Selective REM-Sleep Suppression Increases Next-Day Negative Affect and Amygdala Responses to Social Exclusion

Healthy sleep, positive general affect, and the ability to regulate emotional experiences are fundamental for well-being. In contrast, various mental disorders are associated with altered rapid eye movement (REM) sleep, negative affect, and diminished emotion regulation abilities. However, the neural processes mediating the relationship between these different phenomena are still not fully understood. In the present study of 42 healthy volunteers, we investigated the effects of selective REM sleep suppression (REMS) on general affect, as well as on feelings of social exclusion, emotion regulation, and their neural underpinnings. Using functional magnetic resonance imaging we show that REMS increases amygdala responses to experimental social exclusion, as well as negative affect on the morning following sleep deprivation. There was no evidence that emotional responses to experimentally induced social exclusion or their regulation using cognitive reappraisal were impacted by diminished REM sleep. Our findings indicate that general affect and amygdala activity depend on REM sleep, while specific emotional experiences possibly rely on additional psychological processes and neural systems that are less readily influenced by REMS.


Introduction
5 emotional processing, including emotional reactivity [22,23] and the formation of emotional memories [24,25] . 86 REM-sleep dreaming was also found to attenuate residual emotional load from the day before [26,27] . 87 However, only very few studies have examined the effects of selective suppression of REM sleep during 88 an otherwise normal night of sleep. The existing evidence suggests that the selective deprivation of REM 89 sleep mainly disturbs the consolidation of emotional memories [28][29][30] , whereas the selective suppression 90 of SWS mainly impairs emotionally-neutral declarative memory encoding and consolidation [31,32] . These 91 experimental findings are in line with clinical observations suggesting that REM sleep in particular is 92 closely tied with emotional functioning [19] . 93 In order to better understand the processes underlying the interaction of sleep and emotion, 94 research increasingly addresses the neural mechanisms associated with REM sleep-dependent emotional 95 functioning. At the neural level, a reduction of emotional reactivity in response to previously learned 96 emotional events was found to be accompanied by an overnight decrease of amygdala reactivity, a 97 cluster of nuclei in the temporal lobes and part of the limbic system [27,33] . The limbic system has been 98 widely associated with the processing of affectively laden stimuli and plays an important role in the 99 guidance of behavioral responses to such stimuli [34,35] . Under conditions of total sleep deprivation, next-100 day negative affect is accompanied by increased amygdala reactivity and decreased functional coupling 101 of the medial prefrontal cortex (MPFC) with limbic structures [36] . Since the MPFC is thought to exert 102 inhibitory control on the amygdala [37] , this finding has been interpreted to reflect a failure of top-down 103 control in the regulation of appropriate emotional responsivity [36] . Simon and colleagues directly 104 examined the effect of sleep deprivation on the neural correlates of next day emotion reactivity [17] . 105 Following sleep deprivation, there was no indication for valence-specific processing of affective pictures 106 in the amygdala, however in the sleep-rested control night, a low amount of REM sleep was associated 107 with a decline in anterior cingulate cortex (ACC)-amygdala connectivity, possibly reflecting a specific 108 effect of REM sleep on cognitive control of emotions. 109 Successful cognitive control of emotions is regarded to be an essential prerequisite of mental 110 health [38,39] . In daily life, emotions are constantly regulated either implicitly or explicitly by applying 111 specific cognitive strategies like suppression (e.g. distracting the attention away from unpleasant 112 emotional experiences) or reappraisal (e.g. reinterpreting an unpleasant emotional situation). Emotion 113 regulation thus refers to the ability to "influence which emotions we have, when we have them, and how 114 we experience and express these emotions" (p.497 [40] ). Interestingly, correlational evidence indicates 115 that the success of emotion regulation is associated with sleep quality [41] . In this study, participants were 116 asked to engage in cognitive reappraisal (CRA) [42] , in this case applying a previously learned cognitive 117 strategy to "redirect the spontaneous flow of emotions" (p. 6 [43] ), while watching a sadness-inducing 118 film. The ability to decrease self-reported sadness using CRA compared to baseline was lower in 119 participants who reported poorer sleep quality during the preceding week [41] . 120 Despite their substantial clinical significance, the neural mechanisms of the effect of REM sleep 121 on the efficacy of regulation strategies in ameliorating unpleasant affect remain unknown. To fill this 122 gap, it is important to experimentally induce negative affect that may then be attenuated by applying 123 CRA strategies. In order to induce negative affect in the present study, we simulated social exclusion in a 124 laboratory setting using the so-called Cyberball [44] . Cyberball is a virtual ball-tossing paradigm, where 125 participants are playing with a preset computer program while believing that they are playing with two 126 other human participants. By manipulating the number of ball-tosses towards the participant, the degree 127 of social inclusion can be controlled experimentally. Evidence suggests that the distressing experience of 128 social exclusion might share neural regions and mechanisms with the affective processing of physical 129 pain [45][46][47] . In a neuroimaging study using the Cyberball game, participants showed greater activations in 130 cingulate and prefrontal regions involved in the processing of the affective pain component when 131 excluded compared to being included [48] . Although behavioral consequences and neural activation 132 patterns associated with social exclusion have been studied quite intensively, there is scarce research on 133 intervening cognitive appraisals and coping mechanisms regarding feelings of social exclusion [49] . 134 The aim of the present study was to examine how REM sleep suppression impacts the following 135 day general affect, emotional reactivity, and associated neural mechanisms of emotion regulation during 136 the acute experience of social exclusion. In a between-subjects design we invited participants to a 137 combined polysomnography and fMRI study. After a habituation night allowing regular sleep, for the 138 second, experimental night participants were randomly allocated to either a REM sleep suppression 139 (REMS) group or one of two control groups: a non-suppression control group with regular sleep (CTL) and 140 a high-level control group with similar amounts of awakenings, but where suppression targeted phases 141 of slow wave sleep (SWSS). To assess the impact of REMS on general affect, subjects repeatedly filled in 142 the Positive and Negative Affect Schedule (PANAS) [50] . In the morning after the experimental night, 143 subjects participated in the Cyberball during fMRI scanning to induce feelings of social exclusion. All 144 participants engaged in two sessions of the game. In the first session, participants played the game 145 without any instructions. In the second session, participants were instructed to actively regulate their 146 emotions by applying the previously learned CRA. 147 We hypothesized that selective REMS (vs. SWSS and regular sleep) generally reduces positive 148 and increases negative affect. Furthermore, we expected that selective REMS (vs. SWSS and regular 149 sleep) increases emotional reactivity during social exclusion and dampens the effect of CRA on emotional 150 reactivity. On the level of neural systems we explored whether REMS leads to altered functional activity 151 in (para-)limbic areas such as the amygdala, ACC, insula, and hippocampus during social exclusion and 152 whether neural activity of these regions is modulated by targeted REMS during cognitive reappraisal. 153 8 154 Figure 1.

Experimental Sleep Manipulation Selectively Reduces REM Sleep Percentage 156
The experimental sleep manipulation (see figure 1)   T=4.88, k=4, p=.022, FWE-corrected) as well as the right insula (34,-10,22; T=5.07, k=13, p=.014, FWE-203 corrected) was significantly increased compared to the inclusion condition [51,52] . In addition, across 204 groups, neural activity in the right anterior insula was significantly increased during VIEW as compared to 205 CRA blocks (main effect VIEW>CRA: 30,24,-4; T=4.58, k=1, p=.048, FWE-corrected). Testing whether 206 these two main effects interacted or whether they were modulated by type of sleep suppression did not 207 Nearly everyone can relate to the devastating effects of a sleepless or interrupted night on one's next 223 day mood. While the effect of total sleep deprivation on emotional reactivity has been investigated 224 intensively in the past [53,54] the present study focused on the specific impact of selective REM sleep 225 suppression on general affect, as well as emotion regulation and its neural correlates under conditions of 226 social exclusion. 227 We found that lower amounts of REM sleep across all participants were associated with higher 228 levels of general negative affect in the next morning, a finding that is in line with previous literature 229 implicating REM sleep in emotional functioning [55] . Despite this general effect, however, our findings do 230 not provide evidence for a direct link of REMS with the subjective emotional response to experimentally 231 induced social exclusion. The ability to regulate one's negative emotions during social exclusion was also 232 not affected by prior REMS, which was an unexpected finding. Interestingly though, despite no changes 233 in subjectively reported emotions, neural activity in the limbic system was altered after REMS when 234 participants experienced social exclusion. Precisely, neural responses to passively experienced ostracism 235 were specifically increased in right amygdala after REM sleep was selectively suppressed. The amygdala 236 is strongly associated with emotional processing [56,57] and has anatomical connections to the anterior 237 insula and the anterior cingulate cortex [58,59] . As part of this so-called salience network [60] , the 238 amygdala's assumed function of signaling the relevance of information is central for the domain of 239 affective experiences [57,61] . Previous studies showed that amygdala responses to viewing negative 240 emotional stimuli increased after total sleep deprivation [62] , depended on intact REM sleep in particular 241 [33] , and correlated with autonomic responses to psychosocial stress [63] . Our study connects these 242 previous findings by demonstrating that amygdala activity tracks information relevant to the subjects' 243 social well-being in dependence of REM sleep. 244 We can only speculate why these alterations in brain function after REMS did not manifest on 245 the behavioral level in form of increased emotional reactivity to the experience of social exclusion. 246 Previous research provides several explanations that might be helpful to understand this dissociation. 247 First, the effects of experimental short-term sleep manipulations might be strongest immediately after 248 awakening and may be readily washed out thereafter [64] , and may thus not have lasted until the fMRI 249 session in the present study. This may hold in particular for selective suppression of specific sleep stages 250 rather than total sleep deprivation, which produces stronger and more long-lasting effects [64] . The 251 efficacy of REMS in the present study may have been too small to surface on the behavioral level later in 252 the morning. However, the effects of REMS nonetheless persisted on the level of brain systems. The 253 altered brain activity in the amygdala could indicate that even small changes in REM sleep can influence 254 brain systems implicated in regulating responses to affectively salient stimuli [56,57] , while they are too 255 small to penetrate the level of subjective experiences, which are possibly processed further downstream 256 [63,65] . Yet, what speaks against this explanation are findings by Wiesner and colleagues [30] or 257 Morgenthaler and colleagues [66] who applied even more rigorous REM sleep deprivation, achieving a 258 mean REM sleep percentage of around one percent of TST, but nevertheless could not find REM sleep 259 related behavioral effects during emotion recognition tasks. Similarly, Liu and colleagues compared the 260 effect of a 24h sleep deprivation to regular sleep on the experience of distress following social exclusion 261 in the Cyberball game [67] . As in the present study, the authors did not find an effect of sleep deprivation 262 on the subjective experience of social rejection. 263 Regardless of the timing and potential washout during the day, the lack of significant findings for 264 the experience of social exclusion might also relate to the conceptual breadth and the ecological validity 265 of the affective assessment. While the underlying brain systems show increased reactivity towards 266 potentially threatening conditions and signal greater homeostatic imbalance, the distinct assessment of 267 emotional responses to ostracism, such as feelings of exclusion, might touch a different facet of affective 268 construal. That is, the affective salience of information tracked by amygdala activity [57,61] may be 269 modulated by REMS. However, the construal of emotional experience is assumed not to rely on the 270 activity in single regions, but on the dynamic interaction of various neural systems supporting multiple 271 psychological processes of emotional experience apart from affective salience [68,69] . Hence, amygdala 272 responses to psychosocial stress do not necessarily influence the construal of subjective emotion ratings, 273 that may depend on additional networks involving prefrontal cortical regions, as suggested by previous 274 studies [63,70,71] . In addition, we did not find evidence that REM sleep deprivation modulated the ability to 275 apply cognitive reappraisal on the behavioral level. Therefore, REM sleep deprived subjects may have 276 been able to counteract the impact of increased amygdala responses during social exclusion. A potential 277 factor diminishing the severity of experimental social exclusion is that despite our attempts to create a 278 socially immersive context [72] , laboratory studies per se have limited relevance beyond the experimental 279 situation itself. Thus, even without explicit reappraisal strategies, this limited relevance might have 280 entered into the construal of emotional responses during passive social exclusion, and reduced the 281 influence of altered amygdala responses in the REMS group. 282 A related explanation for the brain-behavior dissociation is that healthy participants have 283 resources for compensation. As the neuroimaging findings advocate that REMS impacts limbic circuit 284 activity during emotional experiences, the question arises whether participants with a priori (sub)clinical 285 peculiarities in reaction to social exclusion, ostracism or labile sense of belonging would also report 286 altered subjective experiences. It has been shown that inter-individual dispositions such as differences in 287 rejection sensitivity [73] , social anxiety [74] , trait self-esteem, depression [75] , attachment style [76] and 288 situational factors of the exclusion experience moderate the effect of social exclusion. Last, it is possible 289 that experimental REM-sleep suppression becomes effective on the emotional level only when applied 290 repeatedly, simulating chronic sleep disturbances that are associated with psychiatric disorders in a 291 more realistic manner. 292 Speculatively, a further explanation could relate to the psychological task used to examine 293 emotional reactivity. Earlier research indicated that sleep deprivation effects might differ depending on 294 whether the focus was on general state-like morning affect (e.g. as measured using the PANAS), the 295 processing and responding to affective material (e.g. emotion recognition, pain processing, reactivity to 296 threatening stimuli) or on the direct induction of emotional states (e.g. inducing the unpleasant 297 experience of social exclusion). One assumption derived from the present data could be that direct 298 emotion-induction tasks are so powerful that a single night of sleep deprivation or selective REMS might 299 not suffice to exert effects on the psychological level. To our knowledge, however, apart from the 300 present study there is no further work that directly examined the effect of selective REMS on 301 experimentally induced emotional states. Second, general unspecific affect may function differently than 302 emotional reactions to specific elicitors [77] , potentially moderating the effect of REMS on these different 303 affective processes. However, at least one study found that general affect was not influenced by REMS, 304 which contradicts our findings [30] . Last, we are not aware of any study systematically comparing the 305 extent to which the different psychological aspects of affective experience are susceptible to selective 306 suppression of sleep stages. Taken together, the specific interaction of selective REMS with general 307 affect, in contrast to more confined emotional responses, demands more in-depths analyses of different 308 kinds of experimental designs and dependent variables. 309 Our findings of increased next morning general negative affect as well as increased limbic activity 310 after REMS are well in line with findings in patients with mental disorders. A wide range of mental 311 disorders, including mood and anxiety disorders, are not only accompanied by profound disturbances of 312 REM sleep [78,79] but also by deficits in generating and controlling emotions in an appropriate way. 313 Posttraumatic Stress Disorder (PTSD) is one of these mental disorders, where a link between REM sleep 314 and emotion regulation has been frequently discussed [80,81] . In these patients, deficits in emotion 315 processing and regulation on the behavioral level are coincided by alterations in brain networks 316 implicated in emotion regulation such as the amygdala, hippocampus, insula, and anterior cingulate [82] . 317 Apart from that, PTSD patients as compared to traumatized individuals who did not develop a PTSD, 318 show a profound REM sleep fragmentation and alterations in REM density accompanied by higher 319 sympathetic drive during this sleep stage [78,79] . Importantly, these REM sleep disturbances early after 320 trauma predict later PTSD development [83,84] , which let researchers speculate that REM sleep alterations 321 are not only a secondary symptom but represent one mechanistic factor contributing to the 322 development and/or the exacerbation of PTSD symptoms [80,81] . However, empirical evidence for such a 323 mechanistic link is scarce. importance from the fact that sleep deprivation does not have negative effects only, but can also have 333 positive effects, as evident in its anti-depressant potentials [87] . 334

Limitations 335
Although in the present study, the use of cognitive reappraisal (CRA) strategies was effective in 336 toning down feelings of social exclusion [88] , CRA was always applied in the second session. Therefore, we 337 cannot rule out that habituation to the task during the second session influenced the regulation of 338 emotionality by CRA. However, counterbalancing the view and reappraisal session would have 339 introduced even stronger confounds, considering that participants would have most likely also engaged 340 in CRA in the second session if they had learned about this strategy in the first session. In order to better 341 discriminate between CRA and habituation effects one could apply a three-session-design and randomly 342 instruct participants to use CRA either in the second or third session. In a larger sample than ours, Mauss 343 and colleagues applied such a design to show that poorer sleep quality over the course of the last week 344 was linked to decreased abilities in engaging in CRA [41] . In conclusion, the present study examined the effect of selective REMS on general affect as well 356 as the neural correlates of task-induced negative emotionality and the ability to regulate emotional 357 experiences by cognitive reappraisal during experimentally induced social exclusion. While we found that 358 REMS predicted general negative affect in the next morning, we did not find experimental evidence for 359 task-induced changes of negative emotionality during the experience of social exclusion. However, limbic an awakening, participants were kept awake for 90 seconds. In case the participant did not wake up, the 406 volume of the acoustic beep was increased, and ultimately the experimenter entered the participant's 407 room to turn on the lights to make sure the target sleep phase was interrupted. To control for the 408 number and lengths of awakenings, both suppression groups were disturbed similarly often during the 409 experimental night. In the REMS group, participants' sleep was disturbed as soon as they entered the 410 REM sleep stage. Awakenings for the SWSS group were selectively carried out during the non-REM sleep 411 stage N3, defined according to AASM guidelines [90] . The control subjects were not awakened at all. PSG 412 recordings were scored by an experienced sleep technician (C.L.) at the sleep laboratory at the 413 Department of Otorhinolaryngology at the University of Lübeck according to AASM guidelines [90] and 414 using Somnologica 3.3.1 (Build 1529). The technician was blinded for group assignments and hypotheses. 415

General Affect Ratings 416
The PANAS [50] was completed shortly before going to bed and after waking up on both nights by all 417 participants to assess the effect of sleep manipulation on general positive and negative affect. 418

Experimental Task 419
Upon waking up after the experimental night, participants were accompanied to the functional magnetic 420 resonance imaging (fMRI) facility. In the MRI, participants were instructed via a screen that they were 421 about to play a ball-tossing game, i.e. the Cyberball task, allegedly with two other persons (see Figure 1; 422 a second paradigm was presented to the subjects afterwards, which involved IAPS pictures to study 423 regulation of basic emotions but is not further described in the present manuscript). All three players, throwing movement and the red ball flying from one avatar to the catching avatar, with the latter 435 moving its arm to catch the ball. While participants believed that the other avatars were controlled by 436 two other anonymous persons sitting in adjacent rooms, their behavior in fact followed a predefined 437 script (see below). Following the instructions, a short sequence of lines of text built up on the screen, 438 making the subjects believe that the computer was being connected to a gaming server of the university 439 on which the experiment was run. This included the request of entering an IP-address, as well as lines 440 stating how many players were logged into the game. Subsequently, participants performed a short 441 training session consisting of seven trials in which they received and threw the ball three times. 442 The task then consisted of two sessions, each of which consisted of four experimental blocks. 443 Before each of the experimental blocks, a line of text was presented on-screen for 1500 ms telling the 444 participant that a new round would start. In every block, a maximum number of 24 ball tosses were 445 completed. In two blocks of every session, the participant was included in the game and repeatedly 446 received the ball throughout the entire block (inclusion blocks, INC), having the possibility to toss the ball 447 to one of the other avatars eight or nine times by clicking a button using the index finger (left player) or 448 middle finger (right player). In the other two blocks (exclusion blocks, EXC), after the participant had 449 received the ball three or four times, the paradigm was programmed so that the two avatars only tossed 450 the ball between one another, thereby effectively excluding the participant from the game. 451 After having completed four blocks during which subjects simply participated in the task without 452 additional instructions (VIEW session), a second session of the task was played. For the second session, a 453 short on-screen text instructed participants to use cognitive reappraisal to regulate their emotions 454 during the game (CRA session; see Fig. 1). Participants were specifically asked to reappraise the situation, 455 in case any negative emotions should arise ("In case negative emotions arise, please try to re-evaluate 456 the situation"). This was accompanied by the instruction to "partake in the game and try to visualize the 457 situation as vividly as possible", which was also presented before the first session. Subjects were asked 458 to press the left button as soon as they were ready, and then a short break varying randomly between 459 1500 and 2500 ms was presented, instructing participants to wait until the other two players had 460 indicated that they were ready. After presentation of a fixation cross for one second, the block started. 461 In each session, two inclusion and two exclusion blocks were presented to the subjects, with the 462 first block of each session always being an inclusion block. In one session, inclusion and exclusion blocks 463 were alternating, while in the other session, the initial inclusion block was followed by two consecutive 464 exclusion blocks and a final inclusion block. Between blocks, participants were asked to rate how they felt about the preceding block 468 regarding the extent to which they felt socially excluded. The low end of the scale was labelled as 469 rejected/despised and the high end was labelled as accepted/familiar. In addition, subjects rated their 470 sadness, anger, and shame, but these ratings were intended to distract from the purpose of the study 471 and not analyzed (sadness was described by: sad, downcast, gloomy; for anger: angry, irritated, furious, 472 mad; for shame: abashed, embarrassed). Each rating was displayed using a 9-point Likert-type scale 473 ranging from 0 (not at all) to 8 (very much). Every rating was initialized at 4 (i.e. a neutral rating), and 474 participants used presses of the right or left response buttons to move the rating to higher or lower 475 values, respectively. The time for ratings was limited to 4 seconds. After each rating and the start of the 476 next block within a session there was a short pause, jittered between 4.3 to 5.3 seconds. Emotion ratings 477 for each condition (EXC, INC) and emotion regulation session (VIEW, CRA) were averaged for each 478 participant and analyzed using repeated measures analyses of variance (rmANOVA). 479

Statistical Analysis 480
Average ratings of feeling excluded for each condition (EXC, INC) and session (VIEW, CRA), as well as the 481 percentages of the different sleep stages for the habituation and experimental nights were analyzed 482 using analyses of variance (ANOVA). Significant interactions and main effects were followed up using 483 paired comparisons. Since we expected that selective REMS would specifically alter affective experience 484 and associated neural responses [22,23] , we furthermore performed an a priori planned contrast of the 485 REMS group against both control groups, the CTL and SWSS. The alpha-level was set to .05, and was 486 adjusted using Bonferroni-correction, in case multiple tests were performed. 487

FMRI Data Acquisition and Preprocessing 488
For each of the two experimental sessions, 130 functional volumes were recorded at 3T (Siemens Trio, 489 Erlangen), of which the first three were discarded to allow for equilibration of T1 saturation effects. 490 Functional volumes consisted of 36 ascending near-axial slices (voxel size=3*3*3 mm, 10% interslice gap, 491 FOV=192 mm) and were recorded with TR=2200 ms, TE=30 ms, FA=90°. In addition, a high-resolution 492 anatomical T1 image was recorded consisting of 176 slices (voxel size=1*1*1 mm, FOV=256 mm, 493 TR=1900 ms, TE=2.52 ms, 9° FA). 494 The MRI data were analyzed using SPM12 in Matlab 2019b. Functional MRI images from each 495 session of the Cyberball paradigm were slice-time corrected to the middle slice and spatially realigned. 496 Subsequently, spatial normalization to MNI space was performed using unified segmentation [93] by 497 estimating the forward deformation fields from the mean functional image and applying these to the 498 realigned functional images. These spatially normalized images were then resliced to a voxel size of 499 2*2*2 mm and smoothed with an 8 mm full-width at half-maximum isotropic Gaussian kernel and high-500 pass filtered at 1/256 Hz. 501

FMRI Data Analysis 502
The preprocessed functional images were statistically analyzed by a two-level mixed effects GLM 503 procedure. For each participant, a statistical model was specified, including data from both experimental 504 sessions (VIEW, CRA). For each session, the INC, EXC and rating phases were modelled as regressors of 505 interest and the six realignment parameters estimated during spatial realignment were included to 506 account for variance in the functional data that was due to head motion. The contrast images obtained 507 from the individual participants were then aggregated in a random effects model on the second level to 508 test for effects of condition and session, and to test for differences between the three experimental 509 groups. To disentangle significant interaction effects, we ran a series of post-hoc tests on the parameter 510 estimates extracted from the peak activation, applying Bonferroni-correction for multiple comparisons. 511

Regions of Interest (ROI) for FMRI Analysis 512
In order to focus our analyses on the limbic system as well as the insula, regions commonly associated 513 with affective processing [94] , we constructed an a priori mask using the Wake Forest University Pickatlas 514 (v. 2.4) [95] . This mask comprised the anterior cingulate cortex, bilateral insula, bilateral hippocampus, as 515 well as left and right amygdala. We thank Konrad Whittaker and Tareq Naji for their help in conducting the polysomnography 523 measurements and Christian Lange for support in scoring the PSG data. 524

Data Availability 525
The datasets generated during and/or analyzed during the current study are available from the 526 corresponding author on reasonable request. 527