1. Neuroscience

Cortical excitability signatures for the degree of sleepiness in human

  1. Chin-Hsuan Chia
  2. Xin-Wei Tang
  3. Yue Cao
  4. Hua-Teng Cao
  5. Wei Zhang
  6. Jun-Fa Wu
  7. Yu-Lian Zhu
  8. Ying Chen
  9. Yi Lin
  10. Yi Wu
  11. Zhe Zhang  Is a corresponding author
  12. Ti-Fei Yuan  Is a corresponding author
  13. Rui-Ping Hu  Is a corresponding author
  1. Department of Rehabilitation Medicine, Huashan Hospital, Fudan University, China
  2. Institute of Neuroscience, State Key Laboratory of Neuroscience, Center for Excellence in Brain Science and Intelligence Technology, Chinese Academy of Sciences, China
  3. Shanghai Center for Brain Science and Brain-Inspired Intelligence Technology, China
  4. Institute of Brain Science, Fudan University, China
  5. Shanghai Key Laboratory of Psychotic Disorders, Shanghai Mental HealthCenter, Shanghai Jiaotong University School of Medicine, China
  6. Co-innovation Center of Neuroregeneration, Nantong University, China
  7. Translational Research Institute of Brain and Brain-Like Intelligence, Shanghai Fourth People’s Hospital Affiliated to Tongji University School of Medicine, China
https://doi.org/10.7554/eLife.65099
Share this article
Cite this article
  1. Chin-Hsuan Chia
  2. Xin-Wei Tang
  3. Yue Cao
  4. Hua-Teng Cao
  5. Wei Zhang
  6. Jun-Fa Wu
  7. Yu-Lian Zhu
  8. Ying Chen
  9. Yi Lin
  10. Yi Wu
  11. Zhe Zhang
  12. Ti-Fei Yuan
  13. Rui-Ping Hu
(2021)
Cortical excitability signatures for the degree of sleepiness in human
eLife 10:e65099.
https://doi.org/10.7554/eLife.65099
  2. Download BibTeX
  3. Download .RIS

Abstract

Sleep is essential in maintaining physiological homeostasis in the brain. While the underlying mechanism is not fully understood, a ‘synaptic homeostasis’ theory has been proposed that synapses continue to strengthen during awake and undergo downscaling during sleep. This theory predicts that brain excitability increases with sleepiness. Here, we collected transcranial magnetic stimulation measurements in 38 subjects in a 34 hr program and decoded the relationship between cortical excitability and self-report sleepiness using advanced statistical methods. By utilizing a combination of partial least squares regression and mixed-effect models, we identified a robust pattern of excitability changes, which can quantitatively predict the degree of sleepiness. Moreover, we found that synaptic strengthen occurred in both excitatory and inhibitory connections after sleep deprivation. In sum, our study provides supportive evidence for the synaptic homeostasis theory in human sleep and clarifies the process of synaptic strength modulation during sleepiness.

Introduction

During sleep, brains undergo profound neurophysiological changes that restore the decline in cognitive functions associated with sleepiness (Harrison and Horne, 2000; Tononi and Cirelli, 2006). While this homeostatic process provides an important opportunity in studying the modulation of cognitive states, the key features of neural circuits underlying wakefulness, sleepiness, and sleep remain to be poorly understood. A synaptic homeostasis theory has been proposed to describe the biophysical change of neural circuits during sleep: wakefulness associates with strengthening of the synaptic connection, while sleep initiates synaptic weight downscaling and facilitates homeostasis (Tononi and Cirelli, 2006; Tononi and Cirelli, 2003). This theory reasons that constant learning and memory activities during awake lead to synaptic potentiation, thus prolonged awake period causes hyperactivity in the neural circuit, enhancing the noise among neural communications and disrupting cognitive functions (Tononi and Cirelli, 2003). It follows that the sleep pressure may be correlated to the degree of this hyperactivity. As increasing noise in the circuits adds barriers for information processing, the organism inevitably falls in sleep to restore the synaptic balance. In a prolonged sleep-deprived state (>24 hr), this hyperactivation may be partially restored, together with the subjective feeling of sleepiness, by circadian modulation (Borbély et al., 2016; Frank and Cantera, 2014).

In agreement with this synaptic homeostasis theory, biochemical markers of synaptic potentiation increase with prolonged wakefulness (Cirelli and Tononi, 2000; Tononi and Cirelli, 2001; Silva, 2003). In addition, resting-state electroencephalogram (EEG) revealed a global increase in theta band power (4–8 Hz) with increased sleepiness (Cajochen et al., 1995; Aeschbach et al., 1997; Vyazovskiy and Tobler, 2005), suggesting sleep-wake state exerts a robust modulation on the neural circuits. However, experiments attempting to pinpoint this modulation by directly measuring brain excitability have yielded less consistent results (Huber et al., 2013; De Gennaro et al., 2007; Manganotti et al., 2001). This is at least partially due to the technical challenges in measuring brain excitability, commonly via transcranial magnetic stimulation (TMS). TMS induces noninvasive activation of local brain region by applying a transient, alternating magnetic field through the skull, the effect of which can be readout through EEG or downstream motor output. Brain excitability can be measured by determining the minimal power required to achieve reliable output or the amplitude of the output given a standardized stimulation (Pascual-Leone et al., 1994). In addition, paired TMS pulses can be applied consecutively within a short temporal interval (e.g., several milliseconds) to induce certain facilitatory or inhibitory effect (Stefan et al., 2000). The nature of these paired-pulse effects has been studied extensively, and some of them have been attributed to the function of certain synaptic receptors (Van den Bos et al., 2018; Daskalakis et al., 2002). While TMS allows noninvasive assessment of cortical excitability in humans, it suffers from relatively high degree of inter-subject variability. Previous studies taking group averages of TMS measurements reported either hyper- (Huber et al., 2013) or hypo- (De Gennaro et al., 2007) excitability associated with sleep deprivation, with inconsistent effects of pair-pulse stimulation results (Manganotti et al., 2001; Kreuzer et al., 2011; Kuhn et al., 2016; Chellappa et al., 2016).

In this study, we attempted to circumvent the above-mentioned person-to-person variability by applying more advanced statistical methods to analyze the relationship between cortical excitability measured by TMS and self-report sleepiness. We collected TMS measurements in 38 healthy subjects in a continuous 34 hr study program. By utilizing a combination of partial least squares (PLS) regression and mixed-effect models, we identified a robust pattern of cortical excitability change that quantitatively predicts the degree of sleepiness in two separate subgroups of subjects. Interestingly, we found that both facilitatory and inhibitory changes in pair-pulse TMS enhance with sleepiness, while the overall excitability moderately decreases. While our results strongly support the synaptic homeostasis theory that synaptic connection strengthens overtime during wakefulness, we showed that the strengthening occurs in both excitatory and inhibitory synapses. Such changes may contribute to the overall maintenance, and subtle decline, of cognitive functions during prolonged wakefulness (Rubin et al., 2017). In sum, our study revealed strengthening of both excitatory and inhibitory synaptic transmission with sleepiness, clarifying and supporting a role of homeostatic modulation of synaptic strength by sleep.

Results

Sleep deprivation program

The awake period started after subjects arrived at the hospital and spent the previous night. During the 34 hr of study, the subjects were kept awake with two researchers and took eight times of Stanford Sleepiness Scales (SSS) evaluations and TMS assessments, and three times of EEG measurements (Figure 1A and B, Figure 1—figure supplement 1). Sleepiness showed an increased after 16 hr (group average SSS scores increased from 2.75 to 4.5), and then a slight decrease after 24 hr (from 4.5 to 3.25) (Figure 1C, Figure 1—figure supplement 2). This decrease in sleepiness agrees with previous sleep deprivation studies (Manganotti et al., 2001), likely due to circadian modulations. Furthermore, consistent with previous reports (Cajochen et al., 1995), EEG showed a stereotypical increase in the power of theta band (4–8 Hz), and a decrease in the alpha band (8–12 Hz) (Figure 1D and E, Figure 1—figure supplement 2). While all brain regions showed consistent global trends, we observed more prominent changes in the occipital areas, consistent with previous studies indicating these areas showed more prominent changes around sleep onsets (Gorgoni et al., 2019; Finelli et al., 2001). We did not observe an increase in the alpha and beta band in frontal regions, a prominent feature associated with sleep onset (Finelli et al., 2001; Marzano et al., 2013), suggesting that the subjects did not fall in and out sleep. In sum, the self-report sleepiness and EEG patterns provided quantitative measurements of the degree of sleepiness in our subjects throughout the study.

Figure 1 with 2 supplements see all
Download asset Open asset
Induction and validation of sleep deprivation in current study.

(A) Schematics of study design. All subjects received eight sessions of transcranial magnetic stimulation (TMS) measurements, but only 15 subjects received electroencephalogram (EEG) measurements (Figure 1—figure supplement 1). (B) Locations of EEG electrodes used in the study. Positionings are labeled following the extended international 10–20 system. (C) Measurements of self-report sleepiness using Stanford Sleepiness Scales (SSS). All individuals received eight measurements. Individuals’ scores were showed in gray and averaged responses were plotted in red curve (applied spline interpolation for smooth visualization). Statistical comparisons between sessions are reported in Figure 1—figure supplement 2. (D) Averaged heatmaps of different frequency bands’ powers in EEG data (N = 15). Interpolations between electrodes were calculated using triangulation-based cubic interpolation method. (E) Quantifications of frequency bands’ powers showed in (D) at different timepoints. Average powers among all electrodes were used for plotting. Red dotted plots show averaged SSS (from C). Data are represented as mean ± SD (in shaded areas). Powers at 21 hr and 28 hr are compared with the 4 hr ones using paired t-test with Bonferroni’s correction for multiple comparisons. Statistically significant points were labeled (*p<0.05; **p<0.01; ***p<0.001).

Raw TMS measurements showed weak correlations to sleepiness

We performed TMS stimulation on subjects’ motor cortices on the left hemispheres and measured motor-evoked potentials (MEPs) in first interosseous dorsal (FDI) muscle on the right hand (Figure 2A and B, Table 1). On average, the minimal power necessary for eliciting reliable MEPs decreased after the subjects woke up and mildly increased after 21 hr (Figure 2C). The delay between TMS and MEP followed a similar pattern (Figure 2E). The increase of minimal power and delay time indicates a reduction in excitability with prolonged wakefulness. However, individuals’ data often do not agree with the average trend (Figure 1—figure supplement 2), and correlations between individuals’ sleepiness score and these parameters showed wide ranges of variabilities (Figure 2D and F). The MT (motor response threshold) and MEP latency are influenced by the excitability along the corticospinal tract. In contrast, the pair-pulse protocols reflect characteristics within the cortex. Among four types of pair-pulse protocols (SICF, SICI, ICF, and LICI; Figure 2G, J, M and P), SICI, ICF, and LICI showed strengthening trends on average, yet none were statistically significant (Figure 2H, K, N and Q), nor any showed consistent correlations to SSS (Figure 2I, L, O and R). In summary, raw TMS measurements showed weak correlation with sleepiness on average, with considerable variance among subjects.

Figure 2
Download asset Open asset
Averages of transcranial magnetic stimulation (TMS) measurements show weak correlations to sleepiness.

(A) Position of TMS targeting. (B) Schematics of the TMS study design. The electroencephalogram (EMG) trace showed is for illustrative purpose only. (C) Averaged measurements of motor threshold. Data are represented as mean ± SD (in shaded areas). ANOVA test was applied to test the differences among timepoints; p-value is showed when statistically significant. (D) Distribution of the correlation coefficients between individuals’ sleepiness scores and motor threshold measurements (median in red line, 25% and 75% quantiles in dotted lines). One-sample t-test was used to compare the distribution to 0; p-value is showed when statistically significant. Equivalent to the formats of (C) and (D), later panels show data for motor-evoked potentials latencies (E, F), short interval intracortical facilitations (SICFs) (H, I), short interval intracortical inhibitions (SICIs) (K, L), intracortical facilitations (ICFs) (N, O), and long interval intracortical inhibition (LICI) (Q, R). For the four pair-pulse protocols, a schematic is showed for the respective protocol, (G) SICF, (J) SICI, (M) ICF, and (P) LICI. For all TMS measurements, N = 38 except for LICI (Q, R, N = 15, Figure 1—figure supplement 1). Statistical comparisons between sessions are reported in Figure 1—figure supplement 2.

Table 1
The result of the Stanford Sleepiness Scales (SSS) and transcranial magnetic stimulation (TMS).
SSSMT (% MSO)MEP latency (ms)SICFICFSICILICI
38383838383815n
2.68 ± 0.1890.49 ± 0.0221.67 ± 0.261.64 ± 0.101.14 ± 0.090.49 ± 0.040.39 ± 0.070 hr
1.76 ± 0.1220.45 ± 0.0221.46 ± 0.271.71 ± 0.131.29 ± 0.090.47 ± 0.050.35 ± 0.074 hr
2.26 ± 0.1800.45 ± 0.022.43 ± 0.251.72 ± 0.081.27 ± 0.080.42 ± 0.040.31 ± 0.0710 hr
2.34 ± 0.1430.45 ± 0.0221.48 ± 0.261.71 ± 0.111.39 ± 0.080.39 ± 0.040.26 ± 0.0716 hr
3.74 ± 0.1490.47 ± 0.0221.91 ± 0.231.58 ± 0.081.46 ± 0.140.38 ± 0.030.24 ± 0.0721 hr
4.50 ± 0.1910.46 ± 0.0221.86 ± 0.261.73 ± 0.111.46 ± 0.090.42 ± 0.030.25 ± 0.0624 hr
3.79 ± 0.210.45 ± 0.0221.66 ± 0.271.71 ± 0.0.91.33 ± 0.090.45 ± 0.040.31 ± 0.0928 hr
3.32 ± 0.2230.45 ± 0.0221.62 ± 0.261.75 ± 0.141.31 ± 0.100.46 ± 0.040.26 ± 0.0734 hr

  1. MT: motor response threshold; MSO: maximal stimulation output; MEP: motor-evoked potential; SICF: short interval intracortical facilitation; ICF: intracortical facilitation; SICI: short interval intracortical inhibition; LICI: long interval intracortical inhibition.

Sleepiness states can be mapped to latent dimensions in TMS measurements

Considering the individual variability in the TMS measurements, we hypothesized that each subject may have certain intrinsic noise pattern and group averages do not separate these noises from underlying correlation between TMS and sleepiness. In order to explore whether TMS measurements encode sleepiness states, we performed a PLS regression using six TMS parameters as predictors and sleepiness scores as responses (15 subjects, Figure 1—figure supplement 2). When we mapped the non-sleepy states (SSS = 1 or 2, blue dots) and sleepy states (SSS = 5, 6, or 7, red dots) on the first two PLS dimensions, we found that the first dimension showed significant separation between these two states, and a mild-sleepy state (SSS = 3 or 4) showed an intermediate peak (Figure 3A). Interestingly, while the first dimension explained only ~13 % of the total data variance, other PLS dimensions did not show separation of sleepiness states, suggesting that indeed the TMS data contain high variance, and a particular linear combination of the parameters captures the association to the degree of sleepiness. In comparison, similar analysis using the power of different EEG frequency bands from the same subjects showed obvious separation between sleepy states and non-sleepy states, yet the mild sleepiness showed a bimodal distribution, rather than forming an intermediate cluster (Figure 3B). Our results indicated that EEG data exhibits a phasic switch to sleepiness, while TMS data showed a correlation to the degree of sleepiness.

Figure 3 with 3 supplements see all
Download asset Open asset
Transcranial magnetic stimulation (TMS) measurements quantitatively predict the degree of sleepiness.

(A) Partial least squares (PLS) regression of TMS measurements to sleepiness scores (15 subjects with eight timepoints each, N = 120 states). The plot shows a scatter plot of each state in the first two PLS dimensions (Stanford Sleepiness Scales [SSS] = 1 or 2 for non-sleepy state, blue dots; SSS = 3 or 4 for mild sleepy state, purple dots; and SSS = 5, 6, or 7 for sleepy state, red dots). The heatmaps show density plots of the blue and red dots. And the distributions along each dimension are plotted on top and right side of the plot. Student’s t-tests were used to compare the distributions between sleepy and non-sleepy states, with p-value indicated if statistically significant. (B) PLS regression of electroencephalogram (EEG) measurements (15 subjects with three timepoints each, N = 45 states), similar to panel (A). (C) Log likelihood of different regression models. Increase in absolute values indicates a worse fit. Dash line indicates the level of full model. All models were compared to the full model using F-test of all four information criteria: Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), log likelihood and deviance; with asterisks showing statistical significance. The red bar indicates the model chosen for later analysis. (D) Mixed-effect linear model fitting results from the training set (15 subjects with eight timepoints each, N = 120 states). Green line shows mean ± SEM of the model predictions. p-Values of each coefficient was showed. (E) Mixed-effect linear model fitting results from the testing set (23 subjects with eight timepoints each, N = 184 states). All coefficients were kept the same with the training set in (D), except allowing a new random effect on intercept for each new subject. p-value was calculated with a F-test of a linear regression between predicted values to reported values to show the slope is significantly different from 0. Inset in (E) shows average performance from fivefold cross-validation compared to shuffled control. (F) Comparison of mixed-effect linear model fittings between TMS and EEG measurements, with balanced dataset (15 subjects with three timepoints each, N = 45 states for both). Heatmaps were calculated by fitting a normal distribution density function with the predicted values at each reported value bin.

In order to examine whether the observed correlations between sleepiness and TMS or EEG measurements are due to a confounding effect of circadian modulation, we conducted three groups of paired analyses for datapoints at 7 a.m. (sleep deprived 0 and 24 hr), 11 a.m. (sleep deprived 4 and 28 hr), and 5 p.m. (sleep deprived 10 and 34 hr). The separation between no sleepy and sleepy states remained primarily consistent, except for the 5 p.m. group with TMS measurements (Figure 3—figure supplements 1 and 2). The lack of correlation in the 5 p.m. group could be due to a lack of sufficient variation in the sleepy states in this timepoint. These data indicate that after controlling for the circadian modulation TMS and EEG measurements still correlate sleepiness.

TMS quantitatively predicts sleepiness

Based on the conclusion above, we further hypothesized that a mixed-effect model of TMS measurements should be able to predict subjects’ sleepiness, while tolerating the high degree of inter-subject variability. We established a linear full model using data from 15 subjects with all six TMS measurements to predict sleepiness scores (Figure 1—figure supplement 2), without considering any interaction among these measurements, and each subject was given a random effect on intercept. The model showed a statistically significant fit. We then performed backward variable selection based on F-test of the standard model information criterion (Figure 3C, plotting log likelihood as representative), and found that SICF and LICI can be eliminated without reducing the model performance. With the updated model, all four TMS parameters (MT, MEP latency, SICI, and ICF) showed statistically significant coefficients, and the predicted sleepiness showed high correlation to the reported sleepiness (Figure 3D). Using the same coefficients on these four TMS parameters, we predicted the other 23 subjects’ sleepiness (Figure 1—figure supplement 2), allowing a new random effect for each subject. Our prediction showed high degree of correlation to their reported sleepiness (Figure 3E), strongly indicating that these coefficients represent a stable pattern in the TMS that robustly correlated with the degree of sleepiness. Using these four parameters and data from all subjects, we performed fivefolded cross-validation of our model (Colby and Bair, 2013). Testing group showed performance close to the training group, while shuffled control generated chance-level prediction, further supporting the validity of our model. Similar analysis using TMS measurements from matched SSS states to predict time of the data showed no significant correlation (Figure 3—figure supplement 2B), indicating that circadian modulation contributed minimally in the TMS variance in our dataset.

In addition to predicting subjective self-reported sleepiness, we explored the group-level correlation between TMS and EEG measurements using canonical correlation analysis (Figure 3—figure supplement 3). We found that the most correlated mode between TMS and EEG is aligned with sleepiness, indicating that sleepiness is the primary covariate linking these two measurements. Finally, we applied the same mixed-effect regression using frequencies power averaged from all EEG electrodes and plotted the variance in the predictions at each degree of reported sleepiness. Consistent with the PLS analysis (Figure 3A and B), averaged EEG showed a robust detection of sleepiness, but the gradient was not distinguishable (Figure 3F). Together, these analyses showed a robust signature of TMS that uniquely reflects the degree of sleepiness in human.

Furthermore, examining the coefficients in the four TMS parameters revealed that all of them were positive. Therefore, our data indicate that sleepiness is associated with increased MT and MEP latency, both pointing to a decrease in the corticospinal tract excitability. Sleepiness is also associated with enhancement in both facilitatory (ICF) and inhibitory (SICI) pair-pulse effects, suggesting the strengthening of both excitatory and inhibitory synapses during prolonged wakefulness.

Discussion

The present study reported that a combination of four TMS parameters (rMT, MEP latency, SICI, ICF) could efficiently predict subjective sleepiness. The values of rMT and MEP latency together directly reflect and measure the excitability states of motor cortical region. Enhanced rMT has been reported using similar sleep-deprivation paradigm (Manganotti et al., 2001). And increases in cortical excitability were reported using local EEG measurements of TMS-evoked responses (Huber et al., 2013). On the other hand, the pair-pulse parameters, SICI and ICF, reflect local changes in the cortex. SICI is believed to be mediated by GABAAR, and ICF is mediated by glutamatergic transmission, potentially through NMDAR (Van den Bos et al., 2018). Therefore, our study suggested that both excitatory and inhibitory synapses are strengthening with prolonged wakefulness. Previous studies demonstrated that cortical excitability can be sensitive to many variables including age (Gaggioni et al., 2019), attention (Cardone et al., 2021), and circadian rhythm (Chellappa et al., 2016; Ly et al., 2016). Among these, effects of circadian rhythm are challenging to disambiguate since sleepiness correlates to circadian factors (Figure 1). Thanks to the prolonged durations of our study, we were able to compare timepoints 24 hr apart (Figure 3—figure supplement 2), which were not available to previous studies (Chellappa et al., 2016; Ly et al., 2016). We found that the correlation between TMS measurements and sleepiness remains significant after controlling for the circadian factors; but TMS measurements do not show correlation to circadian timing within each SSS subgroup. These data strongly suggest that the variance in cortical excitability is primarily correlated with sleepiness in our dataset. However, in other situations with short or no sleep deprivation, it is still likely that circadian factors are correlated with cortical excitability.

This finding agrees with the general theme of the synaptic homeostasis theory of sleep: wakefulness is associated with continuous strengthening of the synapses. Importantly, we provided an additional supplement to this theory that the synaptic strengthening does not only happen in excitatory synapses, but also in inhibitory synapses. Lacking the exact mechanism of these changes, it is possible the observed strengthening in both types of synapses was independently modulated by sleepiness or is due to the circuit response to changes in one type of the synapse in order to maintain the excitation/inhibition balance (E/I balance). Thus, in contrast to the original prediction of the theory that sleepiness leads to overt hyperactivity, we showed that there is a robust E/I balance. Nevertheless, the strengthening of synapses while maintaining balance can still have a detrimental effect on the information processing in the circuits, as demonstrated by multiple studies simulating these effects. Deviating from the optimal range of the synaptic strengths will lead to a reduction in the signal-to-noise ratio in individual neurons (Rubin et al., 2017), as well as the suboptimal sparseness and noise correlations in the population-wide coding (Zhou and Yu, 2018). Sleep, then, may depress these synaptic weights and restore the circuits to an optimal state (Kuhn et al., 2016).

While the above-mentioned modification of the synaptic homeostasis theory does not change the general framework of sleep function, it is worth noting that the difference is not just a pedantic discussion. Our finding predicts that general excitability will not be a reliable marker for sleepiness. In addition, if any intervention were to develop to relief sleep drive, our finding suggests that prolonged synaptic suppression will not serve a restorative function. These predictions differ from the ones posited from the original synaptic homeostasis theory and will have important implications in the future research on the monitoring and modulation of sleep states.

Finally, we reported that a linear combination of TMS parameters can predict the gradual extent for the degree of sleepiness, while using the averaged EEG data only categorically separated non-sleepy and sleepy states. Although EEG was only recorded on three timepoints, limiting the TMS model using the same amount of data still produced a gradient prediction for the degree of sleepiness (Figure 3), suggesting that there is likely to be some intrinsic different modes of modulation to the sources of EEG and TMS measurements by sleepiness. As many frequency bands are empirically and potentially mechanistically linked to neural network function (Başar et al., 2001; Klimesch, 1999), the binary switch of EEG states may reflect a robust scheme of neural computation up to a high degree of sleepiness, and the change in the EEG state is potentially more predictive to the cognitive function decline related to sleepiness, compared to changes in the TMS measurements. Yet the synaptic changes revealed by TMS measurements may represent a mechanism underlying the EEG state switch as the synaptic strengths could impose biophysical boundaries and stable-state landscapes that force the switch in rhythmic activities states. On the other hand, a previous EEG and intracranial recording study showed that sleep onset is associated with a strong theta rhythm in the occipital regions and an increase in alpha and beta rhythm in the frontal regions (Marzano et al., 2013), suggesting that sleep deprivation may lead to different changes in rhythmic activities and potentially excitabilities in different brain regions. It remains to be seen whether our current finding represents a general feature across the brain.

Although our sleep deprivation paradigm exceeded the duration of many previous studies, one limitation of the current study is the limited datapoints of level 6–7 sleepiness. While we observed robust correlation with SSS levels of 1–5, the statistical power for very sleepy states is lower. This may partially explain the lack of correlation in one of the subsets in our results (Figure 3—figure supplement 2).

In conclusion, our study showed that the strengthening of both excitatory and inhibitory synaptic connections in the cortex can quantitatively predict the degree of sleepiness. This finding indicates a modification of the synaptic homeostasis theory of sleep and furthers our understanding of how sleepiness state modulates brain functions.

Materials and methods

Participants

Thirty-eight healthy participants (21 males, 17 females, age: 22.46 ± 0.29 years, age range: 20–27 years, Figure 1—figure supplement 1) with no history of sleep disorders (Pittsburgh Sleep Quality Index; Buysse et al., 1989), head trauma, psychiatric conditions, or any other chronic disease were recruited for present study. The subjects were free of medication and tobacco use. All participants wrote informed consent before the experiment; the study has been approved by the Ethics Committee of Huashan Hospital (2017-410).

Study flow

Request a detailed protocol

The subjects were asked to stay in their regular sleep pattern at least 1 week before the experiment, and the experiment would postpone if the participants have a cold or any other uncomfortable situation. The subjects arrived at the laboratory 1 day before the experiment for preparation and to adapt to the environment. They slept in the separate and quiet room for one night. They were asked not to use the alarm and are wakened by the researchers at 7 a.m. of the second day.

The study lasts for 34 hr, during which the subjects received eight times for SSS evaluation and TMS assessments, as well as three times resting-state EEG measurements (Figure 1). SSS was used in our study due to its wide usage in research, previous validation against objective measurements, and brevity (Hungs, 2012). During the study period, the subjects were not allowed engaging in vigorous exercise or taking coffee or tea, and were maintained awake by two independent researchers.

EEG recording and data processing

Request a detailed protocol

The EEG signal was recorded by EEG system of BrainAmp MR32, BrainProducts, using a 32-channel EEG cap. A specific electrode was used as the reference channel placed between Cz and Fz. Reference would then be transferred offline into overall average reference. Subjects were seated in a comfortable chair with a computer screen in front of them showing instructions when recording the resting state EEG. Subjects were asked to focus on a cross symbol on the screen for 5 min to record their open-eye EEG and followed by another 5 min of close-eye EEG epoch and then repeated for totally 20 min EEG recording with two episodes of open and close eye EEG, respectively. The EEG was acquired at 4, 21, and 28 hr from awake.

The collected EEG signals were processed offline using MATLAB (The MathWorks Inc, Natick, MA). Primary processing included eye-blink correction, artifact rejection, segmentation, and band-pass filter. The eye-blinks were removed from the EEG signal in other channels to eliminate the effects of electro-ocular artifacts. High-frequency artifacts, such as muscle activity, and high-amplitude slow wave were rejected based on the automatic removal algorithm with an initial threshold of 70 μV, which would continuously increase 5 μV at a time if more than 20% of the data was rejected until the threshold reached 150 μV. According to the study protocol, the EEG signal was then segmented by 1 min and following band filter of 0.5–49 Hz as beta wave was often interested in sleep deprivation studies.

The EEG signals were decomposed into the delta (1–4 Hz), theta (4–8 Hz), alpha (8–12 Hz), and low-beta (12–20 Hz) frequency components, spectral power (μV2) of which was analyzed separately via fast Fourier transform (FFT) for overall average at the three timepoints, respectively.

TMS procedures

Request a detailed protocol

The subjects were seated on a comfortable chair in a silent environment. TMS studied was conducted with OSF–priming TMS (YRD CCY-IA, Yiruide Co., Wuhan, China) connected with a 70 mm figure-of-8 coil (CCY-I TMS instrument, Yiruide Co.). The motor hotspot for FDI muscle on the contralateral hemisphere of primary motor cortex (M1) was defined with maximum MEP value. The MT is defined as the lowest density that can be evoked at least 5 MEP, with amplitude more than 50 μV in 10 times of successive stimulation with the relaxed recording muscle (Rossini et al., 2015). Then, the intensity of the test stimulation (TS) was determined when the average amplitude reached at 400–1000 μV. The TS was given 10 times continuously, and their latency and amplitude were recorded.

The paired-pulses TMS (ppTMS) included short interval intracortical inhibition (SICI), long interval intracortical inhibition (LICI), short interval intracortical facilitation (SICF), and intracortical facilitation (ICF). The ppTMS also included three controllable parameters, TS, conditioning stimulation (CS), and interstimulus intervals (ISI). The setting of ppTMS in this study is shown in Table 2 (Rossini et al., 2015; Lazzaro and Ziemann, 2013). Every index was repeated 10 times, and the single pulse of the TS was inserted between the pulses to guarantee the accurate stimulation site and the angle of the coil. The pause was set within 5–7 s between the pulses to make sure the excitability would not have been interfered by the last stimulation. The format of the index of the ppTMS is shown below. The 10 times stimulation of each index were averaged.

ppTMS%MEP=amplitude of ppTMSamplitude of TS×100%
Table 2
The setting of the ppTMS.
SICILICISICFICF
CS (%MEP)701209070
TS (%MEP)Average amplitude reached at 400–1000 μV
ISI21501.515

  1. ppTMS: paired-pulses transcranial magnetic stimulation; SICI: short interval intracortical inhibition; LICI: long interval intracortical inhibition; SICF: short interval intracortical facilitation; ICF: intracortical facilitation; MEP: motor-evoked potential; CS: conditioning stimulation; TS: test stimulation: ISI: interstimulus interval.

Statistical analysis

Request a detailed protocol

The data were repeated measurement variables of single group. One-way ANOVA was used for statistics, and the sample size was calculated based on the data from previous studies. TMS and SSS were repeated seven times and EEG three times. PASS 11 was used to calculate the sample size of 34 (power = 0.9, α = 0.05 [double-sided]). If the lost rate was assumed to be 10%, then the sample size was 34/0.9 = 38 cases. In the actual study, 40 cases were included and lost 2 cases.

All statistical analyses were performed with GraphPad Prism (GraphPad Software LLC, USA). Paired t-tests were used in EEG dataset (Figure 1). One-way ANOVA was applied in the result of TMS measurements and one-sample t-test for the correlations between TMS and SSS (Figure 2). The statistical significance threshold was set as two-tailed, p<0.05. F-tests were used in most of the model-related coefficient comparisons (Figure 3). Data are presented as mean ± standard deviation or mean ± standard error of the mean.

Partial least-square regressions were performed with MATLAB, using SSS as responses. In the case of TMS measurements, all six TMS parameters were used as predictors and output was set with five dimensions. In the case of EEG measurements, all four EEG bands were used as predictors and output was set with three dimensions. The correlation between SSS and the predictor scores for each output dimensions was examined, and the first two dimensions were plotted.

Mixed-effect linear regressions were performed with MATLAB, using SSS as responses. All parameters were normalized to a 0–1 scale according to group extrema before used in the model. Pilot explorations adding sex and age as predictor variables show no effect of these variables. Additionally, using all subjects’ data, all random effects and residuals follow normal distribution and are not correlated, suggesting compliance to the core assumption of the mixed-effect model. No interaction term was allowed, and the random effect was limited to only the intercept for each subject. Backward model selection was performed based on information criteria compared to the full model. It should be noted that parameters not selected may still change with SSS, but would not provide further information in a prediction model. To apply the fitted model in a new group, we calculated the fixed effects of each new subject using the fitted coefficients and then fit a new model adding only a random intercept for each new subject.

For cross-validation of our model, we adopted the methods described in a previous study (Colby and Bair, 2013). In brief, for each iteration, all subjects were randomly assigned into training (23 out of 38) or testing (15 out of 38) group. Within training group, 18 subjects were used to estimate model coefficients. After freezing the fixed-effects coefficients, random effects were estimated for the remaining five subjects and residuals from these five subjects were used as model performance. This within-training group sampling was repeated 500 times and the best model is selected. Using the fixed-effect coefficients from this model, we re-evaluated the random effects from all 23 subjects and calculated the residuals as training group performance. And then we applied this model to the 15 subjects and calculated the random effects and residuals, which were recorded as testing group performance. This concluded one iteration of the cross-validation process. 500 iterations were performed on the dataset, and averaged performance was reported in the results. We conducted the same procedures in randomly shuffled SSS data, and the testing performance in that group is recorded. To estimate chance level, we draw 38 random number from 1 to 7 following a normal distribution as model prediction and calculated the errors. This procedure was repeated 500 times to get an average chance level.

Data availability

All data generated or analysed during this study are included in the manuscript and supporting files.

References

Decision letter

  1. Laura Dugué
    Reviewing Editor; Uni­ver­sité de Paris, France
  2. John R Huguenard
    Senior Editor; Stanford University School of Medicine, United States
  3. Giulia Gaggioni
    Reviewer; University of Liège, Belgium

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

This paper is of potential interest to neuroscientists and sleep researchers, as it links measures of cortical excitability to self-reported sleepiness. The dataset is interesting given all the various measures reported. The authors revised the manuscript in depth according to the reviewers' suggestions, with some new control analyses, including the relative independence from circadian factors.

Decision letter after peer review:

Thank you for submitting your article "Cortical excitability signatures for the degree of sleepiness in human" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and John Huguenard as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Giulia Gaggioni (Reviewer #2).

Based on the reviews, and as detailed below we would like to see a revised version of your manuscript before making a final decision. The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1. The main concern raised by all three reviewers regards the impossibility to disentangle between the contributions of circadian vs. homeostatic factors on the observed effects. The analyses have been performed only considering the homeostatic factor (i.e., the time spent awake). We recommend that you address this concern by the following additional analysis: comparisons between data points collected at the same time-of-day (i.e., controlling the circadian factor): 3 data points for the TMS measures [7 a.m, 11 a.m., and 5 p.m. of the first and second day (before vs. after sleep deprivation)] and 1 data point for EEG measures (11 a.m. of the first and second day).

2. Linking the wakefulness measure to synaptic homeostasis theory may be possible via literature references. It would be important to explain (in the introduction, at least) what synaptic homeostasis theory predicts about a >24h sleep-deprivation intervention including the 'rebound' effect on sleepiness after 24h, since that is what the authors have done.

3. It would also make sense to test the model more rigorously with something like 5-fold cross-validation.

4. Line 352: "Pilot explorations adding sex and age as predictor variables show no effect of these variables." It would be interesting to know if the predictor "session" was also tested.

5. Figures 1.C. and 2.C-E.: if there are any significant differences between sessions, please report it in the figure.

6. Figure 1.D..: could you discuss the higher δ and theta activity observed in occipital areas (instead of fronto-central areas)?

7. Line 128: MT abbreviation without previous definition in the text (but in line 307).

8. Methods / Statistical analysis: specify if the main assumptions required by mixed-effect model where checked and met.

9. Methods / Study flow: give a brief rational why SSS (instead of, for example, KSS, VAS, etc.).

10. An interesting analysis could be to test if the model could also predict the EEG activity (mean or theta), as an objective marker of sleepiness (even though EEG was only recorded 3 times in a subgroup of 15 participants).

11. The study participants do not actually report feeling very sleepy. The key behavioral measure is self-reported sleepiness on the Stanford Sleepiness Scale which is a 7-point scale, with only 6-7 indicating the participant is 'sleepy'. However participants reported level 6-7 sleepiness only at approximately 15 observations (fig3D, grey dots). This is from 38 participants * 8 testing moments i.e., >300 total observations. If participants were only actually sleepy in <5% of observations this could weaken the conclusions about sleep and sleepiness. And the clustering of the observations around the lower points 1-5 on the Sleepiness Scale might reduce confidence in the model fit. This is not necessarily a big issue but it might be sensible to acknowledge it in the discussion.

Reviewer #1:

This study investigates in 38 healthy subjects the association between cortical excitability increases, as assessed by transcranial magnetic stimulation (TMS), and sleepiness induced by a 34-hour protocol of sleep deprivation.

Using advanced statistical methods, the study confirms that the cortical excitability changes are associated to the degree of sleepiness. They also report that synaptic strengthen after sleep deprivation occurred in both excitatory and inhibitory connections.

On the whole, the study is essentially replicative. Differently from what stated by the authors ("Previous studies taking group averages of TMS measurements reported either hyper- [10] or hypo- [11] excitability associated with sleep deprivation, with inconsistent effects of pair-pulse stimulation results [12, 17-19]."), the existing studies are mostly coherent. Refs. #11, #12, and 18 substantially confirm the hypothesis. Refs. #12 and #17 are methodologically weak studies providing inconsistent findings, while Ref. #19 highlights the contribution of both circadian-timing and prior sleep-wake history in regulating cortical excitation/inhibition balance.

On the other hand, the most novel aspect of the current study is represented by the use of advanced statistical methods, which allow to control intrinsic individual variability in the TMS measures. The second novel finding of the study is that the synaptic strengthening does not only happen in excitatory synapses, but also in inhibitory synapses.

Taking into consideration the protocol (please see panel A of Figure 1), TMS measures were collected in eight different datapoints [7 a.m., 11 a.m., 5 p.m., and 11 p.m. of the first day (i.e., without any sleep deprivation) and 4 a.m. 7 a.m. 11 a.m. 5 p.m. of the second day (i.e., after sleep deprivation)]. On the other hand, EEG measures were collected at 11 p.m. of the first day and at 4 a.m. and 11 a.m. of the second day.

The study seems not considering that consolidated models of sleep regulation (Borbély AA, Daan S, Wirz-Justice A, Deboer T. The two-process model of sleep regulation: a reappraisal. J Sleep Res. 2016 Apr;25(2):131-43) state that nightly sleep and diurnal sleep pressure are affected and regulated by circadian-timing (i.e., the circadian process) and prior sleep-wake history (i.e., the homeostatic factor). This means that considering the 8 TMS datapoints and the 3 EEG datapoints intrinsecally mix the effects of circadian and homeostatic factors.

According to this very basic point, the current results can't disentangle in any way the contributions of the two factors. The analyses have been performed only considering the homeostatic factor (i.e., the time spent awake).

Reviewer #2:

In this study, Zhang and colleagues tracked the time course of sleepiness (self-reported), of the EEG, and of 6 indices of cortical excitability (MT, MEP latency, SICF, SICI, ICF, LICI), measured by TMS and paired-pulse TMS at the motor cortex, during 34 hours of sleep deprivation in 38 subjects (EEG and SICI measured in a subgroup of 15 subjects in 3 out of 8 sessions). Only MT and MEP latency showed a significant change with time spent awake. MEP latency was also significantly associated with self-reported sleepiness. The profile of the 6 raw indices showed high variability between individuals. Thus, the authors decided to apply advanced statistical methods such as partial least squares (PLS) regression and mixed-effect models to further analyse the relationship between cortical excitability measures and self-reported sleepiness. Zhang at al. identified a model based on the combination of 4 indices of cortical excitability (MT, MEP latency, SICI, ICF) that quantitatively predicts the degree of sleepiness in two separate subgroups of subjects. Furthermore, based on the regression coefficients of the model, they found that -while corticospinal excitability decreases- both facilitatory and inhibitory changes enhance with sleepiness. These data from a human in vivo protocol add in an interesting twist to the current frame of the SHY theory. The conclusions of this paper are mostly well supported by data driven from the model. However, the raw data are difficult to decipher. Some aspects of the volunteers' recruitment and final claims need to be clarified and extended.

Strengths

The duration and the number of sessions of the protocol are remarkable, as well as the number of participants. The design of the study is articulated and includes questionnaires, EEG and TMS / ppTMS over the motor cortex. Zhang at al. investigated both cortex excitability (i.e. reactivity) through motor-evoked potentials as well as the excitatory and inhibitory balance directly at the motor cortex. In this paper, cortical excitability measures are linked to self-reported sleepiness during an extended wakefulness, possibly representing the neurophysiological substrates of various aspects of human cognitive performance during sleep deprivation.

The authors applied a combination of PLS and mixed model and showed that self-reported sleepiness can be elegantly predicted by a set of indices of cortical excitability. This model allows a fine-grained prediction compared to the bymodal EEG one. This model recalls and further disentangles the positive association between cortical excitability (TEP amplitude) and subjective sleepiness previously reported during a 29 hrs sleep deprivation protocol under constant routine conditions (Ly, J. Q. M. et al. Circadian regulation of human cortical excitability. Nat. Commun. 7:11828 doi: 10.1038/ncomms11828 (2016)).

Weaknesses

Person-to-person variability of the raw data, resulting in weak correlation of the raw data with self-reported sleepiness. This inter-variability could potentially have been minimised, at least in part, by:

1. Designing the sessions based on time elapsed since awake (0h, +4h, +10h, etc.), instead of fixed external clock time (7AM, 11AM, 5PM, etc.), which could have been in contradiction with the individual's normal sleep-wake timing. By designing sessions in function of the individual's timing of awakeness, sessions would have been in line with the sleep-wake pattern of each individual and still comparable between subjects, without the disadvantage of potentially impacting sleepiness and the underlying neurophysiology.

2. Furthermore, since one of the aims was to predict sleepiness, it would have been good practice to administer sleepiness (e.g. Epworth Sleepiness Scale (ESS)) and chronotype questionnaires in the recruitment phase. Ideally, it would have been also better to quantify the sleep-wake pattern of volunteers during the week before the study, ideally with actigraph or at least with sleep diaries.

Clarification and extension of the main conclusion (e.g. as reported in lines 92-94).

1. "While our results strongly support the synaptic homeostasis theory that synaptic connection strengthens overtime during wakefulness, […]". Authors should explain and reconcile the reduction in excitability during sleep deprivation (as inferred from MT and MEP latency) with the synaptic homeostasis theory (and other studies based on TEP).

2. "[…], we showed that the strengthening occurs in both excitatory and inhibitory synapses." Authors should clarify that this conclusion is based on a mixed-effect model and it is not supported by the raw data, which do not show any significant modulation of the excitatory and inhibitory profiles with time spent awake. Furthermore, it would be beneficial for the paper's clarity if the authors report the β values as well as their "R2", and eventually depict this with a graph. Finally, the authors reported "Pilot explorations adding sex and age as predictor variables show no effect of these variables." (line 352). It would be interesting to know if the predictor "session" was also tested in pilot explorations.

Reviewer #3:

Zhang and colleagues investigated the link between cortical excitability and self-reported sleepiness using a combination of resting-state electroencephalogram (EEG), non-invasive brain stimulation (TMS), and a sleep-deprivation intervention. Their key finding is that a subset of TMS measures including the amplitude and latency of the motor-evoked potential, plus measures of intra-cortical inhibition and facilitation, collectively predict a participants' degree of subjective sleepiness across the sleep-deprived period.

The underlying data for this paper is very strong. The authors have collected data from a good number of volunteers (N=38), and have acquired a broad range of TMS measurements; 6 different measurements acquired up to 8 times each. This richness has many advantages: It ensures good statistical power and enables the core mixed-model approach. It also allows the authors to link their results to deeper mechanisms, e.g. by including paired-pulse protocols that probe both cortical excitation and inhibition. That's really important for linking the results to models of sleepiness specified at the synaptic level, rather than just correlating TMS and sleepiness.

However some aspects of the paper require further attention. One is implementational and related to the model fitting. One is conceptual and relates to the behavioural self-reports.

1) The authors claim that the 4 TMS measures predict scores on the sleepiness scale. Therefore I was expecting some kind of cross-validation: partition data into subsets, obtain regression co-efficients from one subset, test the model fit on the other subset, repeat for many partitions. The authors do something a little less robust: they obtain regression co-efficients from 15 subjects and test on the other 23. Only partitioning once seems insufficient, and I couldn't find justifications for the choice of training on n = 15 testing on n = 23. It seems important to rule out that the results depend on the details of their fitting procedure, and to show that the model fit is truly robust.

2) The authors have demonstrated a link between TMS measures of cortical excitability and self-reported sleepiness. However, the link between self-reported sleepiness and synaptic potentiation seems weaker. The authors explain in the intro that learning during wakefulness leads to synaptic potentiation, and eventually circuit hyperactivity that is rebalanced by sleep. In that case, shouldn't longer wake time lead to more hyperactivity? In fact, in their data longer wake time does not lead to greater subjective sleepiness; after 24h of wakefulness the participants report feeling relatively wakeful again (Figure 1C). Is this prediction made by the synaptic homeostasis theory? If so, the paper would be stronger if the authors explained the prediction in the introduction. If not, then it's hard to see how results pertaining to subjective sleepiness support a conclusion about synaptic homeostasis.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Cortical excitability signatures for the degree of sleepiness in human" for further consideration by eLife. Your revised article has been evaluated by John Huguenard (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below:

The authors made a deep and accurate revision, with some new control analyses. They responded to the points that were raised by the reviewers, and the manuscript is now quite improved. Two concerns need further consideration:

1. The authors provide some support to the relative independence from circadian factors. Please discuss the current results within the context of previous studies of cortical excitability as a function of circadian factors (e.g. Chellappa et al. Circadian dynamics in measures of cortical excitation and inhibition balance. Sci Rep. 2016 Sep 21;6:33661. doi: 10.1038/srep33661).

2. Concerning the theta and α activity across sleep onset, please also consider a preliminary finding on intracranial recordings of a pharmaco-resistant patient with epilepsy [supplementary Figure 1 in Marzano et al., 2013 (Sleep Med. 2013 Nov;14(11):1112-22. doi: 10.1016/j.sleep.2013.05.021)]

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Cortical excitability signatures for the degree of sleepiness in human" for further consideration by eLife. Your revised article has been evaluated by John Huguenard (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed. In particular, we note that there were apparently some significant changes in the author list, which is a bit unusual at this stage of the process, as the current revisions were relatively minor. Can you please provide a rationale for each of the authorship changes?

https://doi.org/10.7554/eLife.65099.sa1

Author response

Essential revisions:

1. The main concern raised by all three reviewers regards the impossibility to disentangle between the contributions of circadian vs. homeostatic factors on the observed effects. The analyses have been performed only considering the homeostatic factor (i.e., the time spent awake). We recommend that you address this concern by the following additional analysis: comparisons between data points collected at the same time-of-day (i.e., controlling the circadian factor): 3 data points for the TMS measures [7 a.m, 11 a.m., and 5 p.m. of the first and second day (before vs. after sleep deprivation)] and 1 data point for EEG measures (11 a.m. of the first and second day).

We thank the reviewers for pointing out circadian factors can be a confounding factor in our analyses. Following the suggestions from the reviewer, we now implemented PLS regression (similar to Figure 3A) on matched timepoints with TMS (new Supp. Figure 4A) and EEG data (new supp. Figure 3B) towards SSS. We found that for the 7 a.m. and 11 a.m. timepoints, both TMS and EEG data show obvious separation between sleepiness states, indicating that these measurements indeed represent a strong homeostatic component that is independent of circadian effect. We did not observe good separation in the 5 p.m. TMS group. We argue that this is due to the fact that within this timepoint the average sleepiness showed small difference.

In addition, to further test whether the variance in TMS measurements reflect homeostatic (sleepiness) factors or circadian factors, we grouped similar SSS datapoints and constructed mixed-effect linear models between TMS and hour of the day (new Supp. Figure 4B). In all groups, we did not observe correlation between TMS measurements and circadian factors, corroborating our conclusion that TMS measurements reflect homeostatic factors.

2. Linking the wakefulness measure to synaptic homeostasis theory may be possible via literature references. It would be important to explain (in the introduction, at least) what synaptic homeostasis theory predicts about a >24h sleep-deprivation intervention including the 'rebound' effect on sleepiness after 24h, since that is what the authors have done.

We have now included text introducing the circadian effect in the introduction section (line 61-63). Additional analyses related to circadian effects are now included in the results and Discussion sections.

3. It would also make sense to test the model more rigorously with something like 5-fold cross-validation.

We agree with the reviewers that cross-validation is a common practice for machine learning analysis. In general, however, there is no consensus on how to perform cross-validation for mixed-effect models due to the nature of random effects cannot be predicted for new subjects. It is an area of active research in the field of statistics regarding how to circumvent this issue. Currently, potential solutions including estimating information criteria of the model (Statistical Science, 2013, 28(2), 135–167; Journal of Data Science, 2011, 9, 15-21) or using an iterative method to estimate random effects (Journal of Pharmacokinetics and Pharmacodynamics, 2013, 40, 243–252).

We implemented the second method and used the prediction error as the metrics for measuring model accuracy. Among 38 subjects, we used 60% (23 subjects) for training and 40% (15 subjects) for testing. Within the training subjects, we used a 5-fold random sampling to choose 18 subjects for estimating the initial model, and 5 subjects for evaluating the error. This is repeated for 500 times and the best model is refitted using the 23 subjects for estimating the random effects and calculating the training error. Next, we applied this model to the 15 subjects testing set by freezing the fixed effect and estimating the random effects, and finally calculating the testing error. We repeat this training-testing process for 500 times to get a robust and generalizable result. As a negative control, we used randomly shuffled SSS dataset for the same procedure.

These cross-validation analyses are plotted in the new Figure 3E. The training and testing errors are close to each other, indicating robust predicting power from TMS measurement to SSS. On the other hand, shuffled data produced chance level estimation, further supporting the validity of the model.

4. Line 352: "Pilot explorations adding sex and age as predictor variables show no effect of these variables." It would be interesting to know if the predictor "session" was also tested.

We have tested “session” as a predictor as the reviewers requested. We coded session into numeral 1-8 in the model. The result showed that “session” showed significant correlation to SSS:

Author response table 1
Fixed effects coefficients (95% CIs):
NameEstimateSEtStatDFpValueLowerUpper
{'Session'}0.317130.0430977.35841152.99E-110.231760.4025
{'MT' }0.992750.301343.29441150.0013120.395841.5897
{'MEPL' }1.25780.277014.54071151.39E-050.709111.8065
{'ICF' }0.0998220.311450.320511150.74916-0.51710.71674
{'SICI' }0.481920.301821.59671150.11307-0.115931.0798

However, we do not think “session” is a very meaningful variable as it contains mixed information about homeostatic and circadian information. As shown in Figure 1C, SSS is grossly correlated with session. Thus, this correlation is quite expected.

5. Figures 1.C. and 2.C-E.: if there are any significant differences between sessions, please report it in the figure.

We have now provided the pair-wise post-hoc tests for these panels in Supp. Figure 2.

6. Figure 1.D..: could you discuss the higher δ and theta activity observed in occipital areas (instead of fronto-central areas)?

We have now added this discussion in the result section (line 115-118). In summary, we observed consistent trends with data from individual EEG electrode but indeed see more prominent changes in the occipital regions. This is consistent with previous studies showing that sleep-associated changes happen most prominently in the occipital regions. This may reflect that our subjects are drifting in and out of sleep states throughout the program.

7. Line 128: MT abbreviation without previous definition in the text (but in line 307).

We have added the definition at its first appearance in the text (line 134).

8. Methods / Statistical analysis: specify if the main assumptions required by mixed-effect model were checked and met.

The core assumption of mixed-effect model is that the residuals and random effect coefficients are independent and identically distributed (Methods in Ecology and Evolution 11(9), 1141-1152, 2020). We have plotted these variables in our model:

We have added the description of this calculation in the method section (line 406-421), without including Author response image 1 , as we feel that this might be too technical for the readers.

Author response image 1
Download asset Open asset

9. Methods / Study flow: give a brief rational why SSS (instead of, for example, KSS, VAS, etc.).

We have added this information in the method section (line 310-311).

10. An interesting analysis could be to test if the model could also predict the EEG activity (mean or theta), as an objective marker of sleepiness (even though EEG was only recorded 3 times in a subgroup of 15 participants).

We thank the reviewers for the suggestion. We applied canonical correlation analysis between TMS and EEG measurements due to their multi-dimensional nature. We reported the results in the new Supp. Figure 5. There are 2 modes between TMS and EEG show significant correlations. The first mode primarily aligns with the sleepiness measurements, but not the second mode. This result indicates sleepiness is a major covariate that correlates between EEG and TMS.

11. The study participants do not actually report feeling very sleepy. The key behavioral measure is self-reported sleepiness on the Stanford Sleepiness Scale which is a 7-point scale, with only 6-7 indicating the participant is 'sleepy'. However participants reported level 6-7 sleepiness only at approximately 15 observations (fig3D, grey dots). This is from 38 participants * 8 testing moments i.e., >300 total observations. If participants were only actually sleepy in <5% of observations this could weaken the conclusions about sleep and sleepiness. And the clustering of the observations around the lower points 1-5 on the Sleepiness Scale might reduce confidence in the model fit. This is not necessarily a big issue but it might be sensible to acknowledge it in the discussion.

We agree with the reviewers that it would have been better if more level 6-7 datapoints are in the model. Nevertheless, our model does show a gradient effect that matched well for SSS levels 1-5. We have revised the manuscript to include this issue (line 278-282).

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below:

The authors made a deep and accurate revision, with some new control analyses. They responded to the points that were raised by the reviewers, and the manuscript is now quite improved. Two concerns need further consideration:

1. The authors provide some support to the relative independence from circadian factors. Please discuss the current results within the context of previous studies of cortical excitability as a function of circadian factors (e.g. Chellappa et al. Circadian dynamics in measures of cortical excitation and inhibition balance. Sci Rep. 2016 Sep 21;6:33661. doi: 10.1038/srep33661).

2. Concerning the theta and α activity across sleep onset, please also consider a preliminary finding on intracranial recordings of a pharmaco-resistant patient with epilepsy [supplementary Figure 1 in Marzano et al., 2013 (Sleep Med. 2013 Nov;14(11):1112-22. doi: 10.1016/j.sleep.2013.05.021)]

In this revision, we have made the following changes of the manuscript:

1. We have expanded our discussion on the potential confounding effect of circadian rhythm (lines233-245).

2. Regarding the findings in Marzano et al., 2013, we added two related discussions:

a. On our results of the occipital lobe showing the most prominent power in EEG(lines 123-126);

b. On the potential regional differences among brain regions (lines 285-290).

3. We have made some miscellaneous changes regarding authorship, affiliations and acknowledgements.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

The manuscript has been improved but there are some remaining issues that need to be addressed. In particular, we note that there were apparently some significant changes in the author list, which is a bit unusual at this stage of the process, as the current revisions were relatively minor. Can you please provide a rationale for each of the authorship changes?

We apologize for the confusion caused by the change on the author list made during our last revision. Specifically, we moved Dr. Hua-Teng Cao from 10th to 4th on the author list.

Dr. Cao is a postdoc scholar in Dr. Zhang’s laboratory and has been joining the project since the beginning, assisting Dr. Zhang on the analysis of the dataset. During the first round of revision, Dr. Cao has devoted substantial effort in completing the additional analyses. All corresponding authors (Drs. Zhe Zhang, Ti-Fei Yuan and Rui-Ping Hu) agreed that he should be acknowledged for his contribution. In re-submitting our manuscript, we have already changed Dr. Cao’s authorship in the online system, but failed to update the author list on the manuscript PDF file. We have corrected this error during our second round of revision, leading to the issue described above. Nevertheless, if you check the author list on the online system, you will see that his position was changed during the first round of revision, which correctly reflected his contribution to the study.

https://doi.org/10.7554/eLife.65099.sa2

Article and author information

Author details

  1. Chin-Hsuan Chia

    Department of Rehabilitation Medicine, Huashan Hospital, Fudan University, Shanghai, China
    Contribution
    Conceptualization, Supervision, Investigation, Methodology
    Contributed equally with
    Xin-Wei Tang and Yue Cao
    Competing interests
    none
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3343-9541

  2. Xin-Wei Tang

    Department of Rehabilitation Medicine, Huashan Hospital, Fudan University, Shanghai, China
    Contribution
    Conceptualization, Investigation
    Contributed equally with
    Chin-Hsuan Chia and Yue Cao
    Competing interests
    none

  3. Yue Cao

    Department of Rehabilitation Medicine, Huashan Hospital, Fudan University, Shanghai, China
    Contribution
    Investigation, Project administration, Writing - original draft
    Contributed equally with
    Chin-Hsuan Chia and Xin-Wei Tang
    Competing interests
    none

  4. Hua-Teng Cao

    1. Institute of Neuroscience, State Key Laboratory of Neuroscience, Center for Excellence in Brain Science and Intelligence Technology, Chinese Academy of Sciences, Shanghai, China
    2. Shanghai Center for Brain Science and Brain-Inspired Intelligence Technology, Shanghai, China
    Contribution
    Software, Writing - review and editing, Writing - original draft
    Competing interests
    none

  5. Wei Zhang

    Institute of Brain Science, Fudan University, Shanghai, China
    Contribution
    Project administration, Data curation
    Competing interests
    none

  6. Jun-Fa Wu

    Department of Rehabilitation Medicine, Huashan Hospital, Fudan University, Shanghai, China
    Contribution
    Formal analysis
    Competing interests
    none

  7. Yu-Lian Zhu

    Department of Rehabilitation Medicine, Huashan Hospital, Fudan University, Shanghai, China
    Contribution
    Software
    Competing interests
    none

  8. Ying Chen

    Department of Rehabilitation Medicine, Huashan Hospital, Fudan University, Shanghai, China
    Contribution
    Supervision, Project administration
    Competing interests
    none

  9. Yi Lin

    Department of Rehabilitation Medicine, Huashan Hospital, Fudan University, Shanghai, China
    Contribution
    Investigation
    Competing interests
    none

  10. Yi Wu

    Department of Rehabilitation Medicine, Huashan Hospital, Fudan University, Shanghai, China
    Contribution
    Visualization
    Competing interests
    none

  11. Zhe Zhang

    1. Institute of Neuroscience, State Key Laboratory of Neuroscience, Center for Excellence in Brain Science and Intelligence Technology, Chinese Academy of Sciences, Shanghai, China
    2. Shanghai Center for Brain Science and Brain-Inspired Intelligence Technology, Shanghai, China
    Contribution
    Software, Writing - review and editing, Project administration, Methodology, Writing - original draft
    For correspondence
    zhezhang@ion.ac.cn
    Competing interests
    none
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0899-8077

  12. Ti-Fei Yuan

    1. Shanghai Key Laboratory of Psychotic Disorders, Shanghai Mental HealthCenter, Shanghai Jiaotong University School of Medicine, Shanghai, China
    2. Co-innovation Center of Neuroregeneration, Nantong University, Nantong, China
    3. Translational Research Institute of Brain and Brain-Like Intelligence, Shanghai Fourth People’s Hospital Affiliated to Tongji University School of Medicine, Shanghai, China
    Contribution
    Resources, Funding acquisition
    For correspondence
    ytf0707@126.com
    Competing interests
    None
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0510-715X

  13. Rui-Ping Hu

    Department of Rehabilitation Medicine, Huashan Hospital, Fudan University, Shanghai, China
    Contribution
    Funding acquisition
    For correspondence
    rphu79@163.com
    Competing interests
    none
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7626-2061

Funding

National Key Research and Development Program of China (2018YFC2001700)

  • Yi Wu

Medicine and Engineering Interdisciplinary Research Fund of Shanghai Jiao Tong University (ZH2018ZDA30)

  • Ti-Fei Yuan

Natural Science Foundation of China (81822017)

  • Ti-Fei Yuan

Natural Science Foundation of China (32071010)

  • Zhe Zhang

Natural Science Foundation of China (31771215)

  • Ti-Fei Yuan

Shanghai Science and Technology Development Foundation (18411962300)

  • Yu-Lian Zhu

Shanghai Municipal Commission of Health and Family Planning Foundation (201840225)

  • Rui-Ping Hu

Key Clinical Research Program of Southwest Hospital (s.shslczdzk02702)

  • Yi Wu

Shanghai Science and Technology Development Foundation (20412420200)

  • Yi Wu

Shanghai Pujiang Program (20PJ1415000)

  • Zhe Zhang

Shanghai Municipal Science and Technology Major Project (2018SHZDZX05)

  • Zhe Zhang

State Key Laboratory of Neuroscience

  • Zhe Zhang

Shanghai Municipal Education Commission - Gaofeng Clinical Medicine Grant Support (20181715)

  • Ti-Fei Yuan

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

This work was supported by the National Key R&D Program of China (2018YFC2001700), National Natural Science Foundation of China grants (81822017, 31771215, 32071010), Medicine and Engineering Interdisciplinary Research Fund of Shanghai Jiao Tong University (ZH2018ZDA30), the Key Projects of Shanghai Science and Technology on Biomedicine (18411962300), Shanghai Health and Family Planning Commission project (201840225), Shanghai Municipal Key Clinical Specialty (shslczdzk02702), the Projects of Shanghai Science and Technology (20412420200), Shanghai Municipal Education Commission - Gaofeng Clinical Medicine Grant Support (20181715), Shanghai Pujiang Program (20PJ1415000), Shanghai Municipal Science and Technology Major Project (2018SHZDZX05), and the State Key Laboratory of Neuroscience.

Ethics

Human subjects: 1. That informed consent, and consent to publish, was obtained 2. This study was designed as a prospective self-controlled study. The Ethics Committee of Huashan Hospital approved the study (2017-410) and was registered on the Chinese Clinical Trial Registry (ChiCTR1800016771).

Senior Editor

  1. John R Huguenard, Stanford University School of Medicine, United States

Reviewing Editor

  1. Laura Dugué, Uni­ver­sité de Paris, France

Reviewer

  1. Giulia Gaggioni, University of Liège, Belgium

Publication history

  1. Received: November 22, 2020
  2. Accepted: July 26, 2021
  3. Accepted Manuscript published: July 27, 2021 (version 1)
  4. Version of Record published: August 18, 2021 (version 2)

Copyright

© 2021, Chia et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,371
    Page views
  • 253
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Chin-Hsuan Chia
  2. Xin-Wei Tang
  3. Yue Cao
  4. Hua-Teng Cao
  5. Wei Zhang
  6. Jun-Fa Wu
  7. Yu-Lian Zhu
  8. Ying Chen
  9. Yi Lin
  10. Yi Wu
  11. Zhe Zhang
  12. Ti-Fei Yuan
  13. Rui-Ping Hu
(2021)
Cortical excitability signatures for the degree of sleepiness in human
eLife 10:e65099.
https://doi.org/10.7554/eLife.65099

Categories and tags

Research organism

Further reading

    1. Computational and Systems Biology
    2. Neuroscience

    Multilayer brain networks can identify the epileptogenic zone and seizure dynamics

    Hossein Shahabi, Dileep R Nair, Richard M Leahy
    Research Article Updated

    Seizure generation, propagation, and termination occur through spatiotemporal brain networks. In this paper, we demonstrate the significance of large-scale brain interactions in high-frequency (80–200Hz) for the identification of the epileptogenic zone (EZ) and seizure evolution. To incorporate the continuity of neural dynamics, here we have modeled brain connectivity constructed from stereoelectroencephalography (SEEG) data during seizures using multilayer networks. After introducing a new measure of brain connectivity for temporal networks, named multilayer eigenvector centrality (mlEVC), we applied a consensus hierarchical clustering on the developed model to identify the EZ as a cluster of nodes with distinctive brain connectivity in the ictal period. Our algorithm could successfully predict electrodes inside the resected volume as EZ for 88% of participants, who all were seizure-free for at least 12 months after surgery. Our findings illustrated significant and unique desynchronization between EZ and the rest of the brain in the early to mid-seizure. We showed that aging and the duration of epilepsy intensify this desynchronization, which can be the outcome of abnormal neuroplasticity. Additionally, we illustrated that seizures evolve with various network topologies, confirming the existence of different epileptogenic networks in each patient. Our findings suggest not only the importance of early intervention in epilepsy but possible factors that correlate with disease severity. Moreover, by analyzing the propagation patterns of different seizures, we demonstrate the necessity of collecting sufficient data for identifying epileptogenic networks.

    1. Neuroscience

    Model-based whole-brain perturbational landscape of neurodegenerative diseases

    Yonatan Sanz Perl, Sol Fittipaldi ... Enzo Tagliazucchi
    Research Article

    The treatment of neurodegenerative diseases is hindered by lack of interventions capable of steering multimodal whole-brain dynamics towards patterns indicative of preserved brain health. To address this problem, we combined deep learning with a model capable of reproducing whole-brain functional connectivity in patients diagnosed with Alzheimer’s disease (AD) and behavioral variant frontotemporal dementia (bvFTD). These models included disease-specific atrophy maps as priors to modulate local parameters, revealing increased stability of hippocampal and insular dynamics as signatures of brain atrophy in AD and bvFTD, respectively. Using variational autoencoders, we visualized different pathologies and their severity as the evolution of trajectories in a low-dimensional latent space. Finally, we perturbed the model to reveal key AD- and bvFTD-specific regions to induce transitions from pathological to healthy brain states. Overall, we obtained novel insights on disease progression and control by means of external stimulation, while identifying dynamical mechanisms that underlie functional alterations in neurodegeneration.

    1. Neuroscience

    Distinct roles of forward and backward alpha-band waves in spatial visual attention

    Andrea Alamia, Lucie Terral ... Rufin VanRullen
    Research Article Updated

    Previous research has associated alpha-band [8–12 Hz] oscillations with inhibitory functions: for instance, several studies showed that visual attention increases alpha-band power in the hemisphere ipsilateral to the attended location. However, other studies demonstrated that alpha oscillations positively correlate with visual perception, hinting at different processes underlying their dynamics. Here, using an approach based on traveling waves, we demonstrate that there are two functionally distinct alpha-band oscillations propagating in different directions. We analyzed EEG recordings from three datasets of human participants performing a covert visual attention task (one new dataset with N = 16, two previously published datasets with N = 16 and N = 31). Participants were instructed to detect a brief target by covertly attending to the screen’s left or right side. Our analysis reveals two distinct processes: allocating attention to one hemifield increases top-down alpha-band waves propagating from frontal to occipital regions ipsilateral to the attended location, both with and without visual stimulation. These top-down oscillatory waves correlate positively with alpha-band power in frontal and occipital regions. Yet, different alpha-band waves propagate from occipital to frontal regions and contralateral to the attended location. Crucially, these forward waves were present only during visual stimulation, suggesting a separate mechanism related to visual processing. Together, these results reveal two distinct processes reflected by different propagation directions, demonstrating the importance of considering oscillations as traveling waves when characterizing their functional role.