Northwest Labs B103 and Zoom
Timing via mental navigation in the entorhinal cortex
Mehrdad Jazayeri
McGovern Institute for Brain Research, Department of Brain and Cognitive Sciences, MIT
Numerous experiments have characterized the neural mechanisms that enable the brain to track short time intervals as a single continuous scalar variable. However, in many routine timing tasks such as counting seconds or estimating the duration of a journey or a song, we track time by moving mentally over a succession of memorized temporal events. Despite the prevalence of this form of timing in cognitive functions, we know very little about its neural underpinnings. Here, we address this question using behavior, electrophysiology, and circuit-level modeling in a novel “mental navigation” task in monkeys. We exposed animals to a 1-dimensional path punctuated by six equidistant landmarks while they solved a simple match-to-sample task. After learning the match-to-sample task, animals were asked to produce the time needed to travel between random pairs of landmarks without seeing the intervening landmarks. Animals were able to learn the task and generalized readily to unseen pairs suggesting that they had memorized and relied on the temporal structure of the landmarks. Next, we recorded from the entorhinal cortex (EC) to characterize the underlying neural computations. We focused on EC because neurons in this area are modulated by landmarks, can represent relational structures, and are sensitive to temporal order and duration. We discovered a region of EC associated with attractor dynamics where neurons exhibited hallmarks of timing via mental navigation: ramping activity punctuated by local bumps. Strikingly, the time between bumps matched the interval between the landmarks and thus reflected the activation of an internal model of the underlying temporal structure. Moreover, the produced times were correlated with the time of the last activity bump in single neurons. These findings establish a computational link between EC dynamics and timing via mental navigation. Finally, we analyzed a simplified continuous attractor network model of EC augmented with internal landmarks to better understand the underlying computations and their implications for behavior. Simulation of the model revealed that landmark inputs transiently slow down the network dynamics and act as local resets, which in turn reduce behavioral variability. Remarkably, further analysis of behavior revealed that the animals’ timing variability was accurately captured by a dynamic process punctuated by landmark-induced reset events, as predicted by the model. Together, these findings provide a first glimpse at the computational principles and neural mechanisms that may underlie timing via mental navigation.