Extended Data Fig. 10: Spatial and temporal variations in TE expression across learning.

a, Dynamic regression model estimates of trial-by-trial reward rates (black) and TE coefficients for two example fibers across all trials and sessions through learning (see Methods). Dashed lines indicate the trials in which the TE coefficients and reward rates reached significance (p < 0.01, shuffle test). Fiber locations indicated in the coronal atlas schematic. Black line, reward rate, averaged with a moving average window of 50 trials. b, Histogram of lags between the earliest trial where TE and reward rate reached statistical significance for each fiber in the three included mice. Colours indicate different mice. Negative values indicate that TE becomes significant prior to reward rate. Mouse 1:t₂₇ = −10.9, p = 7.0 × 10⁻¹¹, n = 28 sites; Mouse 2: t₂₇ = −76.2, p = 1.4 × 10⁻³², n = 28 sites, Mouse 3: t₂₉ = −5.4, p = 2.6 × 10⁻⁵, n = 30 sites, t-test on difference in trials between significant trajectory error encoding and significant performance, p-values Bonferroni corrected for 3 mice. c, Non-negative matrix factors (NNMF) captured from trial-by-trial TE trajectories (estimated with dynamic regression) across sites within each mouse representing early (blue) and late (red) TE increases across training for each of the three mice recorded across learning (see Methods for details). d, TE coefficients across trials averaged across recording sites with a larger weight for the early (blue) or late (red) component and normalized to the maximum TE coefficient, for each mouse. Grey line, reward rate averaged with a moving average window of 50 trials. e, Horizontal (left) and sagittal (right) plane maps of the differences in the weights for the late and early components of the NNMF for each recording site (bubble) across mice (n = 88 fibers, 3 mice). f, Spatial coefficients representing the magnitude and direction of the variation in the difference between early and late NNMF weights in e. Data points represent individual mice. ML: t₈₄ = 4.1, p = 3.11 × 10⁻⁴; AP: t₈₄ = 3.6, p = 0.0016; DV: t₈₄ = 1.4, p = 0.54, two-sided t-test on spatial coefficients, p-values bonferroni corrected for 3 dimensions. Error bar, 95% confidence interval around LME model coefficients. g, Vector showing the axis of maximal variation in weight difference, based on the coefficients in g. \(\underline{R}\), Rayleigh’s coefficient, S* = 9.92, p = 0.02. Thin lines represent individual mice. h, Horizontal plane maps of the differences in the weights for the late and early components of the NNMF for each mouse as in e. The bubble sizes indicate the DV position (increasing size indicates more dorsal positions). Shaded regions in all plots are 95% confidence intervals. Brain schematics were adapted from the Allen Mouse Brain Common Coordinate Framework (CCFv3) (https://atlas.brain-map.org/).