Mind In Motion

To Make Sense of How the Brain Controls Movement, Neuroscientist Mark Churchland Had to Toss Out Existing Theories and Listen to His Data
By Sharon Tregaskis | Illustrations by Keith Negley

What, exactly, happens in the milliseconds before you reach for that glass of water on a hot summer’s day? Firing in concert, the millions of neurons inside your motor cortex somehow mobilize the muscles of your shoulder, chest, arm, and fingers as you reach for the glass, grasp it, and lift it to your lips. But how, specifically, do those electrical impulses coordinate and control movement?

Most of us grab a drink without thinking, neither crushing the vessel nor dropping it, neither pouring its contents down our shirtfronts, nor inadvertently tossing the cup over our heads. And yet computer scientists are unable to program a robot to execute this task with the same level of smoothness and effortlessness. Nor are doctors or engineers yet able to offer prosthetics that come close to the flexibility and precision of a lost limb.

For more than a decade, P&S assistant professor of neuroscience Mark Churchland, PhD, has been recording neural and muscular activity in the laboratory, seeking to both explain and predict how the activity of neurons in the motor cortex relates to movement. This goal has long been central to the study of the brain. It is of increasing relevance to human health as the development of neural prosthetics for paralyzed patients has shifted from science fiction to reality over the past two decades. “Think of the motor cortex as a very fancy machine for generating patterns of activity,” says Dr. Churchland, who also co-directs Columbia’s Grossman Center for the Statistics of the Mind. “We’re trying to reverse engineer that machine.”

Dr. Churchland embarked on that reverse engineering project in 2004, as a postdoctoral fellow at Stanford. His earlier graduate work at UCSF focused on the primary visual cortex, where the activity of individual neurons corresponds directly and neatly to cues such as motion, color, and form. “Much of that field was created by men and women who were engineers,” says Dr. Churchland. “They had a very mechanistic perspective and they really did manage to explain things very concretely, and in very satisfying ways.”

Given the early successes of visual neuroscience, it was natural for the same ideas to be applied to the motor cortex. If a single neuron in the primary visual cortex encodes for red and another for green, scientists hypothesized, then perhaps a single neuron in the motor cortex might encode a rightwards reach, and another a leftwards reach. Yet as the accumulating experimental data on neural activity and physical movement became ever more nuanced and detailed, it became increasingly difficult to reconcile facts with theory. “When I moved to the study of the motor cortex,” recalls Dr. Churchland, “I definitely found a lot of the explanations—and even the styles of explanations—not very satisfying.”

Perhaps worse, it seemed like the data that Dr. Churchland was collecting in his own studies more closely approximated an experimental train wreck. The responses of the neurons seemed so complex that the experiments and analyses he had originally expected to perform were clearly inappropriate. “Recording neural activity is like a little ritual,” he says. “You glean a view of every nuance of a neuron’s activity. As I started looking at the motor cortex neurons, I actually thought maybe we weren’t recording in the right area because it didn’t look like it was supposed to look.”

Dr. Churchland was by no means the first to express dismay at the discrepancy between his motor cortex data and what prevailing theories might predict. Researchers started recording and describing the activity of individual neurons in the motor cortex in the late 60s—and they have been debating ever since how to interpret the reams of data that work has generated. “It has become increasingly clear that the accumulating experimental evidence undermines many of our simplistic notions about neural coding,” wrote University of Washington neuroscientist Eberhard Fetz, PhD, in a 1992 literature review for Behavioral and Brain Sciences. “Moreover, the search for neural correlates of motor parameters may actually distract us from recognizing the operation of radically different neural mechanisms of sensorimotor control.” In short, Dr. Fetz was arguing that it was, in fact, fine if no individual neuron was a “rightwards reach” neuron. Often the parts of a machine make little sense in isolation; why should the neural machine of the motor cortex be any different?

The more Dr. Churchland looked at the data, the more enamored he became with the view first proposed by Dr. Fetz, which had also been adopted by some of his other colleagues. Increasingly, he had a hunch that nothing short of a theoretical overhaul—Dr. Fetz’s “radically different” approach—would be required to make sense of it all. “Getting from dissatisfaction to satisfaction was a seven-year journey,” he says. “It was a long time where we had to be OK with the fact that the data were fascinating and interesting, but it wasn’t clear what they meant.”

Monkeys and Their
Billie Jean Task

On the video monitor, trees, brick walls, and other features of an animated landscape whiz by. A whimsical little dragon is barely visible on a hilltop in the distance. In the foreground of that virtual world, a square target appears. Standing in for a soundtrack, a white noise machine supplies the cacophony of the rainforest. Suddenly the target glows and on cue, Drake turns the hand crank to his right, progressing through the three-dimensional tableau on the monitor.

Like his companion, Cousteau, Drake works for Tang, dispensed from a contraption similar to a hamster’s hanging water bottle. Through a straw in easy reach, the players get one sip for reaching the first target, seven sips on reaching the second.

Drake and Cousteau are specially trained 7-year-old rhesus monkeys fitted with an implant smaller than a penny that records their neural activity as they alternately crank toward the glowing target or pause, awaiting their next cue. They have logged months of training and they take their work seriously. “It’s really very challenging,” says second-year PhD student Abigail Russo, who leads the team performing these experiments in the lab of neuroscientist Mark Churchland.

“It is a hard task if you haven’t practiced,” says Ms. Russo, who tried the task herself, with limited success, when she was training to work with the monkeys. On a typical day, Drake and Cousteau log hundreds of trials in the span of an hour or two. “They are experts: They only fail a trial—and miss out on a reward—if they get distracted or if they just aren’t in the mood to play the game.”

To avert that hazard, Ms. Russo pays close attention to the monkeys’ cues. “Sometimes they’re really in the mood for it and sometimes not,” she says. “It’s like having a coworker. They have ups and downs, so you have to work around their moods and inclinations.” To reward their participation or coax greater enthusiasm on a down day—monkeys and researchers alike have the lowest motivation on Monday mornings—she doles out snacks.

“It’s part of building the relationship with the monkey,” says Ms. Russo, who freely dispenses dried strawberries, bananas, and mangoes, among other favored treats. “You want it to be a very positive experience when they come to the lab.”

“The first thing most people ask me when the find out I work with monkeys is: Do they really like bananas?” says Dr. Churchland. “Indeed, they do, though they also like grapes, apples, figs, and pretty much anything fruit, nut, or vegetable that you or I would like.’

The game created by Dr. Churchland and colleagues is known around the lab as the “Billie Jean” task—a reference to Michael Jackson’s 1982 music video, with light-up sidewalk panels evocative of the targets that cue the monkeys’ virtual trek. The game provides the scientists opportunities to observe how neural activity varies as the monkey cranks forward (when the landscape features green grass) and backward (when it’s brown) and how rhythms of neural activity ebb and flow as the monkey awaits the glowing cue that it may commence turning the hand crank that propels it through the landscape. “We now know that movement generation depends upon the onset of dynamics that lawfully transform an initial ‘preparatory neural state’ into the pattern of activity that drives movement,” says Dr. Churchland. “Yet a fundamental question remains: What is the nature of the neural trigger signal that recruits those dynamics and in doing so causes movement?”

At the time, self-described “data geek” John Cunningham, PhD, was a PhD student in the same Stanford lab as Dr. Churchland. Both were captivated by the opportunity to make sense of that fascinating and interesting data. “The brain is the biggest mystery in our universe right now,” says Dr. Cunningham, now an assistant professor of statistics at Columbia and a collaborator with Dr. Churchland in the Grossman Center. “We know how to fix many bones and organs, know about deep space features we haven’t seen, but we haven’t yet figured out how your brain controls your arm to pick up a glass of water.”

Dr. Churchland and Dr. Cunningham have focused much of their investigation on the milliseconds of neural activity that precedes a voluntary movement. Imagine an orchestra, in the moment before the first note is struck: The conductor raises her baton and the musicians ready their instruments. This is the preparatory phase. When the baton drops, the symphony commences—action. In the motor cortex, other scientists had hypothesized that within each motor neuron, the milliseconds preceding action featured a crescendo of electrical activity. Once enough neurons reach their threshold, motion begins. Dr. Churchland, Dr. Cunningham, and their Stanford colleagues speculated that by monitoring the activity of a single neuron, scientists had focused too narrowly in their search for cause and effect—as though reading the score for the canon in Tchaikovsky’s “1812 Overture” could reveal the orchestral complexity of which it is a part. Rather, they hypothesized in a 2010 paper for the journal Neuron that the pattern of activity within a single neuron in the milliseconds before we move is merely the “first cog in a dynamical machine.” “Our results,” they wrote, “suggest that preparatory activity may not represent specific factors and may instead play a more mechanistic role.”

Dr. Churchland had been at P&S only a few months when, in June 2012, Nature published a follow-up work—again with Dr. Cunningham and their collaborators at Stanford—elucidating that dynamical machine. In “Neural Population Dynamics During Reaching,” Dr. Churchland and Dr. Cunningham focused on groups of neurons behaving something like two children on a seesaw at the playground. Differences in the phase and amplitude of oscillations among those pairs produce a rhythmic pattern that drives movement. “Think of the motor cortex as a machine for generating patterns of activity,” says Dr. Churchland. “If you want to make a movement, what your brain produces isn’t that movement, but a pattern of activity in the muscles. To get the right movement, the brain has to generate the right pattern of muscle activity.”

This mechanistic view implied, as Dr. Fetz and others had suggested, that it was perfectly acceptable if individual neurons made little sense on their own. “This was not a radical idea at a conceptual level,” says Dr. Churchland, “but in practice it changes almost everything about how one should analyze neural data. The fundamental unit of data in our field is the response of one neuron, which naturally creates the temptation to try and understand each neuron on its own.” By analogy, neuroscientists had attended closely to individual, easily quantified elements of the machine—the odometer, the steering wheel, the pistons—hoping that each would behave sensibly. “But if you saw a piston or a spark plug by itself, would you be able to explain the series of movements that a car makes?” asks Dr. Cunningham. “Motor cortex neurons are like that, too, understandable only in the context of the whole.”

In the wake of the Nature paper, Dr. Churchland received several honors, including an NIH Director’s New Innovator Award. More much-needed research funding arrived from key foundations, including Kavli, Simons, Sloan, Searle, McKnight, and Burroughs Wellcome. That October, he was named co-founding director of the Grossman Center, charged with furthering our understanding of the brain by applying quantitative methods to uncover deep and meaningful structure in large neural datasets. Dr. Cunningham also received awards and funding from the Simons and Sloan foundations.

Using their own data, as well as observations collected by other laboratories, Dr. Churchland and his team continue to explore the extent to which their dynamical model corresponds to actual measurements, fleshing out their theory as they go. “If you record the activity of a piston, it’s hard to relate that to the straight-line movement of a car,” says Dr. Churchland. “Any engine has moving parts that don’t directly relate to the vehicle’s final movement. For the neural engine, we’re asking, ‘What do the moving parts look like, and can we identify them?’” Instead of focusing on the final command—reach right, move fast—his team is interrogating the neural dynamics that generate those commands, teasing out the mathematical rules that govern those patterns. “The dynamical model might initially seem less intuitive,” says Dr. Churchland, “but it explains a number of features of the data that are otherwise troublesome or ‘messy.’”

As with their earlier work, their current projects depend on data generated by specially trained monkeys, fitted with the same kind of microchips used to detect epileptic seizures in humans. In Dr. Churchland’s lab, the monkeys participate in tasks designed to elicit specific movements—reaching for small targets or manipulating a handle forward or backward, for example—in response to cues that allow the scientists to monitor the animal’s neural state and its muscular activity. “When the body is in motion, the brain is constantly changing its activity as it creates movement,” says Dr. Churchland. “By measuring the movements of the animal, and the changes in its muscles and motor cortex, we can begin to track how neural activity is generated and how it produces muscle activity and, thence, movement.”

"When the body is in motion, the brain is constantly changing its activity as it creates movement."

Over the course of the past six years, Dr. Churchland and Dr. Cunningham have pursued their collaboration across institutions, time zones, and even continents. Reconvening at Columbia has been well worth the wait, says Dr. Cunningham, who accepted his Columbia faculty post in 2013, one year after Dr. Churchland arrived at P&S. “I’m a great believer in very close collaborations between computational and experimental researchers,” says Dr. Cunningham. “I want to sit with the experimentalist and have that person tell me what questions he or she is thinking about, what analyses should be done. And I want to say the same thing about the experiments. I don’t think that Mark’s and my papers would have been nearly as interesting without that close collaboration.”

Dr. Cunningham was the first faculty recruit of the newly established Grossman Center. “Once we had secured the funding and the space, there was no doubt regarding who we wanted to hire,” says Dr. Churchland.

That closed-loop approach between experiment design and data analysis is not universally celebrated, says Dr. Cunningham. “The academic community historically has been largely focused on silos, what one investigator can do,” he says. “At Columbia, I feel encouraged to be a collaborative statistician and a collaborative neuroscientist.” Through initiatives including the Mortimer B. Zuckerman Mind Brain Behavior Institute, the Kavli Institute for Brain Science, the Center for Theoretical Neuroscience, the Data Science Institute, and the interdisciplinary NeuroTechnology Center launched in October, institutional support for collaborations among neuroscientists with appointments across Columbia’s schools has never been stronger.

The building momentum has an array of benefits, from opportunities for postdoctoral fellows to split time with more than one principal investigator and serve as intellectual bridges between labs to large-scale grant proposals where investigators have a wide array of intellectual resources available right on campus, in the event that they hit a bump on the path to discovery. “Biology, psychology, even chemistry, and certainly engineering have always been routes into neuroscience,” says Dr. Churchland. “If we’re to understand how the brain remembers, feels, emotes, and perceives, we need every tool that we can get and that means the work is intrinsically, highly interdisciplinary.”