Intake: Caffeine

Caffeine is an antagonist at the adenosine A1 and A2A receptors, occupying them without switching them on. The A2A subtype is the one human sleep-EEG and pharmacogenetic evidence points to. Blocking the signal is not the same as clearing it, so when the caffeine wears off the accumulated adenosine is still there, which is the flat hour in the late afternoon.

The largest curated dataset pools 141 publications, 500 subject groups and 4,714 individuals, and its authors decline to publish one canonical half-life because between-person variation is too large for an average to describe anyone. Any single figure, the familiar five hours included, is a point pulled from a wide distribution. What the data does pin down is which things move that number, which is the part you can act on.

In the original 26-person study, mean salivary half-life was 3.5 hours in smokers against 6.0 in non-smokers, with body clearance 155 against 94 mL/kg/h. The direction is reproduced in the modern pooled dataset four decades later. It rests on intact groups, not randomization, and smokers differ from non-smokers in diet, alcohol and habitual dose, so the design alone cannot separate the smoke from the smoker; the enzyme mechanism is why it still holds.

In matched women, mean elimination half-life was 7.88 hours on low-dose estrogen oral contraceptives against 5.37 in controls, about 47% longer, driven by slower clearance. It is 18 women on 1985 formulations, and modern pills carry less estrogen, so the exact size is an extrapolation while the direction is corroborated in the pooled data. It should not be read onto progestogen-only or non-oral contraception, which were not tested.

Working in cultured skeletal-muscle cells, researchers showed caffeine raised the autophagy marker LC3b-II and drove vesicle formation through calcium-triggered AMPK activation, and blocking either AMPK or the upstream calcium kinases shut the effect down. AMPK and autophagy are central to how cells handle energy stress and clear damaged parts, which is why this pathway draws longevity interest. It sits at the preliminary, mechanistic tier: the concentrations, timing, and single cell type in a dish do not tell us what caffeine does to lifespan or disease in animals or humans.

Pooling 24 controlled crossover studies: total sleep time down 45.3 minutes, sleep efficiency down 7.0%, onset latency up 9.1 minutes, wake after sleep onset up 11.8 minutes, light N1 sleep up 6.1, and deep N3 plus N4 sleep down 11.4. Most studies used polysomnography, three used activity monitors that overestimate sleep, and the deep-sleep figure rests on 127 participants. The authors caution it may not generalize to teenagers, over-65s, caffeine-naive or heavy habitual users.

Two independent meta-analyzes found no effect on REM: duration mean difference -4.4 minutes (95% CI -10.5 to 1.6, p = 0.127) and REM proportion essentially zero. This is an absence of a detected effect in modestly sized pools, the REM duration analysis resting on 163 participants, not proof none exists, but it is why late caffeine feels survivable while it is costing deep sleep.

Across 22 controlled crossover trials and 956 participants, all polysomnography, total sleep time fell 34.67 minutes, sleep efficiency 4.74%, slow-wave sleep proportion 1.01%, onset latency rose 8.35 minutes and REM proportion was unchanged. Its age and dose subgroup contrasts did not reach significance, so the age and dose angle is weaker than the title suggests; its value is convergence with the 2023 pool about ten minutes apart.

A meta-regression across 30 effect sizes and 262 participants recovered 2.8 minutes of sleep per extra hour before bed and lost 0.2 minutes per extra milligram, giving cut-offs of none for black tea (47 mg), 8.8 hours for coffee (107 mg) and 13.2 hours for a 217.5 mg pre-workout dose. These are group means projected onto a 10 p.m. bedtime, and individual half-life runs from about 2 to 10 hours.

In 23 men measured with at-home polysomnography, 100 mg produced no significant effect on any objective or subjective sleep outcome at any timepoint tested, including four hours before bed. A null in 23 young, caffeine-tolerant men is a small trial not finding an effect, not proof of none, and it does not describe women on the pill or in pregnancy, whose clearance is slower.

In the same 23-man crossover, 400 mg four hours before bed cut total sleep time by an estimated 50.6 minutes and deep N3 sleep by 29.7 minutes (both p < 0.001) and dropped perceived quality 34%. At 8 and 12 hours before bed it still altered objective sleep while perceived quality was intact, which is the most practical finding: the cost is taken whether or not you feel it. 400 mg is a deliberately high test dose.

Drawing on genome-wide data for caffeine metabolism (9,876 people) and sleep traits in up to 453,379, faster metabolism showed no clear effect on sleep duration (p = 0.603) or insomnia (OR 1.040, p = 0.248), though it reduced daytime napping. This estimates lifelong average exposure, so it is nearly silent on what one afternoon cup does, and the sleep outcomes are self-reported in European-ancestry cohorts.

400 mg taken 0, 3 or 6 hours before bed cut measured total sleep time by 1.1 to 1.2 hours at every timepoint, but by sleep diary the six-hour dose cost only 41 minutes and missed significance, so people substantially under-reported a real loss. It was funded by a sleep-technology company, Zeo Inc., and a co-author was its vice president for scientific affairs, so we give it with the funding attached; larger recent work points the same way and puts one coffee's cut-off nearer nine hours.

In the 23-man dose-and-timing trial, there was no significant main effect of CYP1A2 rs762551 or of the adenosine-receptor variant ADORA2A rs5751876 on any objective or subjective sleep outcome. The authors state it was not powered to detect genotype effects, so this is a small study not finding a difference, not showing there is none.

Both concern ADORA2A c.1083T>C (rs5751876). A 2007 controlled sleep-EEG study found caffeine disturbed sleep only in C-allele carriers; a 2017 population study found the caffeine-sleep association only in T-allele carriers. The designs differ, a controlled dose with EEG versus self-reported intake with one night of polysomnography, both results stand, and neither has been replicated in a design able to settle the direction.

Pooling 16 randomized trials of at least a week and 1,010 people, systolic blood pressure rose 2.04 mmHg and diastolic 0.73 overall; caffeine alone gave a larger rise (4.16 over 2.41) than coffee (1.22 over 0.49), with negligible heart-rate change. This is regular intake over a week or more, not one cup, and the authors call the blood-pressure effect of coffee minor. It does not describe people with an arrhythmia or poorly controlled hypertension, who were not studied.

Two large UK Biobank analyzes point the same way. In the JCEM study of 172,315 disease-free adults, about 3 cups of coffee a day carried a hazard ratio of 0.519 for new-onset cardiometabolic multimorbidity and 200 to 300 mg of caffeine a hazard ratio of 0.593, so moderate intake sat near the bottom of the risk curve. A companion multi-state study of 185,112 people traced the whole path of disease and found coffee drinkers were less likely to develop a first condition and less likely to progress from heart disease or stroke to a second one. Because both draw on the same cohort and watch habits, not assigning them, the number is an association at a single population, best read as promising and awaiting replication elsewhere.

Pooling more than 720,000 people and 30,000 atrial fibrillation events, the 2022 meta-analysis found each extra cup a day tied to a relative risk of 0.98, with risk estimates from 1 to 7 cups drifting slightly downward, not up. The per-cup drop is small and its confidence interval reaches 1.00, and the dose-response is not clearly linear, so the solid message is reassurance: habitual coffee does not increase atrial fibrillation risk, reversing decades of caution. The one real limit is dose. Caffeine above about 400 mg in a sitting can transiently raise blood pressure, so this applies to steady moderate coffee, not large boluses, and caffeine-sensitive people may still notice palpitations.

Caffeine is an adenosine-receptor antagonist. Adenosine builds up in the brain across a long or sleepless day and signals fatigue by slowing neural firing; caffeine blocks those receptors, so the brakes on arousal and attention come off. That is why the measurable benefit is largest precisely when adenosine is high, meaning during sleep deprivation and toward the end of a long shift. Because the receptor system adapts with regular exposure, habitual heavy users get a smaller acute lift than occasional users, and part of a morning coffee's effect is reversing overnight caffeine withdrawal.

The JAMA cohort followed more than 131,000 US nurses and health professionals for up to 43 years, one of the longest running looks at this question. Those in the top quarter of caffeinated-coffee intake developed dementia at about 141 per 100,000 person-years versus 330 in the lowest quarter, a hazard ratio of 0.82. They also reported less subjective memory decline and scored modestly better on cognitive tests. The signal was strongest around 2 to 3 cups daily and was specific to caffeinated coffee and tea; decaf showed nothing. Set against this, a 2018 dose-response meta-analysis of eight cohorts found no straight-line relationship per cup for dementia or Alzheimer's, which is why moderate intake, not maximum intake, is the fair takeaway.

Caffeine's ergogenic effect comes mainly from the central nervous system: by blocking adenosine receptors it lowers the perceived effort and pain of a hard effort, so athletes can hold a higher pace before hitting their subjective limit. Older explanations centered on caffeine sparing muscle glycogen by mobilizing fat, but current evidence points more strongly to the brain and to effects on muscle calcium handling and motor-unit recruitment. Timing is usually about 45 to 60 minutes before exercise for capsules, and habitual users still benefit though sometimes slightly less.

In the brain, caffeine blocks adenosine A2A receptors, which are concentrated in the basal ganglia, the motor-control circuit that degenerates in Parkinson's. In animal models, blocking A2A receptors protects dopamine-producing neurons and reduces motor deficits, which gives the epidemiology a plausible mechanism and is why A2A antagonist drugs have been developed as Parkinson's treatments. That blood caffeine measured before diagnosis tracks the lower risk strengthens the case, but human proof would require a prevention trial that has not been done.

Coffee is a complex drink: beyond caffeine it delivers chlorogenic acids, diterpenes, and other polyphenols with antioxidant and anti-inflammatory activity, and chlorogenic acid in particular is thought to improve glucose handling and insulin sensitivity, which fits the strong and dose-dependent diabetes signal seen even for decaf. For mortality, the association is non-linear, meaning the benefit plateaus and the curve flattens; more coffee stops adding much. Because these are cohort studies, residual confounding from lifestyle and smoking cannot be excluded, and the fair reading is a consistent, biologically plausible association that stops short of proof.

Coffee's liver signal is one of the more consistent in nutrition and shows a dose-response, meaning more cups line up with lower risk down to a plateau. Proposed mechanisms include caffeine's blockade of adenosine receptors on hepatic stellate cells, the cells that drive scarring when activated, plus anti-inflammatory and antioxidant effects of chlorogenic acids and the diterpenes cafestol and kahweol, which influence liver enzymes and detoxification pathways. Some benefit appears with decaffeinated coffee, indicating non-caffeine compounds contribute. Because the data are observational and vulnerable to reverse causation, professional bodies describe coffee as promising for liver health, not yet an established treatment.

Across a dozen observational studies covering nearly 347,000 people and about 8,000 depression cases, higher coffee intake carried a pooled relative risk of 0.76, and caffeine in prospective studies a relative risk of 0.84. The dose-response was J-shaped, bottoming out near 400 mL a day, so more coffee did not mean steadily less risk. Because this watches habits, low mood may itself curb coffee drinking, and the finding speaks to risk of depression in the population, not to treating a diagnosed illness.