Grid cells accurately track movement during path integration-based navigation despite switching reference frames

Grid cells, with their periodic firing fields, are fundamental units in
neural networks that perform path integration. It is widely assumed that
grid cells encode movement in a single, global reference frame. By
recording grid cell activity in mice performing a self-motion-based
navigation task, we discovered that grid cells did not have a stable grid
pattern during the task. Instead, grid cells track the animal movement in
multiple reference frames within single trials. Specifically, grid cells
re-anchor to a task-relevant object through a translation of the grid
pattern. Additionally, the internal representation of movement direction
in grid cells drifted during self-motion navigation, and this drift
predicted the mouse’s homing direction. Our findings reveal that grid
cells do not operate as a global positioning system but rather estimate
position within multiple local reference frames.