The four stages, briefly

A normal night cycles through four stages roughly every 90 minutes, four to six times. The labels are old-fashioned but worth keeping clear because they map to different biology.

  • N1 (light sleep, transition). The handful of minutes between drifting off and "really asleep". Mostly accounting filler - you may not even register it.
  • N2 (light sleep, the bulk). Roughly half the night. Heart rate falls, body temperature drops, and the EEG shows the sleep spindles and K-complexes that are thought to gate sensory input and consolidate procedural memory.
  • N3 (slow-wave sleep, deep sleep). The deepest stage. EEG slows into the 0.5–4 Hz delta band. This is where the glymphatic system flushes the brain, where growth hormone peaks, and where the body does the bulk of its physical-recovery work.
  • REM (rapid eye movement). The dreaming stage. The brain is paradoxically active - close to waking levels - but skeletal muscles are paralysed except for the eyes and respiratory muscles. Where emotional memory is reconsolidated and where neural circuits are remapped.

The proportions in a healthy adult night are roughly: 5% N1, 50% N2, 15–20% N3, 20–25% REM. Deep sleep is concentrated in the first half of the night; REM in the second. A short night does not amputate sleep evenly across stages - it loses disproportionately from REM, because the REM-heavy cycles are at the end.

Deep sleep: the brain's overnight maintenance

The most striking single discovery about deep sleep in the last fifteen years is the glymphatic system. Maiken Nedergaard's lab at Rochester showed in 2013, in mice, that the brain's extracellular space expands by roughly 60% during sleep, and that cerebrospinal fluid then flushes through that space, clearing metabolic waste - including beta-amyloid, the protein that aggregates in Alzheimer's disease [1]. The clearance happens almost entirely during slow-wave sleep; during waking, the extracellular space is too compressed for the bulk flow.

Subsequent human studies have supported the mechanism, though with the caveat that direct measurement is harder. PET imaging in healthy adults shows that a single night of total sleep deprivation increases beta-amyloid accumulation in the right hippocampus and thalamus [2]. Cohort data over decades suggests that chronic short sleep and fragmented sleep are associated with elevated dementia risk - the Whitehall II cohort followed 7,959 British civil servants for 25 years and found that persistent six-hours-or-fewer nights in midlife associated with a 30% increased dementia risk [3]. The causal interpretation has to be cautious - sleep loss is also a symptom of incipient neurodegeneration - but the mechanistic and epidemiological lines converge.

The second function of slow-wave sleep is endocrine. The largest growth-hormone pulse of the 24-hour cycle is locked to the first deep-sleep episode of the night, typically within the first two sleep cycles [4]. Growth hormone drives muscle protein synthesis, mobilises lipids, and supports tissue repair. The decline of slow-wave sleep with age tracks the decline of nocturnal growth hormone release - by age 60, slow-wave sleep duration in men has fallen roughly 70% from young-adult levels, with a proportional drop in nocturnal GH secretion. This is a meaningful share of the "somatopause" of aging.

Deep sleep also consolidates declarative (factual) memory. Sleep-spindle activity during N2 and slow-wave activity during N3 are necessary for the hippocampus-to-neocortex transfer of the day's learning [5]. People deprived selectively of slow-wave sleep - even with total sleep time held constant - show measurable next-day memory deficits.

REM: emotional re-coding and the dreaming brain

REM has the most distinctive neurobiology of any stage. The brain's metabolic rate is comparable to waking, but its neurochemical environment is profoundly different: acetylcholine is high, serotonin and norepinephrine are essentially zero, and the body is paralysed. This particular combination appears to be what allows the unique processing function of REM - the integration of emotional memory without the autonomic arousal that the same content would produce while awake.

The strongest evidence is from selective deprivation studies. People who are repeatedly woken at REM onset (but allowed deep sleep) show next-day deficits in emotional regulation, exaggerated amygdala reactivity to negative stimuli on fMRI, and impaired consolidation of emotional memories [6]. The hypothesised mechanism, sometimes called the "overnight therapy" function, is that REM-state low norepinephrine allows the brain to revisit the day's emotional content and consolidate the memory of what happened while dialling down the somatic charge attached to it.

REM is also where motor-skill consolidation appears to peak [7] - a finding that is robust enough to have changed how athletes and rehabilitation clinics think about training-day sleep. The mortality data on REM is weaker than on deep sleep but suggestive: a 2020 analysis of the MrOS Sleep Study and Wisconsin Sleep Cohort found that, in older men, each 5% reduction in REM percentage was associated with roughly 13% higher all-cause mortality over a median 12-year follow-up, independent of total sleep time [8]. That is correlational and the dose-response is gentle, but it makes the same general point as the deep-sleep literature: stage architecture, not just total time in bed, matters.

What the mortality data actually says

The bluntest finding in the sleep epidemiology is that total sleep time has a U-shaped relationship with all-cause mortality. The Cappuccio et al. 2010 meta-analysis pooled 16 prospective cohorts (over 1.3 million participants) and found that both short sleep (under 6 hours per night) and long sleep (over 9 hours) were associated with 12% and 30% higher mortality respectively, with the nadir of risk near 7 hours [9]. The long-sleep tail is almost certainly inflated by reverse causation - people in poor health sleep longer - but the short-sleep tail has held up under sensitivity analyses.

The Van Dongen et al. controlled experiment is the cleanest evidence that chronic short sleep degrades function more than people realise [10]. Subjects given six hours in bed for fourteen nights showed cognitive-vigilance deficits equivalent to two full nights of total sleep deprivation by the end of the second week. Crucially, self-reported sleepiness barely changed past day five - the subjective sensation of "I'm used to it now" tracks adaptation of self-perception, not adaptation of cognitive function.

The interaction between sleep duration and stage architecture is where the more interesting story sits. Short nights don't lose time evenly across stages - they amputate REM disproportionately, because REM-dominant cycles cluster in the last third of the night. A person sleeping 6 hours instead of 8 may lose 40% of their REM, but only 10% of their deep sleep. That asymmetry shows up in the cognitive consequences: short sleep degrades emotional regulation and memory integration first, motor coordination later, declarative recall later still.

The age trajectory and what you can change

Slow-wave activity declines roughly 2% per decade from age 30 in healthy adults [11]. By age 60, men have lost about 70% of their young-adult slow-wave time; women lose less, but still meaningfully. The largest single anatomical contributor is atrophy of the medial prefrontal cortex, which generates the slow oscillations. Sleep fragmentation - measured as the count of brief arousals - climbs in parallel.

The decline is partially modifiable. Three interventions have replicated evidence for increasing slow-wave activity in middle-aged adults.

Regular aerobic exercise. Three to four sessions a week of zone 2 cardio increases both slow-wave time and EEG slow-wave amplitude in previously sedentary adults [12]. The effect emerges within 6–8 weeks and is dose-dependent. Our zone 2 piece walks through the dose-response curve.

Evening cooling. Core body temperature drops by roughly 1°C from the early evening into deep sleep; that thermal drop is part of the trigger for slow-wave generation. A cool bedroom (16–19°C), a warm shower 60–90 minutes before bed (the post-shower vasodilation accelerates the heat dump), and avoiding heavy late meals all support the thermal trajectory.

Alcohol avoidance in the four hours before sleep. Alcohol is sedating but suppresses both REM and slow-wave sleep, particularly in the first half of the night, and produces a rebound wakefulness in the second half as it metabolises. The effect is dose-dependent: one drink with dinner has measurable but small effects; three drinks within four hours of bed measurably destroys slow-wave architecture for the entire night.

How to read your wearable's sleep-stage data

The honest first point: wearable sleep-stage classification is good enough for trends, not good enough for nightly verdicts. The largest comparative validation studies - the SLEEP cohort comparisons across Whoop, Oura, Garmin, and Fitbit against simultaneous polysomnography - find total sleep time accurate to within 10–20 minutes on average and per-stage agreement of 60–75% at the epoch level [13]. That is excellent compared to what was possible a decade ago, but it means a single-night percentage of deep sleep can be off by several points in either direction.

Three practical rules for reading the data without over-fitting it.

  • Trust the trend, not the night. A 7-day rolling average of deep-sleep percentage is far more meaningful than yesterday's number. Wearable algorithm noise averages out across a week; a real biological shift does not.
  • Sleep efficiency is your most reliable number. Sleep efficiency (the fraction of in-bed time actually spent asleep) is one of the things wearables measure most accurately, and it is also the metric that responds fastest to lifestyle changes. Aim for 88% or higher; rolling averages below 80% are worth investigating.
  • The biggest lever is duration, not architecture. If your wearable shows low deep sleep and you are sleeping six hours, the answer is not a stage-optimisation supplement - it is more time in bed. Stage architecture follows from adequate duration; you cannot optimise around insufficient quantity.

For the broader picture of how sleep stacks into your bio-age trajectory alongside cardiovascular fitness and body composition, our daily-habits piece walks through the effect sizes side by side.

What the evidence does not support

Three claims worth flagging because they show up often and are weakly supported.

  • "Polyphasic sleep is as restorative as monophasic." The controlled studies on Uberman, Everyman, and similar schedules all show progressive cognitive degradation indistinguishable from chronic partial sleep deprivation. The protocols work for a few weeks because of compensatory REM intrusion early in naps, but slow-wave debt accumulates and is not repaid.
  • "I'm a short sleeper and feel fine on 5 hours." The DEC2 short-sleeper mutation is real but rare - well under 1% of the population. Most people who self-identify as short sleepers show normal cognitive deficits on objective testing.
  • "Sleeping pills give you proper sleep." Benzodiazepines and Z-drugs (zolpidem, eszopiclone) suppress slow-wave sleep specifically; subjective recall of sleep improves but stage architecture is degraded, and the cognitive consequences over months are non-trivial. The newer dual orexin antagonists (suvorexant, lemborexant) are stage-neutral but the long-term safety data is still thin.

The takeaway

Sleep stages are not interchangeable. Deep sleep does the brain's overnight metabolic cleanup, drives the night's largest growth-hormone pulse, and consolidates declarative memory. REM consolidates emotional memory, supports motor-skill learning, and is the most fragile to short nights. Total sleep duration is the single biggest lever - the U-shaped mortality curve bottoms out near 7 hours and stage architecture follows from adequate duration. Wearable stage data is reliable as a weekly trend, unreliable on a single night. The interventions with the strongest evidence for increasing slow-wave activity are regular zone 2 cardio, cool bedroom temperature, and avoiding alcohol in the four hours before sleep.

If you want sleep duration, stage architecture, HRV, and the rest of your overnight-recovery markers tracked together and rolled into a coherent bio-age picture, have a look at Thier.

Frequently asked questions

How much deep sleep is enough?

Healthy adults spend roughly 13 to 23 percent of a full night in slow-wave (deep) sleep, which is about 60 to 110 minutes on a 7.5-hour night. Deep sleep percentages decline by roughly 2 percent per decade after age 30, so an absolute target is less useful than a relative one: aim to keep your weekly average from drifting downward. The bigger lever is total sleep duration - short sleep crowds out deep sleep first because slow-wave sleep is concentrated in the first half of the night.

How much REM sleep should I aim for?

REM occupies about 20 to 25 percent of total sleep in healthy adults, roughly 90 to 120 minutes on a 7.5-hour night, concentrated in the last third of the night. REM is the stage most aggressively truncated by alcohol, by waking too early, and by short sleep generally. A two-hour-too-short night does not lose two hours evenly across stages - it loses disproportionately from REM.

Is wearable sleep-stage data accurate?

Compared to polysomnography (the lab gold standard), consumer wearables correctly classify total sleep time within roughly 10 to 20 minutes and distinguish deep, REM, and light sleep with about 60 to 75 percent epoch-level agreement - usable for tracking week-to-week trends but unreliable on a single-night basis. The most reliable signal a wearable produces is your total sleep time, your time-in-bed efficiency, and your wake-after-sleep-onset count; the per-stage breakdown is best read as a smoothed weekly trend, not a nightly verdict.

Why does my deep sleep drop with age?

Slow-wave activity (the EEG signature of deep sleep) declines roughly 2 percent per decade from age 30 onward in healthy adults. The largest contributor is structural change in the medial prefrontal cortex - the region that generates the slow oscillations - which atrophies with age. Sleep efficiency (the fraction of time in bed actually spent asleep) also falls and fragmented arousals increase. The decline is partially modifiable: regular cardio, evening cool-down, consistent bed and wake times, and alcohol avoidance all measurably increase slow-wave activity in middle-aged adults.

Is six hours of sleep enough if I feel fine?

Self-reported alertness after chronic short sleep is unreliable - the largest controlled study (Van Dongen 2003) showed that after two weeks of six-hour nights, performance on cognitive vigilance tasks degraded to the level of two full nights of total sleep deprivation, even though subjects rated themselves only mildly impaired. The published mortality data converges on roughly 7 hours as the optimum, with a U-shaped curve: under 6 and over 9 both carry elevated all-cause mortality. The truly short-sleeper genotype (DEC2 mutation carriers) is real but rare - under 1 percent of the population.

References

  1. Xie L, Kang H, Xu Q, et al. Sleep drives metabolite clearance from the adult brain. Science. 2013;342(6156):373-377. PubMed
  2. Shokri-Kojori E, Wang GJ, Wiers CE, et al. β-Amyloid accumulation in the human brain after one night of sleep deprivation. Proceedings of the National Academy of Sciences. 2018;115(17):4483-4488. PubMed
  3. Sabia S, Fayosse A, Dumurgier J, et al. Association of sleep duration in middle and old age with incidence of dementia. Nature Communications. 2021;12(1):2289. PubMed
  4. Van Cauter E, Latta F, Nedeltcheva A, et al. Reciprocal interactions between the GH axis and sleep. Growth Hormone & IGF Research. 2004;14 Suppl A:S10-17. PubMed
  5. Rasch B, Born J. About sleep's role in memory. Physiological Reviews. 2013;93(2):681-766. PubMed
  6. Goldstein AN, Walker MP. The role of sleep in emotional brain function. Annual Review of Clinical Psychology. 2014;10:679-708. PubMed
  7. Diekelmann S, Born J. The memory function of sleep. Nature Reviews Neuroscience. 2010;11(2):114-126. PubMed
  8. Leary EB, Watson KT, Ancoli-Israel S, et al. Association of Rapid Eye Movement Sleep With Mortality in Middle-aged and Older Adults. JAMA Neurology. 2020;77(10):1241-1251. PubMed
  9. Cappuccio FP, D'Elia L, Strazzullo P, Miller MA. Sleep duration and all-cause mortality: a systematic review and meta-analysis of prospective studies. Sleep. 2010;33(5):585-592. PubMed
  10. Van Dongen HP, Maislin G, Mullington JM, Dinges DF. The cumulative cost of additional wakefulness: dose-response effects on neurobehavioral functions and sleep physiology from chronic sleep restriction and total sleep deprivation. Sleep. 2003;26(2):117-126. PubMed
  11. Carrier J, Land S, Buysse DJ, Kupfer DJ, Monk TH. The effects of age and gender on sleep EEG power spectral density in the middle years of life. Psychophysiology. 2001;38(2):232-242. PubMed
  12. Kredlow MA, Capozzoli MC, Hearon BA, Calkins AW, Otto MW. The effects of physical activity on sleep: a meta-analytic review. Journal of Behavioral Medicine. 2015;38(3):427-449. PubMed
  13. Chinoy ED, Cuellar JA, Huwa KE, et al. Performance of seven consumer sleep-tracking devices compared with polysomnography. Sleep. 2021;44(5):zsaa291. PubMed