Hemodynamic cerebral correlates of sleep spindles
during human non-rapid eye movement sleep
M. Schabus†‡§, T. T. Dang-Vu†¶, G. Albouy†, E. Balteau†, M. Boly†¶, J. Carrier?, A. Darsaud†, C. Degueldre†,
M. Desseilles†,††, S. Gais†, C. Phillips†, G. Rauchs†, C. Schnakers†, V. Sterpenich†, G. Vandewalle†, A. Luxen†,
and P. Maquet†§¶
†Cyclotron Research Centre, University of Lie `ge, B-4000 Lie `ge, Belgium;††Departments of Psychiatry and¶Neurology, Centre Hospitalier
Universitaire de Lie `ge, B-4000 Lie `ge, Belgium;‡Department of Psychology, University of Salzburg, A-5020 Salzburg, Austria;
and?Department of Psychology, University of Montre ´al, Montre ´al, QC, Canada H3C 3J7
Edited by Marcus E. Raichle, Washington University School of Medicine, St. Louis, MO, and approved June 26, 2007 (received for review April 3, 2007)
of spindles during sleep, which differ by their scalp topography and
possibly some aspects of their regulation. To test for the existence of
two different spindle types, we characterized the activity associated
with slow (11–13 Hz) and fast (13–15 Hz) spindles, identified as
discrete events during non-rapid eye movement sleep, in non-sleep-
deprived human volunteers, using simultaneous electroencephalog-
raphy and functional MRI. An activation pattern common to both
spindle types involved the thalami, paralimbic areas (anterior cingu-
late and insular cortices), and superior temporal gyri. No thalamic
difference was detected in the direct comparison between slow and
fast spindles although some thalamic areas were preferentially acti-
vated in relation to either spindle type. Beyond the common activa-
tion pattern, the increases in cortical activity differed significantly
between the two spindle types. Slow spindles were associated with
increased activity in the superior frontal gyrus. In contrast, fast
spindles recruited a set of cortical regions involved in sensorimotor
recruitment of partially segregated cortical networks for slow and
fast spindles further supports the existence of two spindle types
during human non-rapid eye movement sleep, with potentially
different functional significance.
electroencephalography (EEG)/functional MRI (fMRI) ? light sleep ?
neuroimaging ? sleep physiology
In the early stages of non-rapid eye movement (NREM) sleep,
electroencephalographic recordings show characteristic spindle os-
cillations. In humans, spindles consist of waxing-and-waning 11- to
15-Hz oscillations, lasting 0.5–3 sec. At the cellular level, spindles
are associated with substantial neuronal activity. Spindles arise
from cyclic inhibition of thalamo-cortical (TC) neurons by reticular
thalamic neurons. Postinihibitory rebound spike bursts in TC cells
entrain cortical populations in spindle oscillations (1). These neu-
to shape the processing of information during light NREM sleep
and participate in the alteration of consciousness that characterizes
this stage of sleep.
Little is known on the cerebral correlates of human spindles.
Early positron emission tomography studies reported a negative
relationship between thalamic cerebral blood flow and the power
spectrum in the spindle frequency band (2). However, the low
temporal resolution of positron emission tomography did not allow
for a fine-grained characterization of the cerebral correlates of
human spindles. In addition, two kinds of spindles are described in
fast spindles (?13 Hz) prevail over centro-parietal areas. The
functional differences. These two spindling activities differ by their
circadian and homeostatic regulations, pharmacological reactivity,
development in infancy, evolution during aging, modulation during
uman sleep is associated with a profound modification of
consciousness and the emergence of distinct sleep oscillations.
menstrual cycle and pregnancy (3), and, intriguingly, by their
association with general cognitive capabilities (4). Source recon-
struction of scalp EEG recordings identified two sources, one for
in the precuneus (5). Despite these functional differences, it is still
debated whether slow and fast spindles reflect the activity of
different neural networks or the differential modulation of a single
generator (for review, see ref. 3).
In this article, we first aimed at characterizing the cerebral
correlates of the neural activity associated with spindles, in a group
of normal, young, and non-sleep-deprived volunteers, using simul-
taneous electroencephalography (EEG)/functional MRI (fMRI)
acquisitions. EEG recordings allowed us to precisely identify the
fMRI analysis as in a classical event-related design. Second, we
tested the hypothesis that in humans, there are two distinct spindle
types by estimating the differences between the regional brain
activity associated with slow and fast spindles.
Of 25 subjects, 14 participants maintained stable S2 and S3 sleep
periods (Fig. 1a for a representative EEG recording). In these
subjects, the series of consecutive fMRI volumes corresponding to
steady stage 2 or stage 3 NREM sleep were selected from the
complete fMRI time series. They constituted a ‘‘session’’ and were
considered for further analysis. One to six sessions [mean (SE): 3.5
(0.5)] were selected per subject [duration range: 137.8–2,164.8 sec,
mean (SE): 622.9 sec (63.4)].
Based on previous data-driven analysis of human sleep data, it
appears that the frequency that best separates slow from fast
were identified on band pass-filtered data between 11–13 Hz and
13–15 Hz respectively, using an automatic detection algorithm
inspired from Mo ¨lle et al. (7). The mean (SE) number of detected
slow and fast sleep spindles per subject was 74.6 (10.6) and 95.3
spindles were identified per session. As expected, the power in slow
and fast sigma bands predominated respectively over frontal and
centro-parietal scalp areas (Fig. 1b). Additionally, averaging 0.5–4
Author contributions: M.S., T.T.D.-V., and P.M. designed research; M.S., T.T.D.-V., G.A.,
A.L., and P.M. contributed new reagents/analytic tools; M.S. and T.T.D.-V. analyzed data;
and M.S. and P.M. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
Abbreviations: EEG, electroencephalography; fMRI, functional MRI; NREM, non-rapid eye
movement; SMA, supplementary motor area; TC, thalamo-cortical.
§To whom correspondence may be addressed at: Cyclotron Research Centre, University of
Lie `ge, Belgium. E-mail: firstname.lastname@example.org or email@example.com.
This article contains supporting information online at www.pnas.org/cgi/content/full/
© 2007 by The National Academy of Sciences of the USA
August 7, 2007 ?
vol. 104 ?
Hz filtered EEG recordings with respect to spindle onsets con-
firmed that spindles occurred on the depolarizing phase of the slow
oscillation (7) (Fig. 1c).
We first identified the brain areas responding to both slow and
fast spindles (conjunction analysis: Fig. 2 and Table 1, center
columns). Significant positive responses were identified in the left
and right thalamus in their lateral and posterior aspects. At the
cortical level, significant increases in activity were detected in
paralimbic areas: the anterior cingulate cortex and the left insula.
identified bilaterally in the superior temporal gyrus, in the vicinity
of auditory cortices.
Beyond these common activations, brain activity associated with
slow and fast spindles did not overlap completely [main effect of
each spindle type: Fig. 2 and Table 1, left and right columns;
additional results in supporting information (SI) Tables 3 and 4].
thalami, anterior cingulate, insular, and auditory cortices. There
were two noticeable differences with the common activity pattern.
forebrain/hypothalamus), the midbrain tegmentum, and cerebellar
vermis. Second, at the cortical level, in addition to the common
with slow spindles, in the right superior frontal gyrus.
In contrast to slow spindles, the activity related to fast spindles
appeared limited at the thalamic level but more extended at the
cortical level, relative to the common activity pattern. At the
activity related to fast spindles was detected not only in orbito-
frontal and middle frontal areas, but interestingly also in the
precentral and postcentral gyri, supplementary motor area (SMA),
and in mid-cingulate cortex ventral to the cingulate motor zones
(Fig. 2 and Table 1, right columns).
Finally, we directly compared the activities related to slow and
fast sleep spindles. This differential effect conveys important in-
formation that does not necessarily appear in the simple main
effects reported above. Importantly, we detected no significant
difference in the thalamus (even at a lenient statistical threshold,
P ? 0.01 uncorrected). In contrast, at the cortical level, significant
mainly located in regions reported above as active during fast
spindles, with significant responses identified in the orbital and
middle frontal, precentral and postcentral, and insular cortices
(Table 2; additional results in SI Table 5). Interestingly, larger
activity for fast spindles was also observed within the mesial
prefrontal cortex and hippocampus (Fig. 3).
The only region showing larger activity during slow than fast
observation that thalamic peak voxels for slow spindles were
primarily located in areas compatible with the mediodorsal nucleus
(8), which has highest probability of connection with the prefrontal
cortex (9). Furthermore, peak voxels for fast spindles were com-
patible with the ventral posterior lateral and pulvinar nuclei (8)
having a substantial probability of connection (9) to primary
6). Consistent with these anatomical data, there was a significant
linear relationship between the activity recorded in peak thalamic
voxels [8 ?16 4] for slow spindles and superior frontal gyrus (38 58
20, Z ? 3.38, PSVC? 0.05) as well as between the activity recorded
in peak thalamic voxels for fast spindles [16 ?22 6] and cingulate
and SMA (2 ?8 66, Z ? 4.42, PSVC? 0.01).
The cerebral correlates of slow and fast spindles, taken as discrete
and identifiable neural events, were characterized by using EEG/
fMRI in non-sleep-deprived normal human volunteers during the
points. First, an activation pattern common to both spindle types
was identified, which involved both thalami, the anterior cingulate
cortex, the left anterior insula, and, bilaterally, the superior tem-
poral gyrus. Second, significant differences in cortical activations
were observed between slow and fast spindles. The activity asso-
ciated with slow spindles largely corresponded to the common
activation pattern, with the additional recruitment of the right
superior frontal gyrus. On the contrary, fast spindles were associ-
ated with a number of significant activations beyond the common
pattern in the SMA, sensori-motor, and mid-cingulate cortex. Fast
spindles elicited significantly larger responses than slow spindles in
the left hippocampus the orbito and mesial prefrontal cortex,
sensori-motor cortex, and anterior insula. Third, no significant
difference in thalamic activation was detected when slow and fast
spindles were directly compared with each other. Slow spindles
were associated with a significant activation in the bulk of both
thalami, with the exception of their anterior portion. Fast spindles
were associated with a thalamic activation of smaller extent, al-
of both thalami.
Common Activation Pattern. Asexpected,significantresponseswere
identified in the left and right thalami in keeping with the increase
in firing during spindles reported at the cellular level (1). At the
cortical level, significant increases in activity were detected in
paralimbic areas: the left insula and the anterior cingulate cortex.
recording (stage 2 sleep; 0.1 to 70 Hz) after scanner and pulse artifact correc-
11–13 (left) and 13–15 Hz (right). For display the average normalized spindle
power of all slow and fast sleep spindles (detected on Cz) was computed at
each channel and between 11–13 and 13–15 Hz, respectively. Slow sigma
mainly expressed over centro-parietal areas. Nose is upwards, right is right-
EEG data (0.5–4 Hz) were averaged with respect to the onset of all sleep
spindles. Spindles start (t ? 0) on the depolarizing phase of the oscillation,
which on average are of much smaller amplitude than the classical full blown
slow waves of deep slow-wave sleep. The classical phase lag from frontal (Fz,
black) to central (Cz, red) and parietal (Pz, blue) areas and the maximal slow
wave amplitude at the frontal recording site are also depicted.
EEG characterization of sleep spindles. (a) Example of a typical EEG
Schabus et al.
August 7, 2007 ?
vol. 104 ?
no. 32 ?
The latter result contrasts with the assumption that spindles are
absent in the cingulate cortex (ref 1, page 284). This notion was
based on the observation that in cats, anterior thalamic nuclei,
which project to cingulate cortex, do not receive inputs from the
reticular nucleus (10) and do not show any spindling activity (11).
In contrast, anterior cingulate and insular cortices seem deeply
involved in human NREM sleep rhythms: Not only are they
systematically active during both types of spindles (Fig. 2), but their
activity is known to change significantly during deep NREM sleep
(12, 13). During spindles, the anterior cingulate cortex might be
driven by medio-dorsal (14) or intralaminar thalamic inputs. In the
neocortex, significant responses common to all spindles were only
identified in bilateral superior temporal gyri, in a location corre-
sponding to auditory cortices. One might be tempted to attribute
this activation to processing of scanner noise. This assumption
would suggest that transmission of auditory inputs during sleep is
modulated by spindles. Auditory information has been shown to be
processed up to the cortical level during NREM sleep (15).
However, changes in responses evoked by sounds presented during
spindles, relative to outside spindles, suggest that spindles inhibit
information processing and protect the sleeper from intrusive
external stimuli (16). These results are consistent with the view,
from the external world during NREM sleep because of synaptic
characterizing the cerebral correlates of sound processing during
Difference in Cortical Activity Associated with Slow and Fast Spindles.
The cortical activity associated with slow spindles was essentially
limited to the common activation pattern. However, an additional
are readily recorded from all prefrontal cortices in humans by
electro-corticography (17), we expected a larger recruitment of
frontal cortices in relation to slow spindles. However, in scalp EEG
recordings, spindles show variable topographical dynamics over
frontal regions (18), suggesting that any given prefrontal region
the probability of detection in our analysis.
Consistent responses to fast spindles were detected in a number
of brain areas, on top of the common activation pattern. Several of
these areas clustered around sensorimotor regions: sensori-motor
cortices, SMAs, and mid-cingulate cortex. These findings are con-
sistent with scalp EEG data suggesting that fast spindles are
topographically and dynamically limited to central and parietal
scalp areas (5, 18). Interestingly, these sensorimotor areas are also
The sensorimotor (?) rhythm is known as a conspicuous sponta-
neous rhythm of relaxed wakefulness, involving the sensorimotor
and premotor cortices (19, 21). Although they differ in several
respects (1), spindles and ? rhythm seem functionally related.
Indeed, in cats, facilitation of sensorimotor rhythm through con-
ditioning during wakefulness increases spindles and decreases mo-
tor output during subsequent sleep (22). Fast spindles might thus
emerge from the interaction between the oscillatory properties of
these sensorimotor TC loops and the oscillatory context of NREM
sleep, characterized for instance by a slow rhythm (?1 Hz) which
organizes the recurrence of spindles (7, 23).
The differential contrast between the two spindle types also
surprising as the hippocampus is not detected in the simple main
effect of fast spindles. However, it is explained by the opposite time
courses of hippocampal responses during slow and fast spindles (cf.
05 10 1520 2530
05 1015 20 25 30
05 101520 2530
05 10 1520 25
05 10 15 202530
05 10 1520 25
05 101520 25
05 1015 20 2530
05 1015 2025
05 10 15 20 2530
05 10 152025
z = 6
x = -6
x = -36
z = 8
x = -10
x = -6
x = -36
x = 26
x = -4
z = 6
x = -10
x = -2
x = 34
Neurological Institute space (Puncorrected? 0.001). The leftmost panels show peristimulus time histograms (PSTHs) depicting the responses in auditory cortices (circled)
panels show PSTHs depicting the response in superior temporal gyri (j), thalami (k), mid cingulate cortex (circled) and SMA (dotted) (l), and anterior insula (m).
www.pnas.org?cgi?doi?10.1073?pnas.0703084104 Schabus et al.
Fig. 3a). This finding is consistent with the temporal correlation
reported in rodents between hippocampal and mesio-frontal neu-
ronal discharges, respectively structured by hippocampal sharp
waves/ripples and cortical spindles (24). It would be tempting to
relate hippocampal and mesio-frontal activities associated with fast
spindles to memory processing. Indeed, power in the fast spindle
range (?13 Hz) increases after encoding of hippocampal-
dependent declarative memories (25) and after procedural motor
spindles (11.25–13.75 Hz)]. However, we did not submit our vol-
unteers to any systematic training session before sleep. We only
show that, even without any previous training, both the hippocam-
pus and mesial-frontal cortex are preferentially active during fast,
relative to slow, spindles, a condition which could promote their
Activity Associated with Slow and Fast Spindles in Thalami and Other
Subcortical Structures. The activities associated with fast and slow
spindles were much less distinct in thalami than in the cortex. Slow
spindles were associated with increased activity in the bulk of both
thalami. For fast spindles, the situation was more complex. They
elicited significant response only as observed in the common
thalamic activation pattern, restricted to lateral and posterior part
of both thalami. However, at a lower statistical threshold, an
increased activity was detected in a large part of both thalami. This
finding explains why no significant difference could be revealed
between slow and fast spindles. A cautious interpretation of the
results would conclude that the major part of both thalami is
recruited during both fast and slow spindles although to variable
suggest that TC and reticular thalamic neurons constitute two
different populations, each characterized by fairly homogeneous
cellular properties (23). Consequently, thalamic populations are
thought to be able to generate a continuum of frequencies, rather
and posterior parts of thalami is preferentially associated with fast
spindles remains speculative. We assume that the cortical areas
which connect to the corresponding thalamic nuclei might contrib-
ute to modulate spindle-related activity. We found that the ana-
esis. Peak thalamic voxels associated with slow spindles were
primarily located in areas compatible with the mediodorsal nucleus
(9). In contrast, peak voxels related to fast spindles were addition-
(8) and were likely to connect to primary sensorimotor, premotor,
and posterior parietal cortices (9) (cf. SI Table 6).
In addition, we observed a significant relationship between the
activity in peak thalamic voxels for slow spindles and superior
frontal gyrus and between the activity recorded in peak thalamic
Table 1. Brain areas showing significant spindle-related increase in activity
Slow spindlesConjunctionFast spindles
Midbrain tegmentum (13)
Orbito-frontal cortex (33)
Superior Frontal Gyrus (30)
Middle frontal gyrus (13, 34)
Supplementary motor area (33)
Precentral gyrus (33)
Postcentral gyrus (33)
Anterior cingulate cortex
Mid-cingulate cortex (33)0
Anterior insula (13)
?6 3.52 0.029
Posterior insula (13)
Superior temporal gyrus (33)
Coordinates (x, y, z) are expressed in millimeters in the Montreal Neurological Institute; Z scores result from the statistical parametric analysis; pSVCrefers to
the probability of the null hypothesis (i.e., absence of activity change associated with spindles), after correction for multiple comparisons on small volumes of
interest identified in the literature (references in brackets after the name of each brain area). Additional results that do not survive correction for multiple
comparisons can be found in SI Tables 3–5.
Table 2. Differential brain responses to slow and fast sleep
spindles in stereotactic space
Slow ? fast
Superior frontal gyrus (30) 26
Fast ? slow
6032 3.78 0.012
Mesial prefrontal cortex (12)
Orbito-frontal cortex (33)
Middle frontal gyrus (34)
Precentral sulcus (33)
Post central gyrus (33)
Anterior insula (33)
Statistical results are presented as described in Table 1.
Schabus et al.
August 7, 2007 ?
vol. 104 ?
no. 32 ?
voxels for fast spindles and cingulate motor zones and SMA. A
detailed characterization of the neural firing patterns in these TC
loops at the cellular level is required to confirm this hypothesis.
Significant responses associated with slow spindles in basal
forebrain/hypothalamus and midbrain tegmentum were unex-
pected, because it is known that their activity prevents spindle
generation through direct cholinergic projections to the thalamus
(28). The spatial resolution of fMRI does not allow to assign the
detected signal to any given subpopulation of neurons. Neverthe-
less, our results show that basal forebrain and midbrain tegmentum
are systematically recruited during slow spindles. Whether this
recruitment is necessary for spindle generation (possibly in relation
with spindle termination) or is a mere corollary of TC oscillations
should be further characterized at the cellular level. Likewise, the
cerebellum has never been specifically implicated in spindle gen-
eration although it is in position to modulate thalamic activity
through direct cerebello-thalamic projections (29). Further re-
search should therefore specify its functional participation in slow
Methodological Issues. Ourresultsdescribeincreasedregionalbrain
activity in association with spindles and contrast with decreases in
regional cerebral blood flow and glucose metabolism reported
during light NREM sleep (12, 13, 30) or in relation to sigma power
(2) with positron emission tomography. Although the latter tech-
nique characterized the hemodynamic and metabolic changes
averaged over tens of seconds, fMRI because of its better temporal
resolution, was sensitive to transient increases in brain activity
associated with spindles and their corresponding neural events,
relative to baseline activity of light NREM sleep. However, fMRI
data analysis is based on multiple regressions and is sensitive only
to consistent increases in activity associated with spindles. In
contrast to EEG studies (18), it is blind to the particular temporal
dynamics of neural activity during any given spindle.
On the other hand, fMRI benefits of a good spatial resolution
and is sensitive to the activity in deep brain structures which can
only be inferred by source reconstruction in scalp EEG or MEG
data (5, 21). However, assigning activation sites to microscopic
neuronal ensembles or nuclei remains uncertain. For instance, the
probability of detecting activation in the reticular nucleus was small
although the nucleus is deeply involved in spindle generation.
Spindles are the hallmark of light NREM sleep and represent
in animals at the cellular level. Using simultaneous EEG/fMRI, we
were able to identify the cerebral correlates of spindles in non-
sleep-deprived normal volunteers. As a rule, and in contrast to
previous neuroimaging studies, spindles were associated with tran-
sient increases in regional brain activity. In addition, we were able
to specify both commonalities and differences in brain responses to
slow and fast spindles. Intriguingly, we observed substantial differ-
ences in cortical, rather than thalamic, activity between the two
spindle types. These findings corroborate the existence of two spin-
dle types during human NREM sleep and suggest that fast spindles
participate in processing sensorimotor and mnemonic information.
Materials and Methods
Population. Participants were healthy, young subjects (n ? 25; 11
females; mean age, 21.96). A semistructured interview established
the absence of medical, traumatic, or psychiatric history, and of
sleep disorders. All participants were nonsmokers, moderate caf-
feine and alcohol consumers and none were on medication. Par-
ticipants gave their written informed consent and received a finan-
cial compensation for their participation. The study was approved
by the Ethics Committee of the Faculty of Medicine of the
University of Lie `ge.
first visit to the laboratory and were not sleep-deprived. Compli-
ance to the schedule was assessed by using a wrist actigraphy
(Actiwatch; Cambridge Neuroscience, Cambridge, U.K.) and sleep
diaries. Volunteers were requested to refrain from all caffeine and
before participating in the study.
EEG Acquisition and Analysis. EEG was recorded by using two
MR-compatible 32-channel amplifiers (Brainamp MR plus; Brain
Products, Gilching, Germany) and a MR-compatible EEG cap
(Braincap MR; Falk Minow Services, Herrsching-Breitbrunn, Ger-
many) with 64 ring-type electrodes. EEG caps included 62 scalp
electrodes that were online referenced to FCz, and one electroocu-
logram and one electrocardiogram channel. EEG was digitised at
5,000 Hz sampling rate with a 500-nV resolution. Data were
analog-filtered by a bandlimiter low pass filter at 250 Hz (30 dB per
ing to a high pass frequency of 0.0159 Hz. Data were transferred
outside the scanner room through fibre optic cables to a personal
computer where the EEG system running Vision Recorder Soft-
ware v1.03 (Brain Products) was synchronized to the scanner clock.
For analysis, EEG data were low-pass filtered (finite impulse
05 10 1520 25 30
05 1015 202530
05 10 1520 2530
0510 1520 25 30
hippocampus (a), mesial prefrontal cortex (b), precentral gyrus (c), and postcentral gyrus (d). Peristimulus time histograms show mean response of the
corresponding voxels (dotted lines; error bars show SEM) and the corresponding fitted responses (continuous lines).
Differential fMRI activity between fast and slow spindles. Larger brain responses for fast (red) than slow (black) spindles were revealed in the
www.pnas.org?cgi?doi?10.1073?pnas.0703084104Schabus et al.
response filter, ?36 dB at 70 Hz), down-sampled to 250 Hz, and Download full-text
re-referenced to linked mastoids. Scanner artefacts were removed
in Vision Analyzer software, using an adaptive average subtraction
(31). Ballistocardiographic artefacts were removed by using an
algorithm based on independent component analysis (32). Sleep
staging followed standard criteria, and identified periods of stage 2
and stage 3 sleep, free of any artifact, during which the EEG and
fMRI data were analyzed. Sleep spindles were automatically de-
tected on Cz (7). For slow spindle detection, data were bandpass
filtered between 11 and 13 Hz, using linear phase finite impulse
response filters (?3 dB at 11.1 and 12.9 Hz). The root mean square
of the filtered signal was calculated by using a time window of 0.25
sec. Sleep spindles were identified by thresholding the spindle root
mean square signal at its 95th percentile. The same procedure was
Hz (?3 dB at 13.1 and 14.9 Hz).
acquired by using a three-Tesla MR scanner (Allegra; Siemens,
Erlangen, Germany). Multislice T2*-weighted fMRI images were
obtained with a gradient echo-planar sequence, using axial slice
orientation (32 slices; voxel size, 3.4 ? 3.4 ? 3 mm3; matrix size,
64 ? 64 ? 32; repetition time (TR) ? 2460 ms; echo time (TE) ?
were scanned during the first half of the night, starting at around
midnight. They stayed until they indicated by button press that they
would like to go out, or for a maximum of 4,000 scans (?164 min).
The number of scans acquired varied between 1,870 and 4,000
(2,770 ? 852 scans or 113.6 min ? 34.9 min [mean ? SD]). A
structural T1-weigthed 3D MP-RAGE sequence (TR ? 1,960 ms;
TE ? 4.43 ms; inversion time, 1,100 ms; field of view, 230 ? 173
mm2; matrix size, 256 ? 192 ? 176; voxel size, 0.9 ? 0.9 ? 0.9 mm)
was also acquired in all subjects.
Mapping 5 (SPM5; www.fil.ion.ucl.ac.uk/spm/software/spm5). The
series of consecutive fMRI volumes corresponding to a chosen
constituted a session (cf. SI Fig. 4). Selected fMRI time series were
corrected for head motion, spatially normalized (two-dimensional
spline; voxel size, 2 ? 2 ? 2 mm3) to an echo planar imaging
template conforming to the Montreal Neurological Institute space,
and spatially smoothed with a Gaussian Kernel of 8 mm FWHM.
The analysis of fMRI data, based on a mixed effects model, was
conducted in two serial steps, accounting respectively for intrain-
dividual (fixed) and interindividual (random effects) variance. For
were convolved with the three canonical basis functions (hemody-
namic response function, its derivative and dispersion), and used as
regressors in the individual design matrix. The square root of the
energy of the signal in the 0.5–4 Hz frequency band was averaged
over each repetition time, convolved with the hemodynamic re-
sponse function and included as another regressor. Movement
parameters estimated during realignment (translations in x, y, and
z directions and rotations around x, y, and z axes) and a constant
vector were also included in the matrix as a variable of no interest.
Serial correlations in fMRI signal were estimated by using an
autoregressive (order 1) plus white noise model and a restricted
maximum likelihood algorithm. The effects of interest were then
tested by linear contrasts (main effects of each spindle type,
differential response between spindles types), generating statistical
parametric maps [(SPM(T)]. The resulting contrasts images were
then further smoothed (6 mm FWHM Gaussian Kernel) and
entered in a second-level analysis. The second-level analysis con-
sisted of an analysis of variance with the three basis functions and
two spindle types as factors. The error covariance was not assumed
independent between regressors and a correction for nonsphericity
was applied. The resulting set of voxel values constituted maps of
F statistics [SPM(F)]. To correct for multiple comparisons, results
are reported in a priori regions of interest previously identified in
neuroimaging studies of NREM sleep (12, 13, 30, 33, 34) on small
spherical volumes (10 mm sphere, i.e., ?4,000 mm3; SVC). Con-
junction analysis used SPMs of the minimum T-statistic over the
which requires significant regions to be present in all tested con-
ditions (i.e., a logical ‘‘AND’’ conjunction). As above, inferences
were based on P values adjusted for the search volume using
random field theory. Probabilities of thalamus connectivity were
is based on diffusion tensor imaging (www.fmrib.ox.ac.uk/
A more detailed description of materials and methods and
thalamic connectivity data can be found in the SI Materials and
This study was supported by the Belgian Fonds National de la Recherche
the Fondation Me ´dicale Reine Elisabeth, the Research Fund of ULg,
PAI/IAP Interuniversity Pole of Attraction P5/04, a PhD grant from the
French Ministe `re de la Recherche (to G.A.), an Emmy Noether Fellowship
from the German Research Foundation (to S.G.), and an Austrian Science
Fund Erwin-Schro ¨dinger Fellowship J2470-B02 (to M.S.).
2. Hofle N, Paus T, Reutens D, Fiset P, Gotman J, Evans AC, Jones BE (1997) J Neurosci
3. De Gennaro L, Ferrara M (2003) Sleep Med Rev 7:423–440.
4. Bodizs R, Kis T, Lazar AS, Havran L, Rigo P, Clemens Z, Halasz P (2005) J Sleep Res
5. AndererP,Klo ¨schG,GruberG,TrenkerE,Pascual-MarquiRD,ZeitlhoferJ,BarbanojMJ,
Rappelsberger P, Saletu B (2001) Neuroscience 103:581–592.
6. Zygierewicz J, Blinowska KJ, Durka PJ, Szelenberger W, Niemcewicz S, Androsiuk W
(1999) Clin Neurophysiol 110:2136–2147.
7. Mo ¨lle M, Marshall L, Gais S, Born J (2002) J Neurosci 22:10941–10947.
8. Morel A, Magnin M, Jeanmonod D (1997) J Comp Neurol 387:588–630.
9. Behrens TE, Johansen-Berg H, Woolrich MW, Smith SM, Wheeler-Kingshott CA, Boulby
PA, Barker GJ, Sillery EL, Sheehan K, Ciccarelli O, et al. (2003) Nat Neurosci 6:750–757.
10. Steriade M, Parent A, Hada J (1984) J Comp Neurol 229:531–547.
11. Pare D, Steriade M, Deschenes M, Oakson G (1987) J Neurophysiol 57:1669–1685.
12. Maquet P, Degueldre C, Delfiore G, Aerts J, Peters JM, Luxen A, Franck G (1997)
J Neurosci 17:2807–2812.
13. Braun AR, Balkin TJ, Wesenten NJ, Carson RE, Varga M, Baldwin P, Selbie S, Belenky
G, Herscovitch P (1997) Brain 120(Pt 7):1173–1197.
14. Giguere M, Goldman-Rakic PS (1988) J Comp Neurol 277:195–213.
15. Portas CM, Krakow K, Allen P, Josephs O, Armony JL, Frith CD (2000) Neuron 28:991–999.
16. Cote KA, Epps TM, Campbell KB (2000) J Sleep Res 9:19–26.
17. Nakamura M, Uchida S, Maehara T, Kawai K, Hirai N, Nakabayashi T, Arakaki H, Okubo
Y, Nishikawa T, Shimizu H (2003) Neurosci Res 45:419–427.
18. Doran S (2003) Sleep Res Online 5:133–139.
19. Hari R, Salmelin R (1997) Trends Neurosci 20:44–49.
20. Pineda JA (2005) Brain Res Rev 50:57–68.
21. Manshanden I, De Munck JC, Simon NR, Lopes da Silva FH (2002) Clin Neurophysiol
22. Sterman MB, Howe RC, Macdonald LR (1970) Science 167:1146–1148.
23. Steriade M, Amzica F (1998) Sleep Res Online 1:1–10.
24. Siapas AG, Wilson MA (1998) Neuron 21:1123–1128.
25. Schabus M, Ho ¨dlmoser K, Gruber G, Sauter C, Anderer P, Klo ¨sch G, Parapatics S, Saletu
B, Klimesch W, Zeitlhofer J (2006) Eur J Neurosci 23:1738–1746.
26. Milner CE, Fogel SM, Cote KA (2006) Biol Psychol 73:141–156.
27. Schmidt C, Peigneux P, Muto V, Schenkel M, Knoblauch V, Munch M, de Quervain DJ,
Wirz-Justice A, Cajochen C (2006) J Neurosci 26:8976–8982.
28. McCormick DA, Bal T (1997) Annu Rev Neurosci 20:185–215.
30. Kajimura N, Uchiyama M, Takayama Y, Uchida S, Uema T, Kato M, Sekimoto M,
Watanabe T, Nakajima T, Horikoshi S, et al. (1999) J Neurosci 19:10065–10073.
31. Allen PJ, Josephs O, Turner R (2000) NeuroImage 12:230–239.
32. Srivastava G, Crottaz-Herbette S, Lau KM, Glover GH, Menon V (2005) NeuroImage
33. Kaufmann C, Wehrle R, Wetter TC, Holsboer F, Auer DP, Pollmacher T, Czisch M (2006)
34. Nofzinger EA, Buysse DJ, Miewald JM, Meltzer CC, Price JC, Sembrat RC, Ombao H,
Reynolds CF, Monk TH, Hall M, et al. (2002) Brain 125:1105–1115.
35. Nichols T, Brett M, Andersson J, Wager T, Poline JB (2005) NeuroImage 25:653–660.
Schabus et al.
August 7, 2007 ?
vol. 104 ?
no. 32 ?