Fig. 3: Temporal distance coding and signatures of attractor dynamics in the EC.
a, Schematic depiction of how single neurons might encode temporal distance at the joystick onset time (top) or joystick offset time (bottom) and track elapsed time by monotonic change of firing rates. b, Scatter plot showing cross-validated explained variance (R2) for a linear regression model at onset (abscissa) and offset (ordinate). Different colours represent R2 for subsets of neurons with positive temporal distance coding at onset only (blue), at offset only (green), at both onset and offset (turquoise), with no positive coding (light grey) and for a null distribution generated from randomly shuffled regressors (black). c, Schematic depiction of how the neural states across the population of neurons might encode temporal distance at onset (blue axis) and/or offset (green axis). d, Left, projection of held-out neural states onto the axes associated with onset (top, blue) and offset (bottom, green) coding estimated using targeted dimensionality reduction. Right, goodness-of-fit (R2) for onset and offset coding for the two monkeys (A and M) (two-sided Wilcoxon rank-sum test; n = 50 bootstraps, Z = −7.81, P << 0.0001 for monkey A and n = 50 bootstraps, Z = −4.17, P << 0.0001 for monkey M). a.u., arbitrary units. e, Cross-correlation structure of 20 simultaneously recorded periodic neurons with the highest periodicity metric (188 pairs) rank ordered on the basis of activity during the navigation epoch (top: MNAV) and plotted with the same order for the ITI (bottom) for monkey A (see Extended Data Fig. 7 for monkey M).
