Access to this full-text is provided by Springer Nature.
Content available from Scientific Reports
This content is subject to copyright. Terms and conditions apply.
Variations of autonomic arousal
mediate the reportability of mind
blanking occurrences
Paradeisios Alexandros Boulakis1,2, Nicholas John Simos1, Stefania Zoi1,
Sepehr Mortaheb1,2, Christina Schmidt2,3, Federico Raimondo4,5,7 & Athena Demertzi1,2,6,7
Mind blanking (MB) is the inability to report mental events during unconstraint thinking. Previous
work shows that MB is linked to decreased levels of cortical arousal, indicating dominance of cerebral
mechanisms when reporting mental states. What remains inconclusive is whether MB can also ensue
from autonomic arousal manipulations, pointing to the implication of peripheral physiology to mental
events. Using experience sampling, neural, and physiological measurements in 26 participants, we rst
show that MB was reported more frequently in low arousal conditions, elicited by sleep deprivation.
Also, there was partial evidence for a higher occurence of MB reports in high arousal conditions,
elicited by intense physical exercise. Transition probabilities revealed that, after sleep deprivation,
mind wandering was more likely to be followed by MB and less likely to be followed by more mind
wandering reports. Using classication schemes, we found higher performance of a balanced random
forest classier trained on both neural and physiological markers in comparison to performance when
solely neural or physiological were used. Collectively, we show that both cortical and autonomic
arousal aect MB report occurrences. Our results establish that MB is supported by combined brain-
body congurations, and, by linking mental and physiological states, they pave the way for novel
embodied accounts of spontaneous thinking.
‘The stage 1 protocol for this Registered Report was accepted in principle on 02/01/23. The protocol,
as accepted by the journal, can be found at: 10.17605/OSF.IO/SH2YE’ Techniques: Life sciences
techniques, Biophysical methods [Electrocardiography - EKG]; Life sciences techniques, Biophysical
methods [Electroencephalography - EEG]; CTS received date: 27.11.2024.
Keywords Mind blanking, Experience sampling, Brain-body interactions, Machine learning, Spontaneous
thinking
During ongoing mentation, our mind constantly shis across dierent mental states. ese mental states typically
bear some content (“what we think about”) and indicate a relationship towards that content (i.e., perceiving,
fearing, hoping, remembering)1. As we move through the environment, our thoughts uctuate between the
external and internal milieu2,3, resulting in a uid stream of consciousness4. External content is tightly coupled
to the processing of environmental stimuli and task-demanding conditions. Internal content is more associated
with self-referential processing and internal dialogue, widely referred to as “mind wandering” (MW)4. Inclusive
as this external-internal dipole may seem, it does not capture the full scope of the “aboutness” of mental content.
Recent work has highlighted another mental state, where people report that they are “thinking of nothing”
or “their mind just went away”, a phenomenological experience termed “mind blanking”(MB)5. As MB is
relatively new in the landscape of ongoing cognition, the extent of MB episodes in daily and clinical settings
remains widely uncharacterized. For example, a recent study found that MB might be miscategorized as MW in
ADHD symptom evaluation6. erefore, the experience of MB occurrences poses a challenge to our everyday
functioning and our understanding of the continuous nature of the stream of consciousness.
1Physiology of Cognition Lab, GIGA-CRC Human Imaging Unit, GIGA Research, University of Liège, Liège, Belgium.
2Fund for Scientic Research FNRS, Brussels, Belgium. 3Sleep & Chronobiology Lab, GIGA-CRC Human Imaging
Unit, GIGA Research, University of Liège, Liège, Belgium. 4Institute of Neuroscience and Medicine, Research Centre
Jülich, Brain & Behaviour (INM-7), Jülich, Germany. 5Institute of Systems Neuroscience, Medical Faculty, Heinrich
Heine University Düsseldorf, Düsseldorf, Germany. 6Psychology and Neuroscience of Cognition Research Unit,
University of Liège, Liège, Belgium. 7Federico Raimondo and Athena Demertzi contributed equally to this work.
email: a.demertzi@uliege.be
OPEN
REGISTERED
REPORT
Scientic Reports | (2025) 15:4956 1
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Currently, there is no clear answer as to how MB reports are generated. So far, behavioral studies show that
MB can arise aer conscious mental eort to empty our mind7–9, is usually unintentional5,10,11 and gets reported
less frequently during unconstrained thinking compared to MW and sensory/perceptual mental states5,11–13. At
the brain level, the inability to report mental events aer the prompt to “empty the mind” has been associated
with activation of the anterior cingulate/medial prefrontal cortex, and deactivation of inferior frontal gyrus/
Broca’s areas and the hippocampus, which the authors interpreted as the inability to verbalize internal mentation
(inner speech)8. Recently, we found that the functional connectome of fMRI volumes around MB reports was
similar to a unique brain pattern of overall positive inter-areal connectivity12 which was also characterized
by increased amplitude of fMRI global signal (i.e. averaged connectivity across all grey matter voxels), an
implicit indicator of low arousal14–16. For example, the amplitude of the global signal correlated negatively with
EEG vigilance markers (alpha, beta EEG frequency bands), while increases in EEG vigilance due to caeine
ingestion were associated with reduced global signal amplitude14. Our ndings corroborate recent EEG-related
evidence supporting the possibility of “local sleeps” during MB reportability10,17. “Local sleeps” refer to the scalp
distribution of EEG potentials during wakefulness, in the form of high-intensity, slow oscillatory activity in the
theta/delta band, which could dierentiate between MB and MW, with more frontocentral potentials tied to
MW and parietal to MB10. Together, the presence of slow waves preceding MB reports and the high fMRI global
signal hint toward the role of arousal in mental content reportability. Starting from this line of evidence, we infer
that arousal uctuations drive MB reportability.
Arousal is a multidimensional construct generally referring to the behavioral state of being awake and alert,
supporting wakefulness, responsiveness to environmental stimuli, and attentiveness18,19. Anatomically, arousal
is supported by the ascending arousal system, the autonomic nervous system, and the endocrine system18. Early
on, Lacey viewed arousal in terms of behavioral arousal (indicated by a responding organism, like restlessness
and crying), cortical arousal (evidenced by desynchronized fast oscillatory activity), and autonomic arousal
(indicated by changes in bodily functions)20. Cortical arousal is self-generated through the reticulate formation
and propagated through dorsal, thalamic, and ventral subthalamic pathways21, and can be indexed by the alpha,
theta, and delta EEG bands during wakefulness22,23. Lower levels of cortical arousal in the form of slow waves
have been associated with an increased number of missed stimuli in behavioral tasks11,23 and decreased thought
intensity24. Also, lower levels of arousal indexed by pupil size have been correlated with a higher probability of
MB reports in sustained attention tasks11,25,26.
Much as it may have been done in terms of cortical arousal, the present study will focus on how autonomic
arousal inuences MB reportability, which is widely understudied. Our choice is justied by the theoretical
assumption that mental function is tightly linked to peripheral body functions, as expressed by the embodied
cognition stance27. Briey, embodiment holds that cognition is bound to a living body interacting with a
dynamic environment, and conceptualizes cognition as the result of brain-body interactions during dynamic
contexts. From that perspective, modications in autonomic arousal are expected to lead to dierential
reportability of mental states. Autonomic arousal links the body and the brain through spinal cord projections
from peripheral organs to the brainstem and can be indexed by physiological signals reecting sympathetic/
parasympathetic balance, such as heart rate, galvanic skin response, and uctuations in pupil size28. Converging
evidence suggests that aerent physiological signals and biological rhythms, such as the cardiac or the respiratory
phase, play a modulatory role in conscious perception29,30, metacognition31, aective salience of information32,
and perceptual condence of sensory sampling33, both during task performance and in-silico simulations34.
Alterations in autonomic arousal were also found to inuence brain activity in that fMRI volumes characterized
by lower arousal levels (indexed by decreased pupil size) showed reduced in-between network integration and
inter-subject variability in comparison to scans characterized by high arousal levels (indexed by increased pupil
size)35.
Taken together, we here advocated for a link between autonomic arousal and thought reportability. Firstly,
we examined how MB report distribution shied across dierent autonomic arousal conditions. To this end,
we used experience sampling under dierently elicited arousal conditions. Experience sampling is a though-
sampling methodology, where people are probed to report their mental state at random intervals, probed by
an external cue4. We employed this task at three distinct arousal conditions: Baseline, High (post-workout), and
Low (post-sleep deprivation). Our operational hypothesis was that optimal levels of autonomic arousal (xed
variable) are necessary for optimal mental state reportability (dependent variable). We expected that deviations
from optimal levels, such as aer sleep deprivation or intense physical exercise, would alter our stream of thought
and promote more frequent MB reports (Supplementary Table S1 for the full scope of our hypotheses). Secondly,
we opted to identify specic brain-body interaction patterns that would promote MB reportability. To this end,
we utilized multimodal neurophysiological recordings and a machine-learning approach to decode MB reports
from arousal measurements.
Methods
Design
e study included healthy volunteers recruited aer campus poster advertisements, intranet electronic
invitations, and through the ULiège “petites annonces” e-campus platform. Inclusion criteria were: (a) right-
handedness, (b) age>18 years, (c) minimal exercise background (<2h per week), (d) good subjective sleep quality
(Pittsburgh Sleep Quality Index [PSQI] ≤ 536), (e) habitual sleep duration of 8 ± 1 hours. Exclusion criteria
were: (a) history of developmental, psychiatric, or neurological illness resulting in documented functional
disability, (b) severe anomalies in pupil shape or inability to open both eyes preventing pupil measurement37,
(c) analgesic medication which may aect physiological arousal, (d) history of psychiatric illness pertaining to
anxiety disorders or scores < 9 in the General Anxiety Disorder-7 (GAD-7 scale)38 as anxious participants may
Scientic Reports | (2025) 15:4956 2
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
experience biased perceptions of their bodily states39, (e) extreme chronotypes, (f) shi work or traveling over
time zones in the past 3 months.
Experience sampling was utilized in a within-participants repeated-measures design. During an experience
sampling session, participants laid restfully and were directed to let their minds wander, without any specic
instructions towards internal (daydreaming, memories, prospective events) or external thoughts (body
sensations, sensory stimuli in their immediate environment). Auditory probes (total n=40, 500Hz simple tones)
invited participants to report what they were thinking at the moment just preceding the probe. e inter-probe
interval was sampled from a uniform distribution between 110 and 120 seconds. Report times were monitored
online to examine if participants missed the probe or fell asleep due to our experimental manipulation. In case of
a report time > 6s, participants were reminded to report their mental state as soon as they heard the probe and
indicate they were awake via button press. In case of unresponsiveness, the experimenters manually awakened
the participant. Depending on the probes’ trigger times and participants’ reaction times, a recording lasted on
average 70-90 minutes. We chose to present 40 probes (overall length approximately 1hour and 15minutes) to
avoid fatigue/drowsiness and the possibility of participants returning to baseline arousal aer the experimental
manipulations. Also, the relatively large experience sampling interval, compared to previous studies, was used
to record enough samples to accurately estimate physiological markers from slow oscillatory signals, such as
heart-rate variability. Upon the probe, participants had to choose among four distinct choices describing their
mental state: mind blanking (MB), mind wandering (MW), perceptual sensations (SENS), or sleep (SLEEP).
ese response options were chosen to minimize assumptions about what the actual partition of mental states
might be. For example, debates about what can be classied as MW40 refer to whether MW is a coherent cluster
of events1,41 and how it is separated from awareness and processing of environmental stimuli40,42. We believe
that our division respects the literature on internal/external thought-orientation brain networks3,43,44 while
introducing minimum assumptions as to the actual content of each state. e introduction of the sleep option
facilitated the identication of trials where participants fell asleep due to the reduced vigilance. Participants
indicated their responses via button press from a response keyboard placed under their dominant hand. We
repeated the experience sampling task on three distinct days, over the span of two weeks under three conditions:
(a) experience sampling during spontaneous thinking without arousal modulations (Baseline), (b) experience
sampling elicited through short, high-intensity interval training (High Arousal), (c) experience sampling aer
total sleep deprivation (Low Arousal) (Fig. 1). e goal of both arousal manipulations was to promote distinct
changes in physiological and cortical markers associated with arousal mechanisms (Supplementary Table S2).
Monitoring of arousal changes was done with physiological and cortical measurements. In case when participants
did not show distinct cortical and physiological changes aer our arousal manipulations, they were excluded
Fig. 1. Experimental protocol. Top e experience sampling task invited participants to sit idly and relax,
letting their minds wander. Every 110–120s, a 500 Hz auditory cue probed participants to report what
they were thinking at that moment. Participants were able to choose from 4 presented responses: Mind
blanking (MB), Mind wandering (MW), Perceptual Sensations (SENS), and Sleep (SLEEP). Bottom Repeated-
measures autonomic arousal recordings. To test how spontaneous thoughts unfold over time across dierent
arousal conditions, we rst invited people for baseline assessments on Day 1 (Baseline condition). On Day
2 participants underwent a 15-minute high-intensity exercise (High Arousal condition) and on Day 3
they participated in a total sleep deprivation protocol (Low Arousal condition). e High and Low Arousal
conditions were counter-balanced across participants. Multimodal physiological recordings were used to
monitor arousal manipulations. e dataset was constituted of EEG, pupillometry, ECG, EDA, and respiratory
data; the arrows indicate the hypothesized directions of the derived metrics.
Scientic Reports | (2025) 15:4956 3
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
from further analysis. Eect monitoring was done by examining the heart rate in High Arousal as well as the EEG
spectra in both High and Low Arousal.
In High Arousal, participants rst performed high-intensity interval activity in the form of static cycling.
ey started with a warm-up training session of 3 minutes to avoid potential muscle trauma and then cycled for
45 s as fast as possible. A resting period of 15 s followed. A total number of 10 workout cycles was administered.
e choice of this timing protocol rested on previous studies indicating that similar exercise routines produce
distinct and sustained sympathetic activity45,46 and cortical excitation46, which can last between 30 and 90
minutes aer exercise cessation47.
In Low Arousal, participants performed the experience sampling task aer one night of total sleep deprivation.
Sleep deprivation leads to an arousal state that is behaviourally distinct from typical wakefulness48,49, promotes
specic neuronal signatures ("local sleeps” in the delta band)11, and has a distinct physiological expression.
Critically, we do not wish to claim that sleep states are identical to “local sleeps”, nor do we suggest an overlap
between low arousal due to sleep deprivation and unconsciousness during sleep. To acquire estimates of their
mean sleep schedule, participants wore an actimeter for one week before the total sleep deprivation protocol
(Supplementary Fig. S1; available for 24/26 subjects due to data corruption). e total sleep deprivation protocol
was as follows: A week prior to sleep deprivation, participants were provided with an actimetry device to track
wake-sleep schedule, and were instructed to follow a consistent 8h sleep schedule. On the deprivation day,
participants arrived at the lab one hour before their normal sleep time to extract their actimetry baseline data,
estimate the optimal sleep deprivation window, and to provide baseline vigilance, drowsiness, and sleepiness
measurements. Aer a total sleep deprivation of 26h (16h of typical wakefulness, 8h of sleep deprivation, and
a 2h post-sleep deprivation period) participants began the post-sleep deprivation, experience sampling session.
As an example, a participant who typically slept at 12 am would arrive at the lab at 11 pm, start sleep deprivation
at 12 am, nish sleep deprivation at 8 am, and perform the experience sampling task at 10 am. Should slow-wave
activity during wakefulness follow the same circadian modulation it follows during sleep50, a potential confound
that could have lowered the power of our analysis is the time window of the experience sampling task. However,
as suggested in50, the relative time-window we selected did not fall under a critical point of large reductions in the
amplitude of the slow-waves. e 2-hour, post-deprivation waiting window allowed us to match the time of the
experience sampling across the 3 conditions, avoiding potential circadian confounds on experience sampling,
as we could easier match sleep-wake cycles and the time of the experience sampling within each participant. We
chose this sleep manipulation as similar manipulations have been previously used to examine the eects of sleep
pressure51,52, and have been shown to elicit distinct low-arousal cortical proles53,54, as well as changes in the
sympathetic/parasympathetic balance55.
Sleep deprivation was controlled with regard to light inuence (illuminance = 15 lux), caloric intake
(standardized meals every 4 h), and body posture (semi-recumbent position during scheduled wakefulness)
to minimize potential masking eects on the sleep-wake regulatory system. Participants were not allowed to
stand up except for regularly scheduled bathroom visits and did not have any indications of the time of the
day. e experimenters continually monitored participants to keep them awake. In case of a sleep event, the
experimenters rst tried to awaken the participant through an intercom, and in case of failure, they manually
awakened the participant. We also monitored for sleep lapses through the experience sampling tasks. In case
participants closed their eyes for a time period of >30 seconds, they were probed by a tone to wake up. If they
did not, the experimenter in the room would awaken the participant.
An one-week interval took place between sleep deprivation and further recordings in order to minimize
potential carry-over eects of sleep deprivation on our follow-up conditions. In that way, the participants’
sleep schedules would also reset to their respective normal cycles. e order of the three arousal conditions
was randomized. As a post-registration note, we randomized only the order between sleep deprivation and
post-exercise, to add a training session before the baseline that allowed participants to get acquainted with the
protocol, without external task impositions, that might have confounded the protocol understanding.
Sampling plan
We used a Neyman-Pearson frequentist approach to balance false-negative and false-positive rates by setting
power to 95% and establishing a Type I error rate (alpha) of 5%. To estimate the desired sample size, a simulation
approach was utilized: data were generated consistent with a latent binomial regression model, in which one
categorical predictor with 3 levels (Base, High, Low) predicted a binary outcome Y (occurence of MB or not). An
original probability pMB = 0.1 was specied as the underlying generative probability in the baseline model based
on previous research5,11,12. We allowed the random intercepts and slopes to vary freely vary around a normal
distribution with a standard deviation of s.d. = 0.1. Given that no previous study to our knowledge has provided
evidence for the distribution of the eect sizes of arousal on mental reports, and to account for possible reverse
eects (such as decreased MB report probability), we reasoned that a meaningful yet conservative eect for the
Low Arousal condition would be an odds ratio of 1.6 and an odds ratio of 0.55 for the High Arousal condition.
Since our initial hypothesized distribution is expected to yield ~3–5 MB reports per session11,12, this eectively
translates to a small eect size of interest of at least 3 more reports across conditions.
Considering these parameters, for each population sample, ranging from 5 to 50 participants, we sampled 500
datasets, and t a binomial model with the participant ID as a random factor, keeping the regression coecients
for the levels of the predictor constant. Based on the simulation analysis, using a false positive threshold of 0.05,
we required a sample size of 26 participants to achieve a power of 0.95 (Supplementary Fig. S2).
Scientic Reports | (2025) 15:4956 4
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Data analysis
Behavioral data
Statistical analysis was performed using generalized linear mixed-eects models. To address whether arousal
aects MB occurrence, we used a binomial, linear model with arousal as a categorical independent variable,
and the proportion of mental reports across a sampling period (40 trials) as our dependent variable. Data
were binary coded (occurence or not of MB report) and t into the model using a “logit” link. Given that the
underlying distribution was unknown, a Bernoulli distribution minimized the assumptions about the model. In
order to examine whether the multinomial distribution of mental reports itself changes across dierent arousal
conditions, we used the generalized estimating equations (GEE) approach, an extension of generalized mixed-
eects models that can account for correlated, repeated-measures count data from multinomial distributions56,57.
Mental reports were aggregated as counts across participants and conditions, and we examined shis in reaction
time distribution using the three experimental arousal conditions as predictors. We considered as reaction times
the intervals between the response probe and the participant’s report. To examine reaction times as a function of
mental states, we specied a generalized linear mixed-eect model with mental reports and arousal conditions
as categorical variables and used a gamma distribution with an “inverse” link function. As reaction times are
usually an indicator of arousal eects on the task performance, an eect of arousal condition as a covariate
might be informative about a potential shi of the overall slower mental report times distribution and about
the arousal condition of the subject itself. is choice of distribution and link minimizes assumptions about
the model, respects the positive, skewed distribution of reaction times, and was previously found to provide a
better t compared to other link functions58. To examine whether arousal shis the dynamics of mental reports,
i.e. one state might be more likely to be followed by MB in one of the arousal states compared to Baseline,
we estimated dynamical transition probabilities across dierent mental states using Markov models. e
transition probabilities of MB were then compared using a linear model with an identity link, with the transition
probabilities as the dependent variable and the arousal condition as the categorical, independent variable.
All specied models were compared against null models using likelihood ratio tests. We introduced the
participant’s ID as an a-priori random factor, i.e., we allowed the model’s intercept to vary. In case we contrasted
multiple models, p-values were corrected using Bonferroni correction. In case of signicance of a xed predictor,
we used corrected pairwise comparisons to examine the marginal means of the predictors.
Brain-based measures
Physiological and cortical timeseries were segmented based on the response probe time. We considered the
110-second period before the response probe as a meaningful analysis epoch, representing the neuronal and
physiological dynamics that result in a specic mental state. is period was used in subsequent analyses.
We recorded EEG with an EasyCap (64 active electrodes) connected to an actiCHamp system (Brain
Products GmbH) using the 10–20 standard conguration. A ground electrode was placed frontally (Fpz in the
10–20 system). Online, we referenced the electrodes to a frontal electrode. Impedance was kept below 20 kΩ.
As a post-registration note, we originally registered to keep impedance below 10 kΩ. However, we decided to
leverage the strength of active electrodes to follow the research standard of 20 kΩ. To minimize impedance, we
used conductive gel. Data were sampled at a sampling frequency of 500 Hz. Preprocessing included band-pass
ltering (0.1–45 Hz, FIR lter), notch ltering (50Hz), and epoch denition (t_start = 110s preceding the probe,
t_max= probe). As a post-registration note, during EEG preprocessing we observed low-frequency (<1 Hz)
artifacts, such as sweat during the post-exercise session, that contaminated the quality of the signal. erefore,
we decided to reanalyze our data using a 1 Hz high-pass lter to minimize the presence of those artifacts. By
visual inspection, we checked and removed noisy electrodes and epochs. In case of discarding more than 50% of
the total epochs for a single participant, that participant was discarded from future analysis. We then used ICA
decomposition to remove non-neuronal components such as blinks, heartbeats, muscle artifacts, etc. Finally,
channels removed due to rejection were interpolated using neighboring channels, and all channels were re-
referenced to the average.
Based on EEG recordings, we estimated three classes of measures: (1) measures estimating spectral power—
raw and normalized power spectra, median spectral frequency (MSF), spectral edge 90 (SEF90), and spectral
edge 95 (SEF95), (2) measures estimating information content—spectral entropy, Kolmogorov-Chaitin
complexity (K) and permutation entropy, and (3) measures estimating functional connectivity—symbolic
mutualiInformation(SMI) and weighted symbolic mutual information(wSMI). Power spectrum density (PSD)
was computed over the delta (1–4 Hz), theta (4–8 Hz) alpha (8–12 Hz), beta (12–30 Hz), gamma (30–45 Hz)
spectral bands, using the Welch spectrum approximation (segments = 512 ms, overlap = 400ms). Segment
rejections were windowed using a Hanning window and zero-padded to 4096 samples. Kolmogorov-Chaitin
complexity was computed by compressing a discretization of the signal using a histogram approach with 32 bins.
Permutation entropy was obtained by computing the entropy of a symbolic transformation of the signals, within
the alpha, delta, and theta bands. SMI and wSMI were then computed from the same symbolic transformation,
but data was rst ltered using current source density estimates to diminish the volume conduction. SMI and
wSMI were computed in theta, delta, and alpha bands59. From the available connectivity metrics, we chose to
use only wSMI as it is the only one that can detect purely nonlinear interaction dynamics and can be computed
for each epoch60.
Physiological measures
Electrocardiogram (ECG) data were acquired using the BIOPAC MP160 system (BIOPAC SYSTEMS Inc.) and
the BIOPAC ECG100C amplier. e data were sampled at a sampling frequency of 2 kHz and recorded using
the AcqKnowledge v4.4 soware. ECG disposable adhesive skin electrodes were used in a bipolar arrangement
Scientic Reports | (2025) 15:4956 5
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
of two electrodes and ground. e positive electrode was at the non-dominant wrist of the participant and the
negative was on the contralateral ankle. e ground electrode was placed on the ipsilateral ankle.
ECG data were ltered with a notch lter (0.05 Hz) to remove baseline wander artifacts. A Butterworth high-
pass lter was applied (0.5 Hz) to attenuate linear dris and physiological artifacts. Powerline interference was
attenuated with a notch lter (50 Hz). Finally, the data were smoothed with a 3rd-order polynomial Savitzky-
Golay lter. Peaks were detected using the native Neurokit2 algorithm. Finally, data were epoched based on the
partition scheme in the EEG preprocessing section.
ECG metrics were grouped into three domains: time, spectral power, and information content. Time-domain
metrics were (a) heart rate (HR), (b) standard deviation of the RR intervals (SDNN), and (c) root mean square
of successive dierences (RMSSD). Spectral power features were (a) low frequency of the heart rate variability
(LF-HRV), (b) high frequency of the heart rate variability (HF-HRV), and (c) LF/HF HRV ratio. Information
content metrics were (a) approximate entropy (AE), (b) sample entropy (SE), and (c) multiscale entropy (MSE).
Initially, we used the native Neurokit2 algorithm to extract the peaks of the QRS complex. RR intervals were
estimated as the sequential dierence of the peak times. We estimated the time domain features based on the RR
timeseries. For the spectral power metrics, the RR was evenly resampled at 4 Hz. Power spectra were computed
over the LF-HRV (0.04–0.15 Hz) and the HF-HRV (0.15–0.4Hz) frequency bands. e power spectrums were
estimated using the Welch procedure.
Respiration. Respiratory data was acquired using a respiratory belt and amplied through the BIOPAC
DA100C amplier. Data were sampled at a sampling frequency of 2 kHz and recorded using the AcqKnowledge
v4.4 soware.
Respiratory metrics were grouped in the time and information content domain. Time-domain metrics were
(a) respiration rate and (b) respiration rate variability. Information content was estimated based on multiscale
entropy.
Pupillometry. Eye movements and pupil size in both eyes were recorded using oculometric glasses
(Drowsimeter R100;Phasya, S.A) with a sampling frequency of 120 Hz. e eye tracker was calibrated at the
start of each recording. Data was epoched based on the epoching scheme in the EEG preprocessing section. We
identied 100ms blink periods around blinks and removed the whole segment, as pre- and post-blink periods
can introduce pupil dilation artifacts while the eye is recovering to its standard size. We interpolated segments
using 3rd-degree cubic interpolation. Dilation speed outliers were calculated by estimating the median absolute
deviation (MAD) of each value. Samples exceeding the deviation threshold were removed. Pupil dilation was
smoothed using a moving average lter and baseline-corrected with a 100 ms period 2 s aer the probe.
Pupil metrics were grouped in the same three domains: time, spectral power, and information content. Time-
domain metrics were: (1) blink rate, (2) pupil size, and (3) pupil size variability. Spectral power metrics were: (1)
low frequency pupil component (LFC), (2) high-frequency pupil component (HFC). e information content
metric is multiscale entropy. e power spectra were estimated using the Welch procedure. As a post-registration
note, we encountered issues extracting pupil metrics at the Low Arousal condition, as participants tended to have
their eyes closed or partially closed for most of the trials. As our device was not sensitive to capture dilation in
this setting, we additionally estimated (a) blink rate, (b) blink duration, (c) blink rate variability, (d) mean eye
openness, (e) eye openness variability, (f) percentage of 70% eye closure and (g) percentage of 80% eye closure.
As stated below, our registered plan was to reliably estimate all time, frequency, and complexity metrics that can
be of use to our classiers. erefore, while we do not deviate from our original registered protocol, it is of note
that these features could not be estimated reliably.
Electrodermal activity (EDA) data was acquired through skin electrodes on the index and middle nger and
amplied through the BIOPAC EDA100C amplier. Data was sampled at a sampling frequency of 2k Hz and
recorded using the AcqKnowledge v4.4 soware. All EDA metrics originated from the time domain: (a) galvanic
skin response (GSR), (b) tonic EDA, and (c) phasic EDA. Extraction of the phasic and tonic components of
the EDA was conducted with deconvolution of the EDA signal with a biologically plausible impulse response
function with initially xed parameters that are iteratively optimized per participant61.
Pattern recognition
To examine the physiological counterpart of the behavioral shis in MB reports, we employed a supervised
decoding approach. Using the multimodal neurophysiological measurements during the three experience
sampling sessions, we trained multiple classiers to discriminate across MB, MW, and SENS reports and identify
whether MB is supported by a unique brain-body interaction pattern. is approach allowed us to extract
meaningful brain-body interactions from the proposed arousal metrics without being conservative about the
nature of the multiple comparisons between the various brain and body metrics.
As features, we opted to collect meaningful data in the time, frequency, information, and connectivity
domain, unless such measurements could not be reliably estimated within our selected time window. e goal of
the multiple selected metrics was to capture potential diverse spatiotemporal relationships (low-high frequency
interactions, phase-amplitude interactions) that might extend across dierent recording modalities. Overall, we
computed 57 features.
As targets, we used the participants’ mental states (MB, MW, and SENS). Since this creates a multiclass
classication problem, we focused on the binary classication of MB vs other reports. We expected to acquire 40
samples per participant and condition (i.e. baseline and arousal states), giving a total of 1040 (26*40) samples per
condition. We expected that 5% of the samples to correspond to the target report (MB), yielding an imbalanced
problem with only 52 target samples per condition.
As learning algorithms, we tested parametric and non-parametric models, such as Support Vector
Machines (SVM), Random Forests (RF), and Extremely Randomized Trees (ET). SVM is a classication
technique that aims to separate labeled inputs by creating a hyperplane that maximizes the distance of their
Scientic Reports | (2025) 15:4956 6
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
features. Given a set of n-labeled inputs, SVM provides a hyperplane in an n-dimensional space that maximally
separates the dierently labeled groups. An RF classier is a meta-estimator. Various classiers (“decision trees”)
are trained in dierent parts of the input dataset, and each classier uses only that part of the dataset to predict the
label of the input. en, the predictions of each classier are pooled (“bagged”) together, and an optimal decision
is chosen based on the label with the most predictions (“votes”). Finally, an ET classier is a meta-estimator that
employs a similar voting scheme. However, in the case of the ET classier, trees are trained on all the features
and the cuto point of the trees (how the various metric nodes are arranged to reach a decision) is randomized.
Since our problem is highly imbalanced, we also tested outlier detection algorithms (i.e. one-class classiers),
aiming to isolate MB from the other reports by considering MB as either an inlier or outlier.erefore, we tested
the one-class counterparts of the SVM (One-class SVM) and RF (Isolation Forests) algorithms.
For model selection and performance estimation, we employed two dierent cross-validation approaches.
First, we used a 5-fold stratied cross-validation scheme trained with all the samples. is provided us with
performance estimates of classiers aimed at obtaining patterns of brain and body function that can predict MB
reports in known participants. As a second approach, we used a 5-fold group stratied cross-validation scheme,
using participants as groups. In this scenario, each participant was either on the train or on the test set. us,
it aimed at learning general patterns of brain and body function that could predict the report of MB in unseen
participants. In other terms, the rst approach aimed at learning patterns that could discriminate MB from other
reports while accounting for each participant’s variance, while the second strengthened the claim, aiming to
learn general patterns that could be found in unseen participants.
As performance metrics, we report a) recall, b) precision, c) F1-score, d) area under the ROC curve (AUC),
and e) balanced accuracy. Recall is the ratio of how oen an item was classied correctly as a positive outcome
(True Positive/True Positive + False Negative). Similarly, precision is the ability of the model to return only the
data points in the relevant class (True Positive/True Positive + False Positive). F1-score is the harmonic mean of
precision and recall. e AUC curve is another evaluation metric that summarizes how well the classier predicts
a class based on dierent thresholds of true positive and false positive ratios. Finally, balanced accuracy is an
evaluation metric suitable for imbalanced datasets, where one class appears at signicantly dierent frequencies
than the others. Balanced accuracy is useful because it is estimated as the average of specicity and sensitivity,
simultaneously controlling for very high precision due to classifying nothing as the infrequent class and very
high recall due to classifying everything as the infrequent class.
We selected each model’s hyperparameters using nested cross-validation (same scheme as the outer cross-
validation), using the F1-score as our optimization metric.
To evaluate the variance in the classier performance and compare it to chance level, we performed repeated
cross-validation (10 times), while training also a “dummy” classier to obtain the empirical chance level of the
training samples distribution. is type of classier generates predictions based on the distribution of training
samples for each class without accounting for the features.
e decoding analysis was implemented in Python using Julearn62 and Scikit-Learn63. Metrics were estimated
from existing Python libraries: MNE64, NICE65 , Neurokit66, and custom in-lab Python functions.
Results
Participants
To achieve a power of 0.95 at an alpha threshold of 0.05, we acquired 3 sessions of 40 trials per session from 26
participants (mean age=26.38, sd= 4.53, min=20, max=40; female=11). As a post-registration note, in case
participants could not adhere to the strict 3-week protocol (30% total sessions), they were rescheduled to a later
date that respected their sleep schedules to avoid time windows with potential extreme slow-wave activity50. Due
to data corruption, one participant had 30 trials in one of the three sessions, and one participant had 33 trials in
one of the three sessions. e remaining two sessions were completed for both participants.
Behavioral data
Occurrences of mental state reports alter across arousal conditions
We found a main eect for mental states, with MB being reported at signicantly lower rates (Mean
proportions±SD: MW=0.56±0.21, SENS=0.2±.14, MB= 0.12±0.13; Kruskal H=124.07, p= 1.2e-27, eta2= 0.53)
compared to MW (Dunn’s test= -10.75, pFDR = 1.8e-26) and to SENS (Dunn’s test= -2.85, pFDR= 4.3e-03).
Additionally, MW was reported signicantly more frequently compared to SENS (Dunn’s test=7.9, pFDR= 4.3e-
15; Fig. 2). As the study was focused on wakeful mental states, “SLEEP” reports were not included in the analysis
(Mean proportions ±SD: Baseline= 0.03±.05, High Arousal= 0.05±.07, Low Arousal= 0.26±.21, Total= 0.1±.17).
We found that a model including all conditions outperformed a null model with only an intercept (FullLogLik=
−1021, NullLogLik= −1046.83, χ2= 51.57, df= 2, pBonf= 6.1e-12): MB was reported signicantly more frequently
in Low Arousal compared to Baseline (Marginal Mean= −0.79, SE= 0.14, CL= [−1.16, −0.43], pFDR= 1.8e-08)
and to High Arousal (Marginal Mean= −0.97, SE=0.15, CL= [−1.35, −0.59], pFDR= 7.9e-11) (Fig. 3a). However,
MB reports during Baseline and High Arousal were comparable (Marginal Mean= 0.17, SE= 0.15, CL= [−0.21,
0.56], pFDR= 2.4e-01). A visual inspection of the individual marginal means showed that this eect was mostly
consistent across participants and was not driven by extreme cases (Fig. 3b–d).
Additionally, generalized estimating equations (GEE) showed a signicant interaction for MW between
Low Arousal—Baseline (beta= 6, SE= 1.5, CL= [3.06, 8.94], pFDR= 6.4e-05) and Low—High Arousal (beta= 8.23,
SE=1.6, CL= [5.1, 11.36], pFDR= 2.6e-07). We also found signicant interactions in SENS reports, such that
SENS tended to be higher in Baseline compared to High (SENS Baseline—SENS High: beta= 2.54, SE= 0.81, CL=
[ 0.96, 4.12], pFDR= 1.7e-3) and Low Arousal (SENS Baseline—SENS Low: beta= 2.46, SE= 0.77, CL= [0.96, 3.97],
pFDR= 1.3e-3). It is of note that this analysis yielded no signicant results for MB, but the overall trend of the beta
estimates was consistent with our positive results of the logit model above (Supplementary Fig. S3).
Scientic Reports | (2025) 15:4956 7
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
MB was characterized by higher reaction times
ere was a main eect of arousal conditions, with reports during Baseline being reported the fastest and
during Low Arousal the slowest (Fig. 4a). Also, there was a main eect of mental states, with MW reports being
reported the fastest and MB reports the slowest (Fig. 4b). A signicant interaction between MW and arousal
showed that MW was reported the slowest in Low Arousal (Fig. 4c). A signicant interaction between MB and
arousal condition showed that MB was reported the slowest in High Arousal and Low Arousal (Fig. 4e). A model
including both arousal and reaction times outperformed simplied models including only null or main eect
terms (FullLogLik=2889.76, χ2= 47.1, df= 4, pBonf= 1.5e-09; Fig. 4c). For a detailed overview of main eects and
interactions, see Supplementary Table S3.
Transition probabilities showed reduced probability to transition to MW in Low arousal
Markov transition probabilities indicated signicant dierences only between High and Low Arousal conditions
(Fig. 5), such that MW was more likely to be followed by MB (t= 3.26, CI= [0.03,.15], pFDR= 9.7e-03, Cohen’s D=
0.74). Also in Low Arousal, both MW (t= −3.79, CI= [−0.31, −0.9], pFDR= 7.6e-03, Cohen’s D= −0.86) and SENS
Fig. 3. e frequency of mind blanking (MB) reports altered across the three arousal conditions. (a) MB
report probability increased in Low Arousal (aer sleep deprivation) compared to High Arousal (aer intense
exercise) and Baseline. Density kernels indicate overall data dispersion and clustering trends. Point plots
represent participants’ MB report probabilities. Box plots indicate medians and interquartile ranges, whiskers
indicate extreme values, and diamonds indicate data outliers. (b–d) Bar plots denote single-subject marginal
means, comparing MB reports across arousal conditions. Compared to Baseline, there was no signicant
change during High Arousal (b). However, there was a visible trend favoring an increased probability of MB
reports in the Low Arousal condition compared to baseline and High Arousal, signifying that the eect was
present in most participants (c–d).
Fig. 2. Mind blanking (MB) was reported signicantly less frequently compared to mind wandering (MW)
and perceptual sensations (SENS) across all arousal conditions, validating what is generally reported in the
literature. Density kernels show overall data dispersion and clustering trends. Point plots are individual subject
estimates. Box plots show medians and interquartile ranges, while whiskers indicate extreme values and
diamonds indicate outliers.
Scientic Reports | (2025) 15:4956 8
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
(t= −3.43, CI= [0.37, −.0 09], pFDR= 9.5e-03, Cohen’s D= −0.77) were less likely to be followed by MW (Fig. 5;
Supplementary Fig. S4).
Exploratory analysis 1: MB frequency did not correlate with SLEEP frequency
As we wanted to avoid participants confounding MB and SLEEP reports, we opted for a paradigm that allowed
participants to report both. Spearman correlations on each condition examined whether these two states were
correlated. We did not nd any signicant correlation between MB and SLEEP report probabilities across any
arousal condition. (Baseline: r= 0.13, p= 5.3e-01, High Arousal: r= 0.31, p= 1.3e-01, Low Arousal: r=−0.05,
p= 8.2e-01) (Supplementary Fig. S5). To strengthen the claim that MB and SLEEP reports do not covary, we
additionally ran separate equivalence tests on each correlation. No test was able to reject an equivalence claim
(Baseline: z= −0.34, p= 3.7e-01, High Arousal: z= 0.54, p= 7e-01, Low Arousal: z= 0.72, p = 2.3e-01). erefore,
these results remain indeterminate.
Exploratory analysis 2: High arousal:MB reports increased at the start, but not the end, of the experience
sampling session
.While we found that MB reports were more frequently in Low Arousal, we did not nd any signicant eect
of High Arousal. In our original hypothesis (Supplementary Table S1), we registered a potential alternative
explanation for the absence of an eect of high arousal in mental state report frequency.High arousal, as elicited
Fig. 4. Mental states had dierent reaction times depending on arousal conditions. (a) Reaction times at
Baseline arousal were reported the fastestlowest, followed by High (aer exercise) and Low Arousal (aer
sleep deprivation), collapsed across all arousal levels. Point plots show individual subject estimates. Box plots
show medians and interquartile ranges, while whiskers show extreme values. (b) Mind wandering (MW) was
reported the fastest, followed by Sensations (SENS) and mind blanking (MB), collapsed across all mental states.
Point plots show individual subject estimates. Boxplots show medians and interquartile ranges, while whiskers
show extreme values. (c–e) Interaction between arousal condition and mental state reaction times: MW was
reported the slowest in Low Arousal compared to Baseline and High Arousal, while MB was reported the
slowest in the Low Arousal and High Arousal conditions compared to Baseline.
Scientic Reports | (2025) 15:4956 9
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
by high-intensity exercise, might not last for the full session, and our session would represent a gradual return
to Baseline Arousal. To test for potential eects of more frequent MB reports only at the start of the experience
sampling we split the High Arousal session in two parts and compared the count of MB reports across the start
and the end of the experiment. Using a chi-squared test we found a signicant eect, with MB reports being
more frequent (divergence= 4.08, p= 3.2e-02) during the rst half of the High Arousal condition compared to
the second half (MBstart= 93, MBend= 66). We additionally attempted to validate this hypothesis by splitting
the session into 4 and 6 discrete segments of 10 and 7 trials each and replicated the same analysis. However,
this analysis did not reach signicance. Finally, to provide further evidence for reduced occurrences of MB
across time, we considered only the rst and last 10 trials. We found a signicant eect of more frequent MB
occurrences (divergence= 7.39, p= 6.6e-03), with the rst 10 trials of the High Arousal condition inducing more
MB compared to the second half (MBstart= 51, MBend= 27).
Classication of MB reports was outperformed by classication containing both BRAIN-BODY markers
We trained a cohort of dierent classication algorithms and evaluated their capacity to classify MB reports from
mental states with content (MW, SENS) based on 26 BRAIN (EEG) and 31 BODY features (12 ECG, 4 EDA, 8
RSP, 7 EYE), spanning time, frequency, information, and connectivity domains for each mental state report. In
our original report, we registered that these features would be estimated across the 110s pre-probe window, with
bad epochs being dropped. However, across an 110s epoch, even a nonlinearity of 1s would result in epoch
removal, leaving a total clean sample of 25 / 78 sessions (29.4%), and a total of 1060/3120 (33.3%) clean epochs.
erefore, to preserve datapoints and data quality and minimize data discarding due to brief non-linearities, we
opted for an extra step in bad epoch removal. Aer the initial epoch denition of 110s, we followed it up by
partitioning that epoch into 5s sub-epochs, resulting in 22 sub-epochs per epoch. We then proceeded to do bad
epoch removal and EEG marker estimation on those sub-epochs. If an epoch consisted of more than 50% bad
sub-epochs, it was discarded. en, we averaged across within each epoch, resulting in no lost sessions, and a
total of 2734 / 3120 (87.6%) total sample size.
Having a nal 2734 reports x 57 features matrix per report, we trained multiple classiers on the total dataset,
to examine whether a specic brain-body prole would outperform chance level classication of MB reports
(Table 1).
Due to the unbalanced nature of our dataset, we evaluated classier performance based on balanced
accuracy, as it avoids inated performance estimates on unbalanced datasets. Overall, we found that a balanced
random forest (a random forest that undersamples the majority class in each bootstrap to equate class count) had
above-chance performance and outperformed all other examined classiers (Fig. 6a). We additionally examined
whether we could predict unknown subjects, by leaving a subset of subjects out on each iteration. Due to the
high degree of per-fold variance, we do not consider any classier as meaningfully performing above chance
level (Fig. 6b). Importantly, these results were replicated when we trained the classiers in the 1Hz ltered data
(Supplementary Fig. S6a,b; Supplementary Table S4).
Having established that MB reports can be predicted from known subjects, we then examined whether a
brain-body data pattern would outperform classiers trained solely on either BRAIN or BODY features. To this
end, we t and optimized a separate balanced random forest classier on discrete feature combinations of our
dataset. For a full report of the performance on dierent features, see Table 2 and Supplementary Table S5.
Overall, we found that a classier trained on both BRAIN and BODY markers numerically outperformed
classiers trained solely on BRAIN or BODY features across all our performance metrics (Fig. 7a,c; Supplementary
Fig. S7a,c; Table 2; Supplementary Table S5). To evaluate the impact of the number of features on the capacity of
Fig. 5. Aer sleep deprivation (Low Arousal), participants were more likely to transition from mind wandering
(MW) to mind blanking (MB) compared to the condition of physical exercise (High Arousal). Additionally,
participants were less likely to transition to MW, either when departing from reports about sensory perception
(SENS) or from MW itself. Arrows indicate the direction of the mental state transition. Bold font indicates
statistical signicance (FDR corrected).
Scientic Reports | (2025) 15:4956 10
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
the learning algorithm to extract relevant information, we also trained the balanced random forest model using
randomly shued bodily features. EEG features were not altered. e model with the shued values showed a
decline in classication performance, providing evidence that, when classifying mental states, a model trained
on both brain and body data learns unique information from both domains (Fig. 7d; Supplementary Fig. 7d).
For feature importance, we calculated Shapley Additive Explanations(SHAP) values for each feature in our
dataset. SHAP values estimate the marginal contribution of each feature, averaged across every potential feature
combination. In this manner, each value represents how much this feature contributes to the classication, aer
controlling for the impact of other features on this feature’s importance. We found that the model relied mostly
on EEG and EYE openness features to discriminate MB reports when pooling MB occurrences across all three
conditions. (Fig. 7b; For an extensive list of all SHAP values, see Supplementary Fig. S8). Importantly, feature
importance did not substantially change when ltering the data with a 1 Hz lter (Supplementary Fig. S7b; For
an extensive list of all SHAP values, see Supplementary Fig. S9). Overall, the comparable performance of the
Fig. 6. Classication performance was above chance level when mind blanking (MB) reports were pooled
across subjects, but not aer training on a subset of participants and classifying the remaining subset. (a)
A balanced random forest classier provided the highest classication performance across all examined
classiers including known subjects. (b) A balanced random forest classier provided the highest classication
performance across all examined classiers on unknown samples. However, due to the high variance, we
could not consider it meaningful. Individual points indicate performance on the folds of the repeated cross-
validation. Results are ordered based on descending order of performance. Chance-level performance is
indicated by the Dummy classier. RF = random forest; SVM = support vector machine; ET = extreme trees; IF
= isolation forest; OC SVM = one-class support vector machine.
Examined Classier Recall Precision F1 ROC AUC Balanced accuracy
Known subjects
Balanced RF 0.62, [0.6, 0.64] 0.26, [0.26, 0.27] 0.37, [0.36, 0.37] 0.71, [0.7, 0.72] 0.66, [0.65, 0.67]
SVM 0.29, [0.28, 0.31] 0.28, [0.27, 0.29] 0.29, [0.27, 0.3] 0.62, [0.61, 0.63] 0.58, [0.58, 0.59]
ET 0.16, [0.15, 0.17] 0.61, [0.58, 0.64] 0.25, [0.23, 0.26] 0.73, [0.72, 0.74] 0.57, [0.56, 0.58]
RF 0.14, [0.13, 0.15] 0.57, [0.53, 0.6] 0.22, [0.21, 0.23] 0.71, [0.7, 0.72] 0.56, [0.56, 0.56]
IF 0.14, [0.13, 0.16] 0.2, [0.19, 0.22] 0.17, [0.15, 0.18] 0.52, [0.52, 0.53] 0.52, [0.52, 0.53]
OC SVM 0.89, [0.86, 0.92] 0.15, [0.14, 0.15] 0.25, [0.25, 0.25] 0.51, [0.5, 0.51] 0.51, [0.5, 0.51]
DUMMY 0.14, [0.13, 0.15] 0.14, [0.13, 0.15] 0.14, [0.13, 0.15] 0.5, [0.49, 0.5] 0.5, [0.49, 0.5]
Unknown Subjects
Balanced RF 0.46, [0.41, 0.51] 0.18, [0.16, 0.2] 0.25, [0.23, 0.27] 0.55, [0.53, 0.57] 0.54, [0.53, 0.56]
IF 0.23, [0.19, 0.27] 0.18, [0.16, 0.2] 0.19, [0.17, 0.22] 0.53, [0.51, 0.54] 0.53, [0.51, 0.54]
RF 0.05, [0.04, 0.06] 0.36, [0.29, 0.44] 0.08, [0.06, 0.09] 0.54, [0.52, 0.55] 0.51, [0.51, 0.52]
OC SVM 0.87, [0.82, 0.92] 0.14, [0.13, 0.15] 0.24, [0.22, 0.26] 0.51, [0.5, 0.52] 0.51, [0.5, 0.52]
ET 0.03, [0.02, 0.03] 0.36, [0.26, 0.45] 0.05, [0.04, 0.06] 0.53, [0.52, 0.55] 0.51, [0.5, 0.51]
DUMMY 0.15, [0.14, 0.16] 0.15, [0.13, 0.17] 0.14, [0.13, 0.16] 0.5, [0.49, 0.51] 0.5, [0.5, 0.51]
SVM 0.2, [0.17, 0.22] 0.16, [0.14, 0.17] 0.16, [0.15, 0.17] 0.49, [0.47, 0.5] 0.5, [0.49, 0.51]
Tab le 1. A balanced random forest classier outperformed all classiers when compared across balanced
accuracy. Cells indicate mean and 95% CI. RF = Random Forest; SVM = Support Vector Machine; ET =
Extreme Trees; IF = Isolation Forest; OC SVM = One-Class Support Vector Machine
Scientic Reports | (2025) 15:4956 11
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
models, and the high degree of overlap in the ranking of the feature importance point to the robustness of the
models.
Exploratory analysis 3: Feature importance altered across arousal conditions.
e decoding analysis in known samples showed that we can predict MB instances from the combination of brain-
body markers with adequate accuracy when MB instances were aggregated across dierent arousal conditions.
Fig. 7. Classication of MB improves when considering both BRAIN and BODY. (a) Balanced random forest
classier trained on a combination of BRAIN and BODY features outperformed classiers trained solely on
BRAIN or BODY features when evaluated with balanced accuracy. Individual points indicate performance on
the folds of the repeated cross-validation. (b) Subset of the 10 features with the highest mean of the absolute
SHAP values obtained from the balanced random forest classier. (c) e per-fold dierences between the
classier trained on both BRAIN and BODY features and the one trained only on BRAIN data suggest that
incorporating both feature domains provides a slight performance improvement over using BRAIN data alone.
e shaded region indicates better performance for the classier trained on both feature domains. e star
indicates the mean dierence. e solid, horizontal line represents the 95% highest-density intervals of the
distribution. Red dots indicate per-fold dierences. (d) e per-fold dierences between the classier trained
on both BRAIN and BODY features and the one trained on BRAIN and shued BODY data suggest that the
model with both BRAIN and BODY data does not consider the body markers as noise.
Classier Recall Precision F1 ROC AUC Balanced Accuracy
BRAIN + BODY 0.62, [0.6, 0.64] 0.26, [0.26, 0.27] 0.37, [0.36, 0.37] 0.71, [0.7, 0.72] 0.66, [0.65, 0.67]
BRAIN 0.61, [0.59, 0.62] 0.24, [0.24, 0.25] 0.35, [0.34, 0.36] 0.7, [0.69, 0.71] 0.65, [0.64, 0.65]
BODY 0.59, [0.58, 0.6] 0.22, [0.21, 0.22] 0.32, [0.31, 0.32] 0.66, [0.66, 0.67] 0.61, [0.61, 0.62]
EYE 0.57, [0.55, 0.59] 0.21, [0.21, 0.22] 0.31, [0.3, 0.32] 0.64, [0.63, 0.65] 0.61, [0.6, 0.62]
ECG 0.55, [0.54, 0.57] 0.18, [0.17, 0.18] 0.27, [0.26, 0.27] 0.58, [0.57, 0.59] 0.56, [0.55, 0.57]
EDA 0.6, [0.57, 0.63] 0.17, [0.17, 0.17] 0.26, [0.26, 0.27] 0.57, [0.56, 0.58] 0.55, [0.54, 0.56]
RSP 0.52, [0.5, 0.54] 0.15, [0.15, 0.16] 0.24, [0.23, 0.24] 0.53, [0.52, 0.54] 0.52, [0.51, 0.53]
Tab le 2. A classier trained on a combination of BRAIN and BODY features outperformed classiers trained
solely on BRAIN or BODY features, when evaluated with balanced accuracy. Cells indicate mean and 95% CI.
Scientic Reports | (2025) 15:4956 12
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
We were further interested in whether this classication was achieved based on a universal mechanism, or
whether we could detect arousal-dependent brain-body congurations that predict MB. To this end, we trained
a balanced random forest classier solely on data acquired from Baseline, from High, and from Low Arousal. We
found that Baseline had the best performance (0.67, [0.65, 0.68]), followed by Low Arousal (0.64, [0.63, 0.65]),
and nally High Arousal (0.61, [0.6, 0.63]). We retained comparable performance when examining the arousal
partitions of the 1 Hz ltered dataset (Supplementary Table S6-7). Examining the SHAP values for each arousal
state, we saw that the models relied on distinct feature domains. During Baseline, the model relied on markers
from the frequency domain of EEG (Fig. 8a). During Low Arousal, MB classication was obtained using the delta
band power, by far the most dominant marker (Fig. 8b). Finally, in High Arousal, the model did not rely on a
single feature, rather in a combination of eye openness, GSR, and the frequency domain of EEG (Fig. 8c). Similar
feature importances were observed in the 1Hz ltered dataset (Supplementary Fig. S10). However, in the 1 Hz
ltered dataset, we observed that ECG features tended to rank higher (Supplementary Fig. S11–16).
Exploratory analysis 4: Feature importance altered based on the pre-probe analysis window
A potential caveat of utilizing the full pre-probe period of 110 s before a report is that we might capture
multiple mental states, and the actual statistical regularities might be weakened when averaged across. With this
consideration, we examined whether we could improve classication performance when classifying MB from
the last 10s before a report. We dened a secondary brain-body data matrix, with BODY features that could be
estimated from 10s of body activity. Across both 0.1 and 1 Hz lters we retained comparable performance in
the classiers trained on both EEG and bodily markers, as well as solely EEG or BODY markers (Supplementary
Fig. S17–20; Supplementary Table S8 and 9). However, we observed decreased performance in the classier
trained solely in the eye openness data (Supplementary Table S8 and 9). An examination of feature importance
showed that the beta, delta, and theta bands of the EEG frequency domains remained the most important EEG
features, but there was a reduction in the importance of the EYE features and an increase in the importance of
EDA (Supplementary Fig. S17b, 18, 19b, 20). Importantly, our results were not aected by the choice of ltering
parameters, indicating robustness of our results to preprocessing parameters.
Discussion
We used experience sampling combined with EEG and peripheral physiological recordings under dierent
autonomic arousal conditions to determine whether MB reports in neurotypical individuals were supported by
distinct brain-body congurations compared to mental states with reportable content. Overall, our results show
that MB is a mental state that becomes more prevalent in low and partially in high arousal states, and that MB is
driven by both brain and body processes, providing evidence for an embodied account of MB.
Behaviorally, we found that MB was reported at signicantly lower rates compared to sensory experiences
or MW, irrespective of the arousal condition. is nding is in line with past research showing that MB rates
vary between 5 and 10% of total probe instances, across both uninterrupted thinking12 and task engagement11.
We also show that sleep deprivation signicantly increased the frequency of MB occurrences. Sleep deprivation
Fig. 8. Ranking of features by mean absolute SHAP value extracted from the balanced random forest classier
varied across dierent arousal conditions. (a) Magnitude of SHAP values for a balanced random forest
classier trained on MB reports collected during the Baseline Arousal condition. e model relied mostly
on features from the EEG frequency domain. (b) Magnitude for SHAP values for a classier trained on MB
reports collected during the Low Arousal condition (aer sleep deprivation). e model mostly used spectral
power in the EEG delta band. (c) Magnitude for SHAP values for a classier trained on MB reports collected
during the High Arousal condition (aer intense exercise). e model relied mostly on features from eye
openness, EDA, and the EEG frequency domain
Scientic Reports | (2025) 15:4956 13
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
has been shown to induce a low arousal state during which cognitive performance declines67, metabolic and
physiological processes change68, and unique neuronal markers, such as slow-wave activity, emerge69. Aer sleep
deprivation, participants also tend to perform worse in sustained attention tasks70, with results suggesting a true
eect of sleep deprivation on more “misses” (no response when necessary) compared to “false alarms” (response
when unnecessary)71, a nding that was recently shown as a behavioral correlate of MB11. Additionally, sleep
deprivation and mounting sleep pressure have been positively correlated with more MW instances72,73, suggesting
an overall mode shi from task engagement to MW74. Our results challenge these past ndings by showing that
participants were more likely to experience an MB event rather than MW aer sleep deprivation. We also found
that MW was in fact more likely to decrease aer sleep deprivation. is is further supported by the results of
the transition analysis, where MW reports were less likely to be followed by another MW report, and more likely
to be followed by MB. Such discrepancies in the reportability of MW aer sleep deprivation could be possibly
explained by the explicit inclusion of MB as a reportable mental state in the experience sampling that our design
opted for. In other words, it might be that the observed MW occurrence increases aer sleep deprivation can
be accounted for by MB reports, once participants have the chance to opt between these two mental states in a
more ne-grained way.
In terms of high arousal induced by high-intensity exercise, our analysis did not reveal any signicant eects
on MB occurrences. As per the provided registered protocol alternative explanation (Supplementary Table 1), we
hypothesized that this arousal manipulation might not have been overall eective as it could not have produced
eects that would last across the whole experience sampling session. To test whether MB frequency reports
would dier between the beginning and at the end of the session, we split the dataset into two parts. When
split, we indeed found a signicant dierence between the frequency of MB reports. is result was replicated
when considering only the rst and last 10 trials per subject, which maximized the distance between initial and
nal physiological arousal within the session. However, we were not able to nd any dierences when the data
were split into smaller bins. Together, we consider that these results provide partial evidence for our registered
hypothesis, showing that residual high arousal eects aer intense exercise can increase the frequency of MB
reports.
In addition to the frequency of mental states across arousal conditions, we also examined whether rreaction
times dier across arousal conditions and mental states. In general, reports in low arousal tended to be the slowest,
consistent with a wide range of attention tasks that show slower reaction times in sleep deprivation compared to
baseline arousal75. We consider these ndings as additional evidence that the arousal manipulation was eective
in that it lowered overall vigilance levels of the . We also observed a main eect of mental states, such that MB
tended to be reported signicantly slower compared to MW and sensations. Contrary to our current results,
we recently found that MB was reported faster when compared to other mental states when content had to be
evaluated12. is apparent mismatch in results can be explained when considering that MB can be a state devoid
of content, and therefore, there is the binary consideration of “yes/no” when evaluating thought content, which
might be a relatively fast decision. is can be dierent, for example, from the evaluation of content-full mental
states, which demand a sequential evaluation of both content presence (“yes/no”) and content evaluation (“what
is the content about?”). is way, the dierence in results can be explained by the imposition of an additional
cognitive evaluation. Overall, we suggest that these results might reect a gradient of vigilance, with participants
being the most alert at baseline arousal, and progressively declining during high and low arousal conditions,
as well as more vigilant when reporting mental states with content compared to MB. Of note, we observed two
interesting interactions between mental states and arousal conditions. MW tended to be reported slower in low
arousal compared to baseline and high, which is consistent with our interpretation of reaction times as marking
vigilant states. However, as we also observed that MB reports tended to be reported slower in both high and low
arousal conditions, we speculate that this might be preliminary evidence that arousal modulates how engaged
participants are with their current mental states. In this sense, exercise fatigue can lead to an MB state that takes
longer to recover from when probed for a report.
A nal explanatory analysis revolved around the relationship between sleep and MB. We recently posited that
MB is a distinct mental state characterized by a unique phenomenological prole of no content76, and unique
neuronal markers, characterized by high cortical integration and low cortical segregation12. is neuronal
conguration is atypical of wakefulness77, and is more closely reminiscent of brain congurations during deep
sleep78. In conjunction with the presence of slow wave intrusions during wakefulness as a marker of MB reports11,
a classic marker of NREM sleep, an emerging issue is whether MB is a misrepresented instance of sleep. is
issue is further complicated by the postulation that in MB there is no content76, and thus does not functionally
represent a wakeful state where people can report content. To avoid this pitfall, we introduced sleep as a potential
report during experience sampling. We found that people discretely reported MB and sleep, providing evidence
that when provided with such options, people can dierentiate between these two experiences. Additionally,
we did not nd that MB and sleep tended to covary. To strengthen this claim, we ran equivalence tests for each
correlation across arousal conditions. However, no test showed a positive result for equivalence. erefore, these
results remain indeterminate, with a trend for no relationship between MB and sleep.
Having established that MB occurrence varied across dierent physiological arousal conditions, we then
examined whether MB could be decoded by brain and body markers. With the aim of showing single trial
prediction, we trained dierent models on EEG and physiological markers, spanning time, spectral, complexity,
and connectivity domains. Overall, we were able to achieve above-chance-level classication, showing that there
exist unique brain-body patterns that can discriminate MB reports from mental states with content. However,
we were not able to show above-chance-level classication when training classiers on unknown subjects.
erefore, our results are not generalizable to novel populations due to the high amount of variance between
subjects. Of importance is the result that a combination of EEG and physiological markers marginally, but
consistently, outperformed both EEG and physiological markers. Overall, we observed an improvement of 2–5
Scientic Reports | (2025) 15:4956 14
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
% in classication performance in balanced accuracy. is improvement can be attributed to unique information
inherent in body signals, as evidenced by the comparison of the classier trained on both brain and body data
compared to classiers trained solely on brain data or brain and shued body data. e classier trained on
both brain and body data does not consider body features as noise or redundant. Overall, while our results
suggest a high degree of overlap between brain and body information in MB, they indicate that information
about MB extracted from the body is partially independent of the EEG features. Feature importance ranking
derived from the classication model indicates that the low and mid frequencies of the EEG power spectrum and
metrics of eye openness are useful predictors of MB. is nding was consistent across analysis windows and
preprocessing parameters. Importantly, all classiers trained on body markers had above chance performance
with variant degrees of variability, with the highest performing being the EYE (eye openness) and the ECG
(heart-rate variability), providing evidence that MB can be decoded solely from bodily signals.
To further validate our protocol, we ran two exploratory analyses, with the aim to examine whether
classication performance varies based on the analyzed pre-probe window and whether feature importance alters
across arousal conditions (For a full Discussion, see the Supplementary Discussion on Methodology). Overall,
when examining a classier trained on a brief 10s window before MB reports, we found similar performance
compared to the full 110s classier. What was interesting was that, while EEG performance remained the same,
performance on classiers trained solely on body features decreased. As brain-physiology coupling occurs at
varying time delays across cardiac79 and respiratory domains80, we interpret these results as evidence that bodily
contributions on MB are based on slow, oscillatory processes that might not be captured from examining short
pre-probe periods. At the same time, our classication analysis on separate arousal conditions showed distinct
brain-body congurations that can predict MB reports. As our decoding approach did not permit any inference
of the directionality eect, or decomposing interactions within and across physiology modalities, at this stage we
claim that our results point to discrete physiological pathways that elicit MB reports. Overall, we show that our
enhanced classication is retained across dierent analysis windows and dierent arousal conditions.
Similarly, enhanced classication when considering a brain-heart matrix compared to solely brain markers
was also shown for patients with disorders of consciousness, where the inclusion of cardiac features outperformed
classication based solely on EEG markers81. To our knowledge, our results are the rst to extend multivariate
decoding past the brain-heart axis and consider the inclusion of multiple unique bodily aerent sources in
classifying mental states. e overall success of the brain-body decoding paradigm in classifying consciousness
levels and mental states provides evidence that bodily information is not redundant and is not necessarily fully
represented within brain dynamics. Instead, an embodied approach, stressing bidirectional information routes
between brain and body can provide better predictive power and assist in more comprehensive, generative
models of experience34,82.
A neurobiological explanation of our results comes from an integrative model of content, task engagement,
and arousal which suggests that the relationship between thought and arousal can be conceptualized as an
inverted u-curve. is means that an optimal arousal level modeled by the locus coerulius-norepinephrine
(LC-NE) release is necessary to actively engage and control our thoughts, either during task engagement
or MW83. is stance treats thought as an active task, where engagement is necessary for clear content and
control of thought dynamics. As arousal tapers o to non-optimal levels of the inverted U-curve, we experience
concurrent, opposing thoughts that serve exploratory purposes for optimal performance, such as exploring
dierent strategies. is necessitates exibility and malleability of content. We here suggest that our results
supplement this model by providing an account of the extremities of the optimal U-curve. As the model suggests
degradation of thought clarity when we move closer to arousal extremities, we consider MB reports as instances
where no content can be clear or present, extending this unifying framework to the entire arousal U-curve.
Neurophysiologically, this model has translated to investigations of pupil dilation as a function of mental state
and task engagement with pupil size yielding both positive26,84 and null results11 in discriminating on-task
vs o-task mental states, as well as contrasting MB and MW. Part of the ascending arousal network, the LC
modulates cardiac, galvanic, respiratory, and pupillary activity28,85. In addition, the LC innervates projections
responsible for eyelid openness86. e combinatorial high performance of dierent body markers in classifying
MB reports, and the evidence that altered levels of arousal increase MB occurrences provide further support for
the modulatory role of the ascending arousal system in mental states and thought reportability.
From a theoretic perspective, our study challenges the conception that brain information is uniquely suitable
to understand thought reportability and provides support for an embodied account of the mind. Embodiment
moves the seat of mental events away from the brain and reformulates cognition as resulting from brain-body
interactions. An extensive literature has shown how cataloged cardiac, respiratory, gut, and pupillary eects
on perception30, action87, metacognition31 and consciousness81, while the collective interplay of peripheral
systems has discriminatory power for clinical88 and consciousness classication89. We show here that within
embodiment, the body is not only facilitatory, but also might impede access to our mental lives. Under specic
brain-body congurations, we are not able to clearly formulate mental content.
Some limitations pertain to our study. First, the nature of experience sampling discretizes the continuous
nature of ongoing thinking. As there is no consensus on how long a mental state might last, or whether all mental
states last the same length, results might average across dierent mental states. While we attempted to circumvent
this problem by analyzing dierent pre-probe windows, it remains unclear whether all mental states last the
same, and what is their actual duration. Secondly, the post-exercise setup might be suboptimal in examining the
eects of high arousal on ongoing cognition. Neuronal and electrophysiological recordings have shown that the
duration of the eects of exercise on ongoing brain and physiological activity45–47 is highly variant. In addition, it
is unclear whether brain and body recover to baseline states at the same rates, potentially confounding the post-
exercise importance of cortical and physiological markers in cognition. Experience sampling with online probes
during exercise could overcome this challenge.
Scientic Reports | (2025) 15:4956 15
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
In conclusion, our study suggests that MB is an arousal-modulated mental state, with a unique cortical and
physiological prole. We think that our results pave a new paradigm for an embodied account of mental states,
where the phenomenology of our mental lives is expressed based on both our body and our brain state. At the
same time, our results challenge the neurocentric approach to mental state research, putting emphasis on the
constant brain-body interactions that shape our cognition. As MB research continues to evolve, we consider our
ndings elaborative for clinical and experimental accounts of spontaneous thinking, where we move towards a
complex and dynamic conception of our mind.
Data availability
e aggregated raw data in a BIDS format, the trained machine-learning models, experimental and analysis logs,
and result dataframes can be found at https://doi. org/ht tps://doi.or g/10.58119/ ULG/174Q6G.
Code availability
All codes to replicate the power analysis, the experience sampling paradigm, and the present analysis can be
found at h t t p s : / / g i t l a b . u l i e g e . b e / P a r a d e i s i o s . B o u l a k i s / m i n d _ b l a n k i n g _ a r o u s a l. An archived version of the code
at the time of submission can be found at https: //d oi. org/h ttp s:/ /d oi.or g/ 1 0.58119/ULG/174Q6G.
Received: 11 August 2022; Accepted: 25 November 2024
References
1. Christo, K., Irving, Z. C., Fox, K. C. R., Spreng, R. N. & Andrews-Hanna, J. R. Mind-wandering as spontaneous thought: A
dynamic framework. Nat. Rev. Neurosci. 17, 718–731 (2016).
2. Van Calster, L., D’Argembeau, A., Salmon, E., Peters, F. & Majerus, S. Fluctuations of attentional networks and default mode
network during the resting state reect variations in cognitive states: Evidence from a novel resting-state experience sampling
method. J. Cogn. Neurosci. 29, 95–113 (2017).
3. Vanhaudenhuyse, A. et al. Two distinct neuronal networks mediate the awareness of environment and of self. J. Cogn. Neurosci. 23,
570–578 (2011).
4. Smallwood, J. & Schooler, J. W. e science of mind wandering: Empirically navigating the stream of consciousness. Annu. Rev.
Psychol. 66, 487–518 (2015).
5. Ward, A. & Wegner, D. Mind-blanking: When the mind goes away. Front. Psychol. 4, 650 (2013).
6. Van den Driessche, C. et al. Attentional lapses in attention-decit/hyperactivity disorder: Blank rather than wandering thoughts.
Psychol. Sci. 28, 1375–1386 (2017).
7. Kawagoe, T., Onoda, K. & Yamaguchi, S. Dierent pre-scanning instructions induce distinct psychological and resting brain states
during functional magnetic resonance imaging. Eur. J. Neurosci. 47, 77–82 (2018).
8. Kawagoe, T., Onoda, K. & Yamaguchi, S. e neural correlates of ‘mind blanking’: When the mind goes away. Hum. Brain Mapp.
40, 4934–4940 (2019).
9. Winter, U. et al. Content-free awareness: EEG-fcMRI correlates of consciousness as such in an expert meditator. Front. Psychol. 10,
3064 (2020).
10. Andrillon, T. et al. Does the mind wander when the brain takes a break? local sleep in wakefulness, attentional lapses and mind-
wandering. Front. Neurosci. 13, 949 (2019).
11. Andrillon, T., Burns, A., Mackay, T., Windt, J. & Tsuchiya, N. Predicting lapses of attention with sleep-like slow waves. Nat.
Commun. 12, 3657 (2021).
12. Mortaheb, S. et al. Mind blanking is a distinct mental state linked to a recurrent brain prole of globally positive connectivity
during ongoing mentation. Proc. Natl. Acad. Sci. U.S.A 119, e2200511119 (2022).
13. Schooler, J. W., Reichle, E. D. & Halpern, D. V. Zoning out while reading: evidence for dissociations between experience and
metaconsciousness. in inking and seeing: visual metacognition in adults and children 203–226 (MIT Press, 2004).
14. Wong, C. W., Olafsson, V., Tal, O. & Liu, T. T. e amplitude of the resting-state fMRI global signal is related to EEG vigilance
measures. Neuroimage 83, 983–990 (2013).
15. Chang, C. et al. Tracking brain arousal uctuations with fMRI. Proc. Natl. Acad. Sci. U.S.A 113, 4518–4523 (2016).
16. Wong, C. W., Olafsson, V., Tal, O. & Liu, T. T. Anti-correlated networks, global signal regression, and the eects of caeine in
resting-state functional MRI. Neuroimage 63, 356–364 (2012).
17. Vyazovskiy, V. V. et al. Local sleep in awake rats. Nature 472, 443–447 (2011).
18. Beri, D. & K, J. R. Physiological correlates of arousal: A metaanalytic review. J. Neurol. & Neurosci. 10, (2019).
19. Ogilvie, R. D. e process of falling asleep. Sleep Med. Rev. 5, 247–270 (2001).
20. Lacey, J. L. Somatic response patterning and stress: some revisions of activation theory. in Psychological stress (eds. Appley, M. H.
& Trumbell, R.) 14–37 (Appleton-Century-Cros, 1967).
21. Jones, B. E. Arousal systems. Front. Biosci. 8, s438-451 (2003).
22. Cantero, J. L., Atienza, M., Salas, R. M. & Gómez, C. M. Alpha EEG coherence in dierent brain states: An electrophysiological
index of the arousal level in human subjects. Neurosci. Lett. 271, 167–170 (1999).
23. Pinggal, E., Dockree, P. M., O’Connell, R. G., Bellgrove, M. A. & Andrillon, T. Pharmacological manipulations of physiological
arousal and sleep-like slow waves modulate sustained attention. J. Neurosci. 42, 8113–8124 (2022).
24. Perogamvros, L. et al. e phenomenal contents and neural correlates of spontaneous thoughts across wakefulness, nrem sleep,
and rem sleep. J. Cogn. Neurosci. 29, 1766–1777 (2017).
25. Stawarczyk, D., François, C., Wertz, J. & D’Argembeau, A. Drowsiness or mind-wandering? Fluctuations in ocular parameters
during attentional lapses. Biol. Psychol. 156, 107950 (2020).
26. Unsworth, N. & Robison, M. K. Tracking arousal state and mind wandering with pupillometry. Cogn. Aect. Behav. Neurosci. 18,
638–664 (2018).
27. Varela, F. J., ompson, E. & Rosch, E. e embodied mind: cognitive science and human experience. (MIT Press, 2017).
28. Wang, C.-A. et al. Arousal eects on pupil size, heart rate, and skin conductance in an emotional face task. Front. Neurol. 9, (2018).
29. Al, E. et al. Heart–brain interactions shape somatosensory perception and evoked potentials. Proc. Natl. Acad. Sci. U.S.A 117,
10575–10584 (2020).
30. Park, H.-D., Correia, S., Ducorps, A. & Tallon-Baudry, C. Spontaneous uctuations in neural responses to heartbeats predict visual
detection. Nat. Neurosci. 17, 612–618 (2014).
31. Garnkel, S. N. et al. What the heart forgets: cardiac timing inuences memory for words and is modulated by metacognition and
interoceptive sensitivity. Psychophysiol. 50, 505–512 (2013).
Scientic Reports | (2025) 15:4956 16
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
32. Garnkel, S. N. et al. Fear from the heart: sensitivity to fear stimuli depends on individual heartbeats. J. Neurosci. 34, 6573–6582
(2014).
33. Allen, M. et al. Unexpected arousal modulates the inuence of sensory noise on condence. eLife 5, e18103 (2016).
34. Allen, M., Levy, A., Parr, T. & Friston, K. J. In the body’s eye: the computational anatomy of interoceptive inference. PLoS Comput.
Biol. 18, e1010490 (2022).
35. Lee, K. et al. Arousal impacts distributed hubs modulating the integration of brain functional connectivity. Neuroimage 258,
119364 (2022).
36. Buysse, D. J., Reynolds, C. F., Monk, T. H., Berman, S. R. & Kupfer, D. J. e pittsburgh sleep quality Index: A new instrument for
psychiatric practice and research. Psychiatry Res. 28, 193–213 (1989).
37. Koenig, J., Jarczok, M. N., Fischer, J. E. & ayer, J. F. e association of (eective and ineective) analgesic intake, pain interference
and heart rate variability in a cross-sectional occupational sample. Pain Med. 16, 2261–2270 (2015).
38. Spitzer, R. L., Kroenke, K., Williams, J. B. W. & Löwe, B. A brief measure for assessing generalized anxiety disorder: e GAD-7.
Arch. Intern. Med. 166, 1092–1097 (2006).
39. Siess, J., Blechert, J. & Schmitz, J. Psychophysiological arousal and biased perception of bodily anxiety symptoms in socially
anxious children and adolescents: A systematic review. Eur. Child Adolesc. Psychiatry 23, 127–142 (2014).
40. Seli, P., Risko, E. F., Smilek, D. & Schacter, D. L. Mind-wandering with and without intention. Trends Cogn. Sci. 20, 605–617 (2016).
41. Seli, P. et al. Mind-wandering as a natural kind: A family-resemblances view. Trends Cogn. Sci. 22, 479–490 (2018).
42. Smallwood, J. Distinguishing how from why the mind wanders: A process–occurrence framework for self-generated mental
activity. Psychol. Bull. 139, 519–535 (2013).
43. Fox, K. C. R., Spreng, R. N., Ellamil, M., Andrews-Hanna, J. R. & Christo, K. e wandering brain: Meta-analysis of functional
neuroimaging studies of mind-wandering and related spontaneous thought processes. Neuroimage 111, 611–621 (2015).
44. Smallwood, J. et al. e neural correlates of ongoing conscious thought. iScience 24, 102132 (2021).
45. Gourine, A. V. & Ackland, G. L. Cardiac vagus and exercise. Physiology 34, 71–80 (2019).
46. Lin, M.-A. et al. Resistance-induced brain activity changes during cycle ergometer exercises. BMC Sports Sci. Med. Rehabil. 13, 27
(2021).
47. Basso, J. C. & Suzuki, W. A. e eects of acute exercise on mood, cognition, neurophysiology, and neurochemical pathways: a
review. Brain Plast. 2, 127–152 (2017).
48. Gorgoni, M. et al. Topographic electroencephalogram changes associated with psychomotor vigilance task performance aer sleep
deprivation. Sleep Med. 15, 1132–1139 (2014).
49. Borbély, A. A., Daan, S., Wirz-Justice, A. & Deboer, T. e two-process model of sleep regulation: a reappraisal. J. Sleep Res. 25,
131–143 (2016).
50. Lazar, A. S., Lazar, Z. I. & Dijk, D.-J. Circadian regulation of slow waves in human sleep: Topographical aspects. Neuroimage 116,
123–134 (2015).
51. Birchler-Pedross, A. et al. Subjective well-being is modulated by circadian phase, sleep pressure, age, and gender. J. Biol. Rhythms
24, 232–242 (2009).
52. Sagaspe, P. et al. Inuence of age, circadian and homeostatic processes on inhibitory motor control: A go/nogo task study. PLOS
ONE 7, e39410 (2012).
53. Cajochen, C. et al. Dynamics of frontal low EEG-activity and subjective sleepiness under high and low sleep pressure. Sleep 24,
A77–A77 (2001).
54. Wu, J. et al. Decreased resting-state alpha-band activation and functional connectivity aer sleep deprivation. Sci. Rep. 11, 484
(2021).
55. Glos, M., Fietze, I., Blau, A., Baumann, G. & Penzel, T. Cardiac autonomic modulation and sleepiness: physiological consequences
of sleep deprivation due to 40h of prolonged wakefulness. Physiol. Behav. 125, 45–53 (2014).
56. Liang, K.-Y. & Zeger, S. L. Longitudinal data analysis using generalized linear models. Biometrika 73, 13–22 (1986).
57. Fitzmaurice, G. M., Laird, N. M. & Ware, J. H. Applied longitudinal analysis. (Wiley, 2011).
58. Lo, S. & Andrews, S. To transform or not to transform: Using generalized linear mixed models to analyse reaction time data. Front.
Psychol. 6, 1171 (2015).
59. King, J.-R. et al. Information sharing in the brain indexes consciousness in noncommunicative patients. Curr. Biol. 23, 1914–1919
(2013).
60. Imperatori, L. S. et al. EEG functional connectivity metrics wPLI and wSMI account for distinct types of brain functional
interactions. Sci. Rep. 9, 8894 (2019).
61. Benedek, M. & Kaernbach, C. A continuous measure of phasic electrodermal activity. J. Neurosci. Methods 190, 80–91 (2010).
62. Hamdan, S. et al. Julearn: An easy-to-use library for leakage-free evaluation and inspection of ML models. GigaByte 2024,
gigabyte113 (2024).
63. Pedregosa, F. et al. Scikit-learn: Machine learning in python. J Mach. Learn. Res. 12, 2825–2830 (2011).
64. Gramfort, A. et al. MEG and EEG data analysis with MNE-Python. Front. Neurosci. 7, (2013).
65. Engemann, D. A. et al. Robust EEG-based cross-site and cross-protocol classication of states of consciousness. Brain 141, 3179–
3192 (2018).
66. Makowski, D. et al. NeuroKit2: A python toolbox for neurophysiological signal processing. Behav. Res. Methods 53, 1689–1696
(2021).
67. Hudson, A. N., Van Dongen, H. P. A. & Honn, K. A. Sleep deprivation, vigilant attention, and brain function: A review.
Neuropsychopharmacol. 45, 21–30 (2020).
68. Csipo, T. et al. Sleep deprivation impairs cognitive performance, alters task-associated cerebral blood ow and decreases cortical
neurovascular coupling-related hemodynamic responses. Sci. Rep. 11, 20994 (2021).
69. Krause, A. J. et al. e sleep-deprived human brain. Nat. Rev. Neurosci. 18, 404–418 (2017).
70. Qi, J.-L. et al. e eects of 43 hours of sleep deprivation on executive control functions: Event-related potentials in a visual go/no
go task. Soc. Behav. Pers. 38, 29–42 (2010).
71. Gosselin, A., De Koninck, J. & Campbell, K. B. Disentangling specic inhibitory versus general decision-making processes during
sleep deprivation using a go/nogo erp paradigm. Int. J. Psychophysiol. 141, 18–27 (2019).
72. Poh, J.-H., Chong, P. L. H. & Chee, M. W. L. Sleepless night, restless mind: Eects of sleep deprivation on mind wandering. J. Exp.
Psychol. Gen. 145, 1312–1318 (2016).
73. Carciofo, R., Du, F., Song, N. & Zhang, K. Mind wandering, sleep quality, aect and chronotype: An exploratory study. PLOS ONE
9, e91285 (2014).
74. Jubera-Garcia, E., Gevers, W. & Van Opstal, F. Local build-up of sleep pressure could trigger mind wandering: Evidence from sleep,
circadian and mind wandering research. Biochem. Pharmacol. 191, 114478 (2021).
75. Lim, J. & Dinges, D. F. Sleep deprivation and vigilant attention. Ann. N. Y. Acad. Sci. 1129, 305–322 (2008).
76. Boulakis, P. A., Mortaheb, S., van Calster, L., Majerus, S. & Demertzi, A. Whole-brain deactivations precede uninduced mind-
blanking reports. J. Neurosci. 43, 6807–6815 (2023).
77. Demertzi, A. et al. Human consciousness is supported by dynamic complex patterns of brain signal coordination. Sci. Adv. 5, 7603
(2019).
78. El-Baba, M. et al. Functional connectivity dynamics slow with descent from wakefulness to sleep. PLOS ONE 14, e0224669 (2019).
Scientic Reports | (2025) 15:4956 17
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
79. Lin, A., Liu, K. K. L., Bartsch, R. P. & Ivanov, P. C. Delay-correlation landscape reveals characteristic time delays of brain rhythms
and heart interactions. Philos. Trans. A. Math. Phys. Eng. Sci. 374, 20150182 (2016).
80. B artsch, R. P., Liu, K. K., Ma, Q. D. & Ivanov, P. C. ree independent forms of cardio-respiratory coupling: Transitions across sleep
stages. Comput. Cardiol. 41, 781–784 (2014).
81. R aimondo, F. et al. Brain–heart interactions reveal consciousness in noncommunicating patients. Ann. Neurol. 82, 578–591 (2017).
82. Candia-Rivera, D., Catrambone, V., Barbieri, R. & Valenza, G. Functional assessment of bidirectional cortical and peripheral
neural control on heartbeat dynamics: A brain-heart study on thermal stress. Neuroimage 251, 119023 (2022).
83. Mittner, M., Hawkins, G. E., Boekel, W. & Forstmann, B. U. A neural model of mind wandering. Trends. Cogn. Sci. 20, 570–578
(2016).
84. Unsworth, N. & Robison, M. K. Pupillary correlates of lapses of sustained attention. Cogn. Aect. Behav. Neurosci. 16, 601–615
(2016).
85. Iwamoto, M. et al. Respiratory entrainment of the locus coeruleus modulates arousal level to avoid physical risks from external
vibration. Sci. Rep. 13, 7069 (2023).
86. Dauvergne, C. et al. Are locus coeruleus neurons involved in blinking?. Neurosci. Res. 61, 182–191 (2008).
87. Park, H.-D. et al. Breathing is coupled with voluntary initiation of mental imagery. Neuroimage 264, 119685 (2022).
88. Tian, Y. E. et al. Evaluation of brain-body health in individuals with common neuropsychiatric disorders. JAMA Psychiatry 80,
567–576 (2023).
89. Bashan, A., Bartsch, R. P., Kantelhardt, J. W., Havlin, S. & Ivanov, P. C. Network physiology reveals relations between network
topology and physiological function. Nat. Commun. 3, 702 (2012).
Acknowledgements
e experimental work was conducted at the GIGA-CRC-Human Imaging platform of ULiège, Belgium.
Author contributions
Paradeisios Alexandros Boulakis: conceptualization; data curation; formal analysis; investigation; methodology;
project administration; soware; visualization; writing—original dra preparation. Nicholas John Simos: inves-
tigation; soware; validation; writing—review and editing. Zoi Stefania: investigation; project administration;
writing—review and editing. Sepehr Mortaheb: formal analysis; soware; writing—review and editing. Chris-
tina Schmidt: methodology; writing—review and editing. Federico Raimondo: formal analysis; methodology;
soware; validation; supervision; writing—review and editing. Athena Demertzi: conceptualization; methodol-
ogy; supervision; writing—original dra preparation.
Funding
At the time of the research, PAB and SM were FNRS Research Fellows. AD and CS are FNRS Research Associ-
ates. is work was supported by the Belgian Fund for Scientic Research (FRS-FNRS), the European Union’s
Horizon 2020 Research and Innovation Marie Skłodowska-Curie RISE programme NeuronsXnets (Grant agree-
ment 101007926), the European Cooperation in Science and Technology COST Action (CA18106), the Léon
Fredericq Foundation, and the University and of University Hospital of Liège. e funders had no role in study
design, data collection and analysis, decision to publish, or preparation of the manuscript.
Declarations
Competing interests
e authors declare no competing interests.
Ethical approval
e experimental procedure has been approved by the CHU Liège local ethics committee and conforms with
the Declaration of Helsinki and the European General Data Protection Regulation (GDPR). Before the onset
of the protocol, participants provided informed consent for their participation in the study. Participants also
received monetary compensation for their participation in the study.
Protocol registration
e stage 1 accepted-in-principle protocol can be found at https://osf.io/sh2ye. e authors conrm that no
data for the pre-registered study was collected prior to the date of AIP.
Additional information
Supplementary Information e online version contains supplementary material available at h t t p s : / / d o i . o r g / 1
0 . 1 0 3 8 / s 4 1 5 9 8 - 0 2 4 - 8 1 6 1 8 - 1 .
Correspondence and requests for materials should be addressed to A.D.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional aliations.
Scientic Reports | (2025) 15:4956 18
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Open Access is article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives
4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in
any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide
a link to the Creative Commons licence, and indicate if you modied the licensed material. You do not have
permission under this licence to share adapted material derived from this article or parts of it. e images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated
otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence
and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to
obtain permission directly from the copyright holder. To view a copy of this licence, visit h t t p : / / c r e a t i v e c o m m o
n s . o r g / l i c e n s e s / b y - n c - n d / 4 . 0 / .
© e Author(s) 2025
Scientic Reports | (2025) 15:4956 19
| https://doi.org/10.1038/s41598-024-81618-1
www.nature.com/scientificreports/
Content courtesy of Springer Nature, terms of use apply. Rights reserved
1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com
Content uploaded by Paradeisios Alexandros Boulakis
Author content
All content in this area was uploaded by Paradeisios Alexandros Boulakis on Feb 10, 2025
Content may be subject to copyright.
