Abstract
Sleep deprivation (SD) is known to impair cognitive function, but its impact on neural dynamic sremains unclear. This study investigated changes in alpha and theta EEG spectral power, specifically their mean, variance, and various correlations, under SD and normal sleep conditions. Using a publicly available dataset, we analyzed 5-minute eyes-open and eyes-closed resting-state EEG recordings, focusing on occipital alpha and frontal theta bands with a time-frequency analysis. While mean spectral power across time in both alpha and theta bands did not differ significantly between NS and SD conditions, the variance in alpha and theta power across time were significantly different in the SD condition, suggesting a prominence in neural instability. Finally, alpha power variance under NS correlated significantly with subjective sleepiness measures and vigilance test, but not in SD, possibly implying a disruption in the linkage between alpha power and physiology related to fatigue and sleep deprivation. These findings suggest that spectral variance, rather than mean power, more definitively reflects the neural outcomes of sleep deprivation. Thus, elevated alpha and theta variance may serve as a neural marker of attentional and cognitive instability, and its dissociation between behavioral measures highlights the disconnection in the neural-behavioral relationship during fatigue.
Introduction
Sleep deprivation affects everyone. Sleep deprivation is a state in which an individual is significantly lacking in quality sleep (Kilgore, 2010). In an average case, humans require 7-8.5 hours of sleep a day (Alhola et al., 2007), for the body to fully move through the sleep stages and perform the vital processes that prepare the body for the next day. However, sleep duration overall has declined by 20% in the past century (St-Onge et al., 2016). Today, 20% of the adult population is reported to be sleep deprived (Abrams, 2015). Of the five different school levels elementary school, middle school, vocational high school, senior high school, and university students – high school students are shown to have the shortest sleep duration and the poorest sleep quality, showing that sleep deprivation is an alarming problem for not only the adult population but also substantially for students and minors (Wheaton et al., 2018).Sleep deprivation may lead to serious, long-term consequences such as chronic depression and anxiety from exhaustion of cognitive control over emotions and thought processes (Kahn et al., 2013), and disrupted circadian rhythm resulting from repeated cycles of downgraded vigilance and anomalies. Not only that, as sleep deprivation accumulates, there is a higher risk of an increased number of health diseases such as diabetes, obesity, heart attack, and stroke (Colten and Altevogt, 2006). Sleep deprivation also causes detrimental short-term effects. Psychological and neural alterations due to deprived sleep result in increased drowsiness, lack of attention, and decreased vigilance. It is seen that sleep deprivation mainly affects the frontal lobe, thalamus, and prefrontal cortex, which are involved in vigilance and drowsiness (Thomas et al., 2006). Likewise, a few regular examples we can spot easily due to a lack of sleep include heavy drowsiness and lack of vigilance, leading to a loss of cognitive control. One of the most critical and common brain functions that are impacted by eroded cognitive control is deteriorated attention span (Gibbings et al., 2021; Kilgore, 2010). Given that even subjects affected by short amounts of sleep deprivation struggle to hold the ability to sustain attention and that sustained attention is vital for almost all tasks, it is important to analyze the direct and objective consequences that sleep deprivation has on alertness and attention. Brainwaves offer a powerful way to study alertness and attention, as different frequencies reflect distinct cognitive states. Measured using electroencephalography (EEG), brainwaves are categorized by frequency. Most well-known frequency bands related to sleep/awake phases are: delta (0.5–4 Hz), theta (4–8 Hz), alpha (8–13 Hz), beta (13–30 Hz), and gamma (30+ Hz) (Groppe et al., 2013). Delta waves are predominant during deep sleep, when conscious activity is minimal (Hestermann et al., 2024). Theta waves appear in light sleep or drowsiness, such as during daydreaming or meditation, showing prominence in the frontal region of the brain (Borrani et al., 2022). Alpha waves, prominent in the occipital area, are associated with relaxed wakefulness (Britton et al., 2016). As cognitive demand increases, beta waves emerge, reflecting active processing and attention (Ray & Cole, 1985). Gamma waves are associated with intense brain activity, particularly in moments of heavy agitation, heightened information processing, or deep concentration, where the brain is operating at its maximum capacity (Yang et al., 2020). Previous studies of sleep deprivation have primarily focused on two frequency bands: posterior alpha waves and frontal midline theta waves (e.g., Strijkstra et al., 2003; Snipes et al., 2022). Posterior alpha wave rhythms are prominent over occipital and parietal regions and are traditionally associated with relaxed wakefulness, as mentioned above. Frontal midline theta activity is commonly linked to working memory and executive control. Studies have shown that Frontal midline theta power increases during tasks requiring sustained attention, memory encoding, retrieval, and mental calculations (Hsieh et al., 2014). As the brain engages in more complicated tasks that require heightened attention, a decrease in posterior alpha power followed by an increased theta power over the frontal region is commonly observed, taking over alpha waves that were observed the restful wake state (Tan et al., 2024; Womelsdorf et al., 2010; McFerren et al., 2021). However, frontal midline theta oscillations are also known to be more active when sleep deprived (Snipes et al., 2022). From what is known about the posterior alpha and the frontal midline theta, one may speculate that people who are sleep deprived and thus drowsy may exhibit greater posterior alpha and frontal midline theta power in general. However, past studies show conflicting results regarding the mean alpha and theta powers under sleep deprivation. For instance, Strijkstra et al. (2003) conducted a controlled sleep deprivation study using EEG measures during a sustained attention task, and they reported a decrease in posterior alpha power and an increase in frontal-midline theta power as time awake increased. Their interpretation of these results was that a rise in theta power reflects increased cognitive effort and sleep pressure, while the decrease in alpha may signify a decrease in cortical activation. Conversely, Gibbings et al.(2021) reported an increase in alpha power under sleep-deprived conditions, directly contrasting Strijkstra et al. (2003). Additionally, Gibbings and colleagues found an increase in theta power under sleep-deprived conditions, aligning with Strijkstra et al. (2003). Unlike Strijkstra et al. (2003) who used a sustained attention task, Gibbings et al. (2021) used a resting-state EEG design and interpreted the alpha increase as a potential marker of cortical disengagement or passive idling, possibly representing a different manifestation of sleep pressure. Similarly, Snipe and colleagues (2022) showed greater theta power under sleep deprivation, although the activity pattern depended on the tasks. Aside from mean oscillation power differences, past studies of sleep deprivation also report a ‘fluctuation’ between wakefulness and sleep-like states in the brain and variations in alpha and theta powers (Doran et al., 2001; Gibbings et al., 2021). In sleep deprivation, the brain alternates more rapidly between periods of relative engagement and disengagement (Doran et al., 2001). This instability was interpreted as a marker of a compromised or stressed neural system. Doran and colleagues (2001) observed that “[during sleep deprivation] the brain goes back and forth – fluctuates – between wakefulness and sleep-like states between seconds,” noting that sleep-deprived individuals enter an unstable wake state where neural processing fluctuates between alertness and microsleep-like states. Given the conflicting results regarding alpha levels, an exploratory approach focusing on the variability and correlation of these signals may provide new insights into the neural dynamics underlying sleep deprivation. While some studies have reported increases or decreases in mean spectral power following sleep deprivation, others have found no significant changes, highlighting inconsistencies in the results of these effects. These inconsistencies suggest that mean power alone may not fully capture the impact of sleep deprivation on brain function. Instead, examining moment-to-moment fluctuations (i.e., variance) and the relationship between alpha and theta wave powers (i.e., correlation) allows for a more realistic and nuanced understanding of neural instability during SD. In the current study, we aim to investigate how sleep deprivation impacts neural stability by quantifying variance in alpha and theta power and examining the correlations between these two frequency bands.
Methods
Dataset Information
This study utilized the publicly available dataset from Xiang et al. (2024), which consists of resting-state EEG recordings and cognitive assessments from 71 participants. This study used a within-subject design where each participant underwent two sessions: one after the SD (sleep deprived) condition, where individuals were sleep deprived for 24 hours, and the other after the NS (normal sleep) condition, where individuals had a normal night’s sleep. All 71 subjects underwent the 5-minute eyes-open EEG testing, but only about half of the subjects (N=38) underwent the 5-minute eyes-closed EEG testing. Subjects also underwent the Psychomotor Vigilance Test (PVT). They also completed state-scale survey assessments, including PANAS (Positive and Negative Affect Schedule), ATQ (Automatic Thoughts Questionnaire), SAI (State Anxiety Inventory), SSS (Stanford Sleepiness Scale), KSS (Karolinska Sleepiness Scale), Sleep Diary, and completed trait-scale survey assessments, including EQ (Empathy Quotient), BPAQ (Buss-Perry Aggression Questionnaire), PSQI (Pittsburgh Sleep Quality Index) tests. Of these tests and surveys, the following measures were utilized in the current study: The Psychomotor Vigilance Test is a cognitive test that measures how well someone can sustain attention and react to stimuli, specifically used in this study to correlate different brain power measures with post and pre-sleep deprivation attention measures and reaction times. The Stanford Sleepiness Scale (SSS) and Karolinska Sleepiness Scale (KSS) are subjective, self-analytical scales that are used to assess an individual’s perceived level of sleepiness at a given moment. Although the SSS and KSS are different in their scale, with SSS a range of 1-7 and KSS with a range of 1-9, the two are otherwise nearly identical in their function. Therefore, we merged the two scales by rescaling them to a common scale of 1–63, which is the least common multiple of 9 and 7. This combined scale of sleepiness was used to correlate the neural measures. The Pittsburgh Sleep Quality Index (PSQI) is a questionnaire that measures the subject’s sleep quality and occurrences of sleep disturbances, and was also used to correlate the neural measures.
Participants
This study involved 71 subjects: the average age of subjects was 20 years, ranging from 17–23 years, and there were 34 females and 37 males participating in the study. None of the participants showed any history or symptoms of neurological, physical, or psychiatric disorders, and maintained a typical sleep schedule before the study.
Procedure
Participants went through two different sessions—a normal sleep (NS) session and a sleep-deprived (SD) session—the order of which was counterbalanced. In the NS session, participants arrived at the lab dispersed throughout the day after a normal routine of sleep which was ensured by a sleep quality monitor. Then, they underwent different behavior testing measures of PVT and state scales, followed by a 5-minute interval of eyes-open EEG recording, and then a 5-minute interval of eyes-closed EEG recording. Each experimental phase employed 61 active Ag/AgCl electrodes embedded in an elastic cap, positioned according to the extended 10–20 international electrode placement system (Brain Products GmbH, Steingrabenstrasse, Germany). The FCz electrode was used as the online reference. Data were recorded at a sampling rate of 500 Hz, with electrode impedance carefully maintained below 5 kΩ. Lastly, the participants filled out a variety of trait scales. In the SD session, participants arrived at the lab at 9 pm the day before the data collection day and were monitored to be continuously awake through the entirety of the study; they were prohibited from active physical movement and beverages or food containing caffeine or alcohol. With no recovery sleep after the sleep deprivation period, the participants immediately took part in the same procedure as the NS condition. The EEG data was recorded at a different time stamp for each subject, but around the same time they underwent the EEG recording for the NS condition.
Preprocessing and Validation Analysis
In order to validate the data downloaded from the public repository, the spectral power for the alpha band at the electrode Cz was computed using the Welch method using EEGLAB (v2024.2) in MATLAB (R2024b), as in Xiang et al. (2024). The raw continuous five-minute EEG data was preprocessed by band-pass filtering between 0.5 and 45 Hz and was segmented into 75 epochs of 4 seconds. Then, the alpha spectrum at Cz was computed using the Welch method function in MATLAB (pwelch). The results resembled what was shown in Figure 4 of Xiang et al. (2024), which encouraged us to move forward with our analysis.
Time-Resolved Spectral Power
Separately for each subject under NS and SD conditions, we performed a time-frequencyanalysis using the newtimef function with default parameters in EEGLAB to quantify the spectral power of the posterior alpha and frontal theta frequency bands across the 5-minute periods. For each subject’s NS or SD trial, the function computed spectral power from 4 to 12 Hz over 300 time points. To focus on specific frequency bands that were shown to affect sleep, we extracted theta power (4-8 Hz) from frontal channels (‘Fz’, ‘FC1’, ‘FC2’, ‘F1’, ‘F2’, ‘Fp1’, ‘Fp2’, ‘AF3’, ‘AF4’ ) and alpha power (8-12 Hz) from occipital channels (‘O1’, ‘PO7’, ‘P7’, ‘P5’, ‘PO3’, ‘O2’, ‘PO4’, ‘PO8’, ‘P6’, ‘P8’). The selection of these channels was based on previous studies (Snipes et al., 2022 ). See Figure 1 for the spatial location of these channels. To quantify alpha and theta power measures, we took the average of spectral power across the respective electrodes. Then, for each subject and for each condition (NS and SD) and state (eyes-open and eyes-closed), the mean and the variance of the resulting power values across the time pointswere computed.
Behavioral Measures
Multiple sleepiness surveys and behavioral assessments were conducted; however, there were some substantial missing data. For example, SSS was collected from only subjects 1 through 39, 71, and 72 while KSS was collected from subjects 40 through 72, most of whom did not overlap with the participants who completed the SSS survey. To account for these missing data and to increase the reliability of these sleepiness measures, we constructed a composite measure of sleepiness by the following approach. First, we combined the sleep measures of SSS and KSS by adjusting the scale to a coalesce measure – each was rescaled to a common range of 1–63 (the least common multiple of the original ranges, 1–7 for SSS and 1–9 for KSS).
Statistical Analyses
Spectral Power Quantification
For each subject in both conditions (NS and SD) and state (eyes-closed and eyes-open), the mean and variance of alpha and theta powers were computed across the time points. Paired t-tests were then performed to compare these measures between NS and SD conditions separately for eyes-open and eyes-closed data.
Correlation Between Alpha and Theta Power
Pearson’s correlation coefficients were computed using the corr function in MATLAB on a per-subject basis to examine the relationship between frontal theta and posterior alpha power in each condition (NS and SD) separately for eyes-open and eyes-closed recordings. Then, paired t-tests were performed to verify the results and compare the strength of these correlations across conditions.
Neural-Behavioral Correlations
Pearson’s correlation coefficients were computed using the corr function in MATLAB on a per-subject basis to examine the relationship between sleepiness scales and other behavioral measures (composite sleepiness measure – SS and PSQI) and the alpha variance (NS and SD) separately for eyes-open and eyes-closed recordings. Note that this relationship became the target of our investigation after identifying a statistically significant difference in the alpha variance between the NS and SD conditions. Paired t-tests were performed to verify the results and compare the strength of these correlations across conditions.
Results
Comparison of Mean and Variance of Alpha and Theta Power To examine whether the effect of sleep deprivation on alpha and theta powers in the current data is consistent with previous studies, we first sought to quantify summary measures of alpha and theta power for each subject under NS and SD conditions, separately for eyes-open and eyes-closed recordings. Each participant’s (alpha and theta) mean spectral power over the five minute period for each of the four conditions (i.e., eyes open NS, eyes close NS, eyes open SD, eyes closed SD) was collapsed into a single value by taking the mean across time. Likewise, each participant’s variability of the spectral power was computed by taking the variance across time.
Correlation Between Time-Resolved Alpha and Theta Power
While the mean spectral powers were hardly different between NS and SD (see Fig. 2), more differences were observed in the variance of alpha and theta power between NS and SD conditions. Studying the correlation between the mean alpha and theta power could reveal whether these fluctuations observed are part of a broader, interrelated response pattern. This could possibly indicate a significant degradation in the relationship and communication of alpha and theta during sleep deprivation states, not a simple decrease or increase in powers.
Neural-Behavioral Correlations
Finally, to question the behavioral relevance of these neural changes, we correlated the variance of alpha power separately with the composite behavioral sleepiness measure (SS), PSQI, and PVT. Given that the variance of alpha power was the only measure that indicated a difference between NS and SD conditions (see Figure 2), only the variance of alpha power was tested in this neural-behavioral correlations analysis. Table 1 summarizes these results. The positive correlations between the variance of alpha and SS (sleepiness scale) in NS showed to be significant in both the eyes-open and eyes-closed states. Additionally, the positive correlation between the variance of alpha and PVT in NS was shown to be significant only in the eyes-closed state.
Conclusion
In conclusion, this study finds that the average spectral power alone may not adequately capture the neural effects of sleep deprivation. Instead, the variability in brain oscillations – particularly in the alpha and theta power – appears to be more indicative of sleep-deprived states. This heightened variability also shows select associations with behavioral alertness under normal sleep, suggesting it plays a functional role in maintaining cognitive stability. Moreover, the observed changes in alpha and theta variance may be related to the disrupted