Dreaming in NREM sleep: A high-density EEG study of slow waves and spindles

Abstract

Dreaming can occur in both rapid eye movement (REM) and non-REM (NREM) sleep. We recently showed that in both REM and NREM sleep, dreaming is associated with local decreases in slow wave activity (SWA) in posterior brain regions. To expand these findings, here we asked how specific features of slow waves and spindles, the hallmarks of NREM sleep, relate to dream experiences. Fourteen healthy human subjects (10 females) underwent nocturnal high-density EEG recordings combined with a serial awakening paradigm. Reports of dreaming, compared with reports of no experience, were preceded by fewer, smaller, and shallower slow waves, and faster spindles, especially in central and posterior cortical areas. We also identified a minority of very steep and large slow waves in frontal regions, which occurred on a background of reduced SWA and were associated with high-frequency power increases (local “microarousals”) heralding the successful recall of dream content. These results suggest that the capacity of the brain to generate experiences during sleep is reduced in the presence of neuronal off-states in posterior and central brain regions, and that dream recall may be facilitated by the intermittent activation of arousal systems during NREM sleep.
SIGNIFICANCE STATEMENT By combining high-density EEG recordings with a serial awakening paradigm in healthy subjects, we show that dreaming in non-rapid eye movement sleep occurs when slow waves in central and posterior regions are sparse, small, and shallow. We also identified a small subset of very large and steep frontal slow waves that are associated with high-frequency activity increases (local “microarousals”) heralding successful recall of dream content. These results provide noninvasive measures that could represent a useful tool to infer the state of consciousness during sleep.

Full text

Behavioral/Cognitive
Dreaming in NREM Sleep: A High-Density EEG Study of Slow
Waves and Spindles
XFrancesca Siclari,
1
XGiulio Bernardi,
1,3
Jacinthe Cataldi,
1
and Giulio Tononi
2
1
Lausanne University Hospital and University of Lausanne, 1011 Lausanne, Switzerland,
2
Department of Psychiatry, University of Wisconsin-Madison,
Madison, Wisconsin 53719, and
3
MoMiLab Unit, IMT School for Advanced Studies Lucca, 55100 Lucca, Italy
Dreaming can occur in both rapid eye movement (REM) and non-REM (NREM) sleep. We recently showed that in both REM and NREM
sleep, dreaming is associated with local decreases in slow wave activity (SWA) in posterior brain regions. To expand these findings, here
we asked how specific features of slow waves and spindles, the hallmarks of NREM sleep, relate to dream experiences. Fourteen healthy
human subjects (10 females) underwent nocturnal high-density EEG recordings combined with a serial awakening paradigm. Reports of
dreaming, compared with reports of no experience, were preceded by fewer, smaller, and shallower slow waves, and faster spindles,
especially in central and posterior cortical areas. We also identified a minority of very steep and large slow waves in frontal regions, which
occurred on a background of reduced SWA and were associated with high-frequency power increases (local “microarousals”) heralding
the successful recall of dream content. These results suggest that the capacity of the brain to generate experiences during sleep is reduced
in the presence of neuronal off-states in posterior and central brain regions, and that dream recall may be facilitated by the intermittent
activation of arousal systems during NREM sleep.
Key words: consciousness; dream; high-density EEG; sleep; slow wave
Introduction
What determines the level of consciousness during sleep? Why
are we sometimes unconscious, while at other times we have vivid
conscious experiences in the form of dreams? When rapid eye
movement (REM) sleep was first described in humans (Aserinsky
and Kleitman, 1955), the explanation seemed relatively straight-
forward: 74% of subjects woken up during this state reported that
they had been dreaming, compared with only 17% awakened at
other times. It is not surprising that following this observation,
Aserinsky and Kleitman (1955) claimed to have “furnished the
means of determining the incidence and duration of periods of
dreaming.” Indeed, the fast-frequency, desynchronized EEG ac-
tivity of REM sleep, similar to the waking EEG, appeared to offer
the ideal premises for conscious experiences to occur. In the fol-
lowing years, however, the assumption that dreaming and con-
scious experiences were synonymous with REM sleep was
challenged (Solms, 2000). Pharmacological suppression of REM
sleep, for instance, did not eliminate dreaming (Oudiette et al.,
2012), and specific forebrain lesions (Solms, 2000) were shown to
suppress dreaming without affecting REM sleep. In addition, by
changing the question from “Tell me whether you had a dream”
to “Tell me what was going through your mind,” reports of con-
scious experiences in non-REM (NREM) sleep were obtained in
up to 70% of cases (Stickgold et al., 2001). Especially in the early
morning hours, NREM sleep dream reports appeared indistin-
guishable from REM sleep dream reports (Monroe et al., 1965;
Antrobus et al., 1995). This raised the question of whether slow
Received March 29, 2018; revised July 30, 2018; accepted Aug. 1, 2018.
Author contributions: F.S. and G.T. designed research; F.S. and J.C. performed research; F.S. and G.B. analyzed
data; F.S. wrote the paper.
This work was supported by Swiss National Science Foundation Grant PZ00P3_173955 (F.S.), the Divesa Foun-
dation Switzerland (F.S.), the Pierre-Mercier Foundation for Science (F.S.), the Bourse Pro-Femme of the University
of Lausanne (F.S.), an EMBO short-term postdoctoral fellowship (G.B.), National Institutes of Health (NIH)/National
Center for Complementary and Alternative Medicine Grant P01-AT-004952 (G.T.), NIH/National Institute of Mental
Health Grant 5P20-MH-077967 (G.T.), and the Tiny Blue Dot Inc. Grant MSN196438/AAC1335 (G.T.).
The authors declare no competing financial interests.
Correspondence should be addressed to Francesca Siclari, Center for Research and Investigation in Sleep
(CIRS), Lausanne University Hospital (CHUV), Rue du Bugnon 46, 1011 Lausanne, Switzerland. E-mail:
francesca.siclari@chuv.ch.
DOI:10.1523/JNEUROSCI.0855-18.2018
Copyright © 2018 the authors 0270-6474/18/389175-11$15.00/0
Significance Statement
By combining high-density EEG recordings with a serial awakening paradigm in healthy subjects, we show that dreaming in
non-rapid eye movement sleep occurs when slow waves in central and posterior regions are sparse, small, and shallow. We also
identified a small subset of very large and steep frontal slow waves that are associated with high-frequency activity increases (local
“microarousals”) heralding successful recall of dream content. These results provide noninvasive measures that could represent
a useful tool to infer the state of consciousness during sleep.
The Journal of Neuroscience, October 24, 2018 •38(43):9175–9185 • 9175

waves, which decrease across the night as a function of homeo-
static sleep pressure, may interfere with the generation of dream
experiences (DEs). Experimental and theoretical work offers
plausible explanations of why this should be so: neuronal “off-
periods” associated with slow waves disrupt causal interactions
between thalamocortical regions (Massimini et al., 2005;Pigorini
et al., 2015), thereby impairing information integration, which
has been proposed as a prerequisite for consciousness (Tononi,
2008). Initial attempts to relate EEG changes in slow-wave activ-
ity (SWA; spectral power, 1– 4 Hz) to dreaming yielded variable
results (Williamson et al., 1986;Esposito et al., 2004;Chellappa et
al., 2011;Marzano et al., 2011). In recent years, it has become
clear that the alternation between off-periods and “on-periods”
underlying the surface EEG slow wave is not global, but occurs
mostly locally, meaning that it can occur out of phase with respect
to other cortical regions (Nir et al., 2011). It has also become clear
that conventional EEG recordings comprising only a few elec-
trodes would not be able to capture such local SWA changes. In
addition, slow waves with neuronal off-periods have been shown
to occur not only in NREM sleep, but also in superficial cortical
layers of primary sensory cortices in REM sleep in mice (Funk et
al., 2016). To account for these aspects, in a recent study we used
high-density EEG (hd-EEG) to investigate changes in brain activ-
ity associated with dreaming. We demonstrated that reports of
dreaming were preceded by decreases in SWA in both REM and
NREM sleep. These decreases were not global, but spatially re-
stricted to a “posterior hot zone” of the brain, suggesting that
local changes in SWA may account for the presence of dreaming
or unconsciousness across different behavioral states. Here we
aimed to extend these findings. Instead of assessing spectral
power, a rough estimate of both slow-wave number and ampli-
tude, we investigated how specific characteristics of slow waves
relate to dreaming. Indeed, specific slow-wave features have been
shown to reflect different aspects of neuronal oscillations (Esser
et al., 2007;Riedner et al., 2007;Vyazovskiy et al., 2007), includ-
ing the degree of cortical bistability (density), the number of
neurons that simultaneously enter a down-state (amplitude), and
synaptic strength (slope). Based on these criteria, we were re-
cently able to characterize two types of slow waves (Siclari et al.,
2014;Bernardi et al., 2018): widespread, steep, and large-
amplitude type I slow waves in frontocentral brain regions that
are likely generated by a subcorticocortical synchronization pro-
cess; and type II slow waves with a smaller amplitude and slope
and a more variable regional involvement, which are likely gen-
erated by a corticocortical synchronization process. To deter-
mine how these two types of slow waves relate to dreaming, and
to extend our findings to sleep spindles, we analyzed high-density
EEG recordings of 14 healthy participants who underwent a serial
awakening paradigm.
Materials and Methods
Participants and procedure. Fourteen healthy participants (mean ⫾SD
age, 33.6 ⫾8.5 years; age range, 22– 47 years; 10 females) screened for
medical, neurological, and psychiatric disorders participated in the
study. None of the subjects was receiving psychotropic medication. All of
the participants had good sleep quality, as assessed by the Pittsburgh
Sleep Quality Index (⬍5 points), and scored within normal limits on the
Epworth Sleepiness Scale. Dream characteristics and spectral power
changes associated with dreaming in the delta (1– 4 Hz) and gamma
(20 –50 Hz) frequency ranges of subjects 1–7, but not slow-wave or spin-
dle findings, have been previously reported (experiment 2; Siclari et al.,
2017), while subjects 8 –14 were newly included. Subjects 1–7 were
acquired at the University of Wisconsin-Madison, while subjects 8 –14
were studied at the Lausanne University Hospital. Written informed
consent was obtained from each participant, and the study was ap-
proved as part of a larger project by the local ethical committees
(University of Wisconsin Madison and Lausanne University Hospi-
tal). Participants were recorded with hd-EEG (256 electrodes) while
they slept in the sleep laboratory. Throughout the night, they were
awakened repeatedly (Siclari et al., 2013) and asked to report whether,
just before the awakening, they had been experiencing anything (DE),
had been experiencing something but could not remember the con-
tent [DE without recall of content (DEWR)] or had not been experi-
encing anything [no experience (NE); Siclari et al., 2013]. The serial
awakening protocol has previously been reported in detail (Siclari et
al., 2013). Awakenings were performed using a computerized alarm
sound lasting 1.5 s.
All subjects were studied by the first author of this article and under-
went the same serial awakening paradigm (Siclari et al., 2013), except for
the fact that University of Wisconsin-Madison subjects were studied for
5–10 nights while Lausanne University Hospital subjects underwent only
2 study nights. As a consequence, University of Wisconsin-Madison sub-
jects had more total awakenings compared with Lausanne University
Hospital subjects (113.6 ⫾26.0 vs 24.86 ⫾8.23 per subject; p⬍0.001,
unpaired two-tailed ttests). No differences between the two groups were
observed in the proportion of NREM sleep awakenings with respect to
total awakenings (75.00 ⫾5.30% vs 78.81 ⫾2.60%); in the proportion of
N2 or N3 awakenings with respect to NREM (N2 ⫹N3) awakenings
(55.9 ⫾13.69% vs 53.45 ⫾22.71% N2 awakenings, p⫽0.81); and in the
proportion of DEs (30.96 ⫾14.49% vs 34.13 ⫾20.64%, p⫽0.74),
DEWRs (38.55 ⫾13.88 vs 40.30 ⫾21.40, p⫽0.86), and NE (30.50 ⫾
21.75 vs 24.97 ⫾24.69, p⫽0.66) in NREM sleep. The study groups did
not differ significantly in terms of age (31.57 ⫾8.5 vs 35.71 ⫾8.6 years of
age; p⫽0.38) and female/male distribution (4/7 vs 5/7 females/males;
p⫽0.57, Pearson’s
␹
2
test). All of the comparisons reported herein were
performed within subjects.
Recordings. Sleep recordings were performed using a 256-channel hd-
EEG system (Electrical Geodesics). Four of the 256 electrodes placed at
the outer canthi of the eyes were used to monitor eye movements, and
submental electromyography was recorded using the Electrical Geode-
sics polygraph input box. The EEG signal was sampled at 500 Hz, and was
off-line bandpass filtered between 0.3 and 50 Hz for the first set of par-
ticipants (subjects 1–7) and between 0.3 and 45 Hz for the second set of
participants (subjects 8 –14, because of the different frequencies of elec-
trical artifacts in the United States and Europe). All of the analyses were
performed on frequency bands available for both studies (⬍40 Hz). Sleep
scoring was performed over 30 s epochs according to standard criteria
(Iber et al., 2007).
Preprocessing of data. The signal corresponding to the 2 min preceding
each awakening in stages N2 and N3 was extracted and considered for
analysis. Channels containing artifactual activity were visually identified
and replaced with data interpolated from nearby channels using
spherical splines (NetStation, Electrical Geodesic). To remove ocular,
muscular, and electrocardiograph artifacts, we performed indepen-
dent component analysis (ICA) using EEGLAB routines (Delorme
and Makeig, 2004). Only ICA components with specific activity pat-
terns and component maps characteristic of artifactual activity were
removed (Jung et al., 2000).
Slow-wave detection. For slow-wave detection, we used an automatic
detection algorithm that was adapted from a previous study (Siclari et al.,
2014). The EEG signal was referenced to the average of the two mastoid
electrodes, downsampled to 128 Hz, and bandpass filtered (0.5– 4 Hz;
stop-band at 0.1 and 10 Hz) using a Chebyshev Type II filter (Matlab,
MathWorks). Only slow waves with a duration between 0.25 and 1 s were
considered. The algorithm was applied to all channels, and the following
slow-wave parameters were analyzed: density, negative peak amplitude,
slope 1 (between the first zero crossing and the negative peak), slope 2
(between the negative peak and the second zero crossing), and the num-
ber of negative peaks. For an initial, explorative analysis, we considered
large-amplitude slow waves separately from other slow waves. Large-
9176 •J. Neurosci., October 24, 2018 •38(43):9175–9185 Siclari et al. •Dreaming in NREM Sleep

amplitude slow waves (⬎75th percentile) were identified for each subject
separately by determining the amplitude corresponding to the 75th per-
centile of the amplitude distribution of all slow waves (all of the 2 min
segments preceding awakenings in N2 and N3 were concatenated for this
analysis).
Separation between type I and type II slow waves. Then, for a more
precise separation between type I and type II slow waves, we used a
previously described approach (Bernardi et al., 2018), which also has the
advantage that it can classify slow waves irrespective of their topograph-
ical distribution. We first created a single timing reference. To do this, the
negative signal envelope was generated by se-
lecting the fifth most negative sample across all
mastoid-referenced channels for each time
point. The fifth most negative channel was ar-
bitrarily selected instead of the most negative
channel to avoid including residual high-
amplitude artifacts (Siclari et al., 2014;Ber-
nardi et al., 2016). Then, a slow-wave detection
procedure was applied to this composite signal
after bandpass filtering (0.5– 40 Hz). To distin-
guish between type I and type II slow waves, for
each detection we calculated a synchronization
score defined as the scalp involvement (ex-
pressed as the percentage of channels showing
a negative averaged current value less than ⫺5
␮
V in the 40 ms time window centered on the
wave peak) multiplied by the mean slope of the
slow wave (i.e., the mean of slope 1 and slope 2;
Bernardi et al., 2018). The 5
␮
V criterion is an
arbitrary cutoff to exclude small baseline fluc-
tuations that are likely unrelated to slow waves
(Siclari et al., 2014;Bernardi et al., 2016). Based
on the distribution of the synchronization
scores of all the slow waves in N2 and N3, we
defined an absolute threshold that corre-
sponded to 4 median absolute deviations
(MADs) from the median. This threshold was
calculated for each subject separately. Type I
slow waves were defined as slow waves charac-
terized by a synchronization score greater than
the threshold, while the remaining slow waves
were classified as type II.
Spectral power analysis. To calculate spectral
power changes associated with type I and type
II slow waves, we computed power spectral
density (PSD) for each channel in the following
two 6 s periods: a prewave period (ending 1 s
before the first zero crossing of the slow wave
detected in the composite reference signal) and
a postwave period (starting 1 s after the second
zero crossing of the slow wave). Spectral power
was calculated in 2 s Hamming windows using
the Welch’s modified periodogram method
(implemented with the pwelch function in
Matlab). We then averaged prewave and post-
wave spectral power across the three 2 s bins for
each channel. To evaluate changes in spectral
power induced by type I and type II slow waves,
we subtracted postwave power from the pre-
wave power and expressed the changes as a per-
centage, so that positive values represent
increases in postwave with respect to pre-
wave power, while negative values represent
decreases in postwave with respect to pre-
wave power. All of the analyses were per-
formed within subjects first, then averaged
across subjects.
Source modeling. To evaluate the cortical lo-
calization of high-frequency power increases
after type I slow waves, a source modeling anal-
ysis of the 18 –30 Hz bandpass-filtered signal was performed using Geo-
Source 3.0 (NetStation, Electrical Geodesics). A four-shell head model
based on the Montreal Neurological Institute atlas and standard elec-
trode coordinates were used to construct the forward model. The inverse
matrix was defined using the standardized low-resolution brain electro-
magnetic tomography method (Tikhonov regularization,
␭
⫽10
⫺2
).
The source-modeling analysis was performed on 6 s windows preceding
and following type I slow waves, and root mean square values in the
specified frequency range were used to compute relative postwave–pre-
Figure 1. Top row, Topographical distribution of tvalues for the contrast between dream experiences and no experiences for
different slow-wave parameters, averaged over the last 60 s before the awakening. Bottom row, Same as the top row, but
electrodes within a cluster showing a statistically significant effect are marked in white ( p⬍0.05, cluster-based correction for
multiple comparisons, two-tailed paired ttests; n⫽12 subjects).
Figure 2. Top row, Topographical distribution of slow-wave parameters for dream experiences (DEs) (first row) and no expe-
riences (NEs) (second row). Slow-wave parameters were averaged over the last 60 s before the awakening and across 12 subjects.
In the third row, the mean differences between DE and NE (DE ⫺NE) are shown for each parameter, so that red colors indicate
higher values in DE, and blue colors higher values in NE. In the fourth row, pvalues for paired electrode-by-electrode ttests are
shown ( p⬍0.05, uncorrected).
Siclari et al. •Dreaming in NREM Sleep J. Neurosci., October 24, 2018 •38(43):9175–9185 • 9177

wave variations. These values were averaged
across slow waves and subjects.
Spindle detection. Spindle detection was based
on an automatic algorithm, adapted from a pre-
vious study (Ferrarelli et al., 2007). The average-
referenced signal was downsampled to 128 Hz
and bandpass filtered between 11 and 16 Hz
(⫺3 dB at 10 and 17 Hz). The following spindle
parameters were analyzed: density, maximal
amplitude, and frequency. Based on a previous
studies (Anderer et al., 2001;Andrillon et al.,
2011;Molle et al., 2011), we categorized spin-
dles as fast or slow based on an individualized
threshold, which was defined as the intermedi-
ate value between the average spindle fre-
quency in one centroparietal channel (Pz) and
one frontal channel (Fz; Siclari et al., 2014).
Experimental design and statistical analysis. Sta-
tistical analyses were performed in Matlab
(MathWorks). Analyses regarding general slow-
wave properties (see Figs. 7,8C–E) were per-
formed on the totality of the available signal (120
s before the awakening), while analyses referring
to dream reports (i.e., comparisons among DE,
NE, and DEWR shown in Figs. 1,2,3,4,5,6; see
also Figs. 8B,10,11) were performed on slow-
wave and spindle parameters in the 60 s before
the awakening. Topographical analyses were
limited to the 185 innermost channels to
avoid the common artifactual contamina-
tion of electrodes located on the cheeks and
neck.
To compare brain activity between DE and
NE (and DEWR, when applicable), slow-wave
and spindle parameters were first averaged
within the 60 s preceding each awakening for
each subject. We then averaged the slow-wave
and spindle parameters associated with DEs
and NEs within each subject. Whole-brain
group-level comparison on average slow wave
and spindle were performed using a cluster-
based correction for multiple comparisons
(Nichols and Holmes, 2002). Two-tailed
paired ttests with a corrected p⬍0.05 thresh-
old were performed. The number of subjects
for each comparison is indicated in the figure
legends. Correction for multiple comparisons
was ensured using a permutation-based su-
prathreshold cluster analysis (Nichols and
Holmes, 2002;Huber et al., 2004). In brief, for
each comparison new datasets were generated
by randomly relabeling the condition label
from original data and paired ttests were per-
formed (N⫽5000). For each iteration, the
maximal size of the resulting significant clus-
ters of electrodes was stored to generate a cluster size distribution. Then,
the 95th percentile of this distribution was used as the critical cluster-size
threshold to achieve cluster-corrected pvalues corresponding to
␣
⬍0.05.
We did not perform an additional correction for multiple comparisons
based on the number of contrasts for slow-wave and spindle parameters
because many of these are not truly independent, but are highly correlated.
To investigate the correlation between slow-wave amplitude in frontal
and occipital regions (Fig. 7), we first averaged, for each 2 min data
segment, slow-wave amplitude within the following two regions of inter-
est: Fz and Oz (occipital), and immediately adjacent electrodes. We then
correlated these frontal and occipital values within subjects by taking
account of all the NREM segments (Spearman rank correlation). Statis-
tical significance was determined using a permutation-based approach.
Specifically, null distributions with properties that are comparable to
those of the original (real) distributions were generated by randomly
shuffling the two variables of interest (frontal and occipital amplitude of
slow waves) in each subject (using all available data segments) and recal-
culating the group-level mean correlation (n⫽1000). This null distri-
bution was then used to calculate the significance level of real group-level
correlation values ( p⬍0.05). The same procedure was performed to test
the correlation between the amplitude of large frontal slow waves (⬎75th
percentile) and occipital slow waves (without an amplitude threshold).
For the correlation between the synchronization index of type I and
type II slow waves and the preceding SWA (Fig. 8C), we extracted, for
each channel, the power spectral density in the 6 s window preceding
each type I and type II slow wave detected in the reference signal. We then
Figure 3. Top row, Topographical distribution of tvalues for the contrast between dream experiences without recall of content
andnoexperiences for different slow-waveparametersaveraged over the last60s before the awakening.Bottomrow, Same as the
top row, but electrodes within a cluster showing a statistically significant effect are marked in white ( p⬍0.05, cluster-based
correction for multiple comparisons, two-tailed paired ttests; n⫽12 subjects).
Figure 4. Top row, Topographical distribution of tvalues for the contrast between dream experiences and no experiences for
different slow-wave parameters averaged over the last 60 s before the awakening. For each subject, only slow waves larger than
the75thamplitudepercentile (across all slowwavesdetectedin NREM sleep) wereincluded.Bottomrow, Same as the toprow,but
electrodes within acluster showing a statistically significant effect are nowmarked in white ( p⬍0.05, cluster-based correctionfor
multiple comparisons, two-tailed paired ttests; n⫽12 subjects).
9178 •J. Neurosci., October 24, 2018 •38(43):9175–9185 Siclari et al. •Dreaming in NREM Sleep

correlated spectral power and the synchronization index for each subject
and channel using a Spearman rank correlation. Statistical significance
was determined using the same permutation-based approach described
above.
Results
Of the 969 awakenings performed across all subjects, 735 were
performed in NREM sleep stages N2 and N3. Of these, 246
(33.5%) yielded reports of DE, 284 (38.6%) of DEWR, and 204
(27.8%) of NE. Two subjects did not present NE in NREM sleep.
Slow waves and dreaming
Slow waves preceding reports of DE, com-
pared with NE, were significantly less nu-
merous, had a smaller amplitude and
slope (corrected p⬍0.05) and displayed a
trend toward more negative peaks (un-
corrected p⬍0.05). These differences
were relatively widespread, but the stron-
gest statistical effects were observed in
posterior and central brain regions (Fig. 1,
see Fig. 2 for mean absolute values, mean
differences and uncorrected statistics).
Similar results for amplitude and slope
were obtained when comparing DEWR
and NE (Fig. 3), while the contrast between
DE and DEWR did not yield any significant
differences (data not shown), suggesting
that these findings reflect differences in
dreaming, and not in the ability to recall the
content of the dream.
Next, based on our previous work, in
which we identified two types of slow
waves that differ in amplitude, we per-
formed an explorative analysis focusing
exclusively on high-amplitude slow waves,
using an arbitrary minimal threshold cor-
responding to the 75th percentile of the
negative peak. This threshold was calcu-
lated for each channel and subject sepa-
rately, by considering all the detected slow
waves in NREM sleep. Consistent with the
previous analysis (Fig. 1), reports of DE
were preceded by fewer high-amplitude
slow waves compared with NE, particu-
larly in central and posterior brain areas
(Fig. 4). However, DEs were also associ-
ated with significantly larger and steeper
high-amplitude slow waves in frontal re-
gions (corrected p ⬍0.05; Fig. 4, see Fig. 5
for mean absolute values, mean differ-
ences and uncorrected statistics), which
displayed a trend toward fewer negative
peaks (uncorrected p⬍0.05). An addi-
tional analysis (Fig. 6) revealed that fron-
tal high-amplitude slow waves were even
larger and steeper when the content of the
dream could be recalled (DE), as opposed
when to when subjects reported dreaming
but could not remember the content
(DEWR), while no significant differences
were found between DEWR and NE (data
not shown).
Together, these results suggest that
dream experiences are more likely to oc-
cur when slow waves are sparse, small, and shallow, particularly
in posterior and central brain regions. In addition, by applying a
minimal-amplitude threshold, we identified a small minority of
very steep, high-amplitude slow waves in anterior cortical regions
that tend to precede reports of dream experiences with recall of
content.
Since we showed that dreaming tends to occur when slow-
wave amplitude in posterior regions is low (Fig. 1), we asked
whether the frontal high-amplitude slow waves were associated
Figure 5. Top row, Topographical distribution of slow-wave parameters for DEs (first row) and NEs (second row). Slow-wave
parameters were averaged over the last 60 s before the awakening and across 12 subjects. In the third row, the mean differences
between dream experiences (DEs) and no experiences (NEs) (DE ⫺NE) are shown for each parameter, so that red colors indicate
higher values in DE, and blue colors higher values in NE. In the fourth row, pvalues for paired electrode-by-electrode ttests are
shown ( p⬍0.05, uncorrected). Here only slow waves larger than the 75th amplitude percentile were included for each subject.
Figure 6. Top row, Topographical distribution of tvalues for the contrast between dream experiences and dream experiences
without recall of content for different slow-wave parameters averaged over the last 60 s before the awakening. For each subject,
only slow waves larger than the 75th amplitude percentile were included. Bottom row, Same as the top row, but electrodes within
a cluster showing a statistically significant effect are marked in white ( p⬍0.05, cluster-based correction for multiple compari-
sons; n⫽14 subjects).
Siclari et al. •Dreaming in NREM Sleep J. Neurosci., October 24, 2018 •38(43):9175–9185 • 9179

with reductions in slow-wave amplitude in posterior areas. To
this aim, we first tested the correlation between the amplitude of
slow waves (without an amplitude threshold) detected in a fron-
tal region (Fz and immediately neighboring electrodes) and an
occipital region of interest (Oz and immediately neighboring
electrodes) for all 2 min NREM segments in each subject. This
analysis revealed, as expected, a strong positive correlation
between the amplitude of frontal and occipital slow waves
(Spearman rank correlation coefficient r⫽0.86, p⬍0.001),
meaning that, all waves considered, NREM segments with large
frontal slow waves also tend to contain large occipital slow waves.
Next, we correlated the amplitude of the large slow waves (above
the 75th amplitude percentile) in the frontal region of interest
with the amplitude of slow waves in the occipital region of inter-
est (without an amplitude threshold). Here a significant negative
correlation emerged (Spearman rank correlation coefficient r⫽
⫺0.19, p⫽0.002). It therefore appears that the larger the subset
of high-amplitude slow waves in the frontal cortex, the smaller
the slow waves in the occipital cortex.
These results suggest that there are two distinct populations of
slow waves that differentially relate to dream reports and whose
amplitudes are anticorrelated (Fig. 7). To achieve a more accurate
and topography-independent classification of the two types of
slow waves, for further analyses we used a previously described
approach (Bernardi et al., 2018) that divides slow waves into
putative type I slow waves (minority of large and steep slow waves
with frontocentral involvement) and type II slow waves (majority
of smaller, shallower, diffusely distributed slow waves) based on a
synchronization index taking into account their amplitude and
slope (for details, see Materials and Methods). This method is
based on a single temporal reference representing all sources
(“negative envelope” of the signal), and can thus classify slow
waves regardless of their origin and scalp distribution. Because
these analyses are mainly aimed at investigating how distinctive
properties of type I slow waves relate to dream experiences, a
stringent arbitrary threshold was applied to separate type I from
potentially large type II slow waves. More specifically, only slow
waves with a synchronization index ⬎4 standard MADs from the
median of all slow waves were considered type I slow waves, while
the remainder of slow waves were considered type II slow waves.
The large majority of slow waves classified as type I according to
this method (95.0 ⫾4.5%, mean across all channels) had an
amplitude above the 75th percentile (i.e., were classified as
“large” based on the amplitude criterion used in previously de-
scribed analyses (Fig. 8A). In contrast, only about half of type II
slow waves (52.6 ⫾4.1%) had a large amplitude.
We then computed the ratio between type I and type II slow
waves for each NREM period preceding awakenings, and found
that it was significantly different among DE, DEWR, and NE
(one-way repeated-measures ANOVA, F
(2,12)
⫽5.795, p⫽
0.008), with DE (0.10 ⫾0.01) having a significantly higher ratio
compared with DEWR (0.05 ⫾0.01, p⫽0.04, paired two-tailed
ttest) and NE (0.04 ⫾0.01, p⫽0.02), while the latter two cate-
gories did not differ significantly (p⫽0.4; Fig. 8B).
Next, we explored the direction of the negative association we
observed between the amplitude of large frontal slow waves and
occipital slow waves (Fig. 7). More specifically, we wanted to
know whether type I slow waves were more likely to occur on an
EEG background of small slow waves or whether they induced
reductions of slow-wave amplitude, particularly in posterior ar-
eas. To this aim, we first examined the correlation between SWA
in the 6 s preceding each slow wave and the synchronization
index of that slow wave. This analysis revealed that type I slow
waves were more likely to be large and steep (i.e., to have a higher
Figure 7. Correlation between slow-wave amplitude in a frontal and occipital region of interest for two representative subjects. Top row, All slow waves (no amplitude threshold). Bottom row,
An amplitude threshold (above the 75th percentile of all waves for each subject) was applied to frontal slow waves.
9180 •J. Neurosci., October 24, 2018 •38(43):9175–9185 Siclari et al. •Dreaming in NREM Sleep

synchronization index) if they occurred on a background of low
SWA, while for type II slow waves the opposite effect was true
(Fig. 8C). We then examined changes in spectral power induced
by type I and type II slow waves (in the 6 s following the slow
wave). We found that, on average, type I slow waves induced
increases in spectral power in all frequency bands, with a small
peak in the 10 –12 Hz range and a larger peak in the 18 –30 Hz
range (Fig. 8D). A source-modeling analysis revealed that the
18 –30 Hz increases following type I slow waves peaked in fron-
tocentral regions including the medial primary motor cortex and
the mid-cingulate cortex (top 5% of voxels; Fig. 8E). EEG exam-
ples of high-frequency increases following type I slow waves are
shown in Figure 9. For type II slow waves, these changes were less
pronounced and peaked in the fast spindle range (⬃13–14 Hz).
Type I slow waves induced significantly larger power increases in
the 0 –2 Hz and 18 – 40 Hz frequency ranges compared with type
Figure 8. Type I and type II slow-wave characteristics. A, Proportion of slow waves classified as type I (left) or type II (right) based on synchronization efficiency that were also defined as large
based on the channel-by-channel amplitude criterion (⬎75th percentile; for comparison, slow waves were identified in a 250 ms time window centered on the peak of the type I/II slow wave). Each
barinthe graph represents the average(%)across subjects ⫾SD. B,TypeI/typeII ratio for the60s preceding DE, DEWR, andNE(mean and SEM). Asterisksindicatestatisticallysignificant differences
at p⬍0.05 (paired two-tailed ttests). C, Correlation between low-frequency spectral power in the 1– 4 Hz range (in the 6 s preceding the slow wave) and the synchronization index (reflecting
slow-wave amplitude and slope) for type I and type II slow waves (Spearman rank correlation coefficient). Type I and type II slow waves were separated according to their synchronization index (for
details, see Materialsand Methods). Electrodes within a cluster showing astatistically significant effect are marked in white ( p⬍0.05,cluster-basedcorrection for multiplecomparisons).D, Spectral
power changes (%) induced by type I and type II slow waves in a frontal electrode cluster (Fz and immediately neighboring electrodes). Spectral power in the 6 s following the slow wave was
compared with spectral power in the 6 s preceding each slow wave for different frequency bands (1– 40 Hz; resolution of 2 Hz bins). Bottom row, pValues for the comparison between PSD changes
induced by type I and type II slow waves (paired two-tailed ttests). E, Cortical distribution of spectral power changes in the 18 –30 Hz for type I slow waves at the source level.
Figure 9. Representative examples of high-frequency increases following type I slow waves. Time 0 corresponds to the maximum negative peak of the type I slow wave. Four representative
midline channels are displayed [Fz (frontal), Cz (central), Pz (parietal), and Oz (occipital)], referenced to the average of two mastoid channels.
Siclari et al. •Dreaming in NREM Sleep J. Neurosci., October 24, 2018 •38(43):9175–9185 • 9181

II slow waves (Fig. 8D). Importantly, overall, neither type I nor
type II slow waves induced decreases in SWA. These results show
that type I slow waves tend to occur when SWA is low, but do not
appear to consistently induce reductions in SWA. Instead, they
are often followed by high-frequency power increases that look
like “microarousals” (Fig. 9) according to sleep-scoring criteria
(Iber et al., 2007).
We then asked whether the high-frequency increases follow-
ing type I slow waves differed among DE, DEWR, and NE. We
found that type I slow waves induced significantly stronger in-
creases in high-frequency activity (18 –30 Hz) in central regions
(with a right-sided lateralization) when subjects could recall the
content of the dream (DE) as opposed to when they could not
(DEWR; Fig. 10).
Spindles and dreaming
Finally, we examined how sleep spindles relate to dreaming. We
found that spindles preceding DE, compared with NE, were more
numerous and had a higher frequency (Fig. 11). There were sig-
nificantly more fast spindles preceding reports of DE, and these
differences were relatively diffuse. However, a contrast between
DEWR and NE showed that dreaming per se (regardless of the
ability to the recall the content of the dream), was associated with
significantly more fast spindles and fewer slow spindles in a cen-
tral posterior region of the brain. The DE versus DEWR contrast
did not show any significant differences (data not shown).
Discussion
Slow-wave characteristics and dreaming
The EEG slow wave of sleep occurs when thalamocortical neu-
rons become bistable and start to oscillate between two states,
each lasting a few hundred milliseconds: a hyperpolarized
“down-state,” during which neurons are silent (off-period), and
a depolarized “up-state” (on-period) characterized by neuronal
firing (Steriade et al., 2001;Nir et al., 2011). In the present study,
we were able to link specific slow-wave characteristics to dream
experiences. We found that dreaming, as opposed to no experi-
ences (unconsciousness), was associated with fewer slow waves,
which were smaller and shallower, and had a tendency to display
a larger number of negative peaks. Large-scale computer simula-
tions (Esser et al., 2007), local field potential studies in rodents
(Vyazovskiy et al., 2007), and EEG studies in humans (Riedner et
al., 2007) have revealed that the amplitude of slow waves reflects
the number of neuronal populations that simultaneously enter a
down-state and are thus in phase with each other, while the first
and second slopes of the slow waves are a measure of the speed
with which neuronal populations transition into the down-state
and up-state, respectively. Finally, the number of negative peaks
of a slow wave is related to the spatial synchrony of slow-wave
generation, with multipeak slow waves resulting from an asyn-
chronous generation of slow waves in distant cortical regions.
Our findings thus indicate that reports of unconsciousness are
more likely when large neuronal populations simultaneously and
rapidly enter a down-state that is synchronized across central
posterior cortical regions. Several studies have suggested that the
off-period associated with slow waves represents the fundamen-
tal mechanism by which consciousness is lost during sleep. It has
been shown that the slow wave-like response induced by TMS
during sleep leads to a breakdown of cortical effective connectiv-
ity among specialized thalamocortical regions (Massimini et al.,
2005,2010), thereby impairing information integration, which
has been proposed as a prerequisite for consciousness (Tononi,
2008). The neuronal off-period associated with the slow wave-
like response to cortical stimulation leads to an interruption of
deterministic interactions, indicated by the loss of causal effects
between the stimulation and brain activity resuming after the
off-period (measured by a phase locking index; Pigorini et al.,
2015). Our results are also consistent with those of a previous
study (Nieminen et al., 2016) showing that within NREM sleep,
TMS applied to posterior cortical regions evokes a larger negative
deflection and a shorter phase-locked response when subjects
reported unconsciousness compared with when they reported
dreaming.
Topographical distribution of slow waves and dreaming
Differences in slow-wave characteristics between dreaming and
unconsciousness, especially amplitude and slope, were not uni-
formly distributed across the cortical surface, but were most con-
sistent in central and posterior regions. This is in line with our
recent study, in which we showed that dream reports were pre-
ceded by lower SWA in a posterior “hot zone” of the brain, com-
prising the medial and lateral occipital lobe and extending
superiorly to the precuneus and posterior cingulate gyrus (Siclari
et al., 2017). Several imaging modalities converge to indicate that
the posterior cingulate gyrus is one of the most consistent cere-
bral “hubs,” connecting many different brain areas (de Pasquale
et al., 2018). The occurrence of off-states in this region may thus
lead to a breakdown of consciousness because it may disrupt
connectivity with many relatively distant areas of the brain. These
results are also in line with reports of cessation of dreaming after
lesions of the inferior parietal and occipital cortex (Wilbrand,
1892;Murri et al., 1985;Solms, 1997;Bischof and Bassetti, 2004),
and with the observation that in the course of development,
dreaming appears to be closely related with the maturation of
visuospatial skills, which depend on posterior (parietal) cortical
areas (Foulkes, 1999). In the frontal cortex on the other hand,
slow waves were less consistently associated with unconscious-
ness than in central and posterior brain regions. When consider-
ing high-amplitude slow waves, we even found that the opposite
was true: these high-amplitude slow waves were larger and
steeper in frontal regions when they preceded reports of dream
experiences.
Figure 10. Topographical distribution of tvalues for the contrast in 18 –30 Hz changes
induced by type I slow waves between dream experiences (DEs), no experiences (NEs) (left), DE
and dream experiences without recall of content (DEWR) (middle), and DEWR and NE (right),
averaged over the last 60 s before the awakening. Electrodes within a cluster showing a statis-
tically significant effect are marked in white ( p⬍0.05, cluster-based correction for multiple
comparisons, two-tailed paired ttests: n⫽12 subjects for DE; n⫽11 subjects for NE; n⫽13
subjects for DEWR).
9182 •J. Neurosci., October 24, 2018 •38(43):9175–9185 Siclari et al. •Dreaming in NREM Sleep

Type I and type II slow waves
We recently obtained evidence for two types of synchronization
processes underlying slow waves in the transition to sleep: a pre-
sumably subcorticocortical synchronization process giving rise
to isolated, very large and steep slow waves with a frontocentral
involvement (type I slow waves), and a presumably corticocorti-
cal synchronization process underlying smaller slow waves, with
more variable cortical origins (type II slow waves; Siclari et al.,
2014). These two processes are temporally dissociated in the fall-
ing asleep period in adults, but are also present during stable sleep
(Bernardi et al., 2018). The fact that reports of dreaming are
particularly likely when large, steep, and frontal type I slow waves
occur could be explained by several findings. First, type I slow
waves are more likely to be large and steep when they occur on an
EEG background of low SWA (Fig. 8C). Therefore, the presence
of frontal type I slow waves may indicate that most cortical re-
gions, and in particular posterior areas, are in a state of relatively
low bistability and are more likely to contribute to conscious
experiences. Second, several lines of evidence suggest that type I
slow waves may be related to arousal systems. Source-modeling
analyses have revealed that they originate preferentially in senso-
rimotor regions and in the posterior–medial parietal cortex, and
that they involve primarily frontomedial regions (Siclari et al.,
2014). The regions of origin are the brain areas with the highest
noradrenergic innervation in the human and monkey cortex
(Gaspar et al., 1989;Javoy-Agid et al., 1989;Lewis and Morrison,
1989). Cortical projections preferentially targeting the frontal
medial area, consistent with the medial frontal involvement, have
been described for the pontine and mesencephalic reticular for-
mation (Jones and Yang, 1985), another
arousal-promoting structure, and the
ventromedial thalamic nucleus (Desbois
and Villanueva, 2001;Kuramoto et al.,
2015). Very large-amplitude slow waves
are specifically associated with fMRI
brainstem activations including the locus
ceruleus (Dang-Vu et al., 2008), and it is
well known that K-complexes, which
share many properties with type I slow
waves (Siclari et al., 2014), can be induced
by peripheral sensory stimulations. Fi-
nally, type I slow waves tend to be fol-
lowed by increases in high-frequency
spectral power that peak in frontomedial
regions (Fig. 8E), often fulfill the criteria
for microarousals (Iber et al., 2007;Fig. 9),
and herald the successful recall of dream
content (Fig. 10). The topographic distri-
bution of these high-frequency increases
(Fig. 8D) closely matches the distribution
of high-frequency activity that has previ-
ously been shown to distinguish dreams
with and without recalled dream content
(Siclari et al., 2017). Interestingly and in
line with these findings, intrasleep awak-
enings are higher in frequent dream re-
callers compared with infrequent dream
recallers (Ruby et al., 2013;Eichenlaub et
al., 2014). Increases in high-frequency
power may thus reflect the intermittent
activation of arousal systems and in par-
ticular surges in noradrenergic activity oc-
curring in NREM sleep (Aston-Jones and
Bloom, 1981). Notably, noradrenaline favors memory encoding
and consolidation in both wakefulness (Roozendaal and Mc-
Gaugh, 2011) and sleep (Gais et al., 2011).
Spindles and dreaming
Our findings suggest that dreaming is more likely to occur in the
presence of fast spindles in a central and posterior cortical region,
while reports of no experience preferentially occur in the pres-
ence of slow spindles in the same areas. These findings are con-
sistent with the recent observation that subjects who reported a
high number of dreams had faster spindles (Nielsen et al., 2016).
Spindle frequency has been related to the level of thalamic hyper-
polarization and typically decreases in the course of the falling
asleep period (Himanen et al., 2002;Andrillon et al., 2011;Siclari
et al., 2014), along with increasing levels of thalamic hyperpolar-
ization (Andrillon et al., 2011) and SWA. One possibility is that
these findings indirectly reflect differences in slow-wave charac-
teristics. Spindles occurring during the positive-to-negative de-
flection of the slow wave (progressive cortical hyperpolarization)
tend to be slower than spindles occurring during the negative-to-
positive phase of the EEG slow wave (transition to depolarized
cortical up-state; Molle et al., 2011;Siclari et al., 2014). It is thus
conceivable that when slow waves are sparse and shallow and are
associated with longer up-states, fast spindles are more likely to
occur. The fact that both slow-wave and spindle differences were
most consistent over the same central posterior brain areas would
support such an interpretation.
Figure 11. Top row, Topographical distribution of tvalues for the contrast between dream experiences and no experiences for
different spindle parameters (60 s before awakening). Electrodes within a cluster showing a statistically significant effect are
marked in white ( p⬍0.05, cluster-based correction for multiple comparisons, two-tailed paired ttests; n⫽12 subjects) are
marked in white. Bottom row, Same as the top row for dream experiences without recall of content vs no experience.
Siclari et al. •Dreaming in NREM Sleep J. Neurosci., October 24, 2018 •38(43):9175–9185 • 9183

Limitations
It should be noted that our distinction between type I and type II
slow waves, although supported by different observations, is
based on an arbitrary cutoff, as there is currently no better way to
separate the two types of slow waves in stable sleep. Therefore,
future studies using techniques allowing direct imaging of
arousal-related subcortical structures will be necessary to validate
this separation and the results.
Summary and conclusions
Here we show that dreaming in NREM sleep is determined by the
proportion and distribution of two distinct types of slow waves.
Dream experiences are more likely to occur when the majority of
slow waves (type II slow waves) are small, sparse, and shallow,
especially in posterior brain regions. The content of dream expe-
riences is more likely to be reported if, in addition, high-
amplitude slow waves (type I slow waves) occur in frontocentral
brain regions and are followed by high-frequency increases (local
microarousals). These results suggest that the capacity of the
brain to generate experiences during sleep is reduced in the pres-
ence of large type II slow waves in posterior and central brain
regions, and that dream recall may be facilitated by the intermit-
tent activation of arousal systems during NREM sleep (associated
with type I slow waves).
References
Anderer P, Klosch G, Gruber G, Trenker E, Pascual-Marqui RD, Zeitlhofer J,
Barbanoj MJ, 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. CrossRef Medline
Andrillon T, Nir Y, Staba RJ, Ferrarelli F, Cirelli C, Tononi G, Fried I (2011)
Sleep spindles in humans: insights from intracranial EEG and unit record-
ings. J Neurosci 31:17821–17834. CrossRef Medline
Antrobus J, Kondo T, Reinsel R, Fein G (1995) Dreaming in the late morn-
ing: summation of REM and diurnal cortical activation. Conscious Cogn
4:275–299. CrossRef Medline
Aserinsky E, Kleitman N (1955) Two types of ocular motility occuring dur-
ing sleep. J Appl Physiol 8:11–18. CrossRef Medline
Aston-Jones G, Bloom FE (1981) Activity of norepinephrine-containing lo-
cus ceruleus neurons in behaving rats anticipates fluctuations in the sleep-
waking cycle. J Neurosci 1:876–886. CrossRef Medline
Bernardi G, Cecchetti L, Siclari F, Buchmann A, Yu X, Handjaras G, Bellesi M,
Ricciardi E, Kecskemeti SR, Riedner BA, Alexander AL, Benca RM,
Ghilardi MF, Pietrini P, Cirelli C, Tononi G (2016) Sleep reverts changes
in human gray and white matter caused by wake-dependent training.
Neuroimage 129:367–377. CrossRef Medline
Bernardi G, Siclari F, Handjaras G, Riedner BA, Tononi G (2018) Local and
widespread slow waves in stable NREM sleep: evidence for distinct regu-
lation mechanisms. Front Hum Neurosci 12:248. CrossRef Medline
Bischof M, Bassetti CL (2004) Total dream loss: a distinct neuropsycholog-
ical dysfunction after bilateral PCA stroke. Ann Neurol 56:583–586.
CrossRef Medline
Chellappa SL, Frey S, Knoblauch V, Cajochen C (2011) Cortical activation
patterns herald successful dream recall after NREM and REM sleep. Biol
Psychol 87:251–256. CrossRef Medline
Dang-Vu TT, Schabus M, Desseilles M, Albouy G, Boly M, Darsaud A, Gais S,
Rauchs G, Sterpenich V, Vandewalle G, Carrier J, Moonen G, Balteau E,
Degueldre C, Luxen A, Phillips C, Maquet P (2008) Spontaneous neural
activity during human slow wave sleep. Proc Natl Acad Sci U S A 105:
15160–15165. CrossRef Medline
Delorme A, Makeig S (2004) EEGLAB: an open source toolbox for analysis
of single-trial EEG dynamics including independent component analysis.
J Neurosci Methods 134:9–21. CrossRef Medline
de Pasquale F, Corbetta M, Betti V, Della Penna S (2018) Cortical cores in
network dynamics. Neuroimage 180:370–382. CrossRef Medline
Desbois C, Villanueva L (2001) The organization of lateral ventromedial
thalamic connections in the rat: a link for the distribution of nociceptive
signals to widespread cortical regions. Neuroscience 102:885–898.
CrossRef Medline
Eichenlaub JB, Bertrand O, Morlet D, Ruby P (2014) Brain reactivity differ-
entiates subjects with high and low dream recall frequencies during both
sleep and wakefulness. Cereb Cortex 24:1206–1215. CrossRef Medline
Esposito MJ, Nielsen TA, Paquette T (2004) Reduced alpha power associ-
ated with the recall of mentation from stage 2 and stage REM sleep.
Psychophysiology 41:288 –297. CrossRef Medline
Esser SK, Hill SL, Tononi G (2007) Sleep homeostasis and cortical synchro-
nization: I. Modeling the effects of synaptic strength on sleep slow waves.
Sleep 30:1617–1630. CrossRef Medline
Ferrarelli F, Huber R, Peterson MJ, Massimini M, Murphy M, Riedner BA,
Watson A, Bria P, Tononi G (2007) Reduced sleep spindle activity in
schizophrenia patients. Am J Psychiatry 164:483–492. CrossRef Medline
Foulkes D (1999) Children’s dreaming and the development of conscious-
ness. Cambridge, MA: Harvard UP.
Funk CM, Honjoh S, Rodriguez AV, Cirelli C, Tononi G (2016) Local slow
waves in superficial layers of primary cortical areas during REM sleep.
Curr Biol 26:396–403. CrossRef Medline
Gais S, Rasch B, Dahmen JC, Sara S, Born J (2011) The memory function of
noradrenergic activity in non-REM sleep. J Cogn Neurosci 23:2582–2592.
CrossRef Medline
Gaspar P, Berger B, Febvret A, Vigny A, Henry JP (1989) Catecholamine
innervation of the human cerebral cortex as revealed by comparative
immunohistochemistry of tyrosine hydroxylase and dopamine-beta-
hydroxylase. J Comp Neurol 279:249–271. CrossRef Medline
Himanen SL, Virkkala J, Huhtala H, Hasan J (2002) Spindle frequencies in
sleep EEG show U-shape within first four NREM sleep episodes. J Sleep
Res 11:35– 42. CrossRef Medline
Huber R, Ghilardi MF, Massimini M, Tononi G (2004) Local sleep and
learning. Nature 430:78–81. CrossRef Medline
Iber C, Ancoli-Israel S, Chesson A, Quan SF (2007) The AASM manual for
the scoring of sleep and associated events: rules, terminology and techni-
cal specifications, Ed 1. Westchester, IL: American Academy of Sleep
Medicine.
Javoy-Agid F, Scatton B, Ruberg M, L’Heureux R, Cervera P, Raisman R,
Maloteaux JM, Beck H, Agid Y (1989) Distribution of monoaminergic,
cholinergic, and GABAergic markers in the human cerebral cortex. Neu-
roscience 29:251–259. CrossRef Medline
Jones BE, Yang TZ (1985) The efferent projections from the reticular for-
mation and the locus coeruleus studied by anterograde and retrograde
axonal transport in the rat. J Comp Neurol 242:56 –92. CrossRef Medline
Jung TP, Makeig S, Humphries C, Lee TW, McKeown MJ, Iragui V, Sejnowski
TJ (2000) Removing electroencephalographic artifacts by blind source
separation. Psychophysiology 37:163–178. CrossRef Medline
Kuramoto E, Ohno S, Furuta T, Unzai T, Tanaka YR, Hioki H, Kaneko T
(2015) Ventral medial nucleus neurons send thalamocortical afferents
more widely and more preferentially to layer 1 than neurons of the ventral
anterior-ventral lateral nuclear complex in the rat. Cereb Cortex 25:221–
235. CrossRef Medline
Lewis DA, Morrison JH (1989) Noradrenergic innervation of monkey pre-
frontal cortex: a dopamine-beta-hydroxylase immunohistochemical
study. J Comp Neurol 282:317–330. CrossRef Medline
Marzano C, Ferrara M, Mauro F, Moroni F, Gorgoni M, Tempesta D, Cipolli
C, De Gennaro L (2011) Recalling and forgetting dreams: theta and al-
pha oscillations during sleep predict subsequent dream recall. J Neurosci
31:6674–6683. CrossRef Medline
Massimini M, Ferrarelli F, Huber R, Esser SK, Singh H, Tononi G (2005)
Breakdown of cortical effective connectivity during sleep. Science 309:
2228–2232. CrossRef Medline
Massimini M, Ferrarelli F, Murphy M, Huber R, Riedner B, Casarotto S,
Tononi G (2010) Cortical reactivity and effective connectivity during
REM sleep in humans. Cogn Neurosci 1:176–183. CrossRef Medline
Molle M, Bergmann TO, Marshall L, Born J (2011) Fast and slow spindles
during the sleep slow oscillation: disparate coalescence and engagement
in memory processing. Sleep 34:1411–1421. CrossRef Medline
Monroe LJ, Rechtschaffen A, Foulkes D, Jensen J (1965) Discriminability of
REM and NREM reports. J Pers Soc Psychol 2:456–460. Medline
Murri L, Massetani R, Siciliano G, Giovanditti L, Arena R (1985) Dream
recall after sleep interruption in brain-injured patients. Sleep 8:356–362.
CrossRef Medline
Nichols TE, Holmes AP (2002) Nonparametric permutation tests for func-
9184 •J. Neurosci., October 24, 2018 •38(43):9175–9185 Siclari et al. •Dreaming in NREM Sleep

tional neuroimaging: a primer with examples. Hum Brain Mapp 15:1–25.
CrossRef Medline
Nielsen T, Carr M, Blanchette-Carrie`re C, Marquis L, Dumel G, Solomonova E,
Julien S, Picard-Deland C, Paquette T (2016) NREM sleep spindles are associ-
ated with dream recall. Sleep Spindles Cortical Up States 1:27– 41. CrossRef
Nieminen JO, Gosseries O, Massimini M, Saad E, Sheldon AD, Boly M, Siclari
F, Postle BR, Tononi G (2016) Consciousness and cortical responsive-
ness: a within-state study during non-rapid eye movement sleep. Sci Rep
6:30932. CrossRef Medline
Nir Y, Staba RJ, Andrillon T, Vyazovskiy VV, Cirelli C, Fried I, Tononi G
(2011) Regional slow waves and spindles in human sleep. Neuron 70:
153–169. CrossRef Medline
Oudiette D, Dealberto MJ, Uguccioni G, Golmard JL, Merino-Andreu M,
Tafti M, Garma L, Schwartz S, Arnulf I (2012) Dreaming without REM
sleep. Conscious Cogn 21:1129–1140. CrossRef Medline
Pigorini A, Sarasso S, Proserpio P, Szymanski C, Arnulfo G, Casarotto S,
Fecchio M, Rosanova M, Mariotti M, Lo Russo G, Palva JM, Nobili L,
Massimini M (2015) Bistability breaks-off deterministic responses to
intracortical stimulation during non-REM sleep. Neuroimage 112:105–
113. CrossRef Medline
Riedner BA, Vyazovskiy VV, Huber R, Massimini M, Esser S, Murphy M,
Tononi G (2007) Sleep homeostasis and cortical synchronization: III. A
high-density EEG study of sleep slow waves in humans. Sleep 30:1643–
1657. CrossRef Medline
Roozendaal B, McGaugh JL (2011) Memory modulation. Behav Neurosci
125:797–824. CrossRef Medline
Ruby P, Blochet C, Eichenlaub JB, Bertrand O, Morlet D, Bidet-Caulet A
(2013) Alpha reactivity to first names differs in subjects with high and
low dream recall frequency. Front Psychol 4:419. CrossRef Medline
Siclari F, Larocque JJ, Postle BR, Tononi G (2013) Assessing sleep con-
sciousness within subjects using a serial awakening paradigm. Front Psy-
chol 4:542. CrossRef Medline
Siclari F, Bernardi G, Riedner B, LaRocque J, Benca MR, Tononi G (2014)
Two distinct synchronization processes in the transition to sleep: a high-
density electroencephalographic study. Sleep 37:1621–1637. CrossRef
Medline
Siclari F, Baird B, Perogamvros L, Bernardi G, LaRocque JJ, Riedner B, Boly
M, Postle BR, Tononi G (2017) The neural correlates of dreaming. Nat
Neurosci 20:872– 878. CrossRef Medline
Solms M (1997) The neuropsychology of dreams: a clinico-anatomical
study. Mahwah, NJ: Erlbaum.
Solms M (2000) Dreaming and REM sleep are controlled by different brain
mechanisms. Behav Brain Sci 23:843–850. CrossRef Medline
Steriade M, Timofeev I, Grenier F (2001) Natural waking and sleep states: a
view from inside neocortical neurons. J Neurophysiol 85:1969–1985.
CrossRef Medline
Stickgold R, Malia A, Fosse R, Propper R, Hobson JA (2001) Brain-mind
states: I. Longitudinal field study of sleep/wake factors influencing men-
tation report length. Sleep 24:171–179. CrossRef Medline
Tononi G (2008) Consciousness as integrated information: a provisional
manifesto. Biol Bull 215:216–242. CrossRef Medline
Vyazovskiy VV, Riedner BA, Cirelli C, Tononi G (2007) Sleep homeostasis
and cortical synchronization: II. A local field potential study of sleep slow
waves in the rat. Sleep 30:1631–1642. CrossRef
Wilbrand H (1892) Ein Fall von Seelenblindheit und Hemianopsie mit
Sektionsbefund. Deutsche Zeitschrift fu¨r Nervenheilkunde 2:361–387.
Williamson PC, Csima A, Galin H, Mamelak M (1986) Spectral EEG corre-
lates of dream recall. Biol Psychiatry 21:717–723. CrossRef Medline
Siclari et al. •Dreaming in NREM Sleep J. Neurosci., October 24, 2018 •38(43):9175–9185 • 9185