On 17 February 1869, Dmitri Mendeleev nodded off over a game of solitaire. On waking, he found he had solved the conundrum of how the chemical elements could be grouped in a meaningful way. The periodic table, used by every chemist today, was born.

Legend has it that inspiration came to Mendeleev in a dream, although he probably embellished the story for dramatic effect. But most of us will have experienced similar, if less epoch-making, insights following sleep. And our ability to perform complex tasks, such as playing a musical instrument, can be better the morning after a practice session the night before.

Cognitive scientists group phenomena such as learning to solve logical problems, or improving musical skills, under the heading of ‘procedural’ learning. This means learning how to do something, rather than simply accumulating a list of facts. For years, researchers have suspected that one important function of sleep is to strengthen this process. But only now are scientists starting to reveal the specific patterns of brain activity involved. “Sleep is to do with cells and building their connections,” says Giulio Tononi, a psychiatrist at the University of Wisconsin in Madison.

In part, the slow progress so far reflects sleep researchers' long-standing focus on REM, or rapid eye movement sleep (see ‘The mysteries of the night’). This coincides with dreaming, during which our brains show patterns of activity that look very similar to those when we are awake. In addition to the trademark flickering eyeballs, both men and women show signs of sexual arousal during REM.

It is hardly surprising, then, that researchers have been drawn to this all-action phase, rather than to the periods of deeper sleep that surround each bout of REM. This quieter phase is known as slow-wave sleep — after the synchronized waves of electrical activity that can be seen on electroencephalograph (EEG) traces undulating across the brain at fewer than four cycles per second.

But despite years of research, the evidence linking REM sleep to procedural learning is contradictory and confusing1. For example, across species there is no firm correlation between the proportion of sleep spent in REM and learning ability. What's more, people who are taking antidepressants that decrease the amount of REM sleep do not have problems with procedural memory. And rare patients with forms of brain damage that eliminate REM completely, similarly show normal procedural learning.

So in the past few years, researchers interested in procedural learning have started to focus on slow-wave sleep — and in doing so have begun to make progress. Perhaps the most convincing evidence linking the quieter phase of sleep to procedural learning comes from Tononi. His team recently published a paper in Nature2 that for the first time linked local changes in short-wave activity to learning a specific task.

Ride the slow wave

Tononi's human volunteers had 256 electrodes placed over their scalps to take EEG recordings of the electrical activity of brain-cell networks before, during and after sleep. Immediately before the subjects dozed off, they played a computer game requiring them to move a cursor to a position on the screen, believed by the participants to be fixed. But during some sessions, the position was moved for some players, who subconsciously learned how to adapt to this handicap.

Immediately before they dropped off to sleep, the subjects who had learned the compensation task showed heightened slow-wave activity in specific brain areas known to be involved in processing spatial information. This was presumably because of the formation and subsequent strengthening of new connections between nerve cells.

Caught napping: Giulio Tononi's net of sensors recorded hotspots (red dots) in the sleeping brain's electrical activity, revealing a link between slow-wave sleep and learning. Credit: G. TONONI

During slow-wave sleep, this activity continued, gradually dying down. Tononi suspects that this decline in slow-wave activity is important in strengthening procedural learning — perhaps because the waking brain is ‘noisy’, with other signals interfering with the learning process.

After a night's sleep, the volunteers who had learned to compensate for the shifting target were better at the task than they had been the night before. And there was a positive correlation across the subjects between the magnitude of the increase in slow-wave activity and the extent of the sleep-induced gain in performance.

During training, the brain tags a task as something to be improved on later during sleep.

This correlation has excited other researchers in the field3. “This is the first real evidence that the oscillations are for memory. It is very impressive to see the slow waves change locally,” says Jan Born, a neuroendocrinologist at the University of Lübeck in Germany. He and others are now looking afresh at their own experiments on learning, wondering whether it will be possible to tie sleep-induced enhancements to changes in activity in specific parts of the brain.

The past couple of years have brought a rich harvest of findings on sleep and learning that should now be amenable to investigation using EEG and other techniques. Born, for instance, has worked on insights similar to Mendeleev's inspiration about ordering the chemical elements. He asked volunteers to generate a succession of seven sequences of numbers using simple rules. But there was a hidden rule — the second sequence always matched the final one. If the volunteers were allowed sleep between two sessions on the task, they noticed the short cut more quickly4. It seems that their sleeping brains processed information that had already been learned. “There has to be a memory that sleep builds upon,” says Born.

Tag team

The morning after: sleep seems to yield the biggest benefits on the toughest computer tests. Credit: R. STICKGOLD

Robert Stickgold, a psychiatrist at Harvard Medical School in Boston, has perhaps the richest series of findings awaiting physiological investigation. He believes that, during training, the brain tags the task as something that will be improved upon later during sleep.

Even more impressively, Stickgold's unpublished data indicate that the brain tags tasks in ‘chunks’, giving higher priority to those that were the most difficult to learn. When Stickgold asked volunteers to type different number sequences on a keyboard, the improvement with sleep was greatest for the sequences that had been the most difficult before sleep. “The brain grabs weaker connections and strengthens those,” argues Stickgold.

Such tagging may also help in integrating tasks such as putting together the musical phrases in a piano sonata. “You learn things in pieces and then sleep smoothes them out,” Stickgold suggests.

But it seems that memories strengthened during sleep become labile for a period after waking: if volunteers learn to tap their fingers in one sequence, and are the next morning asked to learn a second sequence, they seem to lose the sleep-induced boost in performance on the first5.

Inspired in part by Tononi's work, Stickgold and his colleagues now want to investigate what happens in the brain as it strengthens memories and integrates them with one another. As a first step, Matthew Walker, one of Stickgold's Harvard colleagues, aims to vary the tasks that his volunteers perform before and after sleep, trying to determine which brain region is needed for each task using functional magnetic resonance imaging scans.

Different stages of slow-wave sleep might be more important for different types of learning, such as how to play a piano or drive a car.

Walker is also interested in working out whether different types of slow-wave sleep are involved in different types of learning. Slow-wave sleep is composed of four distinct stages, beginning with initial drowsiness and becoming progressively deeper. Walker suspects that different stages might be more important for different types of procedural learning, such as how to play a piano or how to drive a car. “These are subtly different types of learning,” he says.

Walker and Stickgold have already shown that the sleep-induced boost in performance on a learned typing task correlates with the amount of ‘stage 2’ slow-wave sleep, particularly towards the end of the night6.

The Harvard researchers are also interested in investigating the role of particular EEG traces, known as ‘spindles’, which occur near the onset of sleep. These have been suggested to facilitate long-term changes in networks of brain cells, and so could be involved in learning7. Each spindle lasts for 1–2 seconds, and consists of waxing and waning waves at a frequency of 7–14 hertz.

After this initial activity, a full night's sleep involves several bouts of REM, interspersed with periods of slow-wave sleep, with each cycle lasting about 90 minutes (see graphic, right). Intriguingly, it seems that — at least for learning to discriminate between different visual patterns shown on a computer screen — a nap of about 90 minutes containing all the phases of sleep produces about the same gain in performance as eight hours' sleep8.

Sleepy genes

As well as changes in electrical activity, memory consolidation during sleep may involve shifts in gene expression. Tononi's group has found that, in rats, expression levels of about 100 genes increase during sleep, independently of the time of day9. Some of these genes are already known to be involved in the release of neurotransmitter chemicals, through which brain cells communicate with one another.

These findings fit with the theory that sleep is involved in consolidating procedural memory. The levels of different neurotransmitters in the brain fluctuate during the different phases of sleep. And Stickgold has unpublished data which show that cocaine addicts and schizophrenics, who both have dysfunctional neurotransmitter systems, do not show normal sleep-dependent procedural learning.

Establishing the precise relationship between sleep-induced procedural learning and the underlying changes in gene expression, neurotransmitter mechanisms and patterns of electrical activity will require extensive work in both human volunteers and experimental animals. No one is expecting sleep to give up its remaining mysteries without a struggle. But if researchers in the field ever get stuck in interpreting the results of a particularly perplexing experiment, they at least know to follow Mendeleev's example: just sleep on it.