Sleep and the Brain

The architecture of sleep

Until the middle of the twentieth century, sleep was widely assumed to be a single, uniform condition: the brain quiet, the mind blank, the body at rest. The invention of the sleep laboratory demolished that view. When Aserinsky and Kleitman recorded eye movements and EEG across the night in the 1950s, they found that sleep is not one thing but a rotation between at least two profoundly different states, and that one of them looks, electrically, almost exactly like being awake.

The modern picture divides sleep into two broad categories. Non-REM sleep, or NREM, comprises three stages of progressively deeper sleep, in which the EEG slows and grows in amplitude as cortical populations synchronise. REM sleep, named for the rapid eye movements that accompany it, is the paradoxical state: a fast, low-amplitude, wake-like EEG combined with a body that is essentially paralysed. Across a night, the brain moves between these in an orderly sequence.

Stages are scored from a polysomnogram, which combines EEG with an electro-oculogram for eye movements and an electromyogram for muscle tone. That combination is not a convenience. Because the EEG of REM sleep so closely resembles waking, the eye and muscle channels are what allow the two to be told apart at all.

NREM sleep: N1, N2 and N3

NREM sleep occupies roughly three quarters of a typical adult night. Its three stages form a genuine gradient of depth, measured by how strong a stimulus must be to wake the sleeper.

Older terminology split NREM into four stages; stages 3 and 4 were merged into N3 in the 2007 American Academy of Sleep Medicine criteria. Older papers therefore refer to "stage 4 sleep" for what is now the deepest part of N3.

Slow-wave sleep is where several of the most interesting findings cluster. It is when hippocampal replay occurs, in which sequences of neural firing recorded during learning are re-run, compressed in time, during subsequent sleep. It is also when the coupling between slow oscillations, spindles and hippocampal sharp-wave ripples is tightest, a nesting of rhythms that Rasch and Born and others have argued is the physiological substrate of memory consolidation.

REM sleep and the paradox

REM sleep is one of the genuinely strange objects in biology. Look only at the cortical EEG and you would swear the person is awake: the trace is fast, low in amplitude, desynchronised, exactly the signature of an alert brain. Metabolic rate is roughly at waking levels. The eyes move rapidly under closed lids. Vivid, narrative dreaming is most frequently reported from this stage. And yet the sleeper is profoundly unresponsive, and the skeletal muscles, apart from the eyes and the diaphragm, are actively held in a state of near-total paralysis.

That paralysis, called REM atonia, is not passive relaxation. It is imposed: brainstem circuits actively inhibit the spinal motor neurons. Its purpose appears to be to prevent the motor commands generated during dreaming from being executed. The evidence for that comes from the failure mode. In REM sleep behaviour disorder, the atonia mechanism is damaged, and patients physically act out their dreams, sometimes violently. The paralysis, in other words, is a safety interlock.

Dreaming is not exclusive to REM. Mentation is reported from NREM as well, though it tends to be more thought-like and less narrative and bizarre. The old equation of REM with dreaming was too neat, and it has been substantially revised.

The ninety-minute cycle

Sleep does not proceed in a straight line from light to deep and back. It cycles. A typical adult descends from waking through N1 and N2 into N3, spends time in slow-wave sleep, climbs back up through N2, and enters a REM period. That whole loop takes roughly 90 minutes, and it repeats four to six times across a normal night.

Crucially, the cycles are not identical. The composition shifts systematically as the night goes on.

Deep slow-wave sleep is front-loaded: the first two cycles contain most of it, and by the second half of the night N3 may have disappeared entirely. REM is back-loaded: the first REM period of the night may last only a few minutes, while the last can run to half an hour or more. This has a consequence that is easy to miss. Cutting your sleep short by two hours does not remove a uniform slice of sleep. It removes almost entirely REM sleep and the light N2 that surrounds it, while leaving deep sleep largely intact. Going to bed two hours late does the opposite. The two kinds of short night are not equivalent.

The two-process model

Why do you get sleepy when you do? Alexander Borbely's two-process model, proposed in 1982 and still the standard framework, holds that sleep timing emerges from the interaction of two independent processes.

1. Process C: the circadian rhythm

   An internal clock with a period of close to 24 hours, generated by the suprachiasmatic nucleus of the hypothalamus. It runs independently of how long you have been awake, and it is entrained to the outside world chiefly by light reaching the retina. It produces an alertness signal that rises and falls on a fixed daily schedule regardless of your sleep debt.

2. Process S: the homeostatic drive

   Sleep pressure, which accumulates steadily the longer you stay awake and dissipates during sleep. Adenosine, a by-product of cellular energy use, builds up in the brain across the waking day and promotes sleep; caffeine works precisely by blocking adenosine receptors, masking the pressure without removing it.

3. The interaction

   You fall asleep when the homeostatic pressure is high and the circadian alerting signal is low, and the two normally align at night. When they conflict, you get the familiar pathologies: jet lag, shift work, and the "second wind" of staying up past your usual bedtime, when the circadian signal rises even though sleep pressure is enormous.

The model explains a great deal with very little. It accounts for why a nap reduces sleepiness (it bleeds off process S), why you can be exhausted at 4pm and wide awake at 11pm on the same day (process C), and why recovery sleep after deprivation is deeper rather than merely longer: slow-wave activity in the EEG is the best physiological index of process S, and it rebounds in proportion to how long you were awake.

What sleep is for

Every animal studied sleeps or does something like it, and sleep persists despite being obviously dangerous: an unconscious animal cannot watch for predators. Whatever sleep does, it must be important enough to pay that price. The honest position is that we have several well-supported partial answers rather than one complete one.

Memory consolidation

This is the best-evidenced function. During slow-wave sleep, hippocampal neurons replay the firing sequences that occurred during learning, compressed in time, and this replay is coupled to cortical slow oscillations and thalamic spindles. The prevailing model, reviewed comprehensively by Rasch and Born in 2013, is that this coordinated dialogue transfers newly encoded memories from a fast, temporary hippocampal store to slower, more durable cortical representations. Behavioural studies support it: sleep after learning improves retention, and interventions that boost slow oscillations can enhance that benefit.

Synaptic homeostasis

Tononi and Cirelli's synaptic homeostasis hypothesis offers a competing, or perhaps complementary, account. On this view, waking experience drives a net strengthening of synapses across the cortex, which is metabolically expensive and cannot continue indefinitely. Slow-wave sleep, they argue, globally downscales synaptic strength, preserving relative differences while restoring capacity and improving signal-to-noise. The two hypotheses make different predictions, and the debate between them is genuinely unresolved. Both cannot be wholly right in their strong forms.

Clearance of metabolic waste

Xie and colleagues, publishing in Science in 2013, reported that in sleeping and anaesthetised mice the interstitial space of the brain expands substantially and the exchange of cerebrospinal fluid with interstitial fluid increases, accelerating the clearance of solutes including amyloid beta. This glymphatic clearance is markedly greater in sleep than in waking. The finding is important and has been influential, though the mechanism and its magnitude in humans remain under active investigation, and some subsequent work has challenged aspects of it. It should be presented as a strong and promising line of evidence, not a settled fact. See ventricles and cerebrospinal fluid.

Metabolic and immune function

Sleep restriction reliably impairs glucose regulation and alters appetite-regulating hormones in controlled laboratory studies, and it is associated with reduced antibody response to vaccination. Growth hormone is released in pulses tied to slow-wave sleep. These are not exotic side-effects; sleep is embedded in the body's basic regulatory economy.

The cost of sleep loss

The most reliable and best-replicated consequence of sleep deprivation is not dramatic. It is the lapse: a brief, involuntary failure of attention lasting a second or two, in which the sleep-deprived person simply stops responding. Lapses multiply as sleep debt accumulates, and they are the mechanism behind most of the measured performance cost. On sustained attention tasks such as the psychomotor vigilance test, lapses increase steadily across days of restricted sleep.

What makes this dangerous is a second, well-documented finding: subjective sleepiness plateaus while objective performance keeps declining. In laboratory studies of chronic sleep restriction, people who slept six hours a night for two weeks reported feeling only slightly sleepy, but their performance deficits continued to accumulate. They lost the ability to detect their own impairment. This is why "I function fine on five hours" is such an unreliable claim: the person making it is precisely the person whose self-assessment has been degraded.

Beyond attention, sleep loss impairs the encoding of new material, degrades emotional regulation with amygdala reactivity increasing and prefrontal control weakening, and worsens mood. The effects on learning are bidirectional: a poor night's sleep degrades what you can learn tomorrow and also degrades the consolidation of what you learned today.

Myths about sleep

Related topics