What is sleep exactly? Global and local modulations of sleep oscillations all around the clock

Wakefulness, non-rapid eye-movement (NREM) and rapid eye-movement (REM) sleep differ from each other along three dimensions: behavioral, phenomenological, physiological. Although these dimensions often fluctuate in step, they can also dissociate. The current paradigm that views sleep as made of global NREM and REM states fail to account for these dissociations. This conundrum can be dissolved by stressing the existence and significance of the local regulation of sleep. We will review the evidence in animals and humans, healthy and pathological brains, showing different forms of local sleep and the consequences on behavior, cognition, and subjective experience. Altogether, we argue that the notion of local sleep provides a unified account for a host of phenomena: dreaming in REM and NREM sleep, NREM and REM parasomnias, intrasleep responsiveness, inat-tention and mind wandering in wakefulness. Yet, the physiological origins of local sleep or its putative functions remain unclear. Exploring further local sleep could provide a unique and novel perspective on how and why we sleep.


From global to local sleep
1.1.The many definitions of sleep "It is impossible to give a definition of sleep that would satisfy everybody" (Jouvet, 1967) Sleep is a complex phenomenon.Its ubiquity in the animal kingdom is matched with a variety of implementations and if all animals seem to sleep, they do not all sleep equally (Cirelli and Tononi, 2008;Siegel, 2008).Consequently, sleep can be defined along several complementary definitions (Pennisi, 2021).In humans, sleep is typically defined along three dimensions: behavior, physiology, and phenomenology (Fig. 1).
At the behavioral level, sleep is characterized by a transient reduction of responsiveness so characteristic of sleep that behavioral responses are often considered to be anecdotal, to reflect arousal, or to signal an abnormal or pathological sleep (e.g., parasomnia) (Carskadon and Dement, 2005).From a mechanistic perspective, this unresponsiveness has long been interpreted as resulting from a form of unconnectedness with the external world.Accordingly, the thalamic gating hypothesis posited that sensory inputs during sleep are blocked at the level of thalamic relays, leaving the neocortex unaffected by sensory inputs and unable to process them (McCormick and Bal, 1994).Yet, we now know that neocortical regions encode and process external stimuli during sleep and that sleepers can respond, overtly or covertly, to external stimuli without waking up (Andrillon and Kouider, 2020).
At the physiological level, sleep is defined by clear changes in brain dynamics and other peripheral physiological variables (muscular tone, ocular and limb movement, heart rate, etc.) (Carskadon and Dement, 2005;Iber et al., 2007).The physiological characterization of sleep developed after the invention of the electroencephalogram (EEG), which allowed the monitoring of neural dynamics non-invasively with a high temporal resolution and for long periods of time.The first EEG recordings in humans showed a change of neural activity at sleep onset (defined behaviorally) with wake patterns such as alpha oscillations ([10-13] Hz oscillations over occipital electrodes and maximal when individuals are awake, relaxed with eyes closed) being replaced by lower frequencies and higher amplitudes, such as theta ([5-8] Hz), or delta ([1-4] Hz) waves (Loomis et al., 1935).
Importantly, the neurophysiological investigation of sleep allowed the discovery that sleep is not monolithic but composed of two main phases: Non-Rapid Eye Movement (NREM) and Rapid Eye Movement (REM) sleep (Carskadon and Dement, 2005;Iber et al., 2007).NREM sleep is defined by the presence of key EEG hallmarks: slow waves and K-complexes (high-amplitude delta waves) or sleep spindles (waxing-and-waving [11-16] Hz transient oscillation) (De Gennaro and Ferrara, 2003;Léger et al., 2018).REM sleep typically follows NREM sleep and is characterized by wake-like brain activity (EEG dominated by low-amplitude, high-frequency desynchronized rhythms) coupled with muscular atonia and phasic muscular events such as the eponymous rapid eye movements (Hobson and Pace-Schott, 2002;Nir and Tononi, 2010).Since the first EEG recordings, the neurophysiological investigation of sleep from whole-brain non-invasive to intracellular recordings have uncovered much of the mechanisms associated with sleep's distinctive features (Steriade, 2003).Finally, sleep in humans can also be viewed through its associated phenomenological properties, especially when considering our firstperson experience of our own sleep.Indeed, waking up from sleep, individuals often report vivid mental experiences with little or no association with the immediate environment (i.e., dreams), but can also report emerging from unconsciousness (Foulkes, 1962).Initial findings proposed a simple association between REM sleep and dreaming on one hand, and between NREM sleep and unconsciousness on the other hand (Aserinsky and Kleitman, 1953;Dement and Kleitman, 1957).While this view is still influential in the literature, numerous studies have depicted a more complex relationship between brain activity and subjective experiences in sleep (Foulkes, 1962;Siclari et al., 2013Siclari et al., , 2020)).Likewise, the transition to or from sleep can also be associated with involuntary, perceptual experiences called hypnagogic or hypnopompic imagery (Hayashi et al., 1999;Ogilvie, 2001).Spontaneous mental experiences are typically explored by asking participants what was in their mind prior to an awakening (spontaneous or forced), an approach often referred to as "experience sampling" or "serial awakenings" (Siclari et al., 2013).

Sleep as a global phenomenon
These different dimensions of sleep are complementary and yet not put on an equal footing in sleep research and medicine.The physiological definition of sleep is dominant for the normative characterization of sleep and its substages (Iber et al., 2007).Importantly, the remarkable conservation of the EEG signatures of sleep and wakefulness across mammal species allows using common physiological definitions across human and animal models (Buzsáki et al., 2013).
At first glance indeed, sleep can easily be identified via the presence of features salient to the human eye in EEG signals (slow waves, Kcomplexes, sleep spindles) or non-EEG signals (muscle atonia in EMG or rapid eye movements in EOG).Crucially, this approach gives little importance to the spatial location of EEG features since it was inspired by the conceptualization of sleep as global and stable physiological state that would affect brain activity as a whole (see (Decat et al., 2021) for a more in-depth discussion of this classical sleep scoring approach).In support of this global vision of sleep was the discovery of the reticular activating system (Siegel, 2002) and its control over wake-sleep transitions through neuromodulatory pathways innervating the entire cortex although recent accounts describe a more complex system of independent and redundant wake-promoting structures (Datta and Maclean, 2007).Consequently, current recommendations for the monitoring of sleep include a large variety of recording types (EEG, EOG, EMG, ECG plus other physiological variables depending on the possible presence of sleep or other disorders) but with a small number of scalp electrodes (minimum 3) (Iber et al., 2007).And until recently, there was very little focus on the local aspects of sleep and the way distinct brain regions may sleep differently.
In contrast to neurophysiology, the behavioral and phenomenological definitions suffer from significant pitfalls such as implying to perturb or interrupt sleep and/or requiring the ability to report on internal states.Yet, these additional dimensions offer a unique opportunity to refine our conceptualization of sleep by examining when they agree, and perhaps more importantly, when they disagree.Indeed, there is a large overlap between the physiological, behavioral, and phenomenological definitions of sleep.Changes at the neuronal level indicative of sleep are closely associated with a loss of responsiveness and reports of dreams or unconsciousness.In particular, early investigations of REM sleep emphasized on the tight relationship between REM sleep and dreaming on the one hand, and NREM sleep and the loss of consciousness on the other hand (Hobson, 1988;Nir et al., 2013), which has spurred models and theories to explain how these neural changes could account for the loss of responsiveness (e.g., thalamic gating hypothesis (McCormick and Bal, 1994)), the fading of consciousness (e.g., breakdown of cortical functional connectivity (Tononi and Massimini, 2008)), or the generation of dreams (e.g., Hobson's AIM model (Hobson, 2009)).Yet, these three dimensions of sleep do not perfectly overlap.For example, if dreams are very frequent during REM sleep, they are also common in NREM sleep (Foulkes, 1962;Siclari et al., 2013).NREM dreams are also resistant to the pharmacological suppression of REM sleep (Oudiette et al., 2012).Thus, the mechanisms of dreaming might only partially overlap with those of REM sleep (Foulkes, 1993;Solms, 2000).Similarly, even if infrequent, behavioral responses during sleep are not completely abolished after sleep onset (Canales-Johnson et al., 2020;Jagannathan et al., 2022;Strauss et al., 2022) or even during consolidated sleep (Konkoly et al., 2021;Türker et al., 2022).We argue that these apparent contradictions could be resolved by reconsidering the vision of sleep as a global phenomenon.

Sleep in a dish: isolated cortex and cell cultures
Experiments on isolated cortical islands or lesioned brains showed that deafferented cortical networks can show rhythmic activity such as alpha oscillations (Bremer, 1938;Kristiansen and Courtois, 1949) or high-amplitude slow waves (Gloor et al., 1977) reminiscent of rest or sleep activity.Likewise, cultures of dispersed neurons can also display sleep-like rhythms (Hinard et al., 2012;Jewett et al., 2015) and this spontaneous activity was proposed to be the default state of developing cortical networks, preceding the emergence and maintenance of wake-like activity (Corner, 2013).This cortical behavior was in stark contrast with neurons in the brainstem that did not show fluctuations of their dynamics or a pattern of high-amplitude slow oscillations (Jouvet, 1967).Given also that no brain lesion results in the general suppression of NREM sleep without quickly leading to death (Krueger and Obäl, 1993), these observations implied that NREM sleep is an auto-rhythmic behavior of cortical neurons that not only does not require an external pacemaker (Bremer, 1949) but that also appears whenever external influences are removed.Of course, this view did not exclude that non-cortical brain regions could control the onset and maintenance of sleep (Jouvet, 1967).
Taking stock of these early studies, Krueger and Obal proposed that the slow waves typically observed in NREM sleep represent a fundamental default mode of cortical activity and that sleep, and its regulation, should be better understood at the local rather than global scale (Krueger and Obäl, 1993).Importantly, the local regulation of sleep was hypothesized to be tightly linked to synaptic plasticity and sleep was fundamentally conceptualized as a use-dependent process that would ensure the maintenance of a synaptic infrastructure that is constantly changed during wakefulness.

Uni-hemispheric sleep and parasomnias
Since sleep entails a drastic reduction of responsiveness, falling asleep can represent a significant risk.For some animals, a complete loss of responsiveness or mobility is simply incompatible with their ecological niche.A classic example regards dolphins, which need to reach the surface in order to breathe and cannot afford the immobility typically associated with sleep (Siegel, 2008).Likely because of this vital evolutionary pressure, dolphins display a peculiar form of sleep called uni-hemispheric sleep, whereby one hemisphere can show classical EEG markers of sleep while the other can show wakefulness at the same time (Mukhametov et al., 1977;Mascetti and Gian Gastone, 2016).These patterns can reverse, allowing both hemispheres to sleep and dolphins to keep swimming and breathing.The absence or negligible presence of REM sleep in dolphins (Lyamin et al., 2008), although still debated (Lyamin and Siegel, 2019), highlights differential adaptations of REM and NREM sleep to aquatic life and could reflect the fact that the vital functions of NREM sleep do not necessitate a global form of sleep.
Uni-hemispheric NREM sleep is not restricted to dolphins (Siegel, 2008;Mascetti and Gian Gastone, 2016) and has been observed in birds on land or in flight (Rattenborg et al., 2001;Rattenborg and Ungurean, 2022).Importantly, avian uni-hemispheric sleep can be flexible and modulated by predatory risks (Rattenborg et al., 1999) or following learning (Nelini et al., 2010).Although uni-hemispheric sleep remains a rare phenomenon whose neural mechanisms and costs/benefits balance remains unclear (Mascetti, 2016), it strikingly shows the adaptability of sleep to the environment.It also demonstrates that sleep is not necessarily a whole-brain event and even full-fledged slow wave sleep can coexist with wakeful activity within the same brain.
Humans do not show an equivalent of uni-hemispheric sleep.Yet, the investigation of sleep disorders offers a unique window into the various ways a human brain can sleep.Parasomnias and dissociative states, in particular, illustrate well the concept of local sleep since they show that, under certain circumstances, sleep and wakeful activity can be simultaneously observed in different parts of the brain (Mahowald and Schenck, 2005;Siclari and Tononi, 2017;Scarpelli et al., 2022a).For example, Disorders of Arousal, which regroup Confusional Arousals (CA), Sleep Terrors (ST) and Sleepwalking (SW) in the International Classification of Sleep Disorders (ICSD-3) (Sateia, 2014), are now conceptualized as partial awakenings from slow-wave sleep (Castelnovo et al., 2018;Idir et al., 2022).Namely, disorders of arousal show the emergence of wake-like patterns of neural activity in some regions (e.g., in the motor cortex and cerebellum which could explain the parallel recovery of mobility) whereas other brain regions still show sleep-like activity (e.g.frontal and parietal cortices) (Bassetti et al., 2000;Terzaghi et al., 2009Terzaghi et al., , 2012)).This could explain many of the cognitive characteristics of these NREM parasomnias (automatic behaviors, lack of agency, poor recall or altered consciousness) and their specific phenomenology (Oudiette et al., 2009;Uguccioni et al., 2013;Idir et al., 2022).Although episodes associated with disorders of arousals are by definition abnormal, local modulations of sleep slow waves have been observed in patients even when no episode or no sign of arousal is observed (Castelnovo et al., 2016), suggesting the existence of a continuum between pathological episodes and healthy sleep.The local sleep framework could also shed light on REM parasomnias (Scarpelli et al., 2022a) such as REM sleep Behavior Disorder (RDB; (Arnulf, 2012)) as RBD patients show heightened levels of fast activity (typically associated with wakefulness) in REM sleep (Valomon et al., 2021).Finally, episodes of sleep paralysis in turn reveal elements of REM activity within wakefulness (Mainieri et al., 2021).

The dynamics of sleep onset
The transition from wakefulness to sleep is a complex dynamic process and provides a unique window on local sleep.At the physiological level, human intracranial recordings have evidenced a staged shift from wakeful activity to the slower activity characteristic of NREM sleep with thalamic regions transitioning first, followed by cortical regions (Magnin et al., 2010).Of particular interest is the duration (approximately 10 min) between the first and last brain regions transitioning from wakefulness to sleep (Magnin et al., 2010).In line with these findings, hippocampal spindles have been consistently observed several minutes before markers of sleep onset in scalp recordings (Sarasso et al., 2014).A gradual transition from wakefulness to sleep could explain the behavioral, cognitive, and phenomenological changes associated with the falling asleep period (see (Ogilvie, 2001;Goupil and Bekinschtein, 2012) for reviews).Importantly, the temporal dynamic of sleep onset might vary across animal species.For example, the transition from wakefulness to continuous trains of slow waves is much faster in rodents, in which subdivisions of NREM sleep are not used, although a recent study suggests rodents might have brief equivalents of N1 and N2 sleep stages (Lacroix et al., 2018).
At the behavioral level, falling asleep is associated with alterations in individuals' connectedness to their environment, their responsiveness and subjective experience.These changes precede the electrophysiological sleep onset (first epoch of N1 or N2) as intermittent responses are observed in drowsy but awake participants (Prerau et al., 2014).Systematic biases in decisions have also been reported in this period (Bareham et al., 2014).The disappearance of behavioral responses is well predicted by markers of the wake/NREM transitions such as the alpha/theta ratio, which captures the replacement of alpha oscillations with slower rhythms (typical of N1) (Goupil and Bekinschtein, 2012;Prerau et al., 2014;Slater et al., 2017;Jagannathan et al., 2022).Protocols involving auditory stimuli and the Event-Related Potentials (ERPs) technique revealed that, during the sleep onset period, individuals can still display markers of complex, and possibly conscious, sensory processing, but these markers fluctuate with responsiveness (Strauss et al., 2015(Strauss et al., , 2022)).However, the dynamics of consciousness fluctuations is complex during this period, since the conscious processing of external events can be in competition with endogenous experiences including perceptual experiences termed hypnagogia (Foulkes and Vogel, 1965;Rowley et al., 1998;Hayashi et al., 1999;Ogilvie, 2001;Goupil and Bekinschtein, 2012).Further investigations are needed to understand how consciousness shifts from an outward to an inward focus during sleep onset.It is possible that the fine-grained spatio-temporal description of brain dynamics could help predict these shifts, stressing here again the importance of local aspects of sleep.

Local modulations of sleep within sleep
Initially revealed in certain animal species, sleep disorders or sleep/ wake transitions, there is now growing evidence that sleep and wake brain activity can overlap in what is usually considered stable and consolidated sleep (N2, N3 or REM sleep).

Slow waves and cortical bistability
Although NREM sleep is not limited to the apparition and multiplication of slow waves, this transition to slower dynamics is arguably one of the most fundamental aspects of NREM sleep.Accordingly, slow waves are one of the core mechanisms to account for sleep's consequences on responsiveness and consciousness (Nir et al., 2013;Andrillon and Kouider, 2020) as well as for sleep's memory function (Diekelmann and Born, 2010;Genzel et al., 2014;Tononi and Cirelli, 2014;Girardeau and Lopes-Dos-Santos, 2021).We will review here the role of these slow waves in the emergence of the concept of "local sleep".
The prominence given to slow waves and NREM sleep in the following sections reflects their importance in the literature, but it is important to stress that slow waves do not equate to sleep (Jouvet, 1967) or even NREM sleep as there are, especially in humans, large portion of sleep deprived of slow waves and, conversely, slow waves have been reported in human and rodent REM sleep (Funk et al., 2016;Bernardi et al., 2019b;Nazari et al., 2023).In addition, slow waves can also be observed in response to brain lesions (Walter, 1936), the administration of drugs (Swank and Watson, 1949), or even mere hyperventilation (Guaranha et al., 2005), independently of sleep.
In rodents, invasive intracranial recordings have allowed to better understand how sleep rhythms are generated and to bridge the gap between different levels of brain activity (neurons, neuronal assemblies, whole brain activity).Scalp slow waves correspond at the neuronal level with an alternation of on-periods (i.e., episodes of neuronal activity: spikes) and off-periods (i.e., neuronal silences) (Steriade, 2003;Vyazovskiy and Harris, 2013).On-periods are phase-locked with slow waves' up-states (positive deflections in scalp recordings) and off-periods with slow waves' down-states (negative deflections).This alternation is underlined by the slow oscillations of membrane potentials, which are explored with intraneuronal recordings (Steriade et al., 1993a(Steriade et al., , 1993b;;Neske, 2016).Thus, the bistable dynamic associated with slow waves perturbs continuous deterministic processes and cortico-cortical dialogue (Massimini et al., 2005(Massimini et al., , 2012;;Pigorini et al., 2015) leading to a loss of responsiveness and consciousness (Tononi and Massimini, 2008).
Importantly, this bistable behavior can be reproduced in vitro (Hinard et al., 2012;Jewett et al., 2015), which implies that the building block of sleep can be generated by dispersed neurons or small cortical networks (Krueger et al., 2008(Krueger et al., , 2019)).Of course, this does not exclude the global orchestration of sleep slow waves through the influence of subcortical structures nor the fact that local modulations at the cortical level could be the expression of heterogeneities in sub-cortical structures (Fernandez et al., 2018;Vantomme et al., 2019).Accordingly, the over-representation of slow waves over somatosensory and auditory cortices in REM sleep appears associated with their relatively lower cholinergic innervations (Nazari et al., 2023).In fact, it has been proposed that there are two distinct mechanisms for the generation of slow waves: global slow waves triggered by subcortical inputs, and local slow waves that are generated by small cortical networks (Siclari et al., 2014).In conclusion, examining the generative mechanisms of slow waves shows the potential for the local generation and regulation of this major sleep hallmark.

Local and travelling waves
Intracranial recordings in humans confirm the existence of local cortical slow waves accompanied by the occurrence of local off-periods (Nir et al., 2011).In fact, most of the intracranially recorded slow waves are local and global waves (visible across all recording sites at once) are the exception rather than the rule.These local slow waves can stay local or propagate across different regions, along a fronto-parietal axis (Nir et al., 2011) which is identical to the main direction of propagation observed for scalp slow waves (Massimini et al., 2004;Nir et al., 2011).A study of split-brain patients showed the importance of the corpus callosum in the propagation of slow waves since the section of the callosal fibers appears to prevent the propagation of slow waves across hemispheres (Avvenuti et al., 2020).Accordingly, the occurrence and propagation of slow waves would depend mostly on cortico-cortical connectivity.
Sleep spindles, a NREM hallmark which arises from the reciprocal interactions between thalamic and cortical neurons (De Gennaro and Ferrara, 2003;Steriade, 2003), can also be generated locally (Andrillon et al., 2011;Nir et al., 2011;Bastuji et al., 2020).Electrocorticographic data highlight the existence of an important diversity in terms of sleep spindles' occurrence and properties across cortical regions (Piantoni et al., 2017).Sleep spindles often occur during slow waves' up-states (Mölle et al., 2002) and propagate in the opposite direction (from the back to the front of the brain) (Andrillon et al., 2011).This propagation pattern as well as the lack of inter-hemispheric differences in callosotomized split-brain patients (Bernardi et al., 2021) suggest that, contrary to slow waves, spindles propagate across thalamic nuclei which then project to different portions of the cortex (Andrillon et al., 2011).The heterogeneity of thalamic nuclei could thus influence the expression of local spindles at the cortical level (Fernandez et al., 2018), which suggests that local aspects of sleep are both under the influence of local cortical and subcortical determinants (Vantomme et al., 2019).
These insights from intracranial recordings show that the activity observed at the level of the scalp does not always reflect the activity of all cortical regions (Fig. 2), although scalp activity is typically considered the ground truth to determine an individual's sleep/wake state.In some cases, scalp and local activity can directly contradict.For example, local arousals, characterized by a recovery of wake-like pattern of activity in a given brain region, have been observed while sleep-like activity was preserved in other brain regions or at the scalp level (Nobili et al., 2011(Nobili et al., , 2012;;Sarasso et al., 2015).Interestingly, in a task involving a finger response, a local arousal in the motor cortex during REM sleep (absent from scalp recordings) was even associated with an overt response (Mazza et al., 2014).This finding reinforces the notion that behavior during sleep could be associated with local awakenings over motor cortices, in line with current models of REM and NREM parasomnias (Idir et al., 2022;Scarpelli et al., 2022a).
Recent intracranial studies have stressed the importance of these dissociations in sleep states between brain regions (Fig. 2).For example, when sleep is scored on individual brain regions, the hippocampus and neocortex often appear in different sleep states (Durán et al., 2018;Guthrie et al., 2022).According to human intracranial recordings in epileptic patients, there is a greater proportion of wakefulness and REM sleep in the neocortex and a greater proportion of N2 sleep in the hippocampus (Guthrie et al., 2022).Of note, this asynchronous sleep includes episodes that can last up to ~30 min.The hippocampus often leads the transition to NREM sleep, which is consistent with the observation of sleep spindles in the hippocampus prior to sleep onset (Sarasso et al., 2014).The origin of these discrepancies are still unclear but could be due to the internal organization of subcortical nuclei that control wake/sleep transitions (e.g., locus coeruleus) and their projection to the rest of the brain (Guthrie et al., 2022).

Circadian and homeostatic regulation of sleep
An influential model of sleep regulation identifies two main processes regulating sleep: a homeostatic (S) and circadian (C) process (Borbély, 1982).This model has been built on the investigation of sleep, sleep deprivation, and circadian desynchronization (Borbély et al., 2016).In brief, these studies show that the likelihood to observe sleep increases with time spent awake (often referred as the build-up of sleep pressure) and that the beginning of the night is rich in Stage 3 NREM sleep (N3) whereas late sleep cycles contain little or no N3 sleep but an increased proportion of REM sleep (ultradian regulation).This distribution of sleep stages in time is also under circadian control and transitions toward NREM sleep are easier at certain times of the day (e.g., early afternoon, evening) whereas transitions to REM sleep are more frequent at the end of the night or in the morning.
The forced desynchronization of sleep/wake cycles with circadian time can allow to tease apart the homeostatic and circadian processes (de la Iglesia et al., 2004).For example, maintaining vigilance in the middle of the circadian night is more difficult than during the circadian day, even when equating time spent awake (Lee et al., 2009).Conversely, certain periods of the day are less favorable to sleep which is illustrated by the phenomenon of the "wake maintenance zone" (Dijk and Czeisler, 1994) and the fact that, prior to the dim light melatonin onset, sleep can be hard to achieve and performance can remain high even after extended periods of wakefulness (Zeeuw et al., 2018).The level of NREM sleep pressure can be tracked using EEG indexes such as slow wave activity (SWA, power in the delta band).SWA has the advantage of being a continuous variable, which quantifies the shift toward slower frequencies.
Analyses of SWA across the day and night show that SWA decreases exponentially during the night, evidencing the dissipation of the homeostatic pressure (Achermann and Borbely, 2003).Conversely, SWA increases during the day despite individuals being awake (Cajochen et al., 1999;Finelli et al., 2000).Importantly, these studies focusing on SWA suggest a faster build-up of sleep pressure over frontal regions (Ferrara, 2002) as well as an increase in SWA during both wakefulness and REM sleep following NREM sleep deprivation (Ferrara et al., 2002).Finally, in a forced-desynchronization paradigm, the examination of individual slow waves properties (occurrence, amplitude, slope) rather than SWA showed regional differences in slow-wave regulation, with frontal electrodes being under a larger homeostatic influence (Ferrara and De Gennaro, 2011) and central and parietal electrodes showing a circadian influence equivalent or larger than the homeostatic one (Lazar et al., 2015).Thus, the homeostatic and circadian regulation of sleep affect cortical activity at a global scale but with regional differences (Krueger, 2020).

Use-dependent mechanisms of sleep regulation
In addition to the global influence of time spent awake (homeostatic) or time of the day (circadian), previous wake activity can also modulate sleep locally.Thus, rats maintained in continuous darkness (conditions under which they are more active) show higher levels of SWA during NREM sleep over somato-sensory cortices (but not visual cortices) than rats exposed to 12 h of light and 12 h of darkness (Yasuda et al., 2005).Continuous exposure to light (24 h) has the opposite effect.Leveraging the uni-hemispheric nature of sleep in dolphins, researchers observed that the sleep deprivation of one hemisphere resulted in a sleep rebound only for the sleep-deprived hemisphere (Oleksenko et al., 1992).In humans, during the night following a difficult motor task, enhanced local levels of SWA over the motor cortices can be observed when compared to a simpler version of the same task (Huber et al., 2004).Similarly, tactile stimulations on the right hand during the day lead to a subsequent increase in SWA over the left motor cortices during sleep (Kattler et al., 1994).The reverse phenomenon also occurs and the immobilization of an arm leads to a decrease in SWA over the contralateral motor cortices (Huber et al., 2006;Murphy et al., 2011).Depriving individuals of visual stimuli even for short periods of time also reduces SWA locally over occipital cortices (Bernardi et al., 2019a).The complexity of a task can play a role since mice who run on a wheel will show less SWA over frontal electrodes during sleep than mice who explore their environment, potentially because the former task is less cognitively demanding (Vyazovskiy et al., 2006).
It has been suggested that synaptic plasticity could be responsible for these local, use-dependent modulations of slow wave activity (Tononi and Cirelli, 2014).Accordingly, the induction of synaptic plasticity with transcranial magnetic stimulation (TMS) or its reduction with pharmacological interventions also impacts subsequent SWA locally (Huber et al., 2007;De Gennaro et al., 2008;Carroll et al., 2019).Yet, local modulations of sleep are not only use-dependent, and both the context and sleep environment can also impact sleep locally.For example, sleeping in a new environment (e.g., a sleep lab) can lead to an asymmetry in the amount of SWA between the two brain hemispheres, with the left hemisphere showing lower levels of SWA within the Default Mode Network (Tamaki et al., 2016).This lateralization could have a functional significance as it was associated with an increased sensitivity to external stimuli.Of note, this asymmetry is no longer present on the second night and could be associated with the more general phenomenon of the "first night effect" (Newell et al., 2012).This inter-hemispheric modulation of SWA is not as complete as the uni-hemispheric sleep observed in some animals but suggests an adaptation of sleep at the hemispheric level to allow for a better ability to respond to external stimuli.

Sleep regulation: from the ground, up
Several influential models have been proposed to account for the transition from wakefulness to sleep and vice versa.Early models such as the Ascending Reticular Activating System (ARAS, (Moruzzi and  Magoun, 1949;Steriade, 1996)) and the hypothalamic "flip-flop" switch (Saper et al., 2001) proposed the existence of global switches allowing whole-brain transitions between wakefulness to sleep.However, more recent works have highlighted the existence of regulators of sleep and wakefulness at different scales (Brown et al., 2012).
Eventually, a global state of sleep could also result from circadian and homeostatic regulations at local levels (Roy et al., 2008), bridging the gap between global and local views of sleep (Rattenborg et al., 2012).Indeed, the local, use-dependent regulation of sleep appears to depend on changes (e.g., synaptic, metabolic, inflammatory) that can occur locally as a result of neuronal activity and local sleep could be a direct response to these local changes.Importantly, these synaptic, metabolic or inflammatory changes regard the functions that sleep was proposed to regulate at the global level, which suggests that local sleep could represent a key element of these global functions (Krueger and Obäl, 1993).According to this view, sleep is first and foremost initiated and regulated locally, to preserve the homeostasis of local cortical networks and, consequently, of the whole brain (Krueger and Obäl, 1993;Tononi and Cirelli, 2014).Overall, the existing literature is in favor of a complex and hierarchical modulation of sleep, with global and local determinants for sleep occurrence and intensity (Szymusiak, 2010;Deboer, 2013;Krueger, 2020).

Dreaming and subjective experience
One of the most fascinating aspects of sleep is the occurrence of dreams (Scarpelli et al., 2022b).A dichotomy has been proposed between dreamful and dreamless sleep, which has often been confounded with the REM/NREM dichotomy (Hobson, 1988).This association dreamful/REM and dreamless/NREM makes intuitive sense since REM's activity is closer to wakefulness (a state typically associated with consciousness) than NREM sleep (Hobson, 2009).However, this distinction fails to account for the frequent occurrence of dreaming in NREM sleep (Foulkes, 1962;Solms, 2000;Siclari et al., 2013;Scarpelli et al., 2022b).To explain the occurrence of NREM dreams, the notion of covert REM sleep was put forth, which suggests that NREM dreams could still occur because of a shift to REM activity that is simply not visible at the scalp level (Nielsen, 2000).This covert REM sleep could be akin to the asynchronous sleep discussed above (Guthrie et al., 2022).Accordingly, increasing REM sleep pressure through REM sleep deprivation reinforced the dreamlike quality of NREM sleep experiences (Nielsen et al., 2005).Yet, dreaming can disappear after forebrain lesions that spare the brainstem generators of REM sleep (Solms, 2000).Thus, REM activity could be simply not necessary for the occurrence of dreaming.
Accordingly, NREM and REM dreams share common neural correlates, which consist in a relative decrease of SWA and an increase of the power in higher frequencies over parieto-occipital regions (Siclari et al., 2017) (Fig. 2).In REM sleep, modulations of higher frequency did not only predict the occurrence of dreaming but could also predict the content of these dreams (Siclari et al., 2017).Relevant to the local sleep framework, intracranial recordings in humans have shown wake-like patterns of activity during phasic REM sleep (i.e., REM sleep with Rapid Eye Movements) (De Carli et al., 2015) or immediately around Rapid Eye Movements (Andrillon et al., 2015), which could be directly associated with the generation of specific oneiric contents.In fact, interfering locally with the activity of motor cortices during REM sleep using tDCS can modulate reports of movements during dreams (Noreika et al., 2020).
Local aspects of sleep could also shed light on some aspects of insomnia since local over-activations in wake-promoting structures or regions often associated with conscious awareness have been reported in individuals with insomnia (Nofzinger, 2004;Kay et al., 2017) in line with the hyper-arousal account of insomnia (Riemann et al., 2010;Buysse et al., 2011).More generally, the local sleep view could account for the subjective feeling of being asleep (Andrillon, 2021).Indeed, there are common discrepancies between the physiological and subjective assessments of sleep (Bonnet and Moore, 1982;Schinkelshoek et al., 2020;Valko et al., 2021).These discrepancies often referred to as "sleep state misperception", are particularly prevalent in insomnia, since about 1 / 3 of individuals complaining of insomnia do not show impaired sleep in PSG recordings (Edinger and Krystal, 2003).The reverse phenomenon (overestimation of sleep) is also frequent, particularly in hypersomnia (Valko et al., 2021).A recent study leveraging experience sampling suggested that participants do not experience the "deepest" sleep in N3 (here defined at the subjective level; (Stephan et al., 2021)), despite this stage being classically referred to as "deep sleep".However, sleepers' judgements about their own sleep could be easier when emerging from REM sleep because of the frequent occurrence of dreams that can be clearly distinguished from mental content experienced during wakefulness (Andrillon, 2021).At the EEG level, increased power in higher frequencies and in the density and amplitude of sleep spindles, particularly over frontal electrodes, were the best predictor of sleepers' feeling of being awake while asleep (Stephan et al., 2021) suggesting again that local modulations of sleep can also account from sleepers' subjective experience.

Sensory processing
To explain sleepers' unresponsiveness to external stimuli and the surprisingly low incorporations of these stimuli into dream reports, it had been proposed that sensory inputs were blocked at the level of the thalamic relays (McCormick and Bal, 1994;Andrillon and Kouider, 2020).Not reaching the cortex, these stimuli would not be processed nor answered.Yet, previous reviews show extremely wide estimates (from 0% up to 90%) of sensory incorporations into dreams (Solomonova and Carr, 2019;Salvesen et al., 2023)).Eventually, content-specific sensory incorporations (Hoelscher et al., 1981;Trotter et al., 1988) or modulations of oneiric contents through external stimulations (Schredl et al., 2009;Okabe et al., 2018) imply that external inputs are processed by the sleeping brain.Similarly, studies on arousal thresholds (e.g., (Formby, 1967)) showed that not all stimuli are processed equally by the sleeping brain.
Accordingly, recordings of primary auditory cortices during sleep show a relatively faithful encoding of simple and complex auditory stimuli during sleep (Hennevin et al., 2007;Sela et al., 2016;Hayat et al., 2022).In fact, the analysis of covert auditory processing during sleep via the analyses of EEG responses to sounds (e.g., event related potentials (Bastuji, 1999) or stimulus reconstruction (Crosse et al., 2016)) revealed the maintenance of relatively complex and flexible processes during NREM (mostly N2) and REM sleep (Bastuji, 1999;Andrillon and Kouider, 2020).Thus, without waking up, sleepers can extract the semantic content of a word (Bastuji et al., 2002), detect semantic, arithmetic or probabilistic violations (Ibanez et al., 2006;Strauss et al., 2015;Strauss and Dehaene, 2019), select and prepare motor responses arbitrarily mapped to a given sound (Kouider et al., 2014;Andrillon et al., 2016), and even form new memories (Andrillon et al., 2017).Likewise, sleepers show different covert responses to particularly relevant inputs such as their own name (Perrin et al., 1999), or the familiarity and emotional tone of a voice (Blume et al., 2017(Blume et al., , 2018)).More recently, it has been shown that lucid dreamers can perform voluntary actions in their dreams (LaBerge et al., 2018;Oudiette et al., 2018) and even dialogue with experimenters during lucid dreaming using ocular codes or facial contractions (Konkoly et al., 2021).This ability to converse within sleep with an external experimenter could even be preserved, to a limited extent, in healthy sleepers in N2 and non-lucid REM sleep (Türker et al., 2022).These findings stress the permeability of sleep and beg the question of the mechanisms allowing or preventing sleepers from processing and/or responding to sensory inputs during sleep (Andrillon and Kouider, 2020).
Local modulations of sleep appear here particularly relevant (Fig. 2).Indeed, as mentioned above, inter-hemispheric modulations of SWA when sleeping in a new environment could make the brain more T. Andrillon and D. Oudiette responsive to external inputs (Tamaki et al., 2016).These local modulations can also unfold in time.For example, sleepers exposed to a sound in NREM sleep typically show an increase in sleep rhythms (slow waves, spindles) following the sound, which is typically interpreted as a sleep protective mechanism (Halasz, 2005;Halász, 2016).However, following this initial protective response, a local decrease in the slow wave and spindle bands overlaps in time and space with motor preparation (Andrillon et al., 2016).It has also been proposed that K-complexes, which often occur in response to sensory stimuli (Halász, 2016) and could represent a window of wakefulness within sleep (Destexhe et al., 2007), would allow sleepers to orientate their attention to auditory inputs (Legendre et al., 2019).At longer time scales, fluctuations of responsiveness and indexes of rich cognitive states during sleep suggest the existence of transient windows of reactivity to external stimuli within sleep (Türker et al., 2022).These windows could be organized in time via infra-slow fluctuations (~0.02Hz) of sleep hallmarks such as sleep spindles (Lecci et al., 2017;Lázár et al., 2019) leading to an alternance between states of low and high susceptibility to external stimuli (Lecci et al., 2017).

Sleep-like responses to stimulations
The ability of cortical networks to generate local slow waves opens the possibility of observing these patterns in a context of global wakefulness, defined by the dominance of wake activity in most brain regions and the maintenance of normal wake behavior.This phenomenon of local sleep intrusions mirrors the local arousals discussed in the previous section.These local sleep dynamics in wakefulness have been first evidence in the visual cortex of monkeys (Pigarev et al., 1997) and the somatosensory cortex of rats (Rector et al., 2005(Rector et al., , 2009)).The latter is organized in columns working as individual functional units processing inputs coming from a specific whisker (Malach, 1994).This allows experimenters to activate individual columns by stimulating one whisker at a time.This perturbational approach revealed that awake animals can display sleep-like (high amplitude slow responses) and wake-like (smaller and faster) responses at the same time in different columns (Rector et al., 2009).Importantly, the sleep-like responses are homeostatically regulated (their occurrence increases with time spent awake) and are associated with vascular changes and impaired responsiveness (Rector et al., 2009;Krueger et al., 2019).
Electric stimulations in somatosensory cortices led to similar results, with an increase in the amplitude and slope of responses with time spent awake, whereas a decrease was observed with time spent asleep (Vyazovskiy et al., 2008).These results are in line with the notion that, as sleep pressure builds up, cortical networks will more and more respond to external stimuli with a bistable response that is usually observed during sleep.These changes in cortical responses correlate with changes in SWA, an index of sleep pressure and are resorbed after a period of sleep (Cajochen et al., 1995;Vyazovskiy et al., 2007).Thus, modifications of neurons' responses to external stimuli seem directly related to changes in spontaneous network dynamics as sleep pressure increases.

Spontaneous slow waves and neuronal silencing
Neuronal sleep-like activity is not restricted to responses to external stimulations.Sleep-like cortical bistability has been observed in single or multi-unit recordings in fatigued monkeys and sleep deprived rats in absence of external stimulation (Pigarev et al., 1997;Vyazovskiy et al., 2011b).Episodes of neuronal silencing occur along spontaneous high-amplitude slow waves (in the delta-theta range) in local field potentials (Vyazovskiy et al., 2011b).These sleep-like intrusions were observed in awake animals performing a task, so in animals that were globally awake from a behavioral and electrophysiological perspective.These intrusions could account for the increase in SWA observed in scalp and intracranial recordings with time spent awake (Cajochen et al., 1995;Vyazovskiy et al., 2009Vyazovskiy et al., , 2011a)).Furthermore, a homeostatic regulation was again reported, with these slow waves increasing with time spent awake and decreasing with time spent asleep (Vyazovskiy et al., 2011b).Overall, these findings strongly suggest that slow waves observed in sleep and wakefulness share the same generative mechanisms.
Yet, it remains unclear if episodes of neuronal silencing are associated with a phenomenon of hyperpolarization of cortical neurons as in sleep, which would require intracellular recordings (Steriade et al., 1993a).In addition, even if slow waves detected in LFP signals were associated with a significant drop in neuronal firing rate (Vyazovskiy et al., 2011b), a recent study suggests that, even during NREM sleep, some delta waves detected at the scalp level are not associated with neuronal silences or do not show a homeostatic regulation (El-Kanbi et al., 2022).A better characterization of slow waves across multiple recording scales is needed to understand when slow waves are associated (or not) with a phenomenon of neuronal silencing.

Local sleep and sleep deprivation
When sleep deprived rats were trained to execute a sugar pellet reaching task, sleep intrusions were associated with behavioral errors: missed trials were associated with more off-periods in frontal but not parietal cortices (Vyazovskiy et al., 2011b).Consequently, the presence of these off-periods could provide a simple mechanistic explanation for the behavioral errors often observed following sleep deprivation (Van Dongen et al., 2011;Vyazovskiy et al., 2011b).Indeed, sudden off-periods in wakefulness could perturb the neural computations performed by a given brain region.If these operations are relevant for the task-at-hand, then off-periods would be shortly followed by an error.Importantly, this interpretation implies that off-periods can be observed in all brain regions as time spent awake increase, but only off-periods occurring in brain regions involved in a task will be predictive of behavior in that task (region-specific effects) (Andrillon et al., 2019;D'Ambrosio et al., 2019).
Intracranial recordings in humans also showed how changes at the neuronal level could account for the behavioral consequences of sleep deprivation.Indeed, in classical vigilance tasks such as the Psychomotor Vigilance Task (PVT), an increase in slow reaction times (classically, >500 ms) is observed after sleep deprivation (Lim and Dinges, 2008).The examination of single-neuron activity in the medio-temporal lobe showed that neurons responding to specific images show a form of neuronal lapse after extended wakefulness: (1) neurons respond with delayed and weaker bursts, (2) these changes typically occur in trials with lapses of attention (misses or slower responses, (Nir et al., 2017)).
The effects of cognitive fatigue and sleep deprivation on neural dynamics also manifest at the scalp level with an increase in SWA or individual sleep-like slow waves (Hung et al., 2013;Bernardi et al., 2015;Avvenuti et al., 2021).There is some heterogeneity in the literature about the definition of sleep-like slow waves in wakefulness, with different studies focusing on different frequency bands (delta, theta or a combination of delta and theta) or amplitudes ranges (selection or not of the highest amplitude waves) (e.g., (Hung et al., 2013;Bernardi et al., 2015;Quercia et al., 2018;Andrillon et al., 2021)).Yet, results are remarkably convergent.These slow waves are predominant over frontal regions (Andrillon et al., 2021), which could be particularly sensitive to the accumulation of sleep pressure (D'Ambrosio et al., 2019).These slow waves also increase in a use-dependent fashion with more slow waves in regions under high use (e.g.motor cortices after hours in a driving simulator) (Hung et al., 2013).Finally, the increase in sleep-like activity predicts the behavioral impairment associated with fatigue, again in a region-specific fashion (Hung et al., 2013;Bernardi et al., 2015;Ahlstrom et al., 2017;Avvenuti et al., 2021).It is worth noting the variety of cognitive impairments that can be explained by these so-called sleep intrusions: motor impairment (Hung et al., 2013;Ahlstrom et al., 2017), failures of response inhibition (Bernardi et al., 2015) or emotional regulation (Avvenuti et al., 2021).Consequently, local sleep intrusions could very well account for the diversity of the cognitive impact of sleep deprivation (Fig. 2).

Local sleep in well-rested individuals
Local sleep intrusions appear relevant beyond the effect of sleep deprivation and could account for the fatigue that builds up during a normal day or following intense cognitive efforts.Indeed, sleep-like slow waves have been characterized in protocols without any sleep deprivation procedure (Quercia et al., 2018;Andrillon et al., 2021).However, it is important to pause here on the way local sleep is defined in wake EEG recordings.Indeed, a classical approach is to detect waves in the delta/theta band and to focus on the waves with the largest amplitude.This approach allows to focus on moments in which EEG signals get closer to sleep dynamics without relying on arbitrary thresholds, such as the 75 µV threshold used in sleep recordings despite its known limitations (Feinberg et al., 1967;Parrino et al., 2009).A disadvantage of this approach is that sleep-like slow waves will always be detected, even when there is no reason to expect them.Thus, the mere presence of slow waves has little explanatory value but the relationship of the detected slow waves with time spent awake, task performance or subjective sleepiness can, on the contrary, indicate whether these slow waves are a good index of fatigue.
As a matter of fact, slow waves detected in well-rested, healthy individuals over relatively short but cognitively demanding tasks (Quercia et al., 2018;Andrillon et al., 2021) do increase with time spent on task, and correlate with subjective and objective (e.g., pupil size) signs of fatigue (Andrillon et al., 2021).At the behavioral level, these slow waves can also predict the occurrence of attentional lapses (Andrillon et al., 2021) or learning impairment (Quercia et al., 2018).Importantly, slow waves can predict two different types of lapses: sluggish responses or missed trials on the one hand, and impulsive responses on the other hand (Andrillon et al., 2021).More precisely, regional effects were here again reported with frontal slow waves being associated with impulsive errors and posterior slow waves with sluggish responses.This is particularly interesting since fatigue and inattention are characterized by an increase in behavioral variability, so precisely an increase in both sluggishness and impulsivity (Drummond et al., 2006;O'Connell et al., 2009).Local sleep intrusions could be responsible for both tendencies since the cognitive consequences of sleep intrusions would depend on the cognitive operations conducted by a given brain region (Andrillon et al., 2019).Accordingly, sleep intrusions in frontal cortices would perturb executive functions leading to impulsivity and sleep intrusions in posterior cortices could perturb sensory integration leading to sluggish responses.
Finally, sleep intrusions were also associated with changes in subjective experiences associated with lapses of attention (Andrillon et al., 2021;Wienke et al., 2021).Indeed, humans struggle at maintaining attentional focus on a given task, especially when the task is repetitive or individuals are unmotivated (Smallwood and Schooler, 2015).This limitation translates into episodes of so-called mind wandering (i.e., thinking about something else than the task at hand) or mind blanking (i.e., thinking about nothing or not remembering any ongoing thoughts) (Andrillon et al., 2019).It has been estimated that we spend up to 50% of our waking like mind wandering (Killingsworth and Gilbert, 2010;Seli et al., 2013) and both mind wandering and mind blanking increase following extended periods of cognitive effort or following sleep deprivation (Poh et al., 2016;Zhang and Kumada, 2017;Jubera-Garcia et al., 2021).Accordingly, sleep-like slow waves could predict occurrences of mind wandering and mind blanking (Andrillon et al., 2021).Furthermore, here again regional effects were observed with mind wandering being associated with larger and steeper slow waves over frontal electrodes, and mind blanking with steeper slow waves over parietal electrodes.The occurrence and gradual expansion of sleep-like slow waves could paradoxically increase functional connectivity across the brain (Rocchi et al., 2022), explaining the association between patterns of global hyperconnectivity and decreased conscious awareness (Mortaheb et al., 2022).
Examining the co-occurrence of these slow waves across scalp electrodes indicate that frontal slow waves are usually constrained to frontal electrodes whereas posterior slow waves are more widespread and often co-occur with frontal slow waves (Andrillon et al., 2021).Thus, it is possible that wake slow waves, just as sleep slow waves, occur generally first in frontal cortices and then propagate toward the back of the brain (Massimini et al., 2004;Nir et al., 2011).Consequently, posterior slow waves would represent a more 'global' phenomenon of local sleep which could explain their association with decreased levels of vigilance (compared to mind wandering) and a pause in the stream of consciousness (mind blanking) (Andrillon et al., 2021).The gradual increase in spatial expanse of local sleep intrusions could foretell an imminent sleep onset and bridge the gap between the new concept of local sleep and the current definitions of sleep onset or micro-sleep (Harrison and Horne, 1996;D'Ambrosio et al., 2019;Hertig-Godeschalk et al., 2019).
Finally, sleep-like slow waves could be under the same neuromodulatory pathways that govern sleep/wake transitions since an increase in dopamine and noradrenaline (which are wake promoting, (Jones, 2005;Sara, 2009)) decreases sleep-like slow waves whereas a larger concentration of serotonin (which could facilitate sleep onset, (Oikonomou et al., 2019)) increases sleep-like slow waves (Pinggal et al., 2022).Importantly, the same neuromodulators are involved in the regulation of top-down attention (Sara and Bouret, 2012;Thiele and Bellgrove, 2018) and are targeted by treatments of attentional deficits (Spencer et al., 2009;Hvolby, 2015).Furthermore, in (Pinggal et al., 2022), the density of slow waves across pharmacological conditions was linked to behavioral changes and markers of inattention, again in a region-specific fashion, with frontal slow waves being associated with a reduction of reaction times and posterior slow waves with more misses.In rodents, acting on the serotoninergic system with psychedelics can induce a state of "paradoxical wakefulness" whereby moving animals display trains of sleep-like slow waves (Bréant et al., 2022).In conclusion, sleep-like intrusions could represent one of the mechanistic links that tie the neuromodulation of attention and vigilance (Sara, 2009;Sara and Bouret, 2012).

Local sleep and synaptic plasticity
Why would sleep intrusions occur in the first place?Two main hypotheses for the origins of local sleep have been put forth.First, it has been proposed that local sleep could be a direct consequence of synaptic plasticity (Tononi and Cirelli, 2014).Indeed, during wakefulness, animals (and Humans) often engage in exploratory behaviors that can lead to the creation and strengthening of synaptic contacts through synaptic potentiation.This tendency has been confirmed by the examination of cytoarchitectural (increase in the number and size of dendritic contacts with time spent awake), genetic (expression of genes involved in synaptic plasticity) and molecular (GluA1-containing AMPA receptors, AMPARs) changes (Vyazovskiy et al., 2008;Maret et al., 2011;de Vivo et al., 2017).Importantly, these changes are reversed in sleep, stressing the key role of sleep in the maintenance of neuronal homeostasis (Tononi andCirelli, 2012, 2014).
Computational modelling suggests that an increase in synaptic weights in cortical neurons could favor the occurrence of slow waves (Esser et al., 2007).Accordingly, staying awake for long periods of time would lead to an increase of synaptic strength that would lead to sleep-like dynamics (Vyazovskiy et al., 2007).The increase in sleep-like activity in wakefulness was also associated with a reduction in externally induced long-term potentiation, which could be interpreted as a phenomenon of saturation of the network (Vyazovskiy et al., 2008).This model would not only explain the time-dependence but also the use-dependence of sleep intrusions.Indeed, networks engaged in a given task would likely undergo synaptic changes that would accumulate overtime and lead to sleep intrusions.This model could also explain why some cortical networks (e.g., prefrontal cortices) tend to show more sleep intrusions compared to sensory cortices: the more plastic and labile a network, the more sensitive it would be to extended periods of activity (and plasticity).

Local sleep and metabolism
A second putative origin of sleep intrusions regards neuronal metabolism (Rector et al., 2009;Richter et al., 2014).There are many small molecules (e.g.adenosine, neuropeptides, etc.) that can regulate sleep/wake transitions or sleep rhythms (e.g.SWA) at a global and local level (Porkka-Heiskanen et al., 1997;Zielinski et al., 2015;Pethő et al., 2019), possibly in response to metabolic changes.Indeed, sustained neuronal activity burns energy, leading to a decrease of metabolic resources and an increase of metabolic waste.For example, recent studies using Magnetic Resonance (MR) imaging and spectroscopy showed that inducing cognitive fatigue through the exertion of executive control for extended periods of time can lead to a decrease in task-related activations and an increase in metabolites such as glutamate (Blain et al., 2016;Strasser et al., 2020;Wiehler et al., 2022).These changes were observed in prefrontal cortices (e.g., lateral prefrontal cortex) involved in sustained attention but not over the primary visual cortex.Importantly, these neurophysiological changes were paralleled with an increase in impulsive economic choices, which could be linked to the role of the lateral prefrontal cortex in decision making (Lopez-Persem et al., 2020).In conclusion, sustained activity can lead to changes in the extracellular concentration of metabolites (e.g., glutamate, adenosine) which could provide a molecular index of fatigue and/or sleep need (Landolt, 2008;Krueger et al., 2019).It is important to note that the synaptic and metabolic hypotheses are not mutually exclusive since an increase in synaptic contacts would increase metabolic needs and could also be signaled through an increase in adenosine (Tononi and Cirelli, 2014).Yet, it is unclear how these metabolic changes would affect in turn neural computations, leading to changes in behavior.
We propose here that local sleep intrusions could represent the missing link between metabolic changes and the associated cognitive consequences.Changes in the ionic or molecular composition of the intra or extracellular environment could indeed change neuronal dynamics toward the expression of sleep-like rhythms.An increase in extracellular adenosine could inhibit cholinergic arousal systems (Brambilla et al., 2005) and promote SWA in sleep, although no changes were observed in normal wakefulness (Radek et al., 2004).In addition, molecules involved in the regulation of inflammation, such as interleukin-1 beta (IL1) and tumor necrosis factor alpha (TNF) also regulate sleep at a global and local scale possibly by acting at the synaptic level (Krueger and Obäl, 1993;Churchill et al., 2008;Krueger et al., 2019).At the intracellular level, the concentration of chloride (Cl-) increases with time spent awake and is associated with subsequent slow wave activity in wakefulness (Alfonsa et al., 2021).Mechanistically, high chloride levels could reduce local inhibition, leading to local bistable dynamics.Finally, reducing chloride concentration prevents the increase of slow wave activity with time spent awake (in the targeted brain region but not globally) and restores behavioral performance (Alfonsa et al., 2021).Overall, although the link between metabolic changes and local sleep intrusions remains at this stage speculative, it provides a tantalizing missing link between the known effect of fatigue on metabolism, neural dynamics, and performance.

Local sleep across time and space
The local sleep framework stipulates that sleep and wakefulness are not homogenous, whole-brain phenomena but can be modulated in time and space at a local scale.We have detailed in the previous sections the spatial dimension of local sleep, but it is important to stress its temporal dimension.Both NREM and REM sleep have phasic components (e.g.Kcomplexes, sleep spindle, rapid eye movements), which modulate sleepers' disconnection at a time scale of the order of the second (Andrillon and Kouider, 2020).In wakefulness, the occurrence of sleep-like slow waves can also account for moment-to-moment fluctuations of behavioral performance and subjective experience (Andrillon et al., 2021).
There is evidence that local aspects of sleep are also structured at slower time scales, of the order of the minute, via infra-slow oscillations (Lecci et al., 2017;Lázár et al., 2019) or cyclic alternating patterns (CAP, (Terzano et al., 1985;Parrino et al., 2012)).Complementing EEG recordings, these slow dynamics can now be explored using fMRI recordings and slow (<0.2 Hz) oscillations of the blood-oxygen-level-dependent (BOLD) activity have been proposed as signatures of local sleep (Song and Tagliazucchi, 2020;Song et al., 2022).Markers of sleep and wakefulness can also be locally regulated by ultradian and circadian processes.In fact, it has been proposed that the alternance between NREM and REM episodes observed in sleep would continue during wakefulness (Kleitman, 1982) and be reflected in markers of NREM or REM in wake EEG (Othmer et al., 1969;Hayashi and Hori, 1990).This superposition of different rhythms playing out at different time scales reflects the exquisitely complex regulation of sleep at the global and local level.

Local sleep across species and individuals
The "fluid boundaries" (Sarasso et al., 2015) between wakefulness and sleep can be drawn along different dimensions depending on many exogenous and endogenous factors.The constraints brought by the environment appear to largely shape the way animals sleep and the local sleep framework could provide an interesting new approach to comparing sleep across species (Rattenborg and Ungurean, 2022).
In Humans, local sleep could provide an interesting account for currently unexplained differences in sleep quality and daytime vigilance.Wake intrusions during sleep could potentially explain sleep-state misperceptions and discrepancies between what subjects report and what global indexes of EEG indicate in terms of wake/sleep states (Andrillon et al., 2020;Stephan et al., 2021).Likewise, during the day, some attentional or vigilance deficits could be due to a larger propensity to experience sleep intrusions.For example, ADHD symptoms could be linked to local sleep given the prevalence of sleep and vigilance disorders in the ADHD population (Andrillon et al., 2019;Furrer et al., 2019Furrer et al., , 2020)).Along this line, lower local levels of SWA over prefrontal cortices during sleep have been associated with higher risk-taking behaviors in wakefulness (Studler et al., 2022), which suggests that local aspects of sleep could explain inter-individual cognitive differences (Hudson et al., 2020).
We need to understand better how the local aspects of sleep are impacted by brain development (Fattinger et al., 2017;Timofeev et al., 2020), how ageing affects the boundaries between wake and sleep (Mander et al., 2017;Mander, 2020), or how brain lesions or tumors can lead to an increase in sleep-like oscillations in neighboring brain regions (Walter, 1937;Magnus and Van der Holst, 1987;Sarasso et al., 2020).In conclusion, the local sleep framework could prove extremely useful when considering not only sleep disorders but the many psychiatric, neurological, and psychological disorders for which sleep disorders seem to play a significant role.

Adaptive and restorative benefits of sleep
The fundamental roots of local sleep might be found in the maintenance of the brain's synaptic and metabolic homeostasis at a local level (Krueger et al., 2016).For example, the coupling between cerebrospinal fluid flow and slow wave activity during sleep (Fultz et al., 2019) could allow a larger removal of metabolic waste in specific brain regions (Xie et al., 2013;Fultz et al., 2019).Likewise, the local increase during sleep of slow waves because of the local proliferation of synaptic contacts during wakefulness could favor the local downscaling and pruning of these synaptic contacts (Tononi and Cirelli, 2014;Maier et al., 2019).
But local sleep could have other benefits.Within sleep, local arousals could allow sleepers' to maintain some processing of sensory inputs in order to sleep safely or wake up when needed (Andrillon and Kouider, 2020).In particular, the modulation of slow wave activity between the two brain hemispheres when sleeping in a new environment (Tamaki et al., 2016) could fulfil the same adaptive function as uni-hemispheric sleep in some birds (Rattenborg et al., 2001).The existence of short windows of responsiveness during sleep (Destexhe et al., 2007;Türker et al., 2022) might be fundamental for sleepers to optimally determine whether staying asleep or transitioning to wakefulness (Halasz, 2005).The generation of local sleep hallmarks (e.g.sleep spindles, slow waves) could also participate to the flexible consolidation of memories through their replay and transfer from hippocampal storage to specific cortical networks (Diekelmann and Born, 2010; Girardeau and Lopes-Dos-Santos, 2021).Interestingly, it is possible to influence this process and target the consolidation of specific memories (Oudiette and Paller, 2013) by leveraging local aspects of sleep (Geva-Sagiv and Nir, 2019; Bar et al., 2020).
Conversely, within wakefulness, sleep intrusions could slow down or even reverse the accumulation of metabolic waste or synaptic contacts.In that sense, local sleep could be seen as a safety switch that would prevent any harmful imbalance in the synaptic or metabolic balances but this function lacks so far empirical support.Sleep intrusions could also facilitate the replay of memories, facilitating memory consolidation or future planning ( Ólafsdóttir et al., 2018).Interestingly, it has been proposed that since these replays would occur in a background of wake activity, they could impact conscious experience and form the basis of mind wandering (O'Callaghan et al., 2021).Finally, sleep intrusions could enable the expression of a different cognitive mode that promotes creative thinking (Lacaux, 2022).Indeed, narcoleptic patients, who spend more time at the border of sleep and wakefulness, are more creative than the general population (Lacaux et al., 2019).This potential creative boost associated with brief sleep intrusions or sleep onsets also applies to the general population (Lacaux et al., 2021).

Beyond slow waves
It is worth noting that the episodes of local sleep intrusions in wakefulness described so far are mostly restricted to the intrusions of sleep-like slow waves and thus share some but not all the properties of sleep.A sharp contrast appears with local modulations of sleep within sleep itself where a genuine overlap between wake and sleep patterns (including slow waves, sleep spindles, etc.) has been observed to such extent that, taken individually, different brain regions would be scored as being in different states (Nobili et al., 2012;Sarasso et al., 2015;Emrick et al., 2016;Guthrie et al., 2022).In wakefulness, events such as sleep spindles have not been reported except at the transition between wakefulness and sleep and in specific brain regions (hippocampus) (Sarasso et al., 2014).It is unclear thus if the term "local sleep" is appropriate to refer to the local intrusion of a pattern, cortical bistability, that is observed in sleep but that does not equate to sleep.
Going forward, it would be interesting to precisely map out which sleep patterns can intrude in wakefulness and which patterns cannot.It is likely that this divide would separate patterns that can be generated by local cortical networks (e.g., slow waves) from those that involve larger networks (e.g., sleep spindles).It is worth noting that events such as hippocampal sharp wave ripples can occur both in sleep and quiet wakefulness (Roumis and Frank, 2015) and local slow waves have been observed in REM sleep (Funk et al., 2016;Bernardi et al., 2019b;Nazari et al., 2023), suggesting that the division between wake and sleep patterns is rather complex.Structural differences between brain regions could determine the degree of local overlap between REM, NREM and wakefulness (Nazari et al., 2023).
Although, local aspects of sleep have mostly been investigated with a focus on NREM sleep, there is a growing literature on local aspects of REM sleep.Yet, the notion of local REM sleep itself is challenging since REM sleep is characterized by both neuronal and peripheral changes (e. g., muscle atonia) (Pace-Schott and Hobson, 2002).Interestingly, a key signature of REM sleep, the Ponto-Geniculo-Occipital Waves (PGO, (Jouvet and Michel, 1959;Gott et al., 2017)), can occur in wakefulness (Brooks, 1968).Furthermore, these wake PGO waves increase with the administration of hallucinogenic drugs (Stern et al., 1972;Brooks, 1975).As hypothesized previously, intrusions of REM sleep features in wakefulness could be associated with the hallucinations experienced by patients with Alzheimer or Parkinson's Diseases (Arnulf et al., 2000;Manni et al., 2002;Sinforiani et al., 2007) similarly as those occurring in narcolepsy (Leu-Semenescu et al., 2011).Beyond pathological cases, it has been proposed that the common occurrence of sharp-wave ripples in REM sleep and wakefulness could account for both dreaming and mind wandering (O'Callaghan et al., 2021), placing dreaming and mind wandering on the same spectrum (Fox et al., 2013;Domhoff, 2018).However, since the very definition of REM sleep remains challenging (Blumberg et al., 2020) and many features of REM sleep are not exclusive of REM sleep (e.g., rapid eye movements, sharp-wake ripples PGO waves), the contour of what can be considered REM sleep intrusions should be further investigated.

Fig. 1 .
Fig. 1.The three main dimensions of sleep in humans.In humans, scientists typically use three main definitions of sleep.Since the discovery of the EEG, the physiological investigation of sleep and wakefulness have identified three separate states: wakefulness, NREM and REM sleep.Yet, sleep can also be assessed based on the presence or absence of specific behavioral patterns.Finally, sleep can also be characterized based on the phenomenological properties of the subjective experience reported by sleepers themselves.These three dimensions of sleep are complementary and not exclusive from each other and can jointly characterize sleep.The axes do not index absolute but relative variations along each dimension.

Fig. 2 .
Fig. 2. Local aspects of wakefulness and sleep.Different forms of local modulations of sleep have been evidenced.Asynchronous sleep refers to the coexistence of wake and sleep activity (as classically defined) in separate brain regions.At sleep onset, different brain regions transition to sleep at different times leading to windows of time where some but not all the brain has transitioned to sleep.During sleep itself, local modulations of slow wave activity have been associated with dreaming, the feeling of being awake, or a recovery of responsiveness.In the more contrasted case of (disorders of) arousal, an isolated brain region can show wake activity in the context of a globally sleeping brain.Conversely, during wake, slow waves can intrude in wakefulness, leading to moments of inattention with behavioral and phenomenological consequences.The main EEG patterns that compose wakefulness (highfrequency desynchronized activity), NREM (slow waves, K-complexes, spindles) and REM sleep (sawtooth waves) are shown.