Sleep

Abstract

Sleep, which the Medical Subject Headings classify under psychophysiology, is a naturally recurring, reversible state of reduced responsiveness, distinguished from coma or anaesthesia by the ease with which it can be reversed and by its own orderly internal structure. Far from a passive shutdown, it is an active, tightly regulated process built from alternating stages that recur in cycles across the night. This article defines sleep, sets out its architecture of non-REM and REM stages and their electroencephalographic signatures, and explains the two-process model that regulates when and how deeply we sleep. It then surveys the leading accounts of what sleep is for, from memory consolidation and synaptic renormalization to metabolic clearance, examines the neural circuitry switching the brain between waking and sleep, and reviews the cognitive costs of sleep loss, with three interactive demonstrations.

Keywords: sleep, memory consolidation, circadian rhythm

Sleep occupies roughly a third of human life, yet for most of history it was treated as a mere absence of waking, a nightly void in which nothing of interest occurred. That view collapsed in the twentieth century, when the electroencephalogram revealed that the sleeping brain is not quiet but passes through a reliable sequence of distinct electrical states, and the discovery of rapid eye movement sleep showed that some of the most active brain states of the whole day occur while the body lies still (Aserinsky & Kleitman, 1953). Sleep is now understood as an actively generated, highly organized behaviour under precise homeostatic and circadian control, and one whose disruption impairs memory, attention, mood, and health (Krause et al., 2017).

What Sleep Is

Sleep is a reversible behavioural state defined by reduced responsiveness to external stimuli, a characteristic posture and reduced movement, and a rapid return to full waking when roused. The last of these criteria is what separates sleep from the pathological or induced states it superficially resembles: a sleeping person can be woken in seconds, whereas coma, general anaesthesia, and hibernation cannot be reversed by a touch or a sound. Sleep is not a single condition but a family of states, each with its own physiology, and the behaviour of being asleep is generated by the brain rather than imposed on it by exhaustion (Saper et al., 2005).

Sleep must also be distinguished from the closely related concept of arousal, the general level of activation of the nervous system, of which the sleep-wake axis is one expression. Waking and sleeping are not two points but a graded continuum of brain states, and the transitions between them can be gradual or abrupt. What makes sleep a scientific object rather than a folk category is that these states have precise, measurable electrical signatures, and it is to those signatures that any account of sleep must appeal.

Types of Sleep

The Medical Subject Headings file sleep, the parent term, above a set of narrower descriptors, each naming an aspect or a subdivision of the sleeping state rather than a wholly separate kind of sleep. The grouping is an indexing convenience rather than a strict theory: the categories are not perfectly parallel, some name a phenomenon of sleep (dreams), some a manipulation of it (deprivation), and some a measurable parameter of it (duration, latency). None of these narrower descriptors yet has its own article on this site, so each is listed below as plain text; the parent field under which MeSH places sleep is psychophysiology.

Two cautions apply to this taxonomy. First, the categories are not mutually exclusive levels of a single dimension: dreams occur within sleep stages, deprivation is defined relative to duration, and latency is measured at the boundary of the whole state. Second, a MeSH tree is a classification built for indexing the biomedical literature, not a settled scientific ontology of sleep, so its subdivisions should be read as practical headings rather than as a claim about the natural joints of the sleeping brain. The scientifically fundamental division, treated next, is the one the tree captures only obliquely under sleep stages: the split between non-REM and REM sleep.

Sleep Architecture and the Sleep Stages

The deepest fact about sleep is that it comes in two radically different kinds. Non-REM sleep and rapid eye movement (REM) sleep differ so thoroughly in their brain activity, physiology, and mental content that they are better thought of as two distinct states than as degrees of a single one (Rasch & Born, 2013). Non-REM sleep is conventionally divided into three stages of increasing depth. Stage N1 is the brief, drowsy transition from waking, marked by a slowing of the electroencephalogram and slow rolling eye movements. Stage N2, which occupies about half of a night's sleep, is defined by two transient waveforms, sleep spindles and K-complexes, riding on a background of moderate activity. Stage N3, also called slow-wave sleep, is the deepest stage, dominated by large, slow delta waves; it is hardest to wake from and is concentrated in the first half of the night.

REM sleep is the physiological opposite of deep non-REM. The electroencephalogram resembles that of waking, low in voltage and fast, yet the sleeper is behaviourally asleep and, crucially, the postural muscles are actively paralysed, a state called muscle atonia that prevents the sleeper from acting out the vivid dreams that characterize this stage. The eyes dart beneath closed lids in the rapid movements that give the stage its name, breathing and heart rate become irregular, and the brain consumes as much energy as it does awake. Hobson framed REM sleep as a state of protoconsciousness, an internally generated virtual reality whose distinctive neurochemistry and cortical activation underlie the bizarre, emotional, narrative quality of REM dreaming (Hobson, 2009).

These states do not occur at random but alternate in an orderly cycle of roughly ninety minutes, repeated four or five times a night. The balance within each cycle shifts systematically: the early cycles are rich in deep N3 slow-wave sleep, while the late cycles are rich in REM, so that most deep sleep is obtained in the first few hours and most REM in the last. This ordered progression, captured in the hypnogram of Figure 1, is what is meant by sleep architecture, and its disruption, whether by sleep loss, ageing, or disorder, is a recurring theme in the study of what sleep accomplishes.

The Two-Process Model of Sleep Regulation

Why we sleep when we do is explained by the most influential framework in the field, the two-process model of sleep regulation (Borbély et al., 2016). It holds that sleep timing and intensity arise from the interaction of two independent controls. The first, Process S, is a homeostatic sleep pressure that builds up steadily during waking and dissipates during sleep. The longer one has been awake, the greater the drive to sleep and the more intense the slow-wave activity of the subsequent night; conversely, sleep discharges this pressure, rapidly at first and then more slowly, which is why the deepest sleep comes early. Process S behaves like a debt that accumulates with wakefulness and is repaid by sleep.

The second control, Process C, is a circadian rhythm, an endogenous oscillation of about twenty-four hours generated by the suprachiasmatic nucleus of the hypothalamus and entrained to the day by light. Process C sets a preferred phase for sleep, promoting alertness during the biological day and sleepiness at night, independent of how long one has been awake. The genius of the model is that the two processes are separate: sleep propensity at any moment reflects the gap between the rising homeostatic pressure and the circadian signal for wakefulness. This explains otherwise puzzling facts, such as the afternoon dip in alertness, the ability to stay awake late into the evening despite mounting sleep debt because the circadian alerting signal is then near its peak, and the poor daytime sleep of shift workers whose Process C is misaligned with their schedule. The interaction of the two processes is treated quantitatively in the Worked Example below.

The Functions of Sleep

If sleep is actively generated and tightly regulated, it must do something important enough to justify the cost of hours of vulnerability and inactivity. Three leading accounts, not mutually exclusive, dominate the modern literature. The first and best supported is that sleep serves memory. Newly encoded memories are strengthened and reorganized during sleep, and this benefit is specific to sleep rather than to the mere passage of time: participants who sleep after learning retain more than those who stay awake for an equal interval (Walker & Stickgold, 2004). The active systems consolidation account holds that during slow-wave sleep the hippocampus repeatedly reactivates newly formed memory traces and replays them to the neocortex, gradually transferring them into stable long-term storage, a process orchestrated by the interplay of slow oscillations, sleep spindles, and hippocampal sharp-wave ripples (Diekelmann & Born, 2010; Klinzing et al., 2019). Sleep does not merely preserve memories but restructures them, extracting regularities and integrating new information with old (Stickgold, 2005).

A second account addresses a problem created by learning itself. If waking experience continually strengthens synapses, their total strength would grow without bound, saturating the capacity to learn and consuming ever more energy and space. The synaptic homeostasis hypothesis proposes that a core function of sleep, particularly slow-wave sleep, is to renormalize synaptic strength, scaling down the connections potentiated during the day so that the brain begins each morning with restored capacity and an improved signal-to-noise ratio (Tononi & Cirelli, 2014). On this view the memory benefit of sleep is partly a by-product of this downscaling, which preferentially preserves the strongest, most salient traces while pruning the weakest.

A third account concerns the brain's housekeeping. Waking metabolism generates by-products that must be cleared, and the spaces between brain cells expand during sleep, allowing cerebrospinal fluid to flush metabolic waste, including the proteins implicated in neurodegeneration, more efficiently than during waking (Xie et al., 2013). These three functions, memory consolidation, synaptic renormalization, and metabolic clearance, are complementary rather than competing, and together they recast sleep as a period of essential neural maintenance rather than idle rest.

Neural Circuitry of Sleep and Wakefulness

Sleep and waking are produced by identifiable brain circuits that switch the whole forebrain between states. Wakefulness is maintained by an ascending arousal system, a set of nuclei in the brainstem and hypothalamus that release neurotransmitters, acetylcholine, noradrenaline, serotonin, histamine, and orexin, to activate the cortex and thalamus. Sleep is initiated when inhibitory neurons, chiefly in the preoptic area of the hypothalamus, suppress this arousal system (Saper et al., 2005). Because the sleep-promoting and wake-promoting populations inhibit each other, the system behaves like an electrical flip-flop switch: mutual inhibition makes intermediate states unstable and drives rapid, complete transitions between waking and sleep, avoiding the dangerous limbo of being half-asleep (Scammell et al., 2017).

The neuropeptide orexin (also called hypocretin) stabilizes this switch, holding it firmly in the waking position during the day. Its loss produces narcolepsy, a disorder of unstable state boundaries in which the switch flips inappropriately, intruding sleep and its features, including the muscle atonia of REM, into waking. The switching between non-REM and REM sleep is governed by a second, faster flip-flop within the brainstem, whose reciprocal populations alternate across the night. This circuit-level understanding connects the abstract architecture of the hypnogram to concrete cellular machinery, and it grounds the two-process model in physiology: Process C acts on the switch through the suprachiasmatic nucleus, while the homeostatic Process S accumulates sleep-promoting signals that bias the switch toward sleep.

Worked Example

The two-process model can be made concrete with the exponential equations that define Process S, and doing so shows why deep sleep is front-loaded into the night. Process S is expressed on a scale from 0 to 1. During sleep it decays toward a lower asymptote of 0 with a time constant of about 4.2 hours, following S(t) = S₀ · e^(−t / 4.2). During waking it rises toward an upper asymptote of 1 with a slower time constant of about 18.2 hours, following S(t) = 1 − (1 − S₀) · e^(−t / 18.2). These time constants are the standard values used in quantitative implementations of the model.

Begin at sleep onset with a high homeostatic pressure of S₀ = 0.90, typical after a long day awake. After the first 2 hours of sleep, S = 0.90 · e^(−2 / 4.2) = 0.90 · e^(−0.476) = 0.90 · 0.621 = 0.559. In just 2 hours the pressure has fallen by 0.341. Continue to the end of an 8-hour night: S = 0.90 · e^(−8 / 4.2) = 0.90 · e^(−1.905) = 0.90 · 0.149 = 0.134. Over the whole night the pressure falls from 0.90 to 0.134, a drop of 0.766, but note how unevenly: the first quarter of the night discharged 0.341 of that debt, nearly half of the total, while the final 2 hours (from 6 to 8 hours) discharge only 0.90 · (e^(−1.429) − e^(−1.905)) = 0.90 · (0.240 − 0.149) = 0.082. Because slow-wave activity tracks Process S, the model predicts that the deepest, most intense slow-wave sleep is concentrated early, exactly as the hypnogram shows.

Now follow the pressure back up through the waking day. Starting the morning at S₀ = 0.134, after 16 hours awake S = 1 − (1 − 0.134) · e^(−16 / 18.2) = 1 − 0.866 · e^(−0.879) = 1 − 0.866 · 0.415 = 1 − 0.360 = 0.640. The pressure has climbed from 0.134 to 0.640 over the day, restoring most of the debt the coming night will again discharge. The asymmetry of the two time constants, fast decay in sleep and slow accumulation in waking, is what allows roughly 8 hours of sleep to offset roughly 16 hours of waking, and the `TwoProcessModelDemo` above lets these curves be traced directly against the circadian rhythm of Process C.

Discussion

Sleep has been transformed over a single century from a blank in the record of behaviour into one of the most structured and heavily regulated processes the brain undertakes. The discovery of REM sleep showed that sleep is not one state but at least two, each as distinct from the other as either is from waking; the electroencephalographic staging of non-REM sleep revealed an orderly architecture; and the two-process model gave that architecture a quantitative, predictive basis in the interaction of homeostatic and circadian controls (Borbély et al., 2016). The circuitry that implements the switch between states is now mapped in detail, connecting the phenomenology of the hypnogram to the cells and transmitters that generate it (Scammell et al., 2017).

The question of function remains the most active and the most consequential. The evidence that sleep serves memory is now strong and mechanistic, tracing consolidation to specific oscillations of slow-wave sleep, and the synaptic homeostasis and glymphatic clearance accounts add complementary reasons why the brain cannot do without it (Diekelmann & Born, 2010; Tononi & Cirelli, 2014; Xie et al., 2013). What unifies these accounts is that they make sleep a service the brain performs for itself, offline, at the cost of engagement with the world, which is why the impairments of sleep loss are so broad: when the maintenance is skipped, memory, attention, and emotional regulation all degrade together (Krause et al., 2017). Sleep is not the opposite of the mind's work but a phase of it.

Current Directions

Much of the current excitement concerns turning the correlation between sleep oscillations and memory into causal control. Because slow oscillations, spindles, and ripples appear to orchestrate consolidation, researchers are testing whether enhancing or timing them, through auditory stimulation locked to the slow oscillation, transcranial stimulation, or targeted memory reactivation with cues presented during sleep, can strengthen specific memories, work that promises both a test of the active systems consolidation theory and a practical tool (Klinzing et al., 2019). The synaptic homeostasis hypothesis is being probed with direct measurements of synaptic density across the sleep-wake cycle, sharpening the debate over how much of sleep's memory benefit is renormalization rather than selective strengthening (Tononi & Cirelli, 2014).

A second front follows the discovery that the sleeping brain clears metabolic waste more efficiently, which has reframed sleep loss as a candidate contributor to neurodegenerative disease and made the glymphatic system an active target of study (Xie et al., 2013). A third pursues the cognitive and affective toll of the chronic sleep restriction typical of modern life, using neuroimaging to map how the sleep-deprived brain misregulates attention and emotion and asking whether the deficits of successive short nights accumulate without full recovery (Krause et al., 2017). Across these threads the guiding aim is the same: to move from describing what sleep looks like to demonstrating, causally, what it does.

Commonly Confused With

Arousal

Arousal is the overall level of activation of the nervous system, ranging from deep sleep through drowsiness to alert waking and high excitement; sleep is one region of that continuum, not a synonym for its low end. To tell them apart, ask whether the fact concerns the graded dimension of activation (arousal) or the specific, actively generated, cyclically structured state entered at night (sleep). A person can be low in arousal yet awake, or high in brain activation yet firmly asleep, as in REM.

Common Misconceptions

Sleep is a passive state in which the brain shuts down.

Sleep is actively generated by dedicated brain circuits and is far from quiet: in REM sleep the brain is as active as in waking, and throughout the night specific oscillations carry out memory consolidation and maintenance (Saper et al., 2005; Rasch & Born, 2013).

Dreaming happens only in REM sleep.

Vivid, narrative dreaming is most characteristic of REM sleep, but mental activity and dream-like experience also occur in non-REM sleep, typically more thought-like and less bizarre. REM and dreaming are closely linked but not identical (Hobson, 2009).

Lost sleep can be fully repaid by one long sleep afterward.

Recovery sleep discharges homeostatic pressure and restores some function, but the deficits of chronic sleep restriction accumulate and are not entirely reversed by a single recovery night, and cognitive performance recovers more slowly than subjective sleepiness (Krause et al., 2017).

Everyone needs exactly eight hours of sleep.

Sleep need varies with age and between individuals and is better thought of as a range than a single number. What the two-process model fixes is not a universal duration but the dynamics by which pressure builds and dissipates (Borbély et al., 2016).

Glossary

Circadian rhythm.

An endogenous biological oscillation of about twenty-four hours, generated by the suprachiasmatic nucleus and entrained by light, that in the two-process model constitutes Process C.

Hypnogram.

A graph of sleep stage against time across a night, displaying the cyclic architecture of non-REM and REM sleep.

Memory consolidation.

The process by which newly encoded, labile memories are stabilized and reorganized into durable long-term storage, much of it occurring during sleep.

Muscle atonia.

The active paralysis of postural muscles during REM sleep, which prevents the sleeper from enacting the movements of a dream.

Non-REM sleep.

The sleep comprising stages N1 to N3, characterized by progressively slower brain activity and reaching its deepest form in slow-wave sleep.

Process C.

In the two-process model, the circadian component that sets the preferred phase for sleep and wakefulness independent of prior time awake.

Process S.

In the two-process model, the homeostatic sleep pressure that accumulates during waking and dissipates during sleep, tracked by slow-wave activity.

REM sleep.

Rapid eye movement sleep, a state of waking-like brain activity, rapid eye movements, muscle atonia, and vivid dreaming.

Sleep architecture.

The organized pattern of sleep stages and their cyclic recurrence across a night, including the front-loading of deep sleep and the late predominance of REM.

Sleep spindle.

A brief burst of oscillatory activity in stage N2 sleep, generated by the thalamus and implicated in memory consolidation.

Slow-wave sleep.

The deepest non-REM sleep (stage N3), dominated by large, slow delta waves and concentrated early in the night; also the stage most tied to homeostatic sleep pressure.

Suprachiasmatic nucleus.

A small hypothalamic structure that serves as the master circadian clock, entraining the body's rhythms to the light-dark cycle.

Synaptic homeostasis hypothesis.

The proposal that a core function of sleep is to renormalize the synaptic strength potentiated during waking, restoring the capacity to learn.

Two-process model.

The framework explaining sleep timing and depth as the interaction of a homeostatic Process S and a circadian Process C.

Key Researchers

Eugene Aserinsky. Physiologist (1921–1998); as Kleitman's graduate student he discovered rapid eye movement sleep in 1953, observing the periodic eye movements that mark the dreaming brain and co-authoring the paper that founded the modern study of sleep stages. Wikipedia - Wikidata

Jan Born. Neuroscientist at the University of Tübingen; his work established the active systems consolidation account of sleep-dependent memory, showing that slow-wave sleep reactivates and redistributes hippocampal memories to neocortex and that slow oscillations play a causal role. ORCID - Faculty Page

William C. Dement. Pioneer of sleep medicine (1928–2020); he characterized the cyclic architecture of NREM and REM sleep across the night, founded the first sleep disorders clinic at Stanford, and helped establish narcolepsy and sleep apnea as clinical entities. Wikipedia - Wikidata

Nathaniel Kleitman. Physiologist (1895–1999) regarded as the father of modern sleep research; he founded the first dedicated sleep laboratory, wrote the foundational Sleep and Wakefulness, and with Aserinsky discovered REM sleep. Wikipedia - Wikidata

Giulio Tononi. Neuroscientist and psychiatrist at the University of Wisconsin–Madison; he proposed the synaptic homeostasis hypothesis, that sleep renormalizes synaptic strength built up during waking, and developed integrated information theory of consciousness. ORCID - Faculty Page - Wikipedia

Matthew P. Walker. Neuroscientist at the University of California, Berkeley; his work maps the cognitive and emotional consequences of sleep and sleep loss, including sleep's role in memory and the effect of deprivation on the amygdala and prefrontal cortex. Faculty Page - Wikipedia

Frequently Asked Questions

What is sleep?
Sleep is a naturally recurring, reversible state of reduced responsiveness to the environment, generated actively by the brain and organized into cycles of distinct stages. It is distinguished from coma and anaesthesia by how readily it can be reversed and from mere rest by its own internal electrical structure (Saper et al., 2005).

What are the stages of sleep?
Sleep has two fundamentally different kinds: non-REM sleep, divided into stages N1, N2, and the deep slow-wave stage N3, and REM sleep, marked by waking-like brain activity, rapid eye movements, muscle atonia, and vivid dreaming. These alternate in cycles of about ninety minutes across the night (Rasch & Born, 2013).

Why do we sleep?
Sleep serves the brain in several complementary ways: it consolidates and reorganizes memory, renormalizes the synaptic strength built up during waking, and clears metabolic waste from neural tissue. These functions recast sleep as essential neural maintenance rather than idle rest (Diekelmann & Born, 2010; Tononi & Cirelli, 2014; Xie et al., 2013).

What is the two-process model of sleep regulation?
It explains when and how deeply we sleep as the interaction of two controls: a homeostatic Process S that builds sleep pressure the longer we are awake and dissipates it during sleep, and a circadian Process C that sets the preferred phase for sleep independent of prior wakefulness (Borbély et al., 2016).

How does sleep help memory?
During slow-wave sleep the hippocampus repeatedly reactivates newly encoded memories and replays them to the neocortex, transferring them into durable storage in a process coordinated by slow oscillations, sleep spindles, and sharp-wave ripples. People who sleep after learning retain more than those who stay awake (Walker & Stickgold, 2004; Klinzing et al., 2019).

What happens in the brain during REM sleep?
In REM sleep the brain shows fast, low-voltage activity resembling waking while the postural muscles are actively paralysed, the eyes move rapidly, and dreaming is most vivid. It has been described as a state of protoconsciousness, an internally generated experience with its own distinctive neurochemistry (Hobson, 2009).

What are the effects of sleep deprivation?
Sleep loss degrades attention, working memory, learning, and emotional regulation, and it impairs the brain's ability to consolidate new memories. The deficits of chronic sleep restriction accumulate across nights and are not fully reversed by a single recovery sleep (Krause et al., 2017).

What controls the switch between sleeping and waking?
Wakefulness is maintained by an ascending arousal system, and sleep begins when inhibitory neurons in the hypothalamus suppress it. Because the two populations inhibit each other, the system acts like a flip-flop switch that makes fast, complete transitions, stabilized in the waking position by the neuropeptide orexin (Scammell et al., 2017).

References