Effects of lemborexant and zolpidem on sleep electroencephalography in older adults with insomnia: a randomized trial

Abstract

How sleep medications affect brain physiology over time remains largely unknown. Prior pharmacodynamic work has established distinct spectral signatures of GABA-A receptor modulators vs. DORAs in younger or mixed adult samples. Here, we report how sleep EEG effects evolve from acute (days 1–2) to chronic (days 29–30) exposure in older adults with insomnia meeting DSM-5 criteria, and how within-person changes relate to Insomnia Severity Index (ISI). This analysis used data from a randomized, double-blind, placebo-controlled, active-comparator phase III trial (May 2016–January 2018). A subset of 249 participants was analyzed for EEG outcomes. Participants received 5 mg or 10 mg LEM (LEM5, LEM10), 6.25 mg extended-release zolpidem (ZOL), or placebo nightly for 30 days. Polysomnography was collected at baseline, after acute exposure, and after chronic exposure. Acute ZOL exposure produced significant increases in non-rapid eye movement (NREM) slow oscillation (SO), sigma, beta, and gamma power, with decreases in delta and theta compared to all treatment groups. These effects persisted with chronic exposure, alongside an alpha increase from acute to chronic exposure. LEM10 showed minimal acute effects but, with continued use, increased SO and decreased alpha power. The degree of SO power increase from acute to chronic exposure predicted improvements in ISI scores, driven by the LEM10 group. ZOL showed similar REM effects, whereas LEM5, but not LEM10, reduced REM sigma and beta power. These findings demonstrate that ZOL and LEM differentially modulate sleep EEG, with sustained SO enhancement under LEM10 linked to symptom improvement.

Statement of Significance.

Though insomnia is prevalent in older adults, the neurophysiological impact of current pharmacological treatments is poorly understood. This study compared the effects of two sleep medications on sleep microarchitecture following both acute and chronic exposure in older adults with insomnia. Similar to previous reports, zolpidem broadly altered sleep microarchitecture across sleep stages, enhancing high frequency activity, with this effect increasing in magnitude during non-rapid eye movement (NREM) sleep as exposure continued. Lemborexant showed NREM increases in slow oscillation (SO) power and decreases in alpha power following chronic exposure. Decreases in insomnia severity index scores across treatment were associated with increased SO power. Pharmacological treatment for insomnia across 30 days is thus associated with dynamic changes in sleep microarchitecture in a medication-specific manner.

Introduction

Insomnia symptoms, including difficulties falling asleep, staying asleep, and/or waking earlier than desired, as well as experiencing unrestful, nonrestorative sleep despite having adequate sleep opportunity, affect up to 30%–48% of older adults, particularly those with age-related neurodegenerative conditions [1]. Chronic insomnia contributes to daytime dysfunction and increased risk of comorbidities, including cognitive impairment and Alzheimer’s disease [2, 3]. While cognitive behavioral therapy for insomnia (CBT-I) is the recommended first-line treatment, many older adults rely on pharmacological interventions to manage their symptoms [4]. However, these medications differ in their mechanisms of action and may have distinct effects on sleep neurophysiology, which could influence long-term cognitive and health outcomes.

Benzodiazepine receptor agonists (BZRAs), such as zolpidem (ZOL), are commonly prescribed hypnotics that enhance GABAergic inhibition to promote sleep. Although effective in increasing sleep duration, they have been associated with next-day cognitive impairment, increased risk of falls, and potential links to dementia in epidemiological studies [3, 5, 6]. In contrast, dual orexin receptor antagonists (DORAs), such as lemborexant (LEM), suppress wake-promoting orexin signaling and may provide similar benefits without the same cognitive risks [7]. However, the differential impact of these medications on sleep architecture and electroencephalography (EEG) oscillations remains poorly understood, particularly in older adults with insomnia.

Given the emerging evidence that sleep oscillations play a crucial role in learning, memory consolidation, synaptic plasticity, and toxin clearance, alterations in sleep EEG induced by these pharmacological treatments may have significant implications for the long-term cognitive and physical health of older adults with insomnia [8–12]. Therefore, it is critical to understand how insomnia treatments differentially impact sleep neurophysiology, as reflected by changes in sleep EEG, since these oscillatory patterns are fundamental to cognitive processes such as memory consolidation and brain waste clearance. While daytime side effects of some hypnotics may be primarily due to residual sedation [13], chronic alterations in sleep architecture and neurophysiology, such as sustained increases in beta/gamma activity or reductions in slow wave expression, may have longer-term consequences for sleep quality, cognitive function, and brain health, although these effects remain to be directly tested.

Prior work distinguishes the neurophysiological signatures of pharmacological sedation from those of natural sleep, with sedative agents inducing patterns that differ from physiologic non-rapid eye movement (NREM) oscillatory activity [14]. In contrast, DORAs have been shown in clinical trials to improve insomnia symptoms while preserving overall sleep architecture more effectively than GABAergic hypnotics [15]. These mechanistic differences have led to the hypothesis that DORAs may exert sleep-promoting effects with fewer disruptions to sleep microarchitecture. However, existing EEG studies have largely examined acute drug exposure in healthy young or middle-aged adults, limiting their relevance to older adults and to the chronic hypnotic use typical of many insomnia patients. Moreover, the clinical relevance of pharmacologically induced changes in sleep oscillatory activity remains poorly understood, as few studies have directly examined whether such neurophysiological alterations correspond to meaningful improvements in insomnia symptoms. However, these studies have largely been conducted following acute exposure only in healthy young or middle-aged adults, limiting their relevance to the older population most at risk for cognitive decline and chronic hypnotic use. However, the clinical relevance of these neurophysiological changes remains poorly understood. Whether pharmacologically induced alterations in sleep oscillatory activity corresponds to subjective improvements in insomnia symptoms has not been systematically examined. Establishing this link is important for identifying EEG-based markers that may reflect meaningful therapeutic effects.

To address these gaps, we investigated the acute (1–2 days) and chronic (29–30 days) effects of placebo (PBO), extended-release ZOL (6.25 mg), and LEM (5 mg and 10 mg) on sleep EEG spectral power in older adults meeting criteria for insomnia disorder and experiencing moderate to severe insomnia symptoms. We hypothesized that zolpidem would significantly alter EEG spectral power in NREM and REM sleep compared to baseline, whereas LEM would have a minimal impact on sleep oscillations. Additionally, we examined whether EEG changes varied between acute and chronic exposure, and whether they correlated with insomnia symptom severity.

Materials and Metheds

Clinical trial protocol oversight and study participants

Participant recruitment and screening procedures of the parent phase III clinical trial sponsored by Eisai Co., Ltd., (registered as E2006-G000-304; NCT02783729) are described in detail by Rosenberg and colleagues (2019) [16]. To summarize, the parent study, SUNRISE 1, was a randomized, double-blind, placebo-controlled, active-comparator study conducted at 67 sites across North America and Europe from May 2016 to January 2018. The study enrolled 1006 women over 55 years old and men over 65 years old. The trial protocol was approved by the relevant institutional review boards or ethics committees at each site. Eligible participants were required to meet Diagnostic and Statistical Manual of Mental Disorders, fifth edition (DSM-5) [17] criteria for insomnia disorder, have documented sleep maintenance difficulties, and a score ≥ 13 on the Insomnia Severity Index (ISI) [18]. Further, eligible participants needed to report wake after sleep onset time (WASO) of ≥60 minutes at least three nights per week for a duration of three months or more, maintain regular bedtimes, and spend 7–9 hours in bed per day. Although zolpidem is often characterized as a short-acting hypnotic with stronger effects on sleep onset, prior work has shown that in older adults its effective half-life is longer and it can reduce WASO and support sleep maintenance. The ≥60-minute WASO criterion was therefore used to ensure clinically meaningful insomnia symptoms across participants while allowing evaluation of treatment effects on both onset and maintenance symptoms [19]. Participants underwent a two-week PBO run-in period to confirm eligibility and provide baseline data. Polysomnography (PSG) recordings obtained during the run-in period were used to confirm eligibility and establish baseline sleep parameters prior to randomization. During the run-in, participants completed two baseline PSG study nights. The ISI was administered at Screening, during the Run-in phase (Study Day -7, seven days before receiving the first dose of the study drug), at Baseline (the night before the first dose), and at the end of the Treatment Period (days 29 and 30). After the baseline phase, participants were randomized to receive LEM5, LEM10, ZOL, or PBO in a double-blind, double-dummy design. Drug or PBO treatment lasted 30 nights, with PSG-recorded nights conducted after acute (days 1–2) and chronic (days 29–30) exposure (see Figure 1S in the Supplement). The present study represents a secondary analysis of a subset of data collected during the SUNRISE 1 trial (Figure 1).

Figure 1.

Study flowchart for the post hoc analysis.

Data from a randomized subsample of male and female participants over 65 years old were utilized in the current report (Table 1 and Figure 1). From an initial pool of 450 eligible participants, anonymized data from 249 subjects were selected using predetermined criteria to ensure balance across treatment arms for sex (male:female ratio) and race (White/Black representation). To mitigate floor and ceiling effects on cognitive performance, participants were also selected from the middle two quartiles (25th–75th percentile) of baseline composite memory scores on the Performance Assessment Battery. In addition, participants were selected within a restricted range of baseline insomnia symptom severity (ISI; range [13, 28]) and stratified by objectively measured total sleep time at baseline (≤6 h vs. >6 h on PSG) to ensure representation of objectively verified short sleep duration insomnia phenotypes. During subsample selection, participants were also stratified based on objectively measured total sleep time at baseline (≤6 h vs. >6 h on PSG) to ensure representation of short-sleep insomnia phenotypes. All selected participants met the pill count adherence criterion of at least 80% during the 30-day period. Preprocessing and visual inspection of EEG signal quality, blind to medication group, was then conducted to remove data that could not provide accurate spectral estimates. Participants with and without analyzable EEG did not differ (all ps > .81) in age, sex, baseline ISI, race/ethnicity and country, indicating that data exclusion due to EEG quality was not systematically related to key demographic or clinical characteristics (see Table 2S). All inclusion criteria and baseline stratification factors were applied uniformly across randomized treatment arms, and primary analyses focused on within-person change from baseline, supporting valid cross-condition comparisons.

Table 1.

Demographic and baseline characteristics of participants across treatment groups

Treatment group	Mean age (Std)	Total participants	Female No. (%)	Hispanic No. (%)	Non-hispanic No. (%)	Europe No. (%)	North America No. (%)	White race No. (%)
PBO	70.12 (4.46)	41	33(80)	5(12)	36(88)	4(10)	37(90)	33(80)
ZOL	72.64 (4.79)	36	28(78)	4(11)	32(90)	7(20)	29(81)	25(69)
LEM5	70.48 (5.05)	40	32(80)	5(12)	35(87)	3(7)	37(92)	34(85)
LEM10	70.95 (4.36)	41	33(80)	5(12)	36(88)	7(17)	34(83)	38(93)

Abbreviations: PBO, Placebo; ZOL, Zolpidem; LEM5, Lemborexant 5 mg; LEM10, Lemborexant 10 mg; Std, standard Deviation.

Sample size was 158 participants.

PSG and EEG data

PSG procedures are detailed in Rosenberg et al. (2019) [16]. In short, EEG (i.e. F3, F4, C3, C4, O1, and O2), electromyography, electrooculography, and electrocardiography were used to score sleep using standard criteria [16]. An additional clinical PSG was performed, as described [16], prior to the run-in period to screen for sleep disorders other than insomnia. Recordings were reviewed for 249 participants, including EEG preprocessing and visual data quality inspection. Using custom MATLAB (Mathworks, Inc., R2017b) scripts with the EEGLAB toolbox (http://sccn.ucsd.edu/eeglab/), EEG data were notch-filtered (59.5–60.5 Hz) and bandpass-filtered (0.3–35 Hz), then underwent visual inspection and artifact removal. Spectral estimates were computed using 30-second windows with a 5-second sliding step, consistent with previously reported spectral analyses of EEG data collected during sleep [20]. Artifacts were identified on 30-s epochs, and contaminated epochs were excluded. When artifacts spanned multiple channels simultaneously (e.g. movement, electrical noise), the affected vertical time segment was removed across all channels. Channels with >50% unusable data were excluded, and entire nights were discarded if >50% of epochs failed quality control. Automated rejection of high-power artifacts (99th percentile threshold) and manual removal of artifacts lasting >1 second were performed, followed by manual log-linear power-frequency plot reviews to exclude low-quality signals. EEG data of adequate quality for reliable quantitative analysis were available for all visits for 158 out of 249 subjects in NREM sleep (mean age 71.77 ± 0.37, 126 female) and 128 subjects in REM sleep (mean age 70.81 ± 0.37, 99 female). The NREM and REM analytic samples were partially overlapping, as REM data were more frequently excluded for artifact contamination. In addition, phasic eye movement artifacts during REM sleep were also identified and excluded prior to spectral estimation. Each sleep stage was analyzed independently, and comparisons were restricted within stage. Given the robust sample sizes for both NREM (n = 158) and REM (n = 128) sleep, unequal Ns are not expected to bias interpretability or statistical power. Spectral analyses were performed using a Fast Fourier Transform with the multitaper method with 11 discrete prolate spheroidal sequences tapers, a 0.25 Hz bandwidth, zero-order padding, and a 30s window sliding every 5 s to estimate absolute and relative spectral power (0.5–35 Hz). Analyses focused on mean relative power for canonical frequency bands (slow oscillation [SOs]: 0.5–<1 Hz, delta: 1–<4.5 Hz, theta: 4.5–<7.5 Hz alpha: 7.5— < 11 Hz, slow sigma: 11— < 13 Hz, fast sigma: 13— < 16 Hz, beta: 16— < 28 Hz, and gamma: 28—35 Hz), normalizing for overall power to enable condition comparisons. To evaluate the robustness of the primary findings to power normalization strategy, we conducted sensitivity analyses using log-transformed absolute spectral power on the frontal electrode site, applying the same repeated-measures modeling framework; these results are reported in the Supplement Table 2S. For each visit (baseline, acute, chronic), spectral estimates were averaged across both PSG recording nights unless one night was excluded for poor quality. Prior to averaging, we confirmed that spectral power did not differ significantly between the two nights within each condition (all ps > .14), supporting the use of averaged values to improve reliability. Analyses in the main text focus on data averaged across frontal EEG derivations (i.e. F3, F4), as frontal regions are particularly sensitive to alterations in sleep neurophysiology associated with insomnia and its pharmacological treatment [21, 22]. Averages across central (C3, C4) (see Figure 2S and Table 3S for NREM, and Figure 2S and Table 4S for REM sleep) and occipital (O1, O2) (see Figure 4S and Table 5S for NREM, and Figure 2S and Table 6S for REM sleep) derivations are reported in the Supplement. In addition, sleep stage architecture measures across treatment groups are reported in Table 7S in the Supplement. To reduce night-to-night variability and maximize reliability, spectral estimates were averaged across both nights at each visit. While some studies exclude the first study night to minimize the ‘first-night effect [23], the participants were familiar with PSG procedures from screening and randomization visits, reducing the likelihood of pronounced adaptation effects. Importantly, high-density EEG work has shown that although first-night recordings may feature modest increases in high-frequency activity and reduced slow-wave power, these spectral changes tend to be of small to medium size, highlighting the overall robustness of spectral measures across nights [24]. Averaging both nights thus enhances precision and avoids unnecessarily discarding valid data. In addition, participants with analyzable NREM and REM EEG did not differ from those excluded in age, sex distribution, or baseline ISI scores (all p>.10), indicating minimal risk of selection bias. Further, the NREM and REM analytic sets were partially overlapping because REM data were more frequently excluded for artifact contamination. All analyses were performed within each sleep stage, and interpretation was limited to the available data for that stage. Because each stage was analyzed independently, unequal Ns are not expected to affect interpretability.

Statistical analyses

To address potential baseline differences in sleep and ISI between treatment groups, proportional changes from baseline were calculated for acute (1–2 days) and chronic (29–30 days) exposure to LEM5, LEM10, ZOL, and PBO using the following formulas: for acute and chronic changes in sleep measures such as spectral power in each frequency band and for ISI. Negative values indicate a proportional decrease from baseline for sleep measures, while positive values indicate a proportional increase. In addition to baseline-referenced acute and chronic contrasts, evolving treatment effects were assessed using chronic-to-acute change scores, allowing direct evaluation of time-dependent drug effects while minimizing redundant comparisons. For ISI, negative values indicate improvements, and positive values indicate worsening. For statistical analysis of NREM and REM sleep, repeated measures multivariate ANCOVA models were implemented to assess group differences in log-transformed (to account for normality) ratios of change in sleep measures across acute and chronic exposure incorporating Treatment Intervention (PBO, ZOL, LEM5, LEM10) and Sex (Female, Male) as between subjects factors, Age as a between subjects covariate, and Visit (Acute Change Ratio, Chronic Change Ratio) and Frequency Band (SO, Delta, Theta, Alpha, Slow Sigma, Fast Sigma, Beta, Gamma) as within subjects factors. Because ratios could take negative values, we applied a log₁₀(1 + ratio) transformation to normalize distributions and retain both increases and decreases in power while ensuring mathematical validity of the log function. Fisher’s Least Significant Difference (LSD) [25] post hoc testing was implemented to identify which treatment groups differed significantly and to examine the directionality of significant interactions. Further, we used Univariate ANCOVA to assess proportional changes in ISI. We also calculated log-transformed power ratio differences from acute to chronic exposure to evaluate EEG power progression with continued drug use (i.e. the evolution of the drug effect) with one-sample t-tests with LSD adjustments determined whether changes significantly deviated from zero within each treatment group for NREM and REM sleep. We chose this window (i.e. acute to chronic change) because ISI reflects symptoms over the preceding two weeks, making the chronic assessment most representative of the treatment period after acute dosing, and the acute timepoint captures immediate pharmacodynamic responses, while this window indexes maintenance beyond first-dose effects. Partial correlation analyses examined associations between drug effects on sleep EEG and ISI, with post hoc tests identifying the driving treatment groups. Significance was defined as p<.05 following Bonferroni multiple comparisons correction, while 0.05 ≤ p≤.1 were considered trending towards significance. Effect sizes are reported as partial eta squared (ηp²) and Cohen’s d for t-test, reflecting the proportion of variance attributable to each effect after accounting for other model terms. All analyses were conducted in SPSS version 28.0.0.0 (IBM SPSS Statistics, Inc., Chicago, IL).

Results

Final analytical sample sizes after EEG quality control were n = 158 for NREM and n = 128 for REM analyses. Participant characteristics are summarized in Table 1. The analyzed sample had a mean age of approximately 70 years, was 65 % female, and exhibited moderate insomnia severity (mean ISI = 18 ± 4).

Acute and chronic effects of sleep medication on NREM spectral power

We found a significant main effect of Treatment Intervention [F(3 118) = 5.42, p=.002, ηp² = 0.09], with significant interaction effects observed for Frequency Band × Treatment Intervention [F(21 826) = 6.70, p<.001, ηp² = 0.15], Frequency Band × sex [F(7 826) = 2.28, p=.03, ηp² = 0.02], Visit × Treatment Intervention [F(3 826) = 2.21, p=.05, ηp² = 0.01], and Visit × Frequency Band × Treatment Intervention [F(21 826) = 2.69, p<.001, ηp² = 0.06].

Post-hoc comparisons revealed significant treatment-related differences in several frequency bands by both acute and chronic exposure.

During the acute exposure, compared to PBO, ZOL produced significantly increased power, from baseline, in SOs (p=.001), slow sigma (p<.001), fast sigma (p<.001), beta (p=.002), and gamma (p<.001) frequency ranges, as well as decreased, from baseline, in delta (p<.001) and theta (p<.001) (see Figure 2A). Compared to PBO, LEM5, and LEM10 showed no significant differences in change in power in SO, delta, theta, alpha, slow sigma, or fast sigma frequency bands (all ps > .08). However, LEM5 demonstrated lower beta (p=.05) and gamma power (p=.02) relative to PBO. Further, ZOL consistently differed from both LEM5 and LEM10 across multiple frequency bands. Specifically, LEM5 showed lower SOs (p=.003), slow sigma (p<.001), and fast sigma power (p=.002), but higher delta (p=.004) and theta power (p<.001) compared to ZOL. Similarly, LEM10 showed lower SOs (p=.01), slow sigma (p<.001), fast sigma (p=.006), and gamma power (p=.04), but higher delta (p=.02) and theta power (p<.001) relative to ZOL. No significant differences were observed between LEM5 and LEM10 (all ps > 0.11) (see Figure 2A).

Figure 2.

Proportion change from baseline in relative EEG power across frequency bands during NREM sleep. Panel A illustrates the proportion change in EEG relative power for various frequency bands SO: 0.5–<1 Hz, delta: 1–<4.5 Hz, theta: 4.5–<7.5 Hz alpha: 7.5— < 11 Hz, slow sigma: 11— < 13 Hz, fast sigma: 13— < 16 Hz, beta: 16— < 28 Hz, and gamma: 28—35 Hz during NREM sleep under acute drug exposure for PBO (black), ZOL (yellow), LEM5 (light blue), and LEM10 (dark blue). Panel B shows changes under chronic drug exposure. Significant differences between groups are denoted by an asterisk (^*p<.05), and the p-value is reported if there is a trending effect. Data are shown as mean ± standard error of the mean (SEM).

During the chronic visit, compared to PBO, ZOL produced significantly higher power in SOs (p=.04), slow sigma (p<.001), fast sigma (p<.001), and beta (p<.001) frequency ranges, along with significantly lower power in delta (p<.001) and theta (p=.01; Figure 1B). Relative to the LEM groups, ZOL continued to exhibit a distinct spectral profile during chronic exposure. Specifically, ZOL showed greater SO, slow sigma, fast sigma, and beta power, and lower delta and theta power compared to both LEM5 and LEM10 (Figure 1B). In contrast, the LEM groups showed more limited deviations from PBO. Compared to PBO, LEM5 demonstrated reduced slow sigma (p<.001) and fast sigma power (p<.001), whereas LEM10 showed a selective increase in SO power (p=.04). No significant differences were observed between LEM5 and LEM10 across any frequency band during the chronic visit (all ps > .13).

Next, following the significant Visit × Treatment × Frequency Band interaction observed in NREM sleep, we examined changes in the log-transformed chronic-to-acute ratio to assess evolving treatment effects (Figure 3A–D), positive values indicate effects that increased with prolonged exposure, negative values indicate effects that decreased, and zero reflects stable treatment effects after one month of continued exposure. Continued ZOL exposure was associated with progressive increases in theta [t(25) = 2.48, p=.02, d = 0.49)], alpha [t(25) = 3.16, p=.004, d = 0.62)], and fast sigma power [t(25) = 2.16, p=.04, d = 0.42)], with a trend toward increased slow sigma power [t(25) = 1.95, p=.06, d = 0.38)]. In contrast, continued LEM5 exposure resulted in no significant changes across any frequency band (all ps > .27). Continued LEM10 exposure, however, showed a distinct pattern characterized by a significant increase in SO power [t(34) = 2.68, p=.01, d = 0.45)], accompanied by significant decreases in theta [t(34) = −2.19, p=.03, d = 0.37)], alpha [t(34) = −2.89, p=.007, d = 0.49)], slow sigma [t(34) = −2.39, p=.02, d = 0.40)], and fast sigma power [t(34) = −2.01, p=.05, d = 0.34)]. Finally, PBO exposure was associated with a significant increase in gamma power across visits [t(32) = 2.65, p=.01, d = 0.46)].

Figure 3.

Change in NREM relative EEG power across frequency bands from acute to chronic drug use for each treatment group. Panels A–D display changes in relative EEG power during NREM sleep across frequency bands SO: 0.5–<1 Hz, delta: 1–<4.5 Hz, theta: 4.5–<7.5 Hz alpha: 7.5— < 11 Hz, slow sigma: 11— < 13 Hz, fast sigma: 13— < 16 Hz, beta: 16— < 28 Hz, and gamma: 28—35 Hz from acute to chronic drug use: (A) PBO (black), (B) ZOL (yellow), (C) LEM5 (light blue), and (D) LEM10 (dark blue). Significant changes are marked with an asterisk (^*). Notably, ZOL shows an increase in alpha power, while LEM10 exhibits increases in SO power and decreases in alpha power. Trending towards significance is shown via reported p-values.

Acute and chronic effects of sleep medication on REM spectral power

We next examined the acute and chronic effects of sleep medications on REM sleep EEG. There was a significant main effect of Treatment Intervention [F(3 105) = 3.33, p=.02, ηp² = 0.09], a Frequency Band × sex [F(7 735) = 2.13, p=.04, ηp² = 0.02] interaction, and a significant Frequency Band × Treatment Intervention [F(21 735) = 2.19, p=.002, ηp² = 0.06] interaction effect, indicating that drug exposure had frequency-specific effects on spectral power during REM sleep that depended on the specific drug (see Figure 4). However, the lack of a significant Visit×Frequency Band×Treatment Intervention interaction [F(21 735) = 0.58, p=.93, ηp² = 0.02] indicated that treatment effects on EEG in REM sleep did not significantly change with drug exposure. Given this, direct comparisons of acute versus chronic effects of medication exposure on REM sleep were not statistically justified.

Figure 4.

Proportion changes from baseline in relative EEG power across frequency bands during REM sleep. Panel A displays the proportion change in EEG relative power from baseline for various frequency bands SO: 0.5–<1 Hz, delta: 1–<4.5 Hz, theta: 4.5–<7.5 Hz alpha: 7.5— < 11 Hz, slow sigma: 11— < 13 Hz, fast sigma: 13— < 16 Hz, beta: 16— < 28 Hz, and gamma: 28—35 Hz under acute drug exposure for PBO (black), ZOL (yellow), LEM5 (light blue), and LEM10 (dark blue). Panel B shows the same metrics under chronic drug exposure. Significant differences between groups are denoted by an asterisk (^*p<.05), and the p-value is reported if there is a trending effect. Data are shown as mean ± standard error of the mean (SEM).

Post-hoc comparisons during the acute visit revealed that ZOL produced a significant decrease in REM alpha power compared to both PBO (p=.04) and LEM10 (p=.03). ZOL also resulted in significantly higher slow sigma power compared to LEM5 (p=.01). Consistent with this pattern, acute LEM5 exposure showed lower slow sigma power relative to PBO (p=.04). ZOL exposure further resulted in significantly greater fast sigma power compared to LEM5 (p=.004) and LEM10 (p=.02), whereas LEM5 exhibited reduced fast sigma power compared to PBO (p=.04). Finally, beta power was lower in LEM5 compared to ZOL (p=.03) and showed a trend toward reduction versus PBO (p=.06, Figure 4A).

During the chronic visit, ZOL continued to show higher slow sigma power than both LEM5 (p<.001) and LEM10 (p=.02). Chronic ZOL exposure also resulted in significant increases in fast sigma power relative to PBO (p=.02), LEM5 (p<.001), and LEM10 (p<.001), whereas chronic LEM5 treatment resulted in lower fast sigma power compared to PBO (p=.02). Additionally, beta power remained significantly elevated with ZOL compared to both LEM5 (p<.001) and LEM10 (p=.01). Finally, chronic LEM exposure resulted in a decrease in gamma power compared to ZOL (p=.05) and PBO (p=.06-trend, Figure 4B).

Because the Visit × Frequency Band × Treatment interaction was not significant for REM sleep, acute-to-chronic changes were not interpreted as evidence of differential longitudinal treatment effects across groups. However, to provide descriptive context and parallel the NREM analyses, we conducted exploratory within-group one-sample tests on chronic-to-acute change scores. These analyses indicated that ZOL exposure was associated with modest increases in REM alpha (p=.05), fast sigma (p=.03), and beta power (p=.05) over time, whereas no significant longitudinal changes were observed for PBO, LEM5, or LEM10 (all ps > .15).

Acute and chronic effects of sleep medication on ISI

Next, examining the acute and chronic effects of sleep medications on the log-transformed ISI ratio revealed a trend-level main effect of Visit [F(1 146) = 3.63, p=.059, ηp² = 0.024], indicating a modest overall change in insomnia symptom severity from acute to chronic exposure across treatment groups. No significant Visit × Treatment Intervention interaction was observed (p=.347), nor were there significant interactions involving sex (all ps > .32), indicating that changes in ISI over time did not significantly differ across treatment arms. Given the absence of a significant interaction, no post-hoc group comparisons were performed for ISI.

We also examined whether quantitative changes in sleep EEG expression from acute to chronic drug exposure—reflecting evolving drug effects over time—were associated with changes in ISI from acute to chronic drug exposure across treatment groups (Figure 5). In the LEM10 group, greater increases in SOs power were associated with greater reductions in ISI (r = –0.49, p=.004), indicating improved insomnia symptoms with enhanced slow wave activity. Additionally, among LEM10 participants, greater increases in theta power were associated with increased ISI (r = 0.41, p=.02). No significant associations between EEG change and ISI change were observed for PBO, ZOL, or LEM5 (all ps > .15).

Figure 5.

Relationship between changes in NREM SO: 0.5–<1 Hz power and changes in ISI from acute to chronic drug exposure across treatment groups. (A) Represent PBO-black, (B) ZOL-yellow, (C) LEM5-light blue, and (D) LEM10-dark blue. A significant negative correlation was only observed in the LEM10 group only.

Discussion

Our findings align with prior demonstrations that GABAergic hypnotics and DORAs produce distinct spectral signatures [7, 26]; our data newly show dose- and time-dependent evolution in an older insomnia cohort and link SO increases, particularly under LEM10, to symptom improvement. ZOL produced the most pronounced effects across frequencies, particularly during NREM sleep, consistent with previous findings in young adults without insomnia [26, 27]. In contrast, both LEM groups had modest effects on NREM and REM sleep EEG. With continued exposure, ZOL showed progressive increases in theta, alpha, and fast sigma activity, consistent with a profile of accumulating high-frequency activity during sleep over time. In contrast, continued exposure to LEM10 demonstrated increasing SO power but decreasing theta, alpha, and sigma activity over time. These findings suggest that targeting GABA_A versus orexin receptors results in distinct effects on sleep neurophysiology. ZOL maintained its SO enhancement across both acute and chronic exposure, while LEM10 showed a progressive increase in SO power across visits. Importantly, this relationship was specific to the LEM10 group, as no significant associations between SO change and ISI change were observed in the ZOL, LEM5, or PBO groups. Increases in SO power paralleled reductions in ISI, though this relationship was not consistent across all groups. While zolpidem was associated with SO increases that emerged acutely and remained stable across acute and chronic exposure, only the LEM10 group showed a progressive increase in SO power from acute to chronic exposure, and this evolving SO enhancement was associated with reductions in ISI. No significant SO–ISI associations were observed for LEM5 or PBO. This suggests that ZOL and LEM differentially affect sleep EEG and that these effects evolve with continued exposure. To explore whether such neurophysiological changes correspond to improvements in insomnia severity, we examined within-person associations between changes in SO power and ISI scores. Notably, in the PBO group, we observed an increase in gamma power from acute to chronic exposure, suggesting a potential adaptation effect to the recording environment rather than a drug-related change. This increase, without significant changes in other frequency bands, implies that subjective improvements in insomnia may not always align with physiological sleep changes. Since PBO effects were limited to gamma power, these data support the idea that the changes observed in the medication groups were due to pharmacological action rather than subjective bias.

Although our primary focus was on frontal EEG derivations, supplementary analyses at central and occipital sites revealed broadly consistent treatment effects across topographies, with ZOL producing widespread increases in sigma, beta, and gamma power, and LEM showing more circumscribed modulation. Notably, ZOL’s sigma enhancement was strongest over central regions, aligning with the role of central spindles in thalamocortical gating [28]. In contrast, occipital derivations exhibited relatively weaker drug-related changes, particularly for SOs, consistent with the frontal predominance of slow wave generation [29]. These topographical nuances suggest that while both treatments affect global sleep neurophysiology, the most functionally relevant changes may occur in regions specialized for spindle–SO coordination and higher-order cortical integration.

Mechanistic differences and implications for sleep function

ZOL and LEM differ in mechanisms of action, which likely contribute to their differential effects on sleep EEG. ZOL, a BZRA, enhances GABA_A-mediated inhibition, but its effects on sleep neurophysiology resemble sedation more than natural sleep [30]. ZOL is associated with increases in mid- and high-frequency power that continue to accumulate over time, while its effects on SO activity emerge acutely and remain relatively stable with continued exposure. In contrast, LEM exhibited a more selective temporal profile, with SO enhancement emerging primarily following chronic administration, particularly at frontal sites associated with sleep homeostasis. Thus, while both agents increased SO power, they differed in the timing, regional specificity, and broader spectral context in which these changes occurred. In contrast, DORAs, such as LEM, suppress wakefulness by inhibiting orexin signaling, a key regulator of arousal [31]. Unlike GABA_A agonists, DORAs appear to preserve sleep microarchitecture and have been associated with fewer cognitive and psychomotor side effects, lower dependency risks, and minimal withdrawal symptoms [32–34]. Studies suggest that suvorexant, another DORA, effectively manages insomnia in individuals with AD without exacerbating cognitive decline or increasing fall risk [35, 36]. Furthermore, orexin hyperactivity has been implicated in age-related sleep fragmentation, which may contribute to poor sleep quality in older adults and those at risk for AD [37, 38].

The frequency-specific [39] changes observed here may reflect distinct sleep processes and their relevance to insomnia pathophysiology. In NREM sleep, SOs (<1 Hz) and delta activity (1–4 Hz) both occupy the low-frequency range but arise from distinct mechanisms—SOs primarily reflect cortical bistability and homeostatic regulation, whereas delta power reflects broader thalamocortical synchronization that can also accompany sedative drive. SOs are central to synaptic homeostasis and memory consolidation, whereas spindles (sigma activity) support thalamocortical communication and sleep-dependent memory processing. Indeed, increased spindle activity may be as functionally beneficial for cognitive restoration as enhanced SO activity when the two are properly coupled. ZOL robustly enhanced sigma, beta, and gamma activity across both NREM and REM, a pattern resembling sedation-like states and paralleling the elevated high-frequency activity (beta/gamma) [40–42] consistently reported in insomnia disorder as a marker of cortical hyperarousal [43]. In contrast, LEM produced more selective modulation, particularly increasing SOs and reducing alpha activity. Importantly, this SO enhancement under LEM10 was frontal-specific, consistent with the known frontal predominance of homeostatic slow wave generation. This subtler, regionally focused profile may represent a normalization of sleep EEG features, whereas ZOL’s global high-frequency increases resemble the cortical hyperarousal patterns often reported in insomnia [22]. ZOL, by comparison, increased SO power alongside widespread sigma and high frequency enhancement. Thus, while ZOL broadly alters sleep neurophysiology in a manner resembling sedation, LEM10 produced a more localized effect at the sites most sensitive to homeostatic regulation, suggesting a profile more closely aligned with natural sleep processes. However, both mechanisms ultimately promote sleep and reduce insomnia symptoms, indicating that differing EEG profiles do not necessarily imply superiority of one therapeutic pathway over the other. Importantly, these treatment-related spectral patterns were robust across analytic approaches, as comparable effects were observed using both relative and absolute power metrics.

Prior studies in younger adults [44, 45] with insomnia have similarly shown abnormal increases in beta and reductions in slow oscillatory activity, which have been linked to poor sleep continuity and memory outcomes. Although within-person increases in SO in the LEM10 group were associated with reductions in ISI, the overall group-level increase in ISI at the chronic visit suggest that while SO enhancement may contribute to symptom improvement at the individual level, it is not sufficient on its own to drive group-level clinical remission. This suggests that while both medications enhance sleep duration and expression, they may have differing effects on sleep-dependent functions, which may have clinical implications for long-term cognitive outcomes.

Future directions and clinical considerations

By improving sleep while producing more selective, frequency-specific EEG modulation, rather than the widespread high-frequency increases seen with other hypnotics, DORAs may impact sleep function differently from GABAA-acting medications. Future research should directly examine how hypnotic-induced changes in sleep EEG affect long-term cognitive trajectories, dementia risk, and safety in older adults with insomnia. Additionally, it remains critical to determine whether treatment efficacy differs based on insomnia etiology and the presence of baseline neurophysiological hyperarousal, which has been implicated in insomnia pathophysiology.

Limitations and strengths

Our study has several limitations that should be considered when interpreting the results of this study, including an unequal sex distribution, limiting our ability to examine potential sex differences. Additionally, our findings may have limited generalizability to diverse racial, ethnic, and socioeconomic groups. Including a broader age range could reveal whether drug effects vary across the lifespan. In particular, younger or working adults may exhibit different baseline sleep physiology, circadian pressures, and stress-related sleep disruption compared to our predominantly older and likely retired sample, which could meaningfully alter both EEG responses to hypnotics and perceived symptom changes. Examination of a larger group of medications would better elucidate which neurophysiological sleep features are relevant for treatment efficacy and side effects, including in those with different insomnia etiologies and symptom profiles. In addition, examination of the consequences of different sleep medications on functions supported by sleep would address how different treatment mechanisms can affect clinical outcomes. In addition, although the parent trial enrolled individuals meeting DSM-5 criteria for insomnia disorder, the current analyses examine insomnia symptoms (as measured by ISI) and sleep EEG expression. We did not evaluate other diagnostic components (e.g. cognitive/behavioral features, daytime dysfunction), and therefore our findings should be interpreted as reflecting changes in symptom severity and sleep physiology rather than comprehensive treatment effects on insomnia disorder [46]. Finally, the present study implemented a low-density EEG montage, relying on a limited set of frontal, central, and occipital derivations. Although this approach is standard in clinical trials and was sufficient to capture robust drug-related spectral changes, it cannot fully characterize fine-grained topographical patterns of drug effects that may be revealed with high-density EEG. This is important because prior insomnia studies have demonstrated regional specificity of spectral abnormalities, particularly elevated frontal and central beta activity [40] during NREM sleep as a marker of hyperarousal [47], as well as frontally predominant slow wave deficits. Sparse montages may therefore underestimate or miss subtle regional differences. While our frontal findings are consistent with these prior reports, future work using high-density EEG could better resolve the spatial topography of pharmacological effects in insomnia. Future studies using high density EEG arrays may provide a more detailed understanding of regional specificity of medication and insomnia effects. Additionally, because the sample included an older cohort with elevated WASO and balanced representation of individuals with objectively short sleep duration (≤6 hours), the findings may primarily reflect more severe insomnia phenotypes and may not generalize to younger individuals or those with milder or predominantly sleep-onset insomnia. This consideration applies across all treatment arms and does not reflect a mismatch between treatment mechanisms and participant characteristics.

Despite these limitations, our study has several strengths. The clinical trial design includes both acute and chronic exposure to medications. By focusing on older adults with insomnia, a population at high risk for hypnotic use, it provides clinically relevant insights that can inform real-world treatment decisions.

Conclusion

This study highlights the distinct effects of zolpidem and LEM on sleep microarchitecture in older adults with insomnia, following both acute and chronic exposure. Zolpidem induced widespread alterations in NREM and REM sleep EEG, marked by persistent increases in SOs and high-frequency activity, but with a pattern that may resemble sedation more than natural sleep. In contrast, both LEM doses (LEM5 and LEM10) produced minimal changes in sleep EEG relative to PBO, suggesting a more preserved neurophysiological sleep profile.

These findings highlight the differential effects of sleep medications on brain dynamics during sleep and the evolving nature of these effects with continued use. Importantly, the observed changes in sleep EEG were linked to reductions in insomnia severity, particularly in the LEM10 group, suggesting that targeting orexin signaling may provide sleep benefits while minimizing disruptions to sleep neurophysiology. The results emphasize the need for further research into the effects of hypnotic-induced sleep EEG changes, particularly in relation to cognitive trajectories, dementia risk, and treatment safety in older adults with insomnia.

Author contributions

Negin Sattari (Conceptualization [equal], Data curation [equal], Formal analysis [lead], Funding acquisition [equal], Investigation [equal], Methodology [equal], Project administration [lead], Resources [equal], Software [lead], Supervision [lead], Validation [lead], Visualization [lead], Writing—original draft [lead], Writing—review & editing [lead]), Abhishek Dave (Conceptualization [supporting], Formal analysis [supporting], Methodology [equal], Resources [equal], Software [supporting], Visualization [supporting], Writing—original draft [supporting], Writing—review & editing [supporting]), Hamid Niknazar (Conceptualization [equal], Data curation [equal], Formal analysis [equal], Funding acquisition [equal], Investigation [equal], Methodology [equal], Project administration [equal], Resources [equal], Software [supporting], Supervision [equal], Validation [equal], Visualization [equal], Writing—original draft [equal], Writing—review & editing [equal]), Ivy Y. Chen (Conceptualization [supporting], Visualization [supporting], Writing—original draft [supporting], Writing—review & editing [supporting]), Ariel B. Neikrug (Conceptualization [supporting], Formal analysis [equal], Resources [supporting], Validation [supporting], Visualization [supporting], Writing—original draft [supporting], Writing—review & editing [supporting]), Bryce A. Mander (Conceptualization [equal], Formal analysis [equal], Funding acquisition [equal], Investigation [equal], Methodology [equal], Project administration [equal], Resources [equal], Software [equal], Supervision [equal], Validation [equal], Visualization [equal], Writing—original draft [equal], Writing—review & editing [equal]), and Ruth M. Benca (Conceptualization [lead], Data curation [equal], Formal analysis [equal], Funding acquisition [lead], Investigation [equal], Methodology [equal], Project administration [equal], Resources [equal], Supervision [equal], Validation [equal], Visualization [equal], Writing—original draft [equal], Writing—review & editing [equal])

Funding

This work was funded by Eisai Co., Ltd. under study E2006-G000-304 (Principal Investigators: Ruth M. Benca and Bryce A. Mander). Additional support was provided by the National Institute on Aging of the National Institutes of Health under Award Number K01AG068353 (Principal Investigator: Bryce A. Mander) and T32AG000096 (Principal Investigator: Negin Sattari).

Disclosure statement

Financial disclosure: Authors declared that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Dr. Ruth M. Benca has served as a consultant for Alkermes, Biogen, Eisai, Haleon, and Idorsia. Dr. Bryce Anthony Mander currently serves on a scientific advisory board for AstronauTx, Ltd.

Non-financial disclosure: The authors declare that they have no non-financial interests or personal relationships that could be perceived as influencing the work reported in this paper.

Data availability

The data supporting the findings of this study are available from the corresponding author upon reasonable request. Data are not publicly available due to privacy and ethical restrictions, as they contain sensitive information that could compromise participant confidentiality such as EEG.

Role of the Funder/Sponsor

Eisai was involved in the study’s design and conduct, as well as in data collection, management, analysis, and interpretation. The company also contributed to the preparation, review, and approval of the manuscript and participated in the decision to submit it for publication.

Trial Registrations

ClinicalTrials.gov, SUNRISE 1, identifier: NCT02783729 (https://clinicaltrials.gov/study/NCT02783729); EudraCT identifier: 2015-004347-39.