Global dissociation of the amygdala from the rest of the brain during REM sleep


 Rapid-eye-movement sleep (REMS) or paradoxical sleep is associated with intense neuronal activity, fluctuations in autonomic control, body paralysis and brain-wide hyperemia. The mechanisms and functions of these energy-demanding patterns remain elusive and a global picture of brain activation during REMS is currently missing. In the present work, we performed functional ultrasound (fUS) imaging at the whole-brain scale during hundreds of REMS episodes to provide the spatiotemporal dynamics of vascular activity in 259 brain regions spanning more than 2/3 of the total brain volume. We first demonstrate a dissociation between basal/midbrain and cortical structures, the first ones sustaining tonic activation during REMS while the others are activated in phasic bouts. Second, we isolated the vascular compartment in our recordings and identified arteries in the anterior part of the brain as strongly involved in the blood supply during REMS episodes. Finally, we report a peculiar activation pattern in the amygdala, which is strikingly disconnected from the rest of the brain during most but not all REMS episodes. This last finding shows that amygdala undergoes specific processing during REMS and may be linked to the regulation of emotions and the creation of dream content during this very state.


Introduction 36 37
Several studies have shown a strong link between REM sleep (REMS) and emotions. 38 According to Gujar and colleagues, REMS is strongly linked to the recalibration of the human 39 brain's reactivity towards specific emotions with a decreased reactivity towards fearful memory 40 but a significant enhancement toward happy memories (Gujar et al., 2011). Moreover, several 41 studies tend to show that the emotional memories are better recalled than neutral memories 42 (Phelps, 2004) especially during REM-rich sleep (Groch et al., 2015;Nishida et al., 2009;43 Wagner et al., 2001). It is also now well established that REMS disturbances are often 44 observed in PTSD cases. Mellman and colleagues have observed shorter and more frequent 45 sleep bouts in trauma-exposed patients compared to non-injured ones (Mellman et al., 2002). former is consolidated during REMS, while the latter is de-potentiated, meaning that the factual 53 memory remains embedded in the hippocampus, while the intensity of the feelings attached to 54 this memory is downscaled. This is considered as an "overnight therapy", necessary to cope 55 with distressful events and which is disturbed in cases of emotion-based disorders, such as 56 depression or posttraumatic stress disorder (PTSD), as these disorders are often linked with 57 REMS disturbances as previously stated. However, a few studies go against that model 58 (Wiesner et al., 2015). The biological mechanisms of emotional regulation of memories involve 59 an amygdala-hippocampal-medial prefrontal cortical (mPFC) network, whose 60 intercommunications are enhanced by theta and ponto-geniculo-occipital (PGO) oscillations 61 (or equivalent P-wave in rats), as well as elevated acetylcholine and cortisol levels during 62 REMS (Hutchison and Rathore, 2015;van der Helm et al., 2011). 63 In addition to the regulation and processing of emotions, several other functions have been 64 attributed to REMS, such as, for example, memory consolidation (which depends upon the 65 hippocampal theta oscillations created by activation of GABAergic neurons in the medial 66 septum (Boyce et al., 2016)), or the brain maturation specifically during early life REMS (Marks 67 et al., 1995) and more precisely to aid sensorimotor system development through muscle 68 twitches (Blumberg et al., 2013). Another function attributed to REMS is that of forgetting 69 has been maintained across evolution, it must have some important role for the survival of the 104 animal, which does not seem to have been found yet. 105 As previous studies could not give a global view of REMS's brain activity, we took advantage 106 of fUS versatility to scan more than 250 brain regions over multiple coronal and sagittal planes 107 during more than 600 REM episodes. This study thus provides an exhaustive characterization 108 of global brain hemodynamics during rodent REMS. We demonstrate a clear dissociation 109 between basal/midbrain structures and superficial ones, respectively activated in a tonic and 110 phasic manner. We also disentangle the vascular structures involved in the irrigation of the 111 brain during REMS episodes providing a detailed outlook of blood supply. Finally, we show 112 that brain activity reveals a striking dissociation between the amygdala complex and the rest 113 of the brain regions. 114 115

116
This study aimed at investigating the large-scale hemodynamics during REMS, in particular in 117 the amygdalar network. Using a chronic experimental approach developed previously (Sieu et 118 al., 2015), which included a cranial window and the implantation of a permanent fUS-119 compatible plastic prosthetic skull, that also enabled the attachment of the ultrasound probe 120 holder ( Figure 1A). In this setup, the different regions of the brain were monitored in a series 121 of coronal and sagittal planes, each acquisition lasting 30 minutes for 4-6h per day over the 122 course of several days ( Figure 1B). This resulted in a dataset of 84 recordings in n=8 animals, 123 totalizing 617 REM episodes recorded in 259 brain regions (Supplementary Figure S2), 124 together with hippocampal local field potentials (LFP) recordings, accelerometer, and neck 125 electromyogram (EMG) ( Figure 1C, Supplementary Table 1). 126

Distribution of CBV changes across brain regions during diverse arousal states 127
Measurement of the CBV in n=3 rats, in a total of 72 recordings, in which the animal is 128 spontaneously going through different arousal states (quiet wake -QW, active wake -AW, 129 non-REM sleep -NREMS, REM sleep -REMS) revealed a quiescent level of CBV fluctuations 130 in QW and NREMS (Figure 2A). AW however is associated with increased cortical CBV levels 131 especially in the primary sensory areas, while REMS is characterized by increased CBV in all 132 brain regions with strongest effect in the hippocampal and limbic structures. 133 As we aimed at studying in detail the hemodynamic changes in various parts of the brain during 134 REMS, we next focused on the changes in a large number of regions of interests located under 135 our various imaging planes, by computing the percentage of CBV change during REMS. 136 Calculations were performed using 1-3 min of either the QW or the AW for the baseline (Figure  137   2B). This double analysis shows consistently a large range of CBV changes in association with 138 REM between different parts of the brain in both analyses. While the hippocampal formation, 139 the periaqueductal grey (PAG), the superior colliculus (SC) and some parts of the cortex, (such 140 as the cingulate and retrosplenial cortices) present a strong percentage of CBV increase during 141 REMS, areas of the hypothalamus and laterally located cortices (auditory, rhinal, piriform 142 cortices) present modest CBV increases during REMS ( Figure 2B). This combined analysis 143 demonstrates that REMS hyperemia is not only a state of intense activation with respect to 144 QW and NREMS, which are known to quiescent states, but also to AW in all brain regions, 145 with strongest effects in the hippocampus and midbrain structures. Detailed mean values of 146 the CBV distributions in all regions across the different vigilance states are details in 147 Supplementary Tables 2 & 3.  148 149 Dissociation between the basal brain and the superficial brain areas during tonic and 150 phasic activations 151 We previously described strong hemodynamic changes, composed of both phasic and tonic 152 components (Bergel et al., 2018). By thresholding vascular activity during REMS, we were 153 able to extract a binary variable that accounted for the phasic component of REMS (seed 154 phasic-REM, equals 2 during phasic activity, 1 during REM, 0 otherwise), that we used as a 155 'seed' for correlation analyses and compared it with another variable accounting for the tonic 156 component of REMS (seed REM, equals 1 during REMS, 0 otherwise) ( Figure 3A). Individual 157 voxels taken in the superficial or deep structures of the brain show different activation profiles, 158 voxels in basal brain structures showing a very tonic activation (sustained during a single 159 REMS episode) while superficial pixels were active intermittently by phasic bouts, which was 160 captured by the different cross-correlations functions obtained with seed-phasic REM and 161 seed-REM ( Figure 3B). This phenomenon was clearly visible on all correlation maps (each 162 pixel displays the maximum of the cross-correlation function shown in 3B) generated with 163 either seed: cortical structures were more strongly associated with REMS-phasic than with 164 REMS (black arrows) on all brain sections ( Figure 3C). This effect was confirmed in regional 165 analysis across individuals and interestingly the timing associated with either seed variable 166 yielded different information. Interestingly, timings associated with seed-REM captured the 167 broad inter-episode fluctuations while those associated with REM phasic, revealed a precise 168 sequence of activation between brain regions and captured the intra-REM fluctuations (  drop at the end of the REM episode. This increased CBV in arteries was more pronounced (2-194 fold increase) in the arteries that vascularize the rostral part of the brain (acer, azac and azp), 195 compared to the arteries that vascularize the medial and posterior parts of the brain (ach, mcer, 196 pcer), confirming a general phenomenon of increased blood supply during REMS, but also an 197 emphasis of this enhanced blood flow in the rostral part of the brain. Further analysis shows a 198 significant propagation delay along the anterior branch with acer peaking earlier than azac and 199 azp (acer: t1 = -3.14 +/-3.26s, azac: t1 = 1.73 +/-6.50s, azp: t1 = 1.81 +/-3.03s) ( Figure 4G) 200 Such REM-associated increased CBV was observed at a lower level in veins with a surprising 201 antagonist activity between two side-by-side veins: the longitudinal hippocampal vein (lhiv) and 202 the azygos internal cerebral vein (azicv) (Supplementary Figure S3). 203 204

Atypical amygdala activity during REM sleep 205
When assessing inter-regional correlations in the CBV signal, the most striking pattern of 206 activity was found in the amygdala and consisted of a robust disconnection from the rest of the 207 brain, which was clearly visible on 'functional connectivity' matrices averaged over all REMS 208 episodes ( Figure 5A) and in the temporal fluctuations of individual recordings ( Figure 5B). This 209 effect is consistent with observation from previous figures: the amygdala showing both a 210 relatively low-level of hyperemia during REMS compared to other regions ( Figure 2) and low-211 correlation scores ( Figure 3). Strikingly, the amygdala's activity during REMS, showed a 212 remarkably unique activation profile compared to the rest of the brain and long periods of 213 strong fluctuation when the remainder of brain activity was silent ( Figure 5B, second part of 214 the episode). This effect was confirmed and strengthened using a seed-based approach taking 215 either the regional whole-brain activity as a reference ( Figure 5C, Supplementary Figure S4 This study provides a whole-brain characterization of the cerebral and vascular structures 227 involved in the atypical and large-amplitude vascular surges occurring during REMS. This 228 study goes significantly deeper in the understanding of REMS-associated hyperemia, as it 229 imaged a very large number of brain regions (257 regions) over hundreds of REMS episodes. 230 We implemented fUS imaging in 2D imaging planes with light ultrasonic probes as it is 231 compatible with both unrestrained movement and naturally induced sleep studies. 2D fUS 232 imaging enables us to ensure that the animal is not restrained, behaves almost perfectly 233 normally, and sleeps spontaneously. It is primordial as stressed and head-restrained animals 234 are less eager to sleep and deprivation protocols are often used to acquire sleep data, which 235 affects both the structure and nature of sleep episodes. As each imaging session could only 236 image on one single 2D plane, we had to repeat the experiment a large number of times in 237 order to achieve an almost full 3D coverage of the brain's regional activity during REMS. 238 Although this approach of multiple 2D planes has the disadvantage to lose the temporal 239 information regarding the coactivity of brain regions from different planes, we solved this 240 difficulty by imaging from both coronal and sagittal planes, thus relying on a respectable 241 number of co-activated regions in each single session. Such hyperemic activity might be physiologically important as it was kept throughout evolution, 258 despite its energy consumption. Moreover, a clear picture of global brain activity during REMS 259 is still currently missing. 260 In this study, we used functional ultrasound imaging to gather data on more than 250 brain 261 regions in both coronal and sagittal planes, thus providing a very exhaustive characterization 262 of global brain hemodynamics during rodent REMS. We demonstrate a clear dissociation 263 between basal/midbrain structures and superficial ones, respectively activated in a tonic and 264 phasic manner. We also disentangle the vascular structures involved in the irrigation of the 265 brain during REMS episodes providing a detailed outlook of blood supply. Finally, one of the 266 most noteworthy result of this work is the striking global dissociation of the amygdala activity 267 from the rest of the brain during the REM episodes. 268 269 Massive hyperemia observed across the whole brain and neurovascular coupling 270 A previous study has already shown a hyperemic activity during REMS in humans in some 271 brain regions in human using positron emission tomography (Maquet et al., 1996). However, 272 this study only presented a higher vascular activity correlated with REMS in pontine 273 tegmentum, left thalamus, both amygdaloid complexes, anterior cingulate cortex and right 274 parietal operculum, and some regions with a negative correlation with REMS mainly in cortical 275 areas. 276 One of the key findings of the present work is that hyperemia is global and spans throughout 277 all of the forebrain that we were able to image (2/3 of total brain volume). Additionally, it was 278 more sustained in the deep/midbrain structures (in particular in the hippocampus) than in the 279 cortex, which activated in phasic bouts. Thus, REMS can be described as a state of tonic 280 hyperemia in the forebrain that only partially spreads to the cortex. Also, activity in the different 281 cortices were strongly heterogeneous, with strongest activations in the retrosplenial, limbic, 282 motor and visual cortices but close to the levels of wake in the other sensory cortices 283 (somatosensory, piriform). This is surprising as rats preferentially use odor and texture rather 284 than vision. Hence, it is possible that hyperemia is associated with the reactivation of visual 285 networks (geniculate, colliculus, cortex) or in link with memory (retrosplenial, septum, and 286 hippocampus). Vascular hyperactivity specific to REMS in rats divides into tonic and phasic 287 regimes, the latter exhibiting transient brain-wide hyperemic patterns, which we called vascular 288 surges (VS). Bergel et al showed that these VS outmatched wake levels occasionally reaching 289 up to a 100% increase in the cortical and hippocampal regions compared to a quiet wake state.  Moreover, it is known for many years that the amygdala is electrophysiologically active during 343 REMS (White and Jacobs, 1975) and was also confirmed more recently during NREMS 344 (Girardeau et al., 2017). Our hypothesis is that during REMS, while the amygdala is active, it 345 is only strongly activated when the rest of the brain is not and especially the mPFC resulting 346 in the lower correlation observed in the connectivity matrix. Rats might need to regularly go 347 through phases of emotional regulation during REMS to cope with every day's accumulation 348 of strong emotional memories. 349 We hypothesize that we image here this downscaling at play using functional ultrasound 350 imaging during REMS. Finally, such a portable and wide field-of-view neuroimaging modality, 351 as functional Ultrasound imaging provides an extensive picture of brain function and the 352 interaction between large scale brain networks during sleep in rodents. Electrodes sites' locations were verified post mortem via histology to reconstruct the tract of 409 electrode bundles in the tissue. Each rat was euthanized and perfused with paraformaldehyde 410 4% to preserve the brains. Each brain was then cut using a vibratome to make 100µm-thick 411 slices. The slices were then contrasted using hematoxyline/eosine coloration and scanned 412 using a nanoscan. We then compared the slices with plates from Paxinos and defined the 413 trajectory of implantation using the marks left by the electrode in the brain. Knowing the 414 distances between the recording points, we could then define their position. 415 416

Recording sessions 417
After a recovery week following the surgical procedure, the animals were fit to be used in data 418 acquisition. After applying a generous amount of centrifuged ultrasonic gel, the ultrasonic 419 probe was put in place using a magnetic probe holder (home-made 3D designed and printed) 420 and the headstage for LFP recordings was plugged onto the connector. The animal was then 421 placed inside a box under an infra-red camera (to monitor the behavior) and the data 422 acquisition started. The ultrasonic probe was placed randomly across the day and its position 423 milliseconds (with a sufficient axial velocity) which is a good estimate of local cerebral blood 494 flow (CBF). We thus can build Doppler movies with a sampling frequency of 2.5 Hz, which can 495 even be increased if needed up to the pulse repetition frequency (here 500 Hz) through the 496 use of a temporal sliding window. To derive CBV maps from the raw Doppler movies, we 497 performed voxel-wise normalization from a baseline period: depending on the analysis done 498 afterwards, we either used 2 minutes of quiet wake, 2 mins of active wake, or the 20 seconds 499 preceding the onset of a REM episode. We extracted the distribution for each voxel during this 500 baseline period and computed a mean value, leading to one value for each voxel of the image. 501 To derive a signal similar to ΔF/F in fluorescence microscopy, we subtracted the mean and 502 divided by the mean for each voxel in the Doppler movie. This allowed normalization and 503 rescaling of ultrasound data, yielding to an expression in terms of percent of variation relative 504 to baseline (CBV % change). Each voxel was normalized independently before performing 505 spatial averaging. 506 507 Atlas registration 508 Coronal recordings were registered to two-dimensional sections from the Paxinos atlas 509 (Paxinos and Watson, 2017) using anatomical landmarks, such as cortex edges, hippocampus 510 outer shape and large vessels below brain surface as a reference. We performed manual 511 scaling and rotation along each of the 3 dimensions to recover the most probable registration. 512 Once performed, regions of interest were extracted using binary masks. This process allowed 513 us to derive vascular activity in 259 brain regions. 514 515 Cross-correlation analysis 516 We used Pearson's cross-correlation score to quantify the association between REM episodes 517 and the CBV activations in brain regions, or between brain regions, or between LFP signals 518 and CBV activations in brain regions. To do so, we performed the cross-correlation 519 computation on a large temporal window, depending on the couple analyzed. Regarding the 520 CBV activations, we either used pixels ( To assess the association between LFP events and CBV variables, we searched for 525 correlations between each possible combination of LFP band-pass filtered signals and regional 526 CBV variables. As neurovascular processes are not instantaneous, we considered possible 527 delays between electrographic and vascular signals and thus computed cross-correlations 528 functions between the two signals for any LFP-CBV pair and any lag in a given time window (-529 1.0 s to 5.0 s). We performed this analysis over pixel and regional variables, but only regional 530 variables allowed for statistical comparison across recordings. 531 532

Identification of vascular structures 533
Another technical difficulty we encountered was the recognition of the imaging planes / 534 registrations of these planes in a 3D map. Our laboratory recently developed such approach 535 in mice, by co-registering a vascular atlas on the atlas from the Allen Brain Institute (Nouhoum 536 et al., 2021). Such Approach will be available in the near future for rats, using recently 537 published vascular and MRI atlas in rats. But at the time of these experiments, this was not 538 available. Instead we used the few vascular atlas currently available in rats and mice (Scremin,  Student's table with n-2 degrees of freedom. Statistical testing for correlation distributions were 555 computed after Fischer transformation. Multiple comparison for regional analyses were 556 accounted for using Bonferroni correction. Due to the difficult experimental constraints (difficult 557 surgical procedure, precise electrode implantation, habituation and training required for the 558 locomotion task) no replication attempt was performed in this study, but the results were robust 559 and observable across individuals and recordings.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. MateietalSupplementalFiguresandTables.pdf