(Originally published on Scientific American.com)
“One hundred repetitions three nights a week for four years – sixty-two thousand four hundred repetitions make one truth.” These are the thoughts of Bernard Maxwell as he reflects on The World State’s sleep-teaching technique, hypnopaedia, in Aldous Huxley’s Brave New World, before concluding: “Idiots!”
Learn While You Sleep?
Huxley was using the idea to explore social conditioning and control in a dystopian future, rather than what we might call “useful” learning, but the promise of effortless learning while we sleep is an idea that refuses to go away, as evidenced by the continued existence of dubious sleep learning “courses”. The possibility was dismissed scientifically in the 1950s after an experiment showing that people who were played the answers to a list of questions while they slept could not recall any of them the next day, unless they had also shown electrical brain activity indicating they were waking up.
But evidence is now growing that the sleeping brain can, in fact, be taught in certain, limited ways. The most striking demonstration of this comes from a recent study published in Nature Neuroscience, in which people learned to associate sounds with smells while they were asleep. Pleasant and unpleasant odours were paired with different sounds played to sleeping participants and their “sniff responses” were measured. Pleasant smells provoked stronger sniffs and when the sounds paired with these smells were later played alone they still provoked stronger sniffs than those that had been paired with unpleasant odours. This was true both while the participants were still asleep and after they awoke and, unsurprisingly, they had no awareness of having learned anything. This is a limited form of learning known as conditioning, famous since Pavlov and his dog, and it can’t be used for learning anything as complex as, say, language vocab.
Learn Before You Sleep!
But it seems as though learning performed while we are awake can be boosted during sleep. There have been many studies showing that a period of sleep after learning improves memory compared to spending the time awake. Some have even shown that targeting memories during sleep, using a technique known as cueing, increases the beneficial effect on those specific memories. People trained to learn the positions of objects on a screen while sounds are paired with each object do better at recalling the positions of objects whose corresponding sounds are also played to them while they sleep between training and testing. Similarly, the smell of roses wafted under the noses of sleepers improves their ability to remember the positions of card-pairs, as in the well-known memory game, but only if they were smelling roses while learning the card positions. So sounds and smells linked to specific information during learning seem to be able to influence the sleeping brain. The associated memories seem to be “picked out” for greater consolidation during sleep.
The dominant theory to explain what is going on here involves the spontaneous reactivation of groups of neurons in the hippocampus, a part of the brain known to play a crucial role in the formation of new memories. According to this theory, while we’re awake, experiences are “encoded” simultaneously in both the hippocampus and the cortex (where most of our actual higher thinking happens), but the memory traces left in the cortex are fragile and easily disrupted, so when we sleep the two brain structures enter into a kind of dialogue that reorganises and transfers information between them. It is this communication between the two key memory structures of the brain that is believed to be behind the effects of sleep on certain kinds of memory.
During sleep the brain cycles through stages characterised by different biochemical states and electrical fluctuations, or “brain waves”. We first drop down through the light stages of non-REM sleep, (1 and 2), into the deeper stages (3 and 4) known as “slow wave sleep” (SWS), then drift back up through the stages into rapid eye movement (REM) sleep, before sinking back down again and repeating the cycle, at roughly 90 minute intervals. But we spend far more time in the SWS stages than the REM stage in the first half of a normal period of sleep, whereas the opposite is true in the latter half of the session. So early sleep contains a lot of SWS, and late sleep contains a lot of REM.
Slow Wave Sleep:
During SWS, three kinds of brain waves are seen. Slow oscillations from the cortex seem to synchronise the timing of both “spindles” and “sharp wave-ripples” so that they occur during the “up-states” of the slow oscillations, when neurons are firing at similar levels to when we’re awake. Sharp wave-ripples are spikes of electrical activity in the hippocampus, which seem to go hand-in-hand with the reactivation of groups of hippocampal neurons that were previously activated by a waking experience. These “hippocampal replay events” are thought to drive the reactivation of the corresponding groups of neurons in the cortex, which sounds a lot like memory traces being shunted from the hippocampus to the cortex. It’s even been shown that supressing hippocampal ripples with electrical pulses reduces memory consolidation.
Spindles are faster “bursts” of oscillation in the circuitry connecting the thalamus and cortex, which might activate biochemical processes that flag these cortical memory traces for later consolidation. And because both spindles and ripples occur during the slow oscillation up-states, all this happens while the cortex is receptive to the incoming information.
Meanwhile, all this activity is accompanied by changes in the levels of key brain chemicals, without which none of this could take place. During SWS levels of the inhibitory neurotransmitter, acetylcholine are very low, which is thought to enable replay events by reducing the inhibition of certain neurons in the hippocampus (inhibition is activity which reduces, rather than increases, the probability of a neuron firing). Levels of the hormone, cortisol, which would otherwise block the flow of information from the hippocampus to the cortex, are also lowered.
So what about REM sleep? Well, all of the above is what some neuroscientists refer to as “system consolidation”, because it involves the reorganisation of memory within the system, but this doesn’t tell us how those memories are then actually strengthened.
When a connection between two neurons is repeatedly activated, the strength of that connection gets a long-lasting boost. This is called “long-term potentiation” and neuroscientists believe it is fundamental to learning and memory. So it seems plausible that after groups of neurons in the cortex are reactivated by replay events and “tagged” by spindles during non-REM sleep, the relevant connections are then strengthened by potentiation during REM sleep.
This is still really just speculation, but support comes from studies of rats exposed to learning-rich environments before sleep. Increases in certain genes related to “synaptic plasticity” (which just means they increase the susceptibility of neuronal connections to changes) were seen in the cortex during REM sleep, but only in rats exposed to the enriched environment, suggesting REM sleep plays a role in strengthening the neural connections activated by learning. This is called “synaptic consolidation”.
The idea is also supported by the fact that during REM sleep acetylcholine is raised back to waking levels, as this promotes the activity of these plasticity-related genes. The different types of brain waves seen during REM sleep might also play a part, and it’s worth noting that the electrical activity in different brain regions lacks the synchronisation seen in SWS or while awake. This could be an indication that memory systems are disengaged during REM so that synaptic consolidation can happen, as this can take place locally rather than requiring communication between distant brain regions.
Sleep Before You Learn!
An implication of all this would seem to be that the process frees up the hippocampus during sleep so it can do the same job again the next day. But although there have been plenty of studies demonstrating the beneficial effects of sleeping after learning, little work had been carried out investigating the effects of sleep before learning, until a study last year did exactly that.
Neuroscientists at UC Berkeley tested volunteers on their memory of 100 names and faces. They then had one group sleep for 100 minutes while the other group went about their daily business. They later tested both groups on a new set of names and faces and the people who hadn’t slept between tests scored an average of 12% lower than they had previously, whereas the group who’d had a nap not only did better than this, they tended to do better than they had earlier. On average, they did 20% better than the people who had stayed awake.
This suggests that even a short nap refreshes our ability to learn, which is what you’d expect if sleep somehow refreshes the capacity of the hippocampus to form new memories. The team also monitored the brain waves of the sleeping volunteers and found that the more spindles their brains produced, particularly in the prefrontal cortex, the better they did on the later test.
The Big Picture:
So what we have is a picture of memory consolidation during sleep that involves two different stores and a two stage process. The hippocampus is the fast-learning, but low-capacity, short-term store, whereas the cortex is the slower-learning, but high-capacity, long-term store. Experiences are initially encoded in both, but not everything we experience warrants being stored long-term, so the memory traces in the cortex are initially fragile, until groups of neurons are reactivated by replay events in the hippocampus during SWS and then strengthened during REM sleep. This gradually shifts certain memories from short-term to long-term storage, presumably integrating them into existing networks of similar and related memories. How the brain decides which memories will be processed is anybody’s guess at this stage, but since the cortex seems to conduct the whole affair it seems a safe bet the answer lies there.
This is sometimes referred to as “active system consolidation”, and it’s a compelling theory that explains a lot of results and suggests that sleep plays a crucial role in the formation of long-term memories. Much of it however, is still just that – a theory. The other main contender is something called “synaptic homeostasis”. The idea here is that the strength of all connections between neurons is reduced during sleep – a process referred to as “synaptic downscaling” – so that weak connections are eliminated, which, relatively speaking, would enhance the significance of the remaining connections. But this implies that any memories that are only weakly encoded during waking would be deleted during any decent snooze, which isn’t really borne out by either experience or experiments.
The two theories are not necessarily mutually exclusive however. In fact, it’s quite likely that both processes occur, with synaptic downscaling ensuring that overall, the brain’s capacity to encode new information is refreshed, while system consolidation actively strengthens specific, presumably important, memories.
One aspect of the active system consolidation theory that has been mostly speculation up until now is the question of whether the spontaneous activity in the hippocampus actually causes memory consolidation. Although replay events have been seen in the hippocampus, solid evidence for their role in memory consolidation has been elusive. The smell of roses experiment mentioned earlier provided some support because it only worked when the smell was presented during SWS, not REM sleep (i.e. when replay events mostly happen). It also only worked for the card-pair memory task, not for another type of learning which doesn’t rely on the hippocampus.
Psychologists make a distinction between “declarative” and “procedural” memory. Declarative memory includes memories of events and facts (episodic and semantic memory), and seems to depend crucially on the hippocampus. Procedural memory is more about learning physical skills, such as riding a bike, and involves various areas including the motor cortex, striatum and cerebellum.
When the same experiment was run using a procedural memory task, in which people had to repeatedly tap a sequence of five keys as quickly and accurately as they could, a period of sleep after learning did increase people’s speed, but the smell of roses while they slept made no difference either way. So smell only enhanced the kind of memory that depends on the hippocampus. Both types of memory seem to be processed during sleep, but – unsurprisingly given that different brain structures are involved – the details differ.
The researchers also used functional magnetic resonance imaging (fMRI) to show that after the declarative memory task, the hippocampus was activated in response to the smell being presented during SWS. The trouble with this is that replay activity plays out in a specific sequence over time and although fMRI works well for pinning down the location of activity in the brain, it’s particularly bad at pinpointing exactly when it happened. So all the researchers could really say was something was going on in the hippocampus, without being able to definitely say it was replay events.
But another recent paper in Nature Neuroscience presents the most compelling evidence yet that this activity is part of the process of memory consolidation. The study showed that activity seen in a rat’s hippocampus during training could be reactivated during sleep by the sounds used in the training.
Rats were trained to run to the right side of a track in response to one sound, and to the left side in response to another sound. The researchers recorded the activity of neurons in the hippocampus during the task (rather than just using an imaging technique like fMRI) and saw different patterns of activity depending on which side of the track the rats were on. They then played the sounds to the rats during non-REM sleep and found that the neurons that had fired on the right side of the track while awake fired more in response to the right-hand training sound and neurons that had fired on the left side fired more in response to the left-hand training sound. This shows that cueing, which we know enhances memory during sleep, also provokes specific replay events, providing strong evidence for their role in memory consolidation.
The number of replay events wasn’t changed by cueing however – only the pattern of activity was affected, suggesting that cueing doesn’t boost the amount of memory consolidation, so much as just influence how the brain’s limited resources are allocated to determine which memories are consolidated.
This ties in nicely with the results of another recent study in which people who had learned to play two melodies were then played one of the melodies during the SWS phase of an afternoon nap. People performed better on both melodies after snoozing, but the improvement was significantly higher for the melody that had been played while they slept. The performance of a separate group who were not played anything while they slept fell somewhere between the scores for the “cued” and “uncued” melodies from the first group. This led the authors to suggest: “sleep may provide a finite capacity for memory processing such that cued reactivation tends to produce a bias rather than a pure gain”.
So in other words, it might not be possible to boost the amount of memory consolidated while we sleep, but we might be able to influence what is consolidated, although possibly only at the expense of other memories. If more research confirms this, it could still have applications in focussing memory processing during sleep on a desired result, or even indirectly supressing undesirable memories, such as in post-traumatic stress disorder patients.
Sleep On It!
I wasn’t happy with my original final paragraph, but was drawing a blank on how best to wrap-up the piece. So I woke up this morning intending to just spell-check and publish, but then it occurred to me that all this might be related to another line of research I read about recently. Most of us have had experiences like this – of being stumped on something we’re working on, only to wake the next morning with the solution fully-formed and nudging for our attention.
Why is this relevant? Well, if all this theory is correct, the end result of the processing that goes on while we slumber is that some of what we experience during the day is gradually shifted into long-term memory networks, where it is presumably put into context, sitting in the appropriate place connected to the web of similar or related memories that were already there.
That’s where some slightly different, but probably very relevant, research comes in. There is some evidence that sleep can lead to deeper insights into material we’ve just been exposed to. For instance, people taught simple one-to-one relationships between pairs of items, such as A > B and C > D, show no ability to work out the deeper structure that the relationships are drawn from (e.g. A > B > C > D etc.) when tested immediately after learning (e.g. is A > C?), but this increases over time and, if people sleep, they are significantly better at inferring the most distant relationships (e.g. A > D). This might also explain the results of a recent study showing that sleep increases our ability to solve problems we’d been unable to tackle previously – but only for difficult problems. This could be the result of integrating the new information into existing networks during sleep, so that relationships become more apparent and insight seems to have magically appeared upon waking.
It looks like there might be much more to the old adage “sleep on it” than we ever thought.