Extended Data Fig. 3: Sensory reactivation during memory consolidation is essential for engram selectivity, and analysis of stimulus-evoked neuronal firing rates during training.

a-i, Simulation with blockage of training-activated stimulus neurons during consolidation. a, Schematic of simulation protocol with blockage of training-activated stimulus neurons during consolidation. Network and simulation parameters as in Fig. 1b. b-d, Post-encoding evolution of engram cells in a. Means and 99% confidence intervals are shown. n = 10 trials. b, Engram cell ensemble as a fraction of all neurons (see Methods). Dashed line indicates engram cell ensemble at the end of training. c, Ensemble overlap between probing-activated engram cells at consolidation time = t and t-1 h as a fraction of engram cells at consolidation time = t-1 h. Dashed line indicates ensemble of neurons that remained part of the engram in all sampled time points (that is, consolidation time = 0, 1, …, 24 h) as a fraction of engram cells at consolidation time = 0 h (that is, training-activated engram cells). d, Ensemble overlap between engram cells activated during both probing and training as a fraction of training-activated engram cells (top), probing-activated engram cells (middle) and all neurons in the network (bottom). e-h, Analysis of memory recall in a. Means and 99% confidence intervals are shown. n = 10 trials. Color denotes stimulus as in Fig. 1c. e, Firing rate of engram cells averaged across all cue presentations during recall as a function of consolidation time. Dashed line indicates threshold ζ^thr = 10 Hz for engram cell activation. f, Memory recall as a function of consolidation time. g, Discrimination index between recall evoked by cues of the training stimulus and individual novel stimuli as a function of consolidation time (see Methods). h, Fraction of probing-activated engram cells reactivated during recall as a function of consolidation time. i, Mean weight strength of plastic synapses in the network in a clustered according to engram cell status. Top, feedforward excitatory synapses onto excitatory neurons. Middle, recurrent excitatory synapses onto excitatory neurons. Bottom, recurrent inhibitory synapses onto excitatory neurons. Left, at the end of the training phase. Right, after 24 h of consolidation. Representative trial is shown. j-m, Analysis of the distribution of stimulus-evoked neuronal firing rates in the training phase in Fig. 1b (dashed line indicates threshold ζ^thr = 10 Hz for engram cell activation). j, Cumulative distributions of stimulus-evoked neuronal firing rates in the training phase in Fig. 1b. Red, training-activated engram cells that remained part of the engram in all sampled time points (that is, consolidation time = 0, 1, …, 24 h). Black, all other neurons. Two-sided Kolmogorov-Smirnov test, P = 1.776357 × 10⁻¹⁵. k, Cumulative distributions of stimulus-evoked neuronal firing rates in the training phase in Fig. 1b. Red, training-activated engram cells that remained part of the engram in all sampled time points (that is, consolidation time = 0, 1, …, 24 h). Black, training-activated engram cells that dropped out of the engram in any of the sampled time points (that is, consolidation time = 0, 1, …, 24 h). Two-sided Kolmogorov-Smirnov test, P = 3.330669 × 10⁻¹⁶. l, Cumulative distributions of stimulus-evoked neuronal firing rates in the training phase in Fig. 1b. Red, neurons that were not engram cells at the end of training but dropped into the engram in any of the sampled time points (that is, consolidation time = 0, 1, …, 24 h). Black, all other neurons. Two-sided Kolmogorov-Smirnov test, P = 0.0. m, Cumulative distributions of stimulus-evoked neuronal firing rates in the training phase in Fig. 1b. Red, neurons that were not engram cells at the end of training but dropped into the engram in any of the sampled time points (that is, consolidation time = 0, 1, …, 24 h). Black, neurons that were not engram cells at the end of training and remained non-engram cells in all sampled time points (that is, consolidation time = 0, 1, …, 24 h). Two-sided Kolmogorov-Smirnov test, P = 1.060787 × 10⁻⁴⁰. j–m, *P < 0.05.