Grid cells perform path integration in multiple reference frames during self-motion-based navigation

Abstract

AbstractWith their periodic firing pattern, grid cells are considered a fundamental unit of a neural network performing path integration. The periodic firing patterns of grid cells have been observed mainly during behaviors with little navigational demands, and the firing patterns of grid cells in animals navigating 2D environments using path integration are largely unknown. Here, we recorded the activity of grid cells in mice performing the AutoPI task, a task assessing homing based on path integration. Using artificial deep neural networks to decode the animal’s moment-to-moment movement vectors, we found that grid cells perform path integration over short trajectories and change their reference frames within single trials. More specifically, grid cell modules re-anchor to a task-relevant object via a translation of the grid pattern. The code for movement direction in grid modules drifts as the animal navigates using self-motion cues, and this drift predicts the homing direction of the mouse. These results reveal the computations in grid cell circuits during self-motion-based navigation.