Personalized targeted memory reactivation enhances consolidation of challenging memories via slow wave and spindle dynamics

TL;DR

This study addresses the challenge that fixed targeted memory reactivation (TMR) protocols fail to account for individual learning capacity, limiting gains for difficult memories. It introduces a personalized TMR protocol that adjusts cueing frequency based on retrieval performance and task difficulty during a word-pair task, comparing it to standard TMR and no-stimulation controls. The personalized protocol produced stronger memory gains for the hardest items, accompanied by enhanced slow-wave and spindle activity and stronger SW–spindle coupling, with behavior-EEG correlations and multivariate EEG signatures supporting memory-specific circuit engagement. The findings offer a practical path toward sleep-based, individualized cognitive enhancement and illuminate the neural mechanisms by which SW-spindle dynamics support sleep-dependent memory consolidation.

Abstract

Sleep is crucial for memory consolidation, underpinning effective learning. Targeted memory reactivation (TMR) can strengthen neural representations by re-engaging learning circuits during sleep. However, TMR protocols overlook individual differences in learning capacity and memory trace strength, limiting efficacy for difficult-to-recall memories. Here, we present a personalized TMR protocol that adjusts stimulation frequency based on individual retrieval performance and task difficulty during a word-pair memory task. In an experiment comparing personalized TMR, TMR, and control groups, the personalized protocol significantly reduced memory decay and improved error correction under challenging recall. Electroencephalogram (EEG) analyses revealed enhanced synchronization of slow waves and spindles, with a significant positive correlation between behavioral and EEG features for challenging memories. Multivariate classification identified distinct neural signatures linked to the personalized approach, highlighting its ability to target memory-specific circuits. These findings provide novel insights into sleep-dependent memory consolidation and support personalized TMR interventions to optimize learning outcomes.

Figures (5)

• Figure 1: Experimental setup. a The experiment comprises three main sessions: pre-sleep, sleep, and post-sleep. Participants were randomly assigned to one of three groups during the sleep session: the personalized TMR, TMR, and CNT groups. The questionnaire (Q) was administered during the wakefulness periods, alongside tasks such as word-pair learning and a 5-min eyes-closed resting state (RS) in both pre- and post-sleep sessions. b In the encoding phase of the pre-sleep session, participants were exposed to 104-word pairs presented via audiovisual stimuli. In the retrieval phase (pre- and post-sleep sessions), participants recalled and typed the associated word upon viewing the learned cue. Immediately following each retrieval trial, participants rated their expected difficulty in recalling the word pair both immediately and after a 12-h delay using a three-level scale: L1, L2, and L3. These prospective self-ratings were used to classify word-pair difficulty levels and were not updated or redefined across retrieval rounds. Re-encoding was provided only during the pre-sleep retrieval phase. c A hypnogram displaying the sleep session for one participant shows the NREM 2 and NREM 3 stages, during which the stimuli were administered, indicated by gray shading. d Stimulation protocols varied among groups: the personalized TMR group adjusted the number of presentations (PRES) based on response correctness—1 PRES for L2 correct responses, 2 PRES for L2 incorrect responses, and 4 PRES for L3 correct and incorrect responses. The TMR group received the same stimuli uniformly, while the CNT group received no stimulation.
• Figure 2: Memory accuracy. Memory accuracy for all trials and for each difficulty level (L1, L2, and L3) is shown for the personalized TMR, TMR, and CNT groups. In the top row, the light-colored bars represent pre-sleep session accuracy and the dark-colored bars represent post-sleep session accuracy. The bottom row displays the difference in accuracy between the pre-sleep and post-sleep sessions. Significant differences (two-sample t-tests with Bonferroni correction, p$<$ 0.05) are marked with asterisks.
• Figure 3: Memory transition. Memory transition ratios for four outcomes (correct-correct, correct-incorrect, incorrect-correct, and incorrect-incorrect) are presented for the personalized TMR, TMR, and CNT groups. The top row shows a heatmap of transition ratios, and the bottom row presents a bar plot of mean transition values and standard errors, calculated across participants and recall levels. Significant differences (two-sample t-tests with Bonferroni correction, p$<$ 0.05) are indicated with asterisks.
• Figure 4: Electrophysiological responses to auditory stimuli during sleep. a Average event-related potentials (ERP) are displayed across all channels for the three groups in response to auditory cues. Each auditory cue consisted of a word pair presented sequentially within a 4-s window: the first word was delivered at stimulus onset (0 s), followed by the second word approximately 2 s later. The baseline period (-0.5 to 0 s) is indicated by a black bar, and stimulus onset is marked by a dashed line. Colored lines indicate group-specific ERP waveforms, and horizontal bars above the waveforms denote time intervals showing significant group differences. b Time-frequency representations are averaged across channels, with the top row displaying group-level spectral responses and the bottom row highlighting significant intergroup differences. c Event-related phase-amplitude coupling results illustrate slow wave (SW)-spindle coupling, with the top row presenting group results and the bottom row indicating significant differences. d Comparisons of SW power, spindle power, and SW–spindle coupling across groups reveal significant differences; significant findings from two-sample t-tests with Bonferroni correction (p $<$ 0.05) are marked with asterisks. e Pearson’s correlations between behavioral and EEG results indicate significant interactions, marked with a dagger (p$<$ 0.05, FDR correction).
• Figure 5: Classification results. Multivariate classification performance for distinguishing personalized TMR, TMR, and CNT groups in All and level 3 (L3) conditions based on EEG-derived features (slow wave (SW) power, spindle power, and SW–spindle coupling) is presented. The shaded solid line represents observed decoding performance, while the shaded dashed line shows surrogate decoding results from 250 label shuffles. Accuracy values are presented with means and standard errors. The horizontal black solid line indicates chance-level performance, and the lower horizontal line highlights statistically significant time points, identified via a two-sided cluster-based permutation test with 1,000 randomizations.