Extended Data Fig. 4: Time-fields represent an internally-generated signal, not linked to movement.

(a-d) Only 12.0% of the 133 time-cells [cells × positions] that were recorded together with an accelerometer signal (16/133 cells) showed significant correlation between the trial-to-trial variation in firing-rate and the trial-to-trial variation in the acceleration signal. (a-b) Six typical time-cells (columns) that showed no significant correlation between the trial-to-trial variation in firing-rate and the trial-to-trial variation in the acceleration signal. (a) Top: color-coded raster plot, aligned to the moment of landing (t = 0). Trials (y-axis) are sorted according to the trial duration. Plotted as in main Fig. 1f. Middle: temporal tuning-curves – the average firing-rate across all recorded trials, aligned to the moment of landing of the observer bat. Green shading represents statistically-significant time bins. Bottom: acceleration signal, averaged across trials (gray shading, mean ± SEM). The acceleration signal shown here included flight-data for t < 0, while for t > 0 we only included here data recorded when the bat was on the ball (before takeoff). Note the large acceleration signal prior to landing (prior to t = 0) in all cases, which is caused by the bat’s flight – but then during the significant time bins (green shading) there was basically no acceleration signal. In other words, the bats hardly moved during the firing of the time-cells. All three panels for each cell (top, middle, bottom) are aligned to the landing-moment (t = 0) and to each other. (b) Six example scatter plots (for the 6 cells in panel a), showing that there is no significant correlation between the trial-to-trial variation in peak firing-rate and the trial-to-trial variation in the peak acceleration signal (both the peak firing-rate and the peak acceleration signal were measured inside the green rectangles in panel a; we used here a one-sided test for the Pearson correlation, and not two-sided test, because we assumed that only positive correlations are physiologically meaningful). These six examples represent the typical majority of time-cells that we recorded in experiments with accelerometer signal – which showed no trial-to-trial correlation between firing-rate and acceleration. This indicates that time-cells represent an internally-generated signal, unrelated to movement. (c-d) Examples of two rare neurons (columns) which represent the small minority of time-cells that showed a significant correlation between the trial-to-trial variation in firing-rate and the trial-to-trial variation in the acceleration signal. Plotted as in panels a and b. (c) Color-coded rasters, temporal tuning-curves, and acceleration signals – plotted as in panel a. (d) Scatters, plotted as in b. The example cell on the right showed the highest correlation value among all our neurons (r = 0.67); we note, however, that when removing the outlier point, the correlation became non-significant (r = 0.23, P = 0.16). (e-f) The bats did not perform on the balls stereotypical movements that were similar across trials – suggesting that stereotypical movements could not explain the firing of time-cells. (e) Examples: Three acceleration traces recorded on three different trials on the same day, all from the same ball (a significant time-cell was recorded on that day on the same ball). Note that in these three example traces: (i) the acceleration values were extremely low (<0.1 g, where g is the Earth’s gravity), and (ii) the traces were not similar to each other – indicating that this bat did not exhibit stereotypical movements across trials. (f) Population: Distribution of Pearson correlations between the acceleration signals recorded on different trials of the same day, on the same ball (computed from 0.5-s until trial-end; n = 39,323 trial-pairs) – that is, correlations between acceleration-traces as plotted in panel e. The correlation values were pooled across landing balls A and B and across experimental days and bats – only for days and balls on which a significant time-cell was recorded. The correlation of the acceleration signal between the different trials was very low (mean < r > = 0.052) – indicating that there were no stereotypical movements across trials that could explain the firing of time-cells. (g-i) No relation between time of firing and time of reward. (g) A typical example neuron showing no significant correlation between the time of peak neuronal firing (x-axis) and the time of reward delivery after landing (y-axis; extracted from the raw videos), with dots showing individual trials (Pearson r = 0.23; two-sided t-test, P = 0.16; n = 38 trials). Note there was large variability in the time-of-reward (large spread along the y-axis), as compared to the small variability in the neuron’s time of firing across the trials (small spread along the x-axis). (h) Left panel, scatter plot, showing a similar plot as in panel g (with dots showing individual trials), pooled across all the example cells shown in main Fig. 1. Right panel, same scatter as on the left, but here the x-axis data and y-axis data for each neuron were normalized by the mean for that neuron, in order to expose possible correlations which may be masked due to the high variability of preferred-times across different neurons. Both scatters show a lack of significant trial-to-trial correlation between the time of firing for each time-cell and the time of reward on the same trial. In addition, the timing of reward-delivery was highly variable, arguing against a role for reward in the temporal tuning of time-cells. (i) Histogram showing the distribution of Pearson correlation coefficients between the time of peak firing and the time of reward delivery – like the correlation for the cell shown in panel g – plotted here for all the example cells shown in Fig. 1 (in panels h and i, shown are n = 9 cells for which we also recorded raw video movies in addition to the video-tracking; this raw-video footage was used to measure the time of reward). Almost all these cells (except one) showed non-significant correlation (P > 0.05). (j) Raster of the times of ear-movements (x-axis) that were measured across 10 randomly-chosen landing trials (y-axis); the measurements were performed manually from high-speed camera recordings at 100 frames/second. This raster shows that, first, ear movements are generally repetitive – and hence cannot explain the firing of time-cells, which always fire only once per trial, rather than repetitively; and second, ear movements do not show stereotypical structure across trials (note the lack of vertical bands in this raster) – and therefore ear movements cannot underlie the temporally-reproducible, distinct firing of time-cells.