learn_in_sleep

Perchance to learn
25 Sep 99

                                 Not getting enough shut-eye? Then you're cheating your brain
                                 out of the eight hours it needs to learn properly. Helen Phillip
                                 eavesdrops on the night long seminar inside your head

                                 HAVE you ever gone to bed frustrated that you couldn't solve a
                                 problem, and then seen the answer straight away the next morning in
                                 an eerie moment of clarity? Perhaps it was the final clue of a
                                 crossword puzzle, or a face you couldn't put a name to during the
                                 day. You might notice something similar with a more physical
                                 challenge-a tough piece of music you were trying to learn, say-which
                                 magically seemed much easier after you'd slept on it. Did you assume
                                 you were just too tired to get things right the night before, or did you
                                 conclude that you'd worked things out in your dreams?

                                 Scientists have been suggesting links between dreams and memories
                                 for two centuries, and many are now convinced that memories from
                                 your day become fixed or consolidated as you dream. But by
                                 revisiting these memories while you slumber, can you actually work
                                 out a problem or carry on learning something? Some researchers think
                                 so-and not just during the two or so hours of the night you spend
                                 dreaming.

                                 Robert Stickgold, a cognitive neuroscientist in the department of
                                 psychiatry at Harvard Medical School, even suggests that the
                                 alternating periods of deep sleep and lighter, dreaming sleep we
                                 experience each night (see "Cycling in your sleep") are all vital for
                                 assimilating information, for spotting patterns in our memories, and for
                                 learning and honing skills. While we may think that only practice
                                 makes perfect, he believes sleep may also play a crucial role. What's
                                 more, learning and understanding what we've learnt takes all night. "I
                                 used to say that sleep was just a preferred time for learning," says
                                 Stickgold, "but now I'd say that certain parts of learning can't happen
                                 without sleep."

                                 The strongest evidence ties learning to rapid-eye-movement or REM
                                 sleep, a light sleep during which we dream most. Laboratory rats
                                 seem to have more REM sleep if they spend their waking hours
                                 learning or experiencing new things, and rats that are deprived of REM
                                 do much worse at learning the layout of mazes. In people, the link
                                 between dreaming and learning has been much more difficult to pin
                                 down, although signs that people might learn skills during REM sleep
                                 started to emerge in 1994.

                                 Avi Karni of the Weizmann Institute of Science in Rehovot, Israel,
                                 trained people to report whether they'd seen a short horizontal or
                                 vertical pattern appearing in a textured background on a computer
                                 screen while they focused on a letter in the centre. The task takes
                                 about 100 milliseconds for an untrained eye, and performance doesn't
                                 improve during the practice period. But if the trainees came back to
                                 the test after a night's sleep-with no intervening practice-they were
                                 about 15 milliseconds faster, an improvement that lasted for years
                                 afterwards. If Karni woke the trainees every time they drifted into
                                 REM sleep during the night after training, however, he blocked this
                                 improvement completely, which suggested that REM was crucial.

                                 But Stickgold wondered whether REM was the whole story. What if
                                 preventing REM blocked not the entire learning process, but just a
                                 vital step in a night-long sequence? To find out, he measured how
                                 much people improved over several days of practice at Karni's learning
                                 task. But instead of disturbing his volunteers' sleep, he simply
                                 measured the time they spent in each of the four sleep stages, or
                                 depths, and noted when each occurred. If learning happened during
                                 REM, he predicted, trainees who had more REM should learn more-just
                                 like the rats. In fact, he found that the best learners were not simply
                                 those with the most REM, but those with the most REM during the
                                 final two hours of the night and the most "slow-wave sleep"-the
                                 deepest kind of sleep-in the first two hours. This suggests that the
                                 learning process has two distinct phases, says Stickgold-one in
                                 slow-wave sleep and one in REM. The types of sleep in the middle of
                                 the night didn't seem to matter, so long as people didn't skimp on
                                 their hours. "There is zero improvement after six hours' sleep," he
                                 says.

                                 This delayed learning has a time limit, too. If Stickgold's trainees
                                 pulled an all-nighter right after the training period, they learnt
                                 nothing-even if they had the second and third nights to recover from
                                 their fatigue. "You have to sleep within 24 hours after the initial
                                 training or you won't show any improvement," says Stickgold.

                                 Sleep is probably not necessary for all forms of learning, but it is
                                 especially important for "procedural memory"-that is, learning "how"
                                 rather than "what". "Some tasks you learn instantaneously," says
                                 Stickgold. "If you don't remember a phone number 50 seconds later,
                                 sleep won't help. The procedural memories are more what sleep is
                                 about. If you're trying to learn a piano piece and you just cannot get
                                 it, you might find that you put it aside and come back the next
                                 morning-and you've got it."

                                 You can catch what may be the early stages of this process in your
                                 own mind by noticing the "dreamlets" that can happen as you drift
                                 off to sleep. Last April, at a meeting of the Cognitive Neuroscience
                                 Society in Washington DC, Stickgold and his colleagues described the
                                 dreamlets that people experienced while learning the computer game
                                 Tetris. The aim of Tetris is to rotate differently shaped blocks as they
                                 fall down the screen so that they drop into the spaces left by shapes
                                 stacking up below, leaving as few gaps as possible.

                                 Many of the trainees said that they saw images of the Tetris blocks
                                 as they fell asleep, most vividly on the second night. The images
                                 seemed to represent some salient feature of the game. For example,
                                 one trainee said she frequently saw the piece she had most trouble
                                 placing. Another said he saw the piece he needed most often to fill
                                 gaps. And the trainees who reported the most imagery were also the
                                 ones who were worst at the game when they started-the ones who
                                 seemed to have the most to learn. One woman who had played the
                                 game as a young child even noted that her dreamlets used the
                                 coloured pieces and music of her childhood version, rather than the
                                 black-and-white, silent version the researchers used-a sign that the
                                 brain was integrating old memories with the new learning experience,
                                 says Stickgold.

Warmup exercises

                                 People have reported similar sleep-onset imagery after days spent
                                 practising skills such as skiing, climbing, canoeing and bicycle riding.
                                 Whether the imagery continues through the night is unclear-REM
                                 dreams tend to be more bizarre, rather than reruns of our
                                 experiences-but Stickgold suggests that the dreamlets are a sign
                                 that the memory consolidation machinery is being cranked up,
                                 activating processes that are about to come online during sleep.

                                 But so far Stickgold has found only correlations, not clear
                                 cause-and-effect links, cautions Carlyle Smith from Trent University in
                                 Peterborough, Ontario. "We have to do the next step," he says-to
                                 interrupt slow-wave and REM sleep selectively and see what happens
                                 to learning.

                                 When he did this, Smith found that both REM and some parts of
                                 non-REM sleep are indeed important for skill learning. He looked at two
                                 related tasks-simply tracing between the double outlines of a shape,
                                 and tracing the figure while looking at it in a mirror. The mirror group
                                 needed REM sleep to learn the test well, and interruptions during
                                 other sleep stages had no effect. People learning the straightforward
                                 tracing task, on the other hand, did fine if REM was disturbed but very
                                 poorly when the lighter non-REM sleep known as stage II sleep was
                                 interrupted. Smith concluded that simple skills involving only small
                                 refinements of previously learnt skills require stage II sleep, while
                                 learning a completely new task calls for REM sleep.

                                 Which is the most important sleep period depends on what you're
                                 trying to learn, agree Werner Plihal and Jan Born from the University of
                                 Bamberg in Germany. Interrupting slow-wave sleep, they find, makes
                                 it harder for people to learn spatial tasks or associations between
                                 pairs of words-tasks for which the hippocampus, the part of our brain
                                 that records spatial memories, and all the events of the day, seems to
                                 be vital. REM, on the other hand, seems to be more important for
                                 learning procedures or skills-a process that need not involve the
                                 hippocampus.

                                 Sleep may be especially important for transferring the memories of our
                                 day from short-term storage in the hippocampus to a more permanent
                                 store in the cerebral cortex. "The hippocampus can't hold all the
                                 information," says Smith. "It must be sent to a much larger storage
                                 bin like the cortex." Even procedural memories, like how to ride a bike,
                                 that can form without the involvement of the hippocampus may
                                 benefit from this transfer, because our memory of when and how we
                                 learnt that task might help to supply context and thus a deeper
                                 understanding.

                                 This transfer seems to happen during slow-wave sleep, according to
                                 Gyorgy BuzsÁki from Rutgers University in Newark, New Jersey. In
                                 1989, he recorded electrical activity in rat brains and found that
                                 sensory signals flow from the cortex into the hippocampus while the
                                 rats were awake and exploring, and from the hippocampus back to the
                                 cortex during slow-wave sleep (see Diagram).

                                                  In 1994, Matthew Wilson and Bruce McNaughton,
                                                  then both at the University of Arizona in Tucson,
                                                  suggested what these flowing signals might be.
                                                  Neurons called place cells in a rat's hippocampus
                                                  represent every position in its environment. When
                                                  a rat runs around a maze, or around any open
                                                  space, its route is logged by a sequence of place
                                                  cells firing, in response to signals flowing in from
                                                  the sensory areas of the cortex. Wilson and
                                                  McNaughton found that short snippets of the
                                                  place cell sequences seem to replay during
                                                  slow-wave sleep. They concluded that this
                                 playback is part of the process of transferring information from the
                                 hippocampus to the cortex. This rehearsal might also be the reason
                                 that Plihal and Born find that performance on tasks that depend on
                                 the hippocampus improves after slow-wave sleep.

Pillow talk

                                 Soon afterwards, BuzsÁki spotted a flow of information in the opposite
                                 direction during REM sleep, and described a night's sleep as a period
                                 of dialogue between the hippocampus and cortex. Now Gina Poe, a
                                 neuroscientist from Washington State University in Pullman, has found
                                 what the cortex seems to be saying to the hippo-campus. Like Wilson
                                 and McNaughton, Poe recorded the firing patterns of place cells, but
                                 this time during REM sleep. During this sleep stage, the hippocampus
                                 keeps up a regular rise and fall in activity, between 4 and 10 times a
                                 second-the theta rhythm-just as it does when the rats are awake and
                                 exploring their world. Poe found that over the first couple of nights
                                 after the rats start learning a new maze, the cortex seems to drive
                                 the sequence of place-cell firing to coincide with the peaks of the
                                 theta rhythm. This synchrony, she thinks, helps to consolidate the
                                 memory by strengthening the synapses which neurons use to
                                 communicate. It happens during REM, she says, because REM sleep is
                                 the only time of night when the brain has high levels of acetylcholine,
                                 a chemical transmitter needed to strengthen synapses.

                                 But after a week, when the rats have got the hang of the maze, the
                                 place-cell firing begins to coincide with the troughs of the theta
                                 wave, which Poe says would weaken or even eliminate the synapses.
                                 By this time, the memory is well established in the cortex, and so can
                                 be erased from the hippocampus.

                                 Stickgold and Poe both suggest that something else might be
                                 happening during slow-wave sleep. While our waking hours and REM
                                 have the right biochemistry to establish synapses, and can strengthen
                                 them temporarily by pumping up the amount of acetylcholine they
                                 use, slow-wave sleep might start the process of strengthening them
                                 on a more permanent basis. For this to happen, you need to build
                                 bigger synapses-a job that calls for protein manufacture. "If you block
                                 protein formation or provide little energy to the brain after making
                                 new associations, you won't get the long-term aspects of learning,"
                                 Poe says. "The formation of new, stable synapses is an
                                 energy-intensive building process. Evidence is accumulating that
                                 slow-wave sleep is both a time for restoring energy levels and for
                                 building proteins."

                                 But our brains are doing more than just transferring memories during
                                 REM sleep. Stickgold and his colleagues think we're also busy exploring
                                 the links between old and new memories as we dream, which may
                                 help explain how we can sometimes solve problems in our sleep. One
                                 way to measure the strength of such associations is to use a "priming"
                                 test. If you were asked to judge whether the second of a pair of
                                 words you saw was in fact a real word, and you saw "hot" then "cold",
                                 you'd be very quick to make that judgment, because the well-learnt
                                 link between them primes you to conjure up the second word almost
                                 automatically as you're reading the first. If the words were less
                                 strongly associated, such as "thief" and "wrong", you'd be a bit
                                 slower. You'd be slower still if the words weren't linked at all-say,
                                 "car" and "apple". And if the second wasn't even a word-say, "car"
                                 and "blim"-you'd be slower yet. But this all seems to change as we
                                 sleep, the researchers found (Journal of Cognitive Neurosciences, vol
                                 11, p 182).

                                 You can't take the priming test while sleeping, of course, but
                                 Stickgold tried the next best thing. Most researchers agree that for a
                                 short time after waking, people still show most of the biochemical and
                                 sometimes even the brainwave characteristics of the sleep stage
                                 they were roused from. So Stickgold and his colleagues tested people
                                 as they were awakened from REM and stage II non-REM sleep.

                                 Immediately after non-REM sleep, people were a little slower to make
                                 the word-nonword judgment than when they were fully awake, but
                                 people were still quickest to answer when the two words were
                                 strongly associated. Unlike when they were fully awake, however,
                                 weaker associations were no help. But when people were woken from
                                 REM sleep, strong associations no longer speeded up the
                                 word-nonword judgment and weak associations became much more
                                 helpful-almost as effective at priming the answer as strong
                                 associations are when people were fully awake.

                                 Stickgold believes REM sleep may temporarily block the strong
                                 associations and allow the brain to strengthen the weaker links
                                 between less clearly associated memories. The high acetylcholine
                                 levels of REM sleep make each neuron's activity a little more sporadic
                                 or noisy, so your train of thought is more likely to bounce off its
                                 normal track of obvious connections and onto some different route.
                                 This could explain the bizarre dream narratives that we experience
                                 during REM sleep-it's not that we want to learn or rehearse the
                                 bizarre tales themselves, but that they might reflect the brain's
                                 exploring some of the tangential connections between our old and
                                 new memories.

                                 "There are different ways to know something, and we're just beginning
                                 to understand those different ways," says Stickgold. "It's the deeper,
                                 more profound understanding that needs sleep. When I think about
                                 how I would design a brain, I would want it to stop and look for
                                 distant associations. This is what most people would call creativity."

                                 Taken as a whole, he says, "a night's sleep is like five sessions with a
                                 therapist." The hippocampus is the patient, troubled with memories of
                                 the day, and the cortex is the therapist. "In the first part, during the
                                 slow-wave sleep, they talk about what happened, replaying the
                                 autobiographical memories from the hippocampus. During REM the
                                 cortex replies, and they look at how this information fits together.
                                 More slow-wave sleep, more explanation. The two memory systems
                                 talk back and forth trying to come to a consensus on what these
                                 memories mean."

                                 As with any form of therapy, dropping out before you've had the full
                                 course won't do you much good. Burning the candle at both ends
                                 cheats your brain out of its full night's conversation and makes the
                                 whole process futile. As Stickgold advises, "you should definitely get
                                 your eight hours."

Cycling in your sleep

                                  A night's sleep is made up of 90-minute cycles of
                                  rapid-eye-movement (REM) sleep and a series of non-REM sleep
                                  states. Sleep researchers can recognise these stages from
                                  differences in brainwaves, eye movements and muscle tone. Most
                                  dreams seem to happen during REM (stage I) sleep, quite a light
                                  sleep. Non-REM sleep, in turn, is made up of shallow stage II sleep
                                  and deeper slow-wave sleep (divided into stages III and IV). The
                                  first sleep cycle may contain only 2 or 3 minutes of REM sleep,
                                  preceded and followed by the deepest sleep of the night. Later in
                                  the night the REM periods are longer, lasting up to 45 minutes, and
                                  the non-REM periods are mostly stage II sleep.

                                  Robert Stickgold from Harvard Medical School believes that these
                                  alternating bouts of REM and non-REM sleep allow the hippocampus
                                  and cortex to hold a series of "conversations" during which they
                                  transfer the day's memories to long-term storage, explore their
                                  meaning and integrate them with earlier memories.