key: cord-0264396-fbg1lwe6
authors: Atlan, Gal; Matosevich, Noa; Peretz-Rivlin, Noa; Yvgi, Idit; Chen, Eden; Kleinman, Timna; Bleistein, Noa; Sheinbach, Efrat; Groysman, Maya; Nir, Yuval; Citri, Ami
title: Claustral Projections to Anterior Cingulate Cortex Modulate Engagement with the External World
date: 2021-06-18
journal: bioRxiv
DOI: 10.1101/2021.06.17.448649
sha: 7283e3577c3fbe6defac32ae7b64a0fb4810cb68
doc_id: 264396
cord_uid: fbg1lwe6

Engagement is a major determinant of performance. Hyper-engagement risks impulsivity and is fatiguing over time, while hypo-engagement could lead to missed opportunities. Even in sleep, when engagement levels are minimal, sensory responsiveness varies. Thus, maintaining an optimal engagement level with the environment is a fundamental cognitive ability. The claustrum, and in particular its reciprocal connectivity with executive regions in the frontal cortex, has been associated with salience, attention and sleep. These apparently disparate roles can be consolidated within the context of engagement. Here we describe the activity of claustro-frontal circuits in a task imposing a tradeoff between response inhibition and sensory acuity (‘ENGAGE’). Recording calcium fiber photometry during >80,000 trials, we characterize claustrum recruitment during salient behavioral events, and find that a moderate level of activity in claustro-cingulate projections defines optimal engagement. Low activity of this pathway is associated with impulsive actions, while high activity is associated with behavioral lapses. Chemogenetic activation of cingulate-projecting claustrum neurons suppressed impulsive behavior and reduced the engagement of mice in the task. This relationship became even clearer upon addressing individual variability in the strategy mice employed during the ENGAGE task. Furthermore, this association of claustrum activity and engagement extends into sleep. Using simultaneous EEG and photometry recordings in the claustrum, we find that cingulate projecting claustrum neurons are most active during deep unresponsive slow-wave sleep, when mice are less prone to awakening by sensory stimuli.

particular its reciprocal connectivity with executive regions in the frontal cortex, has been associated 23 with salience, attention and sleep. These apparently disparate roles can be consolidated within the 24 context of engagement. Here we describe the activity of claustro-frontal circuits in a task imposing a 25 tradeoff between response inhibition and sensory acuity ('ENGAGE'). Recording calcium fiber 26 photometry during >80,000 trials, we characterize claustrum recruitment during salient behavioral 27 events, and find that a moderate level of activity in claustro-cingulate projections defines optimal 28 engagement. Low activity of this pathway is associated with impulsive actions, while high activity is 29 associated with behavioral lapses. Chemogenetic activation of cingulate-projecting claustrum neurons 30 suppressed impulsive behavior and reduced the engagement of mice in the task. This relationship 31 became even clearer upon addressing individual variability in the strategy mice employed during the 32 ENGAGE task. Furthermore, this association of claustrum activity and engagement extends into sleep. 33

Using simultaneous EEG and photometry recordings in the claustrum, we find that cingulate projecting 34 claustrum neurons are most active during deep unresponsive slow-wave sleep, when mice are less prone 35

to awakening by sensory stimuli. 36

Engagement is a crucial determinant of behavior. Sensory events that are normally ignored can become 38 highly salient, depending on attentional state and engagement with the external world. In the 2011 39

World Championships, reigning champion and world record holder Usain Bolt committed a false start 40 on the 100 meter final and was disqualified from the competition. It cannot be said that Bolt lacked 41 experience or skill, as he is widely regarded as the best sprinter of all time 1 . In fact, his reaction time 42

is considered slow, an indication of his confidence and natural sprinting ability 2 . As he was waiting for 43 the gun to start the race, the slightest twitch of a muscle from his compatriot and eventual race winner, 44

Yohan Blake, was arguably the trigger that sent him off prematurely. Bolt's hyper-sensitivity likely 45 reflects his vigilant concentration and anticipation of the start signal, amplified by the high-pressure 46

occasion. 47

In challenging tasks or under pressure, performance scales with engagement only up to a limit, 48

following a bell-shaped curve known as the 'Yerkes Dodson' law 3 . Optimal performance is achieved 49 when a balance between vigilance and caution is achieved (being "in the zone"). Hyper-engagement 50 reduces task performance by increasing the propensity for impulse errors, while hypo-engagement 51

("zoning out") leads to missed opportunities. Furthermore, maintenance of heightened engagement over 52 time eventually leads to exhaustion and lapses in attention 4,5 . Engagement with the external world can 53 also be addressed in sleep, as deeper sleep is associated with a reduced propensity to be awoken by 54 sensory stimuli 6,7 . 55

Prefrontal regions of the cortex, such as the anterior cingulate cortex (ACC), and the orbitofrontal cortex 56

(OFC), have been implicated in regulating multiple attentional processes such as vigilance and impulse 57 control, positioning them as prime candidates for modulating engagement with the external world 8-11 . 58

Prefrontal cortex is heavily modulated by global arousal signals, widely attributed to the action of 59 neuromodulators and to subcortical structures such as the thalamus and the claustrum, due to their 60 capacity to synchronously signal to broad cortical territories [12] [13] [14] [15] . In this study, we identify a role for 61 claustral neurons projecting to the ACC in defining the degree to which mice engage with the external 62 world. 63

The claustrum is a thin neuronal structure, enclosed between the insular cortex and the striatum in the 64 mammalian brain 15-17 . It has been proposed to mediate cortical synchronicity, salience and attention 65 15,18-22 through strong claustro-cortical feed-forward inhibition 14,19,23 . The most prominent connectivity 66 of the claustrum is with prefrontal cortical structures such as the ACC and OFC 24-28 . Axons from these 67 frontal regions reciprocally innervate most of the claustrum, in contrast to constrained sensory zones 68 defined by afferents from sensory cortices 26,27,29-33 . Further anatomical division of the claustrum into 69 modules is supported by its internal organization into a 'core' and 'shell', as well as by mapping of its 70

projections 24,25,32,34,35 . Such modules could potentially play distinct roles in modulating executive 71 function and sensory processing, particularly given recent studies associating claustrum activity with 72 behavioral performance 19,20,36 . However, the association between claustral modules and physiology and 73

function is yet to be clearly demonstrated 37 . Particularly, data regarding the response patterns of 74 claustral populations during behavior are scarce 18,36,38 , and the rules governing behaviorally-relevant 75 claustrum recruitment remain unexplored. 76

Here we employed fiber photometry from anatomically defined claustral projection networks, recording 77 calcium transients in behaving mice. Our results demonstrate that ACC-projecting (ACCp) and OFC-78 projecting (OFCp) claustrum neurons form distinct modules, differing in their anatomical distribution 79

as well as in their spontaneous activity and their recruitment during behavior. Utilizing an automated 80

behavioral training system, we trained mice on a cognitively-engaging task ('ENGAGE'), imposing a 81 tradeoff between response inhibition and engagement. We find that claustro-frontal populations were 82 recruited during the task, responding transiently to multiple salient sensory events and motor actions. 83

Importantly, the activity of ACCp neurons, but not OFCp neurons, reflects the level to which mice 84 engage with the task. Thus, low ACCp activity corresponds to hyper-engagement, while high ACCp 85 corresponds to disengagement. Chemogenetic elevation of the activity of ACCp neurons was sufficient 86

to suppress impulsive responses. Furthermore, we observed that mice exhibited distinct strategies for 87 coping with the ENGAGE task, which related to their degree of ACCp recruitment. Finally, by applying 88 simultaneous EEG and photometry recordings, we found that the association between ACCp activity 89 and engagement extends to sleep. Claustrum activity increased during periods of maximal slow wave 90 activity (SWA) in NREM sleep, and correlated with the potential of a mouse to maintain its sleep in the 91 presence of awakening tones. Taken together, our results reveal the role of a sub-network of claustral 92

neurons projecting to the ACC in controlling engagement with the external world ( Figure S1 ). ACCp and OFCp claustral neurons exhibit differential projections to frontal targets ( Figure 1D ). In 104 addition, ACCp axonal arborization was more prominent within sensory cortical areas ( Figure 1E , S2E). 105

To address whether the anatomical segregation between ACCp and OFCp neurons extends also to their 106 physiology, we employed an intersectional viral approach to record population calcium transients from 107

ACCp or OFCp neurons in head-restrained mice running on a linear track ( Figure 1F ; Methods). ACCp 108 activity exhibited less frequent but longer-lasting spontaneous calcium events in comparison to OFCp 109 ( Figure 1G ). We next recorded concurrently from both populations in the same animal, using dual-color 110 fiber photometry ( Figure 1H , I). Spontaneous activity was correlated between ACCp and OFCp neurons 111 ( Figure 1J ). However, these windows of correlation were relatively short, as cross-correlations of ACCp 112

and OFCp activity decayed within a second ( Figure 1K ). Together, these results establish the ACCp 113

and OFCp subpopulations as partially overlapping, yet largely independent claustro-frontal networks. 114 115 ACCp activity bi-directionally reflects task engagement 116

We next proceeded to investigate the recruitment and function of the ACCp and OFCp populations 117 during behavior. We developed 'ENGAGE', a novel biphasic form of a randomized cue delay task, 118

supporting the investigation of multiple aspects of attentive behavior, including impulsivity, sensory 119 detection and selection, and sustained attention 40 . Trial onset (initiated by the mouse during training, 120

or every 20s during recording sessions, see below) was indicated by a brief broadband noise (BBN), 121 followed by a randomized delay period (of 0.5-3s), during which mice were required to withhold licking 122 until an auditory 'go' cue was played. Premature ('impulsive') licking during this initial stage of the 123 task resulted in trial abortion. Timely licks (<1.5s) following the go-cue were rewarded ('hit'). Trials in 124 which the mouse did not lick in a timely fashion were defined as 'miss' trials and were not rewarded 125 (Figure 2A ; Methods). Trial difficulty was determined by a combination of several conditions: go-cue 126 tone (four intensities), a tone cloud distractor (presence, absence), and a visual aid, presented together 127

with the auditory go-cue (presence, absence). Conditions were randomized across trials, while 128 maintaining fixed proportions of each condition throughout sessions ( Figure 2B ). Mice were trained in 129 a custom automated behavioral setup, supporting simultaneous individualized training of multiple mice 130

in their home cage ( Figure S3 and see Methods). Upon completion of the automated training protocol 131

in their home cage, mice transitioned to a head-restrained setup for individual recordings, allowing for 132

well-controlled photometry recordings over numerous trials (~100,000 trials from 30 mice in sum, see 133  supplementary table T1) . Mice reliably transferred their learning from the automated training to the 134 head-fixed condition ( Figure 2C ). The ENGAGE task was designed to probe fluctuations of hyper-and 135

hypo-engagement, supported by high proportions of both impulsive and omission errors (Figures 2D) .

On average, the performance of mice was impacted by all trial variables: improving psychometrically 137 as a function of increased intensity of the auditory cue, while performance at low cue intensities 138

benefitted from the addition of the visual aid. The tone cloud contributed to overall attentional load, 139

resulting in reduced hit rates, primarily in trials with intermediate cue intensities, as well as directly 140

increasing impulsive error rates ( Figure 2E ). 141

Within the ENGAGE task, both ACCp and OFCp claustrum populations were recruited by impulsive 142 as well as correct licks ( Figures 2F-G, S4A , B). In contrast, 'miss' trials provided a window into the 143 claustral representation of the go-cue in the absence of confounding lick events ( Figures 2H, S4C ). In 144 order to quantify the degree to which discrete temporal epochs contributed to the activity of claustral 145 networks, we fit a linear encoding model to the data 41 , creating a time-varying event kernel to relate 146 each event to its corresponding neural signal. We then compared the cross-validated explained variance 147 (CvR 2 ) for each event independently, as well as the unique contribution (ΔR 2 ) of that event to explaining 148 the total calcium signal (Figure 2I, S5A, B; supplementary table T2 and Methods) . This analysis served 149

to quantify claustral activity with relation to particular behavioral events, revealing that, as reported in 150 the cortex 41,42 , claustrum activity in both ACCp and OFCp networks was evoked by spontaneous 151 locomotion ( Figure S5C ); task-related licking ( Figure S5D) ; and, in some mice, by sensory stimuli 152 ( Figure S5E ). Strikingly, trial onset was strongly represented in ACCp, but not in OFCp activity ( Figure  153 2J, K). Indeed, trial onset appears to have been a significant catalyst of impulsivity, as 82% of impulsive 154 errors occurred within 1 second of the BBN signaling trial onset ( Figure 2F ). The unique coupling of 155

the ACCp signal to this major determinant of impulse errors suggested that that the ACCp network may 156 function to modulate response inhibition. 157

We therefore wished to determine whether ACCp activity leading up to trial onset may vary in 158 preparation for this predictable, yet challenging aspect of trial structure. Plotting ACCp activity by trial 159 outcome ( Figure 2L ), we observed that pre-trial ACCp activity was lower on average preceding trials 160 terminated by impulsive errors, compared to hit trials ( Figure 2M ). In fact, pre-trial ACCp activity 161

exhibited an inverse correlation with impulsive errors, such that lower pre-trial ACCp activity 162 corresponded to a higher probability that the trial would result in a premature lick ( Figure 2N ). 163

Intriguingly, the opposite relationship was observed between pre-trial ACCp activity and miss errors, 164

such that trials with higher pre-trial ACCp activity were more likely to result in misses ( Figure 2O ). 165

OFCp pre-trial activity was not significantly lower before impulsive errors ( Figure S6A , B), nor was it 166 correlated with impulse errors ( Figure 2P ) or misses ( Figure 2Q ). Importantly, we observed no 167 correlation between pre-trial ACCp activity and reaction time in hit trials, suggesting that ACCp activity 168 related specifically to the capacity of mice to engage with the trial, rather than correlating with 169 disruptions to perception or action ( Figure S6C , D). In sum, while ACCp and OFCp claustral networks 170 are recruited during multiple stages of the task, ACCp activity was uniquely tied to trial onset and to 171 the propensity of mice to engage in impulsive licking following low activity levels, or misses following 172 high activity. 173 174

In order to address the causal role ACCp activity plays in controlling impulsive behavior and 176 engagement, we co-expressed the excitatory DREADD hM3Dq together with GCaMP6s in ACCp  177 neurons (n=5, see supplementary table T1 ). This enabled a direct measurement of the effects of 178 chemogenetic manipulation on ACCp activity within the context of the task ( Figures 3A, S7A ). Mice 179 underwent behavioral training as described above, and were habituated to saline injections during head-180 restrained behavioral sessions. CNO administration (10mg/kg, i.p.) reliably elevated spontaneous 181 claustrum activity ( Figure 3B -C). Mice were then tested following administration of either saline or 182

CNO on interleaved days. Strikingly, CNO significantly and reversibly reduced impulsive error rates, 183

implicating the ACCp in control of impulsivity ( Figure 3D ). CNO administration did not change 184

impulsive error rates in GCAMP6s controls ( Figure S7B ). It lead to no changes in the representation of 185 task parameters in the ACCp signal in hM3Dq mice ( Figure S7C ), nor did CNO affect the response 186 times of mice in hit trials or their overall success rate ( Figure S7D , E). Intriguingly, CNO administration 187 induced a shift in the distribution of trial outcomes over the course of a session, such that in the first 188

half of the session, impulsive errors were largely replaced by hits, while in the second half of the session, 189 miss trials were more common ( Figure 3E ). Consistent with interpreting an elevation of miss trials as a 190 decrease in engagement, streaks of consecutive miss trials were more common following CNO (Figure 191 S7F). Thus, chemogenetic induction of ACCp activity reduced impulsivity at the cost of increased miss 192 trials over prolonged sessions. 193 194

We next addressed the distribution of behavioral strategies taken by mice in dealing with the ENGAGE 196 task. We divided mice into three behavioral categories based on the degree to which their behavior was 197 affected by task parameters (Figure 4A , B). 5/25 mice exhibited a selective approach, primarily 198 participating in easy trials with prominent cues or a visual aid (selective; cue modulation index >0.5).

A second group of mice (6/25) exhibited behavior that was consistent across trial parameters 200 (consistent; cue modulation index <0.5), and scaled with cue intensity. Both groups were minimally 201 affected by the cloud distractor (cloud modulation index < 0.04). In contrast, the largest group of mice 202 (14/25), was more susceptible to interference by the cloud (erratic; cloud modulation index >0.04). We 203 next addressed whether these groups corresponded to other elements of behavior in the task. Consistent 204 mice exhibited the highest overall success rate (mean rate 51%), distinguished from erratic mice, whose 205

success rate was lowest (mean rate 38%) ( Figure S8A ). The groups also differed in their impulsive error 206 rates, with erratic mice exhibiting a higher probability of performing impulsive licks, which further 207

increased in the presence of the tone cloud ( Figure 4C ). As noted earlier, impulsive errors closely 208 coupled to trial onset. This effect varied with the strategy of mice, and was most prominent in erratic 209 mice, evident in faster response times in impulsive errors ( Figure 4D ). Importantly, response times in 210 hit trials did not differ between groups, suggesting that potential confounds, relating to perception or 211 motor deficiencies across behavioral categories, are unlikely ( Figure S8B ). Thus, erratic mice appeared 212

to be hyper-engaged with the task, exposing them to impulsive erroneous responses to the trial-onset 213 cue and the cloud. 214

Consistent with the coupling of impulsive errors and ACCp activity, we observed an inverse correlation 215 between the response time in impulsive trials and the unique contribution of trial onset to the ACCp 216 signal ( Figure S8C ). Furthermore, erratic mice, which were the most prone to impulsive errors, showed 217 the strongest ACCp response to trial onset ( Figure 4E ). In light of this, we re-addressed ACCp pre-trial 218 activity, specifically in erratic mice. The association between trial outcome to ACCp pre-trial activity 219

was even more pronounced in this group in comparison to all mice ( Figure 4F , G). Maintaining a strong 220 negative correlation between pre-trial activity and impulsivity ( Figure 4H ), the positive correlation of 221

ACCp pre-trial activity in erratic mice with misses was stronger compared to all mice ( Figure 4I ). 222

Erratic mice also made more streaks of consecutive misses compared to the other groups ( Figure 4J ), 223

consistent with the notion that these mice transition between states of extreme engagement, 224 characterized by low ACCp activity, and periods of 'zoning out', characterized by high ACCp activity.

In addition, these streaks (>5 consecutive misses) were preceded by an increased reaction time in hit 226 trials ( Figure S8D ), as may expected by a gradual decrease in engagement ('zoning out'). Thus, by 227 considering individual differences in strategy within the ENGAGE task, we highlight the bidirectional 228 relationship between ACCp activity and hyper-engagement vs. disengagement. 229 230 Claustrum activity fluctuates on ultra-slow scales, together with inputs from auditory cortex 231

To understand whether ACCp activity is driven by cortical projections to the claustrum, we performed 232 recordings of axon-targeted GCAMP6s, expressed in inputs to the claustrum from the ACC (ACCi) and 233 auditory cortex (AUDi), together with jRGECO1a activity in ACCp neurons ( Figure S9A ). Spontaneous 234 correlations of ACCi or AUDi with ACCp were low, suggesting that neither of these inputs is the main 235 driver of spontaneous ACCp activity ( Figure S9B ). During the ENGAGE task, however, a strong 236 correlation emerged between AUDi and ACCp (0.6±0.1, in comparison to spontaneous correlation of 237 0.22±0.1). This contrasted with the correlations between ACCp and both OFCp and ACCi, which were 238 maintained at a similar level in the task as during spontaneous recordings ( Figure S9C -D). The AUDi 239 signal also represented task events similarly to the ACCp with respect to licking ( Figure S9E ), cue 240 responses ( Figure S9F ), and representation of trial onset ( Figure S9G ). However these transient events 241 are unlikely to account for the increase in correlation between ACCp and AUDi as the cross-correlation 242 between the two signals was maintained over prolonged time ( Figure S9C ). In fact, pre-trial activity 243

was correlated between the ACCp and AUDi, while no correlation of pre-trial activity was evident 244

between the ACCp and either the OFCp or ACCi ( Figure S9H , I). These results suggest that the ACCp 245

and AUDi acquired a common source of slow modulation within the context of the ENGAGE task, 246

from which OFCp and ACCi were exempt. Behavioral sessions lasted up to three hours, enabling 247

analyses of slow periodicity of pre-trial activity during continuous task performance. Indeed, we 248

observed that activity of all recorded channels tended to fluctuate at an ultra-slow time scale, on the 249 order of tens of minutes (0.1-0.7 mHz; Figure S9J ). However, not all signals from all mice exhibited 250 these fluctuations, and interestingly, the mice whose ACCp activity lacked ultra-slow fluctuations were 251 associated with the erratic group, which employed the least moderated behavioral strategy (Figure 252 S9K). In sum, a strong correlation emerges between the ACCp and AUDi within the context of the 253 ENGAGE task, maintained across ultra-slow time scales, potentially corresponding to a moderated 254 approach to task performance. and high activity in NREM. Yet unlike ACCp, activity in this network during wake trended towards 268 even higher levels than those observed during NREM ( Figure S10A ). We proceeded to examine whether 269

ACCp activity correlates with specific EEG patterns by dividing ACCp activation into quartiles within 270 each state and examining the corresponding EEG power spectrum ( Figure 5F ; Methods). We observed 271 that different levels of ACCp activity were associated with different profiles of SWA (< 4Hz) and theta 272 frequencies (6 -9Hz). SWA is an established marker of sleep depth 44 , and theta activity is maximal 273 during active exploration 45 . Thus, we used the ratio between SWA and theta power as an EEG index 274

for disengagement, and assessed its relation with ACCp activity. We found that ACCp activation 275 exhibited a positive linear relationship with SWA-to-theta ratio in each behavioral state ( Figure 5F , see 276 also supplementary table T3 and Methods). 277

Given the association between ACCp activity and engagement in the ENGAGE task, together with the 278 tight relation between ACCp activity and SWA (which is associated with the depth of natural sleep), 279

we hypothesized that high ACCp activity levels would also be associated with a deeper disengagement 280 from the sensory environment during sleep, leading to a lower probability of sensory-evoked 281

awakenings. To examine this, we set up an auditory arousal threshold experiment 7 , where we delivered 282 sounds approximately every minute and determined offline whether each trial resulted in sound-evoked 283 awakening ( Figure 5G ; Methods). We found that in 9 of 11 recordings, ACCp pre-trial activity was 284 higher before 'maintained sleep' trials than before trials resulting in awakening ( Figure 5H ). Again, this 285

profile was specific for the ACCp network, as OFCp activity did not significantly differ between events 286 leading to awakening vs maintained sleep ( Figure S10B ). Together, these data establish a specific 287 association between ACCp activity and engagement, where activity in this claustro-cortical projection 288 network is maximal during NREM sleep and in deep sleep when sensory stimuli rarely wake up the 289 animals, and is further associated with higher EEG SWA-theta ratio across behavioral states. 290

Proposals regarding the function of the claustrum have been framed within the context of two seemingly 293 distinct timescales. On the one hand, the claustrum has been proposed to function in the processing of 294 acute sensory events or distractors, in the context of salience and the gating of sensory perception 18-295 21,32,38,46 . On the other hand, claustrum activity has been linked to slow, state-like transitions and 296

oscillations, and even consciousness 13,14,22,47 . The detailed description provided in this work, of the 297 activity of distinct claustrum populations during the ENGAGE task, as well as during natural sleep, 298

bridges acute and prolonged timescales. Our observations are consistent with the majority of 299 experimental observations and hypotheses published to date regarding the function of the claustrum, 300

and identify a role for ACCp claustral neurons in controlling the full continuum of engagement, 301

providing a holistic and consistent framework for unifying the different perspective on claustral 302 function ( Figure S1 ). 303

Our study supports the existence of parallel modules of claustrum function, by defining two 304

anatomically, physiologically, and functionally distinct networks, identified by a projection bias to ACC 305 or OFC. Task events and outcomes are broadcast to both OFC and ACC, while trial onset, the most 306 predicable element of the task, was reported selectively by ACCp neurons. Recruitment patterns of 307 projection-defined claustral neurons identified in our recordings may reflect differences in their passive 308

properties 32,48-50 or genetic identity 19,20,36,46,51 , but may also define an orthogonal dimension of 309

behaviorally-relevant ensembles of claustrum neurons. 310

Whereas OFCp pre-trial activity was uncorrelated with performance, high ACCp pre-trial activity 311 correlated with misses, and low activity correlated with impulsive errors. The significance of these 312 results is enhanced considering previous findings, in which silencing a genetically-defined 313 subpopulation of claustrum projection neurons increased impulsive errors in the presence of sensory 314 load 19 . It is likely that this effect is mediated primarily by ACCp neurons 20 . Modulation of the axis of 315 engagement is of broad clinical significance, ranging from hyper-engagement associated with attention-316 deficit disorders 52 , and schizophrenia 53,54 , to hypo-engagement and apathy, commonly observed in 317 neurodegenerative disorders 55 . The identification of a specific pathway bi-directionally controlling the 318 full extent of the engagement axis is anticipated to serve as the basis for novel approaches for therapeutic 319

intervention. 320

The capacity to explore a large number of trials across a rich parameter space within the context of the 321 ENGAGE task, exposed individual differences in attentional strategies employed by mice. Some mice 322

were discriminatory in their approach, such that selective mice exhibited a bias towards the visual cues 323

and most prominent auditory cues, while consistent mice prioritized the auditory cue over the infrequent 324 visual cue. However, most mice (erratic) exhibited behavior that was less moderated, attempting to 325 respond to all the cues within the task (auditory cues of all attenuations, as well as visual cues). The 326

hyper-vigilant approach of erratic mice exposed them to impulsive actions in response to the trial onset 327 tone, as well as the tone cloud distractor. This approach also appeared to be exhausting, leading to 328 erratic mice to streaks of missed trials. Intriguingly, the ACCp signal of mice within the erratic group 329 exhibited the most prominent representation of the trial onset tone, as well as the strongest bidirectional 330 correlations with impulsive actions and omissions. While it is likely that individual differences in 331 behavioral strategies and corresponding neural signals relate to task engagement, few mechanistic 332 studies have probed these relations 56 . Broader implementation of automated training, supporting the 333 investigation of sophisticated behavioral paradigms in large cohorts of animals, will enable further 334 exploration of individual differences. 335

Likewise, our results support and shed new light on recent work associating claustrum activity and 336

sleep, by establishing that the high ACCp claustral activity is associated with decreased engagement 337 during NREM sleep, and a reduced probability to awaken in response to auditory stimuli. This is in line 338

with the fact that ablation of the claustrum most prominently impacts SWA in the ACC 14 . In addition, 339

our long recordings during undisturbed sleep provide the first characterization of claustrum activity 340 during REM sleep, which constitutes less than 10% of the light period in mice 43 . We found that 341 claustrum activity is predominantly silent during REM sleep, an intriguing observation that is outside 342 the scope of this report, but is worthy of further study. 343

Evidence so far suggests that widespread activation of claustral projections to ACC would provide a 345

feed-forward inhibitory signal 19,23 . However the mechanism through which this would modulate 346

engagement remain an open question. performance, as well as a correlation of ACCp activity and SWA-to-theta ratio of cortical EEG. 365

Furthermore, we observe a reduction in ACCp activity during REM sleep, a behavioral state associated 366

with desynchronized cortical activity. It is therefore possible that the mechanism through which the 367 claustrum modulates engagement is by impacting regional cortical oscillations, which differ during 368 vigilant behavior versus lapses 66 . The limited temporal dynamics of calcium signals do not lend 369 themselves to answer questions relating to fast neural activity and oscillations, and future 370 electrophysiological studies are likely to shed more light on the spectral properties of claustro-cortical 371 activity during behavior. 372

In summary, ascribing a role for claustral neurons projecting to the anterior cingulate cortex in 373 modulating the full axis of engagement, from hyper-vigilance and impulsivity through to 'zoning out' 374 and on to deep sleep, provides a holistic explanation of claustral function. Importantly, this framework 375

implies that the breadth of circumstances during which the claustrum is recruited is wider than 376 previously thought, likely spanning the entire range of events in which salient information is processed, 377

such as learning, rewarding or stressful events, and social behavior. 378 379 Methods 380 Animals: All mice described in this study were male C57BL/6JOLAHSD obtained from Harlan 381

Laboratories, Jerusalem Israel. Mice were housed in groups of same-sex littermates and kept in a 382 specific pathogen-free (SPF) animal facility under standard environmental conditions-temperature (20-383 22°C), humidity (55 ± 10 %), and 12-12 h light/dark cycle (7am on and 7pm off), with ad libitum access 384

to water and food. Mice were randomly assigned to experimental groups. All experimental procedures, 385

handling, surgeries and care of laboratory animals used in this study were approved by the Hebrew 386

University bar was glued to the skull, the incision was closed using Vetbond bioadhesive (3M) and the skull was 405 covered in dental cement and let dry. An RFID chip (ID-20LA, ID Innovations) was implanted 406

subcutaneously. Mice were then disconnected from the anesthesia, and were administered with 407 subcutaneous saline injection for hydration and an IP injection of the analgesic Rimadyl (Norbrook) as 408 they recovered under gentle heating. Coordinates for the claustrum were based on the Paxinos and 409

Franklin mouse brain atlas 67 . Unless noted otherwise, viruses were prepared at the vector core facility 410 of the Edmond and Lily Safra Center for Brain Sciences, as described previously 26 . 411

Supplemental claustrum and prepared for EEG and EMG recordings. Two screws, frontal and parietal (1 mm in 416 diameter) were placed over the right hemisphere for EEG recording. Two additional screws were placed 417 above the cerebellum as reference and ground. Two single-stranded stainless-steel wires were inserted 418

to either side of the neck muscles to measure EMG. EEG and EMG wires were soldered onto a custom-419 made headstage connector. Dental cement was used to cover all screws and EEG/EMG wires. Following 420 validation of photometry signal, mice were transported to Tel Aviv University for further recordings. were used in the analysis pipeline, three channels for each image: DAPI, eGFP and tdTomato or Ruby. 445

The data was analyzed using the CellProfiler v.3.0.0 co-localization pipeline (www.cellprofiler.org), 446

with minor modifications, including feature enhancement and shrink/expand objects 68,69 . For 447 fluorescently labelled retroAAV analysis (n = 3 ACC/OFC; 2 ACC/ACC OFC/OFC mice), a DAPI 448 object mask was generated and objects from eGFP and tdTomato channels that overlap with the mask 449

were considered labelled cell bodies. Overlap was defined as the overlay of a detected cell body from 450 the GFP channel coinciding with a cell body in the tdTomato channel, both coinciding with the DAPI 451 mask. RetroAAV-H2B (6 ACC/OFC; 2 ACC/ACC; 2 OFC/OFC mice) expressed in the nuclei, 452

providing lower background and allowing detection of labelled nuclei directly from eGFP and Ruby 453 channels without a DAPI object mask. In addition, the analysis was modified such that object centroid 454 distances were measured and calibrated such that only objects with a maximal 6 pixel centroid distance 455 between them were considered to be double-labelled cells. Histograms corresponding to the spatial immuno-stained to enhance indicator (GCaMP6s or jRGECO1a, see above) fluorescence in projections.

After alignment of section images to the Paxinos and Franklin mouse brain atlas 67 a manual threshold 462

was set for every brain such that the claustrum area would be saturated, and background minimal, This enabled activation of different task parameters for individual mice based on their performance. 478

Training comprised several stages, and each mouse progressed individually, according to its learning. 479

Mice were then taught to associate the auditory-visual cue with water availability during a lick 480 adaptation period. Entry of a mouse into the port (an RFID reading) initiated a trial, reported to the 481 mouse by a 0.1 sec broadband noise (BBN, intensity = 70.5db SPL) marking trial onset. Trial initiation 482

was followed by a varying delay period in which mice had to withhold lick responses. This delay period 483 lasted 0.1 sec in the adaptation phase and was prolonged in the following training steps. If the mouse 484

successfully withheld licking, a cue was presented at the end of the delay period, consisting of 5 pure 485 tone pips of 6 kHz, 0.1 sec long (spaced 0.1 sec, intensity = 86.1db), a white LED light (these 486 auditory/visual cues were referred to as AudVis). The first lick within a 1.5s window following cue 487 onset was rewarded (10ul of drinking water). Impulsive or late licks were not rewarded, and mice had 488

to exit the port (terminate and reinitiate RFID reading) before a new trial could be initiated. After mice 489 reached satisfactory success rates (50-70% correct, 2.6 days on average) they proceeded to stage 2 490

where the delay was prolonged to between 0.5-2s (2.5 days on average). Mice proceeded to stage 3, 491

which included the full range of possible delays (0.5-3s) and a gradual transition to auditory trials with 492 no visual aid (Aud) in three steps: 30% Aud (Stage 3a), 50% Aud (Stage 3b), and 70% Aud (Stage 3c).

Following stage 3 (4.4 days on average) a pure tone-cloud masking stimulus was introduced (4s of 494 continuous chords assembled from logarithmically spaced pure tones in the frequency range of 1-495 10kHz, excluding the target cue frequency, intensity = 67.5db SPL), lasting from trial onset throughout 496 the delay and cue. The tone-cloud was also introduced gradually. Stage 4a comprised of 70% auditory-497 visual trials with tone cloud (AudVisCloud) and 30% Aud (2.7 days in average). Stage 4b included 50% 498

AudVisCloud trials, 20% auditory cloud trials (AudCloud), 15% Aud trials and 15% AudVis trials. 499

After mice were familiar with the cloud in both visual and non-visual trials, we proceeded to stage 4c, 500

and increased the rate of cloud trials to 65% AudCloud, while the rest of the trials comprised 15% 501

AudVisCloud, 15% Aud, and 5% AudVis (mice spent on average 4.5 days in stages 4b + 4c). Finally, 502

we gradually added 3 attenuations of the target cue. First, in stage 5a, 30% of the trials included the full 503 range of attenuations (Go-Cue trial intensities were (db SPL): #1: 68.75db; #2: 81.2db; #3: 86.1db; #4: 504 91.6db), which increased in stage 5b to 50% of the trials and then in stage 5c to 100% percent of the 505 trials (13 days on average, depending on the availability of the recording system, adding up to a mean 506 total of 29.6 days of training). Response duration and reward size were kept constant throughout 507

training. Due to the COVID-19 pandemic, the training schedule of two mice was altered, and they were 508 thus excluded from panels illustrating training data. 509

Task structure during head-restrained recordings was identical to the automated training, except that 510 trials were initiated automatically every 20 seconds. Behavioral sessions contained blocks of trials 511

containing 15 (in some cases shortened to 8 or 10) occurrences of each possible combination of 512 parameters, in random order. These blocks were repeated 2-4 times as long as mice maintained 513 participation, for a total of up to 1000 trials / mouse / day (sessions typically extended over 240-480 514 trials). Trials in which the mouse licked late (>1.5s) were rare (~2% of all trials) and were thus also 515 labelled as misses. The degree to which different trial parameters (cue intensity, cloud, visual aid) 516 affected behavior was quantified by calculating a modulation index for the effect of each parameter on 517 hit rates in the task (i.e. difference normalized by sum: ( , ) = !"#_%&#'()!"#_%&#'* !"#_%&#'(+!"#_%&#'* ). For cue 518 modulation, we compared hit rates in the second lowest intensity, which was the most variable, to the 519 strongest intensity. 520

In Vivo fiber photometry recordings: Fiber photometry data was collected using a 1-site Fiber 521

Photometry system (Doric Lenses, Canada) adapted to two excitation LEDs at 465nm (calcium-522 dependent GCaMP fluorescence) and either 405nm (isosbestic control channel) or 560nm (for two-523 color recording using jRGECO). Simultaneous monitoring of the two channels was made possible by 524

connecting the LEDs to a minicube (with dichroic mirrors and cleanup filters to match the excitation 525 and emission spectra; FMC4 or FMC5, Doric) via an attenuating patch cord (400 µm core, NA=0.37-526 0.48). LEDs were controlled by drivers that sinusoidally modulated 560nm/465nm/405nm excitation at 527 210/210/330Hz, respectively enabling lock-in demodulation of the signal (Doric Lenses, Canada). 528

Zirconia sleeves were used to attach the fiber-optic patch cord to the fiber implant on the animal. Data 529

were collected using Femtowatt photoreceiver 2151 (Newport) and demodulated and processed using 530

an RZ2 (at TAU) or RZ5P (at HUJI) BioAmp Processor unit and Synapse software (TDT). LED 531 intensities were individually modulated in each mouse to allow the recording of viable signals with the 532 minimal intensity possible. To this end, 465nm LED intensity was gradually increased until robust 533

GCaMP/jRGECO fluctuations were observed above noise. Next, the 405nm (isosbestic control channel) 534

LED intensity was set to allow detection of motion artifacts. The total power at the tip of the patch cable 535 was most often 0.05-0.1mW. The signal, originally sampled at 24414Hz, was demodulated online by 536 the lock-in amplifier implemented in the processor, sampled at 1017.25Hz and low-pass filtered with a 537 corner frequency at 4Hz. All signals were collected using Synapse software (TDT). EEG and EMG 538

were digitally sampled at 1017 Hz (PZ2 amplifier, Tucker-Davis Technologies), and filtered online: 539

both signals were notch filtered at 50/100 Hz to remove line noise and harmonics; then, EEG and EMG 540

were band-pass filtered at 0.5-200Hz, and 10-100Hz, respectively. Due to a technical issue, EEG were 541 also high-pass filtered in hardware > 2Hz but a comparison with full broadband (>0.5Hz) EEG in 542 several animals verified signal differences were minor and did not affect the ability to analyze sleep 543 stages or SWA. Simultaneous video data (used for sleep scoring and for behavioral assessments) were 544 captured by a USB webcam (at TAU) or an IR camera (at HUJI, Basler) synchronized with 545 electrophysiology/photometry data. Offline, EEG and EMG were resampled to 1000 Hz (MATLAB, 546

The MathWorks) for sleep scoring and power spectrum analysis. 547

Behavioral fiber photometry recordings were made in one of three head-restrained conditions: 1) 548

Spontaneous recordings, in which no stimuli were presented, and the mouse was free to run on a 549

treadmill (Janelia 2017-049 Low-Friction Rodent-Driven Belt Treadmill) for 10min (for validation of 550 chemogenetic effects) or 40min (for correlation analyses). 2) Passive auditory sessions, in which 551 broadband noise or frequency sweeps (1-40Khz played at a 100kHz sample rate, through an RP2.1 552 processor, TDT), attenuated between 0-20db (SA1 amplifier, PA5 attenuator, TDT) were played while 553 the mouse was free to run on a treadmill. 3) Task sessions, which consisted of several blocks (1-4), each 554

consisting of 120-180 trials, as described above. 555

Fiber photometry analysis: Unless otherwise noted, all analysis was performed using custom MATLAB 556

scripts. First, to correct for baseline drift due to slow photobleaching artifacts, particularly during the 557 first several minutes of each session, a 5 th order polynomial was fit to the raw data and then subtracted 558 from it. After baseline correction, ΔF/F was computed using the 99 th lowest percentile value as F0 (

, and the resulting trace was z-scored relative to the mean and standard deviation of the entire 560 recording session to normalize between channels and across mice. For 2/30 mice, motion artifacts were 561 corrected by using the z-scored isosbestic control channel as a sample-by-sample F0 for computing ΔF. 562

To correct for small session-to-session fluctuations in the signal, while maintaining quantitation of pre-563 trial activity, we calculated pre-trial activity for every individual trial (four seconds before trial onset), 564

and used the pre-trial signal as a dependent variable in a linear model with recording session and trial 565 outcome as independent variables (baseline ~ outcome + session). A scalar value of the intercept and 566 estimate for each session was then subtracted from the corresponding data set, setting the mean baseline 567

for correct trials for each session at approximately zero. Pre-processed data was then cut into 20 second 568 windows (-5:15 seconds) around each behavioral epoch: trial onset, cue onset, lick onset and run onset, 569

and concatenated for each mouse to form an event-aligned activity matrix together with an information 570 Chemogenetic activation: 30 minutes prior to each recording session, mice received an IP injection of 587 either saline as a control or clozapine-n-oxide (CNO), diluted to a final dilution of 1mg/ml (10 mg CNO 588 in 500ul DMSO and 9.5 ml saline) and administered at a dose of 10 mg/kg. 589

Spectral analysis of pre-trial dynamics: Pre-trial activity was analyzed over individual sessions of 300 590 trials each (shorter sessions were excluded from the analysis, and longer sessions were analyzed only 591 up to trial 300). A fast Fourier Transform (using the fft function in MATLAB) was applied to each 592 session, and the average power spectrum over sessions was compared to a threshold defined by the 593 maximal power obtained in each frequency over 1000 shuffling iterations of the data. The reported 594 fluctuation frequency is the peak of the power spectrum that crosses this threshold. 595 intermittently every 60s (± 0.5 jitter) when mice (n=12) were undisturbed. The sensitivity of the setup 610

was confirmed by verifying that awakening probability was significantly higher for louder sounds (19.8 611 ± 8.4% vs. 8.1 ± 2.6% for 80dB vs. 65dB SPL sounds, respectively, p<0.001, paired t-test). The analysis 612 presented in Figure 5H is based on the louder sound, for which there was a sufficient number of trials 613

in both conditions (maintained sleep and awakening). Whenever COVID19 lockdown restrictions 614 allowed (n=9/12 animals), we performed two separate experimental sessions per animal. 615

Data and code availability: Full data and code used for creating the figures will be uploaded to a public 617 repository prior to publication. represent 95% confidence intervals. Unless noted otherwise, data are mean ± s.e.m. *p < 0.05, 843 **p < 0.01, ***p < 0.001; n.s., not significant. See Supplementary Table 3 for further details of the 844 statistical analyses. 845 otherwise, data are mean ± s.e.m. *p < 0.05, **p < 0.01, ***p < 0.001; n.s., not significant. See 874

Supplementary Table 3 for further details of the statistical analyses. 875 with the cloud in both visual and non-visual trials, we proceeded to stage 4c, and increased the rate of 940

AudVis. Finally, in stage 5, 3 additional attenuations of the target cue were introduced. In stage 5a to 942 30% of the trials, in stage 5b to 50% of the trials, and in stage 5c to 100% percent of the trials (success 943 rate in the full task during training is shown in Figure 2C ). (B) As in A, for hit rate (excluding impulsive 944 trials). Right panel summarizes the hit rate in the full task during training compared to the head-fixed 945 recordings. The increase in missed trials reflects the change from a self-paced task to a constant 20 946 second inter-trial interval. (C) As in A, B for impulsive errors, which were far less prominent during 947 head-fixed recordings (potentially reflecting reduced competition for the port compared to the group-948 Response time in ACCp mice (n=20) is uncorrelated with pre-trial activity. Thick line represents linear 974 fit, dotted lines represent 95% confidence intervals. Unless noted otherwise, data are mean ± s.e.m. 975 *p < 0.05, **p < 0.01, ***p < 0.001; n.s., not significant. See Supplementary Table 3 ACCp mice (n=13) is increased in hit trials immediately preceding miss streaks, compared to all hit 995 trials. Unless noted otherwise, data are mean ± s.e.m. *p < 0.05, **p < 0.01, ***p < 0.001; n.s., not 996 significant. See Supplementary Table 3 for further details of the statistical analyses. 997 

On the performance of Usain Bolt in 643 the 100 m sprint

How Usain Bolt can run faster -effortlessly

The relation of strength of stimulus to rapidity of habit-646 formation

The psychological significance of the concept of 'arousal' or 'activation'. Psychol. 648 Rev

In the zone or zoning out? 650 Tracking behavioral and neural fluctuations during sustained attention

The Yin and Yang of Sleep and Attention. Trends 653 Neurosci

Locus coeruleus norepinephrine activity mediates sensory-evoked awakenings 655 from sleep

Prefrontal executive and cognitive functions 657 in rodents: Neural and neurochemical substrates

Dissociable intrinsic connectivity networks for salience processing and 660 executive control

Top-down versus bottom-up control of attention in the 663 prefrontal and posterior parietal cortices

Thalamic dual control of 665 sleep and wakefulness

A claustrum in reptiles and its role in slow-wave sleep

The claustrum coordinates cortical slow-wave activity

Attention: The claustrum

The claustrum: a historical review of its anatomy, physiology, 673 cytochemistry and functional significance

The claustrum in review

A role of the claustrum in auditory scene 676 analysis by reflecting sensory change

The Claustrum Supports Resilience to Distraction

Claustral Neurons Projecting to Frontal Cortex Mediate Contextual 680 Association of Reward

A role for the claustrum in salience processing?

Hypotheses relating to the function of the 684 claustrum

Inhibitory Control 686 of Prefrontal Cortex by the Claustrum

Complete single neuron reconstruction reveals morphological diversity in 688 molecularly defined claustral and cortical neuron types

Input-output organization of the mouse 690 claustrum

Mapping synaptic cortico-692 claustral connectivity in the mouse

Organization of the connections between claustrum and cortex in the mouse

Claustrum projections to prefrontal cortex in the capuchin monkey (Cebus 696 apella)

Rat's claustrum shows 698 two main cortico-related zones

Unimodal responses prevail within the 700 multisensory claustrum

Sensory maps in the claustrum of the cat

Synaptic Connectivity between the Cortex and 704

Claustrum Is Organized into Functional Modules

Cortical hierarchy governs rat claustrocortical circuit organization

Synaptic Organization of the Neuronal 708 Circuits of the Claustrum

Topographic gradients define the 710 projection patterns of the claustrum core and shell in mice

The Mouse Claustrum Is Required for Optimal Behavioral Performance 713 Under High Cognitive Demand

Claustral Input to the Macaque Medial Posterior Parietal Cortex

The Claustrum is Involved in Cognitive Processes Related to the 717 Classical Conditioning of Eyelid Responses in Behaving Rabbits

A Designer AAV Variant Permits Efficient Retrograde Access to 720 Projection Neurons

Neurobiology of Attention

Single-trial neural 724 dynamics are dominated by richly varied movements

Spontaneous behaviors drive multidimensional, brainwide activity

Sleep-Wake Cycle in Young and Older Mice

The homeostatic regulation of sleep need is under genetic 730 control

Rhythms of the Brain. Rhythms of the Brain

The claustrum-medial prefrontal cortex network controls attentional set-734 shifting

What is the function of the claustrum?

Functional properties, topological organization and 738 sexual dimorphism of claustrum neurons projecting to anterior cingulate cortex

Identification of mouse 741 claustral neuron types based on their intrinsic electrical properties

A functional logic for neurotransmitter co-743 release in the cholinergic forebrain pathway

Proteomic analysis illuminates a novel 745 structural definition of the claustrum and insula

A review of heterogeneity in attention 747 deficit/hyperactivity disorder (ADHD)

Impaired attention as an endophenotype for molecular 749 genetic studies of schizophrenia

Apathy in neuropsychiatric disease: Diagnosis, pathophysiology, and treatment. 755

Mice alternate between discrete strategies during perceptual decision-757 making

A hierarchical model of 759 inhibitory control

A Sensorimotor 761 Circuit in Mouse Cortex for Visual Flow Predictions

The posterior cingulate and medial prefrontal cortex mediate the 763 anticipatory allocation of spatial attention

Response inhibition in the stop-signal paradigm

Long-range and local circuits for top-down modulation of visual cortex 767 processing. Science (80-. )

Predictive Processing: A Canonical Cortical Computation

Prefrontal Parvalbumin 771 Neurons in Control of Attention

Very slow EEG fluctuations predict the 773 dynamics of stimulus detection and oscillation amplitudes in humans

Regional Slow Waves and Spindles in Human Sleep

Selective neuronal lapses precede human cognitive lapses following sleep 778 deprivation

The mouse brain in stereotaxic coordinates

CellProfiler: image analysis software for identifying and quantifying 782 cell phenotypes

Improved structure, function and compatibility for cellprofiler: Modular 784 high-throughput image analysis software

Moving beyond P values: data 786 together with activity of claustrum afferents from ACC (ACCi; top) or AUD (AUDi; bottom) axons, 1000 using axonal-targeted GCaMP6s. (B) Average autocorrelations of ACCi/ACCp or AUDi/ACCp 1001 spontaneous activity, and respective cross-correlations between channels

Average autocorrelations of ACCp/OFCp (left), ACCi/ACCp (middle) or AUDi/ACCp (right) activity 1003 during task recordings, as well as cross-correlations between channels

Summary of overall correlations between channel activity during free recordings (spontaneous)

3 mice for each group), demonstrating an increase 1006 in correlated activity between AUDi and ACCp networks during the task. Gray dots represent the 1007 maximal correlation of shuffled data over 1000 iterations per mouse, averaged across mice

Model quantification of ACC (ACCi) and auditory (AUDi) cortical inputs to the claustrum

Average trace from all axonal recordings in ACCi (green) vs AUDi (gray), aligned 1011 to trial onset, depicting trial onset responses in AUDi, and their absence in ACCi activity. (H-K) Pre-1012 trial activity dynamics. (H) Correlation of average pre-trial activity (5s preceding trial onset) in 1013 representative co-recorded ACCp/OFCp (left)

Bottom panel depicts the magnitudes 1015 of pre-trial ACCp activity and corresponding OFCp, ACCi or AUDi activity during individual 1016 consecutive trials in a representative session. (I) Coefficient of determination (R-squared) of the linear 1017 fit between pre-trial activity (5s prior to trial onset) in co-recorded channels. (J) Frequency of ultra-1018 slow oscillations of pre-trial activity in ACCp (n=20

Oscillation frequency was defined as the peak of the frequency spectrum emerging above a 1020 threshold obtained from 1000 shuffles of the data (see Methods). (K) Division of pre-trial ultra-slow 1021 fluctuations in ACCp mice according to strategy (see Figure 4). Arrow points to mice with no 1022 significant oscillation, all associated with the erratic group

Claustral OFCp activity during sleep. (A) Average OFCp claustrum calcium activity in 1026

Note that in this population, activity during NREM sleep is 1027 not higher than in wakefulness. (B) OFCp baseline activity (y-axis) for trials associated with maintained 1028 sleep (left) vs. awakening (right)