Two different theoretical models have, from a neurophysiological basis, attempted to explain how declarative memory consolidation occurs during SWS. The active system consolidation theory (for reviews, see [12, 80]), incorporating findings from both the animal [61–64] and human [19, 43, 65] literatures, suggests that the same brain regions and functional connective systems active during encoding in wakefulness are reactivated during sleep. Specifically, this theory posits that memory traces encoded initially in both the hippocampus and neocortex are reactivated during subsequent episodes of SWS. During those stages of sleep , slow electrophysiological oscillations generated by the neocortex drive repeated reactivations of memory traces in the hippocampus. The upstate (represented as the peak of the wave formation on EEG recordings) of these slow oscillations synchronizes with sharp-wave ripple from hippocampal- and thalamic-driven spindles [34, 82]. This synchronization promotes the formation of ripple-spindle events, which in turn promote effective transfer of reactivated memory traces from the hippocampus to neocortical structures [41]. Hence, the initially unstable and relatively weak memory traces based in the hippocampus are incorporated increasingly strongly into preexisting networks of knowledge stored in neocortical circuits; during later waking activation, these memory traces are therefore less reliant on hippocampal activation [20].

Various studies support these findings. Healthy participants asked to encode word pairs and then recall them later perform better when the delay is filled with the SWS-rich first half of the night than when it is filled with the REM-rich second half of the night or with normal waking activity [21, 47, 48]. Additionally, when waves specific to SWS (i.e., very slow cortical oscillations of <1 Hz) are induced in the prefrontal cortex (PFC) using direct current stimulation, participants in the stimulation condition have greater word-pair retention than no-stimulation controls [2, 44].

In contrast to the active system consolidation theory, the synaptic homeostasis theory [67, 68] suggests that the slow oscillations during SWS act to decrease, rather than increase, neural connections or synaptic strength. Specifically, this theory posits that whereas during waking, learning and memory processes act to increase synaptic strength, during sleep synaptic connections are downscaled so that circuits do not become saturated and therefore unable to encode information successfully the next day. A result of this downscaling is that while weakly potentiated synapses are eliminated, strongly potentiated synapses remain, thus ensuring better recall of relevant information .

Recent studies focusing on biological markers of synaptic strength such as long-term potentiation (LTP) and long-term depression (LTD) have provided some support for this theory. LTP is one key mechanism explaining long-lasting synaptic strengthening in neural networks; LTD, in contrast, is a process that selectively weakens specific synaptic connections over a period of hours, thus clearing them of old memory traces and allowing the potentiation of new traces [32]. Glutamatergic AMPA receptors containing the subunit GluR1 have been found to play an important role in mediating LTP and LTD processes [42]. Vyazovskiy et al. [77], studying the rat hippocampus and cortex, showed that, during wakefulness, the levels of this receptor were high; during sleep, however, they were low. Further, changes in the expression of this receptor were consistent with LTP during the day and with LTD during the night. Importantly, changes in slow-wave activity in the rat brain during sleep were associated with changes in synaptic efficacy, further supporting the hypothesis that SWS is associated with LTD and therefore with synaptic downscaling .

Are these two different models of the neurophysiological mechanisms underlying memory consolidation during SWS sleep compatible with one another? On the face of it, they are not: The active system consolidation model suggests a strengthening of synaptic connections, and therefore of memory traces , whereas the synaptic homeostasis theory suggests an overall decrease in synaptic connections. Nonetheless, several authors (e.g., [12, 80]) have suggested that these models are in fact complementary and together may explain discrete aspects of the consolidation process during SWS sleep. That is to say, the processes can operate simultaneously: While active system consolidation may strengthen memory traces and integrate them into wider and preexisting networks of information, synaptic downscaling may eliminate superfluous neural connections and refresh synaptic potential for encoding. Furthermore, and with specific regard to memory consolidation, one set of hypotheses [80] suggests that these processes work together to improve signal-to-noise ratio : When superfluous neural connections are eliminated, the connections that remain have greater overall strength, thus ensuring optimal consolidation. In a recent update on the state of the art, Genzel et al. [23] suggest that these two models of sleep-dependent consolidation can be further reconciled by understanding the distinction between light NREM sleep (NREM 1 and NREM 2) and deep NREM (SWS). These authors argue that slow oscillations, sharp-wave ripples generated in the hippocampus, and surface EEG indicators (such as K-complexes and spindles) represent global markers of active system consolidation which is strongly (but not exclusively) associated with light NREM sleep. They furthermore propose that slow-wave activity and delta waves are local markers of synaptic homeostasis, which dominates over replay mechanisms related to active system consolidation in SWS. Each successive cycle of lighter and deeper NREM sleep results in the reorganization of information and removal of superfluous connections, respectively, which leads to optimal memory consolidation. Further research is still needed to demonstrate exactly how these processes may work alongside each other.

There is evidence for two distinct memory consolidation processes, each based on discrete electrophysiological events, during REM. The first process involves ponto-geniculo-occipital (PGO) waves. In rats, PGO density during REM increases following learning [10, 11, 69]. This increase is associated with post-sleep task performance improvements. Furthermore, PGO waves are associated with the expression of immediate early genes (IEGs) , which express during LTP in REM sleep and result in long-term synaptic strengthening in the brain [11]. Following learning, IEG activity is localized to brain regions which were involved in the acquisition of new material [52, 53, 69]. A recent study also demonstrated that if REM sleep is selectively deprived after learning in rats, both memory consolidation and LTP are impaired [50].

The second electrophysiological process that promotes memory consolidation during REM is the expression of the theta rhythm [3, 12]. During waking, theta activity occurs during the encoding of hippocampal-dependent memories [9]. During sleep, there is evidence of neuronal replay, associated with theta activity, of this encoded information in the hippocampus [40, 49]. Furthermore, in mice, silencing of REM-specific theta rhythms only (i.e., leaving intact those rhythms during other stages of sleep) selectively impairs consolidation of, for instance, neutral information related to novel object place recognition [3]. In humans, theta activity during REM increases after the learning of word pairs before sleep [17]. Current research shows that theta rhythm and PGO waves are related, in that PGO waves are phase-locked to theta rhythm [51]. In animal models, eliciting theta waves during REM sleep in the medial septum entrains PGO waves to the theta rhythm. In humans, PGO waves are impossible to measure given the current technology because they occur in medial regions of the brain and are not easily detected via surface EEG [29]. Hence, the relationship between theta rhythm and PGO waves, and their distinct or overlapping contribution to memory consolidation, remains unclear in humans.

In summary, these theoretical frameworks and empirical data help explain why the progression of sleep stages over the night results in optimal memory consolidation. During early-night SWS, relevant memory traces are selected (by the strength of their potentiation at synapses; weak connections are eliminated) and integrated into preexisting networks of knowledge. During REM sleep, these traces are strengthened so that they form long-lasting representations in the brain. Of note here is that IEG activity is correlated with EEG spindle activity during the preceding SWS. One hypothesis, then, is that SWS primes networks for later IEG expression [55].

Numerous empirical studies have shown that acquisition of newly-presented declarative information is more efficient after a session of uninterrupted sleep than after a period of sleep deprivation or after a session of fragmented sleep. For instance, Harrison and Horne [28] showed that healthy participants who had been sleep deprived for 36 h performed significantly more poorly on a face recognition task than those who had slept for the normal 8 h prior to testing. Even when a subgroup of the sleep-deprived participants was administered caffeine to combat non-specific effects of lower alertness, the sleep-refreshed group performed better.

There is a clear neural basis for these between-group differences. Functional MRI studies exploring the sleep deprivation and memory testing paradigm described above have shown that, compared to those who sleep normally, sleep-deprived individuals have less medial temporal lobe (and, specifically, hippocampal) activation and greater (perhaps compensatory) PFC and parietal activation [13, 86]. Because, in these studies and in others (e.g., [14, 59, 87]), successful encoding is associated with extent of hippocampal activation, it appears that healthy sleep is important in preparing the appropriate neural circuits for next-day acquisition of novel information.

These empirical data are consistent with the theoretical frameworks outlined above. By the active consolidation model, one proposed result of the dialogue between the neocortex and hippocampus during SWS is that retrieval of “old” memories is less reliant on hippocampal involvement during subsequent (daytime) reactivations. Similarly, the synaptic homeostasis model posits that, upon waking, hippocampal networks are refreshed as a result of synaptic downscaling during SWS. Both theories, then, predict that the hippocampus will be refreshed and ready for new encoding after a night of healthy, uninterrupted sleep.

Only three recent studies, from three different laboratories, have directly investigated memory processing during sleep in PTSD . First, van Liempt et al. [73] showed that combat veterans with PTSD (n = 13) experienced more fragmented sleep during the first, SWS-rich, half of the night than did trauma controls (i.e., veterans who had experienced trauma but who did not meet diagnostic criteria for PTSD; n = 15) and healthy controls (i.e., individuals who had never experienced a DSM-IV A1 criterion traumatic event; n = 15). Furthermore, relative to healthy controls, veterans with PTSD showed significantly lower plasma levels of growth hormone (GH) during the night. This latter finding is important because: