Item-specific memory reactivation during sleep supports memory consolidation in humans

Abstract

Understanding how individual memory traces are reactivated during sleep is instrumental to the theorizing of memory consolidation, a process during which newly acquired information becomes stabilized and long-lasting. Via targeted memory reactivation (TMR), a technique that unobtrusively delivers learning-related memory cues to sleeping participants, we examined the reactivation of individual memories during slow-wave sleep and how canonical neural oscillations support item-specific memory reactivation. Furthermore, we investigated how pre-sleep testing, which presumably induces fast consolidation before sleep, would modulate sleep-mediated memory reactivation. Applying multivariate representational similarity analysis (RSA) to the cue-elicited electroencephalogram (EEG), we identified significant item-specific representations in two post-cue time windows (620-1350 ms and 2270-3190 ms) for post-sleep remembered items, with only the later item-specific representations contributing to memory consolidation. Untested items (i.e., items that were not tested pre-sleep), but not tested items, elicited higher spindles that predicted stronger item-specific representations and post-sleep memories. Notably, for untested items, the strengths of memory reactivation and spindles were temporally coupled to the up-state of the slow oscillation activity. Together, our results unveiled how item-specific memory reactivation and its link with neural oscillations during sleep contributed to memory consolidation. This knowledge will benefit future research aiming to perturb specific memory episodes during sleep.

Targeted memory reactivation during non-rapid eye movement sleep enhances neutral, but not negative, components of memory

“…For each spindle that was detected following a TMR cue, its amplitude was extracted. To calculate the probability of a spindle occurring following a TMR cue, established procedures were followed (Schechtman et al, 2021;Liu et al, 2023). For each trial, a spindle value of 1 was assigned to the time points where a spindle was detected, and 0 where no spindle was detected.…”