The Sleep Center of the First People’s Hospital of Yunnan Province conducted this cohort study from July 2023 to February 2024. This study included 48 PIs and 30 age-and sex-matched GSCs. PIs and GSCs were selected from men and non-pregnant women aged 18–60. PIs were recruited from outpatient clinics at the First People’s Hospital of Yunnan Province, while GSCs were recruited through word-of-mouth referrals. After signing informed consent forms, all participants underwent questionnaire assessments and PSG examinations (data collection of over 8 h) at the Sleep Center of the First People’s Hospital of Yunnan Province.

The inclusion criteria of PIs was: clinical diagnosis of PIs meeting ICSD-3 diagnostic criteria. The exclusion criteria for PIs were: (1) any other sleep disorders, such as periodic limb movements and abnormalities in circadian rhythms; (2) any secondary insomnia, such as insomnia coinciding with menopause in women aged over 45 years; (3) Apnea-hypopnea index (AHI) > 15 and periodic limb movement arousal index (PMLSI) > 15.

The typical exclusion criteria for all participants were: (1) any severe cardiorespiratory diseases, brain diseases, immune system diseases, pain, thyroid dysfunction, metabolic diseases, and other severe or unstable health conditions; (2) mental illness: neither group meets the criteria for anxiety or depression diagnosis. PIs have any significant history of mental illness in the past 6 months, and GSCs have a history of current or past insomnia or any major mental disorder; (3) any substances or drugs that affect sleep, including antipsychotic medications, smoking, coffee consumption of 4 cups per 24 h and alcohol consumption of 14 drinks per week.

The questionnaire assessments were conducted by a neurologist and a psychiatrist, both professionally trained. The inter-rater reliability was set at 90%. If the consistency falls below 90%, any discrepancies were resolved by a third rater.

The Pittsburgh Sleep Quality Index (PSQI) is one of the most widely used tools for assessing sleep quality globally. A score between 8 and 14 indicates mild insomnia, 14–21 indicates moderate insomnia, and ≥ 21 indicates severe insomnia. The Hamilton Anxiety Scale (HAMA) reflects the severity of anxiety symptoms. A score < 7 indicates no anxiety, and a score ≥ 14 indicates definite anxiety. The Hamilton Depression Scale (HAMD) is the most commonly used scale to assess depressive states. A score < 7 indicates normal, and a score ≥ 17 indicates depression. The Ford Insomnia Response to Stress Test (FIRST) evaluates an individual’s insomnia reactivity to the environment, with a total score ranging from 16 to 80. A score ≥ 18 can predict the onset of new insomnia.

The participants were instructed to arrive at the sleep center around 8: 00 PM each night for PSG examination. Before coming to the sleep center, participants were asked to avoid alcohol, drugs, excessive caffeine, and nicotine. Bedtime and lights-out time were determined based on the participants’ self-reported bedtime over the past 2 weeks. PSG recording lasted for no less than 8 h.

PSG was conducted using an ambulatory Somte PSG (Neurotrace Mukie PSM-A, China). The sampling rate was 512 Hz, and the filter settings were as follows: high pass filter 0.1 Hz; low pass filter 50 Hz; notch frequency 60 Hz. Surface electrodes recorded included six EEG (F4/M1, C4/M1, O2/M1, F3/M2, C3/M2 and O1/M2), two electrooculogram (EOG), submental electromyography (EMG), electrocardiogram (ECG) and two reference electrodes (M1and M2). Other electrodes included a thermistor, nasal cannula, snoring sensor, thoracic and abdominal belt, and oxygen saturation probe. All recordings were acquired according to the American Academy of Sleep Medicine scoring rules guideline (Rundo and Downey III, 2019). The same experienced, registered PSG technologist scored all sleep recordings blind to the subject diagnostic group using the revised AASM 2.2 sleep scoring criteria.

Sleep staging was performed using sleep analysis software, with EEG data segmented into 30 s before and after sleep stage transitions. The analysis excluded artifacts containing body movements, respiration, leg movements, and other awakenings. The artifact-free EEG data during all NREM stages (i.e., N1, N2 and N3) were exported in European Data Format (EDF) and imported into MATLAB software (Version 2022b, Mathworks, United States) for spectral analysis.

Using the EEGLAB toolbox and ERPLAB in MATLAB (Math works, R2023b), raw data were imported at a sampling rate of 512 Hz and down-sampled to 256 Hz. The linked ear electrodes were used as a reference and segmented the 30-s NREM sleep epochs into 4-s (4-s) mini-epochs (Buysse et al., 2008; Kang et al., 2018). Semi-automatic artifact rejection procedures were used to remove channels and epochs with high-amplitude noise (Liu et al., 2021). Using the absolute threshold method, trials were selected to exclude those exceeding ±80 μV for each channel. Power spectrum in these ranges across all 4-s NREM epochs for each channel were plotted and visually inspected. When epochs showed significantly higher or lower amplitude power without exceeding the automatic threshold, we lowered the threshold and removed the exceeding epochs. Channels with artifacts affecting a majority of the recording were removed. We utilized additional spectral-based and topographic procedures to eliminate individual channels exhibiting markedly higher power than adjacent channels.

Selected 4-s NREM epochs in each EEG electrode were Hanning-tapered and Fourier-transformed by the FFT algorithm (with 50% overlapping windows) to obtain the power spectrum density at each frequency. We assessed the following seven frequency bands: delta (1–4 Hz), theta (4–8 Hz), alpha (8–12 Hz), sigma (12–16 Hz), beta1 (16–24 Hz), beta2 (24–32 Hz), and gamma (32–50 Hz), with each band inclusive of the lower boundary but not the upper boundary.

SPSS (version 17.0, SPSS, Inc., Chicago, IL, United States) was used for statistical analyses. The data were presented as mean ± standard deviation if they are normally distributed (or approximately normally distributed). If they are not, the data were presented as the median (with interquartile range in brackets). Histograms, q-q plots, and Shapiro–Wilk’s tests assessed the data normality. Normally distributed variables were compared using a one-way analysis of variance or T-test, and the non-normally distributed variables using the Mann–Whitney U test. The Levene test was used to test variance homogeneity. The one-way testing was performed on data with homogeneity of variances and multiple post-hoc comparisons between the six areas using the Scheffe method. Welch’s test for correction was applied for data with heterogeneity of variances, and Tamhane T2 was utilized for post-hoc multiple comparisons. Categorical variables were presented as the frequency distribution and compared using the chi-squared or Fisher’s exact test.

The adjusted p-values were calculated for multiple comparisons using the Least Significance Difference method. A false discovery rate of p < 0.05 was considered statistically significant.

The demographic and clinical characteristics were shown in Table 1. PIs and GSCs differed on the HAMA, HAMD, FIRST and PSQI. The HAMD and HAMA scores of PIs and GSCs fell within the normal range, but the HAMD and HAMA scores of PIs were higher than those of GSCs (p < 0.01). Notably, the median PSQI score for PIs was 16, falling within the range of moderate insomnia. The difference between the two groups was highly significant (p < 0.01). There were no significant differences in age, sex, height, weight, and race between groups.

As shown in Table 2, PIs and GSCs differed in most subjective sleep measures. PIs had more than 1 h of subjective sleep latency (SOL) and total sleep duration (TST) of less than 5 h. The differences between subjective and objective measures showed that PIs significantly overestimated sleep latency and underestimated total sleep duration and efficiency compared to GSCs.

Objective PSG measures of sleep efficiency (SEI), TST, and SOL differed between the two groups (p < 0.01). PIs had poor sleep continuity, as evidenced by many awakenings after sleep onset (WASO; 84.46 ± 54.83; 39.40 ± 30.34, respectively, p < 0.001) and a long total awakening duration (TWT; 123.51 ± 83.58; 51.47 ± 37.10, respectively, p < 0.001). PIs had a lighter sleep, with a higher percentage of N1 (10.25 ± 6.75; 7.11 ± 3.42, respectively, p = 0.008) and a significant reduction in the more profound sleep period, i.e., the N3 period of sleep (6.61 ± 5.51; 14.96 ± 5.18, respectively, p < 0.001), and a decrease in the ratio of REM (12.11 ± 5.92%; 16.99 ± 6.36, respectively, p < 0.001) in PIs compared to GSCs.

As shown in Figure 1 and Table 3. In GSCs, apart from delta power and beta2 power showing differences across various areas, no significant differences were observed in other frequency band’s power between different areas. The delta power of GSCs was highest in F3 and C3.

In PIs, the delta power was also highest in F3. The beta2 power was the highest in C4 and F4, beta1 power in C4 and O2, gamma power in C3, and alpha and sigma powers in O1 and O2. The theta, alpha, sigma, beta1, and gamma power did not show significant differences in different areas in PIs.

As shown in Figures 2, 3 and Table 4 and Supplementary Image 1. The difference in delta power between the two groups was statistically significant in all six areas. The difference in theta power between the two groups was statistically significant in C4, F3, F4, O1 and O2. The alpha power of PIs was lower than GSCs in C3 and higher in O1. The sigma power of PIs was significantly higher than GSCs in F4 and C4. The beta1 power of PIs was substantially higher in C4. The beta2 power of PIs was considerably higher in C4, F4, C3, and F3; meanwhile, C4 and F4 had the highest power in all six areas. The gamma power of PIs was significantly higher than that of GSCs in C3, F3 and C4, and the gamma power of PIs was highest in C3.

As shown in Figure 4 and Supplementary Table 1. In PIs, alpha power was positively correlated with education years across six areas. Beta1 power was positively correlated with age in F3 (r = 0.3283, p = 0.02) and the number of awakenings in C4 (r = 0.3291, p = 0.02) and O2 (r = 0.3018, p = 0.04). Beta2 power was positively correlated with age in C3 (r = 0.3297, p = 0.02) and negatively with WASO in F3 (r = −0.3344, p = 0.02).

In GSCs, reduced delta power in the right hemisphere and increased beta2 power in the right hemisphere suggested heightened alertness (Riemann et al., 2010; Cabrera et al., 2024) during the first night of wearing PSG. However, these changes in alertness did not reach the threshold for awakening, leading to fewer awakening events. In contrast, PIs showed a more pronounced response to environmental adaptation (higher FIRST scores than GSCs; Van Someren, 2021), characterized by significant increases in alertness, particularly in beta 2 power, which was elevated across both hemispheres but more so in the right hemisphere. The right hemisphere exhibited higher sensitivity to alertness during the awakening process (Swarnkar et al., 2007). This was accompanied by an increase in awakening events, with WASO, TWT and WT significantly increased in PIs.

However, no strong correlation between beta2 power and increased alertness was observed in PIs. This might be due to the exclusion of artifacts (e.g., movement artifacts associated with awakenings) during data analysis, which could have impacted the detection of the relationship between beta2 power and awakenings. Future studies incorporating additional alertness indicators, such as cortisol levels and heart rate variability (HRV), along with more artifact-free stereotactic-electroencephalography (SEEG) data, might provide further insights into this phenomenon.

Most previous power spectral analyses on insomnia did not subdivide the beta band into beta1 and beta2. It could not provide a better understanding of the biological significance of EEG (Van Albada and Robinson, 2013).

The sigma power increased on the right side in PIs, which aligned with a hypothesis that sigma power elevation was linked to the activation of emotional memory traces, predominantly in the right hemisphere. According to the hyperarousal theory of insomnia, insomnias require substantial encoding and transformation of emotions during NREM sleep. Amygdala adaptation necessitates stable NA pauses during REM sleep and the reactivation of memory traces during the pre-REM period, indicated by an increase in sleep spindles (sigma) (Sterpenich et al., 2007; Van Der Helm et al., 2011). Hence, alterations in sigma power in insomnias are expected (Spiegelhalder et al., 2012). At the same time, the emotional rhythms prefer to be processed in the right hemisphere (Ahern and Schwartz, 1985).

The clinical treatment strategy of repetitive transcranial magnetic stimulation (rTMS) and transcranial direct current stimulation (tDCS) for insomnia followed the protocol of high-frequency stimulation of the left hemisphere and low-frequency inhibition of the right hemisphere (Perera et al., 2016).

The left side would be a dominant brain area for declarative memory consolidation. The increased gamma power during NREM sleep in PIs could be an active declarative memory consolidation process.

Usually, the gamma oscillation during wakefulness was associated with memory encoding and retrieval. Using deep electrodes in conjunction with scalp electrodes had found spontaneous gamma oscillations, particularly in the depolarized state of slow-wave oscillations (up states) during NREM sleep (Mölle et al., 2009; Valderrama et al., 2012). The local field potentials (LFPs) detected the gamma oscillation in sleeping or anesthetized animals (Mukovski et al., 2007; Mena-Segovia et al., 2008). During NREM sleep, gamma oscillations are associated with memory consolidation, transforming initially unstable memories from wakefulness into more stable representations (Diekelmann and Born, 2010).

Increased gamma power during NREM sleep was detected using the bilateral average montage (Perlis et al., 2001a; Perlis et al., 2001b; Buysse et al., 2008). However, they did not clarify the biological significance. The increase in gamma power in different brain areas would reveal important physiological significance.

The positive correlation between beta1 power and the number of awakenings suggested that individuals exhibiting higher beta1 activity during sleep were more prone to sleep fragmentation. This relationship implied that elevated beta1 power might serve as a biomarker for sleep maintenance difficulties, where increased cortical arousal interrupted sleep, leading to frequent awakenings. Such disruptions could degrade sleep quality and contribute to the subjective experience of non-restorative sleep commonly reported in insomnia (Shi et al., 2022; Hermans et al., 2021). Additionally, awakenings were often characterized by asynchronous information processing between the left and right hemispheres of the brain (Swarnkar et al., 2006; Swarnkar et al., 2007), with the right hemisphere typically showing greater sensitivity to arousal (Liotti and Tucker, 1992).

A limitation of this study was that only six brain areas were compared, which might not provide a comprehensive understanding of the biological mechanisms of insomnia. Future studies using standard PSG combined with high-density EEG (hdEEG) for synchronous recordings in PIs could address this limitation. Comparing the power spectra across different stages of sleep (REM, NREM and wakefulness) would offer more detailed insights.