Slow-wave activity
J Sleep Res. (2014) 23, 255–262
Effect of prolonged wakefulness on electroencephalographic oscillatory activity during sleep E C K E H A R D O L B R I C H 1 , H A N S P E T E R L A N D O L T 2 , 3 , 4 and PETER ACHERMANN2,3,4 1 Max Planck Institute for Mathematics in the Sciences, Leipzig, Germany, 2Institute of Pharmacology and Toxicology, University of Zürich, Zürich, Switzerland, 3Zurich Center for Integrative Human Physiology, University of Zurich, Zurich, Switzerland and 4Neuroscience Center Zurich, University and ETH Zurich, Zurich, Switzerland
Keywords alpha oscillations, delta oscillations, oscillatory events, sleep deprivation, sleep electroencephalography, sleep spindles, spectral analysis Correspondence Dr Eckehard Olbrich, Max Planck Institute for Mathematics in the Sciences, Inselstrasse 22, D-04103 Leipzig, Germany. Tel.: +49-34-1-9959-568; fax: +49-341-9959-555; e-mail: [email protected] Accepted in revised form 23 November 2013; received 10 June 2013 DOI: 10.1111/jsr.12123
SUMMARY
The human sleep electroencephalogram (EEG) is characterized by the occurrence of distinct oscillatory events such as delta waves, sleep spindles and alpha activity. We applied a previously proposed algorithm for the detection of such events and investigated their incidence and frequency in baseline and recovery sleep after 40 h of sustained wakefulness in 27 healthy young subjects. The changes in oscillatory events induced by sleep deprivation were compared to the corresponding spectral changes. Both approaches revealed, on average, an increase in low frequency activity and a decrease in spindle activity after sleep deprivation. However, the increase of oscillatory events in the delta range and decrease in the sigma range occurred in a more restricted frequency range compared to spectral changes. The mean relative power spectra showed a significant increase in theta and alpha activity after sleep deprivation while, on average, the event analysis showed only a weak effect in the theta band. The reason for this discrepancy is that the spectral analysis does not distinguish between diffuse activity and clearly visible temporally localized oscillations, while the event analysis would detect only the latter. Additionally, only a few individuals clearly showed activity in the theta or alpha frequency bands. Conversely, event analysis revealed that some individuals showed an increased rate of sleep spindles after sleep deprivation, a fact that was not evident in the relative power spectra due to a decrease in background activity. The two methods complement each other and facilitate the interpretation of distinct changes induced by prolonged wakefulness in sleep EEG.
INTRODUCTION Oscillatory activity in electroencephalography (EEG) is considered as an emergent property of the thalamocortical system, with specific patterns and dominant frequencies of these oscillations depending on the functional state of the brain (Timofeev and Bazhenov, 2005). Typical patterns in EEG in conjunction with eye movements [electro-oculography (EOG)] and muscle tone [electromyography (EMG)] serve to discriminate non-rapid eye movement (NREM) sleep, rapid eye movement (REM) sleep and waking (Iber et al., 2007; Rechtschaffen and Kales, 1968). Classically, oscillatory activity during sleep is subdivided into delta ª 2013 European Sleep Research Society
(0.5–5 Hz), theta (5–8 Hz), alpha activity (8–11 Hz) and spindle frequency activity (11–16 Hz). EEG can be analysed in many ways, e.g. in the frequency domain by spectral analysis (decomposition of signals into its constituting frequency components) or in the time domain with period-amplitude analysis (information such as incidence and amplitude of waves; Geering et al., 1993; Ktonas and Gosalia, 1981). Time–frequency analysis methods such as wavelet analysis or matching pursuit combine elements of both domains (see e.g. Ktonas et al., 2009). A particular time–frequency analysis method based on autoregressive modelling of short (1 s) overlapping EEG segments was proposed by Olbrich and Achermann (2005) (see also
255
256
E. Olbrich et al.
Methods) to characterize transient oscillatory activity typical for EEG. In the recent past, the effect of sleep deprivation on human sleep EEG was studied mainly using spectral analysis. The main findings were the relative increase of spectral band power in the recovery night compared to the baseline night ly et al., 1981; Finelli et al., for frequencies ≤11 Hz (Borbe 2001a) and decreasing power in higher frequencies (Finelli et al., 2001a). The increase of spectral power in the low frequency bands was most prominent in frontal areas, while the decrease of power in the higher frequency bands appeared in more central and posterior areas (Finelli et al., 2001b; Marzano et al., 2010; Tinguely et al., 2006). Studies investigating the effect of sleep deprivation on specific sleep oscillations focused on sleep spindles (Knoblauch et al., 2003) and slow oscillations (Bersagliere and Achermann, 2010). Sleep deprivation resulted in increased spindle amplitude and a reduction in spindle density, while spindle duration was not affected (Knoblauch et al., 2003). Slow oscillations were redistributed in response to increased sleep pressure: the number of waves/min was reduced below 0.9 Hz and increased above 1.2 Hz (Bersagliere and Achermann, 2010). Our aim was to investigate how oscillatory events in sleep EEG (event analysis) are affected by increased sleep pressure, i.e. after 40 h of sustained wakefulness, and how these changes relate to the previously observed changes in the corresponding power density spectra. Furthermore, we were interested in interindividual variation in the response to sleep deprivation. METHODS Data The analyses were performed on existing data sets of 27 healthy young male participants of previous studies investigating the effects of sleep deprivation on EEG topography [n = 8 (Finelli et al., 2001b)] and of caffeine during prolonged wakefulness [n = 19, placebo condition (Landolt et al., 2004; tey et al., 2006)]. Polysomnographic recordings were Re obtained during an adaptation night, a subsequent baseline and a recovery night after 40 h of sustained wakefulness. Bedtime for all 3 nights was scheduled at 23 h (n = 12) or 24 h (n = 15). Sleep was limited to 8 h for the adaptation and baseline nights and to 12 h (n = 8) or 10.5 h (n = 19) for the recovery nights. Participants were instructed to abstain from alcohol and to adhere to regular bedtimes (8 h time in bed) for 3 days prior to the study, verified by ambulatory activity monitoring and sleep–wake diaries. The study protocols and experimental procedures were approved by the local ethics committees for research on human subjects and participants gave their written informed consent. The sleep EEG data of baseline and recovery sleep after 40 h of sustained wakefulness were analysed. The EEG signals were sampled at 128 Hz (for additional details see Finelli et al., 2001b; Landolt
tey et al., 2006). Sleep stages were scored et al., 2004; Re visually for 20-s epochs (C3A2 derivation), according to the criteria of Rechtschaffen and Kales (1968). The analysis of recovery sleep was restricted to the length of NREM sleep of the corresponding baseline. Event detection Oscillatory events in the sleep EEG were detected in derivation C3A2 using a previously published algorithm (Olbrich and Achermann, 2005). This algorithm is based on modelling overlapping 1-s segments of the EEG time–series by an autoregressive model of order 8 [AR(8)-model]. Thus, the EEG is modelled as a superposition of maximal 4 stochastically driven harmonic oscillators, with damping and frequency varying in time. More precisely, an AR (p) model corresponds to m oscillators and n relaxators with P = 2 m + n. Oscillatory events are detected whenever the damping constant at one or more frequencies is below a predefined threshold c ≤ 6.6 s 1. For a detailed description of the algorithm and its application to human baseline sleep EEG data see Olbrich and Achermann (2005). Detected events in the sleep EEG of a baseline night are shown in Fig. 1. Events were characterized by their time of occurrence, frequency and duration. Analysis of detected events In a first step, we compared the detected events with spectral analysis using the same frequency resolution, i.e. we determined histograms of the events with a frequency bin size of 0.25 Hz (Fig. 1, right panel). For further analysis, the events were grouped into four frequency bands: delta (0– 5 Hz), theta (5–8.5 Hz), alpha (8.5–10.5 Hz) and sigma bands (10.5–16 Hz). These four bands are commonly used and correspond to the peaks in the histogram of the event frequencies in Fig. 1. We subdivided the delta and sigma bands further into two sub-bands: slow delta 0 < f ≤ 2 Hz, fast delta 2 < f ≤ 5 Hz, slow sigma 10.5 < f ≤ 12 Hz and fast sigma 12 < f ≤ 16 Hz. A relevant fraction of the slow delta events were events with f = 0 Hz. While these events are technically non-oscillatory, they were taken into account if the instantaneous frequency during the event was larger than zero on more than one occasion (for details see Olbrich and Achermann, 2005). We distinguished slow and fast delta based on the distinct behaviour of slow (≤2 Hz) and fast delta activity during sleep deprivation and recovery sleep (reviewed in Ferrara and De Gennaro, 2011). The sigma band was divided at f = 12 Hz in order to study possible differences between slow and fast spindles (De Gennaro and Ferrara, 2003; Werth et al., 1997). For each frequency band we calculated the following event properties for Stage 2, slow wave sleep (SWS) and NREM sleep (Stages 1–4): mean event frequency (Hz) by averaging over the frequencies of the single events, event rate as the ª 2013 European Sleep Research Society
Sleep deprivation and sleep oscillations
257
25
20
f [Hz]
15
10
5
Figure 1. Top left: spectrogram and superimposed oscillatory events (black dots) of an individual baseline night. Spectra (derivation C3A2) are colour-coded on a logarithmic scale (0 dB = 1 lV2 Hz 1; 20 dB). Top right: –10 dB frequency distribution of oscillatory events. Bottom: hypnogram. MT: movement time; W: waking; REM: rapid eye movement sleep; 1–4: non-rapid eye movement (NREM) sleep Stages 1–4.
0 0
0
50
100
number of events divided by the time in the particular stage (min 1), mean event duration (s) by averaging the durations of the single events and the event ratio as the ratio between the sum of the durations of the events in this frequency band divided by the total time in the particular stage. The event ratio, combining the duration and the rate, may be a more robust measure of oscillatory activity than the event rate. Spectral analysis Power density spectra were calculated for consecutive 20-s epochs (FFT, Hanning window, average of five 4-s epochs; matched with sleep stages) resulting in a frequency resolution of 0.25 Hz. Artefacts were excluded by visual inspection and semiautomatically (moving average threshold) to exclude high- and low-frequency artefacts (Finelli et al., tey et al., 2006). 2001b; Re To measure rhythmic activity relative to the background of the power density spectrum, a power law function was fit to the spectrum in the 2–5-Hz and 16–25-Hz ranges, excluding the 5.25–15.75-Hz range (containing theta, alpha and spindle peaks) (for details see Geiger et al., 2011) and (Whitten et al., 2011) for a similar method. Activity in the theta, alpha and sigma range was determined as power in the corresponding range minus background power in the same frequency range. Statistics Differences in power density spectra and the distribution of event ratios between baseline and recovery sleep were compared by paired t-tests. Power was log-transformed prior to statistical testing and performed for 0.25-Hz bins. A change was only considered significant if at least three consecutive bins reached significance. Changes in event ª 2013 European Sleep Research Society
50
100
Number of events
MT W REM 1 2 3 4 150
200
250
300
350
400
450
t [min]
ratios and event frequency from baseline to recovery sleep were assessed by boxplots and Wilcoxon’s signed-rank tests. Furthermore, relationships between changes in event properties or events and band power were quantified by Spearman’s rank correlations. RESULTS Fig. 2 illustrates the comparison of the event analysis with the spectral analysis for the data averaged over the subjects. While the blue curves in Fig. 2 show the comparison for NREM sleep, we additionally analysed Stage 2 and SWS (green and red) in order to identify sleep stage-specific changes and to exclude effects of the increased fraction of SWS in recovery sleep. Sleep deprivation resulted in an increase of spectral power from delta to the alpha frequencies and a decrease in the sigma range in NREM sleep, Stage 2 and SWS (Fig. 2). Oscillatory events were increased in the delta range and decreased in the sigma range, i.e. in a more restricted frequency range compared to spectral changes. The peaks in the mean spectra corresponded to peaks in the distributions of the event ratios. The peaks were, however, not exactly at the same frequencies in the spectra and histograms; their positions may have differed by up to 0.5 Hz (Fig. 2). Comparison of Stage 2 sleep (Fig. 2, green lines) with SWS (Fig. 2, red lines) revealed that the main effect of sleep deprivation on the occurrence of oscillatory events was found for delta events in SWS, while in Stage 2 the largest differences in the event ratios were observed for events in the spindle frequency (sigma) range. Additionally, in the average relative power spectra an increase in the alpha band in SWS and in the theta band in Stage 2 was observed. While the latter change was also
258
E. Olbrich et al. Recovery
Baseline
rec/bl
2
10
1.6
10
1.4
Ratio
Power density [μV2 per 0.25 Hz]
2
1.2
0
0
10
10
1 0.8
rec−bl –3
x 10 0.015
0.015
NREM ST2 SWS
0.01
Δ Event ratio
Event ratio
6 0.01
0.005
0.005
4 2 0 –2
0
0
10
f [Hz]
20
0
0
10
20
f [Hz]
reflected by a small peak in the event ratio, the increase of relative alpha power was not reflected in the event analysis. Fig. 3 summarizes the mean changes of the event properties (event rates, durations, event ratios and event frequency) in response to sleep deprivation of different frequency bands. Mainly delta activity (d1, d2) and fast spindles (r2) were affected by prolonged wakefulness. Fast delta events (d2) occurred with a higher rate in Stage 2 and were of longer duration in Stage 2 and SWS, leading to a higher event ratio. Averaged over total NREM sleep, slow delta activity showed, on average, a higher event rate and ratio. This reflects mainly the larger amount of SWS in the recovery night, because the effect vanished in the stage-specific analysis. However, a small but significant increase of event duration resulting in a higher event ratio was found in Stage 2. Moreover, we found increase of event rate, duration and ratio for events in the theta band of Stage 2. This change is reflected more prominently in the relative power spectra (Fig. 2). Individual-level analysis revealed that individual variability in the theta and alpha frequency bands is high (see examples in Fig. 4 and Data S1) and that the changes resulted from a subset of individuals. It is interesting to note that the event frequency in the two delta bands seem to change in opposite directions after sleep deprivation: slow delta events (d1) became faster (higher frequency) while faster events (d2) appeared to become slower. These changes, however, depended to some degree on the frequency applied for subdividing into the two subbands. Furthermore, fast spindles (r2) had lower frequency and event rates in Stage 2, SWS and consequently also in NREM sleep. So far, we have presented results on averaged data over all subjects. However, interindividual variation is considerable, and the individual picture can be quite different from the
0
10
f [Hz]
20
Figure 2. Average power spectra and oscillatory events of baseline and recovery sleep (n = 27), non-rapid eye movement sleep (blue), Stage 2 (green) and slow wave sleep (red). Top: mean power spectrum of baseline (left), recovery sleep (middle) and ratio recovery/baseline (right). Bottom: event ratios of the baseline (left), recovery sleep (middle) and their difference (right). Bars at bottom of right panels indicate significant non-zero mean differences (P < 0.05) between recovery and baseline sleep.
average result. This is exemplified in Fig. 4 with data of three individuals (for data of all individuals see Data S1). Individual MAP16 (top) displays a pattern similar to the average one and can therefore be considered as a ‘typical’ pattern: increasing spectral power in the delta and alpha bands of recovery sleep is accompanied by a corresponding increase of the event ratio and decreasing power in the spindle range by a decrease in the event ratio. The changes in the alpha range appear to be weaker in the event analysis. This is due, however, to a different comparison of recovery and baseline sleep in the two analyses. For the spectra, the ratio recovery to baseline (percentage change) was determined, while for the event analysis the difference of the event ratio had to be calculated, as events did not occur in all frequency bins (event ratio of zero in baseline). Additionally, signatures of a shift in the spindle peak frequency towards lower frequencies and of fast delta events to higher frequencies were evident. The latter, however, could be seen only in the difference of the event ratios. Individual MAP18 (Fig. 4, middle) showed remarkable deviations from the average pattern. The event ratios of both fast delta events and sleep spindles increased after sleep deprivation. The higher incidence of sleep spindles in recovery sleep was evident only in the event analysis. Individual MAP07 (Fig. 4, bottom) represents an example of an individual with a strong increase in oscillatory events in the theta frequency band that was accompanied by peaks both in the spectra of baseline and recovery sleep as well as in the relative power spectrum. Note that for this individual the event analysis revealed a decrease of the event ratio of the fast delta events, while the relative power spectrum showed an increase in this frequency range. Furthermore, the alpha peak in the relative power spectrum was not accompanied by a corresponding peak in the event analysis. ª 2013 European Sleep Research Society
Sleep deprivation and sleep oscillations –1
Δ Evt rate [min ]
δ
1
5
Δ Duration [s] 5
∗
0
∗
∗
1
∗
0 –1
∗
∗
∗
Δ f [Hz] 0.2
0
0
–0.2
–0.2
–5
1 2
Δ Evt ratio 0.2
0
–5
δ
∗
0.02
0
0
–1
–0.02
∗
∗
∗
259
0.5
∗
∗
∗
0 –0.5 –3
x 10
θ
1
0.5
∗
0
1
0
∗
0 –1
–5
–0.5
–3
x 10
α
1 0 –1
1
σ
0
2
0
0
–0.5
∗
∗
∗
0.04
0 –1
∗
0.02
0.04
0
0
–0.02
–0.04
0.04
0
0
0
–2
–0.04
–0.04
NREM ST2 SWS
1
–5
0
–1
σ
5
0.5
1
2
0.5
–0.5
Because it is not feasible to show and discuss all individual patterns, we mention further that some individuals showed a theta peak in the relative power spectrum but no or only a few oscillatory events in the theta frequency range and therefore no peak in the difference of the event ratios. In order to elucidate further the relationship between event and spectral analysis we also studied correlations between spectral band power and event ratios in the corresponding bands in both baseline and recovery sleep. Spectral band power in all frequency bands except the delta bands was estimated in two ways: first in the ‘usual’ way, by integrating spectral power density in the corresponding frequency band, and secondly by additionally subtracting the background power. Spectral band power and event ratios in the corresponding frequency band were correlated in all conditions and frequency bands except for the fast delta band 2–5 Hz (Table 1). In this band the event ratio and the spectral band power were correlated only in Stage 2 of recovery sleep. In general, correlation coefficients were higher in the alpha and sigma bands than in the delta and theta bands. Subtracting the background power increased the strength of the correlation. The most striking increase was observed in SWS, supporting our view that to a large degree the background power reflects SWA. ª 2013 European Sleep Research Society
∗
5
0
–1
Figure 3. Boxplots of differences in events rate (D evt rate), duration, ratio (D evt ratio) and frequency (D f) between recovery and baseline sleep in six different frequency bands: d1…0–2 Hz, d2…2–5 Hz, h…5– 8.5 Hz, a…8.5–10.5 Hz, r1…10.5–12 Hz, r2…12–16 Hz. Boxes represent the lower quartile, median and upper quartile. The notch indicates the 95% confidence interval of the median. Maximum whisker length is 1.5 times of the interquartile range. They extend to the most extreme data value that is not an outlier. Outliers are indicated by dots. The black asterisks indicate significant (P < 0.05) non-zero median differences. NREM: non-rapid eye movement sleep (blue); ST2: Stage 2 (green); SWS: slow wave sleep (red).
∗
NREM ST2 SWS
∗
∗
∗
0.2
∗
∗
∗
0 –0.2
NREM ST2 SWS
NREM ST2 SWS
DISCUSSION A previously developed method for the analysis of sleep oscillations was applied to study the effects of prolonged wakefulness on human sleep EEG and the outcome was compared with the results of spectral analysis of the same data. We performed the analysis separately for Stage 2 and SWS in order to disentangle the changes occurring within NREM sleep from the effect resulting of increased SWS in recovery sleep. The global picture was very similar with both methods: on average, an increase in low frequency activity and a decrease of spindle activity was observed after sleep deprivation. Event ratios and spectral band power as well as the corresponding changes resulting from sleep deprivation were highly correlated. The peaks in the histograms of the event frequencies corresponded roughly to the peaks in the relative power spectra; however, the relative strength of the components was different. Fast sleep spindles (r2) were reduced after sleep deprivation and their frequency became slower (Fig. 3). This slowing of fast spindle activity resulted in a shift of the spindle peak in the power density spectra. This is most evident in the relative spectrum of Stage 2 (Fig. 2, green line), with increased activity at lower spindle frequencies followed by a sharp drop with decreased activity at higher spindle frequencies.
E. Olbrich et al. Baseline
Recovery
Ratio
0
10
1
0.03
0.03
0.02
0.02
0.01
0.01
0
0
–0.01
2
1.4
10
Difference
Event ratio
1.5
0
10
2
2 [μV per 0.25 Hz]
REC vs. BL 2
2
10 MAP16
2
[μV per 0.25 Hz]
2
10
0.01 0
10
1.2
MAP18 0
1
0
10
Ratio
260
10
0.8 –3
0.015 0.01 0.005
0
0
5 0
2
[μV2 per 0.25 Hz]
Difference
0.01 0.005
10
10
2
2.5
0
1.5
10
2
MAP07 0
10
10
Ratio
Event ratio
x 10 0.015
1
0.01 0.02
Difference
Event ratio
0.5 0.02 0 0.01
0.01 –0.01
0
0
10
20
0
0
10
f [Hz]
20
0
f [Hz]
10
20
f [Hz]
Figure 4. Average power spectra and oscillatory events of baseline and recovery sleep in three individuals separated by horizontal lines: non-rapid eye movement sleep (blue), Stage 2 (green) and slow wave sleep (red). For details see Fig. 2.
Table 1 Correlations between the event ratios and the spectral band power P and the spectral band power after subtracting background activity Pdiff2bg Baseline
Recovery
S2 P d d1 d2 h a r r1 r2
0.43 0.48 0.19** 0.33* 0.77 0.65 0.66 0.65
SWS Pdiff2bg
P
0.48 0.83 0.74 0.71 0.73
0.48 0.50 0.16** 0.44 0.76 0.61 0.80 0.53
Baseline–recovery
S2 Pdiff2bg
P
0.65 0.90 0.81 0.79 0.86
0.32* 0.37* 0.51 0.57 0.86 0.67 0.60 0.58
SWS Pdiff2bg
P
0.66 0.90 0.77 0.62 0.66
0.50 0.54 0.26** 0.40 0.61 0.70 0.84 0.50
S2 Pdiff2bg
P
0.66 0.89 0.84 0.78 0.87
0.45 0.53 0.0** 0.86 0.87 0.29** 0.15** 0.49
SWS Pdiff2bg
P
Pdiff2bg
0.89 0.90 0.42 0.33* 0.56
0.47 0.49 0.03** 0.51 0.58 0.02** 0.30** 0.00**
0.53 0.64 0.32** 0.48 0.30**
SWS, slow wave sleep; S2, stage 2. Analysis was performed for baseline and recovery sleep, as well as for the changes induced by sleep deprivation [difference in event ratios correlated with relative power (ratio recovery to baseline sleep)]. Background activity was not subtracted in the d range. All correlations were significant with P < 0.05, except *0.05 < P < 0.1 and **P > 0.1.
The effect of sleep deprivation on alpha and theta activity appeared strongly in the relative power spectrum, but only weakly in the event analysis. Reviewing the data on an individual basis revealed that some individuals showed a
peak only in the relative power spectrum but not in the difference of the event ratios (see e.g. Fig. 4, MAP07). To understand these differences, one has to take into account that the spectral power is a temporal average on 20-s ª 2013 European Sleep Research Society
Sleep deprivation and sleep oscillations segments, while the event analysis works on 1-s segments and picks up only those episodes with strong oscillatory activity. Therefore, the event analysis is specifically sensitive to the variability of the spectral power on shorter time-scales. For instance, the spectral analysis does not distinguish between diffuse oscillatory activity and clearly visible temporally localized burst-like oscillations. The better capability of the event analysis for the characterization of specific temporally localized oscillatory activity became most apparent in the analysis of spindle activity. It emerged that some individuals showed an increased rate of sleep spindles after sleep deprivation (see MAP18 in Fig. 4), a fact that was not evident in the relative power spectra due to a decrease in background activity. While, by definition, rhythmic activity is evident in the oscillatory events, in the power density spectra it is revealed relative to the background only. Thus, a power law function was fitted to the spectrum to determine those components. We suggest that background power results mainly from slow waves or delta activity. Indeed, subtracting background power increased the correlations between spectral band power and event ratios in the corresponding bands (Table 1). The impact of the background power can also be seen in Fig. 4 (MAP18). It shows an individual where both the event ratios of fast delta events and sleep spindles increased after sleep deprivation. This is particularly evident from the event analysis, while in the power spectra this effect was masked by increasing spectral background power originating mainly from delta oscillations. The importance of subtracting background activity also became evident in recent studies investigating theta and alpha activity in the context of dream recall (Marzano et al., 2011). Spectral analysis reveals spectral power as a temporal average. As such, it is less specific than the event analysis. By contrast, the event analysis is sensitive to the fluctuations of the oscillatory activity. Thus, the two methods (spectral analysis and event analysis) could be viewed as complementary and should be used in combination (Fig. 1). One problem common to both methods is their inability to distinguish whether spindles become slower (a general decrease of the frequency of spindles; Fig. 3) or whether there is a relative increase in the occurrence of slower compared to faster sleep spindles. This problem, however, has been recognized in general and has not yet been solved (De Gennaro and Ferrara, 2003). According to our understanding of EEG oscillations as collective modes of neuronal networks, a state space or multivariate approach should be used to address the problem of dissociating whether an oscillatory event has changed or whether a new oscillation has emerged. In a state space model which takes network properties into account, events could be defined based on oscillatory modes of hidden states. Such an event would be reflected differently in different EEG derivations, therefore giving rise to distinct event topographies. We are working on such a method; however, technical and conceptual issues have to be resolved. Simply applying the algorithm used in ª 2013 European Sleep Research Society
261
this paper to multiple derivations would not be appropriate, because it would require additional criteria to determine whether events occurring in temporal proximity to one another across derivations constitute the same or different events. Regarding the oscillatory events in the delta band, our results confirm that a distinction between slower and faster delta oscillations may be meaningful. By dividing the delta band at 2 Hz we found an increased average frequency of slow delta events, while faster delta events became slower in SWS (Fig. 3) with increased sleep pressure. These results, however, are only preliminary because our method for the detection of the oscillatory events is not particularly well suited for the analysis of delta oscillations (refer also to the discussion of f = 0 Hz events in Olbrich and Achermann, 2005). Nevertheless, the increase of the average frequency of slow delta events is in agreement with the finding of shorter half-wave duration of slow oscillations after sleep deprivation (Bersagliere and Achermann, 2010) or comparing power spectra of local field potentials of early (high sleep pressure) and late (low sleep pressure) sleep in rats (Vyazovskiy et al., 2007). We want to emphasize, however, that the slow oscillations are not the only relevant oscillatory activity in the delta frequency band and that oscillations in the range between 2 and 4 Hz deserve closer attention. Steriade and Amzica (1998) discussed two additional sources of oscillations in the delta range: a thalamically generated clock-like oscillation and intrinsic oscillations of cortical neurones; to what extent the oscillatory events in the fast delta band reflect these oscillations needs to be investigated further. Our results demonstrate the usefulness of employing event based methods in addition to spectral analysis. In general, both approaches reveal similar results, but their combination facilitates interpretation of the distinct changes in sleep EEG induced by prolonged wakefulness. ACKNOWLEDGEMENTS We thank Dr Leila Tarokh for comments on the manuscript. The study was supported by Swiss National Science Foundation grants 320000-112674, 320030-130766 (PA) and 3100A0-107874 (HPL). CONFLICT OF INTEREST No conflicts of interest declared. REFERENCES Bersagliere, A. and Achermann, P. Slow oscillations in human nonrapid eye movement sleep electroencephalogram: effects of increased sleep pressure. J. Sleep Res., 2010, 19: 228–237. ly, A. A., Baumann, F., Brandeis, D., Strauch, I. and Lehmann, Borbe D. Sleep deprivation: effect on sleep stages and EEG power density in man. Clin. Neurophysiol., 1981, 51: 483–493. De Gennaro, L. and Ferrara, M. Sleep spindles: an overview. Sleep Med. Rev., 2003, 7: 423–440.
262
E. Olbrich et al.
Ferrara, M. and De Gennaro, L. Going local: insights from EEG and stereo-EEG studies of the human sleep–wake cycle. Curr. Top. Med. Chem., 2011, 11: 2423–2437. ly, A. A. Individual ‘fingerFinelli, L. A., Achermann, P. and Borbe prints’ in human sleep EEG topography. Neuropsychopharmacology, 2001a, 25: S57–S62. ly, A. A. and Achermann, P. Functional topograFinelli, L. A., Borbe phy of the human nonREM sleep electroencephalogram. Eur. J. Neurosci., 2001b, 13: 2282–2290. ly, A. A. Geering, B. A., Achermann, P., Eggimann, F. and Borbe Period-amplitude analysis and power spectral analysis: a comparison based on all-night sleep EEG recordings. J. Sleep Res., 1993, 2: 121–129. Geiger, A., Huber, R., Kurth, S., Ringli, M., Jenni, O. G. and Achermann, P. The sleep EEG as a marker of intellectual ability in school age children. Sleep, 2011, 34: 181–189. Iber, C., Ancoli-Israel, S., Chesson, A. and Quan, S. F. The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology and Technical Specifications, 1st edn. American Academy of Sleep Medicine, Westchester, IL, 2007. Knoblauch, V., Martens, W. L. J., Wirz-Justice, A. and Cajochen, C. Human sleep spindle characteristics after sleep deprivation. Clin. Neurophysiol., 2003, 114: 2258–2267. Ktonas, P. Y. and Gosalia, A. P. Spectral analysis vs. periodamplitude analysis of narrowband EEG activity: a comparison based on the sleep delta-frequency band. Sleep, 1981, 4: 193– 206. Ktonas, P. Y., Golemati, S., Xanthopoulos, P. et al. Time–frequency analysis methods to quantify the time-varying microstructure of sleep EEG spindles: possibility for dementia biomarkers? J. Neurosci. Methods, 2009, 185: 133–142. tey, J. V., To €nz, K. et al. Caffeine attenuates Landolt, H. P., Re waking and sleep electroencephalographic markers of sleep homeostasis in humans. Neuropsychopharmacology, 2004, 29: 1933–1939. Marzano, C., Ferrara, M., Curcio, G. and De Gennaro, L. The effects of sleep deprivation in humans: topographical electroencephalogram changes in non-rapid eye movement (NREM) sleep versus REM sleep. J. Sleep Res., 2010, 19: 260–268. Marzano, C., Ferrara, M., Mauro, F. et al. Recalling and forgetting dreams: theta and alpha oscillations during sleep predict subsequent dream recall. J. Neurosci., 2011, 31: 6674–6683.
Olbrich, E. and Achermann, P. Analysis of oscillatory patterns in the human sleep EEG using a novel detection algorithm. J. Sleep Res., 2005, 14: 337–346. Rechtschaffen, A. and Kales, A. A Manual of Standardized Terminology, Techniques and Scoring System for Sleep Stages of Human Subjects, 1st edn. National Institutes of Health, Bethesda, MD, 1968. tey, J. V., Adam, M., Gottselig, J. M. et al. Adenosinergic Re mechanisms contribute to individual differences in sleep deprivation-induced changes in neurobehavioral function and brain rhythmic activity. J. Neurosci., 2006, 26: 10472–10479. Steriade, M. and Amzica, F. Sleep oscillation, rhythmic K-complexes, and their paroxysmal developments. J. Sleep Res., 1998, 7: 30– 35. Timofeev, I. and Bazhenov, M. Mechanisms and biological role of thalamocortical oscillations. In: F. Colombus (Ed) Trends in Chronobiology Research. Nova Science Publishers Inc, Hauppauge, NY, 2005: 1–47. ly, A. A. and Tinguely, G., Finelli, L. A., Landolt, H. P., Borbe Achermann, P. Functional EEG topography in sleep and waking: state-dependent and state-independent features. NeuroImage, 2006, 32: 283–292. Vyazovskiy, V. V., Riedner, B. A., Cirelli, C. and Tononi, G. Sleep homeostasis and cortical synchronization: II. A local field potential study of sleep slow waves in the rat. Sleep, 2007, 30: 1631–1642. ly, A. A. Spindle Werth, E., Achermann, P., Dijk, D. J. and Borbe frequency activity in the sleep EEG: individual differences and topographical distribution. Electroencephalogr. Clin. Neurophysiol., 1997, 103: 535–542. Whitten, T. A., Hughes, A. M., Dickson, C. T. and Caplan, J. B. A better oscillation detection method robustly extracts EEG rhythms across brain state changes: the human alpha rhythm as a test case. NeuroImage, 2011, 54: 860–874.
SUPPORTING INFORMATION Additional Supporting Information may be found in the online version of this article: Data S1. Spectrograms, averaged spectra and oscillatory events of individual data sets.
ª 2013 European Sleep Research Society
J Sleep Res. (2014) 23, 255–262
Effect of prolonged wakefulness on electroencephalographic oscillatory activity during sleep E C K E H A R D O L B R I C H 1 , H A N S P E T E R L A N D O L T 2 , 3 , 4 and PETER ACHERMANN2,3,4 1 Max Planck Institute for Mathematics in the Sciences, Leipzig, Germany, 2Institute of Pharmacology and Toxicology, University of Zürich, Zürich, Switzerland, 3Zurich Center for Integrative Human Physiology, University of Zurich, Zurich, Switzerland and 4Neuroscience Center Zurich, University and ETH Zurich, Zurich, Switzerland
Keywords alpha oscillations, delta oscillations, oscillatory events, sleep deprivation, sleep electroencephalography, sleep spindles, spectral analysis Correspondence Dr Eckehard Olbrich, Max Planck Institute for Mathematics in the Sciences, Inselstrasse 22, D-04103 Leipzig, Germany. Tel.: +49-34-1-9959-568; fax: +49-341-9959-555; e-mail: [email protected] Accepted in revised form 23 November 2013; received 10 June 2013 DOI: 10.1111/jsr.12123
SUMMARY
The human sleep electroencephalogram (EEG) is characterized by the occurrence of distinct oscillatory events such as delta waves, sleep spindles and alpha activity. We applied a previously proposed algorithm for the detection of such events and investigated their incidence and frequency in baseline and recovery sleep after 40 h of sustained wakefulness in 27 healthy young subjects. The changes in oscillatory events induced by sleep deprivation were compared to the corresponding spectral changes. Both approaches revealed, on average, an increase in low frequency activity and a decrease in spindle activity after sleep deprivation. However, the increase of oscillatory events in the delta range and decrease in the sigma range occurred in a more restricted frequency range compared to spectral changes. The mean relative power spectra showed a significant increase in theta and alpha activity after sleep deprivation while, on average, the event analysis showed only a weak effect in the theta band. The reason for this discrepancy is that the spectral analysis does not distinguish between diffuse activity and clearly visible temporally localized oscillations, while the event analysis would detect only the latter. Additionally, only a few individuals clearly showed activity in the theta or alpha frequency bands. Conversely, event analysis revealed that some individuals showed an increased rate of sleep spindles after sleep deprivation, a fact that was not evident in the relative power spectra due to a decrease in background activity. The two methods complement each other and facilitate the interpretation of distinct changes induced by prolonged wakefulness in sleep EEG.
INTRODUCTION Oscillatory activity in electroencephalography (EEG) is considered as an emergent property of the thalamocortical system, with specific patterns and dominant frequencies of these oscillations depending on the functional state of the brain (Timofeev and Bazhenov, 2005). Typical patterns in EEG in conjunction with eye movements [electro-oculography (EOG)] and muscle tone [electromyography (EMG)] serve to discriminate non-rapid eye movement (NREM) sleep, rapid eye movement (REM) sleep and waking (Iber et al., 2007; Rechtschaffen and Kales, 1968). Classically, oscillatory activity during sleep is subdivided into delta ª 2013 European Sleep Research Society
(0.5–5 Hz), theta (5–8 Hz), alpha activity (8–11 Hz) and spindle frequency activity (11–16 Hz). EEG can be analysed in many ways, e.g. in the frequency domain by spectral analysis (decomposition of signals into its constituting frequency components) or in the time domain with period-amplitude analysis (information such as incidence and amplitude of waves; Geering et al., 1993; Ktonas and Gosalia, 1981). Time–frequency analysis methods such as wavelet analysis or matching pursuit combine elements of both domains (see e.g. Ktonas et al., 2009). A particular time–frequency analysis method based on autoregressive modelling of short (1 s) overlapping EEG segments was proposed by Olbrich and Achermann (2005) (see also
255
256
E. Olbrich et al.
Methods) to characterize transient oscillatory activity typical for EEG. In the recent past, the effect of sleep deprivation on human sleep EEG was studied mainly using spectral analysis. The main findings were the relative increase of spectral band power in the recovery night compared to the baseline night ly et al., 1981; Finelli et al., for frequencies ≤11 Hz (Borbe 2001a) and decreasing power in higher frequencies (Finelli et al., 2001a). The increase of spectral power in the low frequency bands was most prominent in frontal areas, while the decrease of power in the higher frequency bands appeared in more central and posterior areas (Finelli et al., 2001b; Marzano et al., 2010; Tinguely et al., 2006). Studies investigating the effect of sleep deprivation on specific sleep oscillations focused on sleep spindles (Knoblauch et al., 2003) and slow oscillations (Bersagliere and Achermann, 2010). Sleep deprivation resulted in increased spindle amplitude and a reduction in spindle density, while spindle duration was not affected (Knoblauch et al., 2003). Slow oscillations were redistributed in response to increased sleep pressure: the number of waves/min was reduced below 0.9 Hz and increased above 1.2 Hz (Bersagliere and Achermann, 2010). Our aim was to investigate how oscillatory events in sleep EEG (event analysis) are affected by increased sleep pressure, i.e. after 40 h of sustained wakefulness, and how these changes relate to the previously observed changes in the corresponding power density spectra. Furthermore, we were interested in interindividual variation in the response to sleep deprivation. METHODS Data The analyses were performed on existing data sets of 27 healthy young male participants of previous studies investigating the effects of sleep deprivation on EEG topography [n = 8 (Finelli et al., 2001b)] and of caffeine during prolonged wakefulness [n = 19, placebo condition (Landolt et al., 2004; tey et al., 2006)]. Polysomnographic recordings were Re obtained during an adaptation night, a subsequent baseline and a recovery night after 40 h of sustained wakefulness. Bedtime for all 3 nights was scheduled at 23 h (n = 12) or 24 h (n = 15). Sleep was limited to 8 h for the adaptation and baseline nights and to 12 h (n = 8) or 10.5 h (n = 19) for the recovery nights. Participants were instructed to abstain from alcohol and to adhere to regular bedtimes (8 h time in bed) for 3 days prior to the study, verified by ambulatory activity monitoring and sleep–wake diaries. The study protocols and experimental procedures were approved by the local ethics committees for research on human subjects and participants gave their written informed consent. The sleep EEG data of baseline and recovery sleep after 40 h of sustained wakefulness were analysed. The EEG signals were sampled at 128 Hz (for additional details see Finelli et al., 2001b; Landolt
tey et al., 2006). Sleep stages were scored et al., 2004; Re visually for 20-s epochs (C3A2 derivation), according to the criteria of Rechtschaffen and Kales (1968). The analysis of recovery sleep was restricted to the length of NREM sleep of the corresponding baseline. Event detection Oscillatory events in the sleep EEG were detected in derivation C3A2 using a previously published algorithm (Olbrich and Achermann, 2005). This algorithm is based on modelling overlapping 1-s segments of the EEG time–series by an autoregressive model of order 8 [AR(8)-model]. Thus, the EEG is modelled as a superposition of maximal 4 stochastically driven harmonic oscillators, with damping and frequency varying in time. More precisely, an AR (p) model corresponds to m oscillators and n relaxators with P = 2 m + n. Oscillatory events are detected whenever the damping constant at one or more frequencies is below a predefined threshold c ≤ 6.6 s 1. For a detailed description of the algorithm and its application to human baseline sleep EEG data see Olbrich and Achermann (2005). Detected events in the sleep EEG of a baseline night are shown in Fig. 1. Events were characterized by their time of occurrence, frequency and duration. Analysis of detected events In a first step, we compared the detected events with spectral analysis using the same frequency resolution, i.e. we determined histograms of the events with a frequency bin size of 0.25 Hz (Fig. 1, right panel). For further analysis, the events were grouped into four frequency bands: delta (0– 5 Hz), theta (5–8.5 Hz), alpha (8.5–10.5 Hz) and sigma bands (10.5–16 Hz). These four bands are commonly used and correspond to the peaks in the histogram of the event frequencies in Fig. 1. We subdivided the delta and sigma bands further into two sub-bands: slow delta 0 < f ≤ 2 Hz, fast delta 2 < f ≤ 5 Hz, slow sigma 10.5 < f ≤ 12 Hz and fast sigma 12 < f ≤ 16 Hz. A relevant fraction of the slow delta events were events with f = 0 Hz. While these events are technically non-oscillatory, they were taken into account if the instantaneous frequency during the event was larger than zero on more than one occasion (for details see Olbrich and Achermann, 2005). We distinguished slow and fast delta based on the distinct behaviour of slow (≤2 Hz) and fast delta activity during sleep deprivation and recovery sleep (reviewed in Ferrara and De Gennaro, 2011). The sigma band was divided at f = 12 Hz in order to study possible differences between slow and fast spindles (De Gennaro and Ferrara, 2003; Werth et al., 1997). For each frequency band we calculated the following event properties for Stage 2, slow wave sleep (SWS) and NREM sleep (Stages 1–4): mean event frequency (Hz) by averaging over the frequencies of the single events, event rate as the ª 2013 European Sleep Research Society
Sleep deprivation and sleep oscillations
257
25
20
f [Hz]
15
10
5
Figure 1. Top left: spectrogram and superimposed oscillatory events (black dots) of an individual baseline night. Spectra (derivation C3A2) are colour-coded on a logarithmic scale (0 dB = 1 lV2 Hz 1; 20 dB). Top right: –10 dB frequency distribution of oscillatory events. Bottom: hypnogram. MT: movement time; W: waking; REM: rapid eye movement sleep; 1–4: non-rapid eye movement (NREM) sleep Stages 1–4.
0 0
0
50
100
number of events divided by the time in the particular stage (min 1), mean event duration (s) by averaging the durations of the single events and the event ratio as the ratio between the sum of the durations of the events in this frequency band divided by the total time in the particular stage. The event ratio, combining the duration and the rate, may be a more robust measure of oscillatory activity than the event rate. Spectral analysis Power density spectra were calculated for consecutive 20-s epochs (FFT, Hanning window, average of five 4-s epochs; matched with sleep stages) resulting in a frequency resolution of 0.25 Hz. Artefacts were excluded by visual inspection and semiautomatically (moving average threshold) to exclude high- and low-frequency artefacts (Finelli et al., tey et al., 2006). 2001b; Re To measure rhythmic activity relative to the background of the power density spectrum, a power law function was fit to the spectrum in the 2–5-Hz and 16–25-Hz ranges, excluding the 5.25–15.75-Hz range (containing theta, alpha and spindle peaks) (for details see Geiger et al., 2011) and (Whitten et al., 2011) for a similar method. Activity in the theta, alpha and sigma range was determined as power in the corresponding range minus background power in the same frequency range. Statistics Differences in power density spectra and the distribution of event ratios between baseline and recovery sleep were compared by paired t-tests. Power was log-transformed prior to statistical testing and performed for 0.25-Hz bins. A change was only considered significant if at least three consecutive bins reached significance. Changes in event ª 2013 European Sleep Research Society
50
100
Number of events
MT W REM 1 2 3 4 150
200
250
300
350
400
450
t [min]
ratios and event frequency from baseline to recovery sleep were assessed by boxplots and Wilcoxon’s signed-rank tests. Furthermore, relationships between changes in event properties or events and band power were quantified by Spearman’s rank correlations. RESULTS Fig. 2 illustrates the comparison of the event analysis with the spectral analysis for the data averaged over the subjects. While the blue curves in Fig. 2 show the comparison for NREM sleep, we additionally analysed Stage 2 and SWS (green and red) in order to identify sleep stage-specific changes and to exclude effects of the increased fraction of SWS in recovery sleep. Sleep deprivation resulted in an increase of spectral power from delta to the alpha frequencies and a decrease in the sigma range in NREM sleep, Stage 2 and SWS (Fig. 2). Oscillatory events were increased in the delta range and decreased in the sigma range, i.e. in a more restricted frequency range compared to spectral changes. The peaks in the mean spectra corresponded to peaks in the distributions of the event ratios. The peaks were, however, not exactly at the same frequencies in the spectra and histograms; their positions may have differed by up to 0.5 Hz (Fig. 2). Comparison of Stage 2 sleep (Fig. 2, green lines) with SWS (Fig. 2, red lines) revealed that the main effect of sleep deprivation on the occurrence of oscillatory events was found for delta events in SWS, while in Stage 2 the largest differences in the event ratios were observed for events in the spindle frequency (sigma) range. Additionally, in the average relative power spectra an increase in the alpha band in SWS and in the theta band in Stage 2 was observed. While the latter change was also
258
E. Olbrich et al. Recovery
Baseline
rec/bl
2
10
1.6
10
1.4
Ratio
Power density [μV2 per 0.25 Hz]
2
1.2
0
0
10
10
1 0.8
rec−bl –3
x 10 0.015
0.015
NREM ST2 SWS
0.01
Δ Event ratio
Event ratio
6 0.01
0.005
0.005
4 2 0 –2
0
0
10
f [Hz]
20
0
0
10
20
f [Hz]
reflected by a small peak in the event ratio, the increase of relative alpha power was not reflected in the event analysis. Fig. 3 summarizes the mean changes of the event properties (event rates, durations, event ratios and event frequency) in response to sleep deprivation of different frequency bands. Mainly delta activity (d1, d2) and fast spindles (r2) were affected by prolonged wakefulness. Fast delta events (d2) occurred with a higher rate in Stage 2 and were of longer duration in Stage 2 and SWS, leading to a higher event ratio. Averaged over total NREM sleep, slow delta activity showed, on average, a higher event rate and ratio. This reflects mainly the larger amount of SWS in the recovery night, because the effect vanished in the stage-specific analysis. However, a small but significant increase of event duration resulting in a higher event ratio was found in Stage 2. Moreover, we found increase of event rate, duration and ratio for events in the theta band of Stage 2. This change is reflected more prominently in the relative power spectra (Fig. 2). Individual-level analysis revealed that individual variability in the theta and alpha frequency bands is high (see examples in Fig. 4 and Data S1) and that the changes resulted from a subset of individuals. It is interesting to note that the event frequency in the two delta bands seem to change in opposite directions after sleep deprivation: slow delta events (d1) became faster (higher frequency) while faster events (d2) appeared to become slower. These changes, however, depended to some degree on the frequency applied for subdividing into the two subbands. Furthermore, fast spindles (r2) had lower frequency and event rates in Stage 2, SWS and consequently also in NREM sleep. So far, we have presented results on averaged data over all subjects. However, interindividual variation is considerable, and the individual picture can be quite different from the
0
10
f [Hz]
20
Figure 2. Average power spectra and oscillatory events of baseline and recovery sleep (n = 27), non-rapid eye movement sleep (blue), Stage 2 (green) and slow wave sleep (red). Top: mean power spectrum of baseline (left), recovery sleep (middle) and ratio recovery/baseline (right). Bottom: event ratios of the baseline (left), recovery sleep (middle) and their difference (right). Bars at bottom of right panels indicate significant non-zero mean differences (P < 0.05) between recovery and baseline sleep.
average result. This is exemplified in Fig. 4 with data of three individuals (for data of all individuals see Data S1). Individual MAP16 (top) displays a pattern similar to the average one and can therefore be considered as a ‘typical’ pattern: increasing spectral power in the delta and alpha bands of recovery sleep is accompanied by a corresponding increase of the event ratio and decreasing power in the spindle range by a decrease in the event ratio. The changes in the alpha range appear to be weaker in the event analysis. This is due, however, to a different comparison of recovery and baseline sleep in the two analyses. For the spectra, the ratio recovery to baseline (percentage change) was determined, while for the event analysis the difference of the event ratio had to be calculated, as events did not occur in all frequency bins (event ratio of zero in baseline). Additionally, signatures of a shift in the spindle peak frequency towards lower frequencies and of fast delta events to higher frequencies were evident. The latter, however, could be seen only in the difference of the event ratios. Individual MAP18 (Fig. 4, middle) showed remarkable deviations from the average pattern. The event ratios of both fast delta events and sleep spindles increased after sleep deprivation. The higher incidence of sleep spindles in recovery sleep was evident only in the event analysis. Individual MAP07 (Fig. 4, bottom) represents an example of an individual with a strong increase in oscillatory events in the theta frequency band that was accompanied by peaks both in the spectra of baseline and recovery sleep as well as in the relative power spectrum. Note that for this individual the event analysis revealed a decrease of the event ratio of the fast delta events, while the relative power spectrum showed an increase in this frequency range. Furthermore, the alpha peak in the relative power spectrum was not accompanied by a corresponding peak in the event analysis. ª 2013 European Sleep Research Society
Sleep deprivation and sleep oscillations –1
Δ Evt rate [min ]
δ
1
5
Δ Duration [s] 5
∗
0
∗
∗
1
∗
0 –1
∗
∗
∗
Δ f [Hz] 0.2
0
0
–0.2
–0.2
–5
1 2
Δ Evt ratio 0.2
0
–5
δ
∗
0.02
0
0
–1
–0.02
∗
∗
∗
259
0.5
∗
∗
∗
0 –0.5 –3
x 10
θ
1
0.5
∗
0
1
0
∗
0 –1
–5
–0.5
–3
x 10
α
1 0 –1
1
σ
0
2
0
0
–0.5
∗
∗
∗
0.04
0 –1
∗
0.02
0.04
0
0
–0.02
–0.04
0.04
0
0
0
–2
–0.04
–0.04
NREM ST2 SWS
1
–5
0
–1
σ
5
0.5
1
2
0.5
–0.5
Because it is not feasible to show and discuss all individual patterns, we mention further that some individuals showed a theta peak in the relative power spectrum but no or only a few oscillatory events in the theta frequency range and therefore no peak in the difference of the event ratios. In order to elucidate further the relationship between event and spectral analysis we also studied correlations between spectral band power and event ratios in the corresponding bands in both baseline and recovery sleep. Spectral band power in all frequency bands except the delta bands was estimated in two ways: first in the ‘usual’ way, by integrating spectral power density in the corresponding frequency band, and secondly by additionally subtracting the background power. Spectral band power and event ratios in the corresponding frequency band were correlated in all conditions and frequency bands except for the fast delta band 2–5 Hz (Table 1). In this band the event ratio and the spectral band power were correlated only in Stage 2 of recovery sleep. In general, correlation coefficients were higher in the alpha and sigma bands than in the delta and theta bands. Subtracting the background power increased the strength of the correlation. The most striking increase was observed in SWS, supporting our view that to a large degree the background power reflects SWA. ª 2013 European Sleep Research Society
∗
5
0
–1
Figure 3. Boxplots of differences in events rate (D evt rate), duration, ratio (D evt ratio) and frequency (D f) between recovery and baseline sleep in six different frequency bands: d1…0–2 Hz, d2…2–5 Hz, h…5– 8.5 Hz, a…8.5–10.5 Hz, r1…10.5–12 Hz, r2…12–16 Hz. Boxes represent the lower quartile, median and upper quartile. The notch indicates the 95% confidence interval of the median. Maximum whisker length is 1.5 times of the interquartile range. They extend to the most extreme data value that is not an outlier. Outliers are indicated by dots. The black asterisks indicate significant (P < 0.05) non-zero median differences. NREM: non-rapid eye movement sleep (blue); ST2: Stage 2 (green); SWS: slow wave sleep (red).
∗
NREM ST2 SWS
∗
∗
∗
0.2
∗
∗
∗
0 –0.2
NREM ST2 SWS
NREM ST2 SWS
DISCUSSION A previously developed method for the analysis of sleep oscillations was applied to study the effects of prolonged wakefulness on human sleep EEG and the outcome was compared with the results of spectral analysis of the same data. We performed the analysis separately for Stage 2 and SWS in order to disentangle the changes occurring within NREM sleep from the effect resulting of increased SWS in recovery sleep. The global picture was very similar with both methods: on average, an increase in low frequency activity and a decrease of spindle activity was observed after sleep deprivation. Event ratios and spectral band power as well as the corresponding changes resulting from sleep deprivation were highly correlated. The peaks in the histograms of the event frequencies corresponded roughly to the peaks in the relative power spectra; however, the relative strength of the components was different. Fast sleep spindles (r2) were reduced after sleep deprivation and their frequency became slower (Fig. 3). This slowing of fast spindle activity resulted in a shift of the spindle peak in the power density spectra. This is most evident in the relative spectrum of Stage 2 (Fig. 2, green line), with increased activity at lower spindle frequencies followed by a sharp drop with decreased activity at higher spindle frequencies.
E. Olbrich et al. Baseline
Recovery
Ratio
0
10
1
0.03
0.03
0.02
0.02
0.01
0.01
0
0
–0.01
2
1.4
10
Difference
Event ratio
1.5
0
10
2
2 [μV per 0.25 Hz]
REC vs. BL 2
2
10 MAP16
2
[μV per 0.25 Hz]
2
10
0.01 0
10
1.2
MAP18 0
1
0
10
Ratio
260
10
0.8 –3
0.015 0.01 0.005
0
0
5 0
2
[μV2 per 0.25 Hz]
Difference
0.01 0.005
10
10
2
2.5
0
1.5
10
2
MAP07 0
10
10
Ratio
Event ratio
x 10 0.015
1
0.01 0.02
Difference
Event ratio
0.5 0.02 0 0.01
0.01 –0.01
0
0
10
20
0
0
10
f [Hz]
20
0
f [Hz]
10
20
f [Hz]
Figure 4. Average power spectra and oscillatory events of baseline and recovery sleep in three individuals separated by horizontal lines: non-rapid eye movement sleep (blue), Stage 2 (green) and slow wave sleep (red). For details see Fig. 2.
Table 1 Correlations between the event ratios and the spectral band power P and the spectral band power after subtracting background activity Pdiff2bg Baseline
Recovery
S2 P d d1 d2 h a r r1 r2
0.43 0.48 0.19** 0.33* 0.77 0.65 0.66 0.65
SWS Pdiff2bg
P
0.48 0.83 0.74 0.71 0.73
0.48 0.50 0.16** 0.44 0.76 0.61 0.80 0.53
Baseline–recovery
S2 Pdiff2bg
P
0.65 0.90 0.81 0.79 0.86
0.32* 0.37* 0.51 0.57 0.86 0.67 0.60 0.58
SWS Pdiff2bg
P
0.66 0.90 0.77 0.62 0.66
0.50 0.54 0.26** 0.40 0.61 0.70 0.84 0.50
S2 Pdiff2bg
P
0.66 0.89 0.84 0.78 0.87
0.45 0.53 0.0** 0.86 0.87 0.29** 0.15** 0.49
SWS Pdiff2bg
P
Pdiff2bg
0.89 0.90 0.42 0.33* 0.56
0.47 0.49 0.03** 0.51 0.58 0.02** 0.30** 0.00**
0.53 0.64 0.32** 0.48 0.30**
SWS, slow wave sleep; S2, stage 2. Analysis was performed for baseline and recovery sleep, as well as for the changes induced by sleep deprivation [difference in event ratios correlated with relative power (ratio recovery to baseline sleep)]. Background activity was not subtracted in the d range. All correlations were significant with P < 0.05, except *0.05 < P < 0.1 and **P > 0.1.
The effect of sleep deprivation on alpha and theta activity appeared strongly in the relative power spectrum, but only weakly in the event analysis. Reviewing the data on an individual basis revealed that some individuals showed a
peak only in the relative power spectrum but not in the difference of the event ratios (see e.g. Fig. 4, MAP07). To understand these differences, one has to take into account that the spectral power is a temporal average on 20-s ª 2013 European Sleep Research Society
Sleep deprivation and sleep oscillations segments, while the event analysis works on 1-s segments and picks up only those episodes with strong oscillatory activity. Therefore, the event analysis is specifically sensitive to the variability of the spectral power on shorter time-scales. For instance, the spectral analysis does not distinguish between diffuse oscillatory activity and clearly visible temporally localized burst-like oscillations. The better capability of the event analysis for the characterization of specific temporally localized oscillatory activity became most apparent in the analysis of spindle activity. It emerged that some individuals showed an increased rate of sleep spindles after sleep deprivation (see MAP18 in Fig. 4), a fact that was not evident in the relative power spectra due to a decrease in background activity. While, by definition, rhythmic activity is evident in the oscillatory events, in the power density spectra it is revealed relative to the background only. Thus, a power law function was fitted to the spectrum to determine those components. We suggest that background power results mainly from slow waves or delta activity. Indeed, subtracting background power increased the correlations between spectral band power and event ratios in the corresponding bands (Table 1). The impact of the background power can also be seen in Fig. 4 (MAP18). It shows an individual where both the event ratios of fast delta events and sleep spindles increased after sleep deprivation. This is particularly evident from the event analysis, while in the power spectra this effect was masked by increasing spectral background power originating mainly from delta oscillations. The importance of subtracting background activity also became evident in recent studies investigating theta and alpha activity in the context of dream recall (Marzano et al., 2011). Spectral analysis reveals spectral power as a temporal average. As such, it is less specific than the event analysis. By contrast, the event analysis is sensitive to the fluctuations of the oscillatory activity. Thus, the two methods (spectral analysis and event analysis) could be viewed as complementary and should be used in combination (Fig. 1). One problem common to both methods is their inability to distinguish whether spindles become slower (a general decrease of the frequency of spindles; Fig. 3) or whether there is a relative increase in the occurrence of slower compared to faster sleep spindles. This problem, however, has been recognized in general and has not yet been solved (De Gennaro and Ferrara, 2003). According to our understanding of EEG oscillations as collective modes of neuronal networks, a state space or multivariate approach should be used to address the problem of dissociating whether an oscillatory event has changed or whether a new oscillation has emerged. In a state space model which takes network properties into account, events could be defined based on oscillatory modes of hidden states. Such an event would be reflected differently in different EEG derivations, therefore giving rise to distinct event topographies. We are working on such a method; however, technical and conceptual issues have to be resolved. Simply applying the algorithm used in ª 2013 European Sleep Research Society
261
this paper to multiple derivations would not be appropriate, because it would require additional criteria to determine whether events occurring in temporal proximity to one another across derivations constitute the same or different events. Regarding the oscillatory events in the delta band, our results confirm that a distinction between slower and faster delta oscillations may be meaningful. By dividing the delta band at 2 Hz we found an increased average frequency of slow delta events, while faster delta events became slower in SWS (Fig. 3) with increased sleep pressure. These results, however, are only preliminary because our method for the detection of the oscillatory events is not particularly well suited for the analysis of delta oscillations (refer also to the discussion of f = 0 Hz events in Olbrich and Achermann, 2005). Nevertheless, the increase of the average frequency of slow delta events is in agreement with the finding of shorter half-wave duration of slow oscillations after sleep deprivation (Bersagliere and Achermann, 2010) or comparing power spectra of local field potentials of early (high sleep pressure) and late (low sleep pressure) sleep in rats (Vyazovskiy et al., 2007). We want to emphasize, however, that the slow oscillations are not the only relevant oscillatory activity in the delta frequency band and that oscillations in the range between 2 and 4 Hz deserve closer attention. Steriade and Amzica (1998) discussed two additional sources of oscillations in the delta range: a thalamically generated clock-like oscillation and intrinsic oscillations of cortical neurones; to what extent the oscillatory events in the fast delta band reflect these oscillations needs to be investigated further. Our results demonstrate the usefulness of employing event based methods in addition to spectral analysis. In general, both approaches reveal similar results, but their combination facilitates interpretation of the distinct changes in sleep EEG induced by prolonged wakefulness. ACKNOWLEDGEMENTS We thank Dr Leila Tarokh for comments on the manuscript. The study was supported by Swiss National Science Foundation grants 320000-112674, 320030-130766 (PA) and 3100A0-107874 (HPL). CONFLICT OF INTEREST No conflicts of interest declared. REFERENCES Bersagliere, A. and Achermann, P. Slow oscillations in human nonrapid eye movement sleep electroencephalogram: effects of increased sleep pressure. J. Sleep Res., 2010, 19: 228–237. ly, A. A., Baumann, F., Brandeis, D., Strauch, I. and Lehmann, Borbe D. Sleep deprivation: effect on sleep stages and EEG power density in man. Clin. Neurophysiol., 1981, 51: 483–493. De Gennaro, L. and Ferrara, M. Sleep spindles: an overview. Sleep Med. Rev., 2003, 7: 423–440.
262
E. Olbrich et al.
Ferrara, M. and De Gennaro, L. Going local: insights from EEG and stereo-EEG studies of the human sleep–wake cycle. Curr. Top. Med. Chem., 2011, 11: 2423–2437. ly, A. A. Individual ‘fingerFinelli, L. A., Achermann, P. and Borbe prints’ in human sleep EEG topography. Neuropsychopharmacology, 2001a, 25: S57–S62. ly, A. A. and Achermann, P. Functional topograFinelli, L. A., Borbe phy of the human nonREM sleep electroencephalogram. Eur. J. Neurosci., 2001b, 13: 2282–2290. ly, A. A. Geering, B. A., Achermann, P., Eggimann, F. and Borbe Period-amplitude analysis and power spectral analysis: a comparison based on all-night sleep EEG recordings. J. Sleep Res., 1993, 2: 121–129. Geiger, A., Huber, R., Kurth, S., Ringli, M., Jenni, O. G. and Achermann, P. The sleep EEG as a marker of intellectual ability in school age children. Sleep, 2011, 34: 181–189. Iber, C., Ancoli-Israel, S., Chesson, A. and Quan, S. F. The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology and Technical Specifications, 1st edn. American Academy of Sleep Medicine, Westchester, IL, 2007. Knoblauch, V., Martens, W. L. J., Wirz-Justice, A. and Cajochen, C. Human sleep spindle characteristics after sleep deprivation. Clin. Neurophysiol., 2003, 114: 2258–2267. Ktonas, P. Y. and Gosalia, A. P. Spectral analysis vs. periodamplitude analysis of narrowband EEG activity: a comparison based on the sleep delta-frequency band. Sleep, 1981, 4: 193– 206. Ktonas, P. Y., Golemati, S., Xanthopoulos, P. et al. Time–frequency analysis methods to quantify the time-varying microstructure of sleep EEG spindles: possibility for dementia biomarkers? J. Neurosci. Methods, 2009, 185: 133–142. tey, J. V., To €nz, K. et al. Caffeine attenuates Landolt, H. P., Re waking and sleep electroencephalographic markers of sleep homeostasis in humans. Neuropsychopharmacology, 2004, 29: 1933–1939. Marzano, C., Ferrara, M., Curcio, G. and De Gennaro, L. The effects of sleep deprivation in humans: topographical electroencephalogram changes in non-rapid eye movement (NREM) sleep versus REM sleep. J. Sleep Res., 2010, 19: 260–268. Marzano, C., Ferrara, M., Mauro, F. et al. Recalling and forgetting dreams: theta and alpha oscillations during sleep predict subsequent dream recall. J. Neurosci., 2011, 31: 6674–6683.
Olbrich, E. and Achermann, P. Analysis of oscillatory patterns in the human sleep EEG using a novel detection algorithm. J. Sleep Res., 2005, 14: 337–346. Rechtschaffen, A. and Kales, A. A Manual of Standardized Terminology, Techniques and Scoring System for Sleep Stages of Human Subjects, 1st edn. National Institutes of Health, Bethesda, MD, 1968. tey, J. V., Adam, M., Gottselig, J. M. et al. Adenosinergic Re mechanisms contribute to individual differences in sleep deprivation-induced changes in neurobehavioral function and brain rhythmic activity. J. Neurosci., 2006, 26: 10472–10479. Steriade, M. and Amzica, F. Sleep oscillation, rhythmic K-complexes, and their paroxysmal developments. J. Sleep Res., 1998, 7: 30– 35. Timofeev, I. and Bazhenov, M. Mechanisms and biological role of thalamocortical oscillations. In: F. Colombus (Ed) Trends in Chronobiology Research. Nova Science Publishers Inc, Hauppauge, NY, 2005: 1–47. ly, A. A. and Tinguely, G., Finelli, L. A., Landolt, H. P., Borbe Achermann, P. Functional EEG topography in sleep and waking: state-dependent and state-independent features. NeuroImage, 2006, 32: 283–292. Vyazovskiy, V. V., Riedner, B. A., Cirelli, C. and Tononi, G. Sleep homeostasis and cortical synchronization: II. A local field potential study of sleep slow waves in the rat. Sleep, 2007, 30: 1631–1642. ly, A. A. Spindle Werth, E., Achermann, P., Dijk, D. J. and Borbe frequency activity in the sleep EEG: individual differences and topographical distribution. Electroencephalogr. Clin. Neurophysiol., 1997, 103: 535–542. Whitten, T. A., Hughes, A. M., Dickson, C. T. and Caplan, J. B. A better oscillation detection method robustly extracts EEG rhythms across brain state changes: the human alpha rhythm as a test case. NeuroImage, 2011, 54: 860–874.
SUPPORTING INFORMATION Additional Supporting Information may be found in the online version of this article: Data S1. Spectrograms, averaged spectra and oscillatory events of individual data sets.
ª 2013 European Sleep Research Society
