A new study by neuroscientists in The Picower Institute for Learning and Memory at MIT shows how the brain encodes information throughout the decision-making process to keep options clearly in mind and to ensure that chosen and unchosen options are remembered.

The key, the researchers show in the journal iScience, is that the brain convenes ensembles to produce coordinated patterns of electrical activity that distinctly represent and sort options, both during consideration and after choice. 

“It keeps different neural ensembles, different thoughts, distinct from one another, preventing interference between them,” says senior author Earl K. Miller, Picower Professor in MIT’s Department of Brain and Cognitive Sciences.

Lead author Huidi Li, a graduate student in Miller’s lab, says the study results show how the brain responds dynamically to meet the challenge of decision-making.

“The brain doesn’t just hold information statically,” Li says. “Throughout the decision process, the brain flexibly reorganizes information representation to meet the changing task demands.”

Decisions decoded

To conduct the study, Li, Miller, and their team trained two animals to play a game in which they had to look in the direction of one of two indicated targets on a screen, based on which target was assigned the higher reward value. Importantly, the two options were presented and their values were assigned in sequence — first one target, then its value, then the other target and then its value. That way, the brain had to juggle multiple representations for each target — for instance, the order of presentation before the decision, and then chosen-or-not after the decision. Meanwhile, each time the animals played the game, researchers measured the electrical activity of hundreds of neurons in the lateral prefrontal cortex, a surface brain region known for having a key role in linking options, values, and actions in decision-making.

Using “declassifier” algorithms to decode the electrical patterns, the researchers found that the neurons acted in functional ensembles whose collective activity clearly indicated decision-related information, including the distinct target directions and their assigned values. To interpret and compare each ensemble’s patterns, the researchers visualized these “subspaces” geometrically as planes on a 3D graph.

The researchers’ key finding was that before the values were assigned and a decision was made, the neural ensemble patterns consistently represented options based on their order of presentation. For example, on the graphs each “target 1” plane was nicely parallel with the others. Similarly, each “target 2” plane was parallel with its brethren, but the target 2s were more orthogonal, or more perpendicular, with the target 1s, showing that they were represented as entirely distinct from each other.

Then, after the decision, new ensembles provided new representations. Now the “chosen” options, whether they had been presented first or second, had parallel representations. The unchosen targets were also represented as parallel with each other, but as orthogonal from the chosen ones. 

Getting one’s neural ducks in a row

In other words, before the decision, the brain convened ensembles of neurons to distinguish targets by their presentation order, and then after the decisions, gathered ensembles to sort them by whether they were chosen or not. This consistent way of representing chosen options, Li and Miller wrote, could aid decision-making by essentially packaging it for downstream circuits responsible for converting the decision into action (in this case, directing the animal’s gaze in the chosen target direction).

“The observed alignment of chosen target representations could allow downstream areas to read out the location of the chosen target with a single decoder, regardless of its initial presentation order,” the authors wrote.

Notably, the researchers also found that round by round of the game, individual neurons could often be recruited in to different ensembles. A neuron that in one round seemed “selective” for option 2 could end up being selective for option 1 the next. The ensembles were therefore not permanent circuits of specialized neurons, but instead were assembled ad hoc among multifunctional neurons.

In other research, Miller has found that the brain uses brain waves to rapidly and flexibly accomplish this goal of ensemble recruitment.

Another clear implication of the data, Miller says, is that the brain maintained distinct memories of each option, whether it was chosen or not. This could be important for assigning credit down the line to facilitate learning. For instance, remembering that choosing target 2 in round 3 earned a reward.

“Our results illustrate the dynamic subspace reorganization supporting option maintenance and selection in economic decisions,” the authors wrote.

In addition to Li and Miller, the paper’s other authors are Nikolaos Chrysanthidis, Scott Brincat, and Jonas Rose.

The U.S. Office of Naval Research, the U.S. Army Research Office, the Freedom Together Foundation, and the National Institutes of Health provided support for the research.