Carlos Brody

Princeton University
Time: March 4th 16:30-17:30 (IRST) / 8:00-9:00 (EST)
Carlos Brody


Individual variability and neural mechanisms of flexible decision-making


Born and raised in Mexico, Brody entered neuroscience through a Ph.D. at Caltech’s Computation and Neural Systems program, with John Hopfield as his advisor. From there he did a postdoc back in Mexico with Ranulfo Romo, a second short postdoc with Tony Movshon at NYU, and then moved to Cold Spring Harbor Laboratory in 2001 to start a computational group. Two of his neighbors there, Zach Mainen and Tony Zador, had started an effort to train rodents in monkey-like tasks and this tempted Brody to work with them and start his own experimental lab along these lines. He moved to Princeton in 2007, and in 2008 became an Investigator of the Howard Hughes Medical Institute. Since then, his group has been engaged in pushing the envelope on the sophistication of tasks that rodents are known to be trainable on. Using these behaviors, the Brody group uses a combination of high-throughput behavioral training, neural recording and perturbation methods, and computational approaches to elucidate neural mechanisms of cognition.


We are using rats to investigate the neural mechanisms underlying flexible, top-down selection of which features of a stimulus are used to drive decisions. For example, if standing at a busy street and you want to flag a taxi, the color of the cars might be the most important feature driving your actions; but if you intend to cross the road, the most important feature might be the motion of the cars. Two broad classes of models are prominent in the literature. In one, top-down signals selectively control, or “gate”, the feedforward pathway from posterior sensory regions to anterior decision-making regions. The gate lets through only information about the selected feature. In the other class of models, no feedforward gating occurs; feature selection is instead instantiated by selectively controlling the nature of recurrent dynamics within frontal regions. Seeking to distinguish between the two types of models, we trained rats to perform a flexible feature-selection task where information is presented in randomly-timed pulses. We developed analyses that use neural responses to such pulses to distinguish between the two main models. Instead of evidence clearly favoring one or the other model, we found that different individuals lay along a continuous spectrum between the two models. Behavioral analysis confirmed the neurally-suggested position on that spectrum for each individual. Our results underscore the importance of recognizing and identifying individual variability in neural mechanisms. The common practice of reporting only findings that are consistent across animals may discard genuine, informative biological variability. This problem may be particularly acute for cognitive processes, where widely differing internal algorithms may produce indistinguishable behavioral responses.