Paper 28 Disc

J. Sleep Res. (1998) 7, Suppl. 1, 3641

Temporal evolution of coherence and power in the human sleep electroencephalogram

PETER ACHERMANN and ALEXANDER A. BORBE LY
Institute of Pharmacology, University of Zurich, Zurich, Switzerland Accepted ???????

SUMMARY

Coherence analysis of the human sleep electroencephalogram (EEG) was used to investigate relations between brain regions. In all-night EEG recordings from eight young subjects, the temporal evolution of power and coherence spectra within and between cerebral hemispheres was investigated from bipolar derivations along the antero-posterior axis. Distinct peaks in the power and coherence spectra were present in NREM sleep but not in REM sleep. They were situated in the frequency range of sleep spindles (1314 Hz), alpha band (910 Hz) and low delta band (12 Hz). Whereas the peaks coincided in the power and coherence spectra, a dissociation of their temporal evolution was observed. In the low delta band, only power but not coherence showed a decline across successive NREM sleep episodes. Moreover, power increased gradually in the first part of a NREM sleep episode, whereas coherence showed a rapid rise. The results indicate that the intrahemispheric and interhemispheric coherence of EEG activity attains readily a high level in NREM sleep and is largely independent of the signal amplitude.
KEYWORDS alpha activity, coherence, delta activity, human sleep, NREM sleep, REM sleep, sleep spindles

Ahed Bhed Ched Dhed Ref marker Fig marker Table marker Ref end
Ref start

INTRODUCTION Slow-wave activity (SWA, power in the 0.754.5 Hz range) and spindle frequency activity (SFA, power in the 1215 Hz range) are markers of processes underlying sleep regulation. Thus SWA is typically high in the first NREM sleep episode and declines over subsequent episodes and its initial level is determined by the duration of prior waking and sleep (Borbely et al. 1981; Dijk et al. 1990a,b; Werth et al. 1996). SWA was the principal variable for delineating the time course of the homeostatic process S in the two-process model of sleep
Figure 1. Intrahemispheric and interhemispheric coherence spectra in NREM sleep plotted in each frequency bin as a ratio of the corresponding REM sleep value. Mean values from eight subjects (the ratios were calculated for individuals and then averaged). Middle diagram indicates the pairs of bipolar derivations from which the spectra were calculated. Black bars above the abscissae indicate for each bin significant differences between NREM sleep and REM sleep (only P 5 0.05 is shown; 1-p is plotted on a linear scale; the P = 0.01 level is indicated by a horizontal line and P = 0.05 corresponds to the abscissa; paired two-tailed t-tests on Fisher's z-transformed coherence values). Correspondence: Dr P. Achermann, Institute of Pharmacology, University of Zurich, Winterthurerstr. 190, CH-8057 Zurich, Switzer land. Tel.: +41 1 635 5954; Fax: +41 1 635 5707; E-mail: acherman@pharma.unizh.ch

regulation (Borbely 1982; Daan et al. 1984; Borbely and Achermann, 1992; Achermann et al. 1993). SFA contrasts with SWA, as it is determined by both homeostatic and circadian factors (Dijk and Czeisler 1995; Aeschbach et al. 1997).

36

#1998 European Sleep Research Society

Paper 28 Disc

Temporal evolution of coherence and power in the sleep EEG

37

Figure 2. Temporal evolution of coherence and power. Left panels: Color coded coherence and power spectra are plotted as a function of time. The values are expressed in each frequency bin relative to the mean REM sleep value of the entire night. NREM sleep episodes were subdivided into 20 equal intervals, REM sleep into 5 equal intervals and the time between lights-off and sleep onset was represented as a separate interval (Werth et al. 1997). Data were aligned with respect to sleep onset (cycle 1) or with respect to the first occurrence of stage 2 after a REM sleep episode (cycles 2 and 3). REM sleep is indicated by horizontal black bars at the bottom. Right panels: Average spectra based on the first 3 cycles. Note, that power is plotted on a logarithmic scale. Averages of eight subjects.

Both SWA and SFA can be related to changes occurring at the cellular level. The gradual hyperpolarization of thalamocortical neurons which occurs with the progression from

waking to deep NREM sleep (Hirsch et al. 1983), is associated with fluctuations in the membrane potential which are in the frequency range of the sleep EEG (Steriade et al. 1993a, 1994).

#1998 European Sleep Research Society, J. Sleep Res., 7 (Suppl. 1), 3641.

Paper 28 Disc

38

P. Achermann and A.A. Borbely

epochs) were computed for the bipolar derivations F3C3, P3O1, F4C4, P4O2, and coherence spectra (20-s epochs; squared coherence) within and between hemispheres were calculated. The frequency resolution was 0.25 Hz and frequencies up to 50 Hz were analyzed. The two lowest frequency bins (0.25 and 0.5 Hz) were excluded from the analysis because of numerous lowfrequency artifacts (due to sweating, etc.). Before computing averages, Fisher's z-transformation was applied to the coherence values (square root of squared coherence values). Retransformed and squared coherence spectra were used for further processing. Data analysis was performed with the software package MATLAB and its signal analysis and statistics toolbox (The Math Works, Inc. MA, USA). For a more detailed description of methodological aspects see Achermann and Borbely (1998). RESULTS Relative coherence spectra in NREM sleep To examine the topographic aspects of sleep state dependent differences, coherence in NREM sleep was expressed relative to REM sleep (Fig. 1). Significant state dependent differences were present mainly in the low delta, alpha and spindle frequency range. The spectra exhibited marked regional differences. The spindle peak was largest in the intrahemispheric spectra, the alpha peak in the anterior interhemispheric spectrum. Posterior interhemispheric coherence differed little between NREM sleep and REM sleep. In the 1.752.5 Hz range, the REM sleep values exceeded those of NREM sleep. Coherence in the high frequency range was higher in REM sleep than in NREM sleep. Bins with significant differences were scattered and infrequent, and more numerous in the interhemispheric spectra. Temporal changes of coherence and power The left panels of Figure 2 represent synoptically the temporal changes of coherence and power in the first three NREMREM sleep cycles. The corresponding average spectra of the first three cycles are shown in the right panels. The data are expressed relative to REM sleep. Relative coherence showed a prominent state specific modulation. Values in NREM sleep exceeded those in REM sleep in the low delta, alpha and sigma bands (warm colors). Increased alpha values were limited to the first NREM sleep episode in the intrahemispheric coherence spectrum, whereas they were present in all three episodes in the anterior interhemispheric coherence spectrum. In the power spectrum, the NREM sleep values exceeded those of REM sleep in the frequency range of 0.7515.75 Hz (warm colors), whereas the inverse relationship prevailed above 18.25 Hz (dark blue). The modulation of power by the NREM-REM sleep cycle is evident. This modulation encompassed almost the entire frequency range, whereas in the

Thus, depending on the level of hyperpolarization, oscillations occur first in the range of sleep spindles, and then in the delta range (0.754.5 Hz). These results indicate that the progression from `spindle sleep' (stage 2) to slow wave sleep (SWS, stages 3 and 4), as well as the partly inverse relationship between EEG power in the spindle and slow-wave ranges (Aeschbach and Borbely 1993) have a counterpart at the level of thalamocortical neurons. In addition to the oscillations in the delta band, fluctuations at the cellular level at frequencies below 1 Hz were shown to have an equivalent in the human sleep EEG (Achermann and Borbely 1997; Amzica and Steriade 1997). Coherence analysis can be applied for studying the functional connectivity between brain regions. Our analysis is based on the local voltage gradients obtained by bipolar derivations which provide an estimate of the strength of the underlying local generators. In a recent study, we determined coherence during sleep within and between hemispheres and demonstrated the presence of distinct peaks in coherence in NREM sleep (Achermann and Borbely 1998). There were both sleep state-dependent and sleep state-independent regional differences. The analysis was restricted to all-night mean values of the sleep states. Presented here are the results on the temporal evolution of coherence and power spectra in the course of the night. METHODS Subjects and data recording Nocturnal sleep (sleep episode: 23.0007.00 h) of 8 healthy young right-handed men (mean age 23.5 years, range 2226 years) was recorded after an adaptation night. A preceding screening night served to exclude sleep apnea and nocturnal myoclonus. The subjects reported to be in good health and free of sleep complaints. They were instructed to abstain from caffeine and alcohol, and also to keep a regular sleep-waking schedule for at least 7 days prior to the study. The EEG (F3, P3, O1, F4, C4, P4, O2, and A2, all against C3), the submental electromyogram (EMG), the electrooculogram (EOG, differential recording), and the electrocardiogram were recorded by a polygraphic amplifier (PSA 24, Braintronics, The Netherlands). Analog signals were conditioned (high-pass filter 73 dB at 0.16 Hz; low-pass filter 73 dB at 102 Hz, 5740 dB at 256 Hz; notch filter at 50 Hz), sampled at 512 Hz, digitally filtered (EEG: low-pass FIR filter, 73 dB at 49 Hz; EMG: band-pass FIR filter, 73 dB points 15.6 and 54 Hz) and stored with a resolution of 128 Hz. Signal analysis Sleep stages were visually scored for 20-s epochs (C3A2 derivation) according to the criteria of Rechtschaffen and Kales (1968). Only artifact-free 20-s epochs were used for further analysis. To avoid spurious coherence values, REM sleep epochs with REM were excluded. Power spectra of consecutive 20-s epochs (FFT routine, Hanning window, averages of five 4-s

#1998 European Sleep Research Society, J. Sleep Res., 7 (Suppl. 1), 3641.

Paper 28 Disc

Temporal evolution of coherence and power in the sleep EEG

39

Figure 3. Temporal evolution of interhemispheric and intrahemispheric coherence (curves) and power (gray areas) in 3 frequency bands. Episodes were subdivided as in Figure 2. Power and coherence: relative values with respect to REM sleep (all-night mean). REM sleep is indicated by horizontal black bars on the bottom. Low delta: 0.751.5 Hz; alpha: 8.5 11.25 Hz; sigma: 13.515.25 Hz.

Figure 4. Time course of relative interhemispheric coherence (F3C3-F4C4; open circles) and power (F3C3; filled circles) in the low delta band (0.75 1.5 Hz). The data are plotted across the first three NREM sleep episodes (left) and for the first 30 min of the first NREM sleep episode. The values are expressed relative to REM sleep. The time course across the first three NREM sleep episodes was based on mean values per episode (1 SEM; n = 8). Horizontal error bars (+ 1 SEM) indicate the variability in the timing of episode midpoints. A linear regression analysis was performed with the pooled data of eight subjects (episode mean values vs. episode midpoint). Power showed a significant linear decline (P 5 0.005; r2 = 0.31). No significant trend was present for coherence (P 4 0.7; r2 = 0.005). The time course within the first NREM sleep episode was calculated for 1-min intervals (mean +1 SEM; n = 8).

coherence spectrum it was present only in three frequency bands. In the average spectrum of the first three cycles the peaks in the low delta, alpha and sigma range are evident. Their amplitude were smaller at the posterior sites (data not shown). Note, that the power spectrum is smoother than the coherence spectra. Dynamics of coherence and power in selected frequency bands To examine whether the temporal evolution of coherence and power differ, three frequency bands were analyzed. The bands

were selected on the basis of significant differences between NREM sleep and REM sleep (Fig. 1). The low delta band (0.751.5 Hz) and the sigma band (13.515.25 Hz) encompasses the significant bins common in the intrahemispheric and anterior interhemispheric spectra, and the alpha band (8.5 11.25 Hz) corresponds to the significant bins in the anterior interhemispheric spectrum. In Figure 3 the variations of coherence (curve) and power (gray area) for the three frequency bands are superimposed. The modulation by the NREMREM sleep cycle was similar for coherence and power. However, a discrepancy was evident

#1998 European Sleep Research Society, J. Sleep Res., 7 (Suppl. 1), 3641.

Paper 28 Disc

40

P. Achermann and A.A. Borbely

waves above and below 1 Hz (Achermann and Borbely, 1997). Whereas the former showed a decline from the first to the second episode, this was not the case for the latter. The separate characteristics for slow waves above and below 1 Hz had been first observed in animal studies (Steriade et al. 1993b, c, d, 1996), and then described also for human sleep (Achermann and Borbely, 1997; Amzica and Steriade, 1997). The increased coherence in the low delta band in the present study is consistent with the observation in the cat that a widespread synchronization of slow oscillations occurs in the cortex (Amzica and Steriade, 1995). Another major finding was the dissociation in the temporal evolution between coherence and power. The typical decline of low-frequency power across successive NREM sleep episodes (Borbely et al. 1981; Dijk et al. 1990a) was not present for coherence, and the rise rate in the first part of NREM sleep episodes was higher for coherence than for power. In theory, coherence is independent of the amplitudes of the contributing signals. However, this is not necessarily the case for physiological signals. Our observation shows that the coherence in the low delta range is to a large extent independent of their amplitude. If the gradual build-up of slow-wave power within a NREM sleep episode is due to the progressive increase in the number of thalamocortical neurons engaged in synchronized activity, the intrahemispheric and interhemispheric coherence is already present at the outset. Thus while the intraepisodic build-up rate of SWA (i.e. power) and its global decline during sleep may reflect the level of `sleep pressure', the on-off type of coherence changes seems to reflect a switch between NREM and REM sleep `modes'. It will be interesting in this context to explore the changes induced by sleep deprivation. A pronounced feature was the coherence peak in the alpha band which was largest in the interhemispheric spectrum of the anterior derivations (Fig. 1) and showed a pronounced modulation by the NREM-REM sleep cycle. The close relationship between alpha and delta waves is evident from the alpha-delta pattern which is a well known feature of SWS (for an overview see Pivik and Harman, 1995). Posterior interhemispheric coherence differed little between NREM sleep and REM sleep (Fig. 1). Our previous analysis had shown that the typical coherence peaks are attenuated at the posterior sites (Achermann and Borbely, 1998). These results point to differences in functional connectivity between anterior and posterior regions of the hemispheres. In conclusion, coherence analysis is a powerful tool for analyzing state-related changes and their temporal evolution in the sleep EEG and provides insights into the functional connectivity of brain regions. ACKNOWLEDGMENTS We thank Drs Dietrich Lehmann and Irene Tobler for comments on the manuscript. The study was supported by the Swiss National Science Foundation grant 3100042500.94 and the Human Frontiers Science Program grant RG-81/96.

in the low delta band where the decline in power over successive NREM sleep episodes was not paralleled by a decline in coherence. Regression analysis confirmed that power in the low delta band declined significantly from episode 13 whereas no significant decline was present for interhemispheric anterior coherence (Fig. 4, left panel). The same result was obtained for intrahemispheric coherence (data not shown). However, a regression analysis of z-transformed values of power and coherence revealed an overlap in the 95% confidence intervals of the slopes. A further difference was seen in the intraepisodic dynamics of low delta activity. There was a typical gradual build-up of power in the first part of NREM sleep episodes (Achermann and Borbely, 1987) and a more rapid rise of coherence. The rise rate of the two variables was estimated by the time required to attain the average level of the last 10-min interval. This interval was shorter for coherence than for power (episode 1: 5.67 + 1.86 [SEM] min, n = 6 vs. 21.0 + 0.82 min, n = 8; due to missing values in the early part of the episode, rise time of coherence could not be reliably estimated in two subjects but were shorter, P 5 0.001; episode 2: 7.5 + 3.18 vs. 21.0 + 1.74, n = 8, P 5 0.05; episode 3: 9.1 + 2.33 vs. 21.0 + 1.96, P 5 0.005). DISCUSSION Prominent peaks in NREM sleep coherence spectra were observed in the low delta, alpha and sigma band. Activity in these bands was modulated by the NREM-REM sleep cycle (for an early report in an individual see Dumermuth et al. 1983). The peaks in the coherence and power spectra coincided. However, the modulation of power by the sleep cycles was present in much broader frequency ranges than for coherence. The most prominent peak in coherence was observed in the frequency range of sleep spindles. It dominated the intrahemispheric NREM sleep spectra and was distinct also in the anterior interhemispheric spectrum. There was no clear dissociation in the temporal evolution of coherence and power. This supports the results from animal studies that sleep spindles occur synchronously over large cortical areas as a result of intrinsic properties and connectivity patterns of thalamic neurons (Contreras et al. 1996). Global coherence of spindle activity was reported to result from corticothalamic projections and not from horizontal intracortical connections. Elevated levels of coherence at the low end of the spectra encompassed only the lowest bins in the delta band (Fig. 1), whereas low-frequency power is known to exhibit state-related variations over a much larger frequency range (see Achermann and Borbely 1997, Fig. 1). This observation could indicate that the slow waves with frequencies of 1.5 Hz and lower have distinct functional characteristics from slow waves above this range. In support of this interpretation, changes of power over the first two NREM sleep episodes differed between slow

#1998 European Sleep Research Society, J. Sleep Res., 7 (Suppl. 1), 3641.

Paper 28 Disc

Temporal evolution of coherence and power in the sleep EEG

REFERENCES
Achermann, P. and Borbely, A. A. Dynamics of EEG slow wave activity during physiological sleep and after administration of benzodiazepine hypnotics. Hum. Neurobiol., 1987, 6: 203210. Achermann, P. and Borbely, A. A. Low-frequency (51Hz) oscillations in the human sleep EEG. Neuroscience, 1997, 81: 213222. Achermann, P. and Borbely, A. A. Coherence analysis of the human sleep EEG. Neuroscience, 1998, 85: 11951208. Achermann, P., Dijk, D.J., Brunner, D.P., and Borbely, A. A. A model of human sleep homeostasis based on EEG slow-wave activity: quantitative comparison of data and simulations. Brain Res. Bull., 1993, 31: 97113. Aeschbach, D. and Borbely, A. A. All-night dynamics of the human sleep EEG. J. Sleep Res., 1993, 2: 7081. Aeschbach, D, Dijk, D. J. and Borbely, A. A. Dynamics of EEG spindle frequency activity during extended sleep in humans: relationship to slow-wave activity and time of day. Brain Res, 1997, 748: 131136 . Amzica, F. and Steriade, M. Short-and long-range neuronal synchronization of the slow (51 Hz) cortical oscillation. J. Neurophysiol., 1995, 73: 2038. Amzica, F. and Steriade, M. The K-complex: its slow (51 Hz) rhythmicity and relation with delta waves. Neurology, 1997, 49: 952959. Borbely, A. A. A two process model of sleep regulation. Hum. Neurobiol., 1982, 1: 195204. Borbely, A. A. and Achermann, P. Concepts and models of sleep regulation: an overview. J. Sleep Res., 1992, 1: 6379. Borbely, A. A., Baumann, F., Brandeis, D., Strauch, I. and Lehmann, D. Sleep deprivation: effect on sleep stages and EEG power density in man. Electroencephalogr. Clin. Neurophysiol., 1981, 51: 483495. Contreras, D., Destexhe, A., Sejnowski, T. J. and Steriade, M. Control of spatiotemporal coherence of a thalamic oscillation by corticothalamic feedback. Science, 1996, 274: 771774. Daan, S., Beersma, D. G. M. and Borbely, A. A. Timing of human sleep: recovery process gated by a circadian pacemaker. Am. J. Physiol., 1984, 246: R161-R183. Dijk, D. J. and Czeisler, C. A. Contribution of the circadian pacemaker and the sleep homeostat to sleep propensity, sleep structure, electroencephalographic slow waves, and sleep spindle activity in humans. J. Neurosci., 1995, 15: 35263538 . Dijk, D. J., Brunner, D. P., and Borbely, A. A. Time course of EEG

41

power density during long sleep in humans. Am. J. Physiol., 1990a, 258: R650-R661. Dijk, D. J., Brunner, D. P. Beersma, D. G. M. and Borbely, A. A. Slow wave sleep and electroencephalogram power density as a function of prior waking and circadian phase. Sleep, 1990b, 13: 430440. Dumermuth, G., Lange, B., Lehmann, D., Meier, C. A., Dinkelmann, R. and Molinari, L. Spectral analysis of all-night sleep EEG in healthy adults. Eur. Neurol., 1983, 22: 322339. Hirsch, J. C., Fourment, A. and Marc, M. E. Sleep-related variations of membrane potential in the lateral geniculate body relay neurons of the cat. Brain Res., 1983, 259: 308312. Pivik, R. T. and Harman, K. A reconceptualization of EEG alpha activity as an index of arousal during sleep: all alpha activity is not equal. J. Sleep Res., 1995, 4: 131137. Rechtschaffen, A. and Kales, A. A manual of standardized terminology, techniques and scoring system for sleep stages of human subjects. UCLA Brain Information Service: Brain Research Institute, Los Angeles, 1968. Steriade, M., McCormick, D. A. and Sejnowski, T. J. Thalamocortical oscillations in the sleeping and aroused brain. Science, 1993a, 262: 679685. Steriade, M., Nunez, A. and Amzica, F. A novel low (51 Hz) oscillation of neocortical neurons in vivo: depolarizing and hyperpolarizing components. J. Neurosci., 1993b, 13: 32523265. Steriade, M., Nunez, A. and Amzica, F. Intracellular analysis of relations between the slow (51 Hz) neocortical oscillation and other sleep rhythms of the electroencephalogram. J. Neurosci., 1993c, 13: 32663283. Steriade, M., Contreras, D., Curro Dossi, R. and Nunez, A. The slow (51 Hz) oscillation in reticular thalamic and thalamocortical neurons: scenario of sleep rhythm generation in interacting thalamic and neocortical networks. J. Neurosci., 1993d, 13: 32843299. Steriade, M., Contreras, D. and Amzica, F. Synchronized sleep oscillations and their paroxysmal developments. TINS, 1994, 17: 199208. Steriade, M., Amzica, F. and Contreras, D. Synchronization of fast (3040 Hz) spontaneous cortical rhythms during brain activation. J. Neurosci., 1996, 16: 392417. Werth, E., Achermann, P. and Borbely, A. A. Fronto-occipital EEG power gradients in human sleep. J. Sleep Res., 1997, 6: 102112. Werth, E., Dijk, D. J., Achermann, P. and Borbely, A. A. Dynamics of the sleep EEG after an early evening nap: experimental data and simulations. Am. J. Physiol., 1996, 271: R501-R510.

#1998 European Sleep Research Society, J. Sleep Res., 7 (Suppl. 1), 3641.