HHS Public Access

Author manuscript
Nat Neurosci. Author manuscript; available in PMC 2017 October 10.
Author Manuscript

Published in final edited form as:

Nat Neurosci. 2017 June ; 20(6): 872–878. doi:10.1038/nn.4545.

The neural correlates of dreaming

F. Siclari1,2,†, B. Baird1,†, L. Perogamvros1,3,4,†, G. Bernardi1,2,5, J. J. LaRocque7, B.
Riedner1, M. Boly1,8, B. R. Postle1, and G. Tononi1,*

1Department of Psychiatry, University of Wisconsin - Madison, Madison, Wisconsin 53519, USA

2Center for Research and Investigation in Sleep (CIRS), Lausanne University Hospital (CHUV)
and University of Lausanne, Lausanne, Switzerland 3Department of Psychiatry, Geneva
Author Manuscript

University Hospitals, Geneva, Switzerland 4Department of Neuroscience, University of Geneva,

Switzerland 5IMT School for Advanced Studies, Lucca, Italy 6Medical Scientist Training Program
and Neuroscience Training Program, University of Wisconsin, Madison, Wisconsin 53519, USA
7Department of Neurology, University of Wisconsin- Madison, Madison, Wisconsin 53519, USA

8Department of Psychology, University of Wisconsin- Madison, Madison, Wisconsin 53519, USA

Abstract
Consciousness never fades during wake. However, if awakened from sleep, sometimes we report
dreams and sometimes no experiences. Traditionally, dreaming has been identified with REM
sleep, characterized by a wake-like, globally ‘activated’, high-frequency EEG. However, dreaming
also occurs in NREM sleep, characterized by prominent low-frequency activity. This challenges
Author Manuscript

our understanding of the neural correlates of conscious experiences in sleep. Using high-density
EEG, we contrasted the presence and absence of dreaming within NREM and REM sleep. In both
NREM and REM sleep, reports of dream experience were associated with a local decrease in low-
frequency activity in posterior cortical regions. High-frequency activity within these regions
correlated with specific dream contents. Monitoring this posterior ‘hot zone’ predicted whether an
individual reported dreaming or the absence of experiences during NREM sleep in real time,
suggesting that it may constitute a core correlate of conscious experiences in sleep.

Keywords
consciousness; sleep; high-density EEG; source modeling; dreams
Author Manuscript

Users may view, print, copy, and download text and data-mine the content in such documents, for the purposes of academic research,
subject always to the full Conditions of use: http://www.nature.com/authors/editorial_policies/license.html#terms
Correspondence: Giulio Tononi, MD PhD, Department of Psychiatry, University of Wisconsin - Madison 6001 Research Park Blvd,
53719 Madison, Wisconsin, USA, [email protected].
†These authors contributed equally to this work.
Supplementary material is linked to the online version of the paper at www.nature.com/nature.
Author Contributions. FS, LP, BB, JL, MB, BR, BP and GT designed the experiments, FS, LP, BB and JL conducted the
experiments, FS, LP, BB and GB analyzed the data, FS, LP, BB and GT wrote the paper.
GT and BR are involved in a research study in humans supported by Philips Respironics. This study is not related to the work
presented in the current manuscript.
The other authors have indicated no financial conflicts of interest.
 Siclari et al. Page 2

An ongoing stream of experiences accompanies every waking moment. Sleep is the only
Author Manuscript

time in which consciousness fades under normal physiological conditions: subjects

awakened from sleep, especially early in the night, report that they were not experiencing
anything up to 30% of the time1. At other times, subjects awakened from sleep report
dreams — a stream of vivid experiences that occur despite being immobile, unresponsive,
and largely disconnected from the environment. Thus, unlike wakefulness, sleep can be
associated with either the presence or absence of conscious experiences. In addition,
experiences in dreams can assume many forms, ranging from pure perceptual experiences to
pure thought, from simple images to temporally unfolding narratives, which are often similar
to awake conscious states but at times can be different in interesting ways2,3.

The discovery of rapid eye movement (REM) sleep — the ‘third state of being’ besides
wake and non-REM (NREM) sleep – led initially to a straightforward view of the neural
correlates of dreaming4: the wake-like, high-frequency, ‘activated’ EEG5,6 of REM sleep
Author Manuscript

was thought to be associated with the presence of dream experiences, and the low-frequency
activity of NREM sleep with the absence of dreaming. However, later studies showed that up
to 70% of NREM sleep awakenings yield reports of dream experiences1. Conversely, in a
small but consistent minority of cases, subjects deny having had any experience when
awakened from REM sleep. Thus, whether one experiences something or not during sleep
cannot be determined simply by assessing one’s behavioral state based on traditional EEG
features or neuroimaging correlates7,8. The paradoxical occurrence of both the presence and
absence of experiences within the same behavioral state of sleep, and across two very
different kinds of sleep (NREM and REM) challenges our current understanding of the
neural correlates of conscious experience in sleep.

Here we investigate the neural correlates of dreaming using a within state paradigm, in both
Author Manuscript

NREM and REM sleep, by performing serial awakenings of subjects recorded throughout
the night with high-density EEG (256 channels)9. The results highlight a posterior cortical
‘hot zone’ where a local decrease in low-frequency EEG activity during both NREM and
REM sleep is associated with reports of experiences upon awakening, whereas a local
increase in low-frequency activity is associated with the absence of experience. These results
hold for both a large group of naïve subjects and a smaller group of individuals trained in
dream reporting. We then show, in a separate group of subjects, that it is possible to predict
whether an individual will report having dreamt or not in the course of a NREM episode by
real-time EEG monitoring of this posterior hot zone. Finally, we show that the location of
high-frequency EEG activity during dreams correlates with specific dream contents, such as
thoughts, perceptions, faces, spatial setting, movement, and speech.
Author Manuscript

Results
Using a serial awakening paradigm9, participants were awakened throughout the night and
were asked to report whether, just before the awakening, they had been experiencing
anything (dreaming experience, DE), experiencing something but could not remember the
content (dreaming experience without recall of content, DEWR), or not experiencing
anything (no experience, NE, see Methods). If subjects reported a DE they were asked to
describe its most recent content (‘the last thing going through your mind prior to the alarm

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 3

sound’), and to rate it on a scale ranging from exclusively thought-like (thinking or

Author Manuscript

reasoning, with no sensory content) to exclusively perceptual (vivid sensory content, without
thinking or reasoning)10. Finally, they had to estimate the duration of the most recent DE,
and to report whether it contained specific categories of content, including faces, a spatial
setting, movement, and speech (examples of reports are presented in Supplementary table
S1).

We studied two different populations in separate experiments. In experiment 1 we studied a

large group of subjects who underwent few awakenings (N=32, 233 awakenings), while in
experiment 2, we investigated a smaller group of trained subjects who underwent many
awakenings (N=7, 815 awakenings) (see Supplementary table S2). Finally, in a third
experiment, we tested whether, based on the results of the first two experiments, we could
predict the presence or absence of dreaming in real time at the single trial level (N=7, 84
awakenings).
Author Manuscript

Dreaming experience vs no experience: low-frequency power

First, we sought to determine whether and how low-frequency activity (1–4 Hz) changes
between the presence and absence of conscious experience during sleep. EEG slow
frequencies between 1 and 4 Hz during sleep are associated with neuronal down-states and
bistability11, which prevent the emergence of stable causal interactions among cortical
areas12, and have been linked to the loss of consciousness13,14. For this analysis we focused
on the first group of subjects (experiment 1) because of the large number of subjects, and on
N2 sleep, since the number of DE and NE reports in this stage were well balanced. By
computing power spectral density at the source level within the 1–4 Hz frequency band we
found that reports of DE, compared to reports of NE, had decreased low-frequency power.
This decrease was restricted to a bilateral parieto-occipital region, encompassing the medial
Author Manuscript

and lateral occipital lobe and extending superiorly to the precuneus and posterior cingulate
gyrus (Fig. 1A). Experiment 2, in which awakenings were carried out both in N2 and N3
stage sleep, confirmed these results (Supplementary Figure 1).

We then compared instances in which the subjects reported that they had dreaming
experiences, but were not able to recall the content of the experience (DEWR), with
instances in which they reported being unconscious (i.e., having no experiences, NE).
Reports of DEWR, contrasted to reports of NE, were again preceded by decreased low-
frequency power in a parieto-occipital region highlighted by the DE/NE contrast (Fig. 1B,
see Supplementary Figure 2 for a conjunction map of the two contrasts). A contrast between
DE and DEWR in the low-frequency range did not yield significant differences. Thus,
decreased low-frequency power in this posterior cortical zone was associated with reports of
Author Manuscript

dreaming experiences, irrespective of the ability to remember the contents of the experience.

Next, we asked whether the contrast between DE and NE would produce similar results also
during REM sleep, a state that is substantially different from NREM sleep in terms of EEG
signatures (slow waves and spindles in NREM sleep, low-voltage fast activity in REM
sleep4), neural activity (widespread bistability between down and up-states in NREM sleep,
mostly tonic depolarization in REM sleep6), neuromodulation (high acetylcholine in REM
sleep, low in NREM sleep15), and regional activations7. Because of the rarity of NE reports

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 4

in REM sleep (Supplementary table S2), for this analysis we combined the data of
Author Manuscript

experiments 1 and 2 (N=10). Despite the notable physiological differences between NREM
and REM sleep states, we again found that brain activity associated with DE, as compared to
NE, had reduced power in the 1–4 Hz band in a parieto-occipital region (Fig. 2A), which
largely overlapped with the topography observed in NREM sleep (Fig. 2B). Thus, a decrease
in low-frequency power in a posterior cortical ‘hot zone’ is highlighted when individuals
reported experiencing a dream compared with reports of an absence of experience during
sleep, irrespective of the ability to recall the experience and of behavioral state (NREM vs.
REM sleep).

Dreaming experience vs no experience: high-frequency power

We then investigated differences in high-frequency band power (20–50 Hz), since high EEG
frequencies often reflect high rates of neuronal firing16,17 and may thus indicate brain
Author Manuscript

regions that show increased activity. In NREM sleep, DEs, compared to NE, were associated
with increased high-frequency power in the same parieto-occipital region that showed
reduced low frequency power. However, differences in high-frequency power extended
superiorly and anteriorly to parts of the lateral frontal cortex and the temporal lobes (Fig.
3A, see also Supplementary Figure 3 for better comparability with REM sleep (25–50 Hz
range)). DE with recall of content, compared to DEWR, were associated with higher high-
frequency power in medial and lateral frontoparietal areas (Fig. 3B). No differences in high-
frequency power were found when comparing DEWR with NE (Supplementary Figure 4).
Finally, in REM sleep DEs compared to NEs were associated with increased high-frequency
(25–50 Hz) power in frontal and temporal regions (Fig. 3C). Given the age variability in our
sample, an additional repeated measures ANCOVA was performed in the significant clusters
using this parameter as nuisance covariate. A significant main effect of report type (DE, NE)
Author Manuscript

was confirmed for both low-frequency (F(1,30)=9.39, p=0.005) and high-frequency activity
(F(1,30)=10.88, p=0.003), while no significant effects of age emerged (low-frequency:
F(1,30)=2.34, p = 0.14; high-frequency: F(1,30)=1.80, p = 0.19).

To test whether our results were characterized by a significant hemispheric lateralization, we

computed a repeated-measure-ANOVA with hemisphere and type of report (DE, DEWR,
NDE) as within-subject factors for all the contrasts (See ‘Online Methods’ and
Supplementary table 5). We did not find significant interactions between the type of report
and hemisphere, except for a barely significant right-sided lateralization (p= 0.046) for the
high-frequency power contrast between DE and NE in REM sleep.

Absolute power values for DE and NE in the posterior hot zone are shown in Supplementary
Figure 5. It should be noted that absolute low-frequency power values are higher in NREM
Author Manuscript

sleep compared to REM sleep, although differences are comparable and significant in both
sleep stages. It is likely that the absolute values in NREM sleep, although obtained from a
region of interest after source modeling the signal, are nevertheless influenced by high-
amplitude slow waves in other parts of the cortex. Indeed, source modeling strongly reduces
but does not completely eliminate the effects of volume conduction. Moreover,
regularization methods commonly applied to reduce the influence of noise in EEG data
introduce a ‘spatial smoothing’ of the EEG sources. Finally, the effect of slow wave

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 5

propagation, which occurs on a scale of tens/hundreds of milliseconds is not affected by

Author Manuscript

source modeling approaches and may contribute to the relatively high level of low-frequency
activity all over the cortical mantle and in brain areas that are not generating slow waves.

Dream content in REM sleep

We further investigated whether differences in high-frequency power could distinguish
specific contents of DEs. For these analyses, we focused on REM sleep awakenings in
trained subjects (experiment 2), because there were many intra-subject trials, reports were
more detailed9 and subjects were more confident about their answers concerning specific
contents.

Since dreams can differ greatly along the dimension ‘perception vs. thought’10, subjects had
been asked to rate their DEs accordingly. We found that high-frequency activity correlated
with the thinking dimension in frontal brain regions and with the perceiving dimension in
Author Manuscript

parietal, occipital and temporal areas (p<0.05 Fig. 4A). This anterior-posterior gradient
suggests that in dreams, just like in wakefulness, anterior cortical regions mediate thought-
like experiences while posterior regions are involved in perceptual aspects of the experience.

We also examined specific perceptual categories of dream reports. As shown in Fig. 4B, DEs
containing faces, compared to DEs without faces, were associated with increased high-
frequency activity in a circumscribed temporo-occipital region that closely matched the
fusiform face area (FFA). A region of interest analysis (see Supplementary Methods)
confirmed that high-frequency activity in the FFA on the right side was significantly higher
for DEs containing faces (p=0.023, paired one-tailed t-test) compared to DEs without faces,
consistent with the specialization of this area in the perception of faces during wakefulness.
Subjects were also asked to determine whether the setting of the most recent DE was
Author Manuscript

indoors, outdoors, or could not be specified. We found that DEs with a definite spatial
setting were associated with increased high-frequency activity in the right posterior parietal
cortex (rPPC) (Fig. 4C), an area involved in spatial perception and visuo-spatial attention,
whose lesion can cause spatial neglect. This finding was confirmed by a region of interest
analysis (See Supplementary Methods, p=0.023, paired one-tailed t-test). Furthermore, DEs
in which subjects reported the sense of moving in the dream were associated with increased
high-frequency activity in a region surrounding the right superior temporal sulcus (Fig. 4C),
an area involved in the perception of biological motion and viewing body movements. A
region of interest analysis confirmed this finding (See Supplementary Methods, p=0.029,
paired one-tailed t-test). Finally, as shown in Fig. 4C, DEs containing speech were
associated with increased high-frequency activity over a left posterior temporal region
corresponding to Wernicke’s area, as confirmed by a region of interest analysis (See
Author Manuscript

Supplementary Methods, p=0.048, paired one-tailed t-test). It should be noted that an SnPM
correction for multiple comparisons was not applied to whole-brain content-based analyses,
because content-specific areas were expected to be highly spatially-restricted. Instead,
region of interest analyses were used to confirm the specific hypotheses on the involvement
of distinct content-related areas.

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 6

Real-time prediction of dream experience in NREM sleep

Author Manuscript

If a decrease in low-frequency EEG power and an increase in high-frequency power in a

posterior hot zone represents a reliable neural correlate of dreaming, it should be possible to
predict whether a person is dreaming or not in real time by detecting ongoing local EEG
activations. Moreover, this should be the case even though the global EEG, as well as the
local EEG in anterior areas, is characterized by slow wave activity typical of NREM sleep.
To test this prediction, seven additional subjects slept for three consecutive nights in the
sleep laboratory while high-density EEG was continuously recorded. Based on the findings
of Experiments 1 and 2, on the experimental (second and third) nights participants were
awoken from NREM sleep when their neural activity surpassed a bispectral threshold in
low-frequency (0.5–4.5 Hz) and high-frequency (18–25 Hz) band power from an electrode
cluster over the posterior hot zone (Fig. 5A; see Methods; the 18–25 Hz range was used
because it captures EEG activations while avoiding artifacts from eye movements and other
Author Manuscript

sources during online EEG monitoring). In total, participants were awakened 84 times
(mean: 12.0 ± 3.4; Supplementary table S3). After removal of trials containing artifact (see
Methods), there remained 36 awakenings initiated by the algorithm with a DE prediction,
out of which 33 were accurate (DE mean accuracy = 91.6%, p=0.00001), and 26 awakenings
with a NE prediction, out of which 21 were accurate (NE mean accuracy = 80.7%,
p=0.0003); total prediction accuracy across all states was 87% (Fig. 5C). No interaction was
observed between DE prediction and time of night (F(1,6)=1.35, p=0.29). Confirming
significant differences in EEG power in DE compared to NE trials in the prediction ROI, we
observed significantly lower low-frequency activity (p=0.001) in DE (10.17±2.5 [mean ±
SD]) compared to NE (103.47±38.16) and significantly higher high-frequency activity
(p=0.001) in DE (0.14±0.03) compared to NE (0.07±0.02) prediction trials (Fig. 5B, see also
Supplementary Figure 6). DE had a higher high/low frequency band power ratio compared
Author Manuscript

to NE (Supplementary Figure 7), especially in cortical regions including bilateral occipital

cortex, precuneus, right superior parietal lobule (SPL), right precentral gyrus, left superior
and middle temporal sulcus (STS), right lingual gyrus and left inferior frontal gyrus (IFG)
(p<0.001, FDR cluster corrected) (Fig. 5D and Supplementary table S4).

Discussion
The three experiments reported here highlight a parieto-occipital ‘hot zone’ as the neural
correlate of dreaming. When this posterior hot zone showed a decrease in low-frequency
EEG activity – traditionally known as ‘EEG activation’5,6 - subjects reported upon
awakening that they did have dream experiences. By contrast, when low-frequency EEG
activity increased in the same area, subjects reported that they had been unconscious. This
result held true whether the subjects had been in NREM sleep or in REM sleep, irrespective
Author Manuscript

of the dominant EEG activity over other cortical regions.

The influential notion that dreaming is virtually synonymous with REM sleep has dominated
neuroimaging work for the past several decades7, despite the increasing number of studies
suggesting that dream reports could be obtained from every stage of NREM sleep, albeit less
often18. The frequent conflation of REM sleep with dreaming was eased by the marked
similarities between the overall EEG of REM sleep and that of waking consciousness, and

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 7

little attention was given to the rare reports of REM sleep awakenings that yielded no
Author Manuscript

reports. Conversely, reports of dreaming from NREM sleep awakenings were hard to
reconcile with EEG patterns characterized by slow wave activity – the opposite of EEG
activation5. Indeed, it was suggested that reports of dreaming after NREM awakenings were
due to ‘covert’ REM sleep19, or even that neural activity in the cortex was unrelated to the
presence or absence of consciousness20. The present results resolve these paradoxes by
showing that in both REM and NREM sleep, dreaming, and more generally, conscious
experience during sleep, requires a localized activation of a posterior hot zone, irrespective
of the EEG in the rest of the cortex. This explains why dreaming can occur in two behavioral
states that differ radically in terms of global EEG signatures.

Of course, since the posterior hot zone was highlighted by averaging EEG recordings
preceding many awakenings from many subjects, the full neural correlates of experiences in
sleep could occasionally include additional brain regions. For example, dreams that include
Author Manuscript

insight into the state (so-called lucid dreams) or dreams in which an individual has some
degree of control over the content of the dream may recruit additional areas of frontal-
parietal cortex21,22. Also, when considering unthresholded t-maps (Supplementary Figure
8), it is apparent that some cortical areas that do not reach significance in the contrast
between DE and NE are nevertheless consistently and smoothly highlighted in both
experiments 1 and 2. These areas include superior temporal and postcentral regions, as well
the midcingulate region, which is also highlighted by contrasts between highly thought-like
and highly perceptual experiences in REM sleep (Figure 4A). Conversely, we cannot rule
out that the posterior hot zone may include, besides brain regions supporting experiences,
areas that might support unconscious aspects of dream generation. However, many studies
contrasting perceiving or not specific sensory stimuli during wakefulness (such as seeing/not
seeing a face, an object, a place, a movement, and so on) also point to regions within the
Author Manuscript

posterior hot zone23. EEG source modeling indicates that, in addition to low and high-level
sensory areas (primarily visual), the posterior hot zone prominently includes precuneus,
posterior cingulate, and retrosplenial cortices: areas on the medial surface of the brain whose
electrical stimulation can induce a feeling of being ‘in a parallel world’24, disconnected from
the environment in a ‘dream-like state’25, and whose involvement in multisensory
integration is well-suited to support the virtual simulation of a world26 and the ‘immersive
spatio-temporal hallucinations’27 that characterize dreams.

The posterior hot zone was also highlighted by the contrast between the absence of
experiences and experiences without recall of specific contents of the experience. This
finding further suggests that the activation of this posterior hot zone is a marker of
experiences themselves, rather than of the recall of experiences. The trustworthiness of
Author Manuscript

subjective dream reports has often been called into question and has generated much
controversy in the past. Indeed, it is particularly delicate to distinguish the absence of
experience from forgotten experiences in this context. Therefore, to reduce potential biases
associated with collecting reports, we followed standardized experimental protocols.
Specifically, to avoid interfering with dream recall or introducing variability in dream recall,
subjects were always interviewed just after being awakened from a dream (short temporal
lag) and in identical experimental conditions (lying on their back, with eyes closed, without
the examiner entering the room). In addition, in experiment 2 subjects were trained in dream

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 8

reporting so as not to introduce a ‘learning bias’ in recalling dream experiences. Our results,
Author Manuscript

which show, for the first time, that reports of absent and forgotten dream experiences carry
distinct EEG signatures, should encourage the current tendency to take subjective reports of
conscious experience at face value, i.e. as the best inference to scientific explanation,
provided that carefully designed experimental protocols are used27. In line with our results,
both lesion studies and neuroimaging results indicate that specific contents of experiences
are supported by posterior cortical areas28, while a different set of brain areas are involved in
encoding and storing memories and in directing attention29. The results we obtained were
also virtually identical in naïve subjects (Experiment 1) and in trained subjects (Experiment
2), suggesting that they characterize a core correlate of dream experience, rather than some
other cognitive function.

In our experiments conscious experiences were consistently associated with low activity in
the delta range (1–4 Hz) and the absence of experience with high delta activity in the
Author Manuscript

posterior hot zone. During NREM sleep, EEG slow waves in the delta range reflect
alternations between periods of depolarization (up-states), during which neurons can fire,
and periods of hyperpolarization (down-states) during which neurons cease firing11. This
bistability of membrane potential causes the breakdown of stable causal interactions among
cortical areas12, which are thought to be necessary for sustaining conscious experiences13,14.
Recently, the occurrence of local slow waves has also been discovered during REM sleep,
especially in the supragranular layers of primary cortical areas30, adding to the evidence that
neuronal activity during sleep can have a local component11. A global slowing of EEG
activity into the delta range had been associated with loss of consciousness in many previous
studies13,14,31,32. However, there are several situations in which subjects can be conscious
and have experiences despite the presence of local and global slow wave activity23,33,34,
including of course NREM sleep itself, where global slow wave activity is present despite
Author Manuscript

the frequent occurrence of dreaming. The present findings suggest that future work in
patients with disorders of consciousness, epileptic seizures, or under general anesthesia
should examine the level of EEG activation in the posterior hot zone to uncover the possible
occurrence of consciousness despite the loss of responsiveness.

In addition to differences in EEG low-frequency activity in the posterior hot zone, we found
that this region also showed increased high-frequency activity when subjects reported
dreaming experiences during sleep. High-frequency activity usually reflects increased
neuronal firing16. A plausible interpretation of this finding is that, as long as the posterior
hot zone is not in a low-frequency mode that prevents experiences, the specific contents of
experiences when dreaming are dictated by which specific groups of neurons fire strongly
and which do not, just as in wake. Consistent with this, we found that specific contents of a
Author Manuscript

subject’s REM sleep dream – such as thoughts, perceptions, faces, places, movement, and
speech – are associated with increased high-frequency EEG activity in specific cortical
areas, which correspond closely to those engaged during waking perception of the same
contents. Although these differences extended only over a few voxels and were statistically
weaker than the contrasts between DE and NE, their temporal distribution (maximal in the
two seconds before the awakening), and spatial localization (similar to waking perception of
the same contents) strongly support our interpretation. In a recent fMRI study, a pattern
classifier trained while subjects perceived natural images during wakefulness35 was able to

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 9

successfully decode specific content categories reported by subjects awakened at the

Author Manuscript

transition between wakefulness and sleep. Our results with high-density EEG extend these
findings beyond falling asleep to periods of consolidated sleep, providing further evidence
that dream reports reflect conscious experience during sleep rather than confabulations upon
awakening36,37.

Increased high-frequency activity associated with reports of dreaming experiences were also
observed in areas that went beyond the posterior hot zone. Thus, increased high-frequency
activity extended to some portions of prefrontal cortex during REM sleep dreaming,
especially for thought-like experiences. Moreover, a frontal cluster (Fig 3B) showed increase
high-frequency activity during NREM sleep when contrasting experiences with and without
recall of content, suggesting that this area may be important for recall of the content of the
experience. Neuronal activity in the gamma range was initially linked to consciousness, but
later studies showed that consciousness and gamma activity can be dissociated23. For
Author Manuscript

example, one can be aware of a stimulus whether or not it triggers gamma activity38, and
gamma activity can be strong during general anesthesia in the presence of low frequency
oscillations39. However, recent studies have reported that gamma-band transcranial
alternating current stimulation is associated with higher scores on the revised coma scale
recovery scale (CRS-R)40 or self-awareness during REM sleep dreams41, suggesting that
high-frequency activity may indeed influence functions related to cognition or awareness. It
is thus an open question whether high- frequency activity in a subset of prefrontal regions
may contribute some specific contents to experience23, such as conscious thought or mind-
wandering states42, emotional/motivational contents43, metacognition44, or insight into one’s
state (i.e., lucid dreaming)21,22. Alternatively, high-frequency activity in prefrontal regions
may mediate cognitive functions that are unconscious but affect experience indirectly, such
as the encoding and storing of long-term memories, the deployment of attention, and the
Author Manuscript

planning and execution of tasks.

Finally, this study suggests that dreaming may constitute a valuable model for the study of
consciousness with implications beyond sleep. First, the direct comparison between
consciousness (the presence of experience) and unconsciousness (the absence of experience)
within the same behavioral state avoids confounds that limit the interpretation of between-
states paradigms. These confounds arise because, in studies contrasting wakefulness with
sleep, anesthesia, or coma, many other aspects of brain function besides consciousness
change along with behavioral state changes45. Second, during sleep subjects are largely
disconnected from the environment on both the input and output side, meaning that in the
majority of cases, sensory stimuli do not influence the content of experience2. Third, during
sleep subjects are not performing an experimental task, avoiding confounds caused by
Author Manuscript

stimulus processing, response preparation, and execution45.

An influential notion derived from neuroimaging studies contrasting wake with NREM
sleep, anesthesia, or coma is that the neural correlates of consciousness would correspond to
a broad fronto-parietal network46. This notion has been reinforced by findings indicating
that the same fronto-parietal network is activated when contrasting brain responses to the
same stimuli when subjects reported seeing them or not47. However, lesion studies suggest
that consciousness persists after widespread frontal lesions23,48 and recent work controlling

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 10

for confounding factors related to stimulus processing, attention, and task performance have
Author Manuscript

cast doubt on a central role of this fronto-parietal network in supporting

consciousness23,49,50. By exploiting a within state, no stimulus, no task paradigm, our
findings indicate that the neural correlate of consciousness, rather than involving the broad
lateral fronto-parietal network, is restricted to a posterior hot zone.

Methods
Experiments 1 and 2
Procedure—The procedure used in this study has been described in detail in a previous
publication9. Awakenings in the sleep laboratory were performed at pseudorandom intervals,
irrespective of sleep stage, using a computerized sound that lasted 1.5 sec, which was
administered through E-Prime (Psychology Software Tools, Pittsburgh, PA). Subjects were
instructed to signal that they had heard the alarm sound and to lie quietly on their back. They
Author Manuscript

then underwent a structured interview via intercom about their mental activity that lasted
between 20 sec and 3.5 min, depending on whether the subject reported a conscious
experience and had to answer additional questions related to the content. Signed informed
consent was obtained from all participants before the experiment, and ethical approval for
the study was obtained from the University of Wisconsin - Madison Institutional Review
Board. Data collection and analysis were not performed blind to the conditions of the
experiments.

Study participants experiment 1—Thirty-two healthy subjects (12 males, age 46

± 13.3 years, 24–65 [mean ± SD, range]), selected from a group of 69 subjects participating
in a larger research project in our laboratory, were included in this experiment. Subjects
were randomly selected from the subset of all participants that had at least one DE and NE
Author Manuscript

report in NREM sleep during the same night. Among the 240 awakenings performed in
these 32 subjects, 5 had to be excluded for technical problems and 2 because subjects were
too somnolent upon awakening to answer questions reliably. Comparisons between DEWR-
NE conditions and DE-DEWR conditions were performed in the subset of these participants
(N=20) that had all three types (DE, DEWR, NE) of report during the same night. In REM
sleep, only 6 out of the 69 participants presented both DE and NE. They were all included in
the analysis.

Study participants experiment 2—Seven healthy subjects were included in this

experiment (3 males, age 31 ± 8.8 years, 21–47 [mean ± SD, range]). None of the subjects in
this experiment participated in experiments 1 or 3. The behavioral results of 6 of these 7
subjects (i.e., dream recall rates and characteristics of dreams across sleep stages), but not
Author Manuscript

the EEG results, have been described in a previous publication9. Data from experiments 1
and 3 has not been reported previously.

Study participants received comprehensive explanations regarding the questionnaire used in

the experiment, and filled it in at home every morning upon awakening for two weeks before
the first study night. During this training phase experimenters communicated repeatedly with
the subjects to ensure that they were proceeding with the training and to answer any
questions the subjects had regarding the questionnaire. Between five and ten overnight

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 11

recordings were performed for each participant, with a maximum of three consecutive
Author Manuscript

nights. To increase the number of awakenings in REM sleep with respect to NREM sleep,
two additional nights were scheduled for three of the subjects, in which awakenings were
only performed in REM sleep. The total number of nights was distributed as follows among
subjects: S1:10 nights, S2:8 nights, S3:6 nights, S4:5 nights, S5:8 nights, S6:6 nights, S7:8
nights. Overall, 836 awakenings were performed, of which 5 had to be excluded because of
technical problems, and 16 because subjects were too somnolent upon awakening to answer
questions reliably.

Sleep recordings—Recordings were made at the Wisconsin Center for Sleep and
Consciousness. They were performed using a 256-channel high-density EEG (hd-EEG)
system (Electrical Geodesics, Inc., Eugene, Ore.). Four of the 256 electrodes placed at the
outer canthi of the eyes were used to monitor eye movements; additional polysomnography
channels were used to record submental electromyography. Sleep scoring was performed
Author Manuscript

over 30s epochs according to standard criteria51.

Preprocessing of data experiment 1—The signal corresponding to the 2 min

preceding each awakening was extracted and considered for analysis. The EEG signal was
sampled at 500 Hz and off-line band-pass filtered between 1 and 50 Hz for the NREM data
and between 0.3 and 50 Hz for the REM data, in order to be merged with the REM data of
Experiment 2. The NREM EEG data was high-pass filtered at 1Hz instead of lower
frequencies, as there were strong sweating artifacts in some of the participants that caused
intermittent high-amplitude slow frequency oscillatory activity around 0.3 Hz. REM data did
not contain any sweating artifacts and was high-pass filtered at 0.3Hz. Channels containing
artifacts were visually identified and replaced with data interpolated from nearby channels
using spherical splines (NetStation, Electrical Geodesic Inc.). To remove ocular, muscular,
Author Manuscript

and electrocardiograph artifacts we performed Independent Component Analysis (ICA)

using EEGLAB routines52. Only ICA components with specific activity patterns and
component maps characteristic of artifacts were removed53.

Preprocessing of data experiment 2—The signal corresponding to the 2 min

preceding each awakening was extracted and considered for analysis. The EEG signal was
sampled at 500 Hz and off-line band-pass filtered between 0.3 and 50 Hz for both NREM
and REM data. Artifact removal was carried out as in experiment 1.

Signal analysis
Source localization: The previously cleaned, filtered and average-referenced EEG signal
corresponding to the 20s before the awakening was extracted and analyzed at the source
Author Manuscript

level. Source modelling was performed using the GeoSource software (Electrical Geodesics,
Inc., Eugene, Ore.). A 4-shell head model based on the Montreal Neurological Institute
(MNI) atlas and a standard coregistered set of electrode positions were used to construct the
forward model. The source space was restricted to 2447 dipoles that were distributed over
7×7×7 mm cortical voxels. The inverse matrix was computed using the standardized low-
resolution brain electromagnetic tomography (sLORETA) constraint54. A Tikhonov
regularization procedure was applied to account for the variability in the signal-to-noise

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 12

ratio54. We computed spectral power density using the Welch’s modified periodogram
Author Manuscript

method (implemented with the pwelch function in MATLAB; The Math Works Inc, Natick,
MA) in 2s Hamming windows (8 segments, 50% overlap) to decompose the source signals
into frequency bands of interest.

Statistical Analysis: No statistical methods were used to pre-determine sample sizes but our
sample sizes were similar to those reported in previous studies, in which sleep was assessed
using high-density EEG recordings55,56. As within-subject contrasts were carried out, no
randomization in the organization of the experimental conditions was performed. Statistical
analyses were carried out in MATLAB. Statistical analyses and thresholds were selected
based on 1) the statistical power of the contrast and 2) the strength of the a priori hypothesis.
To compare brain activity between DE and NE, the source-modeled signal was averaged
within the time periods and frequency bands of interest. We then averaged the signal
associated with DEs and NEs within each subject and for each frequency band and stage
Author Manuscript

(REM and NREM sleep) separately. Whole-brain group level analyses on average absolute
power values were performed separately for each frequency range. Due to the nature of
high-density EEG, differences were expected to extend over a relatively large number of
voxels. For this reason, a statistical analysis with a cluster-based correction for multiple
comparisons was used (Statistical nonparametric mapping, SnPM)57 and a two-tailed t-test
with a corrected p<0.05 threshold was applied.

For analyses of content (faces, spatial setting, movement, speech), we used a more liberal
threshold (p<0.05, uncorrected, two-tailed paired t-tests), given that activations were
expected in restricted cortical regions, comprising only a few voxels. We first verified that
the data were normally distributed using the Lilliefors modification of the Kolmogorov-
Smirnov test (p>0.05, t(5) < 2.57 and t(6) < 2.45, depending on the number subjects
Author Manuscript

included in each contrast). For all planned comparisons (face, setting, movement, speech), a
normal distribution was confirmed for the large majority of voxels (range 81.2–95.0%).
Given the relative robustness of parametric t-tests to violations of normality, and the
relatively small percentage of voxels breaking this assumption, potential differences were
assessed using two-tailed paired t-tests (p<0.05).

Additional region of interest (ROI) based analyses were performed as follows: ROIs were
defined as 15mm radius spheres centered on coordinates (ex. fusiform face area) obtained
from previous imaging studies58–63. Then, high-frequency power values (20–50 Hz) were
averaged within these ROIs and compared between conditions (i.e., face vs. no face). Given
the strong a priori hypothesis about the direction of the contrast (based on the results of the
previous analysis), one-sided paired t-tests were used for these analyses. Larger ROIs
Author Manuscript

(25mm spheres) were used for the figures (Fig. 4) for better visibility. In the figures we
opted for displaying both the p<0.05 and the p<0.1 significance level, to show that the
activations are spatially smooth, as opposed to random, and thus more likely to correspond
to neural activations. Since experience-related brain activity for the content of experiences
was assumed to be very short-lived and subjects were specifically instructed to refer to the
‘very last’ experience before the alarm sound when reporting the presence of specific
contents, we considered shorter timeframes before the awakening when performing content-
specific contrasts, ranging from 0 to −8s with respect to the awakening.

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 13

To investigate the correlation between the thinking/perceiving score and high frequency
Author Manuscript

power (25–50 Hz), we calculated the Spearman rank correlation coefficient for each subject
and voxel, and then averaged the resulting coefficients across subjects to obtain group-level
correlation maps (Fig. 4). Statistical significance was determined using a permutation-based
approach. Specifically, null distributions were generated for each voxel by randomly
shuffling the two variables of interest (thinking/perceiving score and high-frequency power)
in each subject and recalculating the group-level mean correlation (n=1000). A voxel-wise
parametric one-sided t test was subsequently performed (p < 0.05).

Selection of frequency bands: For low frequency power comparisons, we selected the 1–4
Hz band, as this frequency range is associated with the presence of slow waves, which have
been linked to the loss of consciousness. Slow oscillations have been shown to exist not only
in NREM sleep, but also in in REM sleep30. For high-frequency power comparisons, a
slightly lower frequency range (20–50 Hz) was chosen for NREM sleep compared to REM
Author Manuscript

sleep (25–50 Hz), given that NREM sleep is typically characterized by slower frequencies64.
Frequencies above 50 Hz were not included in order to avoid muscle and powerline artifacts.
Frequencies below 20 Hz were avoided because of the potential overlap with fast sleep
spindles, which can reach 18 Hz65,66.

Hemispheric lateralization: To assess lateralization of findings, for each contrast we

evaluated the potential difference in the number of voxels showing a significant effect
between the two hemispheres by calculating a lateralization index (LI) as described in67. To
test for a statistically significant difference between the level of activity observed in
symmetrical areas across the two hemispheres, we calculated the relative difference in power
spectral density (PSD) between two symmetrical ROIs including the voxels that showed a
significant effect in the hemisphere characterized by the stronger LI, but not in the
Author Manuscript

contralateral side. In particular, we explored the potential interaction between hemisphere

(Left, Right) and type of report (DE, NDE, DEWR) using a repeated measure ANOVA.

Experiment 3
Study Participants Experiment 3—Seven additional participants (4 men, 3 women; age
22 ± 4 years, 19–31 years [mean ± SD, range]) not included in Experiments 1 or 2 were
recruited for Experiment 3. All participants were right-handed and had no history of
neurological disorder. Signed informed consent was obtained from all participants before the
experiment, and ethical approval for the study was obtained from the University of
Wisconsin-Madison Institutional Review Board.

Sleep Recordings—Hd-EEG was continuously recorded overnight using a 64-electrode

Author Manuscript

EEG cap (EasyCap, Herrsching, Germany) with equidistantly spaced sintered Ag/AgCl ring
electrodes. The signal was amplified using BrainAmp MR Plus amplifiers (16-bit analog-to-
digital conversion; 0.5-μV resolution; ±16.384-mV operating range; 5,000-Hz sampling; 0.1-
to 250-Hz band pass) (Brain Products, Munich, Germany) and transmitted via fiber-optic
cables. Data recording during the baseline night was made with BrainVision Recorder (v.
1.20) software while experimental nights data was readout in real-time, filtered (0.5–58 Hz),
and relayed to custom MATLAB scripts for online EEG power analysis using OpenVibe

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 14

software68. EEG was digitized at an online sampling rate of 500 Hz and was referenced to
Author Manuscript

the vertex. Electrode impedances were kept below 20 kΩ.

EEG Data Processing—Offline EEG data analysis was conducted with MATLAB
(Mathworks Inc., Natick, MA) using the EEGLAB toolbox52 and custom scripts. Sleep
recordings were re-staged offline according to standard criteria to verify online sleep
staging51. EEG data was epoched at 60 seconds prior to each awakening, consistent with the
time interval used for prediction (see Procedure). Twenty epochs were removed from
analysis due to the presence of EEG artifacts or because sleep stage could not be confirmed.
Data were bandpass filtered between 0.5 and 58 Hz using a Hamming windowed sinc FIR
filter. Bad channels were identified by visual inspection and interpolated using spherical
splines. Spectral power density was computed using Welch’s modified periodogram method.
Source localization was performed in Brainstorm69 with a noise-normalized estimate using
standardized low-resolution brain electromagnetic tomography (sLORETA54). Group-level
Author Manuscript

analysis on EEG sources was conducted using the GLM framework implemented in SPM12
(Wellcome Trust Department of Imaging Neuroscience, University College London).
Whole-cortex analyses were conducted and correction for multiple comparisons was
performed using topological FDR70 (cluster forming threshold of P = 0.0005; cluster size
threshold of P < 0.001; k = 100 vertices).

Procedure—Participants were invited to spend three consecutive nights in the lab. During
the first night (baseline) they were allowed to sleep without interruption. Individual
thresholds for low-frequency (1–4 Hz) and high-frequency (18–25 Hz) band power were
computed from baseline night recordings by binning all NREM sleep into 60s epochs and
computing the top and bottom 15% thresholds from the distribution for each band for each
participant over the entire baseline night (Fig. 5A, middle panel). 18–25 Hz power was used
Author Manuscript

for the high frequency power range of interest because higher frequencies are more prone to
artifact from eye movements and other sources71,72 and applying ICA for artifact removal to
data in an online fashion was technically challenging. On the experimental (second and
third) nights, participants were awoken from NREM (sleep stage 2 or 3) when their neural
activity surpassed a bispectral threshold in low- and high- frequency band power from an
electrode cluster centered over the posterior hot zone. To ensure a stable sleep stage had
been reached, participants were required to be in NREM for a minimum of 120 seconds
before the algorithm initiated an awakening. The bispectral threshold required that for an DE
prediction to be initiated low-frequency power had to be less than or equal to the lower 15%
threshold while high-frequency power had to simultaneously meet the top 15% threshold for
that subject. Likewise, for a NE prediction to be initiated low-frequency power had to be
greater than or equal to the top 15% threshold while high-frequency power had to
Author Manuscript

simultaneously be less than or equal to the bottom 15% threshold for that subject. This
procedure was motivated by the results of experiments 1 and 2, and aimed to predict the
presence of DE during time epochs in which the cortex was not only not bistable but also
displayed correlates of neural firing (and vice versa for predictions of NE13,16). Upon
awakening participants completed a short version of the sleep questionnaire used in
experiments 1 and 2.

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 15

The post-hoc statistical analyses were performed to confirm the posterior maximum of the
Author Manuscript

low-frequency/high-frequency differences. Because extreme ratios were compared,

statistical differences were expected to be much stronger in this experiment than in
experiments 1 and 2. For this reason, a more stringent threshold was used (p<0.001, FDR
cluster corrected). This analysis confirmed the posterior maximum of the low-frequency/
high-frequency differences. For an analysis with SnPM correction with the same threshold
than in experiments 1 and 2, see Supplementary Figure 7.

Data availability
The data that support the findings of this study are available from the corresponding author
upon reasonable request. Code was written using standard Matlab functions and freely
available toolboxes (Brainstorm, EEGlab). It is available upon reasonable request.
Author Manuscript

Supplementary Material
Refer to Web version on PubMed Central for supplementary material.

Acknowledgments
Supported by NIH/NCCAM P01AT004952 (GT), NIH/NIMH 5P20MH077967 (GT), Swiss National Science
Foundation Grants 139778 (FS), 145571 (FS) and 155120 (LP), Swiss Foundation for Medical Biological Grants
151743 and 145763 (FS), NIH/NINDS F32NS089348 (BB), UW Medical Scientist Training Program Grant T32
GM008692 (JJL), NIH Grants MH064498 (BRP) and MH095984 (BRP and JJL).

The authors thank David Bachhuber, Leonardo Barbosa, Emmanuel Carrera, Anna Castelnovo, Amelia Cayo,
Chiara Cirelli, Chadd Funk, Matthew Gevelinger, Jeanne Harris, Armand Mensen, Poorang Nori, Richard Smith,
Laurène Vuillaume, Sophy Yu, Corinna Zennig and the undergraduate research assistants for help with data
collection, sleep scoring, technical assistance and helpful discussions.
Author Manuscript

References
1. Stickgold R, Malia A, Fosse R, Propper R, Hobson JA. Brain-mind states: I. Longitudinal field study
of sleep/wake factors influencing mentation report length. Sleep. 2001; 24:171–179. [PubMed:
11247053]
2. Nir Y, Tononi G. Dreaming and the brain: from phenomenology to neurophysiology. Trends Cogn
Sci. 2010; 14:88–100. [PubMed: 20079677]
3. Schwartz S, Maquet P. Sleep imaging and the neuro-psychological assessment of dreams. Trends
Cogn Sci. 2002; 6:23–30. [PubMed: 11849612]
4. Aserinsky E, Kleitman N. Regularly occuring periods of eye motility, and concominant phenomena,
during sleep. Science. 1953; 118:273–274. [PubMed: 13089671]
5. Moruzzi G, Magoun HW. Brain stem reticular formation and activation of the EEG.
Electroencephalogr Clin Neurophysiol. 1949; 1:455–473. [PubMed: 18421835]
6. Steriade M, Timofeev I, Grenier F. Natural waking and sleep states: a view from inside neocortical
Author Manuscript

neurons. J Neurophysiol. 2001; 85:1969–1985. [PubMed: 11353014]

7. Maquet P. Functional neuroimaging of normal human sleep by positron emission tomography. J
Sleep Res. 2000; 9:207–231. [PubMed: 11012860]
8. Chow HM, et al. Rhythmic alternating patterns of brain activity distinguish rapid eye movement
sleep from other states of consciousness. Proc Natl Acad Sci U S A. 2013; 110:10300–10305.
[PubMed: 23733938]
9. Siclari F, Larocque JJ, Postle BR, Tononi G. Assessing sleep consciousness within subjects using a
serial awakening paradigm. Front Psychol. 2013; 4:542. [PubMed: 23970876]

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 16

10. Foulkes D. Dream reports from different stages of sleep. The Journal of Abnormal and Social
Psychology. 1962; 65:14–25. [PubMed: 13894288]
Author Manuscript

11. Nir Y, et al. Regional slow waves and spindles in human sleep. Neuron. 2011; 70:153–169.
[PubMed: 21482364]
12. Pigorini A, et al. Bistability breaks-off deterministic responses to intracortical stimulation during
non-REM sleep. Neuroimage. 2015; 112:105–113. [PubMed: 25747918]
13. Tononi G, Massimini M. Why does consciousness fade in early sleep? Ann N Y Acad Sci. 2008;
1129:330–334. [PubMed: 18591492]
14. Purdon PL, et al. Electroencephalogram signatures of loss and recovery of consciousness from
propofol. Proc Natl Acad Sci U S A. 2013; 110:E1142–1151. [PubMed: 23487781]
15. Jones BE. From waking to sleeping: neuronal and chemical substrates. Trends Pharmacol Sci.
2005; 26:578–586. [PubMed: 16183137]
16. Steriade M, Contreras D, Amzica F, Timofeev I. Synchronization of fast (30–40 Hz) spontaneous
oscillations in intrathalamic and thalamocortical networks. J Neurosci. 1996; 16:2788–2808.
[PubMed: 8786454]
17. Le Van Quyen M, et al. Large-scale microelectrode recordings of high-frequency gamma
Author Manuscript

oscillations in human cortex during sleep. J Neurosci. 2010; 30:7770–7782. [PubMed: 20534826]
18. Nielsen, TA. Handbook of behavioral state control Cellular and molecular mechanisms. Lydic, R.,
Baghdoyan, HA., editors. CRC; 1999.
19. Nielsen TA. A review of mentation in REM and NREM sleep: “covert” REM sleep as a possible
reconciliation of two opposing models. Behav Brain Sci. 2000; 23:851–866. [PubMed: 11515145]
20. Cavallero C, Cicogna P, Natale V, Occhionero M, Zito A. Slow wave sleep dreaming. Sleep. 1992;
15:562–566. [PubMed: 1475572]
21. Voss U, Holzmann R, Tuin I, Hobson JA. Lucid dreaming: a state of consciousness with features of
both waking and non-lucid dreaming. Sleep. 2009; 32:1191–1200. [PubMed: 19750924]
22. Dresler M, et al. Neural correlates of dream lucidity obtained from contrasting lucid versus non-
lucid REM sleep: a combined EEG/fMRI case study. Sleep. 2012; 35:1017–1020. [PubMed:
22754049]
23. Koch C, Massimini M, Boly M, Tononi G. Neural correlates of consciousness: progress and
problems. Nat Rev Neurosci. 2016; 17:307–321. [PubMed: 27094080]
Author Manuscript

24. Balestrini S, et al. Reply: The dorsal cingulate cortex as a critical gateway in the network
supporting conscious awareness. Brain. 2016; 139:e24. [PubMed: 26719381]
25. Herbet G, et al. Disrupting posterior cingulate connectivity disconnects consciousness from the
external environment. Neuropsychologia. 2014; 56:239–244. [PubMed: 24508051]
26. Revonsuo A. The reinterpretation of dreams : an evolutionary hypothesis of the function of
dreaming. Behav Brain Sci. 23:877–901. discussion 904–1121 (2000).
27. Windt, JM. Dreaming - a conceptual framework for philosophy of mind and empirical research.
1st. MIT press; 2015.
28. Christophel TB, Hebart MN, Haynes JD. Decoding the contents of visual short-term memory from
human visual and parietal cortex. J Neurosci. 2012; 32:12983–12989. [PubMed: 22993415]
29. Postle BR. The cognitive neuroscience of visual short-term memory. Curr Opin Behav Sci. 2015;
1:40–46. [PubMed: 26516631]
30. Funk CM, Honjoh S, Rodriguez AV, Cirelli C, Tononi G. Local Slow Waves in Superficial Layers
of Primary Cortical Areas during REM Sleep. Curr Biol. 2016; 26:396–403. [PubMed: 26804554]
Author Manuscript

31. Schiff ND, Nauvel T, Victor JD. Large-scale brain dynamics in disorders of consciousness. Curr
Opin Neurobiol. 2014; 25:7–14. [PubMed: 24709594]
32. Sitt JD, et al. Large scale screening of neural signatures of consciousness in patients in a vegetative
or minimally conscious state. Brain. 2014; 137:2258–2270. [PubMed: 24919971]
33. Gokyigit A, Caliskan A. Diffuse spike-wave status of 9-year duration without behavioral change or
intellectual decline. Epilepsia. 1995; 36:210–213. [PubMed: 7821280]
34. Wikler A. Pharmacologic dissociation of behavior and EEG “sleep patterns” in dogs; morphine, n-
allylnormorphine, and atropine. Proc Soc Exp Biol Med. 1952; 79:261–265. [PubMed: 14912089]

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 17

35. Horikawa T, Tamaki M, Miyawaki Y, Kamitani Y. Neural Decoding of Visual Imagery During
Sleep. Science. 2013; 340:639–642. [PubMed: 23558170]
Author Manuscript

36. Malcolm N. Dreaming and skepticism. Philosoph Rev. 1957; 65:14–37.

37. Dennet D. Are dreams experiences? Philosoph Rev. 1976; 9:673–690.
38. Aru J, et al. Local category-specific gamma band responses in the visual cortex do not reflect
conscious perception. J Neurosci. 2012; 32:14909–14914. [PubMed: 23100413]
39. Murphy M, et al. Propofol anesthesia and sleep: a high-density EEG study. Sleep. 2011; 34:283–
291A. [PubMed: 21358845]
40. Naro A, Bramanti P, Leo A, Russo M, Calabro RS. Transcranial Alternating Current Stimulation in
Patients with Chronic Disorder of Consciousness: A Possible Way to Cut the Diagnostic Gordian
Knot? Brain Topogr. 2016; 29:623–644. [PubMed: 27062669]
41. Voss U, et al. Induction of self awareness in dreams through frontal low current stimulation of
gamma activity. Nat Neurosci. 2014; 17:810–812. [PubMed: 24816141]
42. Christoff K, Gordon AM, Smallwood J, Smith R, Schooler JW. Experience sampling during fMRI
reveals default network and executive system contributions to mind wandering. Proc Natl Acad Sci
U S A. 2009; 106:8719–8724. [PubMed: 19433790]
Author Manuscript

43. Perogamvros L, Schwartz S. The roles of the reward system in sleep and dreaming. Neurosci
Biobehav Rev. 2012; 36:1934–1951. [PubMed: 22669078]
44. Fleming SM, Dolan RJ. The neural basis of metacognitive ability. Philos Trans R Soc Lond B Biol
Sci. 2012; 367:1338–1349. [PubMed: 22492751]
45. Hohwy J. The neural correlates of consciousness: new experimental approaches needed?
Conscious Cogn. 2009; 18:428–438. [PubMed: 19345590]
46. Baars BJ, Ramsoy TZ, Laureys S. Brain, conscious experience and the observing self. Trends
Neurosci. 2003; 26:671–675. [PubMed: 14624851]
47. Dehaene S, et al. Cerebral mechanisms of word masking and unconscious repetition priming. Nat
Neurosci. 2001; 4:752–758. [PubMed: 11426233]
48. Hebb DO, Penfield W. Human behavior after extensive bilateral removal from the frontal lobes.
Arch Neurol Psychiatr. 1940; 42:421–438.
49. Frassle S, Sommer J, Jansen A, Naber M, Einhauser W. Binocular rivalry: frontal activity relates to
introspection and action but not to perception. J Neurosci. 2014; 34:1738–1747. [PubMed:
Author Manuscript

24478356]
50. Kamphuisen A, Bauer M, van Ee R. No evidence for widespread synchronized networks in
binocular rivalry: MEG frequency tagging entrains primarily early visual cortex. J Vis. 2008; 8(4):
1–8.
51. Iber, C., Ancoli-Israel, S., Chesson, A., Quan, SF. The AASM Manual for the Scoring of Sleep and
Associated Events: Rules, Terminology and Technical Specifications. 1st. American Academy of
Sleep Medicine; 2007.
52. Delorme A, Makeig S. EEGLAB: an open source toolbox for analysis of single-trial EEG
dynamics including independent component analysis. J Neurosci Methods. 2004; 134:9–21.
[PubMed: 15102499]
53. Jung TP, et al. Removing electroencephalographic artifacts by blind source separation.
Psychophysiology. 2000; 37:163–178. [PubMed: 10731767]
54. Pascual-Marqui RD. Standardized low-resolution brain electromagnetic tomography (sLORETA):
technical details. Methods Find Exp Clin Pharmacol. 2002; 24(Suppl D):5–12. [PubMed:
Author Manuscript

12575463]
55. Huber R, Ghilardi MF, Massimini M, Tononi G. Local sleep and learning. Nature. 2004; 430:78–
81. [PubMed: 15184907]
56. Huber R, et al. Arm immobilization causes cortical plastic changes and locally decreases sleep
slow wave activity. Nat Neurosci. 2006; 9:1169–1176. [PubMed: 16936722]
57. Nichols TE, Holmes AP. Nonparametric permutation tests for functional neuroimaging: a primer
with examples. Hum Brain Mapp. 2002; 15:1–25. [PubMed: 11747097]

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 18

58. Rossion B, Schiltz C, Crommelinck M. The functionally defined right occipital and fusiform “face
areas” discriminate novel from visually familiar faces. Neuroimage. 2003; 19:877–883. [PubMed:
Author Manuscript

12880816]
59. Corbetta M, Miezin FM, Shulman GL, Petersen SE. A PET study of visuospatial attention. J
Neurosci. 1993; 13:1202–1226. [PubMed: 8441008]
60. Malhotra P, Coulthard EJ, Husain M. Role of right posterior parietal cortex in maintaining attention
to spatial locations over time. Brain. 2009; 132:645–660. [PubMed: 19158107]
61. Grossman E, et al. Brain areas involved in perception of biological motion. J Cogn Neurosci. 2000;
12:711–720. [PubMed: 11054914]
62. Puce A, Allison T, Bentin S, Gore JC, McCarthy G. Temporal cortex activation in humans viewing
eye and mouth movements. J Neurosci. 1998; 18:2188–2199. [PubMed: 9482803]
63. Narain C, et al. Defining a left-lateralized response specific to intelligible speech using fMRI.
Cereb Cortex. 2003; 13:1362–1368. [PubMed: 14615301]
64. Gross DW, Gotman J. Correlation of high-frequency oscillations with the sleep-wake cycle and
cognitive activity in humans. Neuroscience. 1999; 94:1005–1018. [PubMed: 10625043]
65. Andrillon T, et al. Sleep spindles in humans: insights from intracranial EEG and unit recordings. J
Author Manuscript

Neurosci. 2011; 31:17821–17834. [PubMed: 22159098]

66. Sirota A, Csicsvari J, Buhl D, Buzsaki G. Communication between neocortex and hippocampus
during sleep in rodents. Proc Natl Acad Sci U S A. 2003; 100:2065–2069. [PubMed: 12576550]
67. Seghier ML. Laterality index in functional MRI: methodological issues. Magn Reson Imaging.
2008; 26:594–601. [PubMed: 18158224]
68. Renard Y, Lotte F. OpenViBE: an open-source software platform to design, test, and use brain-
computer interfaces in real and virtual environments. Presence: teleoperators and virtual
environments. 2010; 19:35–53.
69. Tadel F, Baillet S, Mosher JC, Pantazis D, Leahy RM. Brainstorm: a user-friendly application for
MEG/EEG analysis. Comput Intell Neurosci. 2011; 2011:879716. [PubMed: 21584256]
70. Chumbley J, Worsley K, Flandin G, Friston K. Topological FDR for neuroimaging. Neuroimage.
2010; 49:3057–3064. [PubMed: 19944173]
71. Keren AS, Yuval-Greenberg S, Deouell LY. Saccadic spike potentials in gamma-band EEG:
characterization, detection and suppression. Neuroimage. 2010; 49:2248–2263. [PubMed:
Author Manuscript

19874901]
72. Muthukumaraswamy SD. High-frequency brain activity and muscle artifacts in MEG/EEG: a
review and recommendations. Front Hum Neurosci. 2013; 7:138. [PubMed: 23596409]
Author Manuscript

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 19
Author Manuscript
Author Manuscript
Author Manuscript

Figure 1. Dreaming experience vs. no experience in NREM sleep (low-frequency power)

A. Cortical distribution of t-values for the contrast between DEs and NEs at the source level
for low-frequency power (1–4 Hz) in NREM sleep (20s before the awakening). p<0.05, after
correction for multiple comparisons (two-tailed, paired t-tests, 32 subjects, t(31) > 2.04). B.
Same as A for the contrast between NE and DEWR (two-tailed, paired t-tests, 20 subjects,
t(19) > 2.09).
Author Manuscript

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 20
Author Manuscript
Author Manuscript
Author Manuscript

Figure 2. Dreaming experience vs. no experience in REM sleep (low-frequency power)

A. Cortical distribution of t-values for the contrast between DEs and NEs at the source level
for low-frequency (1–4 Hz) power in REM sleep (20s before the awakening). p<0.05 after
correction for multiple comparisons (two-tailed, paired t-tests, 10 subjects, t(9)>2.26). B.
Conjunction maps: differences and overlap compared to the same contrast performed in
NREM sleep.
Author Manuscript

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 21
Author Manuscript
Author Manuscript
Author Manuscript

Figure 3. Dreaming experience vs. no experience in NREM and REM sleep (high- frequency
Author Manuscript

power)
A. Cortical distribution of t-values for the contrast between DEs and NEs at the source level
for high-frequency power (20–50 Hz) in NREM sleep (20s before the awakening). p<0.05
after correction for multiple comparisons (two-tailed, paired t-tests, 32 subjects, t(31) >
2.04). B. Same as A for the contrast between DE and DEWR in NREM sleep (two-tailed,
paired t-tests, 20 subjects, t(19) > 2.09) C. same as A for high-frequency power (25–50 Hz)
in REM sleep (two-tailed, paired t-tests, 10 subjects, t(9) > 2.26).

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 22
Author Manuscript
Author Manuscript
Author Manuscript

Figure 4. The content of dream experiences in REM sleep

Author Manuscript

A. Correlation between the thinking/perceiving score and 25–50 Hz power (last 8s, 7
subjects). Left: mean Spearman rank correlation coefficients (7 subjects). Right: significant
voxels p<0.05 (one-tailed permutation test, r>0.14). B. Left: 25–50 Hz power differences
(DE with face minus DE without face). ROI contrast R FFA p=0.023; one-tailed paired t-test
(7 subjects, t(6) = 2.52). Right: fusiform face area19 (red). C. Upper row: 25–50 Hz average
power differences between DEs with and without a spatial setting (6 subjects, t(5) > 2.57).
Right: right posterior parietal cortex21. Middle row: movement vs. no movement (7

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 23

subjects, t(6) > 2.45). Right: superior temporal sulcus22. Bottom row: speech vs. no speech
(7 subjects, t(6) > 2.45). Right: Wernicke’s area24. Two-tailed paired t-tests. p<0.05 (red)
Author Manuscript

and p<0.01 (yellow).

Author Manuscript
Author Manuscript
Author Manuscript

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.

 Siclari et al. Page 24
Author Manuscript
Author Manuscript

Figure 5. Real-time prediction of dream experience

Author Manuscript

A. Awakenings were performed in NREM when neural activity surpassed a bispectral

threshold in low- frequency (LF; 0.5–4.5 Hz) and high-frequency (HF; 18–25 Hz) power
over the posterior hot zone. B. DE trials had significantly lower LF activity and significantly
higher HF activity in DE compared to NE prediction trials. The bottom and top of the boxes
show the 25th and 75th percentile (the lower and upper quartiles), the inner band shows the
median, and the whiskers show the upper and lower quartiles +/− 1.5 * IQR (inter quartile
range). Asterisks indicate significant differences (p=0.001; two-tailed t-test) between CE and
NCE trials. C. Prediction accuracy for DE (55) and NE (27) trials. Asterisks indicate
significant differences (p<0.001; one sample t-test two-tailed) between CE and NCE
prediction accuracy and chance (50%). D. DE had a higher HF/LF power ratio compared to
NE in cortical regions including bilateral occipital, medial and lateral parietal, medial
temporal and inferior frontal cortex (p<0.001; two-tailed paired t-test, FDR cluster
Author Manuscript

corrected).

Nat Neurosci. Author manuscript; available in PMC 2017 October 10.