Sleep

Sleep architecture — what each stage actually does

A night of sleep is not a single state. It is a repeating cycle of stages, each running roughly 90 minutes, each doing different physiological work. Polysomnography (PSG = the gold-standard sleep lab measurement, combining EEG, EOG, EMG, and respiratory channels) divides sleep into non-REM stages N1, N2, and N3, plus REM (Rapid Eye Movement) sleep [AASM Scoring Manual, 2023].

N1 is the transitional drift into sleep — a few minutes, easily disrupted, and not particularly load-bearing. N2 accounts for roughly 45–55% of total sleep time and is where sleep spindles and K-complexes appear; these are increasingly linked to memory consolidation and synaptic homeostasis. N3 — slow-wave sleep, often called deep sleep — is where growth hormone pulses, glymphatic clearance of metabolic waste accelerates, and physical recovery concentrates [Xie et al., Science 2013]. REM is where dreaming clusters and where emotional memory processing and creative recombination appear to occur [Walker, Why We Sleep, 2017].

The stages are not interchangeable. Cut sleep short on the back end — a 5 a.m. alarm after a midnight bedtime — and you lose REM disproportionately, because REM density rises through the night. Cut sleep short on the front end — staying up past your circadian sleep gate — and you lose N3, because slow-wave sleep front-loads. Alcohol selectively suppresses REM. Cannabis suppresses REM and N3. Benzodiazepines suppress N3 while increasing N2 — which is why they make people feel rested in a brittle, non-restorative way.

CBT-I — why it beats every pill for chronic insomnia

CBT-I (Cognitive Behavioral Therapy for Insomnia) is the first-line treatment for chronic insomnia in every major clinical guideline, including the American Academy of Sleep Medicine (AASM) and the American College of Physicians [Qaseem et al., Ann Intern Med 2016]. A Cochrane review of CBT-I trials in adults found medium-to-large effect sizes on sleep-onset latency, wake-after-sleep-onset, and sleep efficiency, with durable benefits at one-year follow-up — durability no hypnotic drug class has demonstrated [Cochrane meta-analysis, 2021 update].

The protocol has four components that someone can run without a therapist, though structured programs (Sleepio, CBT-i Coach app, in-person clinicians) are better:

1. Sleep restriction. Set time-in-bed equal to your current average total sleep time (minimum 5.5 hours). Wake at a fixed time daily. When sleep efficiency exceeds 85% for a week, expand time-in-bed by 15 minutes. This is the most effective and most uncomfortable component.
2. Stimulus control. Bed is for sleep and sex only. If awake longer than ~20 minutes, leave the bed, do something low-stimulation in dim light, return when sleepy. Repeat as needed.
3. Cognitive restructuring. Identify and challenge catastrophic beliefs about sleep ("if I don't get 8 hours I'm ruined tomorrow"). These beliefs drive the hyperarousal that maintains insomnia.
4. Sleep hygiene + relaxation. The standard list: consistent schedule, dark cool room, no screens late, caffeine cutoff. Hygiene alone rarely fixes chronic insomnia, but it's a baseline.

For chronic insomnia, the evidence for CBT-I is stronger and more durable than for any hypnotic ever brought to market — and you cannot become dependent on it.

Tiered framework

The supplement stack that holds up

Most "sleep stack" content online is noise. Three compounds have meaningful human trial data and a defensible mechanism.

Magnesium glycinate

Magnesium is a cofactor for GABAergic signaling and NMDA receptor modulation, and intracellular magnesium status is widely under-replete in modern diets. A small RCT in older adults with insomnia found that 500 mg/day of elemental magnesium for 8 weeks improved sleep efficiency, sleep onset latency, and serum melatonin [Abbasi et al., J Res Med Sci 2012]. A 2021 systematic review concluded the evidence is suggestive but limited by small trial sizes and heterogeneous formulations [Mah and Pitre, BMC Complement Med Ther 2021].

Dosing: 200–400 mg elemental magnesium, glycinate or bisglycinate form, 60–90 minutes before bed. Oxide is poorly absorbed and more likely to cause GI distress; citrate is fine but more laxative. Glycine as the chelator is mildly sedating in its own right (see below), which is why glycinate is the consensus form for sleep use.

Apigenin

Apigenin is the flavonoid in chamomile that binds central benzodiazepine receptors at low affinity, producing mild anxiolysis. A randomized trial of chamomile extract (standardized to apigenin) in adults with generalized anxiety disorder showed reductions in anxiety symptoms over 8 weeks [Mao et al., Phytomedicine 2016]. A systematic review on chamomile for sleep concluded modest benefit on sleep quality, though trial quality is mixed [Hieu et al., Phytother Res 2019].

Dosing: 50 mg apigenin (purified) or 400–1600 mg standardized chamomile extract, 30–60 minutes before bed. Apigenin is a CYP enzyme inhibitor in vitro and may affect drug metabolism — a meaningful grey area for anyone on prescriptions metabolized via CYP1A2 or CYP3A4. The human data on this interaction is thin; clinical caution is warranted.

Glycine

Glycine is an inhibitory amino acid neurotransmitter that, taken orally before bed, appears to mildly lower core body temperature — which is itself a sleep-promoting signal. A small RCT in adults with sleep complaints found that 3 g of glycine before bed improved subjective sleep quality and reduced daytime fatigue [Yamadera et al., Sleep Biol Rhythms 2007]. The effect size is modest and the trial pool is small.

Dosing: 3 g (one rounded teaspoon of bulk powder), 30–60 minutes before bed. Sweet taste, mixes in water. Very low side-effect profile.

Melatonin done right — physiologic, not pharmacologic

Melatonin is the most misused sleep supplement in North America. The standard 5–10 mg dose sold over the counter is roughly 10–30 times higher than what the pineal gland produces at peak. Pharmacokinetic studies show that 0.3 mg produces serum melatonin levels in the physiologic range, while 3–10 mg produces supraphysiologic levels that persist into morning, blunt the next night's endogenous pulse, and contribute to grogginess [Zhdanova et al., Clin Pharmacol Ther 1995].

A meta-analysis across 19 trials found that melatonin reduced sleep-onset latency by ~7 minutes and increased total sleep time by ~8 minutes — statistically real, clinically modest [Auld et al., Sleep Med Rev 2017]. For garden-variety adult insomnia, this is not a meaningful effect.

Where melatonin earns its keep is as a chronobiotic — a circadian-shifting agent, not a sedative:

• Jet lag. 0.3–0.5 mg at target-destination bedtime, starting the day of travel, for 3–5 days. Strongest evidence base.
• Delayed sleep phase syndrome. 0.3 mg taken 4–6 hours before current sleep onset, advancing earlier each week. This is a phase-shift use, not a sedation use.
• Shift work disorder. Selective and clinician-supervised; the evidence is more mixed.

Dosing: 0.3 mg, sublingual or fast-release. Most products on the shelf are 3–10 mg, which means either splitting a tablet or sourcing a low-dose product specifically. The popular "extended release" formulations are designed to mimic endogenous secretion but the dose still matters.

Trackers — what Oura, Whoop, and EEG headbands actually measure

Wrist and finger wearables (Whoop, Oura, Apple Watch) infer sleep stages from heart rate, heart rate variability, motion, and skin temperature. They do not measure brain activity. Validation studies against polysomnography show that Oura detects total sleep time and sleep efficiency reasonably well, but stage-classification accuracy — particularly distinguishing N3 from REM — is moderate at best [de Zambotti et al., Behav Sleep Med 2019]. Whoop's published validation shows similar patterns. The trackers are good at trends within a person over time. They are not good at telling you whether last night specifically had 90 or 110 minutes of deep sleep.

EEG-equipped headbands (Dreem, Muse S, Frenz) measure brain activity directly and stage sleep more accurately, but they are uncomfortable for most people to wear nightly and the consumer market for them has been unstable.

What the data is good for: consistency feedback (did I actually go to bed at a regular time this week?), trend detection (sleep latency creeping up over a month suggests something is shifting), and recovery-load context for training. What it isn't good for: chasing a deep-sleep number, diagnosing insomnia, or making clinical decisions. Orthosomnia — the anxiety produced by trying to optimize a sleep-tracker score — is a documented clinical phenomenon [Baron et al., J Clin Sleep Med 2017].

The behavioral foundations

The unglamorous interventions outperform every supplement, and most are free. Stanford neurobiologist Andrew Huberman and UC Berkeley sleep researcher Matthew Walker have both made the case in primary literature: light timing, temperature, and caffeine timing dominate. The order of operations matters.

Light

Bright light in the first hour after waking — ideally outdoors, 10,000+ lux, for 10–30 minutes depending on cloud cover — anchors the circadian phase and sets the timing of melatonin onset that evening. Dim, warm, low-overhead light in the 2–3 hours before bed protects endogenous melatonin onset. Polychromatic short-wavelength (blue-rich) light in the evening suppresses melatonin in a dose-dependent way; whether blue-blocker glasses meaningfully offset this is mixed, but reducing intensity reliably does.

Temperature

Core body temperature drops ~1°C during sleep onset and through the night. A bedroom in the 16–19°C / 60–67°F range supports this. A warm shower or bath 60–90 minutes before bed paradoxically promotes the post-bath drop in core temperature and shortens sleep-onset latency [Haghayegh et al., Sleep Med Rev 2019].

Caffeine

Caffeine has a half-life of 5–6 hours and a quarter-life of 10–12. A 200 mg coffee at 2 p.m. still has ~50 mg circulating at midnight in a typical metabolizer. The conservative cutoff is 8–10 hours before bed; for slow metabolizers (CYP1A2 variants), 12+. The dose-response is real and underappreciated.

Consistency

Fixed wake time, every day including weekends, is the single highest-leverage free intervention. Sleep timing variability — "social jet lag" — independently predicts metabolic and mood outcomes [Wittmann et al., Chronobiol Int 2006]. You can fix bedtime drift, but you cannot fix it by sleeping in.

When to consider a sleep study

Behavioral and supplement work assumes the underlying sleep is structurally sound. When it isn't, no protocol will fix it. Clinical thresholds that should trigger a conversation about polysomnography or a home sleep apnea test (HSAT):

• OSA (obstructive sleep apnea) red flags: witnessed apneas, loud habitual snoring, morning headaches, unrefreshing sleep despite adequate time-in-bed, daytime sleepiness with Epworth score ≥10, BMI ≥30 with hypertension. STOP-BANG ≥3 has good sensitivity for moderate-to-severe OSA [Chung et al., Anesthesiology 2008].
• RLS (restless legs syndrome) features: an urge to move the legs at rest, worse in the evening, relieved by movement. Iron studies (ferritin, transferrin saturation) are the first lab pull; ferritin <75 ng/mL is a treatable contributor [Allen et al., Sleep Med 2018].
• Narcolepsy features: excessive daytime sleepiness with cataplexy, sleep paralysis, or hypnagogic hallucinations. Requires a multiple sleep latency test (MSLT).
• REM behavior disorder: acting out dreams, particularly in adults over 50. This has prognostic implications for neurodegenerative disease and should not be dismissed.

Untreated moderate-to-severe OSA carries cardiovascular, cognitive, and metabolic consequences that no supplement stack will offset. If the red flags are there, the sleep study is the protocol.