The effects of up- and down-state targeted stimulation in Slow Wave Sleep

(1)

M

ASTER

T

HESIS

-

S

JOERD

M

ANGER

___________________________________________________________ The effects of Up- and Down-state targeted stimulation

in Slow Wave Sleep.

____________________________________________________________________

University of Amsterdam

_____________________________________________________________________

STUDENT

Name: Sjoerd Manger ID-cart number: 10195815 Adress: Binnenweg 77a Telephone: 06-21333844

E-mail adress: sjoerdmanger@gmail.com SUPERVISORS

First reader: Lucia M. Talamini Second reader: Thomas Meindertsma

Location: Sleep and Cognition lab / REC L/G WORDS:5.298

DATE: 10-07-2016

(2)

Abstract

Memory reactivation lies at the heart of memory consolidation. It is thought that sleep forms an ideal moment for such a process. By understanding the reactivation process in sleep, we might be able to improve memory for patients suffering from memory disorders like Alzheimer’s or amnesia. Previous research has shown that auditory- or odor cues during non-REM sleep facilitated memory consolidation. It has been hypothesized that Slow Wave Sleep is the essential sleep stage at which this consolidation process takes place. More specifically, Slow Oscillations are argued to be associated with this distinctive feature. In this study we hypothesized that stimulations during the up- and the down-state of a Slow Oscillation have differential effects on memory performance. To study this, we used a phase predicting algorithm. This algorithm gave us the ability to deliver stimuli at any desired state of a Slow Oscillation. We hypothesize that stimulation during the up-states is essential and that this state is a unique contributor to the facilitating effects of memory reactivation. Concurrently we will examine the effect of down-state stimulation, since there seems to be little research about this specific oscillatory state.

(4)

Introduction

How are new experiences stored into long-lasting memories? It has been proposed that there are two separate stores for processing new information, the neocortex and the hippocampus (Marr, 1971; Buzsáki, 1996). New information is initially processed in both the neocortex and the hippocampus. This process is also known as the encoding phase. The hippocampus can be characterized as a temporary storage, whereas the neocortex is known for the ability to store for a longer term. The temporary storage of the hippocampus is prone to decay of information. Memory traces are therefore in need to be stabilized. This stabilization of memory traces is also known as memory consolidation. The fact that both the hippocampus and the neocortex are actively encoding new information might cause interference with the consolidation process. Consolidation during offline periods, such as sleep, is not prone to this interference. Sleep thus seems to serve as an ideal moment for the consolidation process (Rash and Born, 2012; Paller and Voss, 2015).

It has been hypothesized that reactivations of memory traces underlie the process of memory consolidation (Walker, Brakefield, Hobson, Stickgold, 2003; Ji and Wilson, 2006). Memory traces are reactivated in the hippocampus in order to stabilize information. This drives the neocortex to reactivate memory traces in its storage. Consequently, the new information is gradually becoming redistributed in the neocortical regions, such that they form long-lasting memories. This process has been shown to occur spontaneously during specific sleep stages (Dang-vu et al., 2008). Sleep can be mapped into five distinct sleep stages, stage 1,2,3,4 and REM-sleep (Rechtschaffen and Kales, 1968). Consolidation of episodic memory has mostly been associated with “Slow Wave Sleep” (SWS), corresponding to sleep stage 3 and 4 (Steriade and McCarley, 2005; Steriade, Nunez and Amzica, 1993; Steriade, Timofeev and Grenier, 2001).

It has been shown that external auditory- or odor cues facilitate declarative memory consolidation (Rash, Büchel, Gais and Born, 2007; Rihm, Diekelmann, Born and Rasch, 2014; Rudoy, Voss, Westerberg and Paller, 2009). Moreover, cued memory reactivation has also been shown to facilitate procedural skill learning (Cousins, El-Deredy, Parkes, Hennies and Lewis, 2014; Antony, Gobel, O’Hare, Reber and Paller, 2012). Interestingly, Schreiner and Rasch (2014) showed that verbal cues, presented during non-REM sleep, facilitated vocabulary learning, putatively by

(5)

reactivating associated memories. This suggests that complex stimuli (i.e. beyond the more basic odor- or auditory cues) can be used to reactivate memories during sleep. Although the results of the latter experiment are propitious, Schreiner and Rasch did not account for the neuronal processes that underlie memory reactivation.

Memory (re-) consolidation is associated with a number of oscillations:

spindles, which occur at a frequency of ± 11 – 16Hz, sharp wave-ripples of

approximately 200 Hz and Slow Oscillations (Dang-vu et al., 2011; Buzsáki, Horvath, Urioste, Hetke and Wise, 1992; Achermann and Borbely, 1992). Slow Oscillations, or SOs, are the result of global fluctuations on a cellular level. These oscillations occur at a frequency of about 0.8 Hz. SOs are characterized by so-called “up-states”, where a global neuronal depolarization can be observed and “down-states”, typified by a global hyperpolarization (Steriade, Nunez and Amzica, 1993).

Slow Oscillations gained considerable interest in memory research because of their functional importance in consolidation. Previous research has shown that it is possible to induce SOs via transcranial stimulation, resulting in an enhancement of retention of memories (Massimini et al., 2007; Marshall, Helgadóttoir, Mölle and Born, 2006). Importantly, these experiments did not take naturally ongoing oscillations into consideration. Ngo, Martinetz, Born and Mölle (2013) did take the endogenous oscillations into account and showed that auditory stimulation in phase (i.e. during the up-state) with ongoing slow oscillations enhanced its rhythm, phase coupled spindle activity and the consolidation of declarative memory. One can thus hypothesize that the up-state of a slow oscillation is an important contributor to the beneficial effects of memory trace reactivation found in the above-mentioned

experiments. The effect of down-targeted stimulation, however, is yet to be explored.

Cox, Korjoukov, de Boer and Talamini (2014) examined the effects of auditory stimulation in the up- and down state of slow oscillations on sleep-learning. They found that stimulation in the up-state and the down-state resulted in an early theta response (± 5- 10 Hz) and a later gamma response (± 10-35 Hz) when compared to naturally ongoing SOs. Up-targeted stimuli elicited a greater spindle/beta (± 15-30 Hz) band activity when compared to down-targeted stimuli. However, the authors found no behavioral differences in memory performance for words that were presented during SO up- or down states. This suggests that the auditory stimulation did not lead to formation of stable memory traces, leading to an absence of sleep-learning.

(6)

In the current experiment we continued the line of research of Cox, Kourjokov, de Boer and Talamini but took a unique turn by exploring phase specific stimulation effects in a memory reactivation study. Moreover, we investigated these effects in a full-night sleep, with (complex) verbal cues. We were determined to show that the up-state of a slow oscillation contributes to the foundation of memory consolidation, and additionally find the neuronal processes that were lacking from the study of Scheiner and Rash (2014). Concurrently, we examined the effects of stimulation in the down-state, to show that the up-state is unique for its facilitating effect on memory consolidation.

During sleep, a phase predicting algorithm delivered stimuli at the desired up- and down states of a slow oscillation. We theorized that stimulation in the up-state of a slow oscillation would elicit a fronto-central hyperpolarization around 500 ms after stimulus onset, and a depolarization around 900 ms after stimulus onset, similar to the findings by Ngo and colleagues (2013). Additionally, we would expect to see an improvement of memory performance on words that were stimulated during the up-state of a slow oscillation. Here we will show that this effect is specifically a result of stimulation during the up-state of a slow oscillation.

(7)

Methods and Materials

Subjects

Participants in this experiment were all Dutch students from the University of Amsterdam. All students provided written informed consent to participate in this study. None of the subjects in this study knew Danish or any other North Germanic

language. The ethical committee of the Department of Psychology approved this

study. A total of 17 subjects participated in this study, of which 10 were used for analysis. Subjects had to be excluded because of above-average performance in the memory-encoding task. For these subjects it was not possible to set up two equal pools of 40 words per condition (see Sleep procedure). Another reason for exclusion was due to the algorithm performance. In some cases, stimulation occurred outside the preferable sleep stages, which resulted in a reduction in the amount of presentations during sleep (since in these cases there were less opportunities for predictions to occur).

Auditory Stimuli

A total of 120 Danish words were used as auditory stimuli. All stimuli were singular, concrete nouns with one to three syllables. A native Danish speaker pronounced all words. Every single Danish word was translated into Dutch and to be presented visually. Given the temporal frequency of a Slow Oscillation (i.e. +/- one cycle per second), the Danish auditory stimuli were edited such that the middle of each sound clip would fit with the SO peak or trough. Moreover, the duration of the sound clip would then fit the duration of the desired phase. All stimuli were edited to be of equal length (500 ms). All stimuli were presented at on average 50dB (i.e. calibrated between 45 and 55 decibels).

Questionnaires

Most of the items from these questionnaires consisted of original items, meaning that they were not yet validated. The Stanford Sleepiness Scale (SSS) was used to assess each subject’s energy level or sleepiness (Hoddes, Zarcone, and Dement, 1972). The SSS has been positively evaluated and is still used in recent sleep research as a general tool of sleepiness assessment (Hoddes, Zarcone, Smythe, Philips, and Dement, 1973; Glevile and Broughton, 1979).

(8)

Procedure

Memory-Encoding

Subjects arrived at the sleep lab of the University of Amsterdam at 08:00 p.m. and were prepared for EEG and polysomnographic recordings. During these preparations, subjects were asked to fill in the first questionnaire (see Appendix A). This questionnaire assessed subject’s sleeping habits as well as their sleepiness. At approximately 10:00 p.m. subjects were instructed to take place in front of the stimulus monitor to start the memory task. During this memory task, all participants learned a total of 120 Danish words. The memory task was designed such that all words were randomly distributed over three blocks of 40 words each. Each of these blocks was presented twice, once in a passive- and once in an active learning form. Therefore, each subject was exposed to a total of six memory-encoding rounds.

The first encoding round was the first passive learning round. This first round started with a presentation of a fixation cross for 500 ms, followed by a Dutch word on the monitor (2000 ms). Subjects were then presented with the Danish translation by means of an auditory stimulus. This procedure was repeated until each of the 40 words was presented once.

An active learning round followed the passive learning round. In this active learning round subjects went through the same words as previously learned in the passive round, although all words were first randomly shuffled. This time, after a brief presentation of a fixation cross, the Danish word was first presented through the speakers. Subjects were then asked to type the correct Dutch translation. Each block of two rounds ended with a one-minute break. Memory performance was assessed in a final test of all 120 learned words (in randomized order) after all three blocks of two rounds were finished. This final test was similar to the procedure of the active learning round, except that subjects now had to indicate their confidence level on a 5-point Likert scale after typing their answer.

Sleep cueing

After memory performance was assessed, subjects went to bed between 23:00 and 23:30 p.m. (lights-off). All participants were woken at 08:00 a.m. the next morning. While subjects were asleep, a selection of 80 Danish words was presented as auditory stimuli. This selection of words was made based on performance on the final

(9)

memory test. A total of 20 correctly answered words were selected and labeled “Up” (i.e. these words were to be presented at the “up-state” of a slow oscillation). An addition of 20 incorrectly answered words was added to the pool of “Up” labeled words. For stimuli that were to be presented at the “down-state” we set up another pool of 40 selected words (i.e. 20 correctly answered and 20 incorrectly answered). Consequently, we were left with 40 words that were not presented during sleep. These words will henceforward be referred to as “uncued words".

Auditory stimulation started once the experimenter visually detected Slow Wave Sleep in the ongoing EEG signals. An algorithm was then activated that presented stimuli at a specific phase of ongoing slow oscillations (see SO Phase

Targeted Stimulation). Subjects showed signs of SWS at around 12 p.m., on average.

Auditory stimulation was discontinued three hours after the first stimulus was presented. This way, all subjects were given a fixed time window in which the algorithm could predict the phase of SOs. In some cases, for instance when stimulation inadvertently occurred during light-sleep, auditory stimulation led to signs of movement or wakefulness. In those cases, stimulation was halted and, when necessary, volume levels were lowered by 5 decibels. Stimulation would then continue after the experimenter visually detected new signs of SWS.

Memory retrieval

Subjects were woken the next morning at 08:00 am. After recovery from sleep inertia (± 30 minutes), and after a prepared breakfast, subjects were first asked to fill in a sleepiness questionnaire (see Appendix B). This second questionnaire assessed subjective sleep quality on the night before, as well as overall sleepiness. All subjects were then prepared for a memory recall test. This test consisted of the 120 previously learned words and an addition of 30 novel words. Subjects were presented with a fixation cross for 500 ms, followed by an auditory stimulation of a Danish word (1500 ms). Before any answers were typed, subjects were presented with a two-alternative forced choice. Here, subjects had to indicate whether the word was an old word or a new one. Subjects were then asked for the Dutch translation. Each trial ended with a Likert 5-point confidence scale. The recall memory performance test ended after performance for all 120 words was scored. This experiment ended after subjects filled in one last questionnaire (see Appendix C). Here we asked whether subjects noticed anything out of the ordinary during the preceding night

(10)

EEG and polysomnography

High-Density EEG activity was measured with a 64-channel WaveGuard cap, which was connected to a 72-channel REFA amplifier (ANT International, Enschede, The Netherlands; TMS International, Oldenzaal, The Netherlands). Two sintered mastoid electrodes were used for rereferencing the data. Additionally, two bipolar electrodes were used for electrooculography (one horizontal and one vertical recording) and one bipolar electrode was placed on the chin for electromyography recordings. Incoming signals were sampled at 512 Hz. During sleep an additional REFA amplifier was used, which doubled as the main input on which the algorithm could forecast brain activity. Note that only the signal from channel Fpz, referenced to M1, was used and send to a dedicated algorithm PC (see fig.1). Impedance levels were kept below 10 kΩ. EEG signals were recorded with ASALab software (ANT Neuro, 2015).

SO phase targeted stimulation

Stimuli were presented from a dedicated PC running software from OkazoLab called EventIDE (version 11.11.2015, 2015). This stimulus computer was connected to the algorithm computer, which in turn also ran the EventIDE software.

Figure 1. A schematic representation of the experimental setup. Here one can see all three PCs that were used in this experiment (noted as A, B, and C). A denotes the stimulus PC from which stimuli were presented and from which the trigger information is sent to amplifier 1. B denotes the algorithm PC. This PC generates predictions based on the input from amplifier 2. Also, B is connected to amplifier 1 in that it sends triggers that are stored, together with the EEG signal, in PC C. Amplifier 1 and 2 are connected via three jumped electrodes (Fpz, M1 and Patient Ground). The subject is denoted as “g.”.

(11)

This experiment uses a previously developed algorithm for targeting stimuli to selected phases of a slow oscillation (Cox, Korjoukov, de Boer and Talamini, 2014). The goal of the current study was to present stimuli at a specific phase of naturally ongoing slow oscillations. In order to do so, it was necessary to reconstruct this natural signal. We were specifically interested in the phases that correspond to the peak and the trough of a slow oscillation. To present stimuli at these desired phases, the incoming EEG-signal was band-pass filtered (order 1 Butterworth, 1Hz bandwidth) around 0.6 to 1.2 Hz. Next, a Fast Fourier Transform (FFT) was performed to a moving analysis window holding the last 5000 data samples (+/- 10 seconds of data). The Fourier coefficients of all reconstructed positive frequencies were rotated one-quarter cycle counter clockwise (i.e. -90°), and all negative frequencies were rotated one-quarter cycle clockwise (i.e. 90°). This procedure shifted all real-components (as defined by a cosine) into imaginary components (as defined by a sine). These shifted Fourier coefficients were then added back onto the original Fourier coefficients. Finally, the complex analytic signal was calculated by taking the inverse FFT of the result of the latter sum of coefficients. With this reconstructed analytic signal, we were able to calculate the power and phase of all frequencies from our input EEG signal. This allowed us to target and deliver stimuli at any desired phase. In order to do so, a sine wave was fitted on the analytic signal and forecasted into the future. Please note that we only used the analytic signal between 72 and 92% of the moving analysis window for sine-wave fitting. Whenever the algorithm predicted a desired phase along the extrapolated sine wave, a stimulus was presented at the first occurrence of that phase.

Predictions were made only once a set of parameter thresholds were surpassed. First, because we aimed for phases of ± 1Hz oscillations, the algorithm restricted its predictions at frequencies in the range from 0.6 to 1.2 Hz. Secondly, predictions were only made once the power of the signal surpassed a threshold of 0.6 (note that this is normalized power over the selected analytic window). Power was normalized to eliminate interpersonal differences in the level of power. The third parameter was a fitting threshold. This defined how good the fit between the sine wave and the analytic signal should be. We set the threshold for this fitting error to 0.1. As soon as the above-mentioned criteria were met, the algorithm was all set for sending out predictions. The fourth parameter was Crunch time. This parameter was mainly added

(12)

to aid the temporal precision of predictions. We set up a minimum of 70 ms and a maximum of 500 ms, meaning that predictions were stalled for at least 70 ms, up to a maximum of 500 ms. The predictions were stalled for 70 ms to account for any hardware lag and analysis time. We argued that 500 ms should be the maximum since the presentation would then be half a cycle away from the moment the prediction was made. The algorithm was looking for both the up and down state phases. The phase that was the first to be predicted was used for stimulus presentation.

Data analyses

Analyses of EEG data were done with MATLAB (R2012b; MATLAB, The MathWorks Inc., Natick, MA, 2012) and the version 13 EEGLAB toolbox (Delorme en Makeig, 2004). EEGLAB was mainly used for its pre-processing and visualization functions. Behavioral statistics were calculated in SPSS (Version 22; IBM Corp., Armonk, NY, 2013)

Sleep scoring

Galaxy software was used to characterize subjects’ sleep structure (PHIi, Amsterdam, the Netherlands). Data was scored according to standard criteria, on epochs of 30 seconds (Rechtschaffen and Kales, 1968).

Preprocessing

After EEG data acquisition was completed all data were high-pass filtered at 0.1 Hz to minimalize slow drift; an additional notch filter was applied at 50 Hz to counter power line noise. Data were referenced to the average of M1 and M2. The continuous EEG data files were epoched in segments from -1000 ms before stimulus onset to 3000 ms after. All trials were then baseline corrected on a time window from -200 ms to 0 ms (i.e. stimulus onset). Trials that contained artifacts or too many noisy channels were rejected based on visual inspection. Channels that were noisy on a fairly large period of time were interpolated using spherical interpolation. Only the trials that were scored in sleep stages 3 and 4 were included in the analyses (see Discussion).

(13)

Event-related potentials

ERPs were calculated by averaging over condition per subject. Then, averages were made over all subjects. Note that ERPs are included in this report only for visual inspection, to aid interpretation of the time frequency. Statistical analyses of the ERP data will be reported elsewhere.

Time-frequency analyses

A family of complex Morlet wavelets was used for time-frequency decomposition. Wavelets (n = 35) were logarithmically spaced from 1 to 100 Hz. Data was analyzed from 1 to 100 Hz to identify most of the frequency-band spectrum (i.e. from lower-delta to gamma). Even though we hypothesized auditory stimulation would elicit theta (for down-state) and lower-beta (for up-state) responses, we analyzed a broader spectrum of frequencies in favor of assessing frequency band specificity.

The width of the wavelets was set to -2 to 2 seconds in steps of 1/sampling-rate. Time frequency maps were visually inspected for edge artifacts that resulted from time-frequency decomposition. This led to a removal of 500 ms of data at both ends of the time spectrum. EEG power was baseline normalized according to decibel conversion (dB = 10*log10 (power/baseline)). This normalization was done on a baseline window from -500 to -200 pre stimulus onset. Time frequency decomposition was done for both phase conditions separately, yet on the same temporal baseline window, since we hypothesized that both conditions would elicit their activity in the same temporal window before stimulation.

Statistics

Behavioral statistics

Memory performance was statistically evaluated by comparing the performance from the encoding test with performance from the retrieval test. This was done for both the up- and down targeted words. This way one can compare the possible different effects of stimulation in the down- and up states on memory performance.

(14)

Time-frequency statistics

Non-parametric permutation tests were used for the statistics of time-frequency data. Non-parametric permutation testing was favored over parametric testing in view of the skewed, non-normal distribution of our data (see Results). Parametric testing would thus not be appropriate since assumptions about the theoretic distribution of the data would be violated. Instead, test statistics were derived from distributions that are created from the data itself.

To create a distribution of test statistic values under the null hypothesis a random selection of trials was labeled “up” and another random selection was labeled “down”. If the null-hypothesis were true, meaning that there is no difference between the time frequency power responses of up- and down targeted stimuli, one would expect that the labeling of these two conditions should not differ. From there on normalized power was calculated for each “fake condition”. Finally, the dB-normalized power from the “down condition” was subtracted from the “up condition”, for each time-point. This procedure was repeated for each permutation (n = 1000). P-values were computed via a Z-transformation or otherwise known as the PZ-method. All values from the time-frequency map were transformed to Z-values according to 𝑍𝑍 =𝑋𝑋− 𝜇𝜇_𝜎𝜎 , where x is the observed value. The observed value in this case was the dB-normalized power difference between “real” up labeled trials minus “real” down trials. The mean of the permuted values (denoted as 𝜇𝜇) were subtracted from the observed value and divided by the standard deviation (denoted as 𝜎𝜎) of those permuted values. The time-points with the highest values per permutation were stored in order to correct for multiple comparisons (i.e. pixel-based).

For group-level permutation statistics, subjects were shuffled instead of condition labels. On each permutation (n = 1000), a random number of subjects (between 1 to 9 subjects) were selected (e.g. 3 subjects). For each of those three subjects the difference in dB-normalized power between conditions was calculated by subtracting up normalized power from down (i.e. inverting the normal subtraction). The other subjects (in this case 7 subjects) would still be calculated on the “real” difference, meaning subtracting down normalized power from up normalized power. Finally, averaging over these 10 subjects produced one TF data set with a “fake” difference between up and down conditions. Z-values were calculated via z =

(15)

observed mean – permuted mean / std(permuted values). As in the within-subject statistics, all maximum values per permutation were stored in order to correct for multiple comparisons (pixel-based). Using an alpha level of 0.05, pixels were considered significant at P < alpha, since we used one-sided testing.

Results

Sleep characteristics

None of the participants noticed anything out of the ordinary during their sleep. Subjects’ sleepiness (before starting the task), or SSS score, was 3-4 on average, meaning that they indicated that they were relaxed, and a little drowsy, but still sufficiently active for concentration on the task at hand. We argued that this score is not influencing task-performance and that their sleepiness was normal for that time of the evening.

Subjects slept for 7.66 hours on average (Std = 0.68). When compared to the recording time, one can see that subjects had a sleep efficiency of 94.45% on average, indicating that there were little problems falling asleep. This efficiency level is what one could expect from a healthy subject (Moser et al., 2009; Rechtschaffen and Kales, 1986).

Table 1. Sleep stage proportions

Behavioral performance

Overall memory performance, collapsed across all cueing conditions, was analyzed through a paired-sample t-test. Memory performance at the end of encoding (𝑥𝑥̅ = 65.2) and on the retrieval test was not significantly different (𝑥𝑥̅ = 65.3, P = 0.92). However, when comparing across conditions, one may conclude that overall performance on the test was higher on the up-condition (𝑥𝑥̅ = 20.7) compared to the down condition (𝑥𝑥̅ = 19.7). Memory performance data were statistically analyzed trough another paired-sample t-test. Although the overall performance was higher for up-targeted stimuli, no significant differences were found in memory performance between the two conditions (P = 0.38).

Sleep Stage 1 2 SWS REM

Average 13,35 52,45 24,10 11,46 SEM 1,74 1,51 5,11 0,74

(16)

Figure 2. Hits in recall per condition. Bar charts indicating no significant differences in raw memory performance between conditions up (red) and down (purple). Please note that the score for the recall test for the up condition was higher than the score for the encoding test (20 correct words), whereas the overall score for the down condition was lower on the recall test when compared to the encoding test (20 correct words). Error bars were calculated with the standard error of the mean.

Event Related Potentials

All channels were visually inspected but only one frontal channel (Fz) is plotted here (see Fig. 3). From this figure, one may conclude that presentation during the up-state of a slow oscillation let to a destabilization of the oscillation, in that the ongoing wave seemed to be discontinued. Stimulation in the down-state, on the other hand, seemed to evoke a post-stimulation positivity. The two different conditions seem to realign after about 400 ms, suggesting that stimulation for both conditions resulted in similar evoked potentials, post-stimulus.

Figure 3. Average ERP for channel Fz. The red line corresponds to the Up-targeted oscillation

whereas the blue line denotes a down-targeted oscillation.

17 18 19 20 21 22 # cor re ct w or ds Up Down

(17)

Time-Frequency power

To justify the use of permutation testing, a distribution of the data has been made (see fig. 4). As can be seen in the figure, the data is skewed to the left, indicated by the fact that most of the data points are left to the central point.

Figure 4. Distribution of the dB-normalized power data. This histogram shows that the data is

skewed to the left, meaning that it is not normally distributed. The y-axis shows the amount of counts per bin. The x-axis shows the dB-normalized power. A total of 200 bins were used for visual inspection of the distribution.

Time frequency power was very similar across conditions. When averaged over subjects, both conditions show a power increase in the theta range at approximately 500 ms. Followed by later power increase at approximately 1000 ms (see fig. 5). More interestingly, up-targeted stimulation did differ from down-targeted stimulation. After subtraction of dB-normalized power (up minus down), one may notice several significantly enhanced power clusters (P < 0.05). The clusters that are most worthy noting are the power enhancement around 3 Hz (from 550 ms to 1200 ms), and the three clusters around the frequency band from 17 to 30 Hz (from 615 ms to 1000 ms). This activity is mostly found in the frontal areas of the brain for the lower theta range, and centrally for the spindle/gamma range (see fig. 7 and 8). The overall topography of the stimulus induced TF response does not seem to be very different between both conditions, although down-state stimulation seems to elicit more power in the right hemisphere when compared to up-state stimulated slow oscillations, both in the theta and spindle/gamma range (fig. 7 and 8).

(18)

Figure 5. Time-frequency power map for both conditions. The left plot shows TF-power for the up

condition and the right plot shows the power from the down condition. Both plots show power from electrode Fz. Note that the y-axis is logarithmically spaced.

Figure 6. Difference in power between Up and Down. Time-frequency power map from channel Fz.

Contours show the significant areas where power from the “up” condition was greater than the power from the down-targeted stimulation condition. Note that the x-axis shows data from t = 0 in favor of interpretation. Note that the y-axis is logarithmically spaced.

(19)

Figure 7. Topography of power in the up condition. The upper row shows activity from 2-3 Hz and

the lower row shows activity at approximately 25 Hz (derived from activity at 17-30 Hz).

Figure 8. Topography of power in the down condition. The upper row shows activity from 2-3 Hz

(20)

Discussion

We set out to show that the up-state of a slow oscillation contributes to the foundation of memory consolidation, and additionally find the neuronal processes that underlie this phenomenon. Concurrently, we were examining the effects of stimulation in the down-state, to show that the up-state is unique for its facilitating effect on memory consolidation.

However, the data fail to show such a distinct difference between the up- and down-state. As can be concluded from the ERPs, the topographical plots and the time-frequency plots of the individual conditions, there seems to be rather little difference between the two conditions. However, when statistically tested, up-targeted slow oscillations do seem to have increased power in the theta and spindle/gamma range, two frequency bands which one would expect to see during memory consolidation. Moreover, the spatial location (central; Pz, and frontal; Fz, Fpz) of the elicited activity does seem to be in line with other research (Ngo, Martinetz, Born and Molle, 2013; Cox, Kourjokov, de Boer and Talamini, 2014). Nonetheless, the apparent difference in power between the two conditions does not contribute to a difference in memory performance for words that were cued in a down or an up state. Consequently, from this study we cannot show the facilitating effect of word cueing during sleep, nor show that this effect is indeed unique to up-state word cueing.

There are several points that might contribute to the lack of significance in the behavioral data and the unclear distinction between power in the up and the down condition. First, the amount of word triggers was not 80 words per subjects. At the end, since we only used SWS trials, some trials (with unique word cues) were left out, resulting in an unequal amount of cued words per subject. Consequently, the amount of up- and down triggers was not similar across subjects either. This was not only due to the use of SWS trials but also because of the algorithm and its performance. The algorithm was set up such that the probability of a cue in the down-state should be close to the probability of a cue in the up-state. However, the average amount of cues in the up-state was 113.9 compared to 136.9 in the down-state (P = 0.086). Finally, we were left with a total of 10 participants. The availability of a larger sample of subjects would most likely diminish the effects of the inequality of the amount of words and triggers. Finally, in this study we were not controlling for naturally ongoing slow waves. For future research it is sensible to send out predictions of up- and down-states

(21)

but hold actual auditory presentation. This way one could argue for the specific effect of stimulation versus no stimulation.

Despite the fact that (a part of) our hypotheses were not accepted for this study, research in this particular field is still important. Memory consolidation is a fundamental process that drives a range of different cognitions. Previous research has shown that examining memory reactivation can form a gateway to memory consolidation. Understanding this memory reactivation process forms the essential step towards setting up real-life applications in medical use, such as improving memory for patients suffering from Alzheimer’s, amnesia, or any other memory disorder.

(22)

References

Antony, J.W., Gobel, E.W., O’Hare, J.K., Reber, P.J., & Paller, K.A. (2012). Cued memory reactivation during sleep influences skill learning. Nature

Neuroscience, 15, 1114-1118.

ASAlab [Computer Software]. (2015). Retrieved from https://www.ant-neuro.com/products/asa-lab.

Buzsáki, G. (1996). The hippocampo-neocortical dialogue. Cerebral Cortex, 6, 81-92.

Buzsáki, G., Horvath, Z., Urioste, R., Hetke, J. & Wise, K. (1992) High-frequency network oscillation in the hippocampus. Science, 256, 1025–1027.

Cousins, J.N., El-Deredy, W., Parkes, L.M., Hennies, N., & Lewis, P.A. (2014). Cued Memory Reactivation during Slow-Wave Sleep Promotes Explicit Knowlegde of a Motor Sequence. The Journal of Neuroscience, 34, 15870-15876.

Cox, R., Korjoukov, I., de Boer, M., & Talamini, L.M. (2014). Sound Asleep: Processing and Retention of Slow Oscillation Phase-Targeted Stimuli. PloS

ONE, 9, 1-12.

Dang-vu, T.T., Bonjean, M., Schabus, M., Boly, M., Darsaud, A., Desseiles, M., … Maquet, P. (2011). Interplay between spontaneous and induced brain activity during human non-rapid eye movement sleep. Proceedings of the National

Academy of Sciences, 108, 15438-15443.

Delorme, A., & Makeig, S. (2004) EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics. Journal of Neuroscience Methods, 134, 9-21.

Diekelmann, S., & Born, J. (2010). The memory function of sleep. Nature, 11, 114- 126.

(23)

Glenville, M. & Broughton, R.J. (1979). Reliability of the Stanford Sleepiness Scale compared to short duration performance tests and the Wilkinson auditory vigilance task. Pharmacology of the States of Alertness, Pergamon, Oxford, 1979, pp 235-244.

Hoddes, E., Zarcone, V., & Dement, W. (1972). Development and use of Stanford Sleepiness scale (SSS). Psychophysiology, 9, 150.

Hoddes, E., Zarcone, V., Smythe, H., Phillips, R., & Dement, W.E. (1973).

Quantification of sleepiness: A new approach. Psychophysiology, 10, 431-436.

Iber, C., Ancoli-Israel, S., Chesson, A.L., & Quan, S.F. (2007). The AASM Manual for the Scoring of Sleep and Associated Events. American Academy of Sleep

Medicine, Westchester, United States of America.

IBM Corp. Released 2013. IBM SPSS Statistics for Macintosh, Version 22.0. Armonk, NY: IBM Corp.

Ji, D. & Wilson, M.A. (2006). Coordinated memory replay in the visual cortex and hippocampus during sleep. Nature Neuroscience, 10, 100–107.

Korjoukov, I. (2015) EventIDE, Okazolab Ltd [Computer Software]. London.

Marr, D. (1971). Simple memory: a theory for archicortex. Philosophical

Transactions of the Royal Society B: Biological Sciences, 262, 23–81.

Marshall, L., Helgadóttir, H., Mölle, M., and Born, J. (2006). Boosting slow oscillations during sleep potentiates memory. Nature, 444, 610–613.

Massimini, M., Ferrarelli, F., Esser, S.K., Riedner, B.A., Huber, R., Murphy, M., Peterson, M.J., and Tononi, G. (2007). Triggering sleep slow waves by transcranial magnetic stimulation. Proceedings of the National Academy of

(24)

MATLAB and Statistics Toolbox Release 2012b, The MathWorks, Inc., Natick, 2012.

Ngo, H.V., Martinetz, T., Born, J., & Molle, M. (2013). Auditory Closed-Loop Stimulation of the Sleep Slow Oscillation Enhances Memory. Neuron, 78, 545– 553.

Nir, Y., Staba, R.J., Andrillon, T., Vyazovskiy, V.V., Cirelli, C., Fried, I., & Tononi, G., (2011). Regional Slow Waves and Spindles in Human Sleep. Neuron, 70, 153–169.

Paller, K.A. & Voss, J.L. (2015). Memory reactivation and consolidation during sleep.

Learning and Memory, 11, 664-670.

Schabus, M., Dang-Vu, T.T., Heib, D.P.J., Boly, M., Desseilles, M., Van de Walle, G., . . . Maquet, P., (2012). The fate of incoming stimuli during NREM sleep is determined by spindles and the phase of the slow oscillation. Frontiers in

Neurology, 3, 1-11.

Schreiner, T., & Rasch, B. (2014). Boosting Vocabulary Learning by Verbal Cueing During Sleep. Cerebral Cortex, 1-11.

Steriade, M., & McCarley, R.W. (2005). Brain Control of Wakefulness and Sleep.

Springer, New York.

Steriade, M., Nunez, A., & Amzica, F. (1993). A novel slow (1 Hz) oscillation of neocortical neurons in vivo: Depolarizing and hyperpolarizing components.

Journal of Neuroscience, 13, 3252–3265.

Steriade, M., Timofeev, I., & Grenier, F. (2001). Natural waking and sleep states: A view from inside neocortical neurons. Journal of Neurophysiology 85, 1969– 1985.

Rasch, B., Büchel, C., Gais, S., Born, J. (2007). Odor cues during slow-wave sleep prompt declarative memory consolidation. Science, 315, 1426–1429.

(25)

Rechtschaffen, A., & Kales, A. (1968). A manual of standardized terminology, techniques and scoring system of sleep stages in human subjects. Los Angeles: Brain Information Service/Brain Research Institute, University of California.

Rihm, J.S., Diekelmann, S., Born,J., Rasch, B. (2014). Reactivating memories during sleep by odors: odor specificity and associated changes in sleep oscillations. Journal of Cognitive Neuroscience, 23, 1–14.

Rudoy, J.D., Voss, J.L., Westerberg, C.E., Paller, K.A. (2009). Strengthening individual memories by reactivating them during sleep. Science, 326, 1079.

Walker, M.P., Brakefield, T., Hobson, J.A. & Stickgold, R. (2003) Dissociable stages of human memory consolidation and reconsolidation. Nature, 425, 616–620.

(26)

APPENDIX

____________________________________________________________________ A. (Questionnaire – Evening, before memory encoding task)

(27)

_____________________________________________________________ B. (Questionnaire – morning, before Recall test)

(28)

___________________________________________________________________ C. (Questionnaire – after the experiment was finished)