Why the moment of the cue matters
During deep sleep, the brain produces slow oscillations—large, rhythmic waves of electrical activity. A cue that arrives on the up-going slope of one of these waves is more likely to trigger a measurable reactivation of the associated memory than a cue that arrives at other phases [1]. In one experiment, words linked to sounds were replayed during sleep; words cued at the UP state of the slow oscillation showed significantly less overnight forgetting than words cued at the DOWN state [2]. The UP-state cueing also strengthened the ongoing brain wave and was followed by more spindle activity—brief bursts of faster brain waves that are thought to support memory consolidation [2].
The timing effect is not just about the slow oscillation's phase. Reactivations were observed either immediately after the cue or about one second later, suggesting that the brain's response to a cue unfolds over a short window rather than instantly [1]. This means that a cue delivered at the right phase can set off a cascade that takes a moment to develop.
The cue must be linked to the memory—and overlap can backfire
For TMR to work, the cue needs to have been associated with the learning content beforehand. In a typical study, participants learn word–image pairs while hearing a sound linked to the image; later, that sound is played during sleep to reactivate the specific memory [2]. The cue is not a generic reminder—it is a specific key to a specific memory.
But the relationship between cue and memory is not always simple. When participants learned overlapping pairs of associations—where the same cue could point to more than one memory—cueing one pair during sleep increased retention for the first-learned pair but decreased memory for the second-learned pair [5]. This forgetting effect appeared only for pairs that had not been tested before sleep; memories strengthened by prior retrieval did not show the same decline [5]. So the cue-to-item condition matters: a cue that reactivates one memory can interfere with a related, overlapping memory, especially if that second memory has not been recently reinforced.
What else shapes whether TMR helps
The brain's response to a cue also depends on the presence of spindles and the features of the slow oscillation at the moment of cueing; these can be used to predict whether a cue will successfully elicit a classifiable reactivation [1]. This suggests that the optimal timing is not a single fixed point but a combination of conditions—phase of the slow wave, spindle activity, and perhaps the cue's own timing relative to these events.
TMR can also be used to promote forgetting, not just strengthening, depending on the memory domain and the specific conditions [3]. The technique has been studied in declarative, procedural, and emotional memory, and its effects can differ across these domains [3]. In addition, TMR modulates cortical networks differently during sleep than during wake, with increased theta-band connectivity after cues in slow-wave sleep compared to wake periods [4]. These network-level differences may help explain why the same cue can have different effects depending on when it is delivered.
What this means for using TMR
If the goal is to strengthen a memory, the evidence points to delivering the cue at the up-going transition or UP state of the slow oscillation, when the brain is most receptive [1][2]. The cue should be one that was previously paired with the target memory, and the memory should not be part of a set of overlapping associations that could compete with each other—or if it is, the target memory should be strengthened by retrieval before sleep to avoid the forgetting effect [5]. The timing window is short: reactivations occur immediately after the cue or about a second later, so precision matters [1]. These findings are still being translated into wearable technologies, and the optimal conditions may vary with the memory type and the individual's sleep patterns [1][3].
Sources used in this answer
Targeting targeted memory reactivation: Characteristics of cued reactivation in sleep
Cues falling on the up-going transition of the slow oscillation are more likely to elicit a classifiable memory reactivation, and reactivations occur either immediately after the cue or one second later.
Shaping overnight consolidation via slow-oscillation closed-loop targeted memory reactivation
Cueing at slow-oscillation UP states reduced overnight forgetting compared with DOWN-state cueing, enhanced the ongoing UP state, increased spindle power, and led to reinstatement of target image representations.
An update on recent advances in targeted memory reactivation during sleep
Recent advances in TMR include effects on strengthening and forgetting across declarative, procedural, and emotional memory domains, as well as technical approaches to determine optimal timing and cue specificity.
Effects of Targeted Memory Reactivation on Cortical Networks
TMR cues during slow-wave sleep increased theta-band connectivity between frontal and other cortical regions and between hemispheres compared with wake, and hand-specific connectivity could be identified after cues during sleep.
Targeted memory reactivation during sleep can induce forgetting of overlapping memories
Cueing overlapping associative memories during sleep increased retention for first-encoded pairs but decreased memory for second-encoded pairs, and this forgetting effect was present only for pairs not tested prior to sleep.