Figure 2.

Grid Cell Firing Patterns Transition from a Local to a Global Representation with Increasing Experience

(A and B) Fits of local and global models to grid cell firing patterns in the two compartments. The local model was an ideal grid constrained to replicate between the two compartments, whereas the global model was a single continuous grid spanning both compartments. Each row is one cell in one trial: the underlying rate maps in the left and right columns are the same. The white rings overlaid indicate the best fitting local and global models in the left and right columns, respectively. Fit values show the spatial correlations between the local or global models and the data, normalized by the independent model’s fit.

(A) Examples of grid cells recorded during early sessions, where the local model best fit the data.

(B) Example grids recorded during late sessions, where the global model best fit the data.

(C and D) The fit between grid cell firing patterns and ideal local and global grids, respectively, as a function of experience of the environment.

(E) The difference in the fit (global fit − local fit) between the global and local models across sessions.

In (C), (D), and (E), each data point represents the average fit for all cells with an independent fit >0.45, recorded from one animal in one session.

(F) The proportion of 1,000 ideal grids, with random phase offsets between compartments, with a better fit to the data than the local or global models. Values are mean + SEM across all cells with an independent fit >0.45 in the first or last five sessions. Wilcoxon signed-rank tests (WSRTs) compare observed values to an expected median of 0.5. ^∗∗p < 0.01; ^∗∗∗p < 0.001.