In 1857, just two years before Charles Darwin published his great work, The Origin of Species, the brilliant but notoriously cranky British anatomist Richard Owen made a claim that would inspire one of the most celebrated controversies in the history of science. Owen, seeking to shore up the tottering doctrine of special creation that said each species was separately and divinely crafted, asserted that human brains possess three structures not found in any other animal—a claim that the equally brilliant and combative Thomas Henry Huxley would handily demolish.
In retrospect, Owen’s position appears especially bizarre considering that one of the three structures was the hippocampus. This curlicue structure, thought to be the seat of learning and memory, has been poked and prodded in all manner of animals—from sea slugs to rats and monkeys—in the century and a half since the Owen–Huxley encounter. Most of these explorations have been conducted neuron-by-neuron, using single cell recordings. In the early 1990s, a team of MIT researchers devised a way to record from ensembles of neurons in the rat hippocampus and, in particular, from a group known as place cells. These store spatial memory, each cell firing only when the rat is in a particular location. The researchers found that they could predict where a rat was in a simple maze by looking at which place cells were firing. But their predictions were plagued by a high margin of error.
In the mid-1990s, Emery Brown had an insight that would dramatically improve their odds and, at the same time, open the door to a more accurate understanding of how neurons behave. Researchers knew that neurons, like most things in nature, are subject to a mixture of deterministic and random forces. “They respond in a predictable way that is probabilistic,” said Brown. Whether or not a hippocampal place cell fires depends on where an animal is in space, but it is also the result of variables—the buildup of neurotransmitters, the opening of ion channels, mechanical forces affecting neurons—each of which contains an element of randomness.
“Neural systems are dynamic. So right there it says that if you want to try analyzing them you should analyze them with methods that can capture these dynamics,” said Brown, the Massachusetts General Hospital professor of anesthesia at HMS and MGH.
But the methods used by most neuroscientists fell short of the task. Drawing on an approach developed decades ago in the physical sciences, Brown realized that a key to understanding how a rat hippocampal neuron was responding at a given moment, firing or not firing, required looking at the individual neuron’s past: how had it behaved before under similar circumstances? He formulated a set of equations, a mathematical filter, that essentially interpreted the observed behavior of individual neurons through this historical, or dynamical, lens. Sure enough, when they plugged their electrophysiological data into this mathematical filter, the MIT researchers were able to predict the rat’s position with much greater precision.
Over the past 10 years, Brown, who was recently elected to the Institute of Medicine, and his colleagues have been applying this approach to a variety of problems, from how rats use their hippocampus to navigate their environment to how monkeys learn simple associations. Working with Brown, Lakshminarayan Srinivasan and Uri Eden, graduate students in the Harvard–MIT Division of Health Sciences and Technology, along with colleagues have recently turned their attention to another region of the brain, the motor cortex, but with a twist. In a paper in the October Journal of Neurophysiology, they present a design for a mathematical filter that monitors how a person’s intention to move, say, an arm causes particular neurons in the motor cortex to fire. These intention-driven motor signals, as measured by EEGs and other methods, are then converted to inputs for prosthetic devices, such as robotic arms or computer screen cursors, to be used by people with spinal cord injuries or neurodegenerative diseases.
In September, Brown, who practices clinical anesthesiology once a week, received an NIH Director’s Pioneer Award to investigate how anesthesia works in the brain. “I think now one of the most interesting questions in neuroscience was sitting right in front of me and I never looked at it—that is, what happens in the brain under anesthesia,” he said.
The effort could help heal a rift between his clinical and research interests, a split that dates back to his student days. Raised in Florida, Brown entered the HMS MD–PhD program in 1979 as part of the newly launched diversity program initiated by HMS professors Leon Eisenberg, Edward Kravitz, David Potter, and Marshall Wolf, who all received the recent Harold Amos Diversity Award, as well as Jon Beckwith and Alvin Poussaint. Brown became intrigued by the clinical practice of anesthesiology. “It’s real-time physiology, real-time pharmacology. Rapid decision-making. Also, it requires you to have good people skills,” he said. Fascinated by probability theory, he embarked on a PhD that applied statistical methods to understanding circadian rhythms.
In the early 1990s, and looking for a change, Brown began seeking a research area that would most likely attract funding. The NIH had identified neuroscience as one such area. But, as in many careers, serendipity also played a role. Loren Frank, a graduate student working in the lab of MIT professor Matthew Wilson, happened to take Brown’s HMS statistics class. Wilson had developed a method for simultaneously monitoring the activity of groups of hippocampal neurons. “It was a tremendous tour de force,” said Brown. Yet Wilson and Frank were looking for help interpreting their data.
Like many neurophysiologists, they had treated the firing of neurons as discrete events. “Neural systems are giving information like this—bip, bip, bip,” said Brown. “But their properties are changing dynamically, and you want to be able to track that just from these little bips.” In the 1960s, a Hungarian-born mathematician, Rudolph Kalman, had worked out a formula for tracking physical systems as they change through time. According to the Kalman filter, estimating the location of an object such as a plane or satellite depends on two things—where it appears to be (observation) and some understanding of how the system evolves over time (state equation), which in turn is derived by looking at how the system has behaved in the past. But the variables determining a satellite’s position such as height, velocity, and wind are continuously changing.
What Brown recognized—his true insight—is that the firing of neurons is not discrete or continuous but, instead, a point process. “Something builds up, reaches a threshold, discharges, and the system resets somehow and the process goes on again,” said Brown. The challenge for him was to adapt the Kalman filter to the point process behavior of neurons.
In their original experiments, Wilson and colleagues, recording from approximately 30 neurons, were able to determine an animal’s position to within 30 centimeters. Using Brown’s filter they cut down the error to 8 centimeters, results they published in 1998. They would whittle that figure down to 5.5 centimeters over the next few years by more accurately characterizing the properties of the hippocampal neurons, and especially their receptive fields (see video: http://neurostat.mgh.harvard.edu/brown/emeryhomepage.htm). “Essentially the same experiments, but the information extraction increased,” Brown said.
Having seen how much extra information his mathematical filter has squeezed out of the hippocampal, and other data, he is eager to spread the word. “I think one of the big areas that I have to take on now is pedagogy. People still don’t understand this—some people think it’s esoteric,” Brown said. Begging to differ, he said, “It’s actually really simple.” And soon he will be able to test this proposition.
“I got what I think is probably one of my most prestigious invitations to date, and that is to talk about this stuff in front of my son’s sixth-grade class,” he said. “I want them to come away from this and say, ‘Gosh, we want to do that—we want to read a rat’s mind.’”
