System consolidation of memory during sleep

Abstract

Over the past two decades, research has accumulated compelling evidence that sleep supports the formation of long-term memory. The standard two-stage memory model that has been originally elaborated for declarative memory assumes that new memories are transiently encoded into a temporary store (represented by the hippocampus in the declarative memory system) before they are gradually transferred into a long-term store (mainly represented by the neocortex), or are forgotten. Based on this model, we propose that sleep, as an offline mode of brain processing, serves the 'active system consolidation' of memory, i.e. the process in which newly encoded memory representations become redistributed to other neuron networks serving as long-term store. System consolidation takes place during slow-wave sleep (SWS) rather than rapid eye movement (REM) sleep. The concept of active system consolidation during sleep implicates that (a) memories are reactivated during sleep to be consolidated, (b) the consolidation process during sleep is selective inasmuch as it does not enhance every memory, and (c) memories, when transferred to the long-term store undergo qualitative changes. Experimental evidence for these three central implications is provided: It has been shown that reactivation of memories during SWS plays a causal role for consolidation, that sleep and specifically SWS consolidates preferentially memories with relevance for future plans, and that sleep produces qualitative changes in memory representations such that the extraction of explicit and conscious knowledge from implicitly learned materials is facilitated.

Fig. 1

Active system consolidation during sleep. a During slow-wave sleep (SWS) memories newly encoded into a temporary store (i.e. the hippocampus in the declarative memory system) are reactivated to be redistributed to the long-term store (i.e. the neocortex). b System consolidation during SWS relies on a dialogue between neocortex and hippocampus under top-down control by the neocortical slow oscillations (red). The depolarizing up phases of the slow oscillations drive the repeated reactivation of hippocampal memory representations together with sharp-wave ripples (green) in the hippocampus and thalamo-cortical spindles (blue). This synchronous drive allows for the formation of spindle-ripple events where sharp-wave ripples and associated reactivated memory information becomes nested into single troughs of a spindle (shown at larger scale); in the black-and-white version of the figure red, green and blue correspond to dark, middle and light grey

Fig. 2

Stabilizing and labilizing effects of memory reactivation during sleep and wakefulness, respectively. Subjects learned the locations of card pairs (similar to the game ‘Concentration’) while odour was presented as an associative context stimulus. Learning was followed by retention intervals filled with either wakefulness (Wake) or sleep (Sleep; i.e. NonRem sleep as no REM sleep occurred in these intervals). During these intervals the subjects were re-exposed to the odour to reactivate the card location memories. Vehicle was presented in the non-reactivation control condition. Periods of memory reactivation (and vehicle administration) were followed by learning an interference task (with the same cards but different locations) to probe stability of the originally learned card location. a Final recall (mean ± SEM) of the originally learned card-pair locations tested 30 min after interference learning. Performance is given as percentage, with the number of card locations recalled at learning before the retention interval set to 100%. Note, reactivation stabilized memories (i.e. made them resistant to interference learning) when the odour was re-exposed during SWS, but labilised memories when the odour was re-exposed during wakefulness. b Odour-induced memory reactivation during wakefulness increased activity in the lateral prefrontal cortex (left) whereas reactivation during SWS strongly activated the left anterior hippocampus (right) Thresholded at P < 0.001 uncorrected; superimposed on a T₁-template image (adapted from Diekelmann et al. 2011)

Fig. 3

a Effects of retrieval expectancy on retention of declarative word pair memories across 9-h intervals filled with nocturnal sleep, night-time wakefulness and daytime wakefulness. Retention performance is indicated by the percentage of word pairs recalled at retrieval with performance on the criterion trial during learning set to 100%. Memory performance was enhanced after post-learning sleep in comparison with wakefulness only when subjects expected the retrieval test (expected, black bars) but not when the retrieval was unexpected (unexpected, empty bars). b During post-learning sleep, slow oscillation power (0.68–1.17 Hz) within the first 120 min of NonREM sleep (indicated for successive 20-min intervals, 0–20, 20–40 min, etc.) was enhanced when subjects expected retrieval testing (solid lines) in comparison to subjects who did not expect the retrieval (dotted lines). c Slow oscillation power during the first 20-min period of post-learning NonREM sleep was highly correlated to retention of word pairs when subjects expected retrieval testing (filled circles, solid line), but not when retrieval testing was unexpected (empty circles, dotted lines) *P < 0.05, **P < 0.01, ***P < 0.001 (adapted from Wilhelm et al. 2011)

Fig. 4

Extraction of explicit knowledge from an implicitly learned task during slow-wave sleep (SWS). a Subjects implicitly learned a number reduction task (NRT) before retention intervals of nocturnal sleep, daytime wakefulness, or night-time wakefulness. Percentage of subjects gaining explicit insight into the hidden rule of the task after the retention intervals is indicated. The percentage of subjects gaining insight was more than twofold higher when initial practice was followed by sleep in comparison with wakefulness. b If some implicit knowledge about the rule underlying the NRT had been obtained at initial practice, those participants who gained insight after sleep showed significantly higher power in the slow spindle band (8–12 Hz, shaded area) and in the beta frequency band (17–25 Hz) during post-practice SWS (red, upper line) compared with those participants who failed (black, lower line, adapted from Wagner et al. , and Yordanova J., Kolev V., Wagner U. et al., submitted)