognitive dysfunction in patients with OSA has not been examined in detail.

Neuroimaging studies have been widely applied to explain these cognitive deficits and have revealed that patients with OSA show alterations in multiple brain regions, which are responsible for cognitive, affective, autonomic, and sensorimotor control (10–13). According to recent resting-state functional magnetic resonance imaging (rs-fMRI) studies, patients with OSA exhibited significant global and regional connectivity deficits, particularly in the default mode network (DMN) (14), salience network (SN), central executive network (CEN) (15).

The DMN is critical for maintaining brain function in the resting-state and experiences progressive deactivation as the brain engages in goal-directed activity. The DMN is a largescale network that includes a set of highly interconnected brain regions, such as the posterior cingulate cortex (PCC), precuneus, medial prefrontal cortex, and the medial, lateral and inferior parietal regions, which contribute to internal mentation, attention, and adaptive functions (16). In previous studies, patients with OSA showed significant regional deficits in spontaneous activity in DMN subregions (17–19). In addition, Zhang found patients with OSA exhibited structural and functional deficits in the anterior DMN and functional compensation in the posterior DMN (20) using independent component analysis (ICA). Moreover, Li et al. observed altered functional connectivity (FC) between eight pairs of DMN subregions, which was associated with cognitive impairment (21). Patients with OSA show abnormal deactivation in the DMN during working memory tasks. The deactivation of DMN regions is significantly associated with behavioral performance and episodic memory impairments, plays a role in cognitive impairment in patients with OSA (14). However, these previous studies were limited to the spontaneous abnormalities in local brain regions and did not directly assess important topological changes in the DMN of patients with OSA.

Accumulating evidence implicates aberrant activity in the DMN in cognitive impairments and symptoms associated with neuropsychiatric disorders, such as mild cognitive impairment (22), social anxiety disorder (23), primary insomnia (24), and depression (25). Functional alterations in the DMN have been proposed as a quantitative MRI assessment that may facilitate the clinical prognosis and diagnosis (26). Previous study that utilized graph theory approaches revealed alterations in the topological properties of the gray matter volume (GMV) structural network (27) and the brain functional network (28) in individuals with OSA. However, whether the FC is altered within the subregions of the DMN and the topological changes that occur in the DMN in patients with OSA remain unclear.

Here, we hypothesized that the cognitive impairment observed in patients with OSA might be attributed to disrupted FC and the topological configuration of the DMN, and the topological reorganization may probably related to abnormal DMN FC. To test our hypothesis, we applied graph theory approaches to analyze FC and the topological organization of the DMN in male patients with untreated severe OSA and examined the relationships between neuroimaging measures and clinical index.

### MATERIALS AND METHODS

### Participants

Fifty male patients with newly diagnosed untreated severe OSA and 46 male education- and age-matched good sleepers (GSs) were recruited from the Sleep Monitoring Room of the Respiratory Department at the First Affiliated Hospital of Nanchang University, China, from June 2015 to February 2017. Sex differences, depression, obesity, and anxiety may affect spontaneous brain activity, and female OSA patients exhibited a lower apnea–hypopnea index (AHI), which was frequently accompanied by depression and anxiety (29–32). To improve the credibility of our study, we only recruited untreated male patients with severe OSA to rule out potential confounders of sex differences, severity of OSA, depression, and anxiety. The inclusion criteria for patients with OSA and GSs were (1) OSA: an AHI greater than or equal to 30; GSs: an AHI less than 5; (2) male sex; (3) right-handedness; and (4) aged older than 20 years but less than 60 years. The exclusion criteria for all participants were (1) a history of other sleep disorders, such as insomnia or sleep-related eating disorder; (2) identifiable focal or diffuse abnormalities in structural MR images; (3) a history of neurological or mental illnesses (e.g., head injury, depression, psychosis, neurodegenerative diseases, hypothyroidism, and epilepsy); (4) a history of addiction; (5) a history of cerebrovascular disease; and (6) MRI contraindications, such as claustrophobia, metallic implants, or devices in the body. The study protocol was approved by the Medical Research Ethics Committee and the Institutional Review Board of the First Affiliated Hospital of Nanchang University. The current study was conducted according to the principles of the Declaration of Helsinki and the approved guidelines. Written informed consent was obtained from all participants.

### Overnight Polysomnography (PSG)

Prior to collecting MRI brain scans, overnight PSG was performed on all participants using the Respironics LE-Series Physiological Monitoring System (Alice5 LE, FL, USA) to confirm the OSA/ GS diagnosis and to exclude other sleep disorders. On the day prior to overnight PSG, all participants were required to refrain from using hypnotics and consuming alcoholic beverages or coffee. Overnight PSG was recorded from 10:00 p.m. to 6 a.m. A standard electroencephalogram (EEG, F4/M1, C4/M1, O2/M1, F3/M2, C3/M2, and O1/M2), chin electromyogram, electrocardiogram, electrooculogram, thoracic and abdominal respiratory movements, oral and nasal airflow, oxygen saturation (SaO2), body posture, and snoring were recorded. Studies were scored by a PSG technician and reviewed by a qualified sleep medicine physician according to the American Academy of Sleep Medicine (AASM) guidelines (33). Obstructive apnea was defined as any 10 s or longer decrease in airflow ≥90% with evidence of persistent respiratory effort. Hypopnea was defined as a reduction in airflow ≥30% lasting for more than 10 s, accompanied by 4% or greater oxygen desaturation and/or EEG arousal (33). The AHI was computed as the mean number of apnea and hypopnea events per hour during sleep. The arousal index (AI) was calculated as the average number of EEG arousals per hour of sleep.

### Neuropsychological Assessments

Each participant was evaluated with the Epworth sleepiness scale (ESS) (Chinese version) for excessive daytime sleepiness, which requires the participant to rate his/her probability of falling asleep in eight different situations on a scale of increasing probability from 0 to 3. The aggregate score of the ESS is 24, with a score greater than 6 indicating sleepiness, a score greater than 11 indicating excessive sleepiness, and a score greater than 16 suggesting risky sleepiness. In addition, we used the Montreal Cognitive Assessment (MoCA, Chinese version) (34) as a rapid screening tool to assess cognitive function in all participants, including executive function, calculation, memory, attention, abstraction, language, and orientation. The total MoCA score is 30, with a score less than or equal to 26 indicating the presence of a mild cognitive impairment.

### MRI Data Acquisition

All MRI data were collected on a 3.0-T MRI system (Siemens, Erlangen, Germany) by implementing an 8-channel phased-array head coil at the First Affiliated Hospital of Nanchang University, China. Comfortable fixed foam pads were used to reduce head movements and ear plugs were used to minimize scanner noise. First, each participant underwent conventional T1 and T2-weighted imaging to exclude the presence of massive brain lesions. Then, both an 8-min rs-fMRI scan with an echo planar imaging sequence [repetition time (TR) = 2,000 ms, echo time (TE) = 30 ms, field of view (FOV) = 230 mm × 230 mm, thickness = 4.0 mm, gap = 1.2 mm, flip angle = 90°, matrix = 64 × 64, slices = 30] and high-resolution three-dimensional T1-weighted structural MR images using a magnetization-prepared rapidly acquired gradient echo sequence with generalized autocalibrating partially parallel acquisition (GRAPPA) for K space fill (TR = 1,900 ms, TE = 2.26 ms, FOV = 250 mm × 250 mm, thickness = 1.0 mm, gap = 0.5 mm, flip angle = 9°, resolution matrix = 256 × 256, slices = 176) were collected. During the rs-fMRI scan, all subjects were asked to remain motionless, relax, keep their eyes closed, and avoid thinking systematically or falling asleep. After the MRI scan, the participants were asked whether they fell asleep and/or avoided thinking systematically during the entire scan.

## Functional Magnetic Resonance Imaging Data Preprocessing

Image preprocessing was performed using the Data Processing & Analysis Assistant for Resting-State Brain Imaging (DPABI, Chinese Academy of Sciences, Beijing, China1 ) (35) and Statistical Parametric Mapping (SPM8),2 which is run on the MATLAB R2012a (MathWorks, Natick, MA, USA) platform. Preprocessing included the following steps: (1) the first 10 volumes of each functional time series were discarded; (2) slice timing correction was performed for the remaining 230 volumes; (3) three-dimensional head motion correction was conducted for small head movements; (4) high-resolution T1-weighted structural images were co-registered to the mean realigned functional images for each individual, and the transformed T1 structural images were segmented into gray matter, white matter, and cerebrospinal fluid using a new segment algorithm with the diffeomorphic anatomical registration through exponentiated lie algebra (DARTEL) tool (36), the realigned functional volumes were spatially normalized to the Montreal Neurological Institute (MNI) space using the normalization parameters estimated in DARTEL, and then each voxel was re-sampled to 3 mm × 3 mm × 3 mm; (5) the images were spatially smoothed with a 6-mm full-width at halfmaximum Gaussian kernel; (6) the time series were further linearly detrended, and temporal bandpass filtering (0.01–0.08 Hz) was performed to reduce the effect of physiological high-frequency noise and low-frequency drifts; and (7) to further reduce possible sources of artifacts, the nuisance signal (white matter, cerebrospinal fluid, and global signal) and the Friston 24-parameter model (37) were regressed from the time series of all voxels using multiple regression analyses. The participants were excluded if the maximum head motion of maximum rotation was more than 2.0°, the maximum orthogonal direction displacement was more than 2.0 mm, or the mean relative root mean square was greater than 0.2 mm, according to the criteria (38, 39). Four patients with OSA were excluded. Finally, 46 male patients with untreated severe OSA and 46 male age- and education-matched GSs were included in the current study.

### DMN Construction and Graph Analyses Definition of DMN Subregions

According to a previous study, we focused on the DMN and chose a specific set of 20 regions of interest (ROI) with substantial agreement with the functional and anatomic partitions of the DMN (**Table 1**) (16).

### DMN Functional Connectivity and Graph Analyses

A network is composed of a set of nodes and edges between different nodes. The mean time series for each voxel within the ROI of the DMN was extracted using spherical seeds (6 mm in radius) based on the MNI coordinate system. Next, the Pearson correlation coefficients were computed between each pair of DMN subregions in each participant to generate a 20 × 20 correlation matrix of the DMN. Then, we used the graph theoretical network

<sup>1</sup>http://www.restfmri.net (Accessed: 2008).

<sup>2</sup>http://www.fil.ion.ucl.ac.uk (Accessed: 1994).


*Coordinates are based on the MNI coordinate system, and each region of the DMN was acquired by spherical seeds with a radius of 6 mm.*

analysis (GRETNA)3 toolbox (40) to evaluate the topological organization of the DMN.

### Threshold Selection

In this study, the DMN was modeled based on an undirected, binarized method. The establishment of a sparsity threshold (Sp), which is defined as the fraction of the number of existing edges divided by the maximum possible number of edges in a network, ensured that the resulting networks had the same number of edges and minimized the influence of potential confounders on the overall correlation strength between groups (41). In the present study, we computed the network properties of the DMN over a wide range of sparsity levels (from 0.05 to 0.50 using an interval of 0.01), in which the number of spurious edges was minimized and the small-world parameters could be properly estimated (42).

### Network Metrics

In this study, we used the graph theory approach to calculate the global and nodal network properties of the DMN in patients with OSA and GSs. The area under the curve (AUC) of each network metric was calculated for statistical comparison, which was extracted by thresholding across a range of sparsity values to depict changes in the topological characterization of brain

networks, and which is susceptible to detecting topological alterations of brain disorders (41, 43).

## Global Network Metrics

#### Small-World Parameters

Small-world parameters (1) small-worldness, σ, is a fascinating model for the description of complex brain networks that not only support both specialized and integrated information processing but also facilitates an energy-efficient balance between network segregation and integration. Mathematically, a real brain network is considered a small-world network if it displays a much higher clustering coefficient (*C*p) and a similar characteristic path length (*L*p) (compared with 1,000 matched random networks in our study) and meets the following criteria: normalized clustering coefficients γ = *C C* > p p real rand / 1 and normalized characteristic path length λ = *L L* ≈ p p real rand / 1. The small-worldness, σ = γ/λ, is typically >1 for small-world networks (44, 45); (2) The clustering coefficient of node *i* (*Ci*) is defined as the percentage of the number of existing connections among the node's nearest neighbors and the maximum possible number of connections. The clustering coefficient of network *C*p is the average of *Ci* across nodes, which is a measure of network segregation (44); (3) The characteristic path length, *L*p, is quantified as the average of the shortest path length that links all pairs of nodes in the network, which is the most commonly used measure of network information integration (45). The characteristic path length was calculated as the "harmonic mean" distance between all possible pairs of regions to deal with the possible disconnected graphs dilemma in the present study (46). The largest component sizes of individual networks over the sparsity range of 0.05–0.50 with an interval of 0.01, see in **Figure 1**.

#### Network Efficiency

Network efficiency, including global efficiency, *E*glob, which represents the capacity of parallel information transmission over

<sup>3</sup>http://www.nitrc.org/projects/gretna/ (Accessed: December 12, 2012).

Chen et al. Topological Reorganization of the DMN in OSA

the network, and local efficiency, *E*loc, represents the capacity of a network to transmit information at the local level and measures the fault tolerance of the network (42).

### Regional Network Metrics

The degree for a brain region is defined as the number of edges of a node that connect with the remaining nodes in the network, thus measuring how interactive a particular node is in the network. The nodal betweenness is designated as the fraction of shortest paths between two nodes passing through the area in the network and measures the influence of a region on network communication. Nodal efficiency is defined as the inverse of the harmonic mean of the shortest path length in the network, quantifying the importance of the nodes for communication within the network (42).

### Statistical Analysis

Demographic and clinical characteristics of the OSA and GS groups were compared using independent two-sample *t*-tests with IBM Statistical Package for the Social Sciences 20.0 software (IBM SPSS Inc., Chicago, IL, USA). Independent two-sample *t*-tests were performed to compare group differences in the AUC of global network metrics and ROI-ROI FC of the DMN. We also compared nodal properties between patients with OSA and GSs and Bonferroni correction was performed for multiple comparison. The effects of age, body mass index (BMI), and educational level were diminished by a regression analysis. Abnormal DMN FC was calculated as the average of the correlation coefficients of the DMN in patients with OSA that showed significant betweengroup differences. The relationships between abnormal DMN FC and topological metrics of the DMN, and the relationships between network metrics with significant between-group differences and clinical indices in the OSA group were investigated using a Pearson correlation analysis. *p* < 0.05 was considered statistically significant.

## RESULTS

### Demographic and Clinical Data

As shown in **Table 2**, significant inter-group differences were observed in BMI, AHI, total sleep time, Stage 1, rapid eye movement (REM), AI, SaO2 < 90%, average SaO2, oxygen desaturation index, nadir SaO2, MoCA score, visuospatial/executive, delayed memory, attention, abstraction, orientation and ESS score (*p* < 0.05). No inter-group differences in Stage 2 or Stages 3 + 4 were observed (*p* > 0.05).

### Changes in FC Within the DMN Between Patients With OSA and GSs

Compared to GSs, patients with OSA exhibited significantly decreased FC between the bilateral PCC and the bilateral hippocampal formation (HF) and left retrosplenial cortex (Rsp), between the left temporal pole (TempP) and the dorsal medial prefrontal cortex (dMPFC) and left temporal parietal junction (TPJ), between the left Rsp and the bilateral anterior medial prefrontal cortex (aMPFC) and the left posterior inferior parietal lobule (pIPL), between the bilateral HF and the bilateral Rsp, Table 2 | Comparison of the demographic and clinical data from the patients with OSA and GSs.


*Data are presented as the mean* ± *SD.*

*OSA, obstructive sleep apnea; GSs, good sleepers; BMI, body mass index; AHI, apnea–hypopnea index; REM, rapid eye movement; SaO2* < *90%, percentage of total sleep time spent at an oxygen saturation less than 90%; MoCA, Montreal cognitive assessment; ESS, Epworth sleepiness scale; N, number.*

*\*p* < *0.05, which was considered statistically significant.*

and between the right HF and the right TPJ and the right pIPL. Patients with OSA displayed significantly increased FC in the DMN between the right TempP and the right parahippocampal cortex (PHC) and between the right and left HF, compared to GSs (**Figure 2**; **Table 3**). The abnormal DMN FC was positively correlated with *C*p (*r* = 0.384, *p* = 0.008) and *L*p (*r* = 0.338, *p* = 0.022), and negatively correlated with *E*glob (*r* = −0.565, *p* < 0.001) in patients with OSA (see **Figure 3**).

### Differences in Global Network Measures of the DMN

In the defined wide range of thresholds (here from 0.05 to 0.50), both the patients with OSA and GSs exhibited σ value larger than 1, γ value obviously larger than 1, and λ value of approximately equal to 1 (see **Figure 4**), suggesting that both patients with OSA and GSs have typical small-world topology. However, compared to GSs, patients with OSA showed a significantly decreased *C*<sup>p</sup> (*t*=−2.200, *p*= 0.030) and a decreased *E*loc (*t*=−1.942, *p*= 0.054), which have a trend for difference. There was no significant difference in σ (*t* = 0.412, *p* = 0.483), *L*p (*t* = −0.004, *p* = 0.997) or *E*glob (*t* = −0.035, *p* = 0.972). Global network measures are illustrated in **Figure 5**.

### Group Differences in Regional Network Measures of the DMN

Patients with OSA showed abnormal nodal centrality, which showed significant between-group differences in at least one

Table 3 | Abnormal functional connectivity (FC) within the default mode network (DMN) between patients with obstructive sleep apnea (OSA) and good sleepers (GSs).


*Abnormal FC within the DMN between patients with OSA and GSs (p* < *0.05, uncorrected).*

*L, left; R, right; PCC, posterior cingulate cortex; Rsp, retrosplenial cortex; HF,* 

*hippocampal formation; TempP, temporal pole; dMPFC, dorsal medial prefrontal cortex; TPJ, temporal parietal junction; PHC, parahippocampal cortex; aMPFC, anterior medial prefrontal cortex; pIPL, posterior inferior parietal lobule.*

nodal metric, including nodal betweenness, nodal efficiency, and nodal degree. Compared with the GSs, patients with OSA showed decreased nodal centralities in the left PCC and dMPFC, and increased nodal centralities in the vMPFC and the right PHC (*p* < 0.05, uncorrected). Regional network measures are illustrated in **Table 4**.

### Correlations Between Network Measures With Group Differences and Clinical Variables

Within the OSA group, the abnormal DMN FC was negatively correlated with the MoCA score (*r* = −0.366, *p* = 0.012). *C*p was negatively correlated with the MoCA score (*r* = −0.332, *p* = 0.024) and delayed memory (*r* = −0.306, *p* = 0.039). The nodal degree of the left PCC was positively correlated with the nadir SaO2 (*r* = 0.317, *p* = 0.032), and nodal betweenness of the right PHC was positively correlated with the MoCA score (*r* = 0.309, *p* = 0.037). The nodal betweenness (*r* = 0.297, *p* = 0.045), degree (*r* = 0.358, *p* = 0.015), and efficiency (*r* = 0.334, *p* = 0.023) of the right PHC were positively correlated with delayed memory (see **Figure 6**).

### DISCUSSION

The present study applied graph theory approaches to provide evidence that the cognitive impairments observed in patients with OSA might be attributed to the topological configuration of the DMN, which probably resulted from the abnormal DMN FC. Although the DMN of patients with OSA exhibited small-world properties, patients with OSA showed decreased *C*p and *E*loc, abnormal nodal centralities in the DMN, and abnormal FC within the DMN, implying a disturbance in the functional differentiation of the DMN. In addition, the abnormal DMN FC was related to *C*p, *L*p, *E*glob, and the MoCA score. The disrupted topological properties of the DMN significantly influenced cognitive function, including delayed memory and memory extraction in patients with OSA.

### Abnormal FC Within the DMN in Patients With OSA

The current study revealed significantly decreased FC in the anterior–posterior DMN involving the prefrontal, parietal and

Figure 3 | The relationship between abnormal functional connectivity (FC) and topological metrics of the default mode network (DMN) in patients with obstructive sleep apnea (OSA). The abnormal DMN FC value was significantly correlated with *C*p, *L*p, and *E*glob in patients with OSA. *p* < 0.05, which was considered statistically significant.

Figure 4 | Small-world parameters of default mode network in patients with obstructive sleep apnea (OSA) and good sleepers (GSs). Graphs show that in the defined wide range of thresholds, both the patients with OSA and GSs exhibited normalized clustering coefficient (γ) obviously larger than 1, normalized path lengths (λ) approximately equal to 1, and small-worldness (σ) larger than 1, suggesting that both OSA patients and GSs show typical small-world topology.

Figure 5 | Graphs showing the small-world parameters and network efficiency of the default mode network in patients with obstructive sleep apnea (OSA) and good sleepers (GSs). Although OSA and GSs have typical small-world topology, compared to GSs, patients with OSA showed a significantly decreased *C*p (*t* = −2.200, *p* = 0.030) (*p* < 0.05, uncorrected), and a decreased *E*loc (*t* = −1.943, *p* = 0.054), which have a trend for difference.

Table 4 | Between-group differences in regional network measures of the default mode network (DMN) in patients with obstructive sleep apnea (OSA) and good sleepers.


*Patients with OSA showed abnormal nodal centrality in PCC.L, dMPFC, vMPFC, and PHC.R, which showed significant between-group differences in at least one of the three nodal metrics.*

*# p* < *0.05, uncorrected.*

*\*Bonferroni correction p* = *0.05.*

*PCC.L, left posterior cingulate cortex; dMPFC, dorsal medial prefrontal cortex; vMPFC, ventral medial prefrontal cortex; PHC.R, right parahippocampal cortex.*

temporal regions in patients with OSA. Zhang et al. (20) found that patients with OSA exhibited decreased FC in the anterior DMN and a compensatory increased FC in the posterior DMN. Decreased FC in the anterior–posterior DMN indicated that the

showed the relationship between the aberrant network attribute parameters and clinical index in patients with OSA. The red ball represents the increased nodal centrality and the blue ball represents the decreased nodal centrality. *E*loc, local efficiency; *C*p, clustering coefficient; PCC.L, left posterior cingulate cortex; PHC.R, right parahippocampal cortex.

transmission of information and integration of long-distance connectivity between different regions may be damaged in patients with OSA.

We also observed significantly decreased FC in the posterior DMN, which includes the PCC, HF, temporal, and parietal lobes and the limbic system. *E*loc predominantly reflects short-distance connections between neighboring regions (45). Decreased shortdistance connections that are primarily located in the posterior DMN may lead to decreased *C*p and decreased *E*loc of the DMN in patients with OSA. The PCC and HF are connected anatomically and functionally, and these functional interactions are presumed to underlie normal episodic memory capacity (47). Patients with OSA showed decreased FC between the right HF and the PCC, which is related to delayed memory (21). Based on the results, the OSA group showed decreased FC between the bilateral PCC and the bilateral HF, consistent with previous studies (21). Decreased FC in the anterior–posterior DMN and posterior DMN may further indicate cognitive impairments in patients with OSA (48).

Park observed abnormal FC in various brain regions, and altered FC subsequently resulted in disrupted topological properties in patients with OSA, particularly in the integrative aspects of brain network organization (49). Given the significant association between abnormal DMN FC and *C*p, *L*p, and *E*glob of the DMN in the current study, we believe that disrupted FC within the DMN may contribute to the topological reconfiguration of the DMN in patients with OSA. Furthermore, abnormal DMN FC was associated with the MoCA score. Therefore, the abnormal DMN FC may partially explain the impaired cognitive function and topological reconfiguration in patients with OSA.

### Global Network Measures of the DMN

Patients with OSA have recently been shown to display an abnormal small-world organization in both functional (28, 49) and structural (27) brain networks. In the present study, both patients with OSA and GSs showed efficient economic smallworld organization in the DMN. Although the DMN has small-world properties, our results identified decreased *C*p and *E*loc of the DMN in patients with OSA. Thus, individuals with OSA likely have sparse connectivity and disconnections between adjacent brain regions in the DMN, resulting in decreased *C*<sup>p</sup> and *E*loc. *C*p is a metric that quantifies the strength of network segregation (50). The present results indicate a decline in functional differentiation in the DMN, suggesting that highly local specialization and the integrity of the DMN may be impaired in patients with OSA. *E*loc essentially reflects the fault tolerance of the network and the capacity for transmitting information over local networks (45). Our finding of a decreased *E*loc suggests disrupted DMN architecture in patients with OSA that is characterized by higher vulnerability and a decreased capacity for regional information processing. Moreover, *C*p was negatively correlated with the MoCA score and delayed memory, further illustrating that disrupted global topology of the DMN influence cognitive impairments in patients with OSA, including delayed memory and memory extraction.

### Regional Network Measures

Nodal betweenness centrality, nodal efficiency, and nodal degree were combined to compare the regional topological organization between patients with OSA and GSs in our study. Deceased nodal centrality was identified in the PCC and dMPFC. The PCC has strong reciprocal connections with other structures involved in cognitive function (51), the collection and evaluation of information, attention processing, personal significance, and evoked emotion (16). Previous structural neuroimaging studies have observed decreased GMV (52) and white matter integrity (53) in the PCC in patients with OSA. Furthermore, patients with OSA show decreased brain activation, decreased degree centrality, and FC alterations in the PCC (17, 18, 20, 21, 54). DMN dysfunction is associated with impairments in cognitive performance (55, 56). Intermittent hypoxia is a major factor in DMN dysfunction in patients with OSA (14, 21). We also observed a positive correlation between the nodal degree of the left PCC and nadir SaO2, suggesting that the functional damage of the PCC was related to intermittent hypoxia, which may be a major factor involved in DMN dysfunction and may further explain cognitive dysfunction in patients with OSA.

The dMPFC subsystem includes the dMPFC, temporal parietal junction, lateral temporal cortex, and TempP, which are involved in social cognition, metacognition, and mental state inference (16). Patients with OSA displayed decreased FC and reduced GMV in the MPFC of the anterior DMN, indicating structural and functional deficits (20). The OSA group showed decreased nodal centrality of the dMPFC, but a compensatory increase in nodal centrality of the ventral medial prefrontal cortex in the present study, which may also confirm the deficiency in the dMPFC subsystem of the DMN in patients with OSA.

The PHC plays an important role in episodic memory, autobiographical memory and episodic simulation, spatial memory, scene perception, and spatial navigation (47). Previous voxel-based morphometry studies revealed that atrophy (57) and regional cerebral blood flow were significantly reduced in the bilateral PHC (58), which may be related to cognitive impairments in patients with OSA. Nevertheless, we found a compensatory increase in nodal centrality in the right PHC, which was positively correlated with delayed memory, may partially explain the deficits in memory, spatial learning, memory extraction and attention in patients with OSA.

### Limitations

Several limitations in this study should be addressed. First, we only revealed the small-world properties of the DMN, but patients with OSA exhibited disruptions in the DMN, as well as the SN and CEN (15). Therefore, further investigations of other specific brain networks are necessary. Second, we only specifically selected 20 nodes of the DMN (16) and characterized the DMN using an unbiased seed-based FC approach. More nodes of the DMN should be used to construct the DMN and the present findings should be validated by ICA. Third, the global network measures, nodal centrality, and ROI-ROI FC results were not corrected by multiple comparisons, meaning that this study should be considered an exploratory analysis. In addition, a more detailed neuropsychological assessment questionnaire must be used to obtain more interesting data.

## CONCLUSION

In the current GRAPPA study, patients with OSA showed disrupted FC and topological reorganization of the DMN. Abnormal DMN FC may contribute to the topological configuration of the DMN and cognitive impairment in patients with OSA. These results provide important insights into the neurobiological mechanisms of both disrupted FC and disrupted network properties of the DMN, which may partially account for the impaired cognitive function in patients with OSA.

## ETHICS STATEMENT

The study protocol was approved by the Medical Research Ethics Committee and the Institutional Review Board of the First Affiliated Hospital of Nanchang University. The current study was conducted according to the principles of the Declaration of Helsinki and the approved guidelines. Written informed consent was obtained from all participants.

## 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 Natural Science Foundation of China (Grant No. 81560285), the Graduate Innovation Foundation of Jiangxi, China (Grant No. YC2016- S100), the Natural Science Foundation of Jiangxi, China (Grant No. 20171BAB205070, 20132BAB205100), the Education Department Foundation of Jiangxi, China (Grant No. 700544006), the Science and Technology Support Program of Jiangxi, China (Grant No. 20132BBG70061, 20141BBG70026) and the Doctoral Project Startup Fund (Grant No. 700544005).

## REFERENCES


**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 Chen, Fan, Li, Ye, Yu, Gong, Zeng, Peng and Yan. 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 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.*

# Recursive Partitioning Analysis of Fractional Low-Frequency Fluctuations in Narcolepsy With Cataplexy

Xiao Fulong<sup>1</sup> , Lu Chao<sup>2</sup> , Zhao Dianjiang<sup>3</sup> , Zou Qihong<sup>4</sup> , Zhang Wei <sup>5</sup> , Zhang Jun<sup>5</sup> \* and Han Fang<sup>1</sup> \*

<sup>1</sup> Department of Respiratory and Critical Care Medicine, Sleep Medicine Center, Peking University People's Hospital, Beijing, China, <sup>2</sup> Department of Radiology, Peking University International Hospital, Beijing, China, <sup>3</sup> Center for MRI Research, Academy for Advanced Interdisciplinary Studies, Peking University, Beijing, China, <sup>4</sup> PKU-Upenn Sleep Center, Peking University International Hospital, Beijing, China, <sup>5</sup> Department of Neurology, Peking University People's Hospital, Beijing, China

#### Edited by:

Hengyi Rao, University of Pennsylvania, United States

#### Reviewed by:

Hiroshi Kadotani, Shiga University of Medical Science, Japan Axel Steiger, Max-Planck-Institut für Psychiatrie, Germany Benito de Celis Alonso, Benemérita Universidad Autónoma de Puebla, Mexico

\*Correspondence:

Zhang Jun who626@163.com Han Fang hanfang1@hotmail.com

#### Specialty section:

This article was submitted to Sleep and Chronobiology, a section of the journal Frontiers in Neurology

Received: 11 August 2018 Accepted: 16 October 2018 Published: 02 November 2018

#### Citation:

Fulong X, Chao L, Dianjiang Z, Qihong Z, Wei Z, Jun Z and Fang H (2018) Recursive Partitioning Analysis of Fractional Low-Frequency Fluctuations in Narcolepsy With Cataplexy. Front. Neurol. 9:936. doi: 10.3389/fneur.2018.00936 Objective: To identify narcolepsy related regional brain activity alterations compared with matched healthy controls. To determine whether these changes can be used to distinguish narcolepsy from healthy controls by recursive partitioning analysis (RPA) and receiver operating characteristic (ROC) curve analysis.

Method: Fifty-one narcolepsy with cataplexy patients (26 adults and 25 juveniles) and sixty matched heathy controls (30 adults and 30 juveniles) were recruited. All subjects underwent a resting-state functional magnetic resonance imaging scan. Fractional low-frequency fluctuations (fALFF) was used to investigate narcolepsy induced regional brain activity alterations among adult and juveniles, respectively. Recursive partitioning analysis and Receiver operating curve analysis was used to seek the ability of fALFF values within brain regions in distinguishing narcolepsy from healthy controls.

Results: Compared with healthy controls, both adult and juvenile narcolepsy had lower fALFF values in bilateral medial superior frontal gyrus, bilateral inferior parietal lobule and supra-marginal gyrus. Compared with healthy controls, both adult and juvenile narcolepsy had higher fALFF values in bilateral sensorimotor cortex and middle temporal gyrus. Also juvenile narcolepsy had higher fALFF in right putamen and right thalamus compared with healthy controls. Based on RPA and ROC curve analysis, in adult participants, fALFF differences in right medial superior frontal gyrus can discriminate narcolepsy from healthy controls with high degree of sensitivity (100%) and specificity (88.9%). In juvenile participants, fALFF differences in left superior frontal gyrus can discriminate narcolepsy from healthy controls with moderate degree of sensitivity (57.1%) and specificity (88.9%).

Conclusion: Compared with healthy controls, both the adult and juvenile narcolepsy showed overlap brain regions in fALFF differences after case-control comparison. Furthermore, we propose that fALFF value can be a helpful imaging biomarker in distinguishing narcolepsy from healthy controls among both adults and juveniles.

Keywords: narcolepsy, functional magnetic resonance imaging, fractional low-frequency fluctuations, recursive partitioning analysis, receiver operating characteristic curve analysis

**75**

### INTRODUCTION

Narcolepsy is a chronic sleep disorder, characterized by excessive daytime sleepiness, cataplexy, sleep paralysis, hypnagogic, and hypnopompic hallucination and disturbed nocturnal sleep. A deficient endogenous orexin system due to neuronal degeneration in the hypothalamus is the main pathophysiology of the narcolepsy in the human (1). It is indicated that loss of hypocretin is thought to be an underlying cause to the sleeprelated changes and cataplexy, also deficiency in hypocretin system can result in the abnormal cognition and emotion observed in narcolepsy patients (2).

In the past decades, neuroimaging techniques have played an important role in the understanding of physiology and pathology in human sleep medicine (3, 4). Changes in brain structure and function have been investigated in hypersomnia and narcolepsy (5–8). These studies include the measurement of brain structure, such as voxel-based morphometry, diffusion tensor imaging, and metabolic studies using spectroscopy, as well as functional view, such as positron emission tomography (PET), single photon emission computed tomography (SPECT), and functional magnetic resonance imaging (fMRI). Detection of local dysfunction is crucial to the clinical research and clinical practice. Results from previous neuroimaging studies suggested that reduction of hypocretin can lead to attenuation in both resting state glucose metabolism and perfusion within cortex (9). Abnormal perfusion and glucose metabolism in the hypothalamus and prefrontal cortex has been detected among narcolepsy using PET and SPECT (5, 10). A very recent PET research in a large group of junior narcolepsy patients observed that abnormality in many frontal and subcortical brain areas, exhibited significantly correlation with neuro-cognition performance (7).

Resting state fMRI can provide information about spontaneous brain activity by assessment of blood oxygen level dependent (BOLD) signal fluctuations. The resting BOLD signal fluctuations are thought to represent spontaneous and functional process, although on a slower time response. Brain regions involved in specific task or stimuli display coherent low BOLD signal fluctuations in the resting state. Amplitude of lowfrequency fluctuations (ALFF), in which the square root of power spectrum was integrated in a low-frequency range, was developed for detecting the local intensity of BOLD signal fluctuations (11). ALFF has already been applied to fMRI studies about attention deficit hyperactivity disorder and Alzheimer's disease, also in the exploration of neural mechanism of sleep disorders, such as insomnia, sleep deprivation and sleep apnea (11–14). Although ALFF was considered to be a useful tool in detecting the regional neural activity, physiological noise, such as the repetition times in MRI scan and so on, are not critically considered in the ALFF calculation. Therefore, a modified calculation called fractional amplitude of low-frequency fluctuation (fALFF), which means the ratio of the power spectrum of low frequency (0.01–0.08 Hz) to that of the entire frequency range, has been proven to suppress non-specific noise components and improve the effectiveness in exploring local BOLD signals (15). Considering the robustness and stability of ALFF and fALFF calculation, both the ALFF and fALFF can be indicated as potential biomarkers in neuroimaging studies (16).

Recursive partitioning analysis (RPA) could provide a simple, straightforward and intuitive method to classify subjects or to identify synergistic interaction among numerous factors (17, 18). RPA is considered to be a machine learning method and usually requires a large sample to establish a classification model from a training data and verify this model by another test sample. RPA can be realized through computer and many medical care studies have used RPA to detect prognostic and risk factors (19, 20), as well as diagnosis in imaging study (21). Classification and regression tree (CRT) analysis is a kind of tree-building technique from RPA to the generation of clinical decision rules (22). It is a non-parametric method for multi-model numerical data and categorical predictors, also suitable for managing the interactions between predictors which are crucial in determining the outcome. The CRT is a relatively data-driven machine learning calculation, which produces decision tree model easy to interpret (22).

In the present study, we hypothesized that fALFF has the ability to indicate narcolepsy induced neurobiological mechanism with the location of altered neural brain activity, and further distinguish narcolepsy from healthy controls with excellent sensitivity and specificity. Specifically, classification and regression tree form recursive portioning analysis (RPA) and receiver operating characteristic (ROC) curve analysis were used to investigate and validate the ability of fALFF values in distinguishing narcolepsy from healthy controls.

### MATERIALS AND METHODS

### Participants

Twenty six adult narcolepsy patients and another 25 juvenile patients were recruited as newly diagnosed narcolepsy with cataplexy according to the International Classification of Sleep Disorders (ICSD)-3 (23) from the Sleep Medicine Center of the Respiratory Department at Peking University People's Hospital between November 2016 and February 2018. Another 60 genderand age- matched healthy volunteers (30 juveniles and 30 adults) were recruited from the hospital and community (**Table 1**). None of healthy controls had any consistent psychiatric or neurologic condition producing excessive daytime sleepiness. All narcolepsy cases were the first-time visitors and previously had never taken psychiatric stimulant medications. The clinical diagnosis of narcolepsy was made by a sleep specialist based on both excessive daytime sleepiness lasting more than 3 months and defined history of cataplexy, according to the International Classification of Sleep Disorders criteria for narcolepsy. The final diagnosis of narcolepsy was confirmed by a polysomnogram followed by a next day multiple sleep latency test (MSLT). Detailed information, including the presentation of excessive

**Abbreviations:** fMRI, Functional magnetic resonance imaging; PET, Positron emission tomography; SPECT, Single photon emission computed tomography; BOLD, Blood oxygen level dependent; ROC, Receiver operating characteristic; MSLT, Multiple sleep latency test; RPA, Recursive partitioning analysis; AUC, Area under the curve


The P value for gender distribution in the two groups was obtained by the Chi-square test. The P values for differences in age and years of education in the two groups were obtained by the two-sample t test. Values are expressed as the mean±SD or median (25%quartile, 75%quartile). EDS, excessive daytime sleepiness.

daytime sleepiness and cataplexy, hypnagogic hallucinations, and sleep paralysis, were obtained from patients.

The exclusion criteria for both narcolepsy and normal subjects were as follows: (1) other sleep disorders, such as obstructive sleep apnea, insomnia; (2) diabetes, and chronic obstructive pulmonary disease and heart disease; (3) neurological diseases and structural lesion based on brain MRI findings; (4) psychosis disorder; (5) alcohol, drug, and substance abuse; (6) inborn or congenital diseases; (7) MRI contraindications, such as claustrophobia or foreign implants in the body.

This research was performed in accordance with the ethical guidelines of the Declaration of Helsinki (version 2002) and was approved by the Medical Ethics Committee of Peking University People's University. All participants provided written informed consent.

### Imaging Data Acquisition

MRI examination was performed exactly following the daytime MSLT. MRI data were obtained on 3T (3 Tesla) scanner (Siemens, Skyra, Germany) using an 8-channel brain phased-array coil. Foam pads were used to minimize subject head motion, and headphones were used to reduce scanner noise. Resting BOLD MRI scans were obtained with gradient-echo planar imaging (TR = 2030 ms, TE = 30ms, slice = 33, FA = 90◦ , FOV = 224 × 224 mm, matrix = 64 × 64, voxel size=3.5 × 3.5 × 3.5), after the BOLD MRI scan, a high-resolution T1-weighted structural image was acquired with the following parameters: TR = 1900 ms, TE = 2.55 ms, FA = 9 ◦ , FOV = 240 × 240 mm, thickness = 1 mm. A total 240 brain functional volumes were acquired in the resting BOLD MRI scans. All participants, including patients and controls were asked to resist sleeping in order to remain fully awake (5, 24), not to move and keep eye open during the whole MRI scan, supervised clinically and by video both a physician and a technician during the whole process. In addition, we controlled for the absence of emotional triggering factors during the entire process to avoid cataplexyrelated events.

### Functional Imaging Data Analysis

Functional MRI data preprocessing was performed using the Data Processing & Analysis for Resting State Brain Imaging V2.1 [DPABI V2.1 (25)], which works with Statistical Parametric Mapping (SPM8) implemented in the MATLAB (The Math Works, Inc., Natick, MA, USA) platform. The first 5 functional volume images of each subject's dataset were discarded, then the remaining fMRI data were corrected for slice timing and realigned for motion correction. Participants with head motion exceeding 3 mm in translation and 3◦ in rotation were rejected. Anatomical and functional images were manually reoriented to the anterior commissure, and structural images were co-registered to the functional images for each subject using a linear transformation. Also the transformed structural images were segmented into gray matter, white matter, and cerebrospinal fluid by the new segmentation in SPM8. For adult participants, the functional images were normalized to the standard Montreal Neurological Institute (MNI) space template with a resampling voxel size of 3 × 3 × 3 mm. For juvenile participants, the functional images were normalized to the CCHMC pediatric brain template (irc.cchmc.org, The imaging research center, Cincinnati Children's Hospital Medical Center) (26) with a resampling voxel size of 3 × 3 × 3 mm. The normalized functional images were smoothed using a Gaussian filter 4 mm FWHM. Linear trends were removed within each time series. The covariates were regressed out from the time series of every voxel, including the white matter signal, cerebrospinal fluid signal, Friston 24 motion parameters (27, 28) and the global signal. The calculation of the fALFF have been reported in previous studies (15). After fALFF calculation, the time series were filtered using typical temporal bandpass (0.01–0.1Hz) to reduce low-frequency drift, physiological high-frequency respiratory and cardiac noise. To reduce the global effects of variability across the participants, the individual fALFF map was transformed to Z score (minus the global mean value and then divided by the standard deviation) other than simply being divided by the global mean (15).

### STATISTICAL ANALYSIS

### Demographic Data

The demographic data differences between narcolepsy and healthy controls were computed by independent two sample t-test with the IBM Statistical Package for the Social Sciences 23.0 software (IBM SPSS Inc., Chicago, IL, USA). We set the significance level at P < 0.05. Values are expressed as the mean ± SD or median (25%quartile, 75%quartile).

### Between Group Differences in FALFF

A two-sample t-test was performed between narcolepsy and controls using age, gender, and years of education as nuisance covariates to assess case-control comparison in fALFF among adults and juveniles, respectively, corrected for false discovery rate (FDR, P < 0.05).

### Recursive Partitioning Analysis (RPA)

Narcolepsy cases and healthy controls were randomly split into testing data and validation data in the proportion of 7:3, respectively. Testing data was used to develop decision tree model by recursive partitioning analysis (70%) and validation data was used to test the developed model (30%). In the analysis of between group fALFF differences, the brain regions showing statistically significant in adults or juveniles were selected as ROI seeds, respectively, then the mean fALFF value in the region of interesting (ROI) regions were extracted. Recursive partitioning analysis was performed using mean fALFF values within ROI regions showing group differences in adults or juveniles, respectively. We chose the Classification and Regression Trees (CRT) technique in the process of RPA to define narcolepsy or control. The criteria for splitting node including the following: child nodes derived from a parent node should be as homogeneous as possible with the dependent variables, corresponding cut-off points should result in the minimal P value, provides the minimal P value was ≤ 0.0001 (29). Terminal nodes were identified to a class when the significant level of comparison between 2 terminal nodes was >0.05 (29). As for the validation data, sensitivity, specificity, false positive rate (FPR), false negative rate (FNR), positive predictive value (PPV), negative predictive value (NPV), and accuracy were calculated according to the fALFF value cut-off obtained on the basis of developed decision tree model. ROC analysis was applied to measure the discrimination of the decision tree model. RPA process and ROC curve statistical analysis was performed with R (http://www.R-project.org) and Empower-Stats software (www. empowerstats.com, X&Y solutions, Inc., Boston, MA, USA).

### RESULTS

### Demographic Data

As shown in **Table 1**, there were no significant differences between narcolepsy and healthy controls in age, gender, years of education.

### Differences in FALFF Between Narcolepsy and Healthy Controls

In adult participants, compared with healthy controls, narcolepsy had lower fALFF values in bilateral medial superior frontal gyrus (SFGmed), bilateral inferior parietal lobule (IPL) and left supra-marginal gyrus (SMG). Compared with healthy controls, narcolepsy had higher fALFF values in bilateral sensorimotor cortex (SMC) and bilateral middle temporal gyrus (MTG) (**Figures 1A**, **2A** and **Table S1**). In juvenile participants, compared with healthy controls, narcolepsy had lower fALFF values in bilateral medial superior frontal gyrus, bilateral inferior parietal lobule, left superior frontal gyrus (SFG), and right supramarginal gyrus. Compared with healthy controls, narcolepsy had higher fALFF values in bilateral sensorimotor cortex, right middle temporal gyrus, right putamen and right thalamus (**Figures 1B**, **2B** and **Table S1**).

### Recursive Partitioning Analysis (RPA) of FALFF Values

In adult participants, 18 narcolepsy cases (18/26, 69%) and 21 healthy controls (21/30, 70%) were used as testing data in the recursive partitioning analysis and the developed decision tree model was shown in **Figure 3A**. In juvenile participants, 18 narcolepsy cases (18/25, 72%) and 21 healthy controls (21/30, 70%) were used as testing data in the recursive partitioning analysis and the developed decision tree model was shown in **Figure 3B**. The cut-off fALFF values of these nodes were also shown in the decision tree model (**Figure 3**). In adult participants, 8 narcolepsy cases (8/26, 31%) and 9 healthy controls (9/30, 30%) were used as validation data in the ROC analysis of developed decision tree model (**Figure 4A**). When decision tree model applied to the validation data, it revealed the sensitivity was 100%, and the specificity 88.9%. The FPR was 11.1% and the FNR was 0. Meanwhile the model showed the PPV of 88.9%, the NPV of 100%, and the accuracy of 94.1% (**Table 2**). In juvenile participants, 7 narcolepsy cases (7/25, 28%) and 9 healthy controls (9/30, 30%) were used as validation data in the ROC analysis of developed decision tree model (**Figure 4B**). When decision tree model applied to the validation data, it revealed the sensitivity was 57.1%, and the specificity 88.9%. The FPR was 11.1% and the FNR was 42.9%. Meanwhile the model showed the PPV of 80%, the NPV of 72.7% and the accuracy of 75% (**Table 2**).

### DISCUSSION

This study compared fALFF differences in both adult and juvenile narcolepsy patients with those in a group of matched healthy controls. Specially, compared with healthy controls, we identified some overlap brain regions showing significantly different fALFF values in both adult and juvenile narcolepsy patients, including bilateral medial superior frontal gyrus, bilateral sensorimotor cortex, supra-marginal gyrus, middle temporal gyrus, and bilateral inferior parietal lobule. It has been revealed that utility of ROC curve analysis in neuroimaging can distinguish one group of participants from another group of participants (13, 14). Furthermore, by using recursive partitioning analysis and ROC curve analysis, we speculated that the fALFF values in some brain regions were excellent in discriminating narcolepsy subjects from healthy controls in both adults and juvenile with high AUC value.

Low-frequency fluctuation measures are widely used for the assessment of group differences in many previous restingstate studies focusing on clinical case-control comparison (16). Furthermore, standardization has been identified effective in eliminating the dependency of fALFF values on subjective motion (16), so Z score of fALFF (i.e., standardization of fALFF) was used in the between group comparison. In both the adult and juvenile participants, narcolepsy patients showed decreased fALFF in bilateral SFGmed, bilateral supra-marginal gyrus and bilateral IPL compared with healthy controls, while narcolepsy patients showed increased fALFF in bilateral SMC and bilateral middle temporal gyrus compared with healthy controls. Both the

FIGURE 2 | Mean fALFF values differences in regional brain areas between narcolepsy and healthy controls among adults (A) and juveniles (B). SFGmed, medial superior frontal gyrus; MTG, middle temporal gyrus; SMG, supra-marginal gyrus; IPL, inferior parietal lobule; PoCG, postcentral gyrus; PreCG, precentral gyrus; PUT, putamen; THA, thalamus; SFG, superior frontal gyrus. L, left; R, right.

**79**

FIGURE 3 | RPA process results of classification and regression tree about fALFF values within two brain regions between narcolepsy and healthy controls among adults (A) and juveniles (B). As for each box, the binary value in the top represents group (0, healthy controls; 1, narcolepsy); as for the ratio in the middle, numerator means the size of patients or controls in the box and it is reflected by the top binary value, denominator means sample size in the box; the percentage in the bottom means the percentage of each box sample size in the tree. As for a brain region (or a splitting node), the cut-off fALFF value was also shown. SFGmed, medial superior frontal gyrus; SFG, superior frontal gyrus. L, left; R, right.

FIGURE 4 | The validation ROC curve of each brain region based on results from RPA analysis for distinguishing narcolepsy from healthy controls among adults (A) and juveniles (B). SFGmed, medial superior frontal gyrus; SFG, superior frontal gyrus. L, left; R, right.

medial frontal cortex, supra-marginal gyrus, and parietal lobe are abundant in hypocretin projection (30, 31), which can explain the reduced fALFF value in these regions among narcolepsy due to hypocretin deficiency, consistent with two previous positron emission tomography studies (7, 8). Increased fALFF in bilateral SMC, extending to bilateral paracentral lobule (**Figure 1**), may be a compensation of hypocretin deficiency in motor cortex among narcolepsy, although a contradictory result has been reported hypo-activity in sensorimotor cortex in narcolepsy by transcranial magnetic stimulation (TMS) in a previous study (32). Increased glucose metabolism in temporal lobe has been indicated in previous studies (7, 24), which was consistent with increased fALFF value in middle temporal gyrus in the present study result. Such increased fALFF value or hyper-metabolism in temporal lobe may be related to transient activation of this region, compensation for the hypocretin deficiency.

Meanwhile, especially in juvenile narcolepsy, higher fALFF value in right putamen and right thalamus can be detected compared with healthy controls. Putamen is a component of the salience network (24). The salience network is responsible TABLE 2 | Validation for Decision tree model about fALFF differences in brain regions between narcolepsy and healthy control.


Sn, sensitivity; Sp, specificity; FPR, false positive rate; FNR, false negative rate; PPV, positive predictive value; NPV, negative predictive value; AUC, area under the curve; SFGmed.R, right medial superior frontal gyrus; SFG.L, left superior frontal gyrus.

for integration of sensory and attention information, initiation of responses to significant stimuli as a function of top-down attention and cognitive control process (33, 34). The salience network is also thought to maintain the tonic of alertness, correlated with sympathetic regions (35, 36). Also thalamus is a core brain region responsible for sympathetic regulation, arousal, and wakefulness (35, 36). In our resting-state fMRI study, for drug-free narcolepsy patients, it requires a specific order to resist

sleepiness during the MRI scan. The increased fALFF in putamen and thalamus among juvenile narcoleptic patients reinforced its major role in the reservation of the awaking status and the activated sympathetic nervous system. The relatively increased fALFF may reflect the patients' subjective effort to maintain vigilance, consistent with already reported in obstructive sleep apnea (37), in Kleine-Levin syndrome (38) and in PET narcolepsy study (24).

Interestingly, based on the results of classification and regression tree from recursive partitioning analysis, validated ROC curve analysis indicates that in adult participants the fALFF value in right SFGmed alone could discriminate narcolepsy from healthy controls with high degree of sensitivity, specificity, and accuracy (**Figure 4A**). Also in juvenile participants, the validated ROC curve indicated that the fALFF value in left SFG alone could also discriminate narcolepsy from healthy controls with moderate degree of sensitivity, specificity and accuracy (**Figure 4B**). Although there were many brain regions showing fALFF value differences between groups, just one brain region was necessary to discriminate narcolepsy from healthy controls in adults and juveniles, respectively.

The present study has some limitations. Small sample size and single setting should be the first consideration in limitations, especially in the validation data, small sample size may lead to some bias and confounding. Also participants in this study all come from China, which may potentially be not applicable to other ethnic groups. While being fully awake during the whole examination as controlled clinically and by video, but the vigilance state was not monitored through synchronous EEG recording during the MRI scan. Our design cannot directly confirm the absence of short fluctuations in alertness and even short sleep events during the MRI process. Further simultaneous EEG-fMRI studies based on large samples are necessary to confirm our preliminary results on fALFF value differences between narcolepsy and healthy controls, also to compare narcolepsy patients with other hypersomnia and sleep deprivation in resting wakefulness.

To conclude, compared with healthy controls, both the adult and juvenile narcolepsy showed overlap brain regions in

### REFERENCES


fALFF differences after case-control comparison. Furthermore, we propose that fALFF value can be a helpful imaging biomarker in distinguishing narcolepsy from healthy controls among both adults and juveniles.

### ETHICS STATEMENT

This study was approved by the Ethical Committee of the Peking University People's Hospital.

### AUTHOR CONTRIBUTIONS

ZJ and HF designed the study. XF, LC, ZD, ZQ, and ZW carried out the study. XF performed data analysis and wrote the manuscript.

### FUNDING

This work was supported by the National Natural Science Foundation of China (81700088), National Natural Science Foundation of China (81671765), Key International (Regional) Cooperation Program of National Natural Science Foundation of China (81420108002), National Program on Key Basic Research Project of China (973 Program, 2015CB856405), and Beijing Municipal Natural Science Foundation (7172121).

### ACKNOWLEDGMENTS

We thank all the patients and healthy volunteers for their collaboration.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fneur. 2018.00936/full#supplementary-material

Table S1 | Significant differences in fALFF value between narcolepsy patients and healthy controls.

Table S2 | Original data for testing RPA and validated ROC analysis.


**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 Fulong, Chao, Dianjiang, Qihong, Wei, Jun and Fang. 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.

*Marco Hermesdorf <sup>1</sup> \*, Benedikt Sundermann2 , Rajesh Rawal1 , András Szentkirályi1 , Udo Dannlowski <sup>3</sup> and Klaus Berger1*

*<sup>1</sup> Institute of Epidemiology and Social Medicine, University of Münster, Münster, Germany, 2Department of Clinical Radiology, University Hospital Münster, Münster, Germany, 3Department of Psychiatry, University of Münster, Münster, Germany*

Objective: Previous studies on patients with restless legs syndrome (RLS) yielded inconclusive results in the magnetic resonance imaging (MRI)-based analyses of alterations of subcortical structures in the brain. The aim of this study was to compare volumes as well as shapes of subcortical structures and the hippocampus between RLS cases and controls. Additionally, the associations between the genetic risks for RLS and subcortical volumes were investigated.

#### *Edited by:*

*Kai Spiegelhalder, Universitätsklinikum Freiburg, Germany*

#### *Reviewed by:*

*Martin Gorges, Universität Ulm, Germany Katja Menzler, Philipps-Universität Marburg, Germany*

*\*Correspondence: Marco Hermesdorf hermesdorf@uni-muenster.de*

#### *Specialty section:*

*This article was submitted to Applied Neuroimaging, a section of the journal Frontiers in Neurology*

> *Received: 22 December 2017 Accepted: 01 May 2018 Published: 18 May 2018*

#### *Citation:*

*Hermesdorf M, Sundermann B, Rawal R, Szentkirályi A, Dannlowski U and Berger K (2018) Lack of Association Between Shape and Volume of Subcortical Brain Structures and Restless Legs Syndrome. Front. Neurol. 9:355. doi: 10.3389/fneur.2018.00355*

Methods: We compared volumetric as well as shape differences assessed by 3 T MRI in the caudate nucleus, hippocampus, globus pallidus, putamen, and thalamus in 39 RLS cases versus 117 controls, nested within a population-based sample. In a subsample, we explored associations between known genetic risk markers for RLS and the volumes of the subcortical structures and the hippocampus.

results: No significant differences between RLS cases and controls in subcortical and hippocampal shapes and volumes were observed. Furthermore, the genetic risk for RLS was unrelated to any alterations of subcortical and hippocampal gray matter volume.

interpretation: We conclude that neither RLS nor the genetic risk for the disease give rise to changes in hippocampal and subcortical shapes and gray matter volumes.

Keywords: restless legs syndrome, gray matter volume, subcortical brain structures, genetic risk, risk alleles

### INTRODUCTION

Restless legs syndrome (RLS) is a sensorimotor disorder affecting 2.5–10% of the general population (1). RLS is characterized by unpleasant sensations in the legs or other extremities combined with an urge to move in order to reduce the discomforting sensations. These symptoms typically worsen during periods of rest, thus having a negative impact on sleep and quality of life (2, 3). Genetic factors play an important role in RLS as it has been revealed that several single nucleotide polymorphisms (SNPs) contribute to the development of the disease (4, 5). Furthermore, RLS is believed to be a result of iron insufficiency in the brain, presumably caused by improper iron transportation across the blood–brain barrier leading to dysregulated dopaminergic neurotransmission (6).

Due to the specific role of subcortical structures in dopaminergic neurotransmission (7) and their role in iron deposition and motor function (8), these structures are of particular interest in

**83**

the search for neurobiological correlates of RLS. Previous studies employing magnetic resonance imaging (MRI) of the brain provided conflicting results regarding volumetric changes of subcortical gray matter in RLS cases. In particular, a reduction in gray matter volume has been observed in the left hippocampus (9), while others found a significant increase in left hippocampal gray matter associated with RLS (10). Increased gray matter volume in the pulvinar nuclei located inside the thalamus has also been reported (11). In contrast, several studies found no significant associations between RLS and alterations of subcortical gray matter volume (12–15). Most of these studies applied voxel-based morphometry for the detection of local changes in gray matter volume across the brain (9–14). However, specific methods have been developed to detect localized shape differences in subcortical regions and the hippocampus, considering the specific signal characteristics of these brain structures (16). Only a single study investigated such localized shape differences of the thalamus, but did not detect significant shape differences in patients with RLS versus controls (15). Localized shape differences in subcortical regions other than the thalamus have not been investigated in patients with RLS.

The present study aimed to contrast potential differences in localized shape and overall volume of several subcortical gray matter structures (caudate nucleus, globus pallidus, putamen, and thalamus) as well as the hippocampus between RLS cases and controls, all participants in the BiDirect Study. Additionally, we investigated associations between known genetic risks for RLS and potential alterations of subcortical gray matter volume, since MRI-detectable changes in these subcortical structures might be a mediator in the pathway between genotype and RLS.

### MATERIALS AND METHODS

### Participants

The ongoing BiDirect study is conducted to investigate associations between subclinical arteriosclerosis and depression. For this purpose, the BiDirect study integrates two patient cohorts, one including patients with depression, the other patients with cardiovascular disease, and one general population cohort into one project. Details on methods and design of the BiDirect Study are provided elsewhere (17, 18). Participants in the general population cohort were randomly sampled from the population register of the city of Münster, resulting in 911 individuals included in this cohort. All participants had to be in the age range from 35 to 65 years at recruitment. Informed consent was signed by all study participants in the BiDirect project, which was approved by the ethics committee of the University of Münster and the Westphalian Chamber of Physicians. Within the general population cohort, we performed a nested case-control analysis. Participants from the two patient cohorts of the BiDirect Study were thus not considered in the present analysis. Participants who did not undergo T1-weighted MRI were excluded and RLSstatus was assessed in face-to-face interviews by a set of questions that were based on the criteria established by the International RLS Study Group (19). This question set has previously been validated against a standardized neurological examination and both were in good agreement (20). In addition, a physician diagnosis of RLS in the past was assessed. Study participants who positively answered questions on all minimum criteria or reported a physician diagnosis of RLS were classified as RLS cases. In total, 11 participants had a prior physician-based diagnosis of RLS and 28 participants were screened positive by the question set. Participants without a physician diagnosis of RLS and a negative screening were classified as controls. Controls with a previously diagnosed kidney disease and/or diabetes were excluded from the analysis. Based on the group of RLS cases, controls were frequency-matched one to three by the variables age and sex. This resulted in 39 RLS cases and 117 controls for the nested case-control analysis as depicted in **Figure 1**. Furthermore, we conducted a sensitivity power analysis using G\*Power (21), revealing that we can detect substantial effects (*f* = 0.29) with a power of 95% in the shape analysis.

### Image Acquisition

Magnetic resonance imaging of the brain was performed in all BiDirect participants without contraindications. Structural 3D T1-weighted turbo field echo imaging was performed on a 3 T scanner (Intera, Philips, Best, Netherlands) to obtain 160 sagittal slices with a thickness of 2 mm (reconstructed to 1 mm), resulting in a voxel-size of 1 × 1 × 1 mm (TR = 7.26 ms, TE = 3.56 ms, 9° flip angle, matrix dimension 256 × 256, FOV = 256 × 256 mm).

### Image Preprocessing

Magnetic resonance imaging data were preprocessed using FSL (22) version 5.06. Images were linearly registered to the MNI152 template using FLIRT (23). If necessary, images were cropped or bias-field corrected with fsl\_anat1 to ensure optimal registration. The inverse transformation matrix was then applied to the predefined subcortical shape models provided by FSL. With these predefined shape models in native space, subcortical structures of interest (caudate nucleus, hippocampus, globus pallidus, putamen, and thalamus) were segmented from the participants' native space images using a Bayesian appearance model in FIRST (16) and modeled as surface meshes. In a last step, the subcortical surfaces were aligned to a sample-specific mean shape of the respective surface structures applying a 6 degree of freedom transformation whereby differences in rotation and translation were removed.

For the purpose of a volume-based analysis, subcortical structures were boundary corrected and the respective volumes of interest were extracted. In order to adjust for the overall brain volume in the course of analyses, brain volume (i.e., gray and white matter) was estimated by partial volume estimation in FAST (24).

### Genotyping

Genotyping was conducted using the Illumina PsychChip array (Illumina, San Diego, CA, USA). Several SNPs in *MEIS1*, *BTBD9*, *MAP2K5*, *PTPRD*, and *TOX3/BC034767* (4, 25–27) have previously been associated with RLS and were selected for the study at hand. Imputation was performed using IMPUTE version 2.3.2 (28). SNPs being in linkage disequilibrium (*R*<sup>2</sup> ≥ 0.8) or with a

<sup>1</sup>http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/fsl\_anat (Accessed: 2017).

minor allele frequency below 5% were excluded from the analysis. Statistics regarding linkage disequilibrium were derived from the database of the Broad Institute.2

### Comorbidities

We assessed comorbidities using data from face-to-face interviews as well as laboratory data. Previous physician-based diagnoses of stroke, myocardial infarction, and cancer were collected by self-report. Participants were classified as hypertensive

<sup>2</sup>https://archive.broadinstitute.org/mpg/snap (Accessed: 2017).

if the mean of the second and third blood pressure readings for systolic blood pressure was ≥140 mm Hg or the diastolic blood pressure exceeded 89 mm Hg. Furthermore, participants with a self-reported physician-based diagnosis of hypertension in combination with use of antihypertensive medication according to the Anatomical Therapeutic Chemical (ATC) Classification System (ATC C02A, C02D, C02L, C03, C07, C08, C09) were also defined as having hypertension. Depression was assessed as a previous physician-based diagnosis *via* self-report or if participants scored ≥16 points on the Center for Epidemiologic Studies Depression Scale (29). Body size and weight were assessed and participants with a body mass index larger than 30 kg *m*<sup>2</sup> were classified as obese. The presence of thyroid disease was assessed by self-report of a physician-based diagnosis or intake of relevant medication (ATC H03). Thyroid-stimulating hormone (TSH) and free thyroxine (fT4) levels were used to estimate hypothyroidism (TSH > 4.8 <sup>µ</sup>IU mL and fT4< 13 pmol L ) as well as hyperthyroidism (TSH < 0.3 <sup>µ</sup>IU mL and fT4 <sup>&</sup>gt; 23 pmol L ) and participants in either category were also defined as having thyroid disease. Migraine was assessed as physician-based diagnosis *via* self-report or current use of relevant medication (ATC N02CA, N02CC). A comorbidity index was calculated by summing up the presence of the previously described conditions (stroke, myocardial infarction, cancer, hypertension, depression, obesity, thyroid

### Statistical Analysis

Participants with RLS and controls were compared on orthogonal displacements at each vertex regarding the samplespecific mean surfaces of the subcortical structures of interest. These analyses were conducted with a cluster-based *F*-test implemented in FSL randomize (31) with 5,000 permutations. Statistical threshold for significance was set to *p* < 0.05. Extracted volumes of the subcortical structures were compared across groups by several analyses of covariance (ANCOVAs) while adjusting for overall brain volume. The obtained *p*-values were corrected for false discovery rate (FDR) following the Benjamini–Hochberg procedure (32).

disease, and migraine). A similar index of cumulative disease burden has been used previously in the context of RLS (30).

Genotyping data were available for 137 participants. For each participant, the number of risk alleles per SNP was noted. For each respective SNP, a logistic regression was conducted with RLS as dependent variable and risk allele frequency as predictor along with age and sex as covariates of no interest. A weighted genetic risk score (GRS) was calculated for each SNP by multiplying the risk allele frequency of the respective SNP with the odds ratio obtained by the logistic regression. Each respective GRS was used as a predictor in multiple regression analyses with the subcortical brain volumes as dependent variables while adjusting for age, sex, and overall brain volume. The analyses of extracted subcortical volumes and genotyping data were conducted in SPSS version 22 (IBM, Armonk, NY, USA).

### RESULTS

### Subject Demographics

A comparison of group characteristics is summarized in **Table 1**. Participants with RLS and controls did not differ in terms of age and sex. Distributions of comorbidity load were significantly different across groups and the median comorbidity load was higher in RLS cases. Genotyping data were available for 137 (87.8%) participants. The remaining 19 participants were thus not considered for the analyses of genotyping data.

### Shape Analyses

The comparisons of the shapes in the caudate nucleus, hippocampus, globus pallidus, putamen, and thalamus across groups did not yield significant differences in either hemisphere. The subcortical and hippocampal shapes of the sample are shown in **Figure 2**. The analyses of extracted volumetric data did not reveal significant group differences after FDR correction (**Table 2**).

Table 1 | Comparison of demographic characteristics across groups.


*RLS, restless legs syndrome; IQR, interquartile range.*

*a One participant had incomplete comorbidity data and two participants had missing data on disease duration.*

*Age was compared with a t-test. Distribution of sex was compared with a* χ*<sup>2</sup> test and comorbidity load with a Mann–Whitney U-test.*

Figure 2 | Average shapes of the rigidly aligned subcortical structures in the sample.

### Associations Between Risk allele Frequency and RLS

The logistic regressions yielded significant associations between RLS and the risk allele (G) frequency in rs11635424, located in *MAP2K5*. Due to the small sample size, no significant associations were found for the remaining eight SNPs. However, the magnitude of the odds ratios indicated a higher risk for RLS concerning the majority of the SNPs (**Table 3**), in line with prior reports (4, 25, 27, 33).


*RLS, restless legs syndrome; FDR-sig, false discovery rate corrected significance at q* < *0.05; n.s., not significant.*

*a The group comparison is adjusted for overall brain volume.*

Table 3 | Associations between allele frequency and RLS.a


*RLS, restless legs syndrome.*

*a The logistic regressions are adjusted for age and sex.*


Table 4 | Regression coefficients for the GRS per subcortical region.

*GRS, genetic risk score.*

*The regression analyses were adjusted for age, sex, and overall brain volume.*

### Genetic Risk for RLS and Subcortical Volumes

No significant associations between the odds ratio weighted GRS derived from the respective SNPs and subcortical as well as hippocampal volumes were found. The regression coefficients are presented in **Table 4**.

### DISCUSSION

In this nested case-control study, we examined potential alterations in shape and volume of subcortical structures and the hippocampus in cases with RLS versus controls. While potential volumetric alterations of subcortical structures and the hippocampus have been investigated previously using VBM (9–15), shape differences in the caudate nucleus, hippocampus, globus pallidus, and putamen have not been compared before between cases with RLS and controls. Analyzing shape differences, however, is important in order to determine the exact locations where potential anatomical changes in subcortical structures occur. Knowledge of localized shape differences may also aid the interpretation of the relationship with other anatomical findings, e.g., when localized changes in thalamic shape are associated with adjacent reductions of white matter volume (34). Our analyses revealed no group differences in either shape or volume of the caudate nucleus, hippocampus, globus pallidus, putamen, and thalamus. The lack of volume differences supports previous findings (12–15), suggesting that RLS is not accompanied by any changes of subcortical gray matter. Instead, it seems more likely that alterations of the dopaminergic system (6), possibly induced by genes involved in neurodevelopment [*MEIS1* (35, 36) and *TOX3* (37)], protection of dopaminergic neurons [*MAP2K5* (38)], sleep disturbances [*BTBD9* (39)], modulation of dopaminergic neurotransmission [*PTPRD* (40)], and iron regulation within the brain [*BTBD9* (41)], may lead to changes in functional brain networks. In particular, increased functional connectivity has been reported in sensory-thalamic, basal ganglia-thalamic, and other cortical and subcortical networks in patients with RLS, whereas symptom severity correlated with increased network connectivity (42). Hence, in the absence of gray matter alterations, RLS is more likely to be characterized by inefficient network performance.

Although RLS has previously been associated with several SNPs within regions of the above-mentioned genes (4), the exact mechanisms how these SNPs contribute to the development of RLS are still unknown. Hence, we also explored potential associations between known genetic risk markers for RLS and alterations of subcortical volumes to evaluate if these are a potential mediator of the genotype-disease association. Only SNP rs11635424 was significantly associated with RLS. While most of the remaining RLS-related SNPs indicated risks, i.e., odds ratios larger than 1 for the risk alleles, these associations did not reach statistical significance given the small sample size in our study. The magnitude of effect sizes is largely in line with previous studies (4, 25, 27, 33), suggesting that larger samples are advantageous to detect effects of allele frequency in the context of RLS. With regards to the volume of the subcortical structures and the hippocampus, we did not find a significant association with SNP rs11635424 or any of the other eight SNPs, suggesting that RLS-related variations in the genome do not play an important part in the volumetric appearance of subcortical structures and the hippocampus.

The present study is limited by its sample size which is rather small regarding the search for genetic factors contributing to the development of RLS. However, the primary aim was to compare subcortical as well as hippocampal shapes and volumes between RLS cases and controls and to analyze the influence of the odds ratio weighted genetic risk for RLS on subcortical and hippocampal volumes. Within the field of 3 T MRI-literature, the present study is the largest investigating potential volumetric alterations in RLS cases versus controls.

We conclude that RLS is unrelated to changes in shape and volume of the caudate nucleus, hippocampus, globus pallidus, putamen, and thalamus. The SNP rs11635424 was significantly

### REFERENCES


associated with RLS in our sample. The odds ratio weighted GRS from each of the nine SNPs as well as a summed GRS do not account for any volume alterations of subcortical gray matter.

### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the ethics committee of the University of Münster and the Westphalian Chamber of Physicians with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the University of Münster and the Westphalian Chamber of Physicians.

### AUTHOR CONTRIBUTIONS

Design and concept of the BiDirect Study: KB. Preprocessing and data analyses: MH, RR, and AS. Drafting the manuscript and figures: MH. Technical assistance and commenting on the preprocessing of imaging data: BS and UD.

### ACKNOWLEDGMENTS

The present work was supported by a grant to KB from the German Federal Ministry of Education and Research (BMBF; grant FKZ-01ER0816 and FKZ-01ER1506). UD was funded by the German Research Foundation (DFG, grant FOR2107 DA1151/5-1; SFB-TRR58, Project C09) and the Interdisciplinary Center for Clinical Research (IZKF) of the medical faculty of Münster (grant Dan3/012/17).


arteriosclerosis – rationale, design, and characteristics of the BiDirect Study. *BMC Psychiatry* (2014) 14:174. doi:10.1186/1471-244X-14-174


**Conflict of Interest Statement:** MH, BS, RR, AS, and UD have no conflicts of interest to disclose. For the conduction (2007–2014) of a study on the course of restless legs syndrome, KB has received unrestricted grants to the University of Münster from the German Restless Legs Society and Boehringer Ingelheim Pharma, Mundipharma Research, Neurobiotec, Roche Pharma, UCB Germany, UCB Switzerland, and Vifor Pharma.

*Copyright © 2018 Hermesdorf, Sundermann, Rawal, Szentkirályi, Dannlowski and Berger. 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 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.*

# Imaging Individual Differences in the Response of the Human Suprachiasmatic Area to Light

Elise M. McGlashan<sup>1</sup> , Govinda R. Poudel 1,2,3, Parisa Vidafar <sup>1</sup> , Sean P. A. Drummond<sup>1</sup> and Sean W. Cain<sup>1</sup> \*

<sup>1</sup> Monash Institute of Cognitive and Clinical Neurosciences and School of Psychological Sciences, Monash University, Melbourne, VIC, Australia, <sup>2</sup> Sydney Imaging, The University of Sydney, Camperdown, NSW, Australia, <sup>3</sup> Mary Mackillop Institute of Health Research, Australian Catholic University, Melbourne, VIC, Australia

Circadian disruption is associated with poor health outcomes, including sleep and mood disorders. The suprachiasmatic nucleus (SCN) of the anterior hypothalamus acts as the master biological clock in mammals, regulating circadian rhythms throughout the body. The clock is synchronized to the day/night cycle via retinal light exposure. The BOLD-fMRI response of the human suprachiasmatic area to light has been shown to be greater in the night than in the day, consistent with the known sensitivity of the clock to light at night. Whether the BOLD-fMRI response of the human suprachiasmatic area to light is related to a functional outcome has not been demonstrated. In a pilot study (n = 10), we investigated suprachiasmatic area activation in response to light in a 30 s block-paradigm of lights on (100 lux) and lights off (<1 lux) using the BOLDfMRI response, compared to each participant's melatonin suppression response to moderate indoor light (100 lux). We found a significant correlation between activation in the suprachiasmatic area in response to light in the scanner and melatonin suppression, with increased melatonin suppression being associated with increased suprachiasmatic area activation in response to the same light level. These preliminary findings are a first step toward using imaging techniques to measure individual differences in circadian light sensitivity, a measure that may have clinical relevance in understanding vulnerability in disorders that are influenced by circadian disruption.

#### Edited by:

Kai Spiegelhalder, Klinik für Psychiatrie und Psychotherapie, Universitätsklinikum Freiburg, Germany

#### Reviewed by:

Pablo Torterolo, Universidad de la República, Uruguay Chiara Baglioni, University Hospital Freiburg, Germany

> \*Correspondence: Sean W. Cain sean.cain@monash.edu

#### Specialty section:

This article was submitted to Sleep and Chronobiology, a section of the journal Frontiers in Neurology

Received: 27 July 2018 Accepted: 13 November 2018 Published: 29 November 2018

#### Citation:

McGlashan EM, Poudel GR, Vidafar P, Drummond SPA and Cain SW (2018) Imaging Individual Differences in the Response of the Human Suprachiasmatic Area to Light. Front. Neurol. 9:1022. doi: 10.3389/fneur.2018.01022 Keywords: melatonin suppression, light sensitivity, circadian rhythms, light exposure, BOLD-fMRI

## INTRODUCTION

The human circadian system is responsible for regulating physiological processes across the 24 h day. This includes rhythms in alertness, sleep-wake behavior, metabolism, mood and cognitive function (1–3). The endogenous master clock (the suprachiasmatic nucleus, SCN) generates rhythms of ∼24 h, which are synchronized to the environmental light/dark cycle via retinal light exposure (4).

Disrupting the relationship between the light-dark cycle, behavior and internal rhythms has significant consequences for health. Circadian disruption is a factor in the etiology of mood disorders (5), cognitive decline (6), the onset of metabolic diseases such as diabetes (3, 7), cardiovascular health (8), and is associated with an increased risk for cancer (9). Although

**90**

these health concerns may arise from the uncoupling of rhythms with behavior (e.g., cross-meridian travel, engaging in shift-work), it has also been suggested that an abnormal response to environmental light may lead to the development of circadian disruption in the absence of, or in combination with, behavioral change (10, 11). Both hyper- and hypo- sensitivity to environmental light could lead to the development of abnormal circadian synchronization (10–12). Therefore, an abnormal response of the circadian system to light is a potentially important factor for disease vulnerability.

Better characterization of the function of the SCN (master circadian clock) in response to light cues may provide clinically relevant information, leading to improved interventions. However, our understanding of human SCN function in a clinical context to date has often relied on peripheral measurements of clock function. For example, the most common assessments of SCN function involve measuring the timing of melatonin onset (usually via dim-light melatonin onset; DLMO) for circadian timing [e.g., (13, 14)], and melatonin suppression to assess circadian light responsiveness [e.g., (11, 15)]. However, for patients taking beta-blockers, antidepressants, or sleeping aids such as exogenous melatonin, these assessments will be uninformative due to the pharmacological impact on endogenous melatonin levels, or cross reactivity with existing assays (16, 17). The ability to directly assess the activity of the SCN in response to light cues would overcome these limitations.

There is a substantial neuroimaging literature examining nonvisual light responses in humans. For example, the BOLDfMRI response of the suprachiasmatic area to light during the day, evening, and night has been imaged, showing differential activation across times of day which matches the known rhythm in the responsiveness of the circadian system to light (18). Studies have also shown enhancement of activity in brain areas associated with working memory, alertness and cognition [e.g., (19, 20)] and emotional processing (21) in response to blue light, compared to green. Further, the use of light stimuli which differentially stimulate melanopsin (high- or low-stimulation) during fMRI has been utilized to characterize the cerebral activation associated with non-visual light processes (22). However, the measurement of suprachiasmatic area function in humans has yet to be related to individual responsiveness using established laboratory techniques. In this study we examined, within individuals, the relationship between suprachiasmatic area activation in response to light in an fMRI scanner and melatonin suppression to light in the laboratory. We hypothesized increased activation of the suprachiasmatic area in response to light would be associated with increased melatonin suppression to light.

### MATERIALS AND METHODS

### Participants

Ten healthy young men and women (5 men, Mage = 20.80, SD = 1.87) were recruited. Participants were free of medical and psychiatric conditions and were not taking any medications at the time of the study. Women were naturally cycling (i.e., not using any hormonal contraception).

### In-laboratory Circadian Assessments

All participants completed an in-laboratory assessment of circadian light sensitivity. This involved an assessment of dimlight melatonin levels and a subsequent light exposure of ∼100 lux. Sessions ran from ∼4 h prior to the participants' bedtime, until 1 h after, during which the participant remained awake and seated (other than for bathroom breaks). These two sessions were a minimum of 1 week apart, with the dim-light session occurring first. Participants maintained a strict 8:16 h sleep-wake dark-light schedule for at least 1 week prior to, and in between sessions, whereby >1 deviation of more than 30-min in 1 week would be exclusionary. Adherence to the schedule was monitored using wrist-worn actigraphy (Actiwatch Spectrum Plus or L, Philips Respironics, OR, USA) and sleep diaries. Schedules were selected to be in line with participants typical sleep-wake behavior, an example schedule, with an overview of the protocol is available in the **Supplementary Material**. During test-sessions, hourly saliva samples were taken using salivettes (Sarstedt, Germany), which were then assayed in duplicate for melatonin at the Adelaide Research Assay Facility using radioimmunoassay with the G280 antibody and the [1251]2-iodomelatonin radioligand (LOD 4.3 pMol).

### In MRI Light Exposure and Imaging Procedure

Participants completed an fMRI scan beginning ∼1 h prior to habitual bedtime. For 1 h prior to this they were seated in dimlighting conditions of <10 lux. Prior to their scan, participants provided a urine sample for toxicology to be conducted, with a positive result being exclusionary (n = 0, SureStep 6 Panel, Medvet, South Australia, Australia).

All subjects were imaged using a 3T Scanner (Siemens Magnetom Skyra) with 20 channel head coils. High-resolution anatomical images of the whole brain were acquired using T1 weighted anatomical scans (TE = 2.07 ms; TR = 2.3 s; field of view: 256 × 256 mm; slice thickness: 1 mm). Functional images were acquired using echo-planar-imaging (TR: 2.06 s; TE: 24 ms; field of view: 190 × 190 mm; slice thickness: 3 mm; number of slices: 41; flip angle = 90, number of volumes = 177). The first five images of each session were discarded to allow for T1 equilibration.

Participants were requested to lay supine in the MRI scanner, while an optic-fiber-based light delivery system was fitted on the MRI head coil. This device consisted of a halogen light source (DC950H, Dolan-Jenner Industries, MA, USA), which transmitted light through metal-free fiber optic cables (100 strand cable with 0.75 mm fibers, Optic Fiber Lighting, Sydney, AU) to two circular plastic diffusers (40 mm diameter) positioned ∼50 mm above each eye. The diffusers were designed to bathe each eye in light, achieving an even spread of illumination. Light stimuli had a CCT of ∼2800 K (λp = 650 nm), and was delivered at two intensities, ∼100 lux (42.73 µW/cm<sup>2</sup> ) and ∼1000 lux (392.28 µW/cm<sup>2</sup> ).

Each participant was exposed to a passive light stimulus paradigm in which they were requested to keep their eyes open (other than normal blinking). This was comprised of alternating periods of lights off (darkness, six 30 s epochs) and lights on at a moderate level (100 lux, six 30 s epochs) or bright level

(1000 lux, six 30 s epochs). Moderate and bright blocks (of 6 min total duration each) were delivered separately, with the moderate light exposure block always being presented first. Due to the aversive nature of the 1000 lux bright-light stimuli (which often led to significant eye closures), only data for the moderate light exposures are reported here.

### Data Analysis

### Melatonin Suppression

Area under the curve (AUC) was calculated for the final 2 h of each dim-light control, and each 100-lux light exposure (where melatonin levels were adequate in all participants in our protocol). Average percent suppression across the 2 h was then calculated by determining the percent change in AUC from baseline to the 100-lux light exposure for each individual.

### MRI Data Processing

Detailed information regarding fMRI data processing and analysis can be found in the **Supplementary Material**. Briefly, MRI data were pre-processed using FSL (FMRIB's Software Library, www.fmrib.ox.ac.uk/fsl). For each participant, preprocessed fMRI data were analyzed using first-level general linear models. The linear models included regressors for light on blocks and standard motion parameters (six regressors). To focus our analysis on the suprachiasmatic area of the brain, we generated a mask covering hypothalamic area using a meta-analytic tool NeuroSynth (http://neurosynth. org/analyses/terms/hypothalamus/). This mask (see **Figure 1**) covered both the anterior and posterior hypothalamus including the suprachiasmatic area.

### Statistical Analyses

A correlational analysis was used to assess the relationship between suprachiasmatic area function in response to light (100 lux relative to dark periods) and melatonin suppression. A Spearman's correlation was used due to the small sample size and potential non-normality of the BOLD response.

### RESULTS

There was a significant, strong positive correlation between suprachiasmatic area activity during light exposure periods (relative to dark) and melatonin suppression (**Figure 2**). Increased suprachiasmatic area activation was associated with an increase in melatonin suppression (i.e., greater circadian light sensitivity).

FIGURE 2 | Relationship between the fMRI BOLD response in the suprachiasmatic area (SCA) during 100 lux light exposures (relative to dark), and melatonin suppression.

### DISCUSSION

This study provides preliminary evidence for a relationship between suprachiasmatic area activation in response to light and an established in-laboratory measure of circadian light sensitivity. We found a significant relationship between suprachiasmatic area activation and melatonin suppression, indicating that an increase in fMRI measured suprachiasmatic area activation in response to light related to an increase in circadian light sensitivity. Thus, these are the first data in humans to show a relationship between a proximal measure of activity in the anterior hypothalamus and a functional outcome.

An increase in melatonin suppression relates to larger shifts in circadian phase (23), and has been associated with disease states (10, 11). Our results suggest that increased melatonin suppression findings may reflect increased activation of the SCN in response to environmental light. Light information is received at the retina by intrinsically photosensitive retinal ganglion cells (iPRGCs), which then project to the SCN via the retinohypothalamic tract (RHT), and to other brain areas (24, 25). Light exposure leads to changes in circadian timing, amplitude, levels of alertness and mood (23, 26, 27). The magnitude of the impact of this light on the circadian system will be partly dependent on individual differences in light sensitivity, and our results demonstrate that this interindividual variability may arise from functional differences in the ability of retinal light exposure to activate the SCN.

Circadian dysfunction has been associated with several chronic disease states, including mood disorders (10, 28), metabolic and cardiovascular disease (29) and sleep disorders (11, 14, 30). Abnormalities in circadian light sensitivity may be a trait vulnerability for mood disorders with variable or decreased sensitivity being observed in seasonal affective disorder (12), while hypersensitivity to light has been observed in bipolar disorder (10, 28), and in some sleep disorders or disturbances (11, 31). Imaging of the response to moderate light as used in this study may reveal abnormal SCN function, which could lead to circadian dysfunction.

It should be noted that although a significant relationship was observed here between suprachiasmatic area activation and melatonin suppression, our sample was small, and these data do not indicate that an individual scan of the response to light can currently replace melatonin suppression as an indicator of circadian light sensitivity. The BOLD fMRI response to light in the suprachiasmatic area may instead prove a useful clinical tool for studying changes in light sensitivity associated with either a clinical diagnosis, or pharmacological intervention. Given suggestions that light sensitivity can change across a disease course (12), and may mediate treatment response in mood disorders (32, 33), this has important clinical implications. However, further characterization of the relationship between suprachiasmatic area activation and melatonin suppression is required in order to establish clinically meaningful ways of interpreting individual data.

This study has shown, in a small sample, evidence for a relationship between suprachiasmatic area BOLD-fMRI activation to light and an established measure of circadian light sensitivity. This is a first step in the development of imaging techniques for the assessment of individual differences in circadian function. This is critical given the pervasive nature of circadian dysfunction in disease states.

### ETHICS AND DATA AVAILABILITY STATEMENT

All procedures were approved by the Monash University Human Research Ethics Committee (MUHREC) prior to commencement (Project 4760). Participants gave written, informed consent prior to participation and were reimbursed for their time. The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

### REFERENCES


### AUTHOR CONTRIBUTIONS

The study was conceived by SC. All authors contributed to the study design. GP was responsible for programing the light delivery device and MRI sequences, and completing fMRI data analysis. EM and PV were responsible for recruitment, data collection, and melatonin data processing. EM was responsible for the final data analysis and writing the manuscript. All authors reviewed and contributed to the manuscript prior to submission for publication.

### FUNDING

This project was supported by a project grant from the National Health and Medical Research Council (NHMRC) awarded to SC (Project 1064231), and a Platform Access Grant (PAG) from the faculty of Medicine Nursing and Health Sciences at Monash University awarded to SC, SD and GP (15-0038). EM was supported by an Australian Government Research and Training Program (RTP) scholarship. PV was supported by a scholarship from the National Health and Medical Research Council (NHMRC), via the NeuroSleep Centre for Research Excellence.

### ACKNOWLEDGMENTS

We would like to thank Gilles Vandewalle for his advice and consultation regarding the build of our custom light device, and the Monash Instrumentation Facility for their assistance designing and constructing the device. We would also like to acknowledge the radiographers and staff of the Monash Biomedical Imaging center (MBI) at Monash University for the use of their facilities, and assistance in running our studies out of business hours–in particular Richard McIntyre. Lastly, we would like to thank the staff and students of the Monash University Sleep and Circadian Medicine Laboratory for their assistance in the running of in-laboratory circadian assessments.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fneur. 2018.01022/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 © 2018 McGlashan, Poudel, Vidafar, Drummond and Cain. 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.