An electroencephalographic fingerprint of human sleep Luigi De Gennaro, a, T Michele Ferrara, a,b Fabrizio Vecchio, c,d Giuseppe Curcio, a and Mario Bertini a a Department of Psychology, University of Rome BLa Sapienza[, Via dei Marsi 78, Roma, 00185 Italy b Department of Internal Medicine and Public Health, University of BL’Aquila[, Italy c Department of Human Physiology and Pharmacology, High Resolution Section, University of Rome BLa Sapienza[, Italy d AFaR.-Department of Neuroscience, S. Giovanni Calibita, Fatebenefratelli Isola Tiberina, Rome, Italy Received 29 July 2004; revised 21 December 2004; accepted 13 January 2005 Available online 3 March 2005 Abstract Homeostatic and circadian processes are basic mechanisms of human sleep which challenge the common knowledge of large individual variations in sleep need or differences in circadian types. However, since sleep research has mostly focused on group measures, an approach which emphasizes the similarities between subjects, the biological foundations of the individual differences in normal sleep are still poorly understood. In the present work, we assessed individual differences in a range of EEG frequencies including sigma activity during non-REM sleep (8.0Y15.5 Hz range) in a group of 10 subjects who had participated in a slow-wave sleep (SWS) deprivation study. We showed that, like a Bfingerprint[, a particular topographic distribution of the electroencephalogram (EEG) power along the antero-posterior cortical axis distinguishes each individual during non-REM sleep. This individual EEG-trait is substantially invariant across six consecutive nights characterized by large experimentally induced changes of sleep architecture. One possible hypothesis is that these EEG invariances can be related to individual differences in genetically determined functional brain anatomy, rather than to sleep- dependent mechanisms. D 2005 Elsevier Inc. All rights reserved. Keywords: Sleep spindles; Sigma EEG activity; EEG topography; Individual differences Introduction Notwithstanding the increasing use of neuroimaging techni- ques, electroencephalography is still the most universally employed technique in human sleep research. Quantitative analyses of sleep EEG by spectral analysis have led to the development of the 2-process model of sleep regulation (Borbely, 1982). According to this model, the timing of sleep and wakeful- ness is regulated by the interaction of a homeostatic, sleepYwake- dependent Process S and a circadian, sleepYwake-independent Process C (Borbely, 1982). Well-established evidence on the homeostatic facet of sleep regulation suggests that slow-wave activity (EEG power in the 0.75Y4.5 Hz range) depends on the duration of previous sleep and wakefulness, representing a marker of non-REM sleep intensity (Borbely and Achermann, 2000). This feature of sleep has been consistently shown in a broad range of species, including humans, cats, mice, rats, and squirrels (Tobler, 1995). Manipulations of sleep intensity by means, for example, of sleep deprivation, lead to clear homeostatic recovery processes (Borbely and Achermann, 2000) which do not involve the whole cerebral cortex in the same manner. Indeed, these recovery processes are local in nature, as shown in dolphins (Oleksenko et al., 1992), birds (Rattenborg et al., 1999), mice (Huber et al., 2000), rats (Vyazovskiy et al., 2002), and humans (Ferrara et al., 2002; Finelli et al., 2001a). The most striking regional phenom- enon of the human sleep EEG is the hyperfrontality of low- frequency EEG activity during baseline and post deprivation non- REM sleep, probably due to a high Brecovery need[ of the frontal heteromodal association areas of the cortex (Cajochen et al., 1999; Ferrara et al., 2002; Finelli et al., 2001a). Homeostatic, circadian and regional EEG changes are basic mechanisms of human sleep that seem to challenge the common notion of large individual variations in sleep need or differences in circadian types. These differences are often ignored or considered as experimental noise, to be actively suppressed through the use of statistical methods that emphasize group rather than individual results. Nevertheless, the recent extensive use of blood flow imaging techniques in the neurosciences made it clear that, for example, a large variability in the detected hemodynamic responses across sessions of the same subject and across subjects is actually present (Aguirre et al., 1998; Handwerker et al., 2004; Wei et al., 2004). The importance of taking such individual differences into account has recently been pointed out also in the sleep research field, at least as regards individual variability in the susceptibility to sleep deprivation (Bell-McGinty et al., 2004; Leproult et al., 1053-8119/$ - see front matter D 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2005.01.020 T Corresponding author. Fax: +39 06 4451667. E-mail address: [email protected] (L. De Gennaro). Available online on ScienceDirect (www.sciencedirect.com). www.elsevier.com/locate/ynimg NeuroImage 26 (2005) 114 Y 122
9
Embed
An electroencephalographic fingerprint of human sleep
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
www.elsevier.com/locate/ynimg
NeuroImage 26 (2005) 114Y122
An electroencephalographic fingerprint of human sleep
Luigi De Gennaro,a,T Michele Ferrara,a,b Fabrizio Vecchio,c,d
Giuseppe Curcio,a and Mario Bertinia
aDepartment of Psychology, University of Rome BLa Sapienza[, Via dei Marsi 78, Roma, 00185 ItalybDepartment of Internal Medicine and Public Health, University of BL’Aquila[, ItalycDepartment of Human Physiology and Pharmacology, High Resolution Section, University of Rome BLa Sapienza[, ItalydAFaR.-Department of Neuroscience, S. Giovanni Calibita, Fatebenefratelli Isola Tiberina, Rome, Italy
Received 29 July 2004; revised 21 December 2004; accepted 13 January 2005
Available online 3 March 2005
Abstract
Homeostatic and circadian processes are basic mechanisms of human
sleep which challenge the common knowledge of large individual
variations in sleep need or differences in circadian types. However,
since sleep research has mostly focused on group measures, an
approach which emphasizes the similarities between subjects, the
biological foundations of the individual differences in normal sleep are
still poorly understood. In the present work, we assessed individual
differences in a range of EEG frequencies including sigma activity
during non-REM sleep (8.0Y15.5 Hz range) in a group of 10 subjects
who had participated in a slow-wave sleep (SWS) deprivation study.
We showed that, like a Bfingerprint[, a particular topographic
distribution of the electroencephalogram (EEG) power along the
antero-posterior cortical axis distinguishes each individual during
non-REM sleep. This individual EEG-trait is substantially invariant
across six consecutive nights characterized by large experimentally
induced changes of sleep architecture. One possible hypothesis is that
these EEG invariances can be related to individual differences in
genetically determined functional brain anatomy, rather than to sleep-
Available online on ScienceDirect (www.sciencedirect.com).
ness is regulated by the interaction of a homeostatic, sleepYwake-dependent Process S and a circadian, sleepYwake-independentProcess C (Borbely, 1982). Well-established evidence on the
homeostatic facet of sleep regulation suggests that slow-wave
activity (EEG power in the 0.75Y4.5 Hz range) depends on the
duration of previous sleep and wakefulness, representing a marker
of non-REM sleep intensity (Borbely and Achermann, 2000). This
feature of sleep has been consistently shown in a broad range of
species, including humans, cats, mice, rats, and squirrels (Tobler,
1995). Manipulations of sleep intensity by means, for example, of
sleep deprivation, lead to clear homeostatic recovery processes
(Borbely and Achermann, 2000) which do not involve the whole
cerebral cortex in the same manner. Indeed, these recovery
processes are local in nature, as shown in dolphins (Oleksenko et
al., 1992), birds (Rattenborg et al., 1999), mice (Huber et al.,
2000), rats (Vyazovskiy et al., 2002), and humans (Ferrara et al.,
2002; Finelli et al., 2001a). The most striking regional phenom-
enon of the human sleep EEG is the hyperfrontality of low-
frequency EEG activity during baseline and post deprivation non-
REM sleep, probably due to a high Brecovery need[ of the frontal
heteromodal association areas of the cortex (Cajochen et al., 1999;
Ferrara et al., 2002; Finelli et al., 2001a).
Homeostatic, circadian and regional EEG changes are basic
mechanisms of human sleep that seem to challenge the common
notion of large individual variations in sleep need or differences in
circadian types. These differences are often ignored or considered
as experimental noise, to be actively suppressed through the use of
statistical methods that emphasize group rather than individual
results. Nevertheless, the recent extensive use of blood flow
imaging techniques in the neurosciences made it clear that, for
example, a large variability in the detected hemodynamic responses
across sessions of the same subject and across subjects is actually
present (Aguirre et al., 1998; Handwerker et al., 2004; Wei et al.,
2004). The importance of taking such individual differences into
account has recently been pointed out also in the sleep research
field, at least as regards individual variability in the susceptibility
to sleep deprivation (Bell-McGinty et al., 2004; Leproult et al.,
recovery (REC). Adaptation is the first recording night, usually
disregarded for any data analysis in sleep research since it is
characterized by increased duration of stage 1 and intra-sleep wake,
lowered sleep efficiency, and increased latency to both slow-wave
(SWS) and rapid eye movement (REM) sleep. The second and third
nights were considered as baseline sleep, the only difference
between them being that in the third night the subjects were
awakened twice, and a psychophysiological test battery, lasting
about 13 min, was administered (Ferrara et al., 2000). The first
night-time awakening was scheduled after 2 h and the second after
5 h of accrued sleep. Similar experimental awakenings were present
also during nights 4Y6. During the two selective SWS deprivation
nights, two experimenters continuously monitored the EEG chart
and delivered a tone (frequency: 1000 Hz; intensity: 40Y110 dB) bypressing a button whenever at least 2 delta waves (e4 Hz; N75 AV),determined by visual inspection, appeared in a 15-s recording
interval. The intensity of acoustic stimuli began from the lowest
intensity, and it was increased in steps of 5 dB if no response
occurred (sleep stage shift, K complex, EEG desynchronization,
alpha burst, muscle tone increase, and slow eye movements). In this
manner, we prevented the subject from fully entering stage 3 by
lightening his sleep and carefully avoiding full awakenings.
Finally, during the sixth night, the subjects were allowed a recovery
sleep, undisturbed except for the two abovementioned awakenings.
Every night, subjects arrived in the laboratory at about 9:00
p.m. for electrode hook-up. Lights were turned off and poly-
graphic sleep recordings always started at 11:30 p.m. (T30 min)
and ended after 7.5 h of accumulated sleep. Wrist actigraphic
recordings (AMI motion logger 16 K) monitored the participants
to avoid any napping and strenuous physical exercise throughout
the experiment.
Sleep recording
An Esaote Biomedica VEGA 24 polygraph set at a paper speed
of 10 mm/s was used for polygraphic recordings. EEG signals were
high pass filtered with a time constant of 0.3 s and low pass filtered
at 30 Hz (30 dB/octave); unipolar EEGs were recorded according
to the international 10Y20 system: Fz-A1, Cz-A1, Pz-A1, Oz-A1,
and C3-A2. The submental electromyogram (EMG) was recorded
with a time constant of 0.03 s, and bipolar horizontal and vertical
eye movements were recorded with a time constant of 1 s.
Fig. 1. Group analysis: Mean EEG power values (expressed in z scores, and
averaged over all nights) in the 8.0Y15.5 Hz range on the antero-posterior
scalp locations (shown in different colors). At the bottom of the panel, the
one-way ANOVA results ( F values, df = 3,27) comparing the four EEG
derivations for each frequency bin are also reported. The dotted red line
indicates the level of statistical significance ( P e 0.05).
L. De Gennaro et al. / NeuroImage 26 (2005) 114Y122 117
Results
Group analysis
Although analyzed at a higher frequency resolution and also
taking the adaptation night into account, the present group analyses
of EEG topography strictly parallel those previously published
(Ferrara et al., 2002): hence, they will be briefly summarized here.
The results point to significant differences in a wide range of EEG
frequencies, comparing both the antero-posterior scalp locations
and the different recording nights. Regional changes illustrated in
Fig. 1 point to a prevalence of EEG power from 12.75 to 14.50 Hz
on central and especially on parietal scalp locations2, while the
8.00Y11.75 Hz power is higher on the frontal site3. The first range
mostly corresponds to the frequency limits of sigma rhythm, which
is characterized by a centro-parietal prevalence (De Gennaro and
Ferrara, 2003; Finelli et al., 2001a). The second range corresponds
to the alpha rhythm peaking on frontal sites during sleep (Cajochen
et al., 1999; Ferrara et al., 2002; Finelli et al., 2001a).
Further group analyses took the between-night differences into
account. Deprivation nights were characterized by a complete
selective SWS suppression achieved by acoustic stimulation [mean
intensity of acoustical stimuli to suppress SWS was 56.03 dB
(T14.12) in deprivation (1) and 72.1 dB (T11.45) in deprivation
(2)]. Finally, homeostatic effects increased both visually scored
SWS amount and quantitative EEG slow frequencies during the
recovery night (Ferrara et al., 1999, 2002). The ANOVAs showed
that the between-night differences in the considered EEG range are
due to a power decrease in the 9.00 to 10.50 Hz range during the
two SWS-deprivation nights, while the differences in the
13.25Y14.50 Hz range delineate a more complex picture (Figs.
1b and 2). This frequency range is characterized by two main
phenomena: the power increase during the first SWS-deprivation
night as compared to the other nights and a decrease during the
adaptation night as compared to both SWS-deprivation nights.
Furthermore, the recovery night also showed a decrease at 13.25
Hz as compared to both SWS-deprivation nights.
As an example of factors hugely affecting EEG power in the
considered frequency range, Fig. 3 shows three different functional
changes concerning the slower frequencies recorded at the frontal
site: (a) the power increase as a consequence of SWS deprivation;
(b) the lower power during the second with respect to the first half
of the baseline night; and (c) the anterior predominance as
compared to more posterior sites (i.e., Oz). The homeostatic
effects during the recovery night are confirmed by the alpha power
increase in the 8.00 to 10.25 Hz range, as indexed by the
significant one-way ANOVA comparisons for these frequency
bins. Another way to highlight the effects of sleep propensity on
the EEG at the Fz derivation is by comparing the first and second
half of the baseline night (i.e., the 2nd night), since sleep
3 A short summary of the results of the Scheffe post hoc comparisons ( P G0.05) across this frequency range, within the 8.00Y9.00 Hz range, shows thatEEG power at Fz and Oz is significantly higher than at Cz and Pz, while Fz
showed a significant prevalence as compared to Oz, Pz, and Cz in the
9.25Y11.75 range.
2 A short summary of the results of the Scheffe post hoc comparisons ( P G
0.05) across this frequency range shows that EEG power at Cz, Pz, and Oz
leads was significantly higher than that at Fz; Pz, and also showed a
significant prevalence as compared to Oz, while Pz and Cz were not
significantly different.
propensity, as reflected by SWA, declines in the course of a sleep
night (e.g., Borbely, 1982). In this case, too, the one-way ANOVA
comparisons revealed significant differences in a wide frequency
range from 8.00 to 11.50 Hz, pointing to a decrease of alpha power
during the sleep episodes. Finally, the comparisons between the
frontal versus occipital derivations during an undisturbed sleep
night (i.e., the 2nd night) by one-way ANOVAs clearly point to a
higher power at Fz than at Oz across the whole 8.00Y12.25 Hz
range.
Individual case analysis
The analysis of individual cases showed a high topographical
invariance (Fig. 4). Five subjects (#1, #2, #3, #5, and #9) showed
one spectral peak in the range of spindle frequency (range =
12.25Y13.25 Hz) on all brain sites. The remaining subjects had
two peaks on different sites, with one peak again higher on CzYPz(range = 12.50Y13.75 Hz) and a second one higher on Fz (range =
9.00Y12.25 Hz). This second peak coincided for frequency range
and for topographical distribution with the so-called Bslow[spindles at ¨12 Hz, being more pronounced over the frontal sites
(e.g., Anderer et al., 2001; Werth et al., 1997; Zeitlhofer et al.,
1997; Zygierewicz et al., 1999). In these five subjects, the Bslow[spindle activity shows different bounds in different subjects, and
these frequencies mostly overlap with the limits of the alpha
rhythm. Studies on sigma activity, bulking the lower bound of
spindle frequency in humans from the standard 12 Hz range
(Rechtschaffen and Kales, 1968), and thus including frequency
bins traditionally considered to be part of the alpha band, have
Fig. 2. Group analysis: Mean EEG power values (expressed in z scores, and
averaged over all derivations) of the six nights (shown in different colors) in
the 8.0Y15.5 Hz range. At the bottom of the panel, the one-way ANOVA
results ( F values, df = 5,45) comparing the six recording nights for each
frequency bin are also reported. The dotted red line indicates the level of
statistical significance ( P e 0.05).
Fig. 3. Group analysis: Relative EEG power recorded on the frontal (Fz)
derivation in the 8.0Y15.5Hz range, expressed as the raw differences between
the respective values of the 6th and 3rd night (RecoveryYBaseline), of the 2ndand 1st half of the baseline night (2ndY1st half of the night) and of the frontaland occipital derivations (FzYOz). At the bottom of the panel, the horizontal
red bar shows the significant one-way ANOVA results ( F values, df = 1, 9)
by comparing the recovery and baseline sleep recorded on the Fz derivation
( P e 0.05) for each frequency bin. The blue bar indicates the significant one-
way ANOVA results ( F values, df = 1, 9) by comparing the 1st and 2nd half
of the baseline night on the Fz derivation ( P e 0.05) for each frequency bin.
The green bar shows the significant one-way ANOVA results ( F values, df =
1, 9) by comparing the Fz and Oz power of the baseline night ( P e 0.05) for
each frequency bin.
L. De Gennaro et al. / NeuroImage 26 (2005) 114Y122118
been confused by the use of group rather than individual
measures, since a similar approach on the current individual data
would result in an averaged peak centered at ¨11 Hz (see Fig. 1).
In fact, the frontal peaks of subjects #4, #6, #7, #8, and #10 range
from 9.00 Hz to 12.25 Hz and are centered around a frequency
lower than 12 Hz.
Statistical analysis shows that the scalp locations � EEG
frequency matrices were highly intercorrelated within each
individual. Individual mean correlation coefficients ranged from
r = 0.909 to r = 0.987 [grand mean r = 0.958 (T0.026)]. On the
other hand, the mean of all the possible intercorrelations between
different nights/subjects (N = 1350) yielded a coefficient of r =
0.501 (T0.304). These intercorrelations yielded an estimate of
between-subject similarity in the considered EEG frequency range.
The percentages of variance explained by the within-subject and by
the between-subject similarity were 91.8% and 25.1%, respec-
tively. In other words, a specific stability within each individual
explains more than 50% of variance in the antero-posterior EEG
topography across the 8.00Y15.50 Hz range. The statistical
comparison confirms the striking difference between inter- and
intra-individual EEG changes (z = 11.84; P = 0.22 * 10ej31). This
comparison refers to an overall correlation between the matrices
obtained by averaging (after Fisher’s z-transformation) the
correlation coefficient between all possible combinations of
different matrices/nights (1350 comparisons), excluding the differ-
ent nights of the same subject and the same night of different
subjects.
The further overall correlation between the matrices/nights
obtained by excluding only the different nights of the same subject
(N = 1620) yielded a coefficient of r = 0.502 (T0.300). The
percentage of variance explained by this between-subject similarity
was 25.2%. The statistical comparison again confirms the striking
difference between inter- and intra-individual EEG changes (z =
11.83; P = 0.25 * 10ej31).
Discussion
Individual differences in sigma EEG activity
This study showed that each individual is characterized by a
peculiar shape of the sleep EEG power spectra in the sigma range
and that this shape remains stable across different nights. The
striking invariance in the individual sleep EEG topography pattern
appears more noteworthy in the light of the different and
considerable modifications of sleep characteristics across the six
nights. In fact, the first night was characterized by the well-known
Bfirst night effect[ that usually leads sleep researchers to discard
the adaptation night from the data analysis (e.g., Curcio et al.,
2004). Moreover, the homeostatic mechanisms of sleep regulation
were dramatically challenged in the two SWS-deprivation nights,
in which 328.1 (T167.69) and 739.8 (T314.8) auditory stimuli were
delivered, respectively, in the fourth and fifth night, causing a
large, regionally-graded increase in slow EEG frequency power
Fig. 4. Individual case analysis: Individual EEG power values of the antero-posterior scalp locations (Fz, Cz, Pz, and Oz, using the same color coding as in Fig.
1) in the 8.0Y15.5 Hz range in a 0.25-Hz resolution for non-REM sleep episodes of ten subjects. Absolute power was transformed into z scores [note that the
range of y axis values is different from Fig. 1]. Each row represents one subject. For each subject, the mean Bmatrix correlation[ between frequency and EEG
power matrices (see Materials and methods) is reported on the right. Each column shows the data of one of the six consecutive nights [adaptation, baseline,
baseline with awakenings, deprivation (1), deprivation (2), and recovery].
L. De Gennaro et al. / NeuroImage 26 (2005) 114Y122120
L. De Gennaro et al. / NeuroImage 26 (2005) 114Y122 121
decrease of alpha power from the first to the second half of the
night (Finelli et al., 2001a). Even the increase of the so-called slow
spindles after sleep deprivation is explained by the increase of
alpha power after total (Cajochen et al., 1999; Finelli et al., 2001a)
or selective sleep deprivation (Ferrara et al., 2002).
The results in Fig. 3 strengthen this interpretation. In fact, the
analyses confirm the anterior prevalence of EEG activity corre-
sponding to the slow spindle frequency range, and they clearly
indicate that these frequencies actually show a power decrease
from the first to the second half of the night and a power increase
after SWS deprivation. However, what is similarly clear is that
these regional, intra-night and recuperative changes affect the so-
called slow spindle frequency like parallel identical changes affect
a broader range starting from 8.00 Hz.
Finally, it should be mentioned that also the dissociation as a
function of circadian factors has recently been explained by the cir-
cadian rhythm of spindle frequency (Knoblauch et al., 2003), instead
of by the different circadian oscillation for different sleep spindles.
As a general principle of parsimony, the electrophysiological
evidence of a single thalamocortical mechanism (namely, the
duration of the hyperpolarization-rebound sequence), the observa-
tion of individual differences, the complete overlap between some
functional changes in the alpha range and those at ¨12 Hz, all
converge to highlight that there is only one kind of EEG activity
corresponding to sleep spindles and that the ¨12 Hz EEG changes
(in fact, at frequencies less than 12 Hz) are components of alpha
activity.
Conclusions
Average group measures, which emphasize the similarities
between subjects, are only one, albeit the most used, of the possible
approaches in the study of the neurophysiological correlates of
human sleep. However, this approach does not account for the
repertoire of individual patterns that can be important for under-
standing the variety of ways in which human brain organization
underlies behavior. Our results support the view of a genetic
control of sleep-related oscillations also in humans, and they
provide an important basis for future neurophysiological and
genetic studies, such as comparing the sleep EEG topography of
dizygotic and monozygotic twins.
Acknowledgment
This work was supported by the MIUR grant Finanziamento
per le Ricerche di Ateneo 2003.
References
Aeschbach, D., Borbely, A.A., 1993. All-night dynamics of the human