Why does consciousness fade in early sleep

Perniciously, subjects typically believe themselves to be alert all the time during microsleep without recalling any period of unconsciousness. This misapprehension can be perilous to someone in the driver's seat. Microsleep can be fatal when driving or operating machinery such as trains or airplanes, hour after tedious hour.

During a microsleep episode, the entire brain briefly falls asleep, raising the question of whether bits and pieces of the brain can go to sleep by themselves, without the entire organ succumbing to slumber. In this research, 11 adult rats had microwires implanted into their frontal motor cortex, which controls movement. Inserted into the cortical tissue, the sensors picked up both the voltage called the local field potential LFP , akin to the EEG, in addition to the spiking activity of nearby nerve cells.

As expected, when awake, the LFP was dominated by low-amplitude, fast waves readily distinguishable from the larger and slower waves characteristic of non-REM deep sleep [ see box below ].

At the level of individual neurons, the awake animals' cortical cells chatted away in an irregular, staccato manner over an extended period. This neuronal reticence occurs simultaneously all over the cortex. It alternates with regular on periods, leading to the rising and falling brain waves that are the hallmark of deep sleep.

Knowing all this, the researchers decided to probe further. Instead of letting the rats go to sleep at their usual bedtime, the experimentalists engaged the animals in a rodent version of late-night video gaming, continuously exposing them to toys and other objects to sniff, explore and play with. They tapped on the cage and otherwise prevented them from assuming a sleep posture or becoming drowsy.

After four hours of such excitement, the rats could finally slumber. As expected from previous animal and human studies, by the end of the sleep deprivation phase, the LFP began to shift to lower frequencies, compatible with the idea that the pressure for the animals to sleep steadily built up.

Closer inspection of the electrical signatures, however, revealed something unexpected: occasional, sporadic, silent periods of all or most of the neurons in the recorded brain region [ see box below ] without the animals showing either behavioral or EEG manifestations of microsleep. These short, off-like episodes were often associated with slow waves in the LFP.

The opposite happened during recovery sleep, toward the end of this six-hour period, when the pressure to sleep had presumably abated. At this point, large and slow waves in the LFP became more infrequent, and neuronal activity turned more irregular, as it did during wakefulness. It appears that when awake but sleep-deprived, neurons show signs of sleepiness, whereas after hours of solid sleep, individual neurons start waking up. Careful statistical analysis confirmed these trends: the number of off periods increased during the four hours the rats were forced to stay awake, and the opposite dynamic occurred during recovery sleep.

One question was whether any one neuron fell asleep independent of any other neuron. Or was this occurrence more of a global phenomenon, whereby all neurons simultaneously transition to an off period? Click or tap to enlarge. That is, sometimes neurons in both regions went off together, whereas at other times they did so independently.

Yet as the sleep pressure built up, after several hours of being kept awake, neuronal activity during sleep deprivation did become more globally synchronized as it does in deep sleep. These fluctuations occur in the range between seconds and minutes, in contrast to the functional connectivity changes in the hundreds of milliseconds which are involved in the formation of the dynamical core.

This different scale does not eliminate the possibility that dynamical BOLD connectivity is a manifestation of accumulated changes taking place in a faster temporal range. Furthermore, we have recently shown that these sudden BOLD functional connectivity changes can be traced to fluctuations in frequency-specific local neural synchronization, as indexed by scalp EEG power Tagliazucchi et al. Insights on the dynamical nature of conscious processes could be then obtained by analyzing how BOLD dynamical functional connectivity is modulated across different conscious states.

The dynamical core hypothesis predicts that, while average connectivity can remain the same or be even higher during sleep, its temporal variance will decrease when compared to that of wakefulness.

This result was confirmed for frontal connectivity in subjects undergoing transitions to light sleep Tagliazucchi et al. The resulting variances are shown in Figure 3 B for wakefulness, for subjects undergoing transitions to light sleep and for the difference. It can be expected that capitalizing on new tools to map dynamical changes of network modular structure Bassett et al.

Figure 3. A sliding window procedure is applied to obtain the time-dependent Pearson correlation between all pairs of BOLD signals extracted from any given brain parcellation.

The same sliding window is then applied to obtain the evolution of EEG power in different frequency bands.

Results are for awake subjects and for subjects undergoing transitions to light sleep drowsiness. On the right, the difference between both conditions is plotted. Figure adapted from Tagliazucchi et al. Tellingly, correlations between BOLD synchronization and gamma power were not found for subjects undergoing transitions to early sleep.

Furthermore, synchronization in the gamma range has been proposed as a solution to the binding problem and could be of fundamental importance for the formation of the dynamical core Treisman, Taken together, these results show that large-scale dynamical synchronization between a widespread network of brain regions can be predicted from gamma power and can be modulated by vigilance fluctuations, putting in evidence the importance of dynamical functional connectivity for the maintenance of wakeful rest.

The previous discussions rely on the everyday experience of deep sleep associated with a loss of conscious awareness.

Dream reports obtained after awakenings from REM sleep are full of vivid sensory and emotional content Hobson, In contrast, reports obtained after NREM sleep stages are of a more heterogeneous nature.

Generally, the sleep stages defined by the AASM sleep staging criteria are associated with different behavioral and cognitive changes, as well as with different types of subjective conscious content reports obtained after sudden awakenings. Early sleep N1 stage is characterized by recollections of vivid hypnagogic hallucinations and lucid dreams Domhoff, ; Kusse et al. It is known that the cortex can remain in a state similar to wakefulness for minutes after the thalamic deactivation occurring during sleep onset Magnin et al.

This difference likely underlies preserved frontoparietal connectivity during early sleep Larson-Prior et al. In spite of a dramatically diminished conscious awareness, dream reports are not completely absent from the deepest sleep stages Tononi, These reports are less lengthy than those obtained after REM sleep and in general less vivid and elaborated Antrobus, However, this possibility makes it necessary to monitor conscious content whenever sleep is used as a brain state model for unconsciousness.

Given the difficulty of awakening every subject to ask them for reports of their conscious contents, this fact is often overlooked in sleep neuroimaging studies which aim to address changes associated with the unconscious state. We asked our subjects to complete questionnaires after the EEG-fMRI scanning session, which include numerical ratings of conscious content of different nature, such as perceived visual imagery and inner speech.

This result appears to confirm that conscious sensory content can be present during a sleep stage which is commonly associated with unconsciousness N2 sleep. Figure 4. Subjective reports were collected retrospectively immediately after the fMRI session. This and other results do not suggest by any means that NREM sleep is an inadequate experimental model for unconsciousness.

Caution is in order, however, and there is a need for experimenters to assess to what extent their subjects are truly unconscious. In this review we have discussed the potential of NREM sleep neuroimaging to address fundamental predictions of theories about human consciousness. A relatively recent but constant methodological progress has allowed researchers to extract information from fMRI experiments going well beyond changes in BOLD signal amplitude.

Many of these new methods are particularly useful to test the aforementioned predictions. For example, functional connectivity analyses reveal the disintegration of frontoparietal connectivity during deep sleep.

Global functional connectivity networks studied using graph theoretical methods confirm that the deepest stages of NREM sleep are associated with increased functional modularity, in line with the predictions of the information integration theory.

The recent discovery of a constantly changing BOLD connectivity could be fundamental to shed light on how different states of consciousness modulate the occurrence of transient synchronized states in the thalamocortical system. We have emphasized that human consciousness is an emergent and global property of the human brain.

The distributed nature of neural activity giving rise to consciousness is difficult to capture with neuroimaging methods which do not provide simultaneous coverage of the thalamocortical system with a reasonable spatial resolution.

Currently, fMRI is one of the very few methods at our disposal which partially meets these requirements. However, the excitement about its possibilities should not prevent a discussion of its limitations. The most important is the limited temporal resolution in the range of seconds , especially considering that neural processes giving rise to conscious awareness take place at the scale of hundreds of milliseconds. This leads to a careful interpretation of studies reporting functional connectivity changes, which need to be interpreted as an average over long periods of time.

The number of samples required to obtain reliable connectivity estimates imposes a limit on the minimum temporal length during which functional connectivity can be resolved, a limit which is effectively longer than the actual temporal resolution of fMRI.

Finally, we closed our discussion with a note of caution regarding human sleep as an empirical model of absent consciousness. While human NREM sleep offers many advantages, it is also questionable to what extent it can be considered completely devoid of conscious content.

Reports obtained after awakenings from early sleep confirm the presence of vivid sensory imagery. On the other hand, reports from deeper sleep stages are much shorter and can in fact be lacking any conscious content at all. We approached this problem by asking our subjects to fill a post-scan questionnaire which includes the rating of conscious sensory perception visual and auditory during the experiment.

This result suggests that conscious content can indeed be present during a typical NREM neuroimaging experiment. Thus, researchers should take this possibility into consideration when interpreting their results and, if possible, probe conscious content during the different sleep stages under study.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The authors declare no competing financial interests. Terminology and Technical Specifications. Chicago: American Academy of Sleep Medicine. Antrobus, J. Psychophysiology 20, — Baars, B. A Cognitive Theory of Consciousness. Balduzzi, D. Integrated information in discrete dynamical systems: motivation and theoretical framework. PLoS Comp. CrossRef Full Text. Bassett, D. Dynamic reconfiguration of human brain networks during learning. Bastien, C. EEG characteristics prior to and following the evoked K-complex.

Beckmann, C. Investigations into resting-state connectivity using independent component analysis. B Biol. Blumenfeld, H. Why do seizures cause loss of consciousness? Neuroscientist 9, — Boly, M.

The difficulty that people have detecting the strangeness of these experiences prompted psychologists Clemens and Jana Speth, both at the University of Dundee, to examine reports of hypnagogic intrusions. Similarly, by comparing the hypnagogic state to REM dreaming, a study by the same researchers confirmed the long-noted observation that while dreams often feel fully immersive, hypnagogia tends to be experienced as if we were passive observers—with the hallucinatory thoughts and images occurring as a projection on our existing sense of reality.

Admittedly, studies from this new wave of interest in hypnagogia are small and still tentative but they reflect a growing trend towards understanding sleep not just as a state of rest and consolidation, but also as a scientific tool for observing the components of consciousness as they are stripped away for entry into slumber. What these studies may be hinting at, is that the brain processes involved in sustaining consciousness might also be central to maintaining a stable, insightful experience of the world—in other words, keeping hallucinations in check.

Francesca Siclari , a sleep and consciousness researcher at the University Hospital of Lausanne, in Switzerland, also hopes this new wave of interest will result in more practical benefits. The length of the report was assessed by the total number of words the participants used to report their dreams divided by the maximum number of words the subject used.

The amplitude of the TMS-evoked deflection was measured by subtracting the signal amplitude of the anterior cluster from that of the posterior cluster see Methods and Supplementary Fig. S2c for the individual data. Consistent with the fit in Fig. Correlation between the length of the conscious experience and the amplitude of the TMS-evoked response for all participants. The red dots represent the signal amplitude of the TMS-evoked response and the normalized total word count of individual awakenings.

The blue circle represents the average amplitude of the data associated with no conscious experience. Finally, we conducted a control analysis on the trials occurring 2—2. This supports the hypothesis that the CE and NCE reports captured changes in brain states during the last moments before the awakenings. The red dots represents the signal amplitude of the TMS-evoked responses and the normalized total word count of individual awakenings.

The line represents a linear fit to the data. In summary, TMS applied to a posterior cortical region shortly before subjects are awakened from NREM sleep induces a different EEG response when subjects do or do not report having been conscious before the awakenings.

Specifically, when subjects said they were not conscious, the EEG response was larger—similar to a larger NREM-sleep slow-wave—and the period of phase-locking was shorter.

Importantly, this kind of response is characteristic of cortical networks when they are in a condition of bistability between depolarized up-states and hyperpolarized down-states At the single-cell level, bistability means that any input quickly triggers a stereotypical neuronal down-state, after which neurons enter an up-state and activity resumes stochastically and is thought to be due primarily to depolarization-dependent potassium currents and short-term synaptic depression prevalent in NREM sleep Indeed, the spontaneous slow waves that characterize NREM sleep can be considered as a prime expression of neuronal bistability.

Cortical bistability, as reflected in the loss of phase-locking to a stimulus, leads to a breakdown in the ability of the cortex to integrate information, which is thought to be essential for being conscious 1 , 29 , More specifically, the decrease of phase-locked activity in the alpha band suggests that bistability may interfere primarily with feedback processes among distributed cortical areas, which has also been linked to consciousness 31 , 32 , The inverse relationship between bistability and consciousness is further supported by our observation that the amplitude of the TMS-evoked potential indicative of greater bistability was correlated with the length of the report shorter conscious experiences.

Overall, our findings suggest that local changes in the bistability of posterior cortical networks, as revealed by TMS—EEG responses, can faithfully reflect variations in the level of consciousness.

Moreover, since changes in TMS—EEG responses predict changes in consciousness within the same physiological state—NREM sleep—these findings highlight the importance of bistability while ruling out possible confounding factors due to physiological state changes. However, there is still a caveat, as a truly homogenous state is an abstraction: there is still some variability within NREM sleep and, in fact, one can categorize NREM sleep based on type-1 and type-2 waves 34 as well as the predominant location of the wave front versus back 20 , not to mention the different ways these slow waves can travel Future studies should next look at how TMS—EEG responses and dream reports in NREM sleep depend on the pre-stimulus endogenous oscillatory pattern, as the amplitude of the TMS-evoked response has been shown to correlate with the preceding slow-wave amplitude and specific timing of TMS with respect to peaks and troughs Eleven healthy subjects 2 females, age Prior to the experiment, participants were screened for neurological, psychiatric MINI 5.

Participants had no contraindications for TMS e. They had good sleep quality as assessed by the Pittsburgh Sleep Quality Index We discarded five participants because one displayed TMS-evoked responses with a very low signal-to-noise ratio during the first wakefulness session, two were unable to sleep in the laboratory environment one night and two nights, respectively , one perceived phosphenes during TMS and interrupted the study before the last two nights and the last subject presented auditory evoked responses due to the TMS coil click despite the noise masking.

The data presented here are therefore from six right-handed participants who completed the study 5 males, age Thus, here we report the results using the data of the last four nights that were acquired with the higher intensity. Written informed consent was obtained from each participant. The study was approved by the University of Wisconsin Human Subjects Committee and was carried out in accordance with the Declaration of Helsinki For the two weeks before the first night at the laboratory, participants were asked to fill out a questionnaire see below every morning to become familiar with reporting conscious experience upon awakening.

In total, we performed 46 overnight experiments 29 nights reported here and 13 wakefulness sessions during daytime. Participants were awakened during the night after each TMS session, see below by a computerized alarm sound lasting 1. We also requested from our participants that they inform us if they woke up during the administration of the TMS.

Interviews were then conducted at the bedside with a structured questionnaire and answers were audiotaped and later transcribed. Participants were instructed to report whether they had a conscious experience by delivering a dream report, report that they had had a conscious experience but did not remember its content, or report that they had experienced nothing. In the last two cases, they were further asked whether they were sure about their response.

We also asked each subject to report their subjective level of wakefulness. When subjects reported a conscious experience with recall of content, other questions concerning the content of their experience were also asked, such as the degree of thinking and perceiving not discussed in the present article. Electrical brain activity was recorded using a channel TMS-compatible EEG amplifier Nexstim eXimia, Nexstim Plc, Finland with a sample-and-hold circuit, which prevents the amplifier from saturation.

EEG was referenced to an additional channel on the forehead, filtered 0. After observing a minimum of three minutes of NREM sleep stages 2 and 3 , we applied single-pulse TMS to the medial superior parietal cortex superior parietal lobule and precuneus, see Supplementary Table S1.

We chose to target this area for several reasons. First, previous studies have identified it as a hub in brain networks, it is the area most often involved in neural correlates of consciousness and it includes extensive corticocortical connections 20 , 40 , Second, it can be conveniently stimulated without eliciting muscle artifacts Third, it is easily accessible while participants are asleep.

Finally, its TMS response is well-documented 4. The same target was stimulated in all night sessions. The head-tracker goggles were firmly taped on the EEG cap and the co-registration for the navigation was performed several times during the night to ensure exact targeting. The location of the maximum electric field induced by TMS in the cortex was on the convexity of the targeted gyrus with the induced current perpendicular to it and oriented so that the predominant stimulation direction was in the posterior—anterior direction.

The stimulation parameters were in accordance with published TMS guidelines The stimulation was discontinued at any noticeable sign of arousal. In 78 sessions 44 CE and 34 NCE , subjects woke up during the stimulation and the questions were asked without playing the alarm sound. To avoid auditory evoked potentials due to the TMS coil click, participants wore earphones with noise masking and a thin foam pad was placed between the scalp and the coil 11 , Between the TMS sessions, the coil was cooled using ice packs.

At the end of the experiments, the electrode positions and scalp landmarks nasion, left and right tragus were digitized. The digitization data were used to ensure the alignment of the EEG cap across multiple nights.

In total, we performed sessions in NREM sleep up to 16 sessions per night; 40, 40, 26, 44, 49 and 45 sessions for subjects 1—6, respectively. Bad channels were detected visually and later interpolated 48 ; seven sessions 1 CE and 6 NCE sessions that had over seven bad channels were excluded from the analysis.

The data were band-pass filtered between 1. The data were then epoched with respect to the TMS pulses, baseline-corrected using a ms-long baseline interval and average-referenced. Note that the data of subjects 1 and 4 were flipped with respect to the midline electrodes, as they were stimulated on the left hemisphere as opposed to the others.

In these subjects, we chose to target the left hemisphere because, during the daytime session, it provided better TMS-evoked EEG responses lower artifact level, larger response than sites in the right hemisphere. S1 and Supplementary Fig. We had in total and trials for the CE only including the reports of conscious experiences with and without the recall of content of which subjects were sure; trials for CE with recall of content and trials for CE without recall of content and NCE only including the answers of which subjects were sure conditions, respectively, with maximum 14 trials per session.

The number of trials for each subject is shown in Supplementary Table S2. We also did the same analyses separating NREM sleep into stages 2 and 3, which gave results similar to those reported in the present article. The statistical significance of the difference between the TMS-evoked responses in the CE and NCE conditions was assessed by cluster-based permutation statistics We then thresholded the data by discarding the values corresponding to P -values above 0.

Next, we computed cluster statistics by summing up the t -statistics of the neighboring data points neighbors in time and in the channel space; the discarded points defined the cluster borders. By these means, we found the channels and samples belonging to several candidate clusters.

To determine the significance of the clusters, we pooled the CE and NCE data, drew 10, random permutations and computed the cluster statistics for these randomized datasets as the maximum cluster value of each permutation. Finally, the statistics of the candidate clusters were compared against the statistics distribution of the randomized data to obtain their two-tailed P -values, adjusted for multiple comparisons. For both of the clusters, we averaged the data across the channels belonging to the clusters at the single-trial level.

The statistical significance of the differences between the CE and NCE conditions was again assessed by cluster-based permutation statistics as described above, except that now the neighbors occurred only in time.