It was just another day in the lab in 1991 when Matt Wilson first heard something that no one had ever heard before: brain waves from a dreaming rat.
Wilson, now a professor at MIT and a researcher at MIT's Picower Institute for Learning and Memory, had set up an experiment where he recorded neural signals from rats' brains as they ran a maze in the lab. One day, he left the rats hooked up to the recording equipment after they finished running the maze, while he sat at his bench working on some data analysis.
Soon enough, he started to recognize some of the patterns he was hearing from the resting rats' brains. "Suddenly I realized I could hear brain activity that sounded like the animal was running through the maze, but the animal was asleep," he recalls.
Wilson's groundbreaking discovery that the sleeping rat brain replays the rat's recent activities has added to a growing body of evidence that during sleep, the brain reinforces skills learned during the day by repeating and consolidating memories. While that theory is now generally accepted, it has taken several decades for scientists to start figuring out exactly what's happening in the mysterious sleeping brain.
'Initial excitement'
Scientists have been poking around the sleeping brain since the 1950s, when Eugene Aserinsky and Nathaniel Kleitman discovered REM sleep by observing the eyelids of sleeping children. Soon thereafter, researchers identified other stages of sleep, ranging from the light sleep of Stage 1 to the deep slumber of Stage 4. Most dreaming takes place during REM sleep, which occurs in short bursts throughout the night, alternating with the dreamless sleep of stages 1 through 4.
After those initial findings, sleep research produced few significant results as researchers struggled to make sense of the vivid, nonsensical sequences most dreams consist of. "In the 1960s, there was a lot of initial excitement about the possibilities, but it burnt out. Everyone wandered away from it," says Robert Stickgold, a professor of psychiatry at Harvard Medical School.
The tide turned in the 1990s, when sleep researchers including Stickgold made a flurry of discoveries showing that the sleeping brain appears to sift through information absorbed during waking hours, keeping the most important things and casting aside the irrelevant.
Furthermore, different types of sleep appear to be specialized for specific tasks. For example, useless memories, up to 95 percent of the day's experiences, are thrown out during the slow wave sleep of stages 1 and 2. During REM sleep, the important memories are transferred to long-term memory. That's also when most dreaming occurs, as the brain sorts through new information and connects it to older scraps of memory.
"The brain is trying to make sense of all this nonsense. It pulls information together and makes a story of it," says Subimal Datta, a sleep researcher and professor of psychiatry at Boston University Medical School.
Wilson's work in rats has been critical to these discoveries on the role of sleep in learning and memory, according to sleep researchers. "He provided the physiological evidence that neural reactivation is occurring during slow wave sleep and REM sleep," says Datta.
That reactivation occurs primarily in the hippocampus, a brain region involved in converting short-term memories to long-term storage. Wilson's early rat studies showed that the firing patterns seen in the sleeping and awake rats overlapped so closely that the research team could identify the sleeping rat's location in the maze.
While those maze-running dreams involve simple replay of daily events, Wilson now hopes to provoke more complicated dreams in rats to reveal how those dreams impact learning. To that end, he is now running new experiments in which rats are given richer experiences to draw from in their dreams: Instead of running one maze over and over, they run multiple mazes and perform other complex tasks.
Wilson is also collaborating with MIT Professor Susumu Tonegawa, also a member of the Picower Institute, in genetic studies to look for the molecular basis of memory formation during sleep.
'Many neurons at a time'
Key to Wilson's research is a recording device that can monitor brain activity in large collections of neurons. Wilson, who started his graduate studies as an electrical engineer, built the device along with fellow Caltech graduate student Upinder Bhalla in the late 1980s.
Electrodes that can record from individual neurons must be thinner than a human hair, which makes them difficult to maneuver inside a rat's skull. In Wilson and Bhalla's device, the electrodes are wrapped around each other to add strength, a structure inspired by the design of a multi-pipe oil-drilling rig that Wilson read about in graduate school.
Without that device, large-scale study of rat brain activity would not be possible, says James Bower, Wilson's PhD advisor at Caltech, now professor of neurocomputation at the University of Texas at San Antonio.
"The nervous system does not work one neuron at a time, it works many neurons at a time," says Bower. "If you try to understand a TV show by looking at a single pixel, you'll never understand anything, and that's almost certainly true for the nervous system as well."
Bower and Wilson first met at the University of Wisconsin, where Wilson earned his master's degree. Bower hired Wilson to write software to help model the olfactory system, and was impressed enough that when Bower took an appointment at Caltech, he encouraged Wilson to apply to a first-of-its-kind program there in computational neuroscience, which involves modeling the brain's processing functions.
After finishing his PhD at Caltech and postdoctoral fellowship at the University of Arizona, where he first heard the brainwaves of sleeping rats, Wilson came to MIT in 1994 as the first member of what was then called the Center for Learning and Memory.
Like most of his neuroscientist colleagues, Wilson wants to figure out how the brain does what it does. But true to his roots as an electrical engineer, he also wants to use his scientific discoveries to build artificial intelligence systems that learn the same way animals do.
"As engineers, we take principles and build things," he says. "The things we're doing are not just an effort to understand how the biology works. We also want to use our understanding of biological processes to develop devices capable of intelligent behavior."
