communications biology Article

https://doi.org/10.1038/s42003-024-06533-7

Distinct dynamic connectivity profiles

promote enhanced conscious perception
of auditory stimuli
Check for updates
1 1,2 1 1 2,3
Başak Türker , Dragana Manasova , Benoît Béranger , Lionel Naccache , Claire Sergent &
1
Jacobo D. Sitt

The neuroscience of consciousness aims to identify neural markers that distinguish brain dynamics in
healthy individuals from those in unconscious conditions. Recent research has revealed that specific
1234567890():,;
1234567890():,;

brain connectivity patterns correlate with conscious states and diminish with loss of consciousness.
However, the contribution of these patterns to shaping conscious processing remains unclear. Our
study investigates the functional significance of these neural dynamics by examining their impact on
participants’ ability to process external information during wakefulness. Using fMRI recordings during
an auditory detection task and rest, we show that ongoing dynamics are underpinned by brain patterns
consistent with those identified in previous research. Detection of auditory stimuli at threshold is
specifically improved when the connectivity pattern at stimulus presentation corresponds to patterns
characteristic of conscious states. Conversely, the occurrence of these conscious state-associated
patterns increases after detection, indicating a mutual influence between ongoing brain dynamics and
conscious perception. Our findings suggest that certain brain configurations are more favorable to the
conscious processing of external stimuli. Targeting these favorable patterns in patients with
consciousness disorders may help identify windows of greater receptivity to the external world,
guiding personalized treatments.

Our interactions with the environment are determined by an interplay state functional connectivity. Unlike traditional resting-state connectivity
between endogenous ongoing neural activity and our neural responses to analyses that utilize the entire scan9–12, these studies have enabled the
external stimuli. Each moment, our brains process and integrate a wide identification of recurring brain patterns that vary on a scale of seconds.
variety of internal and external stimuli of different modalities. While some of They revealed distinctive connectivity patterns associated with different
these stimuli are processed consciously and contribute to our subjective states of consciousness. Notably, certain brain patterns with long-range
experiences, most remain unconscious1,2. Neural events correlating with connectivity and negative interactions appear to be characteristic of a
conscious perception are widely investigated in the literature, mainly by conscious state and diminish with the loss of consciousness. Additionally,
comparing the conscious and unconscious perception of the same stimulus thalamic deep brain stimulation that aimed at restoring consciousness in
using various paradigms such as masking3, threshold stimuli presentation4, anesthetized non-human primates has been found to restore the afore-
and attentional blink5. These studies have shown that the same stimulus mentioned connectivity patterns13. However, the functional role of these
with a fixed intensity can induce different brain responses and subjective brain patterns in conscious processing and the formation of subjective
experiences. experience remains unknown. In this study, we investigate whether and how
One of the prerequisites for conscious perception is to be in a conscious ongoing connectivity patterns influence the processing of external infor-
state, such as wakefulness (high arousal and high awareness), as opposed to mation, allowing it to become conscious or not.
unconscious states, such as under anesthesia or coma (no arousal, no The effect of spontaneous baseline brain activity fluctuations on per-
awareness), or disorders of consciousness (arousal with limited awareness). ceptual outcome has been previously explored in different domains by
Recent studies in anesthetized non-human primates6,7, and conscious and contrasting perceived and unperceived trials. Electrophysiological record-
unconscious humans8 have explored the time-varying dynamics of resting ings have shown that the pre-stimulus phase14 and power15,16 of alpha

1
Sorbonne Université, Institut du Cerveau—Paris Brain Institute—ICM, Inserm, CNRS, Paris, 75013, France. 2Université Paris Cité, Paris, 75006, France. 3Inte-
grative Neuroscience and Cognition Center—INCC, UMR 8002, Paris, 75006, France. e-mail: [email protected]; [email protected]

Communications Biology | (2024)7:856 1

https://doi.org/10.1038/s42003-024-06533-7 Article

activity, as well as the phase17 and dynamics18 of (infra-) slow cortical patterns associated with conscious states8 affect the capacity for conscious
oscillations in the task specific regions, correlate with the perceptual out- perception. Using fMRI acquisitions, we showed that participants were
come on trial-by-trial basis. Functional magnetic resonance imaging (fMRI) more likely to detect auditory threshold stimuli when they exhibited con-
studies have found that cue-induced pre-stimulus activity reflects atten- nectivity patterns typical of conscious states8. Additionally, we observed a
tional allocation19 and task preparation20, and predicts task performance. higher occurrence of these favorable connectivity patterns following sti-
Moreover, behavioral performance in Stroop21 and motion discrimination mulus detection, with participants more likely to either maintain or tran-
tasks22 as well as the perceptual outcome of ambiguous vase/face stimuli sition to these patterns after conscious perception. Our findings suggest that
presentation23 seem to vary depending on prior activity fluctuations in task- ongoing brain dynamics and conscious perception have a reciprocal influ-
specific regions such as color-sensitive visual areas, motion-sensitive middle ence on each other and that certain brain configurations provide a window
temporal region, and fusiform face area, respectively. In the nociception of higher receptivity to the external world.
domain, pre-stimulus brain activity in the default-mode and fronto-parietal
networks24, along with the functional connectivity between brain areas Results
involved in pain perception25, are correlated with the subsequent perception We investigated how ongoing brain configurations emerging from the
of pain. And finally, baseline activations in sensory and attentional areas26 coordination of different brain regions influenced the perception of external
and functional connectivity between different brain regions27 have shown to stimuli. 25 participants underwent fMRI recordings during an auditory
predict perceptual performance in an auditory threshold stimulus detection task adapted from Sergent et al.28 (Fig. 1). The task involved
detection task. listening to the French vowel (/a/) that was embedded in continuous noise at
These studies consistently suggest that fluctuations in baseline brain 3 different signal-to-noise ratios (SNR -11, SNR -9, and SNR -7) around the
activity can significantly impact our conscious perception of the external detection threshold. The general sound level was adjusted for each partici-
world. Yet, most research has primarily focused on individual activations pant via a staircase procedure prior to the task to ensure that they could
within specific brain regions associated with a particular task. It’s crucial to detect a stimulus with an SNR of −9 dB in 50% of the trials. They also
recognize that cognitive processes transcend localized regions and manifest underwent a resting-state scan before the task.
through the coordination of various brain networks that process and
exchange information. While some studies have explored the impact of pre- Ongoing brain dynamics are supported by brain patterns con-
stimulus functional connectivity on perceptual outcome25,27, they often focus sistent with those identified in previous research
on pairs of regions rather than considering the overall functional config- Applying the Hilbert transform and k-means clustering, we computed
uration of the brain. Although this approach provides valuable insights into whole-brain connectivity patterns for each fMRI volume acquired during
networks relevant to the ongoing perceptual task, it falls short of offering a the resting-state scan (Fig. 2a). Our clustering procedure resulted in five
comprehensive description of the overall brain states that underpin con- distinct connectivity patterns (Fig. 2b). Subsequently, we utilized the cen-
scious processing. troids of these clusters to label the task data: each task fMRI volume was
The primary objective of the current study is to examine how ongoing assigned to the nearest cluster based on its proximity to the cluster centroids.
brain connectivity states influence the formation of conscious experience, Upon visual inspection, our study’s cluster centroids closely matched those
specifically by affecting the ability to process external information. Our documented by Demertzi et al.8. Patterns 1, 2, 3, and 5 in our study exhibited
study encompasses several goals: (i) describing brain states as global con- a robust similarity in their coherence profiles to those identified in the earlier
nectivity configuration patterns involving different networks simulta- study (Fig. 2c). Importantly, Pattern 1 corresponded to the brain state
neously, (ii) confirming the existence and characteristics of recurrent brain associated with conscious states and diminishing with loss of
patterns observed during resting state8, (iii) investigating the dynamics of consciousness8. On the other hand, Pattern 4 was a new cluster, featuring a
brain states in a time-resolved manner, and (iv) exploring how brain unique coherence profile not observed in the previous study. For a more

Fig. 1 | Experimental procedure. The experimental

session began with a one-up-one-down staircase fMRI Session
procedure which allowed participants to become
familiar with the task and to adjust the stimulus Auditory detecon task
volume such that a stimulus with an SNR of -9 would Resng
Staircase Anat
State B1 ? B2 ? B3 ? B4 ? B5 ? B6 ?
be detected in 50% of the trials. Following a
10minute resting state acquisition, participants 8 mins 10 mins ~ 60 mins 5 mins
completed an auditory detection task, during which
they heard stimuli embedded in a continuous noise
at different SNRs (−7, −9 and −11). They were Experimental Block 1 Post-block subjecve report
instructed to press a button if they detected a sti-
mulus. Stimuli were delivered randomly every 14 s
…

How red were you during the experimental

heard block?
(+/−1 s). At the end of each experimental block,
/a/ SNR -7
participants verbally indicated on a scale of 7: (i)
Connuous background noise

䚉 1 = not red at all

~ 14 secs
how tired they felt, (ii) how successful they think unheard
unhea
ard 䚉 7 = very red
they were during the block, and (iii) their attentional /a/ SNR -9
How good you think you performed in the
focus during the block (1 for complete mind- ~ 14 secs
experimental block?
unheard
unhea
ard
wandering and 7 for complete focus on the task). Catch trial
䚉 1 = performed very badly
~ 14 secs 䚉 7 = performed very well
heard
hear
rd
/a/ SNR -9
Where was your aenonal focus during
~ 14 secs the experimental block?
unheard
unhea
ard
/a/ SNR -11
䚉 1 = complete mind-wandering
Time 䚉 7 = always on task

Communications Biology | (2024)7:856 2

https://doi.org/10.1038/s42003-024-06533-7 Article

a
Sntm Sntm

… …

e
tim

tim
S1t1 S1t1

Resting-state data Detection task data

K-means Label data

1 clustering
2 using clusters

CLUSTER CENTROIDS Patterns found in

b c Demertzi et al. (2019)
Pattern 1

Pattern 1 Pattern 2 Pattern 3 Pattern 4

Pattern 5 Pattern 3 Pattern 2 Pattern 1

Pattern 2 0.65
Patterns found in the

Correlation coefficient
current study

0.81

0.57
Pattern 3

0.59

Pattern 4
d 0.5
0.4

Correlation
coefficient
Pattern 4

0.3
0.2
0.1
Pattern 5 0
Pattern 1 Pattern 2 Pattern 3 Pattern 4

Patterns found in
Demertzi et al. (2019)

Fig. 2 | Clustering procedure and the cluster centroids. a Inter-areal connectivity auditory detection task are assigned to the closest of the five clusters. b Cluster
matrices are computed for each resting state and task fMRI volume. Resting state centroids. c Comparison of the matching patterns. Pearson correlation between the
connectivity matrices are divided into five distinct clusters using k-means clustering coherence values of the patterns revealed a strong correspondence between the
(1). The found cluster centroids are then used to classify the task data (2). By patterns of the two studies. d Formal comparison of Pattern 4 to the patterns found
calculating the distance between each connectivity matrix from the task and the in Demertzi et al. This newly discovered pattern displayed a lower degree of simi-
cluster centroids from the resting state data, the connectivity matrices from the larity to the patterns identified in the earlier study.

formal comparison, we calculated Pearson correlation coefficients between The clustering procedure successfully yielded well-defined cluster
the coherence values of the patterns from the two studies (Fig. 2c). Our centroids (Patterns), showcasing functional connections that respected
analysis revealed very strong correlations between our patterns and those network borders despite the sparse nature of the connectivity configurations
from the earlier study, showcasing a one-to-one matching (Pattern 1: in the input data (Supplementary Fig. 1a). This indicated the capability of
rho = 0.65, p < 0.0001; Pattern 2: rho = 0.81, p < 0.0001; Pattern 3: rho = our clustering procedure to identify commonalities among the connectivity
0.57, p < 0.0001; Pattern 5: rho = 0.59, p < 0.0001). As visually depicted, patterns. To ensure that the similarity between our cluster centroids and
Pattern 4 (the new pattern) demonstrated lower similarity to the patterns those from the prior research was not merely a feature inherent to the
from Demertzi et al., with correlation values ranging from 0.099 to method employed, we conducted a control analysis. We generated a sur-
0.36 (Fig. 2d). rogate dataset and repeated the phase-based connectivity analyses along

Communications Biology | (2024)7:856 3

https://doi.org/10.1038/s42003-024-06533-7 Article

Hypothesis Overall detection rates

a b

Experimental results Threshold stimulus detection rates

c d **
**
*
*

Pattern 1 Pattern 2 Pattern 3 Pattern 4 Pattern 5

Fig. 3 | Detection rates vary with SNRs and pre-stimulus connectivity config- stimulus when displaying Pattern 1 compared to the other patterns. d Detection rates
urations. a Hypothesis of the experiment. We predicted increased detection rates of of the threshold stimulus SNR-09 for different patterns. Pattern 1 was associated
the threshold stimulus (SNR -9) when the consciousness-related brain pattern with increased detection compared to the other patterns. No difference was found
(Pattern 1) was present at the moment of stimulation. b Overall detection rates for between Patterns 2–5. Each point represents one participant (n = 25), and the center
different SNR. c Participant-averaged psychometric curves for each pattern. Error lines of the box plots represent medians.
bars indicate the standard error. Participants were more likely to detect the threshold

with the clustering procedure. By randomly shifting the time-series from would be more likely to detect the threshold stimuli if they had the con-
each ROI and participant, we disrupted the temporal relationship between nectivity pattern (Pattern 1) which was previously shown to be the most
different ROIs while maintaining the temporal order within each time series. typical of conscious states8 (Fig. 3a). Additionally, we considered whether
As anticipated, clustering procedure applied to surrogate data resulted in the perception of a previous stimulus could influence the perception of the
indistinguishable centroids lacking inter-areal connectivity (Supplementary current one. To account for this potential effect, we tested both hypotheses
Fig. 1b). The absence of informative clusters in the surrogate analysis with a linear mixed model with subject ID as a random effect and pattern ID,
demonstrated that the patterns identified in the experimental data truly SNR and previous stimulus detection status (detected vs. undetected) as
reflected the brain’s connectivity state and were not artifacts of the fixed effects.
methodology. Overall detection rates (all patterns considered) were 0.90 for SNR -7,
0.58 for SNR -9, 0.22 for SNR -11 and 0.002 for catch trials (Fig. 3b). As
Certain connectivity profiles are associated with enhanced hypothesized, we found a significant interaction between the SNR and
conscious perception Pattern ID (χ²(12) = 22.67, p = 0.031). Detection rates were significantly
We focused on the task data and hypothesized that the detection of the higher when participants were presenting Pattern 1 (mean = 0.67; med-
threshold stimuli would vary depending on the connectivity pattern present ian = 0.71) compared to Pattern 2 (mean = 0.55; median = 0.50; t = 2.69;
at the time of presentation. More precisely, we predicted that participants p = 0.019 after FDR correction; Cohen’s d = 0.28), Pattern 3 (mean = 0.57;

Communications Biology | (2024)7:856 4

https://doi.org/10.1038/s42003-024-06533-7 Article

a b Occurrence rate of Pattern 1

Post-stimulus pattern occurrence rates ---

Rate
Occurrence rate of Pattern 2

Rate
Occurrence rate

Occurrence rate of Pattern 3

** **

Rate
Occurrence rate of Pattern 4
---

Rate
Occurrence rate of Pattern 5

Rate
Fig. 4 | Conscious perception of threshold stimuli impacts subsequent brain spanning from 3 s before the presentation of the threshold stimulus to 9 s after
dynamics. a Subject-averaged pattern probabilities within the 9 s post-stimulus time stimulation. Shaded areas represent standard errors. Pattern 1 showed higher
window following a detected (D) and undetected (ND) threshold stimulus (SNR-9) occurrence in detected trials (V = 353, p = 0.042, marked with a red line) at the
presentation. The post-stimulus probabilities of Pattern 1 and Pattern 4 varied moment of stimulus presentation (t = 0). It also exhibited a trend of higher occur-
depending on whether the stimulus was detected or undetected. Each point repre- rence in detected trials in the post-stimulus period, particularly 4 s after the sti-
sents one participant (n = 25), and the center lines of the box plots represent med- mulation (V = 254, p = 0.086, marked with an orange dashed line). Conversely,
ians. b Time-resolved pattern probability analysis. Subject-averaged pattern Pattern 4 displayed a trend of lower occurrence in detected trials 5 s following the
probabilities across detected (blue) and undetected (green) trials in a time window stimulus presentation (V = 58, p = 0.067, marked with orange dashed line).

median = 0.62; t = 2.67; p = 0.019; d = 0.26), Pattern 4 (mean = 0.54; med- Stimulus detection increases the occurrence of patterns favor-
ian = 0.57; t = 3.49; p = 0.0042; d = 0.35) and Pattern 5 (mean = 0.57; med- able for conscious perception
ian = 0.60; t = 3.34; p = 0.0042; d = 0.27) only for the threshold SNR -9 To further investigate the dynamic interplay between the ongoing brain
(Fig. 3c–d). No significant differences were found in the other SNRs except patterns and conscious perception, we assessed whether conscious per-
between Pattern 2 and Pattern 5 at SNR -11 (mean: 0.16 vs. 0.25; t = −3.29; ception, in turn, alters ongoing brain configurations. To do so, we calculated
p = 0.01; d = −0.29). When comparing the detection rates at SNR-09 for the occurrence probability of each pattern in a 9 s window following the
each pattern to the overall detection rate resulting from the staircase pro- threshold (SNR -9) stimulus presentation and compared the subject-
cedure (0.58), we found that only Pattern 1 had significantly higher detec- averaged probabilities between detected (D) and not detected (ND) trials
tion rates (one-sided Wilcoxon signed-rank test, V = 231, p = 0.033). The using two-sided Wilcoxon signed-rank test. This time-window allowed us to
detection rates for the other patterns did not differ from the overall detection have enough data points to compute post-stimulus pattern probabilities at
performance (Pattern 2: V = 148, p = 0.66; Pattern 3: V = 174, p = 0.38; the trial level without including patterns from the pre-stimulus period of the
Pattern 4: V = 147, p = 0.66; Pattern 5: V = 170, p = 0.42). Thus, Pattern 1 next trial. The occurrence of Pattern 1 increased following detection, with
was uniquely associated with higher detection rates, while the other patterns participants being more likely to transition to highly connected patterns
did not exhibit this effect. after a stimulus detection (V = 258, p = 0.0088) (Fig. 4a). Conversely, we
We also observed a main effect of previous detection (χ²(1) = 21.20, found an increased occurrence of Pattern 4 when participants did not detect
p < 0.0001), indicating an increased likelihood of detection if the previous the stimuli (V = 69, p = 0.01). The occurrence of the other patterns did not
stimulus was also detected. This effect was observed regardless of the pattern differ between detected and not detected trials (Pattern 2: V = 147, p = 0.69,
(χ²(3) = 6.59, p = 0.086, only a tendency was found) and the SNR Pattern 3: V = 138, p = 0.52, Pattern 5: V = 204, p = 0.27).
(χ²(4) = 7.43, p = 0.11). Importantly, our key findings were not solely These results indicated an overall fluctuation in the occurrence of
dependent on including this covariate in our model. Even without con- Patterns 1 and 4 depending on whether participants consciously perceived
sidering previous detection status, a significant pattern*SNR interaction the stimulus. We performed a time-resolved analysis to further explore the
persisted (χ²(12) = 22.27, p = 0.035). Detection rates were significantly exact timing of these changes. We extracted pattern information within a
higher when participants were presented with Pattern 1 compared to Pat- time window spanning from 3 s prior to the presentation of the threshold
tern 2 (t = 2.66; p = 0.021; d = 0.27), Pattern 3 (t = 2.63; p = 0.021; d = 0.25), stimulus (pre-stimulus period) to 9 s after stimulation (post-stimulus per-
Pattern 4 (t = 3.42; p = 0.0053; d = 0.34) and Pattern 5 (t = 3.28; p = 0.0053; iod) for each trial. Subsequently, we computed, for a given time point, the
d = 0.26) only for the threshold SNR -9, confirming our initial results. probability of pattern occurrence across trials where the stimulus was
Next, we assessed whether the reaction time differed depending on the detected and those where it went undetected. Subject-wise comparison of
ongoing patterns. Although we did not find any significant differences average pattern probabilities between detected and undetected trials
between patterns, we found a main effect of the SNR (χ²(2) = 276.83, revealed no significant differences in occurrence probabilities across con-
p < 0.0001). Not surprisingly, the reaction times were faster for SNR -7 ditions during the pre-stimulus period (Fig. 4b). Conversely, during sti-
(mean = 1.02 s) compared to SNR -9 (mean = 1.10 s; t = −9.43; p < 0.0001; mulus presentation (time = 0), the occurrence probability of Pattern 1 was
d = −0.48) and SNR -11 (mean = 1.20 s; t = −13.16; p < 0.0001; d = −1.00). statistically higher in detected trials compared to undetected trials (V = 353,

Communications Biology | (2024)7:856 5

https://doi.org/10.1038/s42003-024-06533-7 Article

p = 0.0034; after FDR correction p = 0.042), corroborating our initial find- in ongoing brain activity similar to a resting state. Given the technical
ings. In the post-stimulus period, the probability values of Pattern 1 limitations of fMRI (i.e. poor time resolution), this approach was optimal for
remained consistently higher in the detected trials, reaching statistical sig- investigating the effect of baseline brain state on conscious content. Focusing
nificance 4 s following the stimulation (V = 254, p = 0.013; a trend after FDR on later time points would have contaminated our signal with stimulus-
correction p = 0.086). For Pattern 4, a difference between detected and induced activity, thus not purely reflecting baseline brain activity.
undetected trials emerged 5 s after the stimulus presentation (V = 58, It’s worth noting that the increase in performance was only observed
p = 0.0051, a trend after FDR correction p = 0.067). A difference was also for stimuli at threshold; the detection rates for supra- and sub-threshold
noted for Pattern 3 2 s after stimulus presentation, although it did not stimuli remained unaffected by the ongoing brain pattern. This may be
withstand multiple comparisons (V = 72, p = 0.015, not significant after because sub-threshold stimuli do not provide enough bottom-up infor-
FDR correction p = 0.20). mation to trigger conscious processing1 and supra-threshold stimuli are
Finally, we leveraged post-block subjective ratings to delve into the strong enough to provoke conscious processing regardless of the ongoing
relationship between the occurrence of the patterns and participants’ sub- brain pattern. Therefore, threshold stimuli provide a sweet spot where
jective mental states within a given block (Supplementary Text). We found a changes in ongoing brain state can impact the processing of stimuli and give
significant correlation between feelings of tiredness and the occurrence of rise to conscious access when the brain state is favorable.
Pattern 4 (rho = 0.3, p = 0.004, after FDR correction), indicating that par- Previous studies using stimuli at threshold to examine the impact of
ticipants more frequently exhibited Pattern 4 in blocks where they reported baseline brain activity on conscious perception have mainly focused on
feeling tired (Supplementary Fig. 2). variations in local activity within task-related regions14,16,22,23,25. While these
Altogether, our results suggest that ongoing whole-brain functional localized activities can offer insights into the excitability of the task-related
connectivity influences our capacity to consciously perceive external stimuli. regions, establishing their connection to consciousness proves challenging.
Certain whole-brain connectivity profiles that were previously associated Some current theories of consciousness propose that perceptual awareness
with conscious states8 consistently enhanced detection of threshold auditory arises from long-range interactions between different brain regions2,29. To
stimuli. Moreover, the conscious perception of the stimulus also influenced test the predictions of these prevailing theories, it becomes imperative to
the subsequent brain dynamics. Indeed, occurrence probabilities of these examine how the coordination of diverse brain networks impacts conscious
‘favorable’ patterns increased following conscious perception, underscoring perception. Our results align with the global theories of consciousness, as
a reciprocal relationship between ongoing brain dynamics and conscious opposed to local theories of consciousness30,31, by demonstrating that spe-
perception. Our findings demonstrate that processing of external stimuli cific configurations of long-range connectivity are more favorable for con-
may vary depending on the specific brain state we are in and that certain scious perception.
brain states promote greater susceptibility to the external world. Recent research has revealed that the processing of high-level stimuli,
such as audio-visual movies, synchronizes ongoing brain connectivity
Discussion dynamics32. Our findings contribute to this body of literature by illustrating
Our study provides compelling evidence that ongoing brain connectivity how the conscious perception of threshold stimuli influences subsequent
profile influences our ability to process external stimuli. Participants were brain patterns. Specifically, connectivity patterns conducive to conscious
more successful in detecting auditory stimuli at threshold when they perception were more frequently observed following a successful detection.
exhibited brain patterns characteristic of conscious states8 during the sti- This suggests that once the brain consciously processes a stimulus, it tends to
mulation. Previous research has shown that while such patterns frequently remain or transition into states that are favorable for conscious perception.
expressed in healthy individuals, they diminish with loss of consciousness8. This stability in the brain’s receptiveness to the external world over a certain
Our results suggest that these patterns facilitate external information pro- period of time aligns with the concept of perceptual hysteresis33, which refers
cessing, which could explain why they are rarely observed in vegetative state to the influence of an immediately preceding perception on the current one.
patients who lack signs of external awareness. Conversely, these patterns Based on our findings, we could speculate that participants likely transi-
become more frequent in patients transitioning to a minimally conscious tioned into or maintained these favorable states after a detection, thereby
state, where they exhibit transient moments of external awareness. increasing their chances of detecting subsequent stimuli. Indeed, we
One could argue that the pattern observed during stimulus presenta- observed that prior detections increased the likelihood of current detections,
tion might be influenced by residual activity from the previous response. To suggesting a hysteresis effect. Although the interaction with the pattern type
mitigate this potential influence, we implemented unusually long inter-trial was only marginally significant, we believe that all these findings reflect a
intervals (14 s +/−1 s) to minimize carryover effects from the previous trial common hysteresis phenomenon facilitated by the transition to favorable
onto the current one. In our analysis of post-detection occurrence rates, we brain patterns. Further studies are required to explore and validate this
restricted our focus to a 9-s time window rather than the full 14 s post- interpretation thoroughly.
stimulus period. This ensured we had enough data points to calculate post- Another interesting finding was the higher occurrence of Pattern 4
stimulus pattern probabilities at the trial level while avoiding the inclusion of following stimuli that participants failed to detect. This result aligns with
patterns from the pre-stimulus period of the next trial. This strategy helped subjective ratings of tiredness, showing a significant correlation with the
us avoid the potential introduction of confounding factors from overlapping occurrence of Pattern 4 in a given block. If this brain pattern is indeed
time points in both analyses. associated with tiredness, it could elucidate the heightened occurrence of the
Given the hemodynamic response latencies, brain pattern observed at pattern when participants were unable to detect the threshold stimulus,
the moment of stimulation (time 0) reflects brain activity from several possibly due to fatigue. However, we did not observe a predictive effect of
seconds earlier. We chose to focus on this time point rather than 4–5 s post- this pattern on subsequent detection, meaning that displaying such a pattern
stimulation (which would capture the peak of hemodynamic activity pre- did not decrease detection rates. Further research is needed to clarify this
sent at the stimulus presentation) because post-stimulus time points would aspect.
also include activity induced by the stimulus and its detection. This would In a recent study, Mortaheb and colleagues investigated the relation-
result in a mixture of baseline brain activity and stimulus/response-induced ship between ongoing brain connectivity and participants’ ongoing
activity, making it difficult to disentangle their contributions. To avoid this, mentation34. The authors found that mind blanking—a wake state without
we focused on the moment of stimulus presentation, reflecting purely any mental content—was associated with brain patterns exhibiting positive
ongoing baseline brain activity unaffected by the stimulation. We chose time connectivity among different brain regions, a pattern also observed in our
0 over earlier time points (e.g., time -1 or -2) to avoid a larger time lag. Our study (Pattern 3) and in a prior study8. Previous research has linked mind
extended inter-trial intervals allowed hemodynamic responses from the blanking to sluggish responses in sustained attention tasks35,36. In light of this
previous trial to return to baseline, leaving us with spontaneous fluctuations literature, we would have expected to observe longer reaction times when

Communications Biology | (2024)7:856 6

https://doi.org/10.1038/s42003-024-06533-7 Article

participants exhibited this positively connected pattern, indicating potential acquisition. The protocol had been approved by the local ethics committee
mind-blanking. However, we did not find such an effect in our study. It is (promoted by the INSERM, CPP Ile-de-France 6, C13-41). All ethical
important to note some important differences between Mortaheb et al.‘s regulations relevant to human research participants were followed.
study and ours. Unlike Mortaheb’s study, our study wasn’t specifically
designed to explore mind blanking. We did not include any direct probes Experimental design and procedure
about mind blanking, but rather inquired about participants’ mind wan- Participants underwent fMRI recordings while performing an auditory
dering at the end of each block. Moreover, while reaction times can serve as detection task. They were asked to detect a French vowel (/a/) embedded in
an indirect measure of mind blanking, subjective reports offer more direct continuous noise28, at 3 different signal-to-noise ratios (SNR -11, SNR -9
evidence. In the future, combining both our and Mortaheb’s paradigms and SNR -7) around the detection threshold. Stimuli were delivered in a
could offer a more comprehensive understanding of how ongoing brain and randomized fashion every 14 s (+/−1 s) using MRI-compatible head-
mental states interact with the ability to perceive the external world. phones. Participants had their eyes closed in the fMRI scanner and pressed a
Our study takes a distinct approach from prior research in this field. button with their right thumb when detected a stimulus. The sound level was
While previous studies typically compared pre-stimulus brain activity in adjusted for each participant via a staircase procedure prior to the task to
detected versus undetected trials14–18,22–27,37, we opted for a hypothesis-driven ensure 50% detection at SNR -9. This resulted in higher detection rates at
approach by independently labeling brain patterns, irrespective of the trial SNR -7 and lower detection rates at SNR -11. The task consisted of 6 blocks
outcome. Rather than looking for differences in brain activity between of 8 min separated by a small rest period. Thirty stimuli were presented in
detected and undetected trials, we predicted and showed that certain pre- each block (ten per SNR level) in addition to three catch trials. Thus, the
defined brain activity configurations can enhance conscious perception. whole task contained 198 trials (33 per block). After each block participants
This strategy allowed us to illustrate the relevance of specific brain patterns were asked to verbally indicate on a scale of 7: (i) how tired they felt, (ii) how
for both conscious access and conscious states, thereby bridging these two successful they think they were during the block, and (iii) their attentional
areas of research that are often explored in isolation. focus during the block (1 for complete mind-wandering and 7 for complete
Our study also replicated the brain connectivity patterns observed focus on the task). Participants also underwent a 10 min resting state with
during rest in a prior publication8. Despite employing different MRI scan- eyes closed before the task and a 5 min anatomical scan after the task. fMRI
ners and scanning different populations, our findings were consistent with was acquired throughout the whole experimental session including the
the earlier study, suggesting that our method can produce reliable results staircase procedure and the resting state. The total experimental session
despite variations in experimental conditions. Some might argue that the took ~2 h.
similarity between the two studies is due to an inherent feature of our
method rather than to the experimental data. To rule out this possibility, we MRI acquisition parameters
tested our clustering method on surrogate data with the same characteristics All MRI data were acquired on a 3 T Siemens Prisma System. T2*-weighted
as our original dataset. This control analysis produced completely different whole brain resting state images were acquired with a multi-band gradient-
connectivity patterns than the original, which lacked all types of con- echo planar imaging (EPI) sequence (600 volumes, 48 slices, slice thickness:
nectivity, illustrating the robustness of our method in different experimental 3 mm, TR/TE: 1000 ms/30 ms, voxel size: 3 × 3 × 3 mm, multiband accel-
settings. One notable difference between the patterns of the two studies was eration factor: 3, flip angle: 60°). Functional MRI images during the detec-
that our inter-areal coherence values were lower than those in the prior tion task were acquired using the same sequence. The cardiac and
study. This discrepancy might be attributed to the fact that we implemented respiratory activities were also recorded during the fMRI acquisitions. An
additional denoising steps, such as regressing out heart and respiratory anatomical volume was acquired using a T1-weighted MPRAGE sequence
activity from the ROI timeseries. This extra step potentially eliminated the in the same acquisition session (192 or 256 slices, slice thickness: 1 mm,
correlation between different regions induced by physiological artifacts, TR/TE: 2.300 ms/ 2.76 ms, voxel size: 1 × 1 × 1 mm, flip angle: 9°).
retaining only the ‘real’ coherence among brain regions and subsequently
reducing the coherence values. However, the denoising procedure did not fMRI preprocessing
impact the coherence profiles. The raw MRI data underwent preprocessing and denoising using custom
Our research has opened up possibilities for other exciting studies that MATLAB (The MathWorks) scripts. The preprocessing included seg-
examine how fluctuations in the ability to perceive external information mentation using CAT1238, realignment, co-registration, and normalization
from different modalities occur. Given the significance of these brain con- into the MNI152 (Montreal Neurological Institute) space as implemented in
figurations for both conscious access and conscious states, we believe that SPM1239. We did not perform slice-timing correction as our TR was already
the observed results extend beyond the auditory domain alone. We antici- short (1 s). We also avoided spatial smoothing of our data. The
pate that future studies will reveal that the perception of stimuli from other susceptibility-induced off-resonance field distortions were corrected using
modalities, such as visual or somatosensory stimuli, is similarly influenced the topup procedure40 as implemented in FSL41, providing a more accurate
by these ongoing patterns. representation of the brain. Peripheral physiological data recorded during
The findings of our study indicate that ongoing patterns of brain the scans such as respiration and cardiac pulsation were extracted using
connectivity, which are associated with conscious states, may actively PhysIO Toolbox42. White matter masks, realignment parameters as well as
contribute to shaping conscious experiences by altering the capacity to their first and second order derivatives, cardiac and respiratory signals were
perceive external information. In the future, identification of these favorable included as nuisance regressors in the generalized linear model (GLM) in
patterns in real-time in individuals with consciousness disorders could order to denoise the data. Average time-series were extracted from 42
enable us to target periods of increased receptivity to the external world, regions of interest (ROIs) defined as 5 mm radius spheres centered at spe-
paving the way for the development of personalized patient-care protocols. cified MNI coordinates (as listed in Supplementary Table 1).

Methods Time-varying functional connectivity patterns

Participants Resting state. After preprocessing, the extracted ROI time-series were
Twenty-six healthy participants were recruited for this study (13 women, converted into a complex representation using their original signal and
mean age: 24.6 ± 4.2 years, 25 right-handed). All participants were native the Hilbert transform. The instantaneous phase was calculated by taking
French speakers with good hearing and without any neurological or psy- the inverse tangent of the ratio of the imaginary and real parts and then by
chiatric disorders. They gave written consent prior to the experiment and “wrapping” into the [-π,π] interval. This created a series of instantaneous
were remunerated €70 for their participation in the study. One male par- phases for each ROI. Next, the phase differences between each ROI pair
ticipant was excluded from the study due to technical issues during the MRI were determined at each time point using cosine similarity, allowing the

Communications Biology | (2024)7:856 7

https://doi.org/10.1038/s42003-024-06533-7 Article

brain’s connectivity configuration at each time point to be represented in 6. Barttfeld, P. et al. Signature of consciousness in the dynamics of
an 861-dimensional space (each dimension represented the coherence of resting-state brain activity. Proc. Natl Acad. Sci. USA. 112,
a pair of ROIs). Data from all participants were combined and k-means 887–892 (2015).
clustering (with k values of 3, 4, 5, 6, and 7) was applied with 1000 7. Uhrig, L. et al. Resting-state dynamics as a cortical signature of
repetitions using the Manhattan distance to identify recurrent con- anesthesia in monkeys. Anesthesiology 129, 942–958 (2018).
nectivity configurations. The silhouette method determined that five 8. Demertzi, A. et al. Human consciousness is supported by dynamic
clusters provided the best classification. The connectivity configuration at complex patterns of brain signal coordination. Sci. Adv. 5,
each time point (a 42 by 42 phase coherence matrix) was then labeled with eaat7603 (2019).
one of the 5 cluster centroids. Finally, the participants’ brain activity 9. Damoiseaux, J. S. et al. Consistent resting-state networks across
during the scans was represented as a sequence of the five centroids. healthy subjects. Proc. Natl Acad. Sci. USA 103, 13848–13853 (2006).
10. Fox, M. D. et al. The human brain is intrinsically organized into
Detection task. For each task fMRI volume, inter-areal coherence dynamic, anticorrelated functional networks. Proc. Natl Acad. Sci.
matrices were calculated using the Hilbert transform and cosine simi- USA 102, 9673–9678 (2005).
larity as described above. These matrices were then assigned to the closest 11. van den Heuvel, M. P. & Hulshoff Pol, H. E. Exploring the brain
of the five clusters that were computed using the resting state data. network: a review on resting-state fMRI functional connectivity. Eur.
Neuropsychopharmacol. 20, 519–534 (2010).
Statistics and reproducibility 12. Yeo, B. T. T. et al. The organization of the human cerebral cortex
Statistical analyses were conducted in R43 using lme444, emmeans45, and car46 estimated by intrinsic functional connectivity. J. Neurophysiol. 106,
packages. To account for multiple comparisons, all statistics were corrected 1125–1165 (2011).
using the False Discovery Rate (FDR) Benjamini-Hochberg procedure. 13. Tasserie, J. et al. Deep brain stimulation of the thalamus restores
Linear mixed models with subject ID as a random factor were used to signatures of consciousness in a nonhuman primate model. Sci. Adv.
investigate SNR and Pattern ID on detection rates and reaction times. The 8, eabl5547 (2022).
statistics for both detection rates and reaction times were calculated at the 14. Busch, N. A., Dubois, J. & VanRullen, R. The phase of ongoing EEG
individual subject level, and the observations in the model were weighted oscillations predicts visual perception. J. Neurosci. 29, 7869–7876 (2009).
based on the number of trials performed by each participant. Only responses 15. Ergenoglu, T. et al. Alpha rhythm of the EEG modulates visual
with a latency between 400 ms and 2000ms were considered in the analyses. detection performance in humans. Brain Res. Cogn. Brain Res. 20,
Importantly, an arcsine transformation was applied to the detection rates 376–383 (2004).
and an inverse transformation was applied to the reaction times (1/RT) to 16. Wyart, V. & Tallon-Baudry, C. How ongoing fluctuations in human
better meet the model assumptions. The assumptions of the linear models visual cortex predict perceptual awareness: baseline shift versus
were assessed visually through residual distributions and Q-Q plots, and the decision bias. J. Neurosci. 29, 8715–8725 (2009).
significance of individual factors was evaluated using Wald X-tests. Subject- 17. Monto, S., Palva, S., Voipio, J. & Palva, J. M. Very slow EEG
averaged post-stimulus pattern probabilities were compared between fluctuations predict the dynamics of stimulus detection and oscillation
detected and undetected trials using paired Wilcoxon signed-rank tests. amplitudes in humans. J. Neurosci. 28, 8268–8272 (2008).
Finally, the relationship between pattern probabilities and subjective reports 18. Baria, A. T., Maniscalco, B. & He, B. J. Initial-state-dependent, robust,
was assessed using Spearman correlations rather than Pearson correlations transient neural dynamics encode conscious visual perception. PLoS
since they are more suited for ordinal scales such as subjective ratings. Comput. Biol. 13, e1005806 (2017).
19. Sapir, A., d’Avossa, G., McAvoy, M., Shulman, G. L. & Corbetta, M. Brain
Reporting summary signals for spatial attention predict performance in a motion
Further information on research design is available in the Nature Portfolio discrimination task. Proc. Natl Acad. Sci. USA 102, 17810–17815 (2005).
Reporting Summary linked to this article. 20. Ekman, M., Derrfuss, J., Tittgemeyer, M. & Fiebach, C. J. Predicting
errors from reconfiguration patterns in human brain networks. Proc.
Data availability Natl Acad. Sci. USA 109, 16714–16719 (2012).
All data that support the findings of the study can be found in OSF. 21. Coste, C. P., Sadaghiani, S., Friston, K. J. & Kleinschmidt, A. Ongoing
brain activity fluctuations directly account for intertrial and indirectly
Code availability for intersubject variability in stroop task performance. Cereb. Cortex.
Custom analysis scripts can be found in OSF. 21, 2612–2619 (2011).
22. Hesselmann, G., Kell, C. A. & Kleinschmidt, A. Ongoing activity
Received: 3 March 2024; Accepted: 2 July 2024; fluctuations in hMT+ bias the perception of coherent visual motion. J.
Neurosci. 28, 14481–14485 (2008).
23. Hesselmann, G., Kell, C. A., Eger, E. & Kleinschmidt, A. Spontaneous
References local variations in ongoing neural activity bias perceptual decisions.
1. Dehaene, S., Changeux, J.-P., Naccache, L., Sackur, J. & Sergent, C. Proc. Natl Acad. Sci. USA 105, 10984–10989 (2008).
Conscious, preconscious, and subliminal processing: a testable 24. Boly, M. et al. Baseline brain activity fluctuations predict
taxonomy. Trends Cogn. Sci. 10, 204–211 (2006). somatosensory perception in humans. Proc. Natl Acad. Sci. USA 104,
2. Dehaene, S. & Changeux, J.-P. Experimental and theoretical 12187–12192 (2007).
approaches to conscious processing. Neuron 70, 200–227 (2011). 25. Ploner, M., Lee, M. C., Wiech, K., Bingel, U. & Tracey, I. Prestimulus
3. Del Cul, A., Baillet, S. & Dehaene, S. Brain dynamics underlying the functional connectivity determines pain perception in humans. Proc.
nonlinear threshold for access to consciousness. PLoS Biol. 5, Natl Acad. Sci. USA 107, 355–360 (2010).
e260 (2007). 26. Sadaghiani, S., Hesselmann, G. & Kleinschmidt, A. Distributed and
4. Pins, D. & Ffytche, D. The neural correlates of conscious vision. Cereb. antagonistic contributions of ongoing activity fluctuations to auditory
Cortex. 13, 461–474 (2003). stimulus detection. J. Neurosci. 29, 13410–13417 (2009).
5. Sergent, C., Baillet, S. & Dehaene, S. Timing of the brain events 27. Sadaghiani, S., Poline, J.-B., Kleinschmidt, A. & D’Esposito, M.
underlying access to consciousness during the attentional blink. Nat. Ongoing dynamics in large-scale functional connectivity predict
Neurosci. 8, 1391–1400 (2005). perception. Proc. Natl Acad. Sci. USA 112, 8463–8468 (2015).

Communications Biology | (2024)7:856 8

https://doi.org/10.1038/s42003-024-06533-7 Article

28. Sergent, C. et al. Bifurcation in brain dynamics reveals a signature of Acknowledgements

conscious processing independent of report. Nat. Commun. 12, This study was funded by the European Union (ERA PerMed JTC2019
1149 (2021). “PerBrain”, grant to JDS). BT received a PhD grant from French Ministry of
29. Mashour, G. A., Roelfsema, P., Changeux, J.-P. & Dehaene, S. Higher Education and Ecole Normale Supérieure. DM received individual
Conscious processing and the global neuronal workspace funding from Ecole Doctorale Frontières de l’Innovation en Recherche et
hypothesis. Neuron 105, 776–798 (2020). Education–Programme Bettencourt.
30. Lamme, V. A. F. Towards a true neural stance on consciousness.
Trends Cogn. Sci. 10, 494–501 (2006). Author contributions
31. Lamme, V. A. F. How neuroscience will change our view on B.T., C.S., and J.D.S. designed the research; B.T. and D.M. acquired the
consciousness. Cogn. Neurosci. 1, 204–220 (2010). data; B.B. provided technical guidance for the acquisition and pre-
32. Türker, B., Belloli, L., Owen, A. M., Naci, L. & Sitt, J. D. Processing of processing of the data; B.T. and D.M. analyzed the data; B.T., D.M., L.N.,
the same narrative stimuli elicits common functional connectivity C.S., and J.D.S interpreted the data. B.T. wrote the manuscript and all
dynamics between individuals. Sci. Rep. 13, 21260 (2023). authors contributed to its editing.
33. Chambers, C. & Pressnitzer, D. Perceptual hysteresis in the judgment
of auditory pitch shift. Atten. Percept. Psychophys. 76, 1271–1279 Competing interests
(2014). The authors declare no competing interests.
34. Mortaheb, S. et al. Mind blanking is a distinct mental state linked to a
recurrent brain profile of globally positive connectivity during ongoing Additional information
mentation. Proc. Natl Acad. Sci. USA 119, e2200511119 (2022). Supplementary information The online version contains
35. Unsworth, N. & Robison, M. K. Pupillary correlates of lapses of supplementary material available at
sustained attention. Cogn. Affect Behav. Neurosci. 16, 601–615 https://doi.org/10.1038/s42003-024-06533-7.
(2016).
36. Andrillon, T., Burns, A., Mackay, T., Windt, J. & Tsuchiya, N. Predicting Correspondence and requests for materials should be addressed to
lapses of attention with sleep-like slow waves. Nat. Commun. 12, Başak Türker or Jacobo D. Sitt.
3657 (2021).
37. Pincham, H. L. & Szűcs, D. Conscious access is linked to ongoing Peer review information Communications biology thanks Davinia
brain state: electrophysiological evidence from the attentional Bblink. Fernández-Espejo and the other, anonymous, reviewer(s) for their
Cereb. Cortex. 22, 2346–2353 (2012). contribution to the peer review of this work. Primary Handling Editor:
38. Gaser, C. et al. CAT—A computational anatomy toolbox for the Benjamin Bessieres. A peer review file is available.
analysis of structural MRI data. bioRxiv https://doi.org/10.1101/2022.
06.11.495736 (2023). Reprints and permissions information is available at
39. Penny, W. D., Friston, K. J., Ashburner, J. T., Kiebel, S. J. & Nichols, T. http://www.nature.com/reprints
E. Statistical Parametric Mapping: The Analysis of Functional Brain
Images, 656 (Elsevier, 2011). Publisher’s note Springer Nature remains neutral with regard to
40. Andersson, J. L. R., Skare, S. & Ashburner, J. How to correct jurisdictional claims in published maps and institutional affiliations.
susceptibility distortions in spin-echo echo-planar images:
application to diffusion tensor imaging. NeuroImage 20, 870–888 Open Access This article is licensed under a Creative Commons
(2003). Attribution 4.0 International License, which permits use, sharing,
41. Smith, S. M. et al. Advances in functional and structural MR image adaptation, distribution and reproduction in any medium or format, as long
analysis and implementation as FSL. Neuroimage 23, S208–S219 as you give appropriate credit to the original author(s) and the source,
(2004). provide a link to the Creative Commons licence, and indicate if changes
42. Kasper, L. et al. The physIO toolbox for modeling physiological noise were made. The images or other third party material in this article are
in fMRI data. J. Neurosci. Methods 276, 56–72 (2017). included in the article’s Creative Commons licence, unless indicated
43. R Core Team. R: A Language and Environment for Statistical otherwise in a credit line to the material. If material is not included in the
Computing. R Foundation for Statistical Computing, Vienne, Austria. article’s Creative Commons licence and your intended use is not permitted
https://www.R-project.org (2021). by statutory regulation or exceeds the permitted use, you will need to
44. Bates, D., Mächler, M., Bolker, B. & Walker, S. Fitting linear mixed- obtain permission directly from the copyright holder. To view a copy of this
effects models using ime4. J. Stat. Softw. 67, 1–48 (2015). licence, visit http://creativecommons.org/licenses/by/4.0/.
45. Lenth, R. V. emmeans: Estimated Marginal Means, Aka Least-Squares
Means. R Package Version 1.6.2-1. https://cran.r-project.org/web/ © The Author(s) 2024
packages/emmeans (2021).
46. Fox, J. & Weisberg, S. An R Companion To Applied Regression 3rd
edn, 608 (SAGE Publications, 2019).

Communications Biology | (2024)7:856 9