Genetics of human sleep EEG

Sleep characteristics are candidates for predictive biological
markers in patients with severe psychiatric diseases, in particular
affective disorder and schizophrenia. The genetic components of
sleep determination in humans remain, to a large degree,
unelucidated. In particular, the heritability of rapid eye movement
(REM) sleep and EEG bursts of oscillatory brain activity in Non-REM
sleep, i.e. sleep spindles, are of interest. In addition, recent
findings suggest a strong role of distinct sleep spindle types in
memory consolidation, making it important to identify sleep
spindles in slow wave sleep (SWS) and to separate slow and fast
spindle localization in the frequency range. However, predictive
sleep biomarker research requires large sample sizes of healthy and
affected human individuals. Therefore, the present work addressed
two questions. The first aim was to optimize data analysis by
developing algorithms that allow an efficient and reliable
identification of rapid eye movements (REMs) and sleep EEG
spindles. In the second part, developed methods were applied to
sleep EEG data from a classical twin study to identify genetic
effects on tonic and phasic REM sleep parameters, sleep spindles,
and their trait-like characteristics. The algorithm for REM
detection was developed for standard clinical two channel
electrooculographic montage. The goal was to detect REMs visible
above the background noise, and in the case of REM saccades to
classify each movement separately. In order to achieve a high level
of sensitivity, detection was based on a first derivative of
electrooculogram (EOG) potentials and two detection thresholds. The
developed REM detector was then validated in n=12 polysomnographic
recordings from n=7 healthy subjects who had been previously scored
visually by two human experts according to standard guidelines.
Comparison of automatic REM detection with human scorers revealed
mean correlations of 0.94 and 0.90, respectively (mean correlation
between experts was 0.91). The developed automatic sleep spindle
detector assessed individualized signal amplitude for each channel
as well as slow and fast spindle frequency peaks based on the
spectral analysis of the EEG signal. The spindle detection was
based on Continuous Wavelet Transform (CWT); it localized the exact
length of sleep spindles and was sensitive also for detection of
sleep spindles intermingled in high amplitude slow wave EEG
activity. The automatic spindle detector was validated in n=18 naps
from n=10 subjects, where EEG data were scored both visually and by
a commercial automatic algorithm (SIESTA). Comparison of our own
spindle detector with results from the SIESTA algorithm and visual
scoring revealed the correlations of 0.97 and 0.92, respectively
(correlation between SIESTA algorithm and visual scoring was 0.90).
In the second part of the work, the similarity of given sleep EEG
parameters in n=32 healthy monozygotic (MZ) twins was compared with
the similarity in n=14 healthy same-gender dizygotic (DZ) twins.
The author of the current work did not participate in acquisition
of twin study sample. EEG sleep recordings used for the
heritability study were collected and already described by
Ambrosius et al. (2008). Investigation of REM sleep included the
absolute EEG spectral power, the shape of REM power spectrum, the
amount and the structural organization of REMs; parameters of
Non-REM sleep included slow and fast sleep spindle characteristics
as well as the shape of the Non-REM power spectrum in general. In
addition to estimating genetic effects, differences in within-pair
similarity and night-to-night stability of given parameters were
illustrated by intraclass correlation coefficients (ICC) and
cluster analysis. A substantial genetic influence on both spectral
composition and phasic parameters of REM sleep was observed. A
significant genetic variance in spectral power affected delta to
high sigma and high beta to gamma EEG frequency bands, as well as
all phasic REM parameters with the exception of the REMs occurring
outside REM bursts. Furthermore, MZ and DZ twins differed
significantly in their within-pair similarity of non-REM and REM
EEG spectra morphology. Regarding sleep spindles, statistical
analysis revealed a significant genetic influence on localization
in frequency range as well as on basic spindle characteristics
(amplitude, length, quantity), except in the quantity of fast
spindles in stage 2 and whole Non-REM sleep. Basic spindle
parameters showed trait-like characteristics and significant
differences in within-pair similarity between the twin groups. In
summary, the developed algorithms for automatic REM and sleep
spindle detection provide several advantages: the elimination of
human scorer biases and intra-rater variability, investigation of
structural organization of REMs, exact determination of fast and
slow spindle frequency for each individual. Algorithms are fully
automated and therefore well suited to score REM density and sleep
spindles in large patient samples. In the second part of the study,
sleep EEG analysis in MZ and DZ twins revealed a substantial
genetic determination of both tonic and phasic REM sleep
parameters. This complements previous findings of a high genetic
determination of the Non-REM sleep power spectrum. Interestingly,
smaller genetic effects and lower night-to-night stability were
observed for fast spindles, especially in SWS. This is in line with
recent hypotheses on the differential function of sleep spindle
types for memory consolidation. The results from the presented
studies strongly support the application of sleep EEG to identify
clinically relevant biomarkers for psychiatric disorders.