Extended Data Fig. 2: The duration of time-fields increased with the neuron’s preferred-time; analysis of non time-cells; and control for sharp-wave ripples.

(a-d) The temporal resolution of time-cells deteriorated with the passage of time. (a) Examples: Spike rasters (top), color-coded rasters (middle), and temporal tuning-curves (bottom) for a subset of the time-cells from Fig. 1f. Each column represents one time-cell. Top row – Spike rasters: x-axis, elapsed time from the moment the bat has landed (time 0); y-axis, repeated landings (trials). Each raster corresponds to a single location in the room (indicated above the raster), and each line in the raster shows the spiking activity in a single trial; each tick represents one spike. The trials in each raster were sorted according to trial-duration; the thin gray line denotes the trial-end (shown are only spikes contained within the trial). Middle row – Color-coded rasters: arranged as the spike-rasters above, but showing the instantaneous firing-rate instead of raw spikes (100 ms time-bins). Plotted as in Fig. 1e; color-scale ranges from zero (blue) to the maximal firing-rate in each panel (red; maximal rate indicated). Bottom row – Temporal tuning-curve for each cell (black trace), which is the averaged firing-rate of the neuron (average of the color-coded raster above). The preferred-time is indicated above the peak-firing of each cell (marked also by a vertical red line). Green shading represents statistically-significant time bins (Methods). Red curve, width-at-half-height of the time-field. Note how the width of the time-field (duration of the red curve) increases with a neuron’s preferred-time. (b-c) Plots showing that time-cells are aligned to landing, and not to takeoff. (b) Examples: color-coded rasters for the same cells as in panel a, aligned here to the bat’s takeoff. x-axis, elapsed time until the moment the bat took off (time 0); y-axis, repeated landings (trials). Each line in the raster represents the firing-rate for the cell in a single trial. The trials in each raster were sorted according to trial-duration (same sorting as in panel a). Color-scale ranges from zero (blue) to the maximal firing-rate in each panel (red; maximal rate indicated). Note that the peak firing across trials is diagonally tilted, and is aligned to landing and not to takeoff. (c) Distributions of Spearman correlations between the time of peak-firing in each trial and the trial-number (ordered by trial-duration). Cells whose firing is truly aligned to landing are expected to show zero correlation when the rasters are aligned to landing (as seen in the example rasters in panel a) and a negative correlation when the rasters are aligned to takeoff (as seen in the negative correlations in the examples in panel b). The distributions in the current panel were plotted for all the significant time-cell rasters (n = 274 cells × positions), separately when the rasters are aligned to landing (blue) or aligned to takeoff (pink). Note that, as expected, the distribution for rasters that we aligned to takeoff was significantly shifted towards –1, as compared to the distribution for rasters aligned to landing (two-sided t-test: P = 7.7 × 10^–170) – indicating that time-cell rasters show vertical bands when aligned to landing (as in panel a), and are tilted when aligned to takeoff (as in panel b); this means that the time-cells are tuned to the elapsed time from landing, rather than to time-until-takeoff (the small rightward shift in the blue histogram occurs because of late noisy firing in longer trials, as seen for example in panel a, fourth cell, which biases the correlations positively). Furthermore, since the time-cells in this analysis were defined based on the alignment of their firing to landing, we performed an additional analysis without such definition – to test whether takeoff (departure) can also trigger time-sequences, perhaps in a different set of neurons. To this end, we aligned the activity of all the neurons to the takeoff instead of landing, and sought to identify significant responses with this new alignment. We used in this analysis the exact same time-binning and same criteria to detect pure time-cells, contextual time-cells, and social time-cells, as we used for ‘landing-triggered’ time-cells throughout the paper – but now aligned on takeoff. This analysis yielded a substantially lower number of significant time-cells from each class: we found only 13 significant pure time-cells when aligned to takeoff versus 44 pure time-cells when aligned to landing; only 65 contextual time-cells when aligned to takeoff versus 125 contextual time-cells when aligned to landing; and only 28 social time-cells when aligned to takeoff versus 56 social time-cells when aligned to landing (all numbers are cells, not cells × positions). This much-lower percentage of significant cells when aligning to takeoff versus landing, strongly suggests that the relevant trigger for time-cells is landing and not takeoff. (d) Scatter plots of the time-field duration (field width at half-height) versus the preferred time, for all the significant time-fields (dots), in each of the three locations in the room: ball A (left; n = 116 significant time-fields), ball B (middle; n = 98), and Start ball (right; n = 61). All three scatter-plots showed significant positive correlations: ball A: Spearman ρ = 0.41, P = 4.6 × 10^–6; ball B: ρ = 0.57, P = 1.3 × 10^–9; Start ball: ρ = 0.82, P = 1.1 × 10^–15 (two-sided tests) (the significant positive correlations persisted also after eliminating from the correlations those time-cells with preferred time < 0.5-s: ball A: ρ = 0.24, P = 0.01; ball B: ρ = 0.46, P = 2.4 × 10^–5; Start ball: ρ = 0.77, P = 4.6 × 10^–9). This demonstrates that in each of the 3 locations in the room (A, B, Start), the time-resolution of time-fields deteriorated with the passage of time – as reported also for time-cells in rats^7,8,10,17. (e) Distribution of the time differences ΔT between the estimated time of landing from the video data and the estimated time of landing from the accelerometer signal (mean and standard deviation of ΔT: µ = 78.4 ms; σ = 90.7 ms; n = 5695 trials; the video-based landing time [our main estimate of landing-time in this study] was explained in the Methods – and the accelerometer-based landing time was estimated as the peak in the accelerometer signal, which exceeded 1.5 × g (1.5 times the Earth’s gravitational acceleration), and occurred within a time window of ± 300 ms around the video-based landing-time). Note that the standard deviation of this distribution was less than the time-bin resolution (100-ms bins) that we used for computing the temporal tuning-curves of the time-cells – indicating a very precise estimation of the landing-time. (f) Non time-cells. Top row: Temporal firing pattern of all the non-time-cells, plotted as in Fig. 1g: the cells are plotted separately for each of the landing-balls, and are ordered by the time of their peak firing-rate. Bottom row: the distributions of peak z-scores for time-cells (blue curves) and non time-cells (red curves). The firing sequences of non-time cells were clearly very different from the firing sequences of the significant time-cells shown in Fig. 1g: The z-scores were dramatically lower for non time-cells as compared to time-cells. In addition, the sequences of non-time cell tended to fall close to the diagonal in the top row. Both of these differences indicate that non time-cells do not exhibit true temporal tuning. (g-h) Sharp-wave ripples (SWRs) do not generate the temporal responses of time-cells. (g) Examples of two time-cells (rows), showing high similarity when plotted with versus without trials that included SWRs (columns; compare left versus right; example cells are from bat 1 [top row] and bat 2 [bottom row]). (h) Distribution of Pearson correlation coefficients between the temporal tuning-curves of time-cells when computed using all trials versus when computed after removal of trials with SWRs. Blue histogram, correlations for the data for all time cells (n = 274 cells × positions; note that the rate of SWRs was very low and they occurred only on a small subset of the trials: on average 0.97% of the trials). Black line, distribution of correlations for cell-shuffling (correlation between the temporal tuning-curve computed over all trials for cell i and the temporal tuning-curve computed over trials without SWRs for cell j, for i ≠ j). The real data correlations were significantly higher than the shuffles (two-sided t-test with unequal variances: P < 10^–300; t = 485.2; df = 7.4 × 10⁴). Inset: enlarged view of the blue histogram (zoom-in on the x-axis between 0.96 – 1). These high correlations indicate that the temporal tuning of time cells could not be explained by the occurrence of sharp-wave ripples.