On the memory processing role of sleep electrophysiology and a new
hypothesis of sleep electrophysiology
2016/ 06/ 22
Boateng Asamoah
10833595
Supervisor
Ysbrand van der Werf, PhD
Co-assesor
Natalie Cappaert, PhD
Universiteit van Amsterdam
MSc in Brain and Cognitive Sciences
1
Table of contents
1
Abstract ... 3
2
Introduction ... 3
3
Sleep stages... 4
4
Synaptic homeostasis theory ... 5
4.1 Growth of synaptic strength ... 5
4.2 Bringing down synaptic weight ... 6
4.3 Rationale for the brain ... 6
4.3.1 Energy usage considerations ... 6
4.3.2 Space considerations ... 7
4.4 Mechanisms of Downscaling ... 7
4.5 Arguing for downscaling instead of potentiation ... 8
4.6 An issue with the general nature of the downscaling ... 8
5
The role of REMS in experience processing ... 9
5.1 REMS preferentially processes emotional memory ... 9
5.2 Sleep to forget, sleep to remember (SFSR) ... 10
5.2.1 Mechanism and prediction ... 11
6
Are REM and NREM functionally linked? ...12
6.1 Considerations on linkage of REM and affect processing ... 12
6.1.1 No evidence for REM related loss of affective tone ... 13
6.1.2 REM and non-emotional processing ... 13
6.2 SWS related fear processing ... 14
6.3 Interplay of sleep stages ... 15
7
A tandem hypothesis of sleep electrophysiology ...15
7.1 A view on experience encoding ... 16
7.2 Tandem processing ... 17
7.3 Cyclic nature of sleep electrophysiology ... 18
7.3.1 Reasons for the brain to make various copies ... 19
7.4 Considerations on REMS’s role according to the tandem hypothesis ... 20
7.4.1 Is sleep necessary for REM’s role according to the tandem hypothesis? ... 20
7.4.2 REMS and memory consolidation ... 20
7.5 Consequences of the tandem hypothesis ... 21
7.6 Tandem hypothesis in relation to the downscaling theory ... 22
2
7.6.1.1 A solution to the non-specificity problem ... 22
7.7 Tandem hypothesis in relation to SFSR ... 23
7.7.1 Why does REMS seem to preferentially process emotion? ... 24
7.8 Limitations of the tandem hypothesis ... 24
8
Conclusion ...25
9
List of abbreviations ...26
3
On the memory processing role of sleep electrophysiology and a new
hypothesis of sleep electrophysiology
Boateng Asamoah
1 Abstract
Sleep is predominant in the animal kingdom. However, since it entails disengagement from the environment it is perceived as risky behaviour. Sleep literature shows that during sleep there is processing of the waking period experiences; thereby providing a possible necessity for this risky behaviour. Certain aspects of sleep electrophysiology are linked to specific experience processing. The synaptic homeostasis hypothesis explains that, during slow wave sleep, the brain downscales synapses that were built during the waking period. By doing so it raises the signal to noise ratio of the experience and prepares itself for the following waking period. The sleep to forget, sleep to remember theory states that during REM sleep there is divorce between facts and affective tone of an experience. No theory has however linked slow wave sleep to REM sleep, explained why slow wave sleep should normally always precede REM sleep and given an explanation for the repetition of the sleep stages during one sleeping period. In this thesis a new hypothesis that assumes the validity of the downscaling theory will be presented. This tandem hypothesis states that during REM sleep a copy of the downscaled synapses is made for storage. During subsequent slow wave and REM sleep bouts further copies of the same experience are made with ever degrading resolutions and rising abstraction. This way the brain can build a copy of the experience that best fits the inner-model of the outside world.
2 Introduction
Sleep occurs in many species of the animal kingdom among which humans (Purves et al., 2008). It takes up large parts our lives (Carter, 2009). Its behavioural hallmark is the disengagement from the environment, which introduces a risk during this behaviour. The likelihood of becoming victim of a predator rises because there is no longer an active assessment of the environment. In evolution different strategies have been developed to tackle this issue. Whiles some animals (predators like cats) sleep relatively long periods of time, others (prey like the giraffe) do so in short intervals or hardly sleep (Purves et al., 2008). Furthermore, animals that sleep in relatively sheltered places (like bats) sleep longer than animals (like sheep) that sleep in the open (Allison & Cicchetti, 1976). The fact that in general preys sleep less robust furthers the idea that, from a survival perspective, sleep is a risky behaviour. Engaging in this behaviour despite its seemingly obvious risks points to a possible importance for the functioning of the organism.
4
Further demonstrative of the importance of sleep is the observation that sleep deprivation causes fatigue and eventually leads to diminished functioning (Purves et al., 2008). As an example, obstructive sleep apnea syndrome (OSAS) – which among others is typically characterized by dozens of waking periods during sleep and a reduced ability to descend to deeper stages of sleep – gives impairments in memory and executive functions in patients (Lal et al., 2012; Purves et al., 2008; A. Siegel & Sapru, 2011). Overall diminished attention in this patient group may further lead to injuries and even death due mostly to the operation of heavy machinery (Kono et al., 2007; Lal et al., 2012; A. Siegel & Sapru, 2011). Notably, in OSAS sleep electrophysiology there is low occurrence or absence of slow wave sleep (SWS). As sleep progresses towards the waking period, the amount of rapid eye movement (REM) epochs rises (more so than in non-patients) but with a shorter duration per epoch (Lal et al., 2012; Purves et al., 2008). The further associations of sleep with mood disorders and several syndromes (like autism) (Kanady et al., 2015; Robillard et al., 2014; Veatch et al., 2015) implies a relationship between affected sleep and an affected brain.
Sleep shows various electrophysiological characteristics that are divided into stages and are repetitive in nature. These are thought, among other functions, to process memory (Walker & Stickgold, 2006). Different theories have been proposed concerning the relevance of the sleep stages for functioning. They are based on many studies showing a correlation between sleep and memory consolidation. These theories typically focus on one major electrophysiological sleep stage. They do not consider SWS and REM sleep (REMS) together and do not explain why the sleep stages should be repeated during one sleeping period. In this thesis some theories, concerning the relevance of the sleep stages for memory processing, will be reviewed. More specifically the downscaling theory will be clarified; also the sleep to forget sleep to remember (SFSR) theory and its attempt to explain certain sleep dependent memory processing will be reviewed. After this a new hypothesis will be presented. This hypothesis links REM and non-REM (NREM) sleep and explains the necessity for the repetition of sleep stages.
3 Sleep stages
The electrophysiological sleep stages are referred to as I, II, III, IV and REM (Siegel & Sapru, 2011; Squire et al., 2008). The first 4 stages are also grouped under the name NREM sleep (NREMS) and Stages III and IV are typically referred to as SWS (A. Siegel & Sapru, 2011; Walker & van der Helm, 2009). During stage I the EEG frequency goes down and the signal amplitude rises in comparison to the waking state. Descent into stage II is marked by a further decrease and increase in frequency and amplitude respectively. During stage II spindles (bursts of activity) appear. They are of higher frequency than the rest of the signal in this stage. Stage III and IV follow stage II with an even lower frequency, higher amplitude and decrease in spindle amount (Purves et al., 2008). This stage is followed by REMS, which
5
is characterized by a low amplitude, high frequency activity that is not very different from the awake signal. During this stage various physiological properties of the body become more active, but there is a paralysis of the body. During one sleeping period sleep stages are repeated and the amount of SWS typically decreases while the amount of REMS increases. Furthermore, the sleep stages are recurrent in the sense that they are repeated during one sleeping period (Purves et al., 2008; L. R. Squire et al., 2008).
4 Synaptic homeostasis theory
One important hypothesis is the synaptic homeostasis hypothesis (also referred to as the downscaling theory) proposed in (Tononi & Cirelli, 2006). This hypothesis concentrates on the slow wave stage of sleep electrophysiology. In short it states that when SWS is reached, slow waves prune all synaptic connections that were established during wakefulness. All of these connections are pruned in an amount that is relative to the strength of the synaptic connection. Therefore, in principle their relative strength stays the same. However the weakest connections eventually disappear because they do not reach a threshold. The brain supposedly does this in order to return to a baseline. This way it heightens important synaptic connections in relation to the less important connections thus increasing the signal to noise ratio. Without this return to baseline, energy usage of the brain would rise every day and the skull would not be able to keep up with the growth of the brain. This daily reset to baseline aids in the brain’s ability to stay plastic and therefore process more information.
4.1 Growth of synaptic strength
The theory claims that during wakefulness the neuro-modulatory milieu favours storage during experiences. During this engagement with the environment (which is not necessarily linked to any specific learning paradigm), strong presynaptic firing is accompanied by a postsynaptic reaction. Also during wakefulness in the adult, changes in neuronal behaviour will more readily lead to long term potentiation (LTP) than long term depression (LTD). As a result of this the total synaptic strength increases. For example, whisker stimulation for a period of 24 hours leads to a growth of the synaptic density by 36% (Knott et al., 2002). This example strengthens the argument that during wakefulness total synaptic weight rises. The strength of the individual synapses, that are either built or strengthened, are unequal. This is due to the different levels of saliency during the experiences. Because of the growing size of the synapses the capacity of the brain to further increase the total synaptic weight declines. This means that at the end of a waking period, in comparison with its beginning, space is scarcer and energy usage is higher; leading to a lowered ability to lay down new information.
6
4.2 Bringing down synaptic weight
During the sleeping period we disengage from the environment. According to the downscaling theory neuro-modulatory processes drive the brain towards slow oscillations. De- and hyper-polarizations dictate these oscillations across all neurons. The neuro-modulatory molecules also prevent post synaptic reaction following presynaptic activity. This is a self-evident adaptation because there is a disengagement with the environment and neuronal activity is not driven by sensory input. Due to the high synaptic strength at the end of the waking period, neurons are highly synchronized. This gives rise to the high amplitudes of the EEG signal. Therefore, a greater amount of synaptic potentiation during the waking period inevitably leads to a greater occurrence of slow wave activity (SWA).
The brain goes through the stages of sleep and eventually reaches SWS. The constant cycle of depolarization and hyperpolarization leads to an equal percentage downscaling of all existing synapses in the neuronal ensembles involved in the slow oscillation. The synapses are to exceed a certain strength threshold. If this threshold is not reached, the synapse disappears altogether. At the end of the sleeping period all synapses have been maximally brought down in strength and the synapses that did not meet the threshold criterion have disappeared. Since the high connectivity at the end of the waking period was the cause of the synchronized firing, as sleep progresses the synchronization goes down and the higher amplitudes become scarcer. At the end of the sleeping period the total synaptic weight has been brought down to equal the weight of the beginning of the previous waking period. This is somewhat supported by (Knott et al., 2002); showing that days after stimulation, a stimulation dependent rise in synaptic weight had fallen back to pre-stimulation levels. The relative strengths, however, of the remaining synapses have changed and no longer equals relative strengths at the beginning of the previous waking period. Rather it reflects more the relative strengths of the synapses at the end of the previous waking period. This way a trace of the waking period experience is kept intact.
4.3 Rationale for the brain
Building up, strengthening and at least for a relatively short period of time maintaining the synapses costs energy and resources. After this effort, this theory claims that these synapses are partly broken down during the sleeping period. Apart from explaining that this is conducive to the function of the brain, the theory also explains this paradox by stating that there are energy and space benefits associated with the downscaling.
4.3.1 Energy usage considerations
A larger synaptic weight leads to higher energy usage. Synaptic weight correlates with firing rate (Buzsáki et al., 2002) so that an increasing synaptic weight would correlate with increased neuronal
7
firing. Based on a revision of the energy efficiency of neurons by (B. C. Carter & Bean, 2009; Sengupta et al., 2010), (Howarth et al., 2012) shows that action potentials (AP’s) account for 21% of the neuronal energy usage. Synaptic processes (postsynaptic receptors, neurotransmitter recycling, vesicle cycling and presynaptic Ca2+ entry) account for 59% of energy usage. This shows that the relative usage of
energy for AP’s and synapse related processes account for the largest energy usage by neurons. If the overall synaptic weight grows during the waking period, energy usage would rise accordingly. This is further supported by the fact that during a period of wakefulness the global arterial carbon dioxide tension (pCO2) rises as much as 17.5% (Braun et al., 1997). pCO2 may be used as a measure of energy
usage by biological tissues and its changes are directly linked to cerebral blood flow and volume (Reiman et al., 1986). The rise of pCO2 is thus an indication of increased neuronal energy usage.
Therefore, as the synaptic weight increases and firing rates and pCO2 increase accordingly, the energy
usage of the brain should also rise. As a result the brain’s energy efficiency would cumulatively go down if the system did not return to baseline.
4.3.2 Space considerations
The growing synaptic weight also means that brain tissue is built up. The brain, however, exists in a confined space. In a recent review it is argued that mechanical factors, among which pressure, play a role in both the form and function of the brain (Goriely et al., 2015). Furthermore, due to herniation risk and the extreme sensitivity of the brain to pressure a heightened intracranial pressure is considered dangerous. For this reason clinical cases of increased intracranial pressure typically require medical intervention (Goriely et al., 2015). Since there is little space inside the skull for safe expansion, it could be argued that synaptic downscaling is more favourable than cumulative growth. Whether the expected pressure increase will actually take place is still not clear. Computational (in the form of further refined finite element numerical head models) and specific experimental studies could help resolve this.
4.4 Mechanisms of Downscaling
The synaptic homeostasis hypothesis claims that wakefulness associated changes lead to plasticity via the mechanisms of LTP. In support of this (Braun et al., 1997; Knott et al., 2002) show that synaptic density rises upon continuous stimulation of whiskers. Wakefulness leads to activation and upregulation of LTP associated genes and molecules, while they are downregulated during sleep. The induction of these genes (among which are CREB, Arc and BDNF) is dependent on the activity of the noradrenergic system (Cirelli & Tononi, 2000). During periods of wakefulness, neuronal activity is related to external events. Systems, like the noradrenergic, in the brain are active during this external engagement. This system adjusts its firing rates according to the level of environmental engagement (Aston-Jones & Bloom, 1981; Zitnik, 2015). During experiences, especially those with high saliency,
8
noradrenaline (NA) plays a role in the induction of mechanisms which enhance learning (Gary Aston-Jones & Cohen, 2005). Furthermore, synaptic potentiation arises trough the mechanism of LTP and the noradrenergic system plays a role in both protein synthesis dependent and independent LTP (Gelinas, 2005).
On the other hand, low levels of NA, associated with REMS (Mehta et al., 2016), aids in preventing a postsynaptic reaction upon presynaptic input (Aston-Jones & Bloom, 1981). This way they limit synaptic potentiation. Additionally, lesions in the noradrenergic system give rise to slow waves in recorded EEG signals (Lidbrink, 1974). These changes disappear after a period of time which most likely has to do with compensatory mechanisms organised by the brain after lesions (Berridge, 2008). Before this compensation, however, LTP related factors go down in level when noradrenergic connections to the cerebral cortex are destroyed. They permanently go down to levels comparable with sleep levels, even during wakefulness. The reduced ability to induce LTP after this lesioning thus reflects noradrenergic involvement in laying down waking period experiences.
4.5 Arguing for downscaling instead of potentiation
Several studies support the notion of downscaling as the mechanism deployed by the brain during sleep to process the waking period experiences. (Yang & Gan, 2012) showed that during the sleeping period there is a net loss of synaptic connections while there is a relative net gain during the waking period. (Hill et al., 2008; Nere et al., 2013) deployed computer simulations. They showed that for various forms of sleep dependent memory formation, down selection of synapses during sleep will lead to enhanced performance. This however is less true for synaptic potentiation. Based on, among others, these studies (Cirelli & Tononi, 2015) argue that potentiation during sleep is not necessary for experience processing. In disagreement with (Heller, 2014) they argue that (Aton et al., 2014) does not show potentiation of synapses. Rather the published data in (Aton et al., 2014) only shows enhancement. Indeed, it would seem that the studies used by (Heller, 2014) to argue for potentiation mostly show enhancement of performance rather than potentiation of connections.
Together with the benefits associated with downscaling in comparison with synaptic potentiation, interpretation of various studies (e.g. abovementioned) support synaptic downscaling rather than synaptic potentiation as the sleep dependent experience processing mechanism. Notably, (Buzsáki & Mizuseki, 2014) reason, based on the idea of a pre-configured brain, that replay in itself cannot be an argument learning or synaptic plasticity.
4.6 An issue with the general nature of the downscaling
The downscaling hypothesis argues that all synapses are downscaled during SWS. As the organism engages in the cycle of wakefulness and sleep, new experiences are, respectively, encoded and
9
subsequently made efficient by raising their signal to noise ratio. This also means that the connections that represent existing knowledge are scaled down during the recurrent general downscaling (Giuditta, 2014). This problem arises because the downscaling theory does not describe the downscaling of specific connections (Giuditta, 2014). This would lead to a loss of older memories. However this doesn’t happen. This means that the downscaling theory will have to explain how downscaling can be non-specific and still not cause loss of older experiences.
5 The role of REMS in experience processing
SWS reportedly plays an important role in memory enhancement as discussed earlier. Interestingly many studies do not implicate this sleep stage in emotional memory processing. Sleep, however, has a regulatory effect on emotional brain function as well (Goldstein & Walker, 2014). According to some studies, brain areas implicated in emotion processing, such as the amygdala, the hippocampus, the insula and the medial prefrontal cortex, show activity increases during REMS (Dang-Vu et al., 2010; Miyauchi et al., 2009). Apart from its electrophysiology being different from NREMS electrophysiology, there are neurochemical changes that take place. Most notably, NA presence drops to levels lower than that of wakefulness or NREMS. REMS is furthermore specifically linked to emotional memory as will be discussed in the following session.
5.1 REMS preferentially processes emotional memory
In a human study (Payne et al., 2008) showed an improvement of the recognition of emotionally charged objects and scenes. They furthermore showed that memory of the background of emotional scenes had declined in relation to the background of neutral scenes. This sleep dependent processing correlated only with time spent in REMS. In a different study, similar results were obtained using a somewhat different design (Payne et al., 2012). (Ravassard et al., 2015) showed that REMS deprivation in rats leads to, among other effects, an impairment in the consolidation of contextual fear conditioning.
(Wagner et al., 2001) tested emotional memory recall by letting subjects read and recall emotional and non-emotional texts. To this end, two sleep and one awake control conditions were deployed. Subjects in condition 1 read the texts at the beginning of the night and were awoken mid-sleep for testing. Subjects in condition 2 slept at the beginning of the night but were awoken mid-sleep to read the texts and were tested at the end of sleep. Subjects in conditions 3 and 4 read the text at the same moment as in condition 1 or 2 respectively but were kept awake. For condition 2 subjects retention of emotional text proved superior to retention of non-emotional texts. This group also performed better than condition 1 subjects for retention of emotional texts. Furthermore, condition 2 subjects performed better than awake subjects for emotional texts. No benefit of sleep on neutral text was reported. These
10
results hint to the benefits of REMS for emotional memory retention. More recent studies have also pointed to REMS as implicated in emotional experience processing. Using images instead of text but in a similar sleep design as (Wagner et al., 2001), (Groch et al., 2013) showed a higher emotional memory recognition after REM-rich sleep compared to SWS-rich sleep. REMS therefore seems to facilitate the consolidation of emotional memory, more so than SWS.
In (van der Helm et al., 2011) subjects rated twice, with a time interval of 12 hours, the emotional intensity of 150 standardized affective pictures on a scale of 1 – 5. Subjects that slept between the two tests showed a decrease in amygdala reactivity in comparison with subjects that stayed awake between the two tests. A circadian control test confirmed results were not due to time of day. Their data further showed an increase in ventromedial prefrontal cortex (vmPFC) – thought to play a role in fear extinction and by doing so to have a regulatory effect on the amygdala (Milad & Quirk, 2002; Morgan et al., 1993) – functional connectivity with the amygdala in the sleep group while the awake group had a decrease in this connectivity. On the behavioural level, the decrease of amygdala reactivity coincided with a decrease in the ratings of the presented pictures. In the awake group however there was an increase. Additionally, the sleep group showed a significant decrease in the most intense emotional ratings and an increase in neutral ratings. This change pattern did not occur in the awake group. These results thus point to emotional processing by REM activity.
(Walker & van der Helm, 2009) note that as insomnia leads to less sleep, it may also cause an imbalance in memory encoding. When insomnia is linked to depression there may be an impaired ability to form and keep memories with a positive affect. At the same time this insomnia will preserve the long-term formation of negative emotional memory. This would mean that, despite the occurrence of both positive and negative experiences, long term emotional memory would mostly be negative in nature. This would explain the higher incidence of depression among groups suffering insomnia. Studies (Walker & van der Helm, 2009) refers to suggest that when patients suffer depression they go into REMS faster and stay in the REMS stage longer. Deprivation of late night sleep (rich in REMS) leads to diminished depressive symptoms in groups that react positively to total sleep deprivation. After recovery sleep, depressive symptoms reappear quickly. This implies that the occurrence of REMS has effects on the way emotional experiences are processed.
5.2 Sleep to forget, sleep to remember (SFSR)
Based on these effects of REMS (Walker & van der Helm, 2009) suggest a hypothesis in which REMS aids in the proper storage of emotional memories and their subsequent recall. They argue that affective experiences are remembered more robustly than neutral experiences. They then state that this has to do with the specific neurochemical processes associated with these experiences. When
11
these experiences are recalled however, the elicited autonomic activation is far less in comparison to the moment of the actual experience. They argue that, as time passes by, the affective tone that initially accompanied learning during the experience disappears. The information associated with the experience remains and leads to the robust recall. The process of peeling away the affective tone takes place during multiple REMS stages. This does not necessarily have to happen within one sleeping period. Rather subsequent sleeping periods further aid in depotentiating the affective tone of the experience.
This way, a REMS governed, ‘affective therapy’ is attained. The specific electrophysiological characteristics of REMS point to a reactivation of experiences of the waking period. These REMS reactivations occur under an altered neurochemical environment. Notably the noradrenergic input, which is associated with stressful events (Gary Aston-Jones & Cohen, 2005), is brought down to lower levels. The factors playing a role in REMS (electrophysiological and neurochemical) work together to create an environment in which the memory of the informative aspects of an emotional experience is strengthened and stored. At the same time it causes reduced strength of the affective tone that was initially acquired.
5.2.1 Mechanism and prediction
The model’s mechanism is as follows. During the waking period, when there are experiences with varying levels of saliency, the formation of cortical memory which is bound to the hippocampus is facilitated by the amygdala. The neurochemical environment (e.g. relatively high levels of aminergic neurotransmitters) of the brain aids in the formation of the connections that represent these memories. Upon entry of REMS, the connections are replayed. This happens in a changed neurochemical environment. The low levels of aminergic concentration help achieve a depotentiation of the affective tone. At the same time it helps the progression and strengthening of the factual information of the experience in the cortex. The progression of information into the cortex supports the integration of the new information with the existing knowledge base. This way a comprehensive database of integrative information is built, which presumably reflects the process resulting in dreaming. When this process of losing the affective tone fails there are consequences.
These consequences take the form of a persistent affective tone to an experience. This way, co-arising with recall of the experience is the autonomic arousal that coincided with the experience in the first place. This is exactly what is seen in patients suffering from post-traumatic stress syndrome (PTSS). Upon the encounter of a cue associated with the traumatic event, there is hyperarousal in these patients. According to (Goldstein & Walker, 2014) levels of NA are brought down during REMS (this has an effect on the amygdala as well as the PFC). This leads to a low tonic activity in the locus coeruleus
12
(LC) at the beginning of the waking period. The low tonic activity makes for a sharper distinction with the phasic activity that is associated with external emotional events. The lowered levels of NA engage alpha-2 receptors which induce top-down control of the amygdala. This way a better separation can be made between levels of saliency during the experience. In the sleep deprived on the other hand, there is no REMS associated down regulation of LC activity. Because of this the activity of LC stays higher leading to a diminished capacity to distinguish between an external emotional stimulus and the baseline that is now heightened. High levels of NA bind to alpha-1 receptors which then bring down the regulation of PFC on the amygdala. The amygdala now has a higher activity and the saliency detection coincidentally has gone down. A similar mechanism is at work in PTSS patients (Goldstein & Walker, 2014). Here however, the problem starts with the inability to adequately bring down noradrenergic levels during REM. This diminished ability to discern between threatening and non-threatening cues are expected under the hypothesis of (Walker & van der Helm, 2009).
According to (Walker & van der Helm, 2009) this dissociation process preferentially takes place overnight. This means that during the night we sleep to forget the affective tone of the experience. That same sleep further aids in the memory of informative aspects of the emotional experience. By stating that it happens overnight, they hint to a circadian effect of this process. However it is not made clear why the circadian rhythm or more specifically the time of day would play a role. It is not explained why sleep during the day wouldn’t lead to the same process of shedding affective tone and salvaging factual information.
6 Are REM and NREM functionally linked?
Based on presented studies, it would seem that there is a disassociation between the two major sleep stages, namely REM and NREM sleep; where SWS occupies itself with non-emotional memory consolidation. REMS on the other hand deals with emotional memories. It does so by enhancing the memory that gave rise to the arousal and makes sure that the environment of that stimulus becomes less memorable. This division may however not be as strict as it would seem. Despite the discussed body of work mainly linking REM to emotional processing, several studies have questioned this link and others have implicated REMS to non-emotional memory processing.
6.1 Considerations on linkage of REM and affect processing
Methods that were used in early research may have had confounds (Horne & McGrath, 1984). These possible confounds may partly explain the linkage in those studies (Smith & Rose, 1995). Some more recent studies have pointed out that the association between REMS and the processing of emotional memory may not be as strong as previously thought. According to (Kaida et al., 2015) various studies have shown opposite results to each other or were somewhat inconclusive themselves.
13
In the past the flowerpot method (Smith & Rose, 1995) was deployed to induce REMS deprivation. It relies on the fact that upon entry of REMS there is a relaxation of the body. This way the animals (that had been standing on a small platform located in a pool of water) topple into water and have to climb back on the flowerpot. This means that during the experiments there was no control as to the sleep stages the animals entered and how they may have interacted. According to (Horne & McGrath, 1984) the flowerpot method is stressful. Attempts to tackle this issue by enlarging the platform or having control animals that do not fall may not be adequate. Notably, behaviour that is a result of REMS deprivation resembles behaviour induced by stress alone. For this particular reason, (Horne & McGrath, 1984) argue that the described behaviour ascribed to REMS deprivation may be a result of this confound which may have skewed early research on REMS deprivation and its role in certain forms of memory processing somewhat.
Additionally, according to (Walker & Stickgold, 2006) the notion that REMS has no role to play in non-emotional memory may have been based on unfortunate overgeneralizations. Rather they argue that the sleep stages should be viewed as playing different roles in different forms of memory during various stages of memory processing.
6.1.1 No evidence for REM related loss of affective tone
(Groch et al., 2013) concluded that the affective tone of an emotional memory does not seem to be diminished, but rather kept intact during REMS. (Wiesner et al., 2015) deprived subjects from either REMS or SWS. They showed that sleep has beneficial effects for memory consolidation. More interestingly they showed that during recognition, REMS deprived subjects did not have a higher valence or arousal than the SWS deprived group (which did have REMS) or an awake group. Also the decrease of arousal over time was not dependent on whether subjects slept or not. In opposition to SFSR both (Groch et al., 2013) and (Wiesner et al., 2015) found no support for the notion that REMS leads to weakening of the affective tone of an emotional experience. Both groups further reported no memory benefits for neutral memories after sleep.
6.1.2 REM and non-emotional processing
Some studies have in fact shown that REMS is not just implicated in the processing of emotional memory. (Koninck et al., 1989) recorded from students learning a new language in an intensive manner. They showed a correlation between efficiency in acquiring a language and REMS. For proficient learners the percentage of REMS rose during the course and returned to baseline after the course. This was not the case for less proficient language learners. (Smith & Rose, 1995) showed transient impairment of spatial learning in rats. When the platform in the Morris water maze was hidden, REMS deprived rats were less proficient in finding the platform. However only the first REMS
14
deprivation session had an effect. These studies point to an involvement of REMS in non-emotional processing.
6.2 SWS related fear processing
The processing of memory by SWS also seems more complex than described earlier in this thesis. SWS has mostly been implicated in various forms of non-emotional memory. The downscaling theory only refers to the benefits of SWS for non-emotional memory. As will be reviewed in this section, some studies have associated SWS with emotional memories.
Because contextual extinction aids in extinction, (Hauner et al., 2013) hypothesized that the exposure to a fear conditioning context would promote extinction. They made use of the fact that memories created (in the presence of a context) during the waking period can be enhanced during SWS by replaying the same context. Odour was used because of its reliability and the fact that it may be deployed without obstruction of sleep. They combined high resolution functional magnetic resonance imaging (fMRI) with physiological recordings to assess subjects. Subjects were subjected to an olfactory contextual conditioning. Pictures of faces (CS) were paired to electric shocks (US). This was done whiles odorants were present. During the period in which subjects were in SWS, One of the context odours was reintroduced to them in absence of the US. This reintroduction supposedly reactivated the brain correlates of the US. The results suggest that skin conductance response (SCR) declined, in comparison with odorant off periods when subjects were presented the target odour in the course of SWS. When subjects woke from sleep and were retested, SCR had fallen from pre-sleep to post-sleep. This reduction was specific to and greatest for odorant targeted CS. Furthermore a multivariate fMRI pattern analysis showed a decorrelation of pattern activity in the left amygdala. This supports the idea that the context driven extinction changed the reaction of the amygdala so that the cue became less associated with fear leading to a lower skin conductance. This decorrelation was however not significant when the whole amygdala was assessed.
A different study also showed that re-exposure to a contextual cue during SWS leads to the extinction of fear memory (He et al., 2015). Here, subjects received a mild shock (US) that was paired with a tone (CS). The results show that reintroducing the CS during SWS caused the extinction of fear memory. A control audio stimulus ensured that the extinction was not due to the replay of a tone as such. This study furthermore confirmed that re-exposure to context leads to the extinction of fear that was coupled to that context. A half night paradigm was used to ensure slow wave rich sleep and avoid strong REMS involvement. Data of subjects that reported to have noticed the tone were excluded from analysis. It was not clear as to whether noticing the tone affected the sleeping profile of these subjects. The tone may have affected the power of the brain frequencies. Further concerning the spectral
15
analysis, a power analysis may have clarified as to whether the statistically insignificant increase in power over the 10-16 Hz had high statistical power or not. A possible effect here may need to be taken into consideration for interpretation of this and future studies. Nevertheless, this study indicates – in accordance with (Hauner et al., 2013) – that cue replay during SWS may lead to emotional processing.
6.3 Interplay of sleep stages
The discussed literature above implies that SWS is not only implicated in the consolidation of factual information and REMS not only engaged in the processing of emotional memory. Rather it suggests an interplay of both REM and SWS in the processing of memory in general. There may be a relationship between the two states of physiological sleep so that together they ensure processing of memory in a way that leads to consolidation and integration with existing memory and the internal model of the outside world. Notably, (Tononi & Cirelli, 2014) refers to REMS as an important open issue that needs to be addressed. Concerning this, it is of interest that the mammal foetus only has a REM like sleep electrophysiology (Giuditta, 2014; Squire et al., 2008). SWS only occurs after birth and then progresses to the adult like form over time. This implies that REMS is an innate physiological state governing the laying down of brain circuits (Roffwarg et al., 1966). Later in life SWS appears so as to deal with experiences. Another fact that pleads for both SWS and REMS as involved in the process of memory consolidation is the fact that waking subjects from either SWS or REMS may influence memory performance (Stones, 1977).
Further evidence for both SWS and REMS in memory processing comes from a counter-bias study (Hu et al., 2015). Subjects were shown images combined with words that are viewed as socially counter-stereotyped. An audio cue was paired with the images and words. Immediately after training subjects had a lower implicit bias score. The audio cue was unobtrusively replayed during the whole time spent in SWS. Data showed that replay of the audio cue during SWS led to a smaller bias score that was persistent one week later. When there was no replay, bias rose back to pre-training levels. Notably, despite the fact that replay only took place during SWS, the lower bias score only correlated with duration of both REM and SWS. Also (Aton et al., 2014) showed that the response enhancement after sleep correlated with both SWS and REMS. Hence, there is a body of work hinting to an interplay of both REM and SWS in sleep dependent experience processing.
7 A tandem hypothesis of sleep electrophysiology
The fact that both REM and SWS seem to be involved in the processing of memory of various nature calls for a theory that takes this fact into account. This theory should combine both SWS and REMS into one theory of sleep dependent experience processing and storage as is implied by studies reviewed under the section 6 (are Rem and NREM functionally linked?). Before the presentation of this
16
‘tandem like’ hypothesis a general view on the nature of experience encoding will be discussed shortly (box 1). This view will prove necessary to explain why REMS would seem to preferentially process emotional experiences (and divorce affective tone from factual information) whiles such specific processing may not exist as has been argued by (Groch et al., 2013; Walker & Stickgold, 2006; Wiesner et al., 2015). Furthermore, combined with the tandem hypothesis (which includes the downscaling theory) this view on experience encoding can explain the occurrence of PTSS. After presentation of the tandem hypothesis, its relationship to the downscaling theory and consequences for SFSR will be explored. The presentation of this tandem hypothesis shall end with a few predictions.
In short, the tandem hypothesis states that SWS aids to downscale synapses as argued in (Tononi & Cirelli, 2006). After this a copy of the waking period experiences (in their processed form) are built – during the REMS stage – and integrated in the internal model when possible. This is done for each sleep cycle of SWS and REMS. So that, for at least a short period of time, there exists more than one copy of the same experience in memory. The brain combines these copies and builds from them a low resolution high abstract version, of the experience, that contains the necessary information.
7.1 A view on experience encoding
Box 1A cartoon depicting a view on experience encoding by the brain during the waking period. This view states that for all experiences there is a hypothetical storage format containing at least components 𝑝 (facts of the experience) and 𝑞(emotional aspects of the experience). What differs between differently salienced experiences are the synaptic weights associated with the components. The view may best be described by an exponential function 𝑓(𝑥) = 𝑒𝑥∗𝑏+ 𝐶 (fig. a). For each component of the hypothetical storage format, 𝑏 (manipulating 𝑥 and therefore the growth of the function) is unique so that for all values of 𝑥, 𝑓(𝑥) is also unique. Furthermore C is unique so that for each component, 𝑓(𝑥) is unique where 𝑥 = 0. Let 𝐶𝑃> 𝐶𝑞 be true for a specific experience (𝐸). This view then states that 𝑏𝑞> 𝑏𝑝, so that there is a value
for 𝑥 where 𝑓𝑝(𝑥) = 𝑓𝑞(𝑥). It further states that until 𝑓𝑝(𝑥) = 𝑓𝑞(𝑥), 𝑓𝑝(𝑥) > 𝑓𝑞(𝑥), so that the facts are always encoded
stronger than the affective tone. For values of 𝑥where 𝑓𝑝(𝑥) < 𝑓𝑞(𝑥) this relationship is reversed. From this reversed
relationship a possible mechanism for PTSS may be deduced. Also the view can explain why exposure therapy should be effective.
For experiences of various nature the strength of the memory and therefore of the components is different. The specific neurochemical deployment of the brain causes experience to be encoded at a certain value for 𝑥 which then leads to a corresponding synaptic strength. The specific neurochemical deployment in turn is given rise to by the importance of the experience for survival.
17
7.2 Tandem processing
The hypothesis of sleep dependent memory processing combines both SWS and REMS. More specifically, in this tandem hypothesis, during SWS synapses are downscaled as described by the downscaling theory. During the subsequent period of REMS synapses are replayed. These are the same synapses that were built and/ or strengthened during the waking period and subsequently downscaled during SWS. This is also why the electrophysiology of REMS so much resembles that of the waking period (Buzsáki & Mizuseki, 2014). During this period of sleep electrophysiology a copy of the processed experience as such is made in long-term memory. The replay of connections does not lead to strengthening of the synapses. The altered neurochemical environment of the brain during this sleep stage prevents the replay to cause strengthening of synapses. The goal of the replay is the integration of the copy of the processed experience into existing memory and the internal model. This means that there is a stepwise process where REMS must follow SWS – which is the case in healthy adults (Carskadon & Dement, 2011) – to finish the cycle of sleep dependent memory processing. SWS and REMS therefore work together to process the waking day experience by respectively downscaling and then integrating a copy of the processed experience into older memory (fig 1-I).
18
Figure 1. A cartoon of the tandem processing. (I) depicts the first part of the hypothesis which states that after the downscaling a copy of the experience is made during REMS. The experience here is exemplified as colours that have neighbouring wavelengths on the electromagnetic (EM) spectrum. The following copy contains the same information but in an abstracter form (instead of the colour orange it now states the colour in name and EM informs that the information should be viewed in terms of the electromagnetic spectrum). Later copies (e.g 1.3) are even more abstract and may not contain some information that could be important. When this very abstract copy is integrated with an earlier copy the lost information may be retrieved.
During SWS there is downscaling of synapses as explained by the downscaling theory. During the following REM stage a copy (1.1) of the experience as has been processed by SWS is made. After this, SWS further downscales the synapses. During the following REMS a new copy (1.2) containing the same – further scaled down – experience is made. This copy (1.2) however has a lower resolution and higher abstraction than copy 1.1. If needed, information from copy 1.1 is added to copy 1.2 to create another copy (2.1). Copy 2.1 has a level of resolution and abstraction in between copies 1.1 and 1.2; however it is closer to copy 1.2. Copies 1.1 and 1.2 are deleted. During the following REMS (which follows yet another SWS downscaling) a copy (1.3) is made – with an even lower resolution and higher abstraction than all previous copies – and incorporated with copy 2.1 to create copy 3.1. During subsequent bouts of slow wave and REMS copies are created that contain ever decreasing resolutions and increasing abstractions. Towards the end of sleep there is one copy of the waking period experience that is abstract in nature and should best fit the internal model of the outside world.
7.3 Cyclic nature of sleep electrophysiology
Electrophysiological recordings show that during the sleeping period the various stages are repeated, resulting in a cyclic nature of sleep electrophysiology (Purves et al., 2008; A. Siegel & Sapru, 2011). To the best of my knowledge, there is no explanation (in sleep literature) as to why this repetition of sleep stages should occur. This is also the case for both the downscaling and SFSR theories. According to SFSR subsequent REM sleeps cause more loss of the affective tone. This way it has incorporated a necessity for the repetition of at least REMS. However since REMS mediated processing of an
19
emotional experience may also happen during different sleeping periods, there is no need to have multiple REM cycles per sleeping period. Furthermore, SFSR does not explain why one long REMS stage per sleeping period would not work just as well.
The tandem hypothesis explains this repetition of the sleep stages (during one sleeping period) as follows. During the waking day data is gathered among which data that do not serve any specific goal. Once the organism falls asleep and goes into SWS the mechanism of downscaling ensues. The downscaling theory explains that part of the data is lost during SWS. During the subsequent REM stage a copy of the experience is made in long-term memory. After every SWS and subsequent REMS a copy of the same waking period experience is made. The first copy has a high resolution and contains the largest amount of information about the experience. In contrast, the last copy has the lowest resolution of all the copies and also has the highest level of abstraction. This is somewhat supported by (Smith & Rose, 1995) where only deprivation of the first REM stage caused a reduced ability to find the platform in the Morris watermaze. This means that during one sleeping period the brain stores more than one copy of the waking period experiences (fig. 1). As the sleeping period progresses and more data are removed by SWS the necessity to encode the rest of the data becomes that much more important. This would also explain why early sleep is rich in SWS when there is more data, in absolute sense, to delete. During late sleep, when signal to noise ratios are heightened the remaining connections contain relatively less data so that there is less downscaling needed. The late sleep connections may be showing more the essence of the experience. There is therefore a greater necessity to consolidate the remaining data, thus making the role of REMS more important in late sleep.
This also means that for various reasons the first REMS stages should have a shorter duration than later REMS stages which is also the case (Kandel et al., 2000; Purves et al., 2008). Firstly, because the first REMS stage is only to make a copy of the experience its duration would be shorter than that of later REM stages when the copy is compared to an earlier copy. Also during the last REMS, there is a final integration of the experience with older memories and the internal model. Perhaps there would be more time necessary to complete this final integration. Under section 7.8 ‘limitations of the tandem hypothesis’ it will be argued that the REM duration image described here is not necessary as straightforward.
7.3.1 Reasons for the brain to make various copies
For every REM stage the brain makes a copy of the same experience. These copies can be compared and integrated to create one combination out of multiple copies. Possibly the brain does this in order to build from the copies one image of the experience that best fits the model of the outside world as
20
it exists in the brain. This is favourable because a copy that best fits this model makes for smoother integration with existing memories which in their turn have been moulded to fit the model. Under section 7.6.1.1 (a solution to the non-specificity problem) it will be argued that this best fit copy may aid in preventing the loss of older memories during the downscaling as has been noted by (Giuditta, 2014).
The brain also has storage space considerations to take in to account. It is therefore most likely that the brain would pick a low resolution, high abstract copy for storage in long-term memory. During the construction of this best fit copy, information from earlier copies that may be necessary to draw up this best fit copy is taken from earlier copies. Possibly this would be an ongoing process. So, perhaps, the brain doesn’t wait until the end of the sleeping period to construct this best fit copy. Rather it would integrate copies during sleep progress and already do away with data that seems less informative. The copies are therefore built in the first place to serve a short term auxiliary purpose.
7.4 Considerations on REMS’s role according to the tandem hypothesis
7.4.1 Is sleep necessary for REM’s role according to the tandem hypothesis?
According to (Tononi & Cirelli, 2006) there always has to be an explanation as to why a certain processing should preferentially take place during sleep. If indeed sleep is a risky behaviour as is argued in the introduction, then the healthy brain should avoid it whenever possible. So why should the consolidation take place during sleep? The specific neurochemical composition of the awake brain would interfere with consolidation. It is already known that during wakefulness there is some consolidation however it is less thorough than during sleep (Korte & Schmitz, 2016). Synaptic potentiation during wakefulness may be interfering with proper consolidation. From this follows that a brain that is engaged in vigorous encoding (e.g. during an acute stressful event) is less inductive to consolidation. This is somewhat supported by (Uwaya et al., 2016), showing that an acute stressful situation impedes the formation of fear memory. Therefore it would seem that disengagement from the environment is necessary for thorough consolidation.
7.4.2 REMS and memory consolidation
Memory benefits of sleep are mainly ascribed to SWS. However the idea that REM activity would be involved in memory consolidation is not a stretch. (Tononi & Cirelli, 2014) points to a possible memory enhancement goal for REMS by arguing that during REMS there is insertion of AMPAR in synapses that have been scaled down. (Ravassard et al., 2014, 2015) argue that during REMS factors typically implicated in LTP and longlasting LTP (L-LTP) are upregulated. Also sleep dependent decreased memory consolidation is associated with REMS (Gorgoni et al., 2013). These and other studies support the notion that REMS may be implicated in memory processing.
21
A role for REMS in memory processing has been questioned. These doubts are based on data suggesting that there can be long periods of REM deprivation without very obvious consequences (Siegel, 2001). This fact and the role of REMS according to the tandem hypothesis do not contradict each other. This is because there is still some integration outside of sleep (Korte & Schmitz, 2016). The downscaling will ensure that the brain is ready for a new waking period of experiencing. Therefore the negative effects associated with OSAS do not occur. However the consequence of long term REMS deprivation is that it will prove impossible to function at a relatively high level of cognitive and executive demand as has somewhat been shown by (Yang et al., 2008) concerning spatial learning in rats. This is because the waking period experience processing is not as thorough as the sleep dependent processing as argued in section 7.4.1 (is sleep necessary for REM’s role according to the tandem hypothesis?).
7.5 Consequences of the tandem hypothesis
This hypothesis claims that there is a logical flow from SWS to REMS and specifically in that order. This means that an altered SWS may affect REMS. An example of this can be seen in OSAS. Earlier it was mentioned that the sleep electrophysiology of OSAS differs from non-patients. The short lasting but more frequent entries into REMS may be explained by the tandem hypothesis as follows. When OSAS patients cannot enter SWS, the downscaling is kept at a minimum or does not happen at all. Because the tandem hypothesis claims that the processed memory is incorporated in earlier copies and stored during REMS; for the OSAS patient there would be a reduced necessity for REM dependent processing. This is what is seen in the sleep electrophysiology of OSAS patients. Entries of REMS are short lasting. This is because there is less new processed information to be integrated. The tandem hypothesis therefore predicts a functional linkage between SWS and REMS. This means that manipulating SWS to a certain level will lead to an altered electrophysiological profile of REMS. This altered electrophysiological profile is an adaptation to the manipulation of SWS.
Manipulating the profile of sleep electrophysiology will also always have effects for sleep dependent experience processing. When there is no or little SWS, REMS cannot get to a good high abstract copy that fits the internal model well. If a copy is made it will be a high resolution copy unfit for incorporating in existing memory. Alternatively if one was to enforce, experimentally, one large SWS period on the sleeping brain, this would lead to one high abstract copy that may not carry enough information so as to fit the model well and or be an informative memory of the experience. These would be rigorous ways to test the idea that REMS and SWS work together to cause sleep dependent memory consolidation. It needs to be stated that it is not clear as to how safe it would be (in terms of long term effects for the subject) to enforce SWS on the brain.
22
7.6 Tandem hypothesis in relation to the downscaling theory
The tandem hypothesis is related to the downscaling theory in that the former assumes the validity of the latter. Concerning the latter, (Nere et al., 2013) showed that sleeping plasticity may be governed by synaptic potentiation in the case of cuing to retrieve certain memories. Hinting to Ockham’s razor, (Nere et al., 2013) points to the fact that the brain would use two mechanisms to achieve the same thing. However if the second mechanism is deployed under different circumstances then a two-mechanism machinery for experience processing would be feasible. The unequal neurochemical environment of the brain between SWS and REMS gives this difference in circumstances. The different neurochemical environment during REMS, in comparison with SWS, (Walker & van der Helm, 2009) may serve as a foundation for a mechanism that is different from an SWS dependent mechanism. The specific electrophysiology of REM may be a phenomenon displaying and altering the relationship between neurons.
7.6.1 Non specificity of the downscaling
(Giuditta, 2014) points to a possible issue with the general nature of the downscaling theory. It states that, due to this general nature of the downscaling, there is no protection of older memories. The continuous entries in SWS should lead to an ever downscaling of older memories. This is an important point because the downscaling as is will eventually lead to the erasure of especially older experiences. Since the tandem hypothesis assumes the validity of the downscaling theory it should make an attempt to engage this issue. If somehow the long-term memory is protected from the downscaling effects of SWS then the problem would be solved.
7.6.1.1 A solution to the non-specificity problem
In searching for a possible protection of older memories the work by Buzsáki and Mizuseki (2014) could be considered. It argues that the brain may have a pre-configuration that can be measured electro-physiologically. According to this view a certain percentage of the neurons has a pre-configured activity. The model of the outside world in the brain is therefore not only dependent on experience but also on this pre-configuration. They argue that the difference between the pre-experience activity (which is the pre-configuration) of these neurons and the post-experience activity differs very little. Since experience has this little effect on the activity of these neurons it would seem that experience alters pre-configured neuronal activity just a little. When we consider this pre-configuration of a subset of neurons on one hand and on the other keep in mind the experience based occurrence of SWS in mammals (Giuditta, 2014; Roffwarg et al., 1966; Squire et al., 2008), it would be feasible to think that the downscaling activity has a cap set to the pre-configuration of the brain. Because SWS occurs later in life, it is possible that it cannot downscale the pre-configured activity of the brain.
23
This implies that in order for SWS to downscale certain connections there needs to be a large enough difference between the pre- and post-experience activity. Hence when an activity can hardly be distinguished from the preconfigured activity SWS can no longer downscale. This would also explain the necessity for the brain, first of all, to pick the copy with the least data to store but also to add to that copy enough information (from previous copies) so that the experience can be stored in a manner that best fits the internal model. Since the internal model is based on the pre-configuration, a situation transpires where the older memories are protected because they do not differ enough from the pre-configuration so as to be erased. This way the downscaling does not only work to keep energy consumption and space usage down but also aids in the creation of a best fit version of the experience for integration with the internal model. If there is indeed a pre-configuration that is taken into account by SWS then long-term memory is protected from the general nature of the downscaling.
7.7 Tandem hypothesis in relation to SFSR
According to (Groch et al., 2013; Wiesner et al., 2015) there is no evidence for a REM related decrease of affective tone. Furthermore according to (Walker & Stickgold, 2006), the idea that REMS is only involved in emotional memory may have been based on overgeneralizations. It would therefore seem that there is reason to doubt the idea that REMS is actively engaged in the separation of affective tone and factual information of the experience. The tandem hypothesis deals with this issue by stating that all processing of experience takes place during SWS – in accordance with (Tononi & Cirelli, 2012) – . All of the processed experience is then integrated in older memories and therefore also the internal model during REMS. This means that the decrease of affective tone also happens during the general downscaling of SWS. This can explain the sleep-dependent weakening of affective tone as has been shown in (van der Helm et al., 2011).
The view on experience encoding (box 1) states that the factual information of the experience (𝑝) would, normally (until 𝑓𝑝(𝑥) = 𝑓𝑞(𝑥)), be encoded stronger by definition than the affective tone (𝑞). The reason why this should be the case is that the affective tone only serves to draw attention to the experience. It creates the environment for strong connections during the experience. Conceptually, as an experience shifts from neutral to the more emotional variant and the brain’s neurochemical deployment changes accordingly connections are laid down stronger. The affective tone of the memory in this more emotional situation (𝑞𝑒) is also laid down stronger in comparison with the affective tone of the neutral experience(𝑞𝑛). Also component 𝑞 is typically weaker and therefore in
comparison with its 𝑝 companion more prone to being overruled by new experiences. This would explain why the affective tone of a salient event goes down upon repeated exposure.
24
7.7.1 Why does REMS seem to preferentially process emotion?
Considering the mechanism of the tandem hypothesis. SWS erases a part of the synaptic strength and therefore parts of the experience. After this a copy of the processed experience is integrated with long-term memory during REMS. During waking day experiences 𝑞𝑛 is already laid down weaker than 𝑞𝑒 according to the view on experience encoding. Throughout one sleeping period, fewer copies will contain 𝑞𝑛 as in comparison with 𝑞𝑒. This means that in the final copy that is integrated with existing
memory and the internal model there will be more 𝑞𝑒 than 𝑞𝑛. REMS therefore deals with more 𝑞𝑒
than 𝑞𝑛. This is the reason why an emotional memory would correlate more with REMS than a neutral memory would. Furthermore, because 𝑝 connections are stronger than their 𝑞 companions, the latter would reach the threshold (as introduced by the downscaling theory) earlier and disappear. So, when considered together with the view on experience encoding, the downscaling theory has incorporated in itself a possible explanation for why there would seem to be a divorce between affective tone and factual information. Interestingly, if the downscaling is relative as argued by the downscaling theory, then there would be a greater absolute loss of 𝑝 than of 𝑞. This way SWS can correlate more with the factual information than with affective tone. There is, therefore, no need for REMS to preferentially engage in emotional processing alone.
7.8 Limitations of the tandem hypothesis
The tandem hypothesis concentrates on REMS, SWS and their recurrent occurrence. It does not take into account stages I and II of NREM sleep, K-complexes and spindles. These aspects of sleep electrophysiology have processing roles which may or may not entail memory processing (Astill et al., 2014; de Zambotti et al., 2016; Rihm & Rasch, 2015; Stones, 1977). An example of this is the fact that REM electrophysiology seems to be engaged in internal processing (Roffwarg et al., 1966). It is possible that the occurrence of REM electrophysiology, at least partly, indicates the maintenance of the internal model. The new experiences can then be taken into consideration during this maintenance. Also the waking day consolidation would still have to be integrated in the internal model. This means that the dependence of REM sleep on the occurrence of SWS (as argued in the tandem hypothesis) cannot be very stringent. If SWS doesn’t occur or is severely truncated then REM sleep would still occur as has somewhat been stated in (Van Der Werf et al., 2009).
Concerning REMS and its profile during sleep, under section 7.3 (Cyclic nature of sleep electrophysiology) it was argued that later REM episodes should have a longer duration than earlier ones. It needs statement that earlier versions of the experience also have a higher resolution and would need more time to process. Therefore, even though the tandem hypothesis predicts longer REM durations in late sleep, it is not clear as to how this early versus late REMS duration relate to each other.
25
8 Conclusion
Scientific literature on sleep states that during sleep experiences are processed. This processing leads to memory enhancement. The electrophysiological characteristics of sleep have made it possible to divide sleep into stages, which are the subject of research. The replay to potentiate theory has some major questions to answer still. SFSR has to deal with the outcome of some experimental work stating that there is no direct connection between REMS and preferential processing of affective tone. The downscaling theory on the other hand seems to have a high level of validity. It explains what may be happening during SWS and why it is beneficial for the organism. However it joins most theories in that it is concentrated on one major sleep stage. Furthermore, no theory has explained why sleep stages occur many times during one sleeping period.
The tandem hypothesis however concentrates on multiple aspects of sleep electrophysiology. It states that both SWS and REMS work together to process waking period experiences and incorporate them into the internal model of the outside world. The tandem hypothesis assumes the validity of the downscaling theory and states further that during REMS the processed information is integrated with the internal model. The tandem theory does not have a preferential emotional processing role for REMS. This fits some experimental work and the idea that the attachment of REMS to emotional processing may have been based on some overgeneralizations (Groch et al., 2013; Walker & Stickgold, 2006). The tandem theory rather states that REMS has to do with engaging and incorporating the experience that was processed during the pre-occurring SWS. Via this role and with a view on experience encoding, this thesis further attempts to explain why correlations between REMS and emotion processing could be found in experimental data.
Further concerning the downscaling theory, this thesis introduces a cap on how much synaptic strength can be scaled down. This is based on the idea that the brain has a pre-configuration that dictates how experiences are encoded (Buzsáki & Mizuseki, 2014). This solves the problem of the non-specific nature of the general downscaling as has been noted by (Giuditta, 2014).
It would be interesting to investigate whether and if so how SWS and REMS are functionally linked. Will the manipulation of SWS have an effect on the profile of REMS as can be seen in OSAS? Will the manipulation of REMS lead to a diminished ability to incorporate all forms of experiences into existing memories; and is the behavioural effect of deprivation of the first REMS stage different from that of deprivation of the last REMS stage? These are some of the questions that are to be addressed experimentally. Also other characteristics of sleep electrophysiology are still to be taken into account in a comprehensive theory of sleep dependent processing.
