Something goes on in a sleeping brain that nobody expected to be able to watch. Cells that were busy while you were awake become busy again in the dark, in roughly the order they were busy the first time, and faster. Researchers call it replay. The question worth asking is whether that word is doing honest work, or whether it is a nice metaphor that escaped from the lab and grew.
It is honest. That is the short version. The rest of this is about what replay actually means, which is stranger and more modest than the phrase suggests.
The belief we all start with
Sleep looks like downtime. You lie down, the lights go off, and the next thing you are aware of is the alarm. Nothing appears to happen in between, so the sensible conclusion is that nothing does: the brain closes for the night the way a shop closes, and reopens in the morning.
It is a fair belief to hold. Being unconscious is a poor vantage point from which to notice activity, and for most of human history there was no way to check. The instruments arrived very late.
What was actually seen
The founding observation came in 1994. Researchers recorded from large ensembles of place cells in the hippocampus of rats, the neurons that fire when an animal occupies a particular spot, across three periods: sleep before a spatial task, the task itself, and the sleep that followed. Cells that had fired together while the rat was in a given location showed an increased tendency to fire together again during subsequent slow-wave sleep, compared with the sleep that came before.1
Two things should be said plainly about that study. It involved three rats, and it demonstrated correlated co-firing rather than ordered sequence: the same cast reassembling, without evidence yet that they were saying their lines in order. It is also correlational. Its real strength is that it is the observation everything else grew out of, and it has been repeated in many laboratories and several species in the thirty years since.
The order arrived two years later. By examining the temporal relationships between pairs of place cells, the same lab showed that the sequence in which cells fired as a rat ran a track carried over into the slow-wave sleep that followed, and did so more reliably than in the sleep that came before.2 What returns in sleep has structure. It runs in the direction it was lived.
That result comes with its own measure of restraint. The effect is a statistical bias measured across many pairs of cells at short latencies, rather than a single trajectory decoded and watched from start to finish, and the paper puts no number on how much quicker sleep runs the sequence than the rat ran the track.
A sharper method arrived in 2002, when researchers decoded long runs of four or more cells firing in their original order during the slow-wave sleep that followed a session. The runs came in brief bursts of roughly a hundred milliseconds, compressing the stretch of behaviour they stood for by something close to twentyfold.3 That figure is where the talk of compression in this piece comes from, and it is still rats.
Mechanism is not the same as proof
Everything described so far sits alongside memory. It does not show that the sitting-alongside does any work. That test came in 2009, and it is the main reason this piece is graded the way it is.
Replay rides on a particular electrical event in the hippocampus, a brief high-frequency burst called a sharp wave-ripple. A group built a closed-loop system that detected these ripples as they occurred during the rest period after training, and selectively eliminated them. Rats that lost their ripples went on to perform worse on a hippocampus-dependent spatial memory task than rats given the same stimulation at other moments.4 Remove the events, and the memory is measurably poorer.
Precision still matters here. What was suppressed was the carrier oscillation rather than the replayed content specifically, so the study establishes that ripples are necessary for the memory and leaves a small gap between that and proving the sequence inside them is the operative signal. It was published as a short brief communication. It remains the cleanest causal evidence the field has.
What happens in people
Nobody drops electrodes into a healthy sleeping human out of curiosity, so the human work came at the question from the other side. Rather than record replay, researchers tried to trigger it, then measured what survived the night.
In 2007 a group did something elegantly simple. An odour was present in the room while participants learned a set of material, and the same odour was re-presented while they slept. When it arrived during slow-wave sleep, retention of that hippocampus-dependent declarative memory improved. When it arrived during REM sleep, or during wakefulness, or to people who had learned without the odour present at all, it did nothing. Nor did it help a procedural task that does not depend on the hippocampus. Imaging showed the hippocampus activating in response to the smell during slow-wave sleep.5
The controls are what make it convincing. A cue that works only in the right sleep stage, only for the right kind of memory, and only for people to whom the cue meant something, is behaving like a mechanism and not an artefact. Worth noting: this cues reactivation and measures the behavioural consequence. It does not record replay in a human directly.
It has since been done a great many times. A meta-analysis pooled 91 experiments, 212 effect sizes and just over two thousand participants, and found a reliable overall benefit from cueing memories during sleep: Hedges’ g of 0.29, with significant effects in stage 2 and slow-wave sleep and none in REM or wakefulness.6 That stage-specificity is a quiet vote of confidence, because it matches what the rodent recordings say about where replay lives.
Now the honest part, which the number carries on its face. A g of 0.29 is a small effect. It is reliable and it points the same way across a large literature, which is what you want from a mechanism. It is not a lever anyone has learned to pull hard. This evidence supports the statement that the process is real and replicable. It does not support the statement that you can meaningfully upgrade your memory overnight.
The moment it stops being abstract
In 2022 the recording was finally made inside a person. A 36-year-old man with tetraplegia, enrolled in a brain-computer interface trial, had two 96-channel intracortical microelectrode arrays in his left precentral gyrus. He played a colour-and-sound sequence-matching memory game, moving a cursor by neural activity alone. Then he slept, with the arrays and a scalp EEG still recording.
Overnight, decoded with exactly the same filter settings used during the game, his brain produced the target sequence again at a rate significantly above chance, at one to four times the speed at which he had ever played it, and most frequently during slow-wave sleep.7 He learned a game in the evening and rehearsed it in the night, faster than he could move, without any idea he was doing it.
Hold it lightly all the same. One participant, a single case, and the recordings come from motor cortex rather than the hippocampus where most replay research lives. As a demonstration it is extraordinary. As a claim about how everybody’s sleep works, it is one man on one night.
Your brain does not switch off at night. It replays the day, sped up, during deep sleep, and researchers have now recorded that happening inside a sleeping human.
What this might mean for your evening
Very little that you need to do, which is usually the honest answer. There is no protocol in any of this, no smell to bottle, no sequence to time. What the work says is simpler: the hours after you stop paying attention are not empty ones, and the deep, early-night portion of sleep is where most of this activity has been observed.
What follows from that is unglamorous. The last hour before bed is the one part of this whole picture you have any say over. Whether how you spend it shows up as something you would notice about your own memory is well beyond what these studies can tell you, and anyone who says otherwise is selling something. But it is a reason to think of the end of a day as a handover rather than a stop.
Mythrae makes audio for that last hour. It is company for the end of a day, not a treatment and not a technique, and we would not claim it does anything to the processes described above. The Sleep room is where that work lives, and the list below is how you hear when it opens. If it is the memory side of this that caught you, the Memory room is where that thread carries on.
References
- Wilson, M. A. & McNaughton, B. L. (1994). Reactivation of hippocampal ensemble memories during sleep. Science, 265(5172), 676–679.
- Skaggs, W. E. & McNaughton, B. L. (1996). Replay of neuronal firing sequences in rat hippocampus during sleep following spatial experience. Science, 271(5257), 1870–1873.
- Lee, A. K. & Wilson, M. A. (2002). Memory of sequential experience in the hippocampus during slow wave sleep. Neuron, 36(6), 1183–1194.
- Girardeau, G., Benchenane, K., Wiener, S. I., Buzsáki, G. & Zugaro, M. B. (2009). Selective suppression of hippocampal ripples impairs spatial memory. Nature Neuroscience, 12(10), 1222–1223.
- Rasch, B., Büchel, C., Gais, S. & Born, J. (2007). Odor cues during slow-wave sleep prompt declarative memory consolidation. Science, 315(5817), 1426–1429.
- Hu, X., Cheng, L. Y., Chiu, M. H. & Paller, K. A. (2020). Promoting memory consolidation during sleep: a meta-analysis of targeted memory reactivation. Psychological Bulletin, 146(3), 218–244.
- Rubin, D. B., Hosman, T., Kelemen, J. N., Kapitonava, A., Willett, F. R., Coughlin, B. F., Halgren, E., Kimchi, E. Y., Williams, Z. M., Simeral, J. D., Hochberg, L. R. & Cash, S. S. (2022). Learned motor patterns are replayed in human motor cortex during sleep. The Journal of Neuroscience, 42(25), 5007–5020.