- Email: [email protected]
Coherence of the electroencephalogram during the first sleep cycle
Clinical Neurophysiology 116 (2005) 1088–1095 www.elsevier.com/locate/clinph
Coherence of the electroencephalogram during the first sleep cycle R.B. Duckrow*, H.P. Zaveri Department of Neurology, Yale University School of Medicine, New Haven, CT 06520-8018, USA Accepted 11 December 2004 Available online 24 January 2005
Abstract Objective: The increasing amplitude of the electroencephalogram (EEG) during non-rapid eye movement sleep implies a progressive synchronization of neuronal activity. We sought to characterize the spatial relationship of cortical activity at different frequencies during the first sleep cycle, focusing on sleep stages 3 and 4 (slow wave sleep). Methods: Sleep EEGs were obtained at home from six adults using a portable recorder. Signal power and magnitude squared coherence were measured during the first sleep cycle. Spectra obtained from bipolar and common reference derivations were compared. Results: During slow wave sleep, signal power is highest in the delta frequency band and regional coherence below 5 Hz is broadly distributed. Although signal power in the alpha and sigma frequency bands is lower, peaks of regional coherence in those bands are similar to or higher than delta-band coherence. Regional coherence during slow wave sleep is differentially distributed with a 14 Hz component in central and posterior regions and a 10 Hz component in frontal and central regions. Conclusions: Ten and 14 Hz rhythms are an essential component of slow wave sleep. Significance: The interpretation of scalp EEG power and coherence spectra is limited by the lack of a satisfactory recording reference. However, conclusions can be made by comparing and contrasting results from both bipolar and common reference recordings. q 2005 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved. Keywords: Electroencephalography; Sleep; Coherence; Power spectra; Bipolar; Referential
1. Introduction Coherence, the normalized cross-power spectrum, may be interpreted as the frequency-indexed correlation coefficient estimating the linear relationship between two time series (Jenkins and Watts, 1968). It has been applied to the electroencephalogram (EEG) to quantify intra- and inter-hemispheric functional relationships with variable results (French and Beaumont, 1984). Much of this variability appears because of differences in method, but there are also fundamental problems of interpretation that arise because of the lack of a satisfactory recording reference. Sleep is a state characterized by a progressive increase in the amplitude of slow EEG activity * Corresponding author. Address: Department of Neurology, Yale University School of Medicine, New Haven, CT 06520-8018, USA. Tel.: C1 203 785 3865; fax: C1 203 737 2799. E-mail address: [email protected] (R.B. Duckrow).
punctuated by various rhythmic patterns at higher frequencies that indicate a complex synchronization of neuronal activity (Steriade et al., 1993a,b; Steriade et al., 1994). In their careful analysis of the human sleep electroencephalogram, Achermann and Borbe´ly (1998) observed the dominance of slow wave activity in the power spectrum of non-rapid eye movement sleep but did not find an associated peak in the intra-hemispheric coherence spectrum. Their study used longitudinal bipolar electrode derivations, raising the possibility that widely distributed synchronous activity would be underrepresented because of common mode rejection. We addressed the question of regional interrelatedness of cortical activity measured at the scalp during sleep using both bipolar and common reference recordings of the first sleep cycle in humans. Although we analyzed the awake, slow wave and rapid eye movement (REM) sleep stages, we focused our presentation on results from slow wave sleep (SWS, sleep stages 3 and 4).
1388-2457/$30.00 q 2005 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.clinph.2004.12.002
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
2. Methods Data were collected at the University of Connecticut Health Center using methods approved by the Institutional Review Board of that institution. Sleep records were obtained from six adult subjects referred to the neurophysiology laboratory for ambulatory EEG monitoring because of unexplained spells. The records were selected because each 24-h study was normal. The subjects ranged in age from 21 to 60 with a median of 40 years. Four subjects were women. Four were taking no medications. One subject was taking gabapentin and fluoxetine, and one subject was taking enalapril and diltiazem. Signals from the traditional 19 scalp electrode locations of the International 10–20 system and each mastoid were bandpass filtered from 0.5 to 100 Hz and digitized at an effective rate of 200 samples per second (H2O Ambulatory Recorder, Grass-Telefactor, West Warwick, RI, USA). After electrode application the subjects were given a diary to record their activity, including when they slept. They returned home to sleep in their own bed. Continuous data from the first sleep cycle was extracted and quantitative results were derived from data segments visually edited for artifact. The first cycle of sleep was scored as awake, slow wave sleep, or rapid eye movement (REM) sleep using traditional criteria applied to signals from all electrodes (Radtke, 1990). The alpha rhythm prior to sleep onset was used to characterize wakefulness, as it was usually free of movement artifact. Dedicated muscle and eye movement electrodes were not placed but, when present, rapid lateral eye movements and muscle tone were adequately characterized in frontal-temporal (F7, F8) and mid-temporal (T3, T4) signals. The REM sleep period was also evident in plots of frequency band-integrated power, as shown in Fig. 1. Signals from both bipolar and reference derivations were examined. Bipolar derivations included F3–C3 and
1089
F4–C4 (FC, frontal–central), F7–T3 and F8–T4 (FT, frontal–temporal), and P3–O1, and P4–O2 (PO, parietal–occipital). Common reference electrode derivations included F3 and F4 (F, frontal), T3 and T4 (T, temporal), C3 and C4 (C, central), P3 and P4 (P, parietal), and O1 and O2 (O, occipital) referred to the ipsilateral mastoid, A1 or A2. Magnitude squared coherence (MSC or coherence) was estimated by the weighted-overlapped segment averaging method (Carter et al., 1973; Zaveri et al., 1999). The edited EEG of each sleep state was segmented (segment length TZ2 s), linearly detrended, and weighted with a Hann window before calculation of the fast Fourier transform (spectral resolution DFZ0.5 Hz). The number of 2 s sample epochs used to characterize each state ranged from 80 to 960, as shown in Table 1. The minimum value of ndZ80 allows us to claim MSCO0.06 as non-zero with a confidence of aZ0.05 and power of 1KbZ0.8 using a one tailed test of significance (Carter et al., 1973; Zaveri et al., 1999). Power spectral density (PSD or power) was estimated by the Welch method in the same edited data segments used for coherence analysis. Segments were not overlapped during estimation of MSC or PSD. Analysis was restricted to periods of relative signal stationarity, as indicated by time plots of band-integrated power. An example is given in Fig. 1. Accordingly, sleep stages 1 and 2 were excluded from analysis, as these states are characterized by a continuous change in signal power. Spectra from homologous electrode derivation were averaged between hemispheres, as justified by the known interhemispheric symmetry of power and coherence during sleep (Achermann and Borbe´ly, 1998; Finelli et al., 2001), and then across subjects. Peaks were located subjectively as local maxima on the background pattern, and only qualitative statements were made concerning relative frequency at spectral power or MSC peaks or bands. The 95% confidence intervals of relevant local maxima (peaks) were derived from an analytic derivation of the probability distribution function
Fig. 1. EEG samples and a time plot of frequency band-integrated power for a single subject during the first sleep cycle. Alpha-band power (8–13 Hz, using P3– O1) drops with sleep onset followed by the progressive rise of delta-band power (1–4 Hz, using F3–C3) during sleep stages 1 and 2. The stable plateau of maximum delta-band power characterized slow wave sleep and was used for analysis. Sleep stages are indicated in the plot of band-integrated power by horizontal arrows. A, awake; S1, stage 1 sleep; S2, stage 2 sleep; SWS, slow wave sleep, stages 3 and 4 sleep; REM, rapid eye movement sleep.
1090
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
Table 1 Number of 2 s disjoint EEG segments (nd) used for spectral analysis from wakefulness (Awake), slow wave sleep (SWS), and rapid eye movement sleep (REM) Subject
Awake
SWS
REM
A B C D E F
80 140 – 101 120 175
960 600 337 840 600 120
– – 333 292 – 235
of the coherence estimate (Carter et al., 1973; Zaveri et al., 1999). We used ndZ100 to generate these confidence intervals. This produced a conservative estimate of confidence, as most data segments were much larger, as shown in Table 1. The power and coherence spectra for each subject are available as supplementary information. Slow oscillations below 1 Hz (Achermann and Borbe´ly, 1997; Amzica and Steriade, 1997; Mo¨lle et al., 2002; Steriade et al., 1993a,b) were not studied in detail because of restrictions placed by the 0.5 Hz highpass recording filter.
3. Results 3.1. Analysis of sleep measures The first sleep cycle was used for spectral analysis because it was most likely to provide a period of quiet
wakefulness with eyes closed and a stable epoch of slow wave sleep No attempt was made to control the sleep environment or characterize sleep architecture over the entire night. One record did not have an awake epoch without artifact. All subjects entered stable slow wave sleep. Only 3 of the 6 subjects entered into REM sleep, as scored using the signals from available electrodes. 3.2. Analysis of bipolar derivations 3.2.1. Power Power spectral density plots derived from bipolar derivations are presented in the top row of Fig. 2 The dominant 9 Hz peak present during wakefulness attenuated during REM sleep while power in the delta frequency range was similar during wakefulness and REM sleep. Slow wave sleep was accompanied by a marked increase in power below 5 Hz and a broad peak at 8 Hz at all locations. During slow wave sleep there was a broad peak at 14 Hz in frontal– temporal and frontal-central derivations but not the parietal– occipital derivation. 3.2.2. Coherence Coherence spectra derived from bipolar derivations are presented in the bottom row of Fig. 2. Coherence was measured between regions represented by three electrode pairs. Homologous left and right intra-hemispheric measurements were averaged to determine frontal–central to parietal-occipital coherence (FC–PO), frontal–central
Fig. 2. Power spectral density (top row) and coherence spectra (bottom row) derived from bipolar recordings while awake (left column), slow wave sleep (middle column), and rapid eye movement sleep (right column). Bipolar derivations included F3–C3 and F4–C4 (FC, frontal–central), F7–T3 and F8–T4 (FT, frontal–temporal), and P3–O1 and P4–O2 (PO, parietal–occipital). Spectra were averaged across hemisphere and then across subjects. Conservative estimates of the 95% confidence interval of selected peak values are show by vertical lines. Only the lower portion of each confidence interval is shown for clarity.
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
to frontal-temporal coherence (FC–FT), and frontal–temporal to parietal-occipital coherence (FT–PO). During wakefulness, peaks of coherence were present at 9 Hz in FC–FT and FC–PO pairs and possibly the FT–PO pair. During slow wave sleep a peak of coherence was present at 14 Hz in the FC–PO pair and at 12 Hz in the FC–FT pair. The FC–FT pair also had a peak at 1.5 Hz. The FT–PO pair had no significant coherence. During REM sleep, broad-band coherence was present in the FC–FT pair, but there were no clear peaks. A possible peak at w4 Hz was present in the FC–PO pair. During each sleep state, FC–FT coherence was highest, as is expected when activity from nearby scalp regions is compared. Coherence in this pair was present over the entire frequency band and decreased with increasing frequency. This trend was not present when coherence was measured
1091
over longer distances, as the baselines of the other coherence spectra were flat. 3.3. Analysis of common reference derivations 3.3.1. Power Power spectral density plots derived from common reference derivations are presented in the top row of Fig. 3 Spectral power during wakefulness and REM sleep contained patterns similar to those seen with bipolar derivations. Residual 9 Hz activity during REM sleep was only present at the parietal and occipital sites. Slow wave sleep had maximum spectral power below 5 Hz and broad peaks at 8 Hz at all sites. Isolated peaks were present at 10 Hz in the frontal and central sites and at 14 Hz at central, parietal, and occipital sites. Expanded displays of power
Fig. 3. Power spectral density (top row) and coherence spectra (middle and bottom rows) derived from common reference recordings while awake (left column), slow wave sleep (middle column), and rapid eye movement sleep (right column). Common electrode derivations included F3 F4 (F, frontal), T3 T4 (T, temporal), C3 C4 (C, central), P3 P4 (P, parietal), and O1 O2 (O, occipital) referred to the ipsilateral mastoid. Spectra were averaged across hemispheres and then across subjects. Conservative estimates of the 95% confidence interval of selected peak values are show by vertical lines. Only the lower portion of each confidence interval is shown for clarity.
1092
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
spectra are available as supplementary data on the publisher’s web site.
3.4. Comparison of bipolar and common reference derivations
3.3.2. Coherence Coherence spectra derived from common reference derivations are presented in the middle and bottom panels of Fig. 3. For each hemisphere, the ipsilateral mastoid electrode was used as the common reference. Again, homologous left and right hemisphere comparisons were averaged. Intra-hemispheric coherence was measured between all possible pairs of regions sampled by the restricted electrode set described in the methods. For display purposes, sequential comparisons were made of the frontal region to each of the other hemispheric sites with increasing electrode separation, i.e. frontal to central (F–C), frontal to temporal (F–T), frontal to parietal (F–P), and frontal to occipital (F–O). These comparisons are shown in the middle row of Fig. 3. Then, sequential comparisons were made of the occipital region to the other sites with increasing electrode separation (Epstein and Brickley, 1985). These comparisons are shown in the bottom row of Fig. 3. Coherence spectra derived from common reference recordings during wakefulness and slow wave sleep showed a progressive increase in the coherence value beginning between 12 and 15 Hz and rising to a plateau above 30 Hz. Based on our previous work, this pattern was recognized as the signature of reference contamination (Zaveri et al., 2000). However, both distinct peaks of coherence and coherence values indistinguishable from zero were present at frequencies below 15 Hz, indicating that reference contamination was minimal at these lower frequencies. During wakefulness there were peaks of scalp coherence at 3 Hz in all electrode pairs separated by one or two interelectrode distances. There was a consistent peak at 9 Hz in the occipital to frontal pair. During slow wave sleep, there were peaks at 1.5, 8, and 10 Hz when comparing frontal, temporal, central, and parietal sites. The 10 Hz peak predominated. However, these peaks were not present in the occipital to frontal and occipital to temporal pairs, where coherence below w12 Hz could not be distinguished from zero. The 1.5 and 8 Hz peaks were present in the occipital to parietal pair, but a 14 Hz peak replaced the 10 Hz peak. This 14 Hz peak was also present in the occipital to central pair. Again, when the occipital site was compared with the remaining regions, coherence below w12 Hz could not be distinguished from zero. During REM sleep, frontal, temporal, central, and parietal comparisons had coherence peaks at 3 and 7 Hz, but these peaks were less distinct than those found during other states. However, a prominent peak was present at 3 Hz in the occipital to parietal pair. During REM sleep, when muscle tone is reduced, the progressive increase in coherence at frequencies above 15 Hz, ascribed to reference contamination during wakefulness or slow wave sleep, was not present.
During wakefulness, coherence spectra from all bipolar derivations and the occipital to frontal common reference derivation showed a peak at 9 Hz. The difference between coherence spectra from bipolar and common reference derivations during wakefulness was confined to the lack of a substantive peak below 5 Hz in bipolar derivations. During slow wave sleep a well defined coherence peak at 14 Hz in the bipolar frontal–central to parietal–occipital (FC–PO) pairing was mirrored by coherence peaks at the same frequency in common reference derivations that compared the occipital region to the parietal or central regions but not the frontal or temporal regions. During slow wave sleep a coherence peak at 10 Hz was present in common reference recordings that compared the frontal to the central, parietal, or temporal regions but not the occipital region. A coherence peak at 10 Hz was not present in bipolar comparisons. Using common reference derivations there were peaks at 8 Hz and below 5 Hz in all pairs except the occipital to frontal (O–F) pair and the occipital to temporal (O–T) pair. A peak below 5 Hz was present in the bipolar frontal-central to frontal–temporal (FC–FT) comparison, but there was no clear peak at 8 Hz. There was a broad peak in this bipolar comparison at 12 Hz that did not correlate well with any peak in reference derivations. A summary of the distribution of spectral peaks of EEG power and coherence during slow wave sleep is shown in Fig. 4. Only 3 subjects provided REM sleep for analysis, and, except for a 3 Hz peak in the occipital to parietal pair, peaks present in coherence spectra from common reference
Fig. 4. The distribution of spectral peaks of EEG power and coherence during slow wave sleep. During slow wave sleep, significant coherence at 10 Hz is present in frontal, central, temporal, and parietal regions. Significant coherence at 14 Hz is present in central, parietal, and occipital regions. Spectral power at 8 Hz was present diffusely in both bipolar and reference recordings. A peak of coherence was present at 8 Hz in all common reference electrode comparisons except the frontal-occipital pair. There was no coherent activity at 8 Hz in bipolar recordings. MSC, magnitude squared coherence; F, frontal; C, central; P, parietal; O, occipital; T, temporal.
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
recordings were poorly defined. Even less structure was present in coherence spectra from bipolar derivations. The coherence spectra for each subject during wakefulness, slow wave sleep, and rapid eye movement sleep derived from bipolar and reference derivations are available as supplementary data on the publisher’s web site.
4. Discussion Our results confirm the findings of previous investigators who used longitudinal bipolar electrode derivations to study the sleep EEG (Achermann and Borbe´ly, 1998; Dumermuth et al., 1972). During slow wave sleep, EEG power is concentrated below 5 Hz while regional coherence within the hemisphere is present in restricted bands at higher frequencies. Our analysis of slow wave sleep using common reference and bipolar recordings expands these findings by localizing intra-hemispheric coherence at 10 Hz to frontal, temporal, central, and parietal regions and intra-hemispheric coherence at 14 Hz to occipital, parietal, and central regions. While Achermann and Borbe´ly (1998) combined the 4 non-REM sleep stages for analysis, we concentrated on slow wave sleep because of its relative stability. Their bipolar comparison of frontal-central to parietal–occipital regions demonstrated a major coherence peak at 14 Hz and a minor coherence peak at 10 Hz. Our mastoid-reference recordings during slow wave sleep have a spectral power peak at 10 Hz at frontal and central electrodes but not at the remaining locations. Subtraction of activity at this frequency in frontalcentral regions by the bipolar method and lower power at this frequency in posterior regions could explain reduced anterior–posterior coherence at 10 Hz in the bipolar study of Achermann and Borbe´ly and its complete absence in our bipolar findings. This interpretation is supported by our finding coherence peaks at 10 Hz in comparison of common reference recordings from anterior but not posterior electrodes (middle column of Fig. 3). During slow wave sleep a coherence peak at 14 Hz is present in both our bipolar and common reference recordings. It was also present in the bipolar recordings of Achermann and Borbe´ly (1998). This is best explained by a prominent gradient of activity at that frequency centered on central and parietal regions and extending to the occipital region but not the frontal region. This formulation is supported by our power spectra that show a peak at 14 Hz with maximum amplitude in the central lead followed by the parietal and then occipital leads but absent in frontal and temporal leads. Activity at that frequency would appear in the bipolar frontal–central derivation because of its presence in the central lead. Activity at that frequency would appear in the parietal– occipital derivation because of the amplitude gradient, but to a lesser degree than in the frontal region. This formulation is supported by our bipolar power spectra.
1093
The presence of 14 Hz activity during slow wave sleep is consistent with the general observation that sigma-band power is highest during stage 2 sleep and progressively declines during stages 3 and 4. However, sigma-band power is still higher in slow wave sleep than in stage 1 or REM sleep (Dijk et al., 1993; Johnson et al., 1969). The majority of this power is usually attributed to episodic spindle activity appearing in the analysis window of the Fourier transform used to estimate spectral power. Sleep spindles are reported to appear as often in slow wave sleep as in stage 2 (Gaillard and Blois, 1981) but with less individual power (Dijk et al., 1993). However, the opposite finding, equal amplitude but fewer number, has also been reported (Zeitlhofer et al., 1997). This could be explained by the considerable variation in spindle characteristics determined by automated methods. It is more likely that the amplitude of sigma activity is continuously modulated and highly variable during non-REM sleep (Mo¨lle et al., 2002; Zygierewicz et al., 1999), resulting in characterizations that are dependent on the somewhat arbitrary selection of detection parameters. Dumermuth et al. (1972) considered sigma activity the essential component of slow wave sleep, whether it appeared as spindles or as continuous activity. It is accepted that sleep spindles appear at different frequencies with differing topography. Gibbs and Gibbs (1951) described three types of spindles during sleep: 14 Hz spindles dominant in parietal areas; 12 Hz spindles dominant in frontal areas; and 10 Hz spindles that were ‘generalized’ (frontal and parietal). The topographic distribution of the 14 and 12 Hz spindles described by the Gibbs has been confirmed, although most investigators report center frequencies 0.5 Hz lower (Jobert et al., 1992; Scheuler et al., 1990a; Werth et al., 1997; Zygierewicz et al., 1999). Our observations extend the topography of the 14 Hz activity into slow wave sleep and add that this activity has increased regional coherence at central, parietal, and occipital sites. However, we cannot address the continuous or episodic nature or the amplitude envelope of this sigma activity. We do not find a peak of signal power at 12 Hz during slow wave sleep. It is possible that signal power at this frequency is lost or that the center frequency of this activity decreases as sleep advances beyond stage 2. Others studying sigma power during sleep analyzed stage 2 exclusively or combined all stages of non-REM sleep, so the persistence of a 12 Hz rhythm into slow wave sleep cannot be confirmed. Given the within-subject stability of the characteristic frequency of other sleep-related patterns, the correspondence of our 10 Hz peak with the 12 Hz rhythm seem unlikely. However, Zygierewicz et al. (1999), using matching pursuit decomposition of overnight sleep recordings into spindle-shaped envelopes, also found 10.2 Hz frontal spindles in at least one subject. While the designation of 10 Hz activity during sleep as a spindle pattern is controversial (Jankel and Niedermeyer, 1985), the presence of 10 Hz activity with frontal maximum
1094
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
during slow wave sleep is clear (Dumermuth et al., 1972; Hauri and Hawkins, 1973; Pivik and Harman, 1995; Scheuler et al., 1983). However, it may only present a recognizable pattern in 10–30% of subjects (Gibbs and Gibbs, 1951; Scheuler et al., 1990b) and must be differentiated from arousal. We find peaks of mean spectral power at 8 and 10 Hz in common reference recordings during slow wave sleep. The 8 Hz peak is diffuse and the 10 Hz peak is frontal-central. Post hoc inspection of the power spectra from individual subjects showed the 10 Hz peak was exclusive to 4 subjects and the 8 Hz peak was exclusive to the other 2 subjects, indicating that the 10 Hz activity is not a universal finding. The global 8 Hz rhythm could be derived from the 9 Hz awake alpha and represent unrecognized arousals. However, post hoc review of the EEG showed nearly continuous low amplitude alpha-range activity and was inconsistent with arousals. Also, this activity did not match exemplars of alpha-sleep or alphadelta sleep found in the literature (Hauri and Hawkins, 1973; Scheuler et al., 1983). It is possible that the characteristic frequency of the frontal sigma rhythm, nominally 11.5 Hz, drops below the traditional sigma-frequency range during slow wave sleep. Of greater interest is the localization of increased coherence at 10 Hz during slow wave sleep to frontal, temporal, central, and parietal regions in contrast to the localization of increased coherence at 14 Hz to central, parietal, and occipital regions. The segregation of these regional interactions becomes apparent when common reference recordings are used. It is our contention that coherence measurements from bipolar recordings are confounded by the independence of the signals probed by each element of the bipolar pair of scalp electrodes, as we argue for intracranial recording (Zaveri et al., submitted). Scalp coherence measurements may also be confounded when widely distributed cortical rhythms are minimized by common mode subtraction. We accept the argument that common reference recordings allow contamination of the coherence measurement by a common signal. However, this contamination is dependent on the signal-to-noise ratio and can be recognized by its characteristic spectral pattern, allowing conclusions to be drawn from spectral segments that show less or no contamination (Zaveri et al., 2000). The subjects providing data for this analysis were not screened for sleep disorders or habituated to the recording equipment, and no attempt was made to quantify the architecture of their overnight recording. The uncontrolled use of medications or the presence of either an acute or chronic sleep disturbance would affect the spectral character of the sleep recordings. For example, even modest sleep deprivation can vary the duration of sleep stages (Webb and Agnew, 1971), increase the power spectral density in the delta frequency band (Borbe´ly et al., 1981), and change the incidence and magnitude of sleep spindles, depending on their frequency (Knoblauch et al., 2003).
However, the intent of this study was not to detect changes in sleep architecture but to characterize the presence and distribution of the most obvious peaks in the coherence spectra and contrast those derived from bipolar and common reference electrode derivations. To this end, the sustained presence of slow wave sleep was the essential requirement of the recording. It is reassuring that the results from our bipolar recordings reproduce findings from studies with stricter limits on subject recruitment (Achermann and Borbe´ly, 1998; Dumermuth et al., 1972). Of those medications taken by two of the six subjects, fluoxetine had the greatest potential influence on the EEG during sleep (Ebert and Kierch, 1999; Feige et al., 2002; Salinsky et al., 2002). Feige et al. (2002) reported that subchronic administration of fluoxetine increased EEG power at frequencies above 10 Hz during non-REM sleep with more than a 50% increase in EEG power in a distinct spectral peak at 14 Hz. Post-hoc examination of the power spectra of our single subject taking fluoxetine (subject D, spectra available as supplementary data) showed no obvious differences from the spectra of the other subjects. The same was true of the coherence spectra derived from common reference recordings. For example, the coherence peak at 14 Hz in the O–P derivation during slow wave sleep was present in the spectrum of each subject with individual peak values ranging from 0.49 to 0.70. The peak value for the subject taking fluoxetine was 0.55. However, in the coherence spectra derived from bipolar recordings, coherence peak values at 14 Hz in the FC–PO derivation during slow wave sleep ranged from 0.15 to 0.32. The value for the subject taking fluoxetine was 0.22. It is possible that fluoxetine affected the mean amplitude of this peak, but since all other subjects had a peak at that frequency, the presence of fluoxetine did not cause that peak to appear. Also, this peak was reported by Achermann and Borbe´ly (1998), and their subjects were not taking medications. We agree with Dumermuth et al. (1972) that sigma-band activity is an essential component of slow wave sleep. Coherence peaks in that band are organized in central, parietal, and occipital regions and must be contrasted with a separate band of coherent activity at 10 Hz in frontal, temporal, central, and parietal regions. The presence of peaks indicates an underlying structure to EEG patterns. The differential segregation of these coherence peaks suggests the independence of these patterns. The relationship of this 10 Hz activity to frontal spindles and to the thalamocortical circuits presumed to underlie their genesis remains to be elucidated.
Acknowledgements The authors recognize the technical contributions of Kevin Gaski.
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
Appendix. Supplementary Material Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.clinph.2004. 12.002
References Achermann P, Borbe´ly AA. Low-frequency (!1 Hz) oscillations in the human sleep electroencephalogram. Neuroscience 1997;81(1):213–22. Achermann P, Borbe´ly AA. Coherence analysis of the human sleep electroencephalogram. Neuroscience 1998;85(4):1195–208. Amzica F, Steriade M. The K-complex: its slow (!1-Hz) rhythmicity and relation to delta waves. Neurology 1997;49(4):952–9. Borbe´ly AA, Baumann F, Brandeis D, Strauch I, Lehmann D. Sleep deprivation: Effect on sleep stages and EEG power density in man. Electroenceph Clin Neurophysiol 1981;51(5):483–93. Carter GC, Knapp CH, Nuttall AH. Statistics of the estimate of the magnitude-coherence function. IEEE Transactions on Acoustics, Speech, and Signal Processing 1973;21(4):388–9. Dijk D-J, Hayes B, Czeisler CA. Dynamics of electroencephalographic sleep spindles and slow wave activity in men: effect of sleep deprivation. Brain Res 1993;626(1–2):190–9. Dumermuth G, Walz W, Scollo-Lavizzari G, Kleiner B. Spectral analysis of EEG activity in different sleep stages in normal adults. Eur Neurol 1972;7(5):265–96. Ebert U, Kierch W. Effects of captopril and enalapril on electroencephalogram and cognitive performance in healthy volunteers. Eur J Clin Pharmacol 1999;55(4):255–7. Epstein CM, Brickley GP. Interelectrode distance and amplitude of the scalp EEG. Electroenceph Clin Neurophysiol 1985;60:287–92. Feige B, Voderholzer U, Riemann D, Dittmnn R, Hohagen F, Berger M. Fluoxetine and sleep: effect of a single dose, subchronic threatment, and discontinuation in healthy subjects. Neuropsychopharmacol 2002; 26(2):246–58. Finelli LA, Borbe´ly AA, Achermann P. Functional topography of the human nonREM sleep electroencephalogram. Eur J Neurosci 2001; 13(12):2282–90. French CC, Beaumont JG. A critical review of EEG coherence studies of hemisphere function. Int J Psychophysiol 1984;1(3):241–54. Gaillard J-M, Blois R. Spindle density in sleep of normal subjects. Sleep 1981;4(4):385–91. Gibbs FA, Gibbs EL.. 2nd ed Atlas of electroencephalography. vol. 1. Reading MA: Addison-Wesley Publishing Company; 1951 p. 93–94. Hauri P, Hawkins DR. Alpha-delta sleep. Electroenceph Clin Neurophysiol 1973;34(3):233–7. Jankel WR, Niedermeyer E. Sleep spindles. J Clin Neurophysiol 1985;2(1): 1–35. Jenkins GM, Watts DG. Spectral analysis and its applications. San Francisco, CA: Holden-Day; 1968. Jobert M, Poiseau E, Ja¨hnig P, Schulz H, Kubicki S. Topographical analysis of sleep spindle activity. Neuropsychobiology 1992;26(4):210–7.
1095
Johnson L, Lubin A, Naitoh P, Nute C, Austin M. Spectral analysis of the EEG of dominant and non-dominant alpha subjects during waking and sleeping. Electroenceph Clin Neurophysiol 1969;26(4):361–70. Knoblauch V, Martens WLJ, Wirz-Justice A, Cajochen C. Human sleep spindle characteristics after sleep deprivation. Clin Neurophysiol 2003; 114(12):2258–67. Mo¨lle M, Marshall L, Gais S, Born J. Grouping of spindle activity during slow oscillations in human non-rapid eye movement sleep. J Neurosci 2002;22(4):10941–7. Pivik RT, 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(3):131–7. Radtke RA. Sleep disorders. In: Daly DD, Pedley TA, editors. Current practice of clinical electroencephalography. 2nd ed. New York, NY: Raven Press; 1990. p. 574–5. Salinsky MC, Binder LM, Oken BS, Storzbach D, Aron CR, Dodrill CB. Effects of gabapentin and carbamazepine on the EEG and cognition in healthy volunteers. Epilepsia 2002;43(5):482–90. Scheuler W, Stinshoff D, Kubicki S. The alpha-sleep pattern. Differentiation from other sleep patterns and effect of hypnotics. Neuropsychobiology 1983;10(2–3):183–9. Scheuler W, Kubickis, Scholz G, Marquardt J. Two different activities in the sleep spindle frequency band-discrimination based on the topographical distribution of spectral power and coherence. In: Horne J, editor. Sleep 90. Bochum: Pontenagel Press; 1990a. p. 13–6. Scheuler W, Rappelsberger P, Schmatz F, Pastelak-Price C, Petsche H, Kubicki S. Periodicity analysis of sleep EEG in the second and minute ranges-example of application in different alpha activities in sleep. Electroenceph Clin Neurophysiol 1990b;76(3):222–34. Steriade M, McCormick DA, Sejnowski TJ. Thalamocortical oscillations in the sleeping and aroused brain. Science 1993;262:679–85. Steriade M, Nunez A, Amzica F. A novel slow (!1 Hz) oscillation of neocortical neurons in vivo: depolarizing and hyperpolarizing components. J Neurosci 1993;13(8):3252–65. Steriade M, Contreras D, Amzica F. Synchronized sleep oscillations and their paroxysmal developments. Trends Neurosci 1994;17(5):199–208. Webb WB, Agnew HW. Stage 4 sleep: influence of time course variables. Science 1971;174(4016):1354–6. Werth E, Acherman P, Dijk D-J, Borbe´ly AA. Spindle frequency activity in the sleep EEG: individual differences and topographic distribution. Electroenceph Clin Neurophysiol 1997;103:535–42. Zaveri HP, Williams WJ, Sackellares JC, Baydoun A, Duckrow RB, Spencer SS. Measuring the coherence of intracranial electroencephalograms. Clin Neurophysiol 1999;110(10):1717–25. Zaveri HP, Duckrow RB, Spencer SS. The effect of a scalp reference signal on coherence measurements of intracranial electroencephalograms. Clin Neurophysiol 2000;111(7):1293–9. Zaveri HP, Duckrow RB, Spencer SS. On the use of bipolar montages for the time-series analysis of intracranial electroencephalograms. Submitted. Zeitlhofer J, Gruber G, Anderer P, Asenbaum S, Schimicek P, Saletu B. Topographic distribution of sleep spindles in young healthy subjects. J Sleep Res 1997;6(3):149–55. Zygierewicz J, Blinowska KJ, Durka PJ, Szelenberger W, Niemcewicz S, Androsiuk W. High resolution study of sleep spindles. Clin Neurophysiol 1999;110(12):2136–47.
Coherence of the electroencephalogram during the first sleep cycle R.B. Duckrow*, H.P. Zaveri Department of Neurology, Yale University School of Medicine, New Haven, CT 06520-8018, USA Accepted 11 December 2004 Available online 24 January 2005
Abstract Objective: The increasing amplitude of the electroencephalogram (EEG) during non-rapid eye movement sleep implies a progressive synchronization of neuronal activity. We sought to characterize the spatial relationship of cortical activity at different frequencies during the first sleep cycle, focusing on sleep stages 3 and 4 (slow wave sleep). Methods: Sleep EEGs were obtained at home from six adults using a portable recorder. Signal power and magnitude squared coherence were measured during the first sleep cycle. Spectra obtained from bipolar and common reference derivations were compared. Results: During slow wave sleep, signal power is highest in the delta frequency band and regional coherence below 5 Hz is broadly distributed. Although signal power in the alpha and sigma frequency bands is lower, peaks of regional coherence in those bands are similar to or higher than delta-band coherence. Regional coherence during slow wave sleep is differentially distributed with a 14 Hz component in central and posterior regions and a 10 Hz component in frontal and central regions. Conclusions: Ten and 14 Hz rhythms are an essential component of slow wave sleep. Significance: The interpretation of scalp EEG power and coherence spectra is limited by the lack of a satisfactory recording reference. However, conclusions can be made by comparing and contrasting results from both bipolar and common reference recordings. q 2005 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved. Keywords: Electroencephalography; Sleep; Coherence; Power spectra; Bipolar; Referential
1. Introduction Coherence, the normalized cross-power spectrum, may be interpreted as the frequency-indexed correlation coefficient estimating the linear relationship between two time series (Jenkins and Watts, 1968). It has been applied to the electroencephalogram (EEG) to quantify intra- and inter-hemispheric functional relationships with variable results (French and Beaumont, 1984). Much of this variability appears because of differences in method, but there are also fundamental problems of interpretation that arise because of the lack of a satisfactory recording reference. Sleep is a state characterized by a progressive increase in the amplitude of slow EEG activity * Corresponding author. Address: Department of Neurology, Yale University School of Medicine, New Haven, CT 06520-8018, USA. Tel.: C1 203 785 3865; fax: C1 203 737 2799. E-mail address: [email protected] (R.B. Duckrow).
punctuated by various rhythmic patterns at higher frequencies that indicate a complex synchronization of neuronal activity (Steriade et al., 1993a,b; Steriade et al., 1994). In their careful analysis of the human sleep electroencephalogram, Achermann and Borbe´ly (1998) observed the dominance of slow wave activity in the power spectrum of non-rapid eye movement sleep but did not find an associated peak in the intra-hemispheric coherence spectrum. Their study used longitudinal bipolar electrode derivations, raising the possibility that widely distributed synchronous activity would be underrepresented because of common mode rejection. We addressed the question of regional interrelatedness of cortical activity measured at the scalp during sleep using both bipolar and common reference recordings of the first sleep cycle in humans. Although we analyzed the awake, slow wave and rapid eye movement (REM) sleep stages, we focused our presentation on results from slow wave sleep (SWS, sleep stages 3 and 4).
1388-2457/$30.00 q 2005 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.clinph.2004.12.002
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
2. Methods Data were collected at the University of Connecticut Health Center using methods approved by the Institutional Review Board of that institution. Sleep records were obtained from six adult subjects referred to the neurophysiology laboratory for ambulatory EEG monitoring because of unexplained spells. The records were selected because each 24-h study was normal. The subjects ranged in age from 21 to 60 with a median of 40 years. Four subjects were women. Four were taking no medications. One subject was taking gabapentin and fluoxetine, and one subject was taking enalapril and diltiazem. Signals from the traditional 19 scalp electrode locations of the International 10–20 system and each mastoid were bandpass filtered from 0.5 to 100 Hz and digitized at an effective rate of 200 samples per second (H2O Ambulatory Recorder, Grass-Telefactor, West Warwick, RI, USA). After electrode application the subjects were given a diary to record their activity, including when they slept. They returned home to sleep in their own bed. Continuous data from the first sleep cycle was extracted and quantitative results were derived from data segments visually edited for artifact. The first cycle of sleep was scored as awake, slow wave sleep, or rapid eye movement (REM) sleep using traditional criteria applied to signals from all electrodes (Radtke, 1990). The alpha rhythm prior to sleep onset was used to characterize wakefulness, as it was usually free of movement artifact. Dedicated muscle and eye movement electrodes were not placed but, when present, rapid lateral eye movements and muscle tone were adequately characterized in frontal-temporal (F7, F8) and mid-temporal (T3, T4) signals. The REM sleep period was also evident in plots of frequency band-integrated power, as shown in Fig. 1. Signals from both bipolar and reference derivations were examined. Bipolar derivations included F3–C3 and
1089
F4–C4 (FC, frontal–central), F7–T3 and F8–T4 (FT, frontal–temporal), and P3–O1, and P4–O2 (PO, parietal–occipital). Common reference electrode derivations included F3 and F4 (F, frontal), T3 and T4 (T, temporal), C3 and C4 (C, central), P3 and P4 (P, parietal), and O1 and O2 (O, occipital) referred to the ipsilateral mastoid, A1 or A2. Magnitude squared coherence (MSC or coherence) was estimated by the weighted-overlapped segment averaging method (Carter et al., 1973; Zaveri et al., 1999). The edited EEG of each sleep state was segmented (segment length TZ2 s), linearly detrended, and weighted with a Hann window before calculation of the fast Fourier transform (spectral resolution DFZ0.5 Hz). The number of 2 s sample epochs used to characterize each state ranged from 80 to 960, as shown in Table 1. The minimum value of ndZ80 allows us to claim MSCO0.06 as non-zero with a confidence of aZ0.05 and power of 1KbZ0.8 using a one tailed test of significance (Carter et al., 1973; Zaveri et al., 1999). Power spectral density (PSD or power) was estimated by the Welch method in the same edited data segments used for coherence analysis. Segments were not overlapped during estimation of MSC or PSD. Analysis was restricted to periods of relative signal stationarity, as indicated by time plots of band-integrated power. An example is given in Fig. 1. Accordingly, sleep stages 1 and 2 were excluded from analysis, as these states are characterized by a continuous change in signal power. Spectra from homologous electrode derivation were averaged between hemispheres, as justified by the known interhemispheric symmetry of power and coherence during sleep (Achermann and Borbe´ly, 1998; Finelli et al., 2001), and then across subjects. Peaks were located subjectively as local maxima on the background pattern, and only qualitative statements were made concerning relative frequency at spectral power or MSC peaks or bands. The 95% confidence intervals of relevant local maxima (peaks) were derived from an analytic derivation of the probability distribution function
Fig. 1. EEG samples and a time plot of frequency band-integrated power for a single subject during the first sleep cycle. Alpha-band power (8–13 Hz, using P3– O1) drops with sleep onset followed by the progressive rise of delta-band power (1–4 Hz, using F3–C3) during sleep stages 1 and 2. The stable plateau of maximum delta-band power characterized slow wave sleep and was used for analysis. Sleep stages are indicated in the plot of band-integrated power by horizontal arrows. A, awake; S1, stage 1 sleep; S2, stage 2 sleep; SWS, slow wave sleep, stages 3 and 4 sleep; REM, rapid eye movement sleep.
1090
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
Table 1 Number of 2 s disjoint EEG segments (nd) used for spectral analysis from wakefulness (Awake), slow wave sleep (SWS), and rapid eye movement sleep (REM) Subject
Awake
SWS
REM
A B C D E F
80 140 – 101 120 175
960 600 337 840 600 120
– – 333 292 – 235
of the coherence estimate (Carter et al., 1973; Zaveri et al., 1999). We used ndZ100 to generate these confidence intervals. This produced a conservative estimate of confidence, as most data segments were much larger, as shown in Table 1. The power and coherence spectra for each subject are available as supplementary information. Slow oscillations below 1 Hz (Achermann and Borbe´ly, 1997; Amzica and Steriade, 1997; Mo¨lle et al., 2002; Steriade et al., 1993a,b) were not studied in detail because of restrictions placed by the 0.5 Hz highpass recording filter.
3. Results 3.1. Analysis of sleep measures The first sleep cycle was used for spectral analysis because it was most likely to provide a period of quiet
wakefulness with eyes closed and a stable epoch of slow wave sleep No attempt was made to control the sleep environment or characterize sleep architecture over the entire night. One record did not have an awake epoch without artifact. All subjects entered stable slow wave sleep. Only 3 of the 6 subjects entered into REM sleep, as scored using the signals from available electrodes. 3.2. Analysis of bipolar derivations 3.2.1. Power Power spectral density plots derived from bipolar derivations are presented in the top row of Fig. 2 The dominant 9 Hz peak present during wakefulness attenuated during REM sleep while power in the delta frequency range was similar during wakefulness and REM sleep. Slow wave sleep was accompanied by a marked increase in power below 5 Hz and a broad peak at 8 Hz at all locations. During slow wave sleep there was a broad peak at 14 Hz in frontal– temporal and frontal-central derivations but not the parietal– occipital derivation. 3.2.2. Coherence Coherence spectra derived from bipolar derivations are presented in the bottom row of Fig. 2. Coherence was measured between regions represented by three electrode pairs. Homologous left and right intra-hemispheric measurements were averaged to determine frontal–central to parietal-occipital coherence (FC–PO), frontal–central
Fig. 2. Power spectral density (top row) and coherence spectra (bottom row) derived from bipolar recordings while awake (left column), slow wave sleep (middle column), and rapid eye movement sleep (right column). Bipolar derivations included F3–C3 and F4–C4 (FC, frontal–central), F7–T3 and F8–T4 (FT, frontal–temporal), and P3–O1 and P4–O2 (PO, parietal–occipital). Spectra were averaged across hemisphere and then across subjects. Conservative estimates of the 95% confidence interval of selected peak values are show by vertical lines. Only the lower portion of each confidence interval is shown for clarity.
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
to frontal-temporal coherence (FC–FT), and frontal–temporal to parietal-occipital coherence (FT–PO). During wakefulness, peaks of coherence were present at 9 Hz in FC–FT and FC–PO pairs and possibly the FT–PO pair. During slow wave sleep a peak of coherence was present at 14 Hz in the FC–PO pair and at 12 Hz in the FC–FT pair. The FC–FT pair also had a peak at 1.5 Hz. The FT–PO pair had no significant coherence. During REM sleep, broad-band coherence was present in the FC–FT pair, but there were no clear peaks. A possible peak at w4 Hz was present in the FC–PO pair. During each sleep state, FC–FT coherence was highest, as is expected when activity from nearby scalp regions is compared. Coherence in this pair was present over the entire frequency band and decreased with increasing frequency. This trend was not present when coherence was measured
1091
over longer distances, as the baselines of the other coherence spectra were flat. 3.3. Analysis of common reference derivations 3.3.1. Power Power spectral density plots derived from common reference derivations are presented in the top row of Fig. 3 Spectral power during wakefulness and REM sleep contained patterns similar to those seen with bipolar derivations. Residual 9 Hz activity during REM sleep was only present at the parietal and occipital sites. Slow wave sleep had maximum spectral power below 5 Hz and broad peaks at 8 Hz at all sites. Isolated peaks were present at 10 Hz in the frontal and central sites and at 14 Hz at central, parietal, and occipital sites. Expanded displays of power
Fig. 3. Power spectral density (top row) and coherence spectra (middle and bottom rows) derived from common reference recordings while awake (left column), slow wave sleep (middle column), and rapid eye movement sleep (right column). Common electrode derivations included F3 F4 (F, frontal), T3 T4 (T, temporal), C3 C4 (C, central), P3 P4 (P, parietal), and O1 O2 (O, occipital) referred to the ipsilateral mastoid. Spectra were averaged across hemispheres and then across subjects. Conservative estimates of the 95% confidence interval of selected peak values are show by vertical lines. Only the lower portion of each confidence interval is shown for clarity.
1092
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
spectra are available as supplementary data on the publisher’s web site.
3.4. Comparison of bipolar and common reference derivations
3.3.2. Coherence Coherence spectra derived from common reference derivations are presented in the middle and bottom panels of Fig. 3. For each hemisphere, the ipsilateral mastoid electrode was used as the common reference. Again, homologous left and right hemisphere comparisons were averaged. Intra-hemispheric coherence was measured between all possible pairs of regions sampled by the restricted electrode set described in the methods. For display purposes, sequential comparisons were made of the frontal region to each of the other hemispheric sites with increasing electrode separation, i.e. frontal to central (F–C), frontal to temporal (F–T), frontal to parietal (F–P), and frontal to occipital (F–O). These comparisons are shown in the middle row of Fig. 3. Then, sequential comparisons were made of the occipital region to the other sites with increasing electrode separation (Epstein and Brickley, 1985). These comparisons are shown in the bottom row of Fig. 3. Coherence spectra derived from common reference recordings during wakefulness and slow wave sleep showed a progressive increase in the coherence value beginning between 12 and 15 Hz and rising to a plateau above 30 Hz. Based on our previous work, this pattern was recognized as the signature of reference contamination (Zaveri et al., 2000). However, both distinct peaks of coherence and coherence values indistinguishable from zero were present at frequencies below 15 Hz, indicating that reference contamination was minimal at these lower frequencies. During wakefulness there were peaks of scalp coherence at 3 Hz in all electrode pairs separated by one or two interelectrode distances. There was a consistent peak at 9 Hz in the occipital to frontal pair. During slow wave sleep, there were peaks at 1.5, 8, and 10 Hz when comparing frontal, temporal, central, and parietal sites. The 10 Hz peak predominated. However, these peaks were not present in the occipital to frontal and occipital to temporal pairs, where coherence below w12 Hz could not be distinguished from zero. The 1.5 and 8 Hz peaks were present in the occipital to parietal pair, but a 14 Hz peak replaced the 10 Hz peak. This 14 Hz peak was also present in the occipital to central pair. Again, when the occipital site was compared with the remaining regions, coherence below w12 Hz could not be distinguished from zero. During REM sleep, frontal, temporal, central, and parietal comparisons had coherence peaks at 3 and 7 Hz, but these peaks were less distinct than those found during other states. However, a prominent peak was present at 3 Hz in the occipital to parietal pair. During REM sleep, when muscle tone is reduced, the progressive increase in coherence at frequencies above 15 Hz, ascribed to reference contamination during wakefulness or slow wave sleep, was not present.
During wakefulness, coherence spectra from all bipolar derivations and the occipital to frontal common reference derivation showed a peak at 9 Hz. The difference between coherence spectra from bipolar and common reference derivations during wakefulness was confined to the lack of a substantive peak below 5 Hz in bipolar derivations. During slow wave sleep a well defined coherence peak at 14 Hz in the bipolar frontal–central to parietal–occipital (FC–PO) pairing was mirrored by coherence peaks at the same frequency in common reference derivations that compared the occipital region to the parietal or central regions but not the frontal or temporal regions. During slow wave sleep a coherence peak at 10 Hz was present in common reference recordings that compared the frontal to the central, parietal, or temporal regions but not the occipital region. A coherence peak at 10 Hz was not present in bipolar comparisons. Using common reference derivations there were peaks at 8 Hz and below 5 Hz in all pairs except the occipital to frontal (O–F) pair and the occipital to temporal (O–T) pair. A peak below 5 Hz was present in the bipolar frontal-central to frontal–temporal (FC–FT) comparison, but there was no clear peak at 8 Hz. There was a broad peak in this bipolar comparison at 12 Hz that did not correlate well with any peak in reference derivations. A summary of the distribution of spectral peaks of EEG power and coherence during slow wave sleep is shown in Fig. 4. Only 3 subjects provided REM sleep for analysis, and, except for a 3 Hz peak in the occipital to parietal pair, peaks present in coherence spectra from common reference
Fig. 4. The distribution of spectral peaks of EEG power and coherence during slow wave sleep. During slow wave sleep, significant coherence at 10 Hz is present in frontal, central, temporal, and parietal regions. Significant coherence at 14 Hz is present in central, parietal, and occipital regions. Spectral power at 8 Hz was present diffusely in both bipolar and reference recordings. A peak of coherence was present at 8 Hz in all common reference electrode comparisons except the frontal-occipital pair. There was no coherent activity at 8 Hz in bipolar recordings. MSC, magnitude squared coherence; F, frontal; C, central; P, parietal; O, occipital; T, temporal.
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
recordings were poorly defined. Even less structure was present in coherence spectra from bipolar derivations. The coherence spectra for each subject during wakefulness, slow wave sleep, and rapid eye movement sleep derived from bipolar and reference derivations are available as supplementary data on the publisher’s web site.
4. Discussion Our results confirm the findings of previous investigators who used longitudinal bipolar electrode derivations to study the sleep EEG (Achermann and Borbe´ly, 1998; Dumermuth et al., 1972). During slow wave sleep, EEG power is concentrated below 5 Hz while regional coherence within the hemisphere is present in restricted bands at higher frequencies. Our analysis of slow wave sleep using common reference and bipolar recordings expands these findings by localizing intra-hemispheric coherence at 10 Hz to frontal, temporal, central, and parietal regions and intra-hemispheric coherence at 14 Hz to occipital, parietal, and central regions. While Achermann and Borbe´ly (1998) combined the 4 non-REM sleep stages for analysis, we concentrated on slow wave sleep because of its relative stability. Their bipolar comparison of frontal-central to parietal–occipital regions demonstrated a major coherence peak at 14 Hz and a minor coherence peak at 10 Hz. Our mastoid-reference recordings during slow wave sleep have a spectral power peak at 10 Hz at frontal and central electrodes but not at the remaining locations. Subtraction of activity at this frequency in frontalcentral regions by the bipolar method and lower power at this frequency in posterior regions could explain reduced anterior–posterior coherence at 10 Hz in the bipolar study of Achermann and Borbe´ly and its complete absence in our bipolar findings. This interpretation is supported by our finding coherence peaks at 10 Hz in comparison of common reference recordings from anterior but not posterior electrodes (middle column of Fig. 3). During slow wave sleep a coherence peak at 14 Hz is present in both our bipolar and common reference recordings. It was also present in the bipolar recordings of Achermann and Borbe´ly (1998). This is best explained by a prominent gradient of activity at that frequency centered on central and parietal regions and extending to the occipital region but not the frontal region. This formulation is supported by our power spectra that show a peak at 14 Hz with maximum amplitude in the central lead followed by the parietal and then occipital leads but absent in frontal and temporal leads. Activity at that frequency would appear in the bipolar frontal–central derivation because of its presence in the central lead. Activity at that frequency would appear in the parietal– occipital derivation because of the amplitude gradient, but to a lesser degree than in the frontal region. This formulation is supported by our bipolar power spectra.
1093
The presence of 14 Hz activity during slow wave sleep is consistent with the general observation that sigma-band power is highest during stage 2 sleep and progressively declines during stages 3 and 4. However, sigma-band power is still higher in slow wave sleep than in stage 1 or REM sleep (Dijk et al., 1993; Johnson et al., 1969). The majority of this power is usually attributed to episodic spindle activity appearing in the analysis window of the Fourier transform used to estimate spectral power. Sleep spindles are reported to appear as often in slow wave sleep as in stage 2 (Gaillard and Blois, 1981) but with less individual power (Dijk et al., 1993). However, the opposite finding, equal amplitude but fewer number, has also been reported (Zeitlhofer et al., 1997). This could be explained by the considerable variation in spindle characteristics determined by automated methods. It is more likely that the amplitude of sigma activity is continuously modulated and highly variable during non-REM sleep (Mo¨lle et al., 2002; Zygierewicz et al., 1999), resulting in characterizations that are dependent on the somewhat arbitrary selection of detection parameters. Dumermuth et al. (1972) considered sigma activity the essential component of slow wave sleep, whether it appeared as spindles or as continuous activity. It is accepted that sleep spindles appear at different frequencies with differing topography. Gibbs and Gibbs (1951) described three types of spindles during sleep: 14 Hz spindles dominant in parietal areas; 12 Hz spindles dominant in frontal areas; and 10 Hz spindles that were ‘generalized’ (frontal and parietal). The topographic distribution of the 14 and 12 Hz spindles described by the Gibbs has been confirmed, although most investigators report center frequencies 0.5 Hz lower (Jobert et al., 1992; Scheuler et al., 1990a; Werth et al., 1997; Zygierewicz et al., 1999). Our observations extend the topography of the 14 Hz activity into slow wave sleep and add that this activity has increased regional coherence at central, parietal, and occipital sites. However, we cannot address the continuous or episodic nature or the amplitude envelope of this sigma activity. We do not find a peak of signal power at 12 Hz during slow wave sleep. It is possible that signal power at this frequency is lost or that the center frequency of this activity decreases as sleep advances beyond stage 2. Others studying sigma power during sleep analyzed stage 2 exclusively or combined all stages of non-REM sleep, so the persistence of a 12 Hz rhythm into slow wave sleep cannot be confirmed. Given the within-subject stability of the characteristic frequency of other sleep-related patterns, the correspondence of our 10 Hz peak with the 12 Hz rhythm seem unlikely. However, Zygierewicz et al. (1999), using matching pursuit decomposition of overnight sleep recordings into spindle-shaped envelopes, also found 10.2 Hz frontal spindles in at least one subject. While the designation of 10 Hz activity during sleep as a spindle pattern is controversial (Jankel and Niedermeyer, 1985), the presence of 10 Hz activity with frontal maximum
1094
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
during slow wave sleep is clear (Dumermuth et al., 1972; Hauri and Hawkins, 1973; Pivik and Harman, 1995; Scheuler et al., 1983). However, it may only present a recognizable pattern in 10–30% of subjects (Gibbs and Gibbs, 1951; Scheuler et al., 1990b) and must be differentiated from arousal. We find peaks of mean spectral power at 8 and 10 Hz in common reference recordings during slow wave sleep. The 8 Hz peak is diffuse and the 10 Hz peak is frontal-central. Post hoc inspection of the power spectra from individual subjects showed the 10 Hz peak was exclusive to 4 subjects and the 8 Hz peak was exclusive to the other 2 subjects, indicating that the 10 Hz activity is not a universal finding. The global 8 Hz rhythm could be derived from the 9 Hz awake alpha and represent unrecognized arousals. However, post hoc review of the EEG showed nearly continuous low amplitude alpha-range activity and was inconsistent with arousals. Also, this activity did not match exemplars of alpha-sleep or alphadelta sleep found in the literature (Hauri and Hawkins, 1973; Scheuler et al., 1983). It is possible that the characteristic frequency of the frontal sigma rhythm, nominally 11.5 Hz, drops below the traditional sigma-frequency range during slow wave sleep. Of greater interest is the localization of increased coherence at 10 Hz during slow wave sleep to frontal, temporal, central, and parietal regions in contrast to the localization of increased coherence at 14 Hz to central, parietal, and occipital regions. The segregation of these regional interactions becomes apparent when common reference recordings are used. It is our contention that coherence measurements from bipolar recordings are confounded by the independence of the signals probed by each element of the bipolar pair of scalp electrodes, as we argue for intracranial recording (Zaveri et al., submitted). Scalp coherence measurements may also be confounded when widely distributed cortical rhythms are minimized by common mode subtraction. We accept the argument that common reference recordings allow contamination of the coherence measurement by a common signal. However, this contamination is dependent on the signal-to-noise ratio and can be recognized by its characteristic spectral pattern, allowing conclusions to be drawn from spectral segments that show less or no contamination (Zaveri et al., 2000). The subjects providing data for this analysis were not screened for sleep disorders or habituated to the recording equipment, and no attempt was made to quantify the architecture of their overnight recording. The uncontrolled use of medications or the presence of either an acute or chronic sleep disturbance would affect the spectral character of the sleep recordings. For example, even modest sleep deprivation can vary the duration of sleep stages (Webb and Agnew, 1971), increase the power spectral density in the delta frequency band (Borbe´ly et al., 1981), and change the incidence and magnitude of sleep spindles, depending on their frequency (Knoblauch et al., 2003).
However, the intent of this study was not to detect changes in sleep architecture but to characterize the presence and distribution of the most obvious peaks in the coherence spectra and contrast those derived from bipolar and common reference electrode derivations. To this end, the sustained presence of slow wave sleep was the essential requirement of the recording. It is reassuring that the results from our bipolar recordings reproduce findings from studies with stricter limits on subject recruitment (Achermann and Borbe´ly, 1998; Dumermuth et al., 1972). Of those medications taken by two of the six subjects, fluoxetine had the greatest potential influence on the EEG during sleep (Ebert and Kierch, 1999; Feige et al., 2002; Salinsky et al., 2002). Feige et al. (2002) reported that subchronic administration of fluoxetine increased EEG power at frequencies above 10 Hz during non-REM sleep with more than a 50% increase in EEG power in a distinct spectral peak at 14 Hz. Post-hoc examination of the power spectra of our single subject taking fluoxetine (subject D, spectra available as supplementary data) showed no obvious differences from the spectra of the other subjects. The same was true of the coherence spectra derived from common reference recordings. For example, the coherence peak at 14 Hz in the O–P derivation during slow wave sleep was present in the spectrum of each subject with individual peak values ranging from 0.49 to 0.70. The peak value for the subject taking fluoxetine was 0.55. However, in the coherence spectra derived from bipolar recordings, coherence peak values at 14 Hz in the FC–PO derivation during slow wave sleep ranged from 0.15 to 0.32. The value for the subject taking fluoxetine was 0.22. It is possible that fluoxetine affected the mean amplitude of this peak, but since all other subjects had a peak at that frequency, the presence of fluoxetine did not cause that peak to appear. Also, this peak was reported by Achermann and Borbe´ly (1998), and their subjects were not taking medications. We agree with Dumermuth et al. (1972) that sigma-band activity is an essential component of slow wave sleep. Coherence peaks in that band are organized in central, parietal, and occipital regions and must be contrasted with a separate band of coherent activity at 10 Hz in frontal, temporal, central, and parietal regions. The presence of peaks indicates an underlying structure to EEG patterns. The differential segregation of these coherence peaks suggests the independence of these patterns. The relationship of this 10 Hz activity to frontal spindles and to the thalamocortical circuits presumed to underlie their genesis remains to be elucidated.
Acknowledgements The authors recognize the technical contributions of Kevin Gaski.
R.B. Duckrow, H.P. Zaveri / Clinical Neurophysiology 116 (2005) 1088–1095
Appendix. Supplementary Material Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.clinph.2004. 12.002
References Achermann P, Borbe´ly AA. Low-frequency (!1 Hz) oscillations in the human sleep electroencephalogram. Neuroscience 1997;81(1):213–22. Achermann P, Borbe´ly AA. Coherence analysis of the human sleep electroencephalogram. Neuroscience 1998;85(4):1195–208. Amzica F, Steriade M. The K-complex: its slow (!1-Hz) rhythmicity and relation to delta waves. Neurology 1997;49(4):952–9. Borbe´ly AA, Baumann F, Brandeis D, Strauch I, Lehmann D. Sleep deprivation: Effect on sleep stages and EEG power density in man. Electroenceph Clin Neurophysiol 1981;51(5):483–93. Carter GC, Knapp CH, Nuttall AH. Statistics of the estimate of the magnitude-coherence function. IEEE Transactions on Acoustics, Speech, and Signal Processing 1973;21(4):388–9. Dijk D-J, Hayes B, Czeisler CA. Dynamics of electroencephalographic sleep spindles and slow wave activity in men: effect of sleep deprivation. Brain Res 1993;626(1–2):190–9. Dumermuth G, Walz W, Scollo-Lavizzari G, Kleiner B. Spectral analysis of EEG activity in different sleep stages in normal adults. Eur Neurol 1972;7(5):265–96. Ebert U, Kierch W. Effects of captopril and enalapril on electroencephalogram and cognitive performance in healthy volunteers. Eur J Clin Pharmacol 1999;55(4):255–7. Epstein CM, Brickley GP. Interelectrode distance and amplitude of the scalp EEG. Electroenceph Clin Neurophysiol 1985;60:287–92. Feige B, Voderholzer U, Riemann D, Dittmnn R, Hohagen F, Berger M. Fluoxetine and sleep: effect of a single dose, subchronic threatment, and discontinuation in healthy subjects. Neuropsychopharmacol 2002; 26(2):246–58. Finelli LA, Borbe´ly AA, Achermann P. Functional topography of the human nonREM sleep electroencephalogram. Eur J Neurosci 2001; 13(12):2282–90. French CC, Beaumont JG. A critical review of EEG coherence studies of hemisphere function. Int J Psychophysiol 1984;1(3):241–54. Gaillard J-M, Blois R. Spindle density in sleep of normal subjects. Sleep 1981;4(4):385–91. Gibbs FA, Gibbs EL.. 2nd ed Atlas of electroencephalography. vol. 1. Reading MA: Addison-Wesley Publishing Company; 1951 p. 93–94. Hauri P, Hawkins DR. Alpha-delta sleep. Electroenceph Clin Neurophysiol 1973;34(3):233–7. Jankel WR, Niedermeyer E. Sleep spindles. J Clin Neurophysiol 1985;2(1): 1–35. Jenkins GM, Watts DG. Spectral analysis and its applications. San Francisco, CA: Holden-Day; 1968. Jobert M, Poiseau E, Ja¨hnig P, Schulz H, Kubicki S. Topographical analysis of sleep spindle activity. Neuropsychobiology 1992;26(4):210–7.
1095
Johnson L, Lubin A, Naitoh P, Nute C, Austin M. Spectral analysis of the EEG of dominant and non-dominant alpha subjects during waking and sleeping. Electroenceph Clin Neurophysiol 1969;26(4):361–70. Knoblauch V, Martens WLJ, Wirz-Justice A, Cajochen C. Human sleep spindle characteristics after sleep deprivation. Clin Neurophysiol 2003; 114(12):2258–67. Mo¨lle M, Marshall L, Gais S, Born J. Grouping of spindle activity during slow oscillations in human non-rapid eye movement sleep. J Neurosci 2002;22(4):10941–7. Pivik RT, 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(3):131–7. Radtke RA. Sleep disorders. In: Daly DD, Pedley TA, editors. Current practice of clinical electroencephalography. 2nd ed. New York, NY: Raven Press; 1990. p. 574–5. Salinsky MC, Binder LM, Oken BS, Storzbach D, Aron CR, Dodrill CB. Effects of gabapentin and carbamazepine on the EEG and cognition in healthy volunteers. Epilepsia 2002;43(5):482–90. Scheuler W, Stinshoff D, Kubicki S. The alpha-sleep pattern. Differentiation from other sleep patterns and effect of hypnotics. Neuropsychobiology 1983;10(2–3):183–9. Scheuler W, Kubickis, Scholz G, Marquardt J. Two different activities in the sleep spindle frequency band-discrimination based on the topographical distribution of spectral power and coherence. In: Horne J, editor. Sleep 90. Bochum: Pontenagel Press; 1990a. p. 13–6. Scheuler W, Rappelsberger P, Schmatz F, Pastelak-Price C, Petsche H, Kubicki S. Periodicity analysis of sleep EEG in the second and minute ranges-example of application in different alpha activities in sleep. Electroenceph Clin Neurophysiol 1990b;76(3):222–34. Steriade M, McCormick DA, Sejnowski TJ. Thalamocortical oscillations in the sleeping and aroused brain. Science 1993;262:679–85. Steriade M, Nunez A, Amzica F. A novel slow (!1 Hz) oscillation of neocortical neurons in vivo: depolarizing and hyperpolarizing components. J Neurosci 1993;13(8):3252–65. Steriade M, Contreras D, Amzica F. Synchronized sleep oscillations and their paroxysmal developments. Trends Neurosci 1994;17(5):199–208. Webb WB, Agnew HW. Stage 4 sleep: influence of time course variables. Science 1971;174(4016):1354–6. Werth E, Acherman P, Dijk D-J, Borbe´ly AA. Spindle frequency activity in the sleep EEG: individual differences and topographic distribution. Electroenceph Clin Neurophysiol 1997;103:535–42. Zaveri HP, Williams WJ, Sackellares JC, Baydoun A, Duckrow RB, Spencer SS. Measuring the coherence of intracranial electroencephalograms. Clin Neurophysiol 1999;110(10):1717–25. Zaveri HP, Duckrow RB, Spencer SS. The effect of a scalp reference signal on coherence measurements of intracranial electroencephalograms. Clin Neurophysiol 2000;111(7):1293–9. Zaveri HP, Duckrow RB, Spencer SS. On the use of bipolar montages for the time-series analysis of intracranial electroencephalograms. Submitted. Zeitlhofer J, Gruber G, Anderer P, Asenbaum S, Schimicek P, Saletu B. Topographic distribution of sleep spindles in young healthy subjects. J Sleep Res 1997;6(3):149–55. Zygierewicz J, Blinowska KJ, Durka PJ, Szelenberger W, Niemcewicz S, Androsiuk W. High resolution study of sleep spindles. Clin Neurophysiol 1999;110(12):2136–47.