Thursday, April 01, 2004

The Mind at Night

By: Andrea Rock

The lab rat that runs the maze all day reruns it in its dreams. Zebra finches seem to rehearse their songs in their sleep. And people who always had a hunch that a good night’s sleep helped to firm up what they learned before they went to bed were on to something important.

Many lines of research on memory are converging on the conclusion that not only dreaming but every phase of sleep may contribute to consolidating the formation of long-term memories and be an essential step in some kinds of learning. In The Mind at Night, science writer Andrea Rock takes readers on a tour of the brain research laboratories that over the past few decades have explored the brain’s activity during sleep and dreaming, unearthing crucial connections to memory, learning, creativity, and emotional adjustment. In this excerpt from Chapter 5, Rock explains just how busy our brain really is during our supposed “hours of rest.”

Excerpted from The Mind at Night: The New Science of How and Why We Dream by Andrea Rock. © 2004 by Andrea Rock. Published by Basic Books. Reprinted with permission.

The dream is memory itself changing before your eyes.—Bert States 


Matthew Wilson spends his days investigating what rats dream about after a rough day in his lab at the Massachusetts Institute of Technology (MIT). “People ask why I’m interested in rats’ dreams, and I have to say I’m not, but I am interested in how memory is expressed during sleep and how that might relate to what we subjectively experience as dreaming,” says Wilson, who started out as an engineering student studying artificial intelligence. He switched to neuroscience when he realized it was impossible to build truly intelligent robots until we had a better understanding of how the brain itself functions. “We want to understand how what you do during the day gets into your sleeping state and if that has any impact other than giving you something to write about in a dream diary. We now believe it does—that the brain’s activity at night is a fundamental part of learning and long-term memory formation,” he says. 

Pointing to the stacks of paper that cover nearly every available surface in his office at MIT, Wilson wryly says, “The challenge of organizing my office is similar to the challenge the brain faces. The process of sorting through all of this information, selecting what I want to store, and organizing it so that it’s easily accessible when I need it is something I could squeeze in here and there during the day, but it would be much more effective to do that after hours when I’m otherwise undisturbed.” He contends that the hours when the mind slides into sleep and dreaming provide the perfect opportunity for the brain to filter daily experience, evaluate what’s relevant, and then integrate it with the vast repository of previous experience in long-term memory. It’s a time when we are unencumbered by the need to deal with the outside world. 

Wilson’s belief is founded in an experiment that provided one of those rare “Aha” moments that mark the high point of any scientist’s life. To address his questions about how memory works, Wilson decided that working with rats would be more productive than experimenting with humans because he would have greater control over the experiences the rats were exposed to during waking. He could also more precisely measure how they responded by implanting microelectrodes near individual brain cells, allowing him to eavesdrop on what was happening at that level, in both sleep and waking. 

Wilson and his team trained rats to travel a maze in search of chocolate-flavored food rewards. Through the sensors implanted in the rats’ brains, they continuously recorded the firing patterns of clusters of neurons responsible for orienting the animals in space. The neurons the researchers were monitoring are located in the hippocampus, a region of the brain initially involved in memory storage in both rats and humans. 

But they also recorded what happened in those brain cells as the rats slept—and what they discovered was a remarkable mental replay of experience. The same firing patterns of neurons that they’d seen when the rats were running the maze were reproduced in nearly half of the forty-five periods of REM sleep they recorded, when the rats were presumably dreaming—a vivid demonstration of the nightly survival skill rehearsal described by Jonathan Winson as the biological purpose of the dreaming phase of sleep. The replication was so precise that Wilson could pinpoint where in the maze the rat would be if it were awake and whether it was standing still or running. The amount of time it took for the experience to replay in sleep was the same as it had taken to transpire in the first place. 

“Seeing that these animals were literally running through this maze again mentally for over two minutes during sleep was easily the most astonishing thing I have ever experienced and probably ever will experience. What I was seeing was not a report of memory or my guess about memory; it was memory in action. The thrill of science is not confirming your hypotheses but rather finding things like this in your data that you had not anticipated,” says Wilson. 

The results of that study, published in 2001, are a key component of a growing body of scientific evidence indicating that the brain’s activity during dream-laden REM sleep is crucial to consolidating memory. Current research, however, indicates that it is not REM sleep alone that helps convert experience into memory but mental activity during other sleep stages as well. Sleep onset, slow-wave sleep, and REM all may play different roles in processing specific types of memories, or they may interact in an intricately choreographed sequence to encode information in a usable, lasting form. Sleeping hours appear to be an optimal time for integrating new memory, not just because the brain is freed from tasks such as ensuring that we’re not hit by a truck but also because shifts in brain chemical levels and other physiological changes create ideal conditions for the reorganization and strengthening of memory. 


To understand the growing body of evidence about how information processing that occurs in the brain during sleep influences waking behavior, it’s helpful to take a closer look at how memory really works. First, toss out any ideas about memories being literal mental videotapes of everything you’ve ever experienced stored away in some central filing system in the brain. When you experience anything—learning a new computer program, taking a hike in the Maine woods, or simply having a conversation with friends over lunch—the record of that experience is initially held by the hippocampus, a horseshoe-shaped structure in the center of the brain that curves outward and connects with the amygdala, which in turn is crucial both in generating our initial emotional responses and in determining the emotional coloring of memories that we store. The hippocampus takes in all available information about an experience from our senses, as well as from these emotional circuits, thus serving as a kind of super clearinghouse for information needed to construct a memory. 

But to become permanently incorporated as memory, information in the hippocampus must be replayed to the higher-level processing systems in the neocortex, where it can be compared with previously encoded experiences and evaluated. Part of this memory consolidation process also involves discarding what the brain judges to be nonessential. In fact, Nobel laureate Francis Crick and his colleague Graeme Mitchison came up with a theory that we actually “dream to forget.” In the aftermath of his acclaim as co-discoverer of the structure of DNA, Crick turned his attention to investigating the nature of consciousness. Examining the dreaming process as part of that effort, he proposed in 1983 that memories were indeed consolidated and reorganized during sleep. According to Crick and Mitchison’s theory of “reverse learning,” the random stimulation of the forebrain by the brainstem would set this memory-reorganizing process in motion. Superfluous information and nonsensical mental associations that were being pruned from neural networks would make an appearance as dream material on the way out, which would explain bizarre elements in dreams. “In order to optimize storage and recall of memories, the brain has to go through a process that in the world of computers is called garbage collection. Getting rid of inessential facts and invalid associations helps to consolidate the facts that are important for your future behavior. So this reverse learning theory is one variant of the idea that REM dreams are necessary for memory consolidation,” explains Crick’s current collaborator, Christof Koch. 

“In order to optimize storage and recall of memories, the brain has to go through a process that in the world of computers is called garbage collection. Getting rid of inessential facts and invalid associations helps to consolidate the facts that are important for your future behavior.”

Working memory consists of information that you are holding in your consciousness at this very moment—either knowledge you’ve just acquired or something you’ve temporarily brought to mind from long-term memory. Our ability to consciously hold information in this short-term buffer is surprisingly limited. If someone were to present you with a random series of numbers and immediately ask you to repeat them, you’d likely be unable to retain and parrot back more than seven digits at a time—the equivalent of a local telephone number. 

When we are in the act of holding information in memory, what happens physiologically is that groups of interconnected neurons fire together in a particular pattern that ties together all of the elements of that specific memory. When a memory is replayed, it reactivates the firing pattern of those same neurons and causes an anatomical change in which the connections between the neurons actually grow stronger the more they are replayed. As neuroscientists put it, cells that fire together wire together, and it is this “wiring together” that transforms short-term memories into long-term ones. Patients who’ve suffered memory loss due to brain injuries reveal that the most fragile memories are the most recent—facts learned or experiences that occurred in a period of days, weeks, or months preceding brain damage—while older memories are less susceptible to disruption because they have had more opportunity to become consolidated. The more often a memory circuit is reactivated, the more deeply it is ingrained. Over a period that can range from days to years, memories become encoded in the neocortex and no longer depend on the hippocampus to be set in motion. 


We form two basic types of memory. Procedural (also known as implicit) memories, typically involve knowing how to do something, such as riding a bike. This kind of memory can be formed and retrieved outside our awareness. For instance, we don’t have to stop to remember how to put one foot in front of the other in order to walk or consciously think about where to place our fingers on the keyboard for touch-typing once we’ve consolidated those skills in memory. And when we first learn to speak, we learn the rules of language without consciously setting out to do so. 

Most psychologists also would argue that early childhood memories can be activated as procedural memories by some event in the present and can influence our behavior even though the memory activation has occurred outside conscious awareness. For instance, say that a toddler’s parents left him overnight with Aunt Agatha while they were out of town attending a wedding. A transportation glitch delayed their return by a couple of days. It was the little boy’s first separation experience, and his overwhelming emotions were unhappiness and anxiety. He has no conscious autobiographical memory of that weekend, but on the rare occasions when Aunt Agatha comes to visit later in his life, his inexplicable knee-jerk emotional reaction is an overwhelming desire to slam the door, based on procedural memories associated with Aunt Agatha. 

The brain systems that carry out procedural learning in humans have existed throughout the evolutionary history of mammals and work unconsciously not because of some grand design to conceal aspects of our mental life from us, as Freud might have it, but simply because their operation is not directly accessible by the conscious brain, according to Joseph LeDoux, a neuroscientist at New York University. Well known for his research on the biological basis of emotion and memory, LeDoux points out that procedural learning shapes our most basic characteristic traits: our style of walking and talking, what we notice or ignore, and how we react emotionally when things don’t go our way. “Memory does indeed make us who we are,” says LeDoux in his book Synaptic Self. “Keep in mind, though, that the memories involved are distributed across many brain systems and are not always or even mostly available to you consciously.” 

The second broad category of memory that most people think of when the word comes to mind is available consciously and falls into a category called declarative— knowing that rather than knowing how. Declarative memories in turn come in two varieties. Factual (semantic) memory is general knowledge about the world, such as the fact that John Kennedy was shot on November 22, 1963, or that a Volkswagen is an automobile of a certain shape and size. Then there is autobiographical (episodic) memory, which is the record of what’s happened to you personally, such as what you were doing on that particular November day in 1963 or the details of the road trip you took too many summers ago in a battered red Volkswagen with your best friend from college. Declarative memories usually are recalled explicitly—we know that the information is there and we deliberately bring it to conscious awareness, even though we sometimes go through that frustrating experience of having a name or a song title on the tip of the tongue but still inaccessible. Damage to the hippocampus results in amnesia. While amnesiacs have access to procedural memory and to some factual memory—they generally remember how to speak and what to do with a cup or a door or a car—they typically lose autobiographical memory. 

Human autobiographical memory is a more sophisticated version of the memory system that Matthew Wilson’s rats relied upon to replay their travels through the maze. Rats have cells in the hippocampus known as “place cells” that fire when they are in a specific location in space and then fire again whenever they are placed in that same location—or as in Wilson’s study, when they mentally replay the experience of being in that location during sleep. Humans also link memory to location in this way. A brain-imaging study of London taxi drivers revealed that merely showing them maps of routes they frequently traveled activated the same areas of their brain that were engaged when they were actually traveling those routes. But as the human brain evolved in complexity, the hippocampus expanded its role to become a key element in the system for keeping track of emotionally tagged autobiographical memory. 

All of these various categories of memories are stored in neural networks in different regions scattered throughout the brain. As neurologist Antonio Damasio puts it in his book The Feeling of What Happens, “There is no single place in our brain where we will find an entry with the word hammer followed by a neat dictionary definition of what a hammer is.” Instead, there are a number of records in our brain that correspond to our past interaction with hammers: their shape, the hand motion required to manipulate the hammer, the result of the action, and the word that designates it in whatever languages we know. Yet when we bring the image of a hammer to mind, all of those components are seamlessly stitched together. 

Autobiographical memories of events in our lives are stored and recalled in similar fashion. Sounds, sights, and emotions associated with an experience all are encoded in different neural circuits. Calling up the memory of your wedding day or your tenth birthday, therefore, is not a matter of pulling out a single mental snapshot but more like assembling a mosaic with colorful pieces (the smell of the flowers and the sounds of the music in the church; the taste of the chocolate cake, the joy you felt when you saw the puppy with the birthday bow around its neck) pulled from many different storage bins and then instantly welded into a unified memory. 

An experience in the present that activates just one piece of that mosaic can set the entire circuit of interconnected brain cells firing to assemble the entire memory. In his literary masterpiece Remembrance of Things Past, Marcel Proust beautifully illustrated this process in a scene during which the narrator finds that dipping a petite madeleine into a cup of tea suddenly causes him to feel an overwhelming sense of joy. He then realizes that the mere taste of the tea-soaked pastry has triggered the feeling of intense happiness he felt as a child on Sunday mornings when visiting his beloved aunt, who would dip a madeleine in her tea and give it to him as a treat. He hadn’t eaten the pastry since those childhood days, and the taste alone was sufficient to automatically conjure up the emotionally laden memory of those Sunday mornings. “Foreshadowing scientific research by more than a half century, Proust achieved the penetrating insight that feelings of remembering result from a subtle interplay between past and present,” says Daniel Schacter, chair of the Psychology Department at Harvard University, in his book Searching for Memory

If our emotions are aroused during an experience, our memory of it will be strengthened by that emotional tagging. There’s one exception to that general rule, though. Extreme emotional arousal, particularly stress, increases the concentration of a hormone called cortisol, which actually disrupts the activity of the hippocampus and can weaken our ability to form autobiographical memories of the disturbing experience, even though procedural memories may be recorded—a phenomenon common among people who suffer post-traumatic stress disorder (PTSD). Recall of memories linked to strong emotions also can be greatly influenced by our emotional state at the time we retrieve them. Researchers have found, for instance, that we are more likely to remember unhappy events when we are already down in the dumps. And each time we call up an emotional memory, it may be altered slightly by what we’re thinking and feeling at the time we’re bringing it to mind. According to Joseph LeDoux, memories are “constructions assembled at the time of retrieval,” and information stored at the time of the experience is just one of the building blocks used to build a memory. 

What we have seen and heard afterward can also shape what we recall, as commonly occurs when eyewitnesses at crime scenes give accounts that are inaccurate because they are influenced by what others report to have seen. A perfect case in point: when two snipers terrorized the Washington, DC, area in 2002 by randomly shooting victims in public spots such as parking lots and gas stations, an early eyewitness report of a white truck speeding away from the shooting scene led to a string of subsequent reports from witnesses saying they’d spotted the same vehicle at other locations where the snipers had struck. In fact, the snipers’ vehicle turned out to be a battered blue Chevrolet, but the power of that first suggestion was strong enough to result in the search being targeted on a mythical white vehicle.

In a very real physiological sense, what we’ve incorporated in memory from the past also significantly affects how we experience the present and form new memories. “Experiences are encoded by brain networks whose connections have already been shaped by previous encounters with the world,” says Daniel Schacter. “This preexisting knowledge powerfully influences how we encode and store new memories, thus contributing to the nature, texture, and quality of what we will recall of the moment.” We remember only what we have encoded and what the brain decides to encode depends on our past experiences, knowledge, and needs.


While we certainly consolidate memory and adjust our mental models during waking hours, many studies now indicate that a significant portion of this work takes place during our dreaming time and directly affects our waking behavior. “The brain is constantly evaluating new experience to figure out how it fits into the mental model built through previous memories, testing to see how well that model is working to predict novel occurrences and to guide decisions. Much of this revision appears to occur during sleep,” says MIT’s Matthew Wilson. 

The way in which dreams are woven from memory is reflected in two dream accounts offered by neurophilosopher Owen Flanagan in his book Dreaming Souls. The first is a dream he had when he was five years old and the second, a dream he recorded at age forty-eight: 

Dream in 1955: A pack of wolves was chasing me. I was terrified and couldn’t run away fast enough. I awoke breathless, trying unsuccessfully to scream.

Dream in 1997: I was involved in a military maneuver sponsored by the CIA. My unit was badly positioned relative to the enemy and we were pathetically armed. I was very frightened. I tried to explain to my comrades —between normal trips to get the clutch on my car fixed—that our nonautomatic rifles, a cross between a musket and an M-1 but with no magazines, were losers. Then I gave an antiwar speech insisting that we not obey government orders to do battle. I had some supporters and was subject to some ridicule. The chair of my department appeared in a feathered cap and wearing a tartan-plaid kilt, his weapon pointed as if he did not know what to do with it. He was clearly our leader. I was amused and scared. I picked up my car and was congratulated by the auto mechanics on our victory. 

Deconstructing the memory components of the dreams, Flanagan points out that the plot of the five-year-old’s dream is much simpler than the adult version, in part because it is drawing from a much more limited repertoire of experience. It is a typical pursuit dream, and since, at the time, he’d recently learned about wolves from frequently hearing the tales of Three Little Pigs and Little Red Riding Hood, his brain supplied the wolves as the threat pursuing him. The dream he had when he was forty-eight drew upon his far greater memory stores, weaving together elements from several time periods. Coming of age in Vietnam, he had spent time in antiwar protests and then on active duty in the military. He’d had experience dealing with car repairs and at the time of the dream was a university professor, so those elements were also incorporated with earlier memories in constructing the dream. “In both dreams, my mind appears to have put the activated memories and experiences into a story, a narrative structure,” he says. “How exactly this is and why is one puzzle that needs attention.” Flanagan adds that the emotions that permeate the dreams, especially the fear, most likely were prompted by activation of the amygdala, the brain system that triggers the fight-or-flight response. 

Our off-line processing of the day’s events involves incorporating autobiographical memories that have a profound influence on who we are. What we record as autobiographical memory and how we integrate it with past experience contributes to the development of what neurologist Antonio Damasio calls the autobiographical self. That sense of self is based on past experience, but it is also what allows us to imagine and plan for the future. “The autobiographical self hinges on the consistent reactivation and display of selected sets of autobiographical memories,” says Damasio. “The idea each of us constructs of our self, the image we gradually build of who we are physically and mentally, of where we fit socially, is based on autobiographical memory over years of experience and is constantly subject to remodeling. I believe that much of the building occurs non-consciously and that so does the remodeling.” 

An important component of that consistent reworking of autobiographical memory may indeed occur in dreams, usually outside our conscious awareness, though waking life greatly influences which sets of memories are selected for replay as dream material. “It now seems likely that as we sleep, our brains are working hard to save the experiences that we will carry around with us for much of our lives,” says Daniel Schacter. “The important events in our lives that we often review during waking may be frequently ‘replayed’ during sleep. Experiences that receive little attention during waking probably receive fewer nocturnal playbacks, paving the way for forgetting.”


If memory is indeed what dreams are made of, what are the brain’s rules for selecting which life events are processed in sleep, and how are those events then integrated with existing memory? A creative experiment to find out how and when daily experience turned up in dreams was devised by Howard Roffwarg and his associates at the Albert Einstein College of Medicine in 1978. Nine college students were fitted with goggles that filtered out blue and green wavelengths in light, so that everything they perceived appeared to be tinted red. The test subjects wore the goggles during waking hours for a period ranging from five to eight days straight, and they gradually became accustomed to this altered world, which they called goggle colored. 

The subjects spent each night in the sleep lab, where they were monitored by EEG. The researchers hoped that by tagging all incoming visual images in a distinct color,  they could track how the brain processed ongoing experience in dreams via the subjects’ reports about when and how the red coloration was incorporated in their dream images. When awakened during REM, the students reported dreaming in goggle color for about half of the scenes in the first dreams of the night, but not in succeeding ones. On the following nights, goggle color appeared in later periods of REM, as well, with nearly half of later dreams incorporating scenes of red coloration and more than 80 percent of the dreams in the first REM period of the night including such scenes. 

Researchers had assumed that whatever dream material was not colored in red would be derived from memories of pre-goggle wearing experience, but in some instances, events that occurred before the experiment also appeared in goggle color. There also were combinations in a single dream scene: a setting where the room was normal but the scene the dreamer saw when looking out the window was shaded red. When the goggles were removed and the subjects experienced just one day of normal vision, the red tint disappeared from their dreams. The researchers could conclude only that daily experience is quickly incorporated into dreams in a process that involves a complicated interaction between recent experience and memory. Exactly how that dance was choreographed remained a mystery.

Among those in the forefront of solving that mystery is Robert Stickgold, an assistant professor of psychiatry at Harvard. Stickgold came up with a novel way to try to coerce the brain into revealing its rules by examining a stage of sleep that previously had been overlooked by most researchers. As we’re drifting off to sleep, we typically experience what’s known as hypnagogic imagery—hallucinatory visual images and other sensations that usually aren’t threaded together in narrative as most dreams are. More than a decade ago, Stickgold became fascinated by this sleep onset phenomenon while on vacation in Vermont. “After a day of hiking and rock climbing,” Stickgold recalls, “as I drifted off to sleep, I immediately felt I was back on the mountain in one tricky passage where I had to cling to the rocks to pull myself up. I roused myself a couple of times, but each time I dozed off, the feeling of my hands on the rocks returned. Later in the night when I’d awake and try to get those same images back, I couldn’t, but when I was first falling asleep they were unavoidable.” He began taking note of other instances at sleep onset when he’d get similar strong spontaneous replays of the day’s events and found that they tended to occur when his days included out-of-the-ordinary experiences, such as days spent white-water rafting or sailing in rough water. 

Stickgold’s curiosity was both personal and academic. Though he’d started out as a biochemist, Stickgold became interested in neurophysiology while doing postdoctoral work at Harvard. His career shifted entirely when he took a course on the dreaming brain taught by Allan Hobson and, shortly thereafter, in 1990, joined Hobson’s lab. “I wanted to bring the scientific rigors of biochemistry to the study of dreaming, which I view as a route to understanding the waking mind,” says Stickgold.

In an effort to learn more about how the brain selects which memories will be activated and when, Stickgold decided to focus on the sleep onset period to see if he could manipulate the content of imagery dreamers experienced as they were drifting off. Since asking subjects to go mountain climbing or white-water rafting as part of a study would have posed legal liability nightmares, Stickgold opted to see if a somewhat tamer novel experience could also induce such images. Even he was startled by the results. 

In the first experiment, he recruited volunteers to play a computer game called Tetris, in which players assemble geometric puzzle pieces that are falling down the screen. Twenty-seven people played the game for seven hours over a period of three days. Ten of the players were experts because they’d played a Nintendo version of the game previously, while the rest were novices. Stickgold included among the novices five people who were amnesiacs, simply to see whether their dream imagery would include anything from the game—something he considered unlikely. 

When the volunteers were awakened in the first few minutes of sleep and asked to report what was going through their minds during the first two nights, more than 60 percent reported dreaming about Tetris at least once, and all reported the exact same images: falling Tetris pieces. The majority of dream reports occurred on the second rather than the first night of training. “It’s as if the brain needs more time or more play before it decides this is something that needs to be dealt with at sleep onset,” he says. 

To Stickgold’s surprise, the amnesiacs also reported seeing the same kind of Tetris images, even though in waking they had no memory of the game and had to be reintroduced to the researchers from one day to the next. “I was stunned, because we thought if there’s one stage of sleep that depends on episodic (autobiographical) memories, which amnesiacs lack, it would be sleep onset.” 

The fact that the amnesiacs got Tetris images at sleep onset indicates that autobiographical memories—details that link us to reality with specifics such as names, times, and places that we can consciously recall— are not the source of dream images at sleep onset. Instead, the dream imagery is being drawn from the kind of memory that amnesiacs do have—procedural and factual memories generated in the higher levels of the neocortex, where sensory information from experience is first received and associations with existing older autobiographical memories are formed. Scientists had long suspected that this was the source of imagery and memory for the more hallucinatory dreams we experience during REM and also in non-REM sleep periods later in the night. But since sleep onset seemed to incorporate more transparent replicas of real-life events from the day, Stickgold says his finding suggests that all dream imagery comes from the cortex as it associates fragments of recent experience with older memories. “Now we have experimental evidence about where dreams are coming from, and since the process works the same for normals and amnesiacs, it meets the kind of hard scientific standards I had as a biochemist,” says Stickgold. Indeed, when the Tetris study appeared in Science, it marked the first time in thirty years that the respected professional journal had published an article related to dream research. 

The amnesiacs’ results also suggest that these unconscious Tetris memories that showed up in their dreams affected their waking behavior. The amnesiacs had to be taught how to play the game all over again each day, but at the start of one session, a researcher noticed that one of the amnesiacs instinctively placed her fingers on the three keys that are used in playing Tetris: “She did not quite know what she was doing and yet she did,” says Stickgold. “Memories can be activated in our brain that are outside conscious awareness but that nonetheless guide our behavior.” 

The experiment also demonstrated how the brain edits out information it considers irrelevant: none of the sleep onset dreams included the dreamer or any details of the testing room—only the essential images of the task that had been learned were replayed. And the brain was also busy making connections: rather than seeing falling Tetris images in black and white as they appeared on the screen used in the experiment, one expert player dreamed of the pieces in color and accompanied by music, as she’d experienced them in the Nintendo versions on which she’d first learned to play years earlier. The substitution of those older images for the new reveals that the brain wasn’t simply replaying memories of the day’s events but transforming them through association....


A subset of memory consolidation is learning, whether it’s taking piano lessons for the first time or memorizing dates for a history exam. Dream researchers are rapidly building evidence that the dreaming stage of sleep, in a complicated dance with mental activity in other sleep stages, plays a significant role in acquiring new information and skills. “A lot of scientists I know are also musicians, and they frequently have the experience of practicing a difficult new musical piece and not getting it down, but when they come back to it after a couple of nights’ sleep, they suddenly have it, even without practicing in the meantime,” says Dan Margoliash, a professor of biology at the University of Chicago. “What does that mean? We’re obligated to ask those questions now and examine them as rigorously as we do other aspects of behavior.” 

Like Matthew Wilson, Margoliash has looked for answers via animal studies, and he has uncovered evidence that just as rats dreamt of rerunning mazes, birds replay and refine their species’ courtship songs as they sleep. Margoliash studies zebra finches, tiny birds who first learn their species’ song pattern by imitating what they hear from adult birds. “Not only does the bird need to hear itself practicing the song early in life while first learning to sing, but adult birds also have to hear themselves singing regularly to maintain the song correctly. Humans also need to hear their own voices regularly or their quality of speech deteriorates, as it does in people who suffer hearing loss as adults,” explains Margoliash. 

Scientists previously had assumed the auditory feedback a bird requires to keep its song in top form simply occurred while the bird was singing during waking hours, but when Margoliash recorded signals from the neurons that produce singing signals in the brains of both waking and sleeping birds, he discovered something unexpected—the same firing patterns that occurred when the birds were actually singing began to appear once the birds slipped into sleep. The researchers first discovered that the brain cells’ firing pattern was replicated when recordings of the birds’ own singing were played back to them during sleep, but they also found that even when the recordings weren’t playing, those neurons still fired spontaneously in patterns that suggested mental song rehearsal was occurring, primarily during slow-wave sleep. 

And while the auditory signals associated with song replay flowed freely in sleeping birds between different brain areas that control singing, that auditory feedback flow was blocked when the birds woke up, as if a barrier had slammed down. Based on this initial evidence, Margoliash hypothesized that rather than listening to its own song as it is singing and instantaneously doing any mental fine-tuning that might be needed, the finch instead stores the auditory signals from that singing episode in the bird-brain equivalent of the hippocampus to be replayed during sleep—assessing its own performance and adjusting its network of song-producing neurons off-line. In fact, he suggests, it may be difficult for either the human or animal nervous system to modify itself while it is actually in the process of singing or, in the human case, performing a new gymnastic move. 

Margoliash—whose aversion to taking himself too seriously is evident in his e-mail tag, “bigbird”—says that at first he was skeptical of his own hypothesis about bird-song being replayed and tuned up during sleep, because it seemed “a bit wild.” But now he believes rapidly accumulating evidence from his own lab and from experiments by other researchers indicates that both dream-rich REM sleep and slow-wave sleep play an integral role in learning.

The suggestion that a good night’s sleep improves human learning surfaced early on in a scientific report published in 1924, but the results of rounds of experiments conducted after the discovery of REM sleep in the 1950s tossed cold water on that idea. The experimenters required that subjects learn factual information, including memorizing lists of paired words that had no obvious connection, such as cow/stair. They then tested subjects to see whether depriving them of their dreaming time affected their performance. It didn’t, so researchers mistakenly assumed there was no connection between sleep and learning. 

What researchers have since discovered is that different sleep stages are tailored toward different types of learning, explains Carlyle Smith, who began studying the link between learning and sleep back in the early 1970s, when he joined the stream of American researchers who traveled to France to work in the lab of dream research pioneer Michel Jouvet. “We spent a month sawing little rods to make a maze for mice and then recorded their brain activity twenty-four hours a day for ten days straight. The mice who became smarter at running the maze showed big increases in REM sleep, and the others didn’t,” recalls Smith, now a psychology professor at Trent University in Peterborough, Ontario, whose studies have continued over more than three decades. “From that point on, I never doubted that sleep and learning were connected, and now there’s enough supporting evidence to get other researchers interested in it, too,” he says. 

A steady accumulation of research by Smith and others has helped explain how dreaming and cognitive processing in various stages of sleep affect learning. Shortly after falling asleep, we enter the light sleep known as stage II, and it is this phase that appears to be responsible for the improvement in performance that musicians, athletes, and dancers often experience a day or two after practicing a new skill. A 2002 study by Harvard researcher Matthew Walker found that a 20 percent improvement in motor skill tasks was largely dependent on test subjects getting stage II sleep in the final two hours of sleep prior to waking in the morning. “To get the maximum benefit from practice if you’re learning a new sport or a new piece of music, you need to get a full night’s sleep for at least the first night afterwards so that you don’t miss that final period of stage II sleep before waking,” says Smith. 

Following stage II comes slow-wave sleep, a deeper sleep that precedes REM. Slow-wave sleep is more prevalent in the first half of the night, when it occupies up to 80 percent of sleep time. During the second half of the night, the proportion of REM sleep jumps dramatically and alternates with stage II sleep. Slow-wave sleep is important for learning tasks involving factual memory—the kind of rote memorization you would need for a history test, for instance. Our dream-laden REM sleep, in contrast, is critical for procedural learning—the “how to” category that includes learning a new behavioral strategy. Studies have shown that not only does the amount of REM sleep that subjects experience increase immediately after they have been trained on such tasks, but their performance declines if they are deprived of REM after training, particularly on the first night.

In a well-known 1994 study, a team of Israeli scientists led by Avi Karni and Dov Sagi measured the time it took people to carry out a visual discrimination task— identifying the shape of a striped area flashing against the background of a test pattern on a computer screen. They found that speed in carrying out this procedural learning task improved not during the practice session but in the eight hours following it. If subjects were repeatedly awakened during REM, they failed to learn, but their performance didn’t suffer if they were awakened during the deepest stages of slow-wave sleep.

Since then, other researchers have conducted studies using the same learning task featured in the Israeli study, but their conclusions suggest that optimum learning may in fact depend on a combination of both types of sleep, not just REM. One of the studies indicated that improved performance hinged on getting sufficient slow-wave sleep in the first quarter of the night and REM in the last quarter of the night. Matthew Wilson has found the same sort of process occurs in rats, and he surmises that it is during slow-wave sleep that memory traces in the hippocampus are in essence stamped in for further processing later, during dreaming sleep, especially those REM periods in the last part of the night. During late stages of REM, it appears that the hippocampus and related limbic system structures such as the amygdala (which is important in processing emotions) exchange information with the higher-level processing centers in the neocortex in a way that strengthens memory and solidifies learning. 

In fact, molecular biological evidence also supports the idea that the brain is learning as well as spinning dream plot during REM. Every cell contains a collection of genes, each of which has a specific function in the body. When a gene is called into action to perform its DNA-ordained purpose, it becomes activated in a way that can now be measured. This measurable activity is called gene expression. A 2002 study revealed that a specific gene that is expressed during waking when a rat is in the process of learning is expressed again most strongly during late stages of REM, indicating that learning-associated changes on a molecular level occur in this phase of sleep. And when the hippocampus is put out of commission through anesthesia, the gene expression associated with learning simply doesn’t occur at all in the neocortex. 

 “It has been hypothesized for some time that memory traces must be passed from the hippocampus to the neocortex for long-term storage, and our study indicates that this may be what’s occurring during REM sleep. Especially during the later phases of REM, the hippocampus is talking to the neocortex,” says Constantine Pavlides, a neurophysiologist at Rockefeller University who is one of the authors of the study and a protégée of Jonathan Winson, whose theories from the 1970s on the biological function of REM sleep are bolstered by the new molecular biology research. 

The research suggesting that learning takes place after we drift off to dreamland obviously gives “Sleep on it” a whole new meaning.


About Cerebrum

Bill Glovin, editor
Carolyn Asbury, Ph.D., consultant

Scientific Advisory Board
Joseph T. Coyle, M.D., Harvard Medical School
Kay Redfield Jamison, Ph.D., The Johns Hopkins University School of Medicine
Pierre J. Magistretti, M.D., Ph.D., University of Lausanne Medical School and Hospital
Helen Mayberg, M.D., Icahn School of Medicine at Mount Sinai 
Bruce S. McEwen, Ph.D., The Rockefeller University
Donald Price, M.D., The Johns Hopkins University School of Medicine
Charles Zorumski, M.D., Washington University School of Medicine

Do you have a comment or question about something you've read in CerebrumContact Cerebrum Now.