An electroencephalographic fingerprint of human sleep

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-

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-

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).

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.,

L. De Gennaro et al. / NeuroImage 26 (2005) 114Y122 115

2003; Van Dongen et al., 2004). However, although inter-

individual differences in neurobehavioral deficits from sleep loss

constitute a differential vulnerability trait, its neurobiological

correlates have yet to be discovered. The elucidation of some

biological mechanisms for the behavioral trait of morning-

nessYeveningness (Duffy et al., 2001), for individual differences

in the circadian pacemaker program (i.e., long and short sleepers)

(Aeschbach et al., 1996, 2001, 2003) and for the variability of

sleep duration in the general population (Aeschbach et al., 2003),

also encourages a neurophysiological approach to individual

differences in sleep characteristics.

Sleep spindles seem to be a natural candidate for this analysis:

they are one of the hallmarks of non-REM sleep, one of the few

transient EEG events which are unique to sleep, and it has been

reported that their incidence shows great inter-individual differ-

ences in humans (Werth et al., 1997). As far as their electro-

physiological mechanisms are concerned, sleep spindles depend on

variations in membrane potentials of thalamocortical neurons that

oscillate in the frequency range of spindles at an intermediate level

of hyperpolarization (Steriade, 1999). At the macroscopic EEG

level, spindle frequency in humans encompasses the 12Y14 Hz

range (sigma activity1), although many studies bulk these standard

bounds of sigma band (Rechtschaffen and Kales, 1968), including

frequency bins traditionally considered to be part of the alpha band

(for a review, De Gennaro and Ferrara, 2003). Furthermore, large

genotype differences between the relative contribution of power in

the sigma range have been found in the non-REM sleep of inbred

mice (Franken et al., 1998).

Given the large individual differences in spindle frequency

activity in human subjects (Werth et al., 1997) and since it has been

suggested that such differences are invariant within individuals and

could be related to the individual traits of functional anatomy rather

than to sleep-dependent mechanisms (Finelli et al., 2001b), in the

present work, we decided to specifically analyze both group and

individual differences in a range of EEG frequencies including

sigma activity during non-REM sleep [8.0Y15.5 Hz range (0.25-Hz

resolution)] in a group of 10 subjects who had previously

participated in a slow-wave sleep (SWS) deprivation study (Ferrara

et al., 1999). The six consecutive nights of the experimental

paradigm were characterized by profound differences in quantita-

tive EEG measures, as assessed by spectral analysis of the EEG

(Ferrara et al., 2002). Whether or not spindle frequency activity is

an individual EEG-trait, we hypothesize that its invariance within

individuals will be maintained also during nights with a largely

different sleep architecture.

Materials and methods

Subjects

The current study was carried out on sleep recordings of

normal males who had participated in a SWS deprivation study

(Ferrara et al., 2002). Ten normal male subjects [mean age = 23.4

1 Although there are some intrinsic limitations of Fast Fourier Transform

(FFT) analysis in distinguishing phasic activity from background EEG

activity, it has nevertheless been shown that spectral power reflects sleep

spindle activity by comparing visual scoring of spindles, automatic

detection of spindles, and the FFT analysis of sigma activity (Dijk et al.,

1993).

years (SEM = 0.87)] were selected as paid volunteers from a

university student population. They reported drinking less than

three caffeinated beverages per day, usually sleeping 7Y8 h per

night with sleep onset between 11:00 p.m. and 12:00 midnight,

not taking naps during the day, with no excessive daytime

sleepiness, and no other sleep, medical, or psychiatric disorder, as

assessed by a 1-week sleep log, by actigraphic recordings and by a

clinical interview.

All subjects signed an informed consent before participating in

the study, and the study protocol was approved by the local

Institutional Review Board and was conducted in accordance with

the Declaration of Helsinki.

Procedure

The participants were recorded for 6 consecutive nights in a

sound-proof, temperature controlled room: (1) adaptation; (2)

undisturbed baseline (BSL); (3) baseline with awakenings (BLA);

(4) SWS deprivation-1 (DEP1); (5) SWS deprivation-2 (DEP2); (6)

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.

L. De Gennaro et al. / NeuroImage 26 (2005) 114Y122116

Electrode impedance was kept below 5 KV. Left central EEG (C3-

A2), EMG, and horizontal and vertical EOG were used to visually

score sleep stages in 20 s epochs.

Quantitative analysis of signals

The polygraphic signals were analog to digital converted on-

line with a sampling rate of 128 Hz and stored on the disk of a

personal computer. Artifacts were excluded off-line on a 4-s

basis by visual inspection. The EEG power spectra of the antero-

posterior derivations (Fz-A1, Cz-A1, Pz-A1, and Oz-A1) were

calculated by a Fast Fourier Transform routine for consecutive 4

s epochs of non-REM sleep (stage 2 + 3 + 4), across the

8.0Y15.5 Hz range in a 0.25-Hz resolution. Total sleep time in

the six nights ranged from 451 to 478 min, and non-REM

duration (stage 2 + 3 + 4) from 302 to 319 min. In both cases,

differences were not significant. Sleep onset was determined

according to standard criteria (Rechtschaffen and Kales, 1968).

Absolute individual power values were z-transformed based on

the distributions of individual values for each derivation and night

to stabilize intra-individual variance and to remove the influence of

individual mean power in the considered frequency range.

Statistics

The preliminary analyses of EEG topography were carried out

on group measures. For this purpose, one-way analyses of variance

(ANOVAs) compared the four EEG derivations for each frequency

bin. Further ANOVAs on group data compared the six sleep

recordings for each frequency bin. These ANOVAs were not aimed

to elucidate the specific differences between derivations and nights

(and their interactions), since they had previously been demon-

strated in a detailed analysis at a 1-Hz resolution on raw (not z-

transformed) data (Ferrara et al., 2002). The aim of these ANOVAs

was only to point out the range of between-derivation and between-

night differences in group analyses, substantially confirming what

showed at a 1-Hz resolution.

With specific regard to individual case analysis, to evaluate the

correlation between different nights, we implemented a matrix

correlation (Ramsay et al., 1984), which is an extension of the

usual productYmoment correlations for two variables. The usual

correlation coefficient is obtained from the two vectors x (N � 1)

and y (N � 1) which represent N observations for variables X and

Y, respectively. Therefore, the usual correlation coefficient may be

called a Bvector correlation[. When we have p variables X1I Xp

instead of the single variable X, and p variables Y1IYp instead of

the single variable Y, then we have the data matrix X (N � p) and

the data matrix Y (N � p). The matrix correlation is the index

representing the relationships between the two matrices X and Y.

Each row � column element of the two matrices X and Y has been

correlated to the corresponding element of the other matrix; in such

way, both rows and columns have been correlated giving a unique

value indicating the strength of the association between two

bidimensional measures.

In the present study, any night was seen like a matrix. Each

element of the matrix represents the spectral value of a specific

derivation at a given frequency. In each night, we considered the

four different scalp locations, so that each element of a matrix

was correlated with the same element of any other matrix. More

specifically, Xi ,j represents the EEG power for the i derivations

and j frequency bins in one night, while Yi ,j represents the same

variables in a different night. For each night-subject, the thus-

defined matrix was correlated with the corresponding matrices of

the other five nights, yielding 15 matrix-correlation coefficients

for each subject. The correlation coefficient denotes the strength

of the linear association between the different matrices (nights).

The correlation coefficient changes as a function of the

association between the values of the corresponding variable in

two different matrices (scalp location � EEG frequency),

indicating to what extent the antero-posterior EEG topography

across the 8.00Y15.50 Hz range of one specific night is associated

to that of another night. Hence, the matrix correlation values

measure similarities in topographical distribution of EEG power

values, since they correlate the i values (at any frequency bin) of

a specific scalp location of a night and the i values of the

corresponding scalp location of any other matrix. This is

intrinsically linked to the aim of the current study, that is, the

assessment of individual differences in antero-posterior EEG

topography.

The averaged values of these coefficients were then compared

against the overall correlation between the matrices that was

obtained by estimating the mean correlation coefficient between all

possible combinations of different nights (1350 comparisons),

excluding the different nights of the same subject and the same

night of different subjects. A further overall correlation between

the matrices was also obtained by estimating the mean correlation

coefficient between all possible combinations of different nights

(1620 comparisons), excluding only the different nights of the

same subject. These two overall correlations, calculated with or

without the same night of different subjects, reflect different

estimates of the between-subject similarity, that is, either taking or

not taking into account the similarity due to the experimental

condition (i.e., the specific experimental night).

Both the averaging of within-subject correlations and of

between-subject correlations were performed after a Fisher z-

transformation of the coefficients, and then the average values were

back-transformed. The comparison between the mean within-

subject correlation and the two overall correlations of all the other

different nights were not strictly aimed to assess the significance of

that difference (which is, in fact, very likely), but, above all, to

provide an estimate of the exceeding percentage of variance

explained by within-subject similarity as compared to the between-

subject one.

To assess the significance of the abovementioned comparisons

between correlations coefficients, given the null hypothesis that

both samples of pairs show the same correlation strength, i.e., R1 =

R2, each correlation coefficient was transformed with the Fisher z-

transform: Zf = 1/2 * ln [(1 + R) / (1 j R)]. Fisher’s z-

transformation was used for the variance stabilization of the r

values. All the obtained values were averaged in Zf1 (within-

subjects) and Zf2 (between-subjects). The difference, where N is

the sample size: z = (Zf1 j Zf2) / SQRT [1 / (N1 j 3) + 1 / (N2 j

3)], is approximately normally distributed (Haegert et al., 1999).

The z values have been used to assess statistical significance at the

a level of e0.05. Like the correlation coefficients calculated

between unidimensional variables, the z-transformation is used in

testing the significance of the difference between two matrix

correlation coefficients obtained from two independent samples of

N1 and N2 size, under the null hypothesis of a difference between

the two population correlations equal to 0. Hence, the z-trans-

formation was used to compare the distributions of within-subject

correlations with those of between-subject correlations.

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

L. De Gennaro et al. / NeuroImage 26 (2005) 114Y122 119

during the non-REM sleep of the recovery night (Ferrara et al.,

2002)4.

Therefore, a stable, frequency-specific pattern of EEG power

topography along the antero-posterior cortical axis during non-

REM sleep distinguishes each individual like a Bfingerprint[. Asimilar invariance of EEG power distribution during human non-

REM sleep has also been suggested by a full-scalp recording

(Finelli et al., 2001b). Furthermore, the differences between the

long and the short sleepers in the homeostatic response to sleep

loss, with the short sleepers living under a higher sleep pressure

(Aeschbach et al., 1996), also seem consistent with the hypothesis

of individual invariance of EEG power during human non-REM

sleep. Generally speaking, quantitative non-REM sleep EEG

measures are characterized by a high within-subject stability. As

an example, period and amplitude measures of several EEG bands

including sigma activity were shown to be highly consistent across

nights (Feinberg et al., 1980). This consistency was attributed to:

(a) the fact that subjects were studied in the same biological state,

avoiding the variations in mood, arousal, or attention which can

affect EEG measures during waking; (b) the lengthy period of

observation (5Y6 h per subject) that is typical of sleep studies. The

strong trait-like characteristics of the quantitative EEG sleep

variables were later confirmed in different age groups of subjects

(Tan et al., 2000, 2001). Furthermore, the findings by Palagini et al.

(2000) go in the same direction, demonstrating that individual

differences in spectral measures of baseline sleep were signifi-

cantly retained on the third night of temazepam administration,

although the drug altered the EEG in the delta, sigma, and beta

bands. The particularly high inter-night correlation in the sigma

band (r = 0.977) reported by Tan et al. (2000) was attributed to the

marked and stable individual differences in organized spindle

activity (Gaillard and Blois, 1981; Silverstein and Levy, 1976).

Moreover, the particular waveform of spindles more closely

approximate the Fourier assumptions of sinusoidal shape and

stationarity, a factor that may contribute to the high reliability of

the spindle frequency activity (Tan et al., 2000).

Our findings clearly showed that this Bspectral signature[remains substantially invariant during different nights character-

ized by massive changes of sleep structure. Moreover, this

invariance is also topographically specific (at least, on the

antero-posterior axis). Therefore, it can be hypothesized that it

may be related to individual differences in genetically determined

functional brain anatomy rather than to sleep-dependent mecha-

nisms, since it has been already demonstrated that many features of

the normal waking EEG are among the most heritable traits in

humans (Van Beijsterveldt and Boomsma, 1994). In fact, the

monozygotic cotwin covariance for EEG power is greater than

expected for additive gene actions, as compared with dizygotic

cotwin covariance (Lykken, 1982; Stassen et al., 1988). This

pattern of twin covariances could be due to gene interactions, but

also to greater monozygotic than dizygotic environmental cova-

riance (Christian et al., 1996). In any case, it has been concluded

4 Although independent of the homeostatic changes after SWS depriva-

tion, the invariance in the individual EEG topography pattern could

nevertheless be theoretically dependent on circadian factors, since sleep

occurred at only one circadian phase, and the separation of the two

mechanisms is allowed only by forced desynchrony protocols (e.g., Dijk,

1999; Dijk et al., 1995, 1997, 1999). Only specific studies using these

protocols can directly assess this hypothesis.

from the available data that the waking EEG is among the most

heritable human traits (Van Baal et al., 1996). Moreover, as far as

sleep EEG is concerned, a strong genetic control for delta and

sigma oscillations during non-REM sleep, and for theta oscillations

during REM sleep, has been found in inbred mice (Franken et al.,

1998; Tafti et al., 2003).

Individual differences or different spindles?

The present results also provide a definitive contribution to the

Bslow-fast spindle[ issue. On the basis of regional differences in

cortical topography of sleep spindles, two distinct sleep spindle

types have been hypothesized: Bslow[ spindles at ¨12 Hz, more

pronounced over the frontal sites, and Bfast[ spindles at ¨14 Hz,

more evident over parietal and central sites (Jobert et al., 1992;

Werth et al., 1997; Zeitlhofer et al., 1997; Zygierewicz et al.,

1999). While lower sigma frequencies and slow spindles decrease

over consecutive non-REM sleep episodes, higher sigma frequen-

cies and fast spindles progressively rise (Jobert et al., 1992; Werth

et al., 1997). Moreover, other functional dissociations have been

reported as a function of age and maturation (Landolt et al., 1996;

Shinomiya et al., 1999), homeostatic and circadian factors

(Aeschbach and Borbely, 1993; Aeschbach et al., 1997; Dijk and

Czeisler, 1995; Dijk et al., 1997; Wei et al., 1999), menstrual cycle

phase (Driver et al., 1996), pregnancy (Brunner et al., 1994), and

pharmacological agents (Aeschbach et al., 1994; Dijk et al., 1995;

Jobert et al., 1992). These results have been interpreted according

to at least two functionally separate spindle generators. Some

support for this hypothesis has been provided by a study using low-

resolution brain electromagnetic tomography (LORETA), showing

a source in the prefrontal cortex (Brodmann areas 9 and 10) for

slow spindles and in the precuneus (Brodmann area 7) for fast

spindles (Anderer et al., 2001).

On the other hand, electrophysiological studies have attributed

the whole frequency range of sleep spindles to a single mechanism,

namely, the duration of the hyperpolarization-rebound sequence in

thalamocortical neurons: long hyperpolarizations yield slower EEG

frequencies, short hyperpolarizations yield faster EEG frequencies

(e.g., Steriade and Amzica, 1998). Cortical areas on which Bslow[spindles have been detected are related to those thalamic nuclei in

which relay cells display long hyperpolarizations (Steriade,

personal communication).

In our opinion, the present results clearly point to a single

mechanism. In line with the traditional sigma band limits

(12.00Y14.00 Hz), the data show that all subjects have only one

kind of sleep spindle, ranging within the 12.25Y13.75 Hz

frequency, while in some subjects there is a peak of EEG activity

on more anterior brain sites within the 9.00Y12.25 Hz range.

Although, on the basis of a FFT approach, one cannot ultimately

say whether this is linked to spindle activity or not, two main

arguments militate in favor of an interpretation in terms of alpha

activity: (1) In the current individual data, spindle activity, as a

general mechanism discernible in any sleep recordings, should

correspond to the peak within the 12.25Y13.75 Hz frequency

observed in all subjects; (2) A second peak, clearly discernible in

half of the subjects (#4, #6, #7, #8, and #10), ranges from 9.00 Hz

to 12.25 Hz with different bounds in different subjects, and these

frequencies mostly overlap with the limits of the alpha rhythm.

Hence, the main dissociation between slow and fast spindles, that

is, their inverse time course across consecutive sleep episodes

(Jobert et al., 1992; Werth et al., 1997), can be explained by the

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

sleep EEG. J. Sleep Res. 2, 70–81.

Aeschbach, D., Dijk, D.J., Trachsel, L., Brunner, D.P., Borbely, A.A., 1994.

Dynamics of slow-wave activity and spindle frequency activity in the

human sleep EEG: effect of midazolam and zopiclone. Neuropsycho-

pharmacology 11, 237–244.

Aeschbach, D., Cajochen, C., Landolt, H., Borbely, A.A., 1996. Homeo-

static sleep regulation in habitual short sleepers and long sleepers. Am.

J. Physiol. 270, R41–R53.

Aeschbach, D., Dijk, D.J., Borbely, A.A., 1997. Dynamics of EEG spindle

frequency activity during extended sleep in humans: relationship to

slow-wave activity and time of day. Brain Res. 748, 131–136.

Aeschbach, D., Postolache, T.T., Sher, L., Matthews, J.R., Jackson, M.A.,

Wehr, T.A., 2001. Evidence from the waking electroencephalogram that

short sleepers live under higher homeostatic sleep pressure than long

sleepers. Neuroscience 102, 493–502.

Aeschbach, D., Sher, L., Postolache, T.T., Matthews, J.R., Jackson, M.A.,

Wehr, T.A., 2003. A longer biological night in long sleepers than in

short sleepers. J. Clin. Endocrinol. Metab. 88, 26–30.

Aguirre, G.K., Zarahn, E., D’esposito, M., 1998. The variability of human.

BOLD hemodynamic responses. NeuroImage 8, 360–369.

Anderer, P., Klosch, G., Gruber, G., Trenker, E., Pascual-Marqui, R.D.,

Zeitlhofer, J., Barbanoj, M.J., Rappelsberger, P., Saletu, B., 2001. Low-

resolution brain electromagnetic tomography revealed simultaneously

active frontal and parietal sleep spindle sources in the human cortex.

Neuroscience 103, 581–592.

Bell-McGinty, S., Habeck, C., Hilton, H.J., Rakitin, B., Scarmeas, N.,

Zarahn, E., Flynn, J., DeLaPaz, R., Basner, R., Stern, Y., 2004.

Identification and differential vulnerability of a neural network in sleep

deprivation. Cereb. Cortex 14, 496–502.

Borbely, A.A., 1982. A two process model of sleep regulation. Hum.

Neurobiol. 1, 195–204.

Borbely, A.A., Achermann, P., 2000. Sleep homeostasis and models of sleep

regulation. In: Kryger, M.H., Roth, T., Dement, W.C. (Eds.), Principles

and Practice of Sleep Medicine. Saunders, Philadelphia, pp. 377–390.

Brunner, D.P., Munch, M., Biedermann, K., Huch, R., Huch, A., Borbely,

A.A., 1994. Changes in sleep and sleep electroencephalogram during

pregnancy. Sleep 17, 576–582.

Cajochen, C., Foy, R., Dijk, D.J., 1999. Frontal predominance of a relative

increase in sleep delta and theta EEG activity after sleep loss in humans.

Sleep Res. Online 2, 65–69.

Christian, J.C., Morzorati, S., Norton Jr., J.A., Williams, C.J., O’Connor, S.,

Li, T.K., 1996. Genetic analysis of the resting electroencephalographic

power spectrum in human twins. Psychophysiology 33, 584–591.

Curcio, G., Ferrara, M., Piergianni, A., Fratello, F., De Gennaro, L., 2004.

Paradoxes of the first-night effect: a quantitative analysis of antero-

posterior EEG topography. Clin. Neurophysiol. 115, 1178–1188.

De Gennaro, L., Ferrara, M., 2003. Sleep spindles: an overview. Sleep Med.

Rev. 7, 423–440.

Dijk, D.J., 1999. Circadian variation of EEG power spectra in NREM and

REM sleep in humans: dissociation from body temperature. J. Sleep

Res. 8, 189–195.

Dijk, D.J., Czeisler, C.A., 1995. Contribution of the circadian pacemaker

and the sleep homeostat to sleep propensity, sleep structure, electro-

encephalographic slow waves, and sleep spindle activity in humans.

J. Neurosci. 15, 3526–3538.

Dijk, D.J., Hayes, B., Czeisler, C.A., 1993. Dynamics of electroencephalo-

graphic sleep spindles and slow wave activity in men: effect of sleep

deprivation. Brain Res. 626, 190–199.

Dijk, D.J., Roth, C., Landolt, H.P., Werth, E., Aeppli, M., Achermann, P.,

Borbely, A.A., 1995. Melatonin effect on daytime sleep in men:

suppression of EEG slow-frequency activity and enhancement of

spindle frequency. Neurosci. Lett. 201, 13–16.

Dijk, D.-J., Shanahan, T.L., DuVy, J.F., Ronda, J.M., Czeisler, C.A., 1997.

Variation of electroencephalographic activity during NREM and REM

sleep with phase of circadian melatonin rhythm in humans. J. Physiol.

(London) 505, 851–858.

Dijk, D.J., Duffy, J.F., Riel, E., Shanahan, T.L., Czeisler, C.A., 1999.

Ageing and the circadian and homeostatic regulation of human sleep

during forced desynchrony of rest, melatonin and temperature rhythms.

J. Physiol. 516 (Pt 2), 611–627.

Driver, H.S., Dijk, D.J., Werth, E., Biedermann, K., Borbely, A.A., 1996.

Sleep and the sleep electroencephalogram across the menstrual cycle in

young healthy women. J. Clin. Endocrinol. Metab. 81, 728–735.

Duffy, J.F., Rimmer, D.W., Czeisler, C.A., 2001. Association of intrinsic

circadian period with morningnessYeveningness, usual wake time, and

circadian phase. Behav. Neurosci. 115, 895–899.

L. De Gennaro et al. / NeuroImage 26 (2005) 114Y122122

Feinberg, I., Fein, G., Floyd, T.C., 1980. Period and amplitude analysis of

NREM EEG in sleep: repeatability of results in young adults.

Electroencephalogr. Clin. Neurophysiol. 48, 212–221.

Ferrara, M., De Gennaro, L., Bertini, M., 1999. Selective slow-wave sleep

(SWS) deprivation and SWS rebound: do we need a fixed SWS amount

per night? Sleep Res. Online 2, 15–19.

Ferrara, M., De Gennaro, L., Casagrande, M., Bertini, M., 2000. Selective

slow-wave sleep deprivation and time-of-night effects on cognitive

performance upon awakening. Psychophysiology 37, 440–446.

Ferrara, M., De Gennaro, L., Curcio, G., Cristiani, R., Corvasce, C., Bertini,

M., 2002. Regional differences of the human sleep electroencephalo-

gram in response to selective slow-wave sleep deprivation. Cereb.

Cortex 12, 737–748.

Finelli, L.A., Borbely, A.A., Achermann, P., 2001a. Functional topography

of the human nonREM sleep electroencephalogram. Eur. J. Neurosci.

13, 2282–2290.

Finelli, L.A., Achermann, P., Borbely, A.A., 2001b. Individual Ffingerprints_

in human sleep EEG topography. Neuropsychopharmacology 25,

S57–S62.

Franken, P., Malafosse, A., Tafti, M., 1998. Genetic variation in EEG activity

during sleep in inbred mice. Am. J. Physiol. 275, R1127–R1137.

Gaillard, J.M., Blois, R., 1981. Spindle density in sleep of normal subjects.

Sleep 4, 385–391.

Haegert, D.G., Cowan, T., Murray, T.J., Gadag, V., O’Connor, P., 1999.

Does a shift in the T-cell receptor repertoire precede the onset of MS?

Neurology 11, 485–490.

Handwerker, D.A., Ollinger, J.M., D’Esposito, M., 2004. Variation of

BOLD hemodynamic responses across subjects and brain regions and

their effects on statistical analyses. NeuroImage 21, 1639–1651.

Huber, R., Deboer, T., Tobler, I., 2000. Topography of EEG dynamics after

sleep deprivation in mice. J. Neurophysiol. 84, 1888–1893.

Jobert, M., Poiseau, E., Jahnig, P., Schulz, H., Kubicki, S., 1992.

Topographical analysis of sleep spindle activity. Neuropsychobiology

26, 210–217.

Knoblauch, V., Martens, W., Wirz-Justice, A., Krauchi, K., Cajochen, C.,

2003. Regional differences in the circadian modulation of human sleep

spindle characteristics. Eur. J. Neurosci. 18, 155–163.

Landolt, H.P., Dijk, D.J., Achermann, P., Borbely, A.A., 1996. Effect

of age on the sleep EEG: slow-wave activity and spindle

frequency activity in young and middle-aged men. Brain Res. 738,

205–212.

Leproult, R., Colecchia, E.F., Berardi, A.M., Stickgold, R., Kosslyn, S.M.,

Van Cauter, E., 2003. Individual differences in subjective and objective

alertness during sleep deprivation are stable and unrelated. Am. J.

Physiol.: Regul., Integr. Comp. Physiol. 284, R280–R290.

Lykken, D.T., 1982. Presidential address, 1981. Research with twins: the

concept of emergenesis. Psychophysiology 19, 361–373.

Oleksenko, A.I., Mukhametov, L.M., Polyakova, I.G., Supin, A.Y.,

Kovalzon, V.M., 1992. Unihemispheric sleep deprivation in bottlenose

dolphins. J. Sleep Res. 1, 40–44.

Palagini, L., Campbell, I.G., Tan, X., Guazzelli, M., Feinberg, I., 2000.

Independence of sleep EEG responses to GABAergic hypnotics:

biological implications. J. Psychiatr. Res. 34, 293–300.

Ramsay, J., Ten Berge, J., Styan, G.P., 1984. Matrix correlation.

Psychometrika 49, 403–423.

Rattenborg, N.C., Lima, S.L., Amlaner, C.J., 1999. Half-awake to the risk

of predation. Nature 397, 397–398.

Rechtschaffen, A., Kales, A., 1968. A Manual of Standardized Terminol-

ogy, Techniques and Scoring System for Sleep Stages of Human

Subjects. Brain Information Service/Brain Research Institute, Univer-

sity of California, Los Angeles.

Shinomiya, S., Nagata, K., Takahashi, K., Masumura, T., 1999. Develop-

ment of sleep spindles in young children and adolescents. Clin.

Electroencephalogr. 30, 39–43.

Silverstein, L.D., Levy, C.M., 1976. The stability of the sigma sleep

spindle. Electroencephalogr. Clin. Neurophysiol. 40, 666–670.

Stassen, H.H., Lykken, D.T., Bomben, G., 1988. The within-pair EEG

similarity of twins reared apart. Eur. Arch. Psychiatr. Neurol. Sci. 237,

244–252.

Steriade, M., 1999. Coherent oscillations and short-term plasticity in

corticothalamic networks. Trends Neurosci. 22, 337–345.

Steriade, M., Amzica, F., 1998. Coalescence of sleep rhythms and their

chronology in corticothalamic networks. Sleep Res. Online 1, 1–10.

Tafti, M., Petit, B., Chollet, D., Neidhart, E., de Bilbao, F., Kiss, J.Z.,

Wood, P.A., Franken, P., 2003. Deficiency in short-chain fatty acid

beta-oxidation affects theta oscillations during sleep. Nat. Genet. 34,

320–325.

Tan, X., Campbell, I.G., Palagini, L., Feinberg, I., 2000. High internight

reliability of computer-measured NREM delta, sigma, and beta:

biological implications. Biol. Psychiatry 48, 1010–1019.

Tan, X., Campbell, I.G., Feinberg, I., 2001. Internight reliability and

benchmark values for computer analyses of non-rapid eye movement

(NREM) and REM EEG in normal young adult and elderly subjects.

Clin. Neurophysiol. 112, 1540–1552.

Tobler, I., 1995. Is sleep fundamentally different between mammalian

species? Behav. Brain Res. 69, 35–41.

Van Baal, G.C., De Geus, E.J., Boomsma, D.I., 1996. Genetic architecture

of EEG power spectra in early life. Electroencephalogr. Clin. Neuro-

physiol. 98, 502–514.

Van Beijsterveldt, C.E., Boomsma, D.I., 1994. Genetics of the human

electroencephalogram (EEG) and event-related brain potentials (ERPs):

a review. Hum. Genet. 94, 319–330.

Van Dongen, H.P., Baynard, M.D., Maislin, G., Dinges, D.F., 2004.

Systematic interindividual differences in neurobehavioral impairment

from sleep loss: evidence of trait-like differential vulnerability. Sleep

27, 423–433.

Vyazovskiy, V.V., Borbely, A.A., Tobler, I., 2002. Interhemispheric

sleep EEG asymmetry in the rat is enhanced by sleep deprivation.

J. Neurophysiol. 88, 2280–2286.

Wei, H.G., Riel, E., Czeisler, C.A., Dijk, D.J., 1999. Attenuated amplitude

of circadian and sleep-dependent modulation of electroencephalo-

graphic sleep spindle characteristics in elderly human subjects. Neuro-

sci. Lett. 260, 29–32.

Wei, X., Yoo, S.S., Dickey, C.C., Zou, K.H., Guttmann, C.R., Panych, L.P.,

2004. Functional MRI of auditory verbal working memory: long-term

reproducibility analysis. NeuroImage 21, 1000–1008.

Werth, E., Achermann, P., Dijk, D.J., Borbely, A.A., 1997. Spindle

frequency activity in the sleep EEG: individual differences and

topographic distribution. Electroencephalogr. Clin. Neurophysiol. 103,

535–542.

Zeitlhofer, J., Gruber, G., Anderer, P., Asenbaum, S., Scimicek, P., Saletu,

B., 1997. Topographic distribution of sleep spindles in young healthy

subjects. J. Sleep Res. 6, 149–155.

Zygierewicz, J., Blinowska, K.J., Durka, P.J., Szelenberger, W., Niemce-

wicz, S., Androsiuk, W., 1999. High resolution study of sleep spindles.

Electroencephalogr. Clin. Neurophysiol. 110, 2136–2147.