Extended Data Fig. 6: Self time-cells: stability across sessions.

In session 1, the observer bat mimicked the flight-choices of the demonstrator bat; in session 2, the observer mimicked an object (Methods). Session 2 was recorded immediately after session 1. This Extended Data Figure shows that the temporal-tuning was generally conserved between the two sessions; however, it also shows that in session 2, the distribution of preferred-times in the Start location became more similar to that in locations A and B (panels a-b: compare the bottom-left panel in b to the two panels above it and to the bottom-left panel in a; and see also panel e – note in session 2 the distributions of preferred-times became more similar across the 3 landing-balls; see below for Kolmogorov-Smirnov tests). The main change in session 2 was a reduction in the percentage of cells with preferred-times < 1-s on the Start ball (panel f: note in session 2 [S2] the green and purple bars were more similar to each other than in session 1 [S1]). This figure suggests that since a major change between session 1 and session 2 was the presence of the demonstrator bat at the Start location in session 1, versus its absence in session 2, this presence/absence may underlie the observed neural differences in the firing sequences between the Start ball and balls A and B in session 1. (a) The temporal tuning-curve of time-cells was stable across consecutive sessions. Left column, temporal tuning curves of time-cells that were significantly-tuned in session 1. Cells were sorted according to their preferred time of firing. Right column, temporal tuning curves of the same cells which were tuned in session 1, but plotted for session 2; cells were sorted according to their preferred time in session 1. Note the stability of the internally-generated firing sequences across consecutive sessions (compare left and right panels). (b) Same analysis as in panel a, but for significant time-cells in session 2. Panels a and b demonstrate the stability of the sequences over the two sessions. (c) Violin plots of the distributions of Pearson correlations between the temporal tuning-curves in the two sessions; we repeated this calculation for each of the three locations. The correlations between the two sessions were very high at locations A and B (medians: ball A, r = 0.86, n = 51 cells; ball B, r = 0.91, n = 33 cells), and were statistically indistinguishable between balls A and B (two-sided Wilcoxon rank-sum test, P = 0.59; two-sided Kolmogorov-Smirnov test, P = 0.34) – indicating stability of the time representation across the two sessions for balls A and B. By contrast, the across-session correlations for the Start location were significantly lower (median on Start: r = 0.72, n = 17 cells; comparing correlations in A versus Start: two-sided Wilcoxon rank-sum test, P < 0.002; two-sided Kolmogorov-Smirnov test, P < 0.005; comparing correlations in B versus Start: two-sided Wilcoxon rank-sum test, P < 0.002; two-sided Kolmogorov-Smirnov test, P < 0.002) – consistent with the explanation that the presence of the demonstrator bat at the Start location in session 1 was responsible for the difference in the firing sequences in session 1 between the Start ball and the other two locations, A and B; note the demonstrator was removed from the room in session 2. (d) Gray bars, distribution of differences in preferred-times (ΔT) for the same neuron between session 1 and session 2 (at the same location). Plotted for all the time-cells that were significant in session 1; pooled across the 3 balls (n = 174 cells × positions; this number is smaller than the total number of time-cells in this study, because we included here only the significant time-cells where session 2 was run, which was only for a subset of the cells). The sharp peak at ΔT = 0 indicates that the preferred-time of time-cells was stable across sessions. Red line, shuffle distribution (cell shuffling: ΔT for cell i in session 1 minus cell j in session 2, for i ≠ j; two-sided Kolmogorov-Smirnov test of data versus shuffles, P = 6.8 × 10^–5). (e) Cumulative distribution functions (CDF) for the preferred-times of the time-cells in each location, for session 1 (left) and session 2 (right) (ball A: yellow; ball B: cyan; Start ball: green). In session 1 the distribution of preferred-times on the Start ball (green) was quite different from those on balls A or B. By contrast, in session 2, the CDF for the Start ball became statistically indistinguishable from the CDFs for balls A and B (two-sided Kolmogorov-Smirnov test on time segments between t = 0 and t = 4 s: Session 1: Start ball versus ball A: P = 0.047; Start ball versus ball B: P = 0.047; ball A versus ball B: P = 0.74; Session 2: Start ball versus ball A: P = 0.15; Start ball versus ball B: P = 0.37; ball A versus ball B: P = 0.74). Note that we removed the demonstrator bat from the room in session 2, so only in session 1 the observer bat was landing next to the demonstrator bat on the Start ball. Taken together, this suggests that the presence of the demonstrator bat at the Start location in session 1 was responsible for the difference in the firing sequences seen in session 1 between the Start ball and the other two locations, A and B – while in session 2, when the demonstrator was removed from the room, the time-cell sequences became more similar to each other. (f) Percentage of time-cells with short preferred-times (< 1 s). Magenta bars, the percentage of time-cells with short preferred-times on balls A and B was statistically indistinguishable between session 1 (S1; n = 213 cells) and session 2 (S2; n = 132 cells) (two-sided log odds ratio test: P = 0.055). Green bars, same for the Start ball: here, the percentage of time-cells with short preferred-times was significantly smaller in session 2 (S2; n = 50 cells) than in session 1 (S1; n = 61 cells) (two-sided log odds ratio test: P < 10^–5). Together, panels a-c and e-f suggest that the internally-generated firing sequence on the Start ball became more similar to those on balls A and B during session 2, when the other bat (demonstrator) was absent from the Start-ball and from the room altogether. (g) Venn diagram depicting the distribution of time-cell tuning in the different locations, in session 2. Note that we included in this figure only neurons that were stably spike-sorted across both sessions (see Methods section on ‘Spike sorting’).