**3. Sleep and memory**

Likewise the heterogeneous structure of sleep, memory categories believed to exist in human brain are distinctly different. Roughly, they can be divided into declarative and nondeclarative or procedural memory, and the same categorization holds true for the processes of learning (Dienes & Perner 1999). Declarative memory is considered to comprise consciously accessible memories of fact-based information (i.e., knowing "what"). Several subcategories of the declarative system exist, including episodic memory (autobiographical memory for events of one's past) and semantic memory (memory for general knowledge, not tied to specific events, but rather tied to its verbal components) (Tulving 1985). Current neural models of declarative memory formation emphasize the critical importance of structures in the medial temporal lobe, especially the hippocampus, and thus, declarative memory is also known as hippocampus-dependent memory (Eichenbaum, 2000). In contrast, procedural or implicit memory is regarded as non-conscious, comprising memory of knowing "how", such as learning of actions, habits, and skills, as well as implicit learning (Dienes & Perner 1999). Procedural memory formation appears to be less dependent on medial temporal lobe structures, and to include sensori-motor cortices, basal ganglia and cerebellum (Forkstam & Petersson 2005).

#### **3.1 Overview of human EEG data**

Since the discovery of SO (< 1 Hz) in cats (Steriade et al., 1993) and (~ 0.7-0.8 Hz) in humans (Achrermann & Borbely, 1997), this hallmark of human non-REM sleep, and SWS in

suppressed during non-REM sleep (Broun et al., 1997; 1998; Maquet et al., 1996; Miyauchi et al., 2009; Nofzinger et al., 1997). Yet, other brain structures are shown to specifically increase their activation in relation to distinct non-REM sleep stages and their EEG signatures. For example, blood oxygen level-dependent (BOLD) signal from the thalamus is strongest during spindle activity in stage 2 of non-REM sleep (Schabus et al., 2007), whereas during SO in SWS, brain functional activation is strong in the medial temporal cortex, the parahippocampal cortex, and neocortical areas (Dang-Vu et al., 2005; 2008; Maquet et al.,

Notably, from a cognitive point of view, the mental characteristics of sleep-wake stages well correspond to their brain activation patterns found in neuroimaging studies (Fosse et al., 2001; 2004; Hobson & Pace-Schott, 2002; Hobson et al., 2000; Stickgold et al., 2001). Thus, wake is characterized by strongest and most logic thoughts in the presence of sensory input, executive control and goal-directed behavior. Sleep onset is hallmarked by the so called hypnagogic hallucinations, and as sleep deepens from stage 2 of non-REM sleep to SWS, thinking becomes more and more scares, and almost absent during SWS. Yet, there are logic thoughts mostly associated with previous wake experiences (Fosse et al., 2004; Hobson, 2005; Hobson et al., 2000; Stickgold et al., 2001). During REM sleep, mental activity is likely hallucinatory, and is behaviorally expressed in vivid, bizarre and elusive dreams (Fosse et al., 2004; Hobson et al., 2000; Stickgold et al., 2001). Also, REM sleep mentality is characterized by a most salient emotional tone upon lack of sensory input (Hobson et al., 2000) and executive control, as evidenced by the neuroimaging data (suppression of the

Likewise the heterogeneous structure of sleep, memory categories believed to exist in human brain are distinctly different. Roughly, they can be divided into declarative and nondeclarative or procedural memory, and the same categorization holds true for the processes of learning (Dienes & Perner 1999). Declarative memory is considered to comprise consciously accessible memories of fact-based information (i.e., knowing "what"). Several subcategories of the declarative system exist, including episodic memory (autobiographical memory for events of one's past) and semantic memory (memory for general knowledge, not tied to specific events, but rather tied to its verbal components) (Tulving 1985). Current neural models of declarative memory formation emphasize the critical importance of structures in the medial temporal lobe, especially the hippocampus, and thus, declarative memory is also known as hippocampus-dependent memory (Eichenbaum, 2000). In contrast, procedural or implicit memory is regarded as non-conscious, comprising memory of knowing "how", such as learning of actions, habits, and skills, as well as implicit learning (Dienes & Perner 1999). Procedural memory formation appears to be less dependent on medial temporal lobe structures, and to include sensori-motor cortices, basal ganglia and

Since the discovery of SO (< 1 Hz) in cats (Steriade et al., 1993) and (~ 0.7-0.8 Hz) in humans (Achrermann & Borbely, 1997), this hallmark of human non-REM sleep, and SWS in

**2.5 Mental characteristics of the sleep-wake stages** 

dorsolateral prefrontal cortex, e.g., Maquet et al., 1996).

**3. Sleep and memory** 

cerebellum (Forkstam & Petersson 2005).

**3.1 Overview of human EEG data** 

1997).

particular, has been proposed as an essential mechanism underlying the consolidation of hippocampus-dependent memories (Buzsáki, 1989; Marshall & Born, 2007; Mölle et al., 2004; Steriade, 2001). In human SWS, "up" and "down" EEG states of SO are shown be dissimilarly associated with a number of electrophysiological events. Specifically, the "up" state of the SO is marked by increased occurrence of delta slow waves and sleep spindles, whereas during the 'down" state of SO, delta slow wave and sleep spindle activities markedly decrease (Mölle et al., 2002; 2004). Thus, human SO are demonstrated to group both slow waves and spindles. Further, a rapid increase of the underlying neuronal activity (depolarizing state) and a rapid decrease in it (hyperpolarizing state) have recently been shown to characterize human SO "up" and "down" wave forms, respectively (Nir et al., 2011). Finally, human studies have demonstrated time and phase coupling between SO, slow waves, sleep spindles, and hippocampal SWR bursts (Clemens et al., 2007; 2011; Nir et al., 2011). Collectively, these findings strongly indicate that non-REM sleep/SWS SO represent an EEG mechanism involved in plastic changes sub-serving the hippocampusdependent memory consolidation.

Indeed, SO have been shown to be strongly associated with both procedural (Huber et al., 2004) and declarative (Mölle et al., 2004) memory consolidation taking place in human non-REM sleep/SWS. Later, to verify the specific role for SO in hippocampus-dependent memory consolidation, a series of studies, in which brain rhythms have been modulated using trans-cranial direct current stimulation (tDCS), has been conducted in humans. This method is now recognized as a reliable tool for modulating both the internally generated brain rhythms and the activity of underlying neuronal populations, depolarized under anodal tDCS and hyperpolarized under cathodal tDCS, respectively (Fröhlich & McCormick, 2010; Reis et al., 2008).

Initially, weak (not perceived by subjects) anodal tDCS oscillating at 0.75 Hz (slow oscillation stimulation, SOS) has been delivered during the transition from stage 2 of non-REM sleep to SWS after declarative and procedural learning before sleep. Compared with a sham condition, stimulation has selectively produced a gain in only declarative memory after sleep. Importantly, it has also produced a substantial increase in SO (~ 0.75 Hz) and frontal slow alpha spindle (8-12 Hz) activity, possibly by entraining these sleep EEG rhythms (Marshall et al., 2006). These findings have provided strong evidence for the role of SO and/or frontal slow spindle activity for the hippocampus-dependent memory consolidation. However, they have not addressed the question of whether SOS itself or whether endogenous SO boosted by the SOS have resulted in improvement of the consolidation of declarative memory found. This question was addressed in two later studies. In these studies, weak anodal tDCS oscillating at frequencies not common for the respective functional brain states was applied. In particular, SOS oscillating at 0.75 Hz was applied during quiet or resting wake retention period after learning declarative and procedural tasks. SOS did not affect either declarative or procedural memory consolidation, nor did it affect working memory and mood at retest. However, in contrast to its EEG effects during non-REM sleep (Marshall et al., 2006), it produced only a local (at the frontal sites of stimulation) increase in EEG power in SO (0.4-1.2 Hz) frequency band, accompanied by a widespread and strong increase in theta (4-8 Hz) EEG power. Further, when delivered in active wake state during encoding of a verbal learning memory task, the 0.75 Hz SOS produced virtually the same EEG effects as during quite wake (local increase in SO and widespread increase in theta power), but it significantly improved encoding of declarative verbal information (Kirov et al., 2009). Recently, tDCS oscillating at 5 Hz (wake and/or REM

The Memory, Cognitive and Psychological Functions of Sleep:

associated with memory processes.

**3.2 Overview of human neuroimaging data** 

Update from Electroencephalographic and Neuroimaging Studies 163

dependent and experience depended sleep processes. They subtracted sleep spindle EEG activity measured at a "non-learning hemisphere (left)" from that measured at a "learning hemisphere (right)" and were able to demonstrate strong positive correlations with offline memory improvement that were not evident for either hemisphere alone (Nishida & Walker, 2007). Thus, classical sleep spindles appear to represent an EEG pattern reliably

Notably, the first human neuroimaging study almost entirely confirm previous rodent findings, which have shown that sleep-dependent memory consolidation relies on reactivation of the so-called hippocampus place cells during SWS in response to spatial or maze navigating tasks (Wilson & McNaughton, 1994). Thus, Peigneux et al. (2004) using a combination of PET and sleep EEG (PSG) methods, showed for the first time such a reactivation of hippocampal and neocortical regions in humans. At encoding during wakefulness, subjects were trained (or not) to learn and find their way inside a complex three-dimensional virtual town. Compared with the non-trained group, subjects who learned the task at encoding during wake displayed an increase in their regional cerebral blood flow (rCBF) in a bilateral pattern of neural activation, including the right and left hippocampus, the right and left parahippocampal gyri, superior parietal lobules, right and left precuneus, lingual and posterior cingulate gyri, middle and superior occipital cingulate gyri, and the anterior lobes of cerebellum. In the trained group, rCBF was consequently measured during stage 2 of non-REM sleep, SWS and REM sleep. The neural pattern of activity found at training sessions during wakefulness was reactivated only during SWS. This pattern of reactivation during SWS strongly correlated with the level of improvement on the task at recall after sleep (Peigneux et al., 2004). Three years later, Rasch et al. (2007) also used a combination of fMRI and PSG methods to investigate specifically the role for SWS in declarative memory consolidation. The authors first established a robust association between declarative learning (card place location) stimuli and a smell, olfactory stimulus (the smell of a rose). Subjects learned object locations in a two-dimensional (2D) object location memory task in the evening before sleep. During the first two periods of subsequent SWS, the odor cue was presented again (in an alternating 30 s on/30 s off mode). In a control condition, odorless vehicle was delivered. At retrieval testing after sleep, memory of card locations was distinctly enhanced when the odor was presented during SWS as compared to presentation of vehicle alone. Re-exposure to odor during SWS improved retention of memory in a hippocampus-dependent manner: bilateral reactivation of the hippocampus and medial prefrontal lobe, as measured by BOLD signal (Rasch et al., 2007). Further, an fMRI study demonstrated that compared with total sleep deprivation, post-learning sleep enhances hippocampal responses during recall of word pairs (declarative memory) 48 h after learning, thus indicating intra-hippocampal memory processing during sleep (Gais et al., 2007).. At the same time, it was shown that sleep induced a memory-related functional connectivity between the hippocampus and the medial prefrontal cortex. Six months after learning, recalling the same declarative memories reactivated the medial prefrontal cortex even more strongly than they did during encoding before sleep, thus showing that sleep leads to long lasting changes in representation of memories at a system (hippocampal-cortical) level. Although this study does not show

sleep EEG rhythm, theta stimulation) was applied in the first non-REM sleep cycle of overnight sleep, during the transition from stage 2 of non-REM sleep to SWS, after subjects have learned tasks of both declarative and procedural memory before sleep. Notably, theta stimulation disrupted the normal progression of SWS, SO (0.5-1 Hz) and SWA (1-4 Hz) EEG power in the course of the stimulation. Also, it significantly decreased slow spindle activity (8-12 Hz) only frontally. These EEG changes were associated with a strong impairment of only declarative memory consolidation at recall after sleep (Marshall et al., 2011). Collectively, these findings strongly indicate that SO, SWA and frontal slow spindle activity reflect specific EEG mechanisms of hippocampus-dependent memory consolidation, which do not act beyond non-REM sleep/SWS. Moreover, they provide one of the strongest evidence for the assumption that during active wake, encoding of memory is reflected by theta EEG oscillations, which possibly reflect a transfer of information from the cortex to the hippocampus (Sederberg et al., 2003). Essentially, they strongly support the notion that memories encoded during wake undergo off-line consolidation during non-REM sleep/SWS by mechanisms involving an interplay between the hippocampus and the cortex, as reflected by the SO and frontal slow alpha activity (Buzsáki, 2006; Mölle et al., 2002; 2004; Marshall et al., 2006). Another recent investigation clearly showed strong and positive correlations between SO and sleep spindle EEG activities during the earliest portion of overnight non-REM sleep/SWS and rates of off-line improvement of both declarative and non-declarative memories (Wilhelm et al., 2011). These sleep EEG rhythms correlated positively with gain of improvement of only memories that were expected to be of further relevance, and not with memories that were not, with the latter memories being not found affected by sleep. Notably, this study demonstrates that SO and sleep spindles during non-REM sleep/SWS, are important EEG patterns not only for memory consolidation but also for memory reprocessing or reconsolidation (Wilhelm et al., 2011).

Some of the observations in the studies of Marshall et al. (2006; 2011) and Wilhelm et al. (2011) deserve further mentioning. These studies clearly show associations of non-REM sleep/SWS EEG signatures with either memory consolidation or reconsolidation in the first non-REM sleep cycle (Marshall et al., 2006; 2011) and even before the occurrence of the first SWS period (Wilhelm et al., 2011). A more recent study demonstrates that during the first half of overnight sleep, SO and SWA represent a global electric brain event that can be reliably recorded from hippocampus, medial temporal lobe and neocortex, whereas during the second half, both SO and SWA are expressed rather like a local phenomenon without strong phase and time coupling (Nir et al., 2011). Thus, regarding the studies using oscillating tDCS during sleep (Marshall et al., 2006; 2011) together with these of Wilhelm et al. (2011) and Nir et al. (2011), it can be concluded that SO, frontal slow alpha activity and sleep spindles, all represent important sleep EEG signatures reflecting processes of hippocampus-dependent memory consolidation and reconsolidation, which processes take place in the earliest part of non-REM sleep.

Similarly, classical sleep spindles, the major EEG signature of stage 2 of non-REM sleep, have been proposed to reflect mechanisms of brain plasticity that take place during non-REM sleep (Sejnowski & Destexhe, 2000; Steriade, 2001). Many human studies have demonstrated that either number, density, or EEG power spectral activity of spindles during stage 2 of non-REM sleep have been significantly involved in both declarative (Clemens et al., 2005; Gais et al., 2002) and procedural (Clemens et al., 2006; Fogel & Smith, 2006; Fogel et al., 2007b; Nishida & Walker, 2007; Shabus et al., 2004; Tucker & Fishbein, 2009) memory consolidation. Moreover, Nishida & Walker attempted to distinguish between usedependent and experience depended sleep processes. They subtracted sleep spindle EEG activity measured at a "non-learning hemisphere (left)" from that measured at a "learning hemisphere (right)" and were able to demonstrate strong positive correlations with offline memory improvement that were not evident for either hemisphere alone (Nishida & Walker, 2007). Thus, classical sleep spindles appear to represent an EEG pattern reliably associated with memory processes.

#### **3.2 Overview of human neuroimaging data**

162 Neuroimaging – Cognitive and Clinical Neuroscience

sleep EEG rhythm, theta stimulation) was applied in the first non-REM sleep cycle of overnight sleep, during the transition from stage 2 of non-REM sleep to SWS, after subjects have learned tasks of both declarative and procedural memory before sleep. Notably, theta stimulation disrupted the normal progression of SWS, SO (0.5-1 Hz) and SWA (1-4 Hz) EEG power in the course of the stimulation. Also, it significantly decreased slow spindle activity (8-12 Hz) only frontally. These EEG changes were associated with a strong impairment of only declarative memory consolidation at recall after sleep (Marshall et al., 2011). Collectively, these findings strongly indicate that SO, SWA and frontal slow spindle activity reflect specific EEG mechanisms of hippocampus-dependent memory consolidation, which do not act beyond non-REM sleep/SWS. Moreover, they provide one of the strongest evidence for the assumption that during active wake, encoding of memory is reflected by theta EEG oscillations, which possibly reflect a transfer of information from the cortex to the hippocampus (Sederberg et al., 2003). Essentially, they strongly support the notion that memories encoded during wake undergo off-line consolidation during non-REM sleep/SWS by mechanisms involving an interplay between the hippocampus and the cortex, as reflected by the SO and frontal slow alpha activity (Buzsáki, 2006; Mölle et al., 2002; 2004; Marshall et al., 2006). Another recent investigation clearly showed strong and positive correlations between SO and sleep spindle EEG activities during the earliest portion of overnight non-REM sleep/SWS and rates of off-line improvement of both declarative and non-declarative memories (Wilhelm et al., 2011). These sleep EEG rhythms correlated positively with gain of improvement of only memories that were expected to be of further relevance, and not with memories that were not, with the latter memories being not found affected by sleep. Notably, this study demonstrates that SO and sleep spindles during non-REM sleep/SWS, are important EEG patterns not only for memory consolidation but also

for memory reprocessing or reconsolidation (Wilhelm et al., 2011).

place in the earliest part of non-REM sleep.

Some of the observations in the studies of Marshall et al. (2006; 2011) and Wilhelm et al. (2011) deserve further mentioning. These studies clearly show associations of non-REM sleep/SWS EEG signatures with either memory consolidation or reconsolidation in the first non-REM sleep cycle (Marshall et al., 2006; 2011) and even before the occurrence of the first SWS period (Wilhelm et al., 2011). A more recent study demonstrates that during the first half of overnight sleep, SO and SWA represent a global electric brain event that can be reliably recorded from hippocampus, medial temporal lobe and neocortex, whereas during the second half, both SO and SWA are expressed rather like a local phenomenon without strong phase and time coupling (Nir et al., 2011). Thus, regarding the studies using oscillating tDCS during sleep (Marshall et al., 2006; 2011) together with these of Wilhelm et al. (2011) and Nir et al. (2011), it can be concluded that SO, frontal slow alpha activity and sleep spindles, all represent important sleep EEG signatures reflecting processes of hippocampus-dependent memory consolidation and reconsolidation, which processes take

Similarly, classical sleep spindles, the major EEG signature of stage 2 of non-REM sleep, have been proposed to reflect mechanisms of brain plasticity that take place during non-REM sleep (Sejnowski & Destexhe, 2000; Steriade, 2001). Many human studies have demonstrated that either number, density, or EEG power spectral activity of spindles during stage 2 of non-REM sleep have been significantly involved in both declarative (Clemens et al., 2005; Gais et al., 2002) and procedural (Clemens et al., 2006; Fogel & Smith, 2006; Fogel et al., 2007b; Nishida & Walker, 2007; Shabus et al., 2004; Tucker & Fishbein, 2009) memory consolidation. Moreover, Nishida & Walker attempted to distinguish between useNotably, the first human neuroimaging study almost entirely confirm previous rodent findings, which have shown that sleep-dependent memory consolidation relies on reactivation of the so-called hippocampus place cells during SWS in response to spatial or maze navigating tasks (Wilson & McNaughton, 1994). Thus, Peigneux et al. (2004) using a combination of PET and sleep EEG (PSG) methods, showed for the first time such a reactivation of hippocampal and neocortical regions in humans. At encoding during wakefulness, subjects were trained (or not) to learn and find their way inside a complex three-dimensional virtual town. Compared with the non-trained group, subjects who learned the task at encoding during wake displayed an increase in their regional cerebral blood flow (rCBF) in a bilateral pattern of neural activation, including the right and left hippocampus, the right and left parahippocampal gyri, superior parietal lobules, right and left precuneus, lingual and posterior cingulate gyri, middle and superior occipital cingulate gyri, and the anterior lobes of cerebellum. In the trained group, rCBF was consequently measured during stage 2 of non-REM sleep, SWS and REM sleep. The neural pattern of activity found at training sessions during wakefulness was reactivated only during SWS. This pattern of reactivation during SWS strongly correlated with the level of improvement on the task at recall after sleep (Peigneux et al., 2004). Three years later, Rasch et al. (2007) also used a combination of fMRI and PSG methods to investigate specifically the role for SWS in declarative memory consolidation. The authors first established a robust association between declarative learning (card place location) stimuli and a smell, olfactory stimulus (the smell of a rose). Subjects learned object locations in a two-dimensional (2D) object location memory task in the evening before sleep. During the first two periods of subsequent SWS, the odor cue was presented again (in an alternating 30 s on/30 s off mode). In a control condition, odorless vehicle was delivered. At retrieval testing after sleep, memory of card locations was distinctly enhanced when the odor was presented during SWS as compared to presentation of vehicle alone. Re-exposure to odor during SWS improved retention of memory in a hippocampus-dependent manner: bilateral reactivation of the hippocampus and medial prefrontal lobe, as measured by BOLD signal (Rasch et al., 2007). Further, an fMRI study demonstrated that compared with total sleep deprivation, post-learning sleep enhances hippocampal responses during recall of word pairs (declarative memory) 48 h after learning, thus indicating intra-hippocampal memory processing during sleep (Gais et al., 2007).. At the same time, it was shown that sleep induced a memory-related functional connectivity between the hippocampus and the medial prefrontal cortex. Six months after learning, recalling the same declarative memories reactivated the medial prefrontal cortex even more strongly than they did during encoding before sleep, thus showing that sleep leads to long lasting changes in representation of memories at a system (hippocampal-cortical) level. Although this study does not show

The Memory, Cognitive and Psychological Functions of Sleep:

plasticity mechanisms at synaptic and genetic levels.

**4. Sleep and cognitive functions** 

presented below in the respective section.

**4.1 Overview of human EEG data** 

(Tucker & Fishbein, 2009).

Update from Electroencephalographic and Neuroimaging Studies 165

subjects. The increase in functional connectivity during post-training REM sleep additionally suggests that the reactivated brain areas participate in an optimization of a

Altogether, the above neuroimaging findings show that the mechanisms of memory processing during early night non-REM sleep/SWS are distinctly different from those that take place in REM sleep. The former include complex interactions between hippocampus and cortex, thus reflecting brain plasticity mechanisms at system and neural levels. The latter include local patterns of brain reactivation, which are proposed to reflect brain

The strongest evidence for the important role that sleep plays in a variety of cognitive functions comes from many observations of effects of sleep loss and sleep deprivation on cognition (Killgore, 2010; Walker, 2008). The exclusively important role of sleep for cognitive functions has been best demonstrated in a study showing that even one night of total sleep deprivation results in inability to learn facts, i.e. deficient encoding of episodic memories (Yoo et al., 2007b). The neuroimaging correlates of this impaired learning ability will be

Sleep spindles, the major hallmark of stage 2 of non-REM sleep, have been for long proposed to be associated with human individual cognitive abilities or intelligence. For example, Bódizs et al. (2005) found that both grouping of fast sleep spindles by cortical slow oscillation over the left frontopolar derivation (Fp1) and fast sleep spindle density over the right frontal area (Fp2, F4) during stage 2 of non-REM sleep, correlated positively with general mental ability. Further, a robust positive correlations were found between slow (< 13 Hz) and fast (> 13 Hz) spindle activity in stage 2 of non-REM sleep and both individual cognitive abilities and implicit/explicit memory-related abilities (Schabus et al., 2006). Later, Fogel et al. (2007a) showed first that number of spindles in stage 2 of non-REM sleep remains relatively stable within individuals from night to night. Second, the authors demonstrated that the number of spindles and EEG power of sigma (slow spindle) activity were positively correlated with performance intelligence quotient (PIQ), but not with verbal IQ. Also, perceptual/analytical skills measured by the PIQ accounted for most of the interindividual differences in spindles. Interestingly, in the same study, a relationship between rapid eye movements in REM sleep and VIQ in individuals with higher IQ scores was also demonstrated (Fogel et al., 2007a). However, Tucker & Fishbein (2009) demonstrated that while subject's intelligence correlated positively with pre-sleep acquisition and post-sleep retest performance on both procedural and declarative tasks, it did not correlate with over-stage 2 of non-REM sleep spindle events. These findings suggest that intelligence may not be a powerful modulator of sleep's effect on memory performance

One major question arising from the above described findings is whether sleep contributes to human intelligence in a state dependent manner (i.e., by providing neurobiological conditions for memory consolidation and reconsolidation), whether sleep is associated with intelligence in a trait dependent manner (i.e., strictly individual sleep characteristics are

network that subtends subject's visuo-motor response (Laureys et al., 2001).

reactivation of the hippocampus and medial prefrontal cortex during sleep, basing on previous findings, the authors assume an important role of hippocampal-cortical connectivity for sleep-dependent declarative memory consolidation (Gais et al., 2007). The last convincing evidence concerning the role for SWS in memory formation comes from a recent fMRI study where the effect of declarative memory reconsolidation on declarative memory consolidation during SWS and wake was investigated (Diekelmenn et al., 2011). By using an experimental protocol similar to that previously applied by Rasch et al. (2007), the authors aimed at reactivating memories in humans by presenting associated odor cues either during SWS or during wake. During wake, reactivation of memories was followed by an interference task to probe memory stability, and as expected (Brown & Robertson, 2007; Robertson, 2009), this reactivation resulted in destabilized memories. In contrast, reactivation during SWS immediately stabilized memories, thereby directly increasing their resistance to interference. Importantly, BOLD signal revealed quite different patterns of reactivation during SWS and wake. The reactivation during SWS was mainly seen in the hippocampus and posterior cortical regions, whereas reactivation during wake was primarily found in the prefrontal cortical areas, thus showing that reactivation of memory serves distinct functions depending on brain state: wake versus SWS (Diekelmann et al., 2011). It is to be noted that similar differences between effects of oscillating trans-cranial direct current stimulation on both EEG activity and memory processes was also shown during wake when compared to SWS, although these differences distinguished specifically verbal memory encoding from declarative memory consolidation (Kirov et al., 2009; Marshall et al., 2006; 2011).

To our knowledge, only one neuroimaging study was able to provide convincing evidence for reactivation of brain regions during REM sleep in association with sleep-dependent consolidation of procedural memory (Maquet et al., 2000). In this study, three groups of subjects learned a serial reaction time task (SRTT) during wakefulness. To verify which brain regions are activated by SRTT, one group was scanned by using of PET either during training or during the subsequent rest wake period, and was not further examined. Subjects from a second group were trained on the task during two sessions in the afternoon, and then scanned during the night after training, both during wake and during various sleep stages. This group was retested at recall after sleep to verify that sleep-dependent learning had occurred. A third group was scanned at the same time points as the second group was, however, under absence of learning (no learning condition). Sleep architecture in the latter two groups was assessed by a routine PSG, and rCBF was measured across both three groups and all time points. Interestingly, although the third group improved implicit learning component of SRTT at recall after sleep, no any changes in sleep architecture between the learning and the non-learning conditions were found. However, as measured by the rCBF, a specific pattern of brain activation found in the first group during and after learning was reactivated only in the second group, and only during REM sleep. This pattern engaged a set of brain regions located in occipital and premotor cortices (Maquet et al., 2000). A later PET study by the same group (Laureys et al., 2001) confirmed that specific brain reactivation occurs during REM sleep in relation to procedural memory consolidation (Maquet et al., 2000). The authors showed that the left premotor cortex is functionally more correlated with the left posterior parietal cortex and bilateral pre-supplementary motor area during REM sleep in subjects previously trained to a reaction time task relative to untrained subjects. The increase in functional connectivity during post-training REM sleep additionally suggests that the reactivated brain areas participate in an optimization of a network that subtends subject's visuo-motor response (Laureys et al., 2001).

Altogether, the above neuroimaging findings show that the mechanisms of memory processing during early night non-REM sleep/SWS are distinctly different from those that take place in REM sleep. The former include complex interactions between hippocampus and cortex, thus reflecting brain plasticity mechanisms at system and neural levels. The latter include local patterns of brain reactivation, which are proposed to reflect brain plasticity mechanisms at synaptic and genetic levels.
