Extended Data Fig. 9: Analysis of longitudinal calcium imaging experiments.

a, Correlation between freezing discrimination and engram discrimination at different delay times when using non-negative matrix factorization to identify engram cells in the longitudinal calcium imaging experiments in Fig. 4a (see Methods). Left, dynamic engram. Right, stable engram. Discrimination indices computed based on either the animal’s freezing levels or the average ΔF/F signals of engram cells during recall in the training and the neutral contexts (see Methods). Spearman’s rank correlation coefficient and associated two-sided test with P are shown. b-c, Longitudinal imaging reveals engram cell turnover in contextual fear memory. Analysis of engram cell ensembles identified in the longitudinal calcium imaging experiments in Fig. 4a (see Methods). b, Schematic depicting the set of imaged cells (N\({}_{{{{\rm{Im}}}}}\)), the set of engram cells identified at delay = 0 h (that is, fear training) (N₀), the set of engram cells identified at delay = 1 h (N₁) and the set of engram cells identified at delay = 24 h (N₂₄). By comparing N₀ and N₁, we can identify cells that dropped out of the engram, cells that dropped into the engram, and cells that remained in the engram (that is, engram overlap N₀ ∩ N₁) from delay = 0 to 1 h. By comparing N₁ and N₂₄, we can identify cells that dropped out of the engram, cells that dropped into the engram, and cells that remained in the engram (that is, engram overlap N₁ ∩ N₂₄) from delay = 1 to 24 h. c, Longitudinal engram overlap from delay = 0 to 1 h (that is, N₀ ∩ N₁) and from delay = 1 to 24 h (that is, N₁ ∩ N₂₄). Left, (N₀ ∩ N₁) / N₀ measures the fraction of engram cells at delay = 0 h that remained part of the engram at delay = 1 h and (N₁ ∩ N₂₄) / N₁ measures the fraction of engram cells at delay = 1 h that remained part of the engram at delay = 24 h. Middle, (N₀ ∩ N₁) / N₁ measures the fraction of engram cells at delay = 1 h that were also part of the engram at delay = 0 h and (N₁ ∩ N₂₄) / N₂₄ measures the fraction of engram cells at delay = 24 h that were also part of the engram at delay = 1 h. Right, (N₀ ∩ N₁) / N\({}_{{{{\rm{Im}}}}}\) measures the fraction of imaged cells that were engram cells both at delay = 0 and 1 h and (N₁ ∩ N₂₄) / N\({}_{{{{\rm{Im}}}}}\) measures the fraction of imaged cells that were engram cells both at delay = 1 and 24 h. To compare longitudinal engram overlap in mouse 1-4 to overlap at random, we generated ensembles of random cells using the following procedure. First, we took N\({}_{{{{\rm{Im}}}}}\) of an individual mouse and randomly drew \({{{{\rm{N}}}}}_{0}^{{{{\rm{random}}}}}\), \({{{{\rm{N}}}}}_{1}^{{{{\rm{random}}}}}\) and \({{{{\rm{N}}}}}_{24}^{{{{\rm{random}}}}}\) of the same size as N₀, N₁ and N₂₄ of the same mouse, respectively. We then repeated this procedure 10 times for each mouse. Finally, we compared longitudinal engram overlap in mouse 1-4 versus random cells using a two-sided Mann-Whitney U test. (N₀ ∩ N₁) / N₀, U = 108.5, P = 0.250176. (N₁ ∩ N₂₄) / N₁, U = 101.5, P = 0.389608. (N₀ ∩ N₁) / N₁, U = 99.5, P = 0.436077. (N₁ ∩ N₂₄) / N₂₄, U = 82.5, P = 0.934642. (N₀ ∩ N₁) / N\({}_{{{{\rm{Im}}}}}\), U = 97.5, P = 0.486051. (N₁ ∩ N₂₄) / N\({}_{{{{\rm{Im}}}}}\), U = 86.5, P = 0.805871. n = 4 mice. For mouse 1-4, individual data points as well as mean and 95% confidence intervals are shown. For random cells, mean and 95% confidence intervals are shown. Data compared using a Mann-Whitney U test met required assumptions (that is, continuous dependent variable, independent variable consisting of two independent groups, and independence of observations). d-g, Analysis of the distribution of ΔF/F-based discrimination indices of imaged cells in the longitudinal calcium imaging experiments in Fig. 4a (see Methods). Representative animal shown. d, Cumulative distributions of ΔF/F-based discrimination indices of imaged cells in the fear training session in Fig. 4a (dashed line indicates threshold \({\zeta }_{disc}^{thr}\) = 0.2 for engram cell identification). Red: engram cells identified during fear training that remained part of the engram in all imaging sessions. Black: all other imaged cells. Two-sided Kolmogorov-Smirnov test, P = 0.009213. e, Cumulative distributions of ΔF/F-based discrimination indices of imaged cells in the fear training session in Fig. 4a (dashed line indicates threshold \({\zeta }_{disc}^{thr}\) = 0.2 for engram cell identification). Red: engram cells identified during fear training that remained part of the engram in all imaging sessions. Black: engram cells identified during fear training that dropped out of the engram in any of the imaging sessions. Two-sided Kolmogorov-Smirnov test, P = 0.571078. f, Cumulative distributions of ΔF/F-based discrimination indices of imaged cells in the fear training session in Fig. 4a (dashed line indicates threshold \({\zeta }_{disc}^{thr}\) = 0.2 for engram cell identification). Red: cells that were not identified as engram cells during fear training but dropped into the engram in any of the remaining imaging sessions. Black: all other imaged cells. Two-sided Kolmogorov-Smirnov test, P = 0.001345. g, Cumulative distributions of ΔF/F-based discrimination indices of imaged cells in the fear training session in Fig. 4a (dashed line indicates threshold \({\zeta }_{disc}^{thr}\) = 0.2 for engram cell identification). Red: cells that were not identified as engram cells during fear training but dropped into the engram in any of the remaining imaging sessions. Black: cells that were not identified as engram cells during fear training and remained non-engram cells in all imaging sessions. Two-sided Kolmogorov-Smirnov test, P = 0.410507. c–g, *P < 0.05; ns, not significant.