Dr. Shadlen is also a professor of neuroscience at Columbia University.
Neural Basis of Decision Making and Cognition
Michael Shadlen decided to become a physician after reading The Plague by Albert Camus, because he was inspired by the humanity of Dr. Rieux, one of the novel's main characters. "But I got carried away by neuroscience," he admits. He now sees patients for a month each year, and spends the rest of the time investigating how the brain decides on one course of action rather than another. This basic science could eventually help patients with Alzheimer's, autism, or other brain disorders, because it is essential to know how the brain is supposed to work before correcting any malfunctions. "Currently, if a patient comes in with signs of, say, confusion, we don't know what it is about the normal brain that lets us not be confused," Shadlen says.
As a neurology resident at Stanford, Shadlen often treated violent and confused young men in the emergency room. Even across the room, he could see these men bouncing their eyes between two positions. This bizarre movement indicated intoxication with angel dust (phencyclidine), the recreational drug of choice in the early 1990s. The inability to keep the eyes still indicated that the brain stem was having difficulty integrating eye speed signals to achieve a steady eye position. So Shadlen wondered if mental confusion occurred because some part of the cerebral cortex was having trouble integrating other types of signals.
While Shadlen was a postdoc in the lab of William Newsome (now an HHMI Investigator) at Stanford in the mid-1990s, he recorded the electrical activity of individual neurons while macaque monkeys sat in front of a computer and responded to images on the screen. Such activity enables neurons to talk to one another. "For the kinds of things we want to study, we need animals that are capable of pretty sophisticated tasks," Shadlen says. He stresses that the monkeys are perfectly comfortable performing such tasks while they have electrodes in the brain because the brain has no pain receptors.
These experiments provided the first evidence that a collection of neurons in the parietal cortex (toward the rear of the top of the brain) might integrate information to make decisions. They suggested that those neurons accept information from the part of the brain that processes visual information and then decide how other neurons should shift the eyes. "If this ultimately proves to be the case, several fascinating issues in cognitive neuroscience will be brought under rigorous physiological scrutiny," Shadlen wrote in a 1996 paper in Proceedings of the National Academy of Sciences.
After Shadlen established his own lab at the University of Washington in 1995, he continued to study the parietal cortex. This lobe is part of the association cortex, which encompasses the many areas between the regions of the brain that process sensory information or send commands to muscles. Neurons in the association cortex spring into action when sensory information needs to be interpreted, which is much of the time. People who have a tumor or stroke in the parietal cortex can process information from their surroundings, but have trouble using it in a rational way.
Shadlen and his students discovered that decision-making neurons in the parietal cortex can store and integrate information for at least a few seconds, unlike neurons that control muscles and give commands in split seconds. The group's later work revealed that neurons also obey a termination rule. "They figure out when they have enough information to make a decision, and then pull the trigger," Shadlen says, adding that the neurons reach this point regardless of whether the evidence is good or bad, the task is easy or hard, or the decision is right or wrong. He discovered that decision-making ends when the rate at which neurons fire electrical impulses reaches a critical level.
The next question was how neurons accumulate information. About 6 years ago, Shadlen read a book about the British cryptographers who cracked the Enigma code that Germany used to encrypt radio messages during World War II. This alphabetical code changed with every letter entered, and the key to deciphering it lay in the settings of rotors in the Enigma device, which were adjusted daily. To guess the settings, Alan Turing devised a formula to look for pairs of matching letters in two encoded messages intercepted on the same day. As he went along, he added up the logarithms of the probabilities (the likelihoods) that the Enigma machines that had encoded the messages shared identical rotor settings. After reading the book, Shadlen had a way-out idea: What if neurons take a similar approach when making decisions?
The experiments for testing this idea were based on the weather prediction test that is sometimes used to investigate human learning and memory. Subjects are shown some cards that say how likely it is to rain and others that say how likely it is to be sunny. After seeing a few cards, they are asked to predict sunshine or rain.
One of Shadlen's postdocs, Tianming Yang, was able to train two monkeys to perform a similar task. In one type of experiment, an animal sees a sequence of four shapes over 2 seconds. Each shape indicates a different probability that a reward will be given if the animal chooses either a red or green target. For example, one shape might indicate a 1 in 10 chance that the reward is associated with the red target (a red symbol on the computer screen), whereas another might indicate a 1 in 100 chance that the reward is associated with the green target. After viewing the four shapes, the monkey moves his eyes to one of the targets, and the probability that he will get the reward is governed by the four shapes. There is always a better choice, but as with the weather, the outcome is not guaranteed.
As reported in a recent Nature paper, the monkeys were able to accumulate and synthesize the probability information conveyed by the sequence of shapes. As well as taking information from one shape, they had to combine it with information from the other shapes, compare the various pieces of evidence, and decide which were the most convincing. "We found that monkeys do indeed learn probabilities and add them up just in the manner prescribed by Turing," Shadlen says.
Recordings from individual neurons in the parietal cortex showed the decision-making process in action. As each set of shapes came on the screen, the neurons changed the rate at which they fired. "To observe single neurons doing addition and subtraction is astounding," Shadlen says (see Probabilistic reasoning…).
The same parietal neurons register how much time has elapsed. "They anticipate the time that something, such as an eye movement, should happen," Shadlen says. He has evidence that neurons can measure time for as long as 2 seconds, but says there is reason to believe this form of timekeeping can last from tenths of seconds to 10 seconds. Over the past 7 years, neuronal timekeeping has become a second line of inquiry in his lab. "At the core of just about any higher cognitive function, there is the need to keep track of elapsed time," Shadlen says. "That is what allows us to infer cause and effect, and it gives the brain the freedom to process on a time frame that is not dictated by immediate changes in the sensory environment or the need to control body musculature."
In the long-term, Shadlen would like to discover how to preserve neurons' ability to integrate information and track time. "My hypothesis is that, in 10 years time, we will recognize that many disorders of higher brain function have a common failure mode," he says. "The ultimate payoff will come when we take these insights and combine them with molecular developments to devise therapies that preserve or replace key brain functions."