Too Expensive Not to Matter

Consider what sleep costs an animal.

For hours at a stretch it cannot eat, cannot mate, cannot defend territory, and cannot notice a predator approaching. For a wild animal this is close to the worst arrangement imaginable, and evolution has had several hundred million years to get rid of it.

It has not been got rid of anywhere anybody has looked. Fish sleep. Insects sleep: fruit flies have sleep states that respond to caffeine and to sleep deprivation much as ours do. Nematodes with 302 neurons have a sleep-like state.

Where the cost is intolerable, the solution has been to sleep more cleverly rather than to stop. Dolphins and some birds sleep one hemisphere at a time, keeping one eye open, which lets them keep swimming or flying. Frigatebirds sleep in flight, in short bursts.

Rats kept totally awake die within two to three weeks, faster than they die of starvation.

Something that expensive and that universal is doing something essential, and the honest position is that we still do not fully know what. What is well established is the structure and some of the functions.

Sleep is not one state. It alternates between two that are as different from each other as either is from waking.

In REM sleep the cortex shows fast desynchronised activity that looks much like waking, the eyes move, and the skeletal muscles are actively paralysed by inhibition from the brainstem. That paralysis matters: when it fails, people act out their dreams, which is REM sleep behaviour disorder and can be dangerous. It is also frequently an early sign of Parkinson's disease, appearing years before any movement symptoms.

In deep non-REM sleep the cortex shows large slow waves of synchronised firing, sweeping across the brain in a coordinated rhythm. This is the state most associated with feeling rested, and it dominates the early part of the night.

The cycle runs roughly every ninety minutes, with deep sleep concentrated early and REM increasing before waking. So a shortened night does not remove sleep evenly. It removes REM disproportionately, because that is when REM would have happened.

The best-evidenced function is memory.

During slow-wave sleep, hippocampal place cells that fired in sequence along a route during the day replay the same sequences, compressed in time. This is observable directly in recordings from rodents, and disrupting the replay events impairs the memory afterwards.

In humans, learning followed by sleep beats learning followed by an equivalent waking interval, and the effect is specific rather than general tiredness. The sharpest demonstration is targeted memory reactivation: pair some learned material with a sound or a smell, then present that cue quietly during subsequent deep sleep, and recall for that specific material improves relative to material not cued.

That is a causal manipulation of what gets consolidated while somebody is unconscious, which is a strange thing to be able to do.

The most publicised recent hypothesis concerns waste clearance, and it is worth reporting with its caveats.

Maiken Nedergaard's group reported in 2013 that the interstitial space in the brains of sleeping mice expands substantially, allowing cerebrospinal fluid to flush through more effectively and clear metabolites, including amyloid-beta, which accumulates in Alzheimer's disease.

The glymphatic hypothesis generated enormous attention, partly because it offered a mechanism linking poor sleep to dementia. It has since been challenged: a 2024 study reported reduced rather than increased clearance during sleep, and the measurement techniques are contested. It remains an active area rather than a settled result, and the popular accounts have generally not caught up with the argument.

Two practical findings are on firmer ground.

The first concerns whether people adapt to short sleep. Restriction studies keep participants at four to six hours for a fortnight and measure both performance and subjective sleepiness. Performance on attention and working memory degrades cumulatively, day after day, with no sign of levelling off. Subjective sleepiness rises for a few days and then plateaus.

So people stop feeling worse while continuing to get worse, which is precisely the combination that makes chronic restriction dangerous. The people most confident that they function fine on five hours are the ones the data says are least able to assess it. Genuine short sleepers exist (a mutation in DEC2 and a handful of other genes produces them) and they are rare, and almost nobody claiming to be one is.

The second concerns adolescents, and it is one of the clearest cases of a policy following directly from a biological finding.

Melatonin release shifts roughly two hours later during adolescence and returns earlier in the twenties. This is measurable, cross-cultural, and not a matter of discipline or screens, though screens do not help. Teenagers also need more sleep than adults rather than less.

A school day starting at seven in the morning therefore asks adolescents to be alert at a time their biology has designated as the middle of the night. Districts that have shifted start times later report improved attendance and grades, and several studies have found reductions in adolescent car crash rates.