Sleep as a window into brain health: clinical implications of sleeping brain wave-structure associations

What do brain waves during sleep tell us about brain health? In clinical practice, they are barely used for sleep staging. A gap exists between extensive sleep electroencephalograph (EEG) research findings and their applied clinical value. With the goal of improving clinical utility, we read the Kozhemiako et al. paper [1] (SLEEP 2025 in press) with great interest. They identified several EEG-structure associations. The study represents an important step toward understanding how sleep reflects brain health and morphology in a community-dwelling population, beyond a previous study using a clinical dataset [2]. Here, we discuss the clinical implications of structure-EEG associations in various diseases. We argue that sleep is a window into brain health, including suggesting specific structural pathologies, tracking disease progression, and measuring treatment responses.

Sleep disturbances are linked to cognitive impairment [3–6] and neurodegenerative diseases [7–11]. Alzheimer’s disease (AD) leads to hippocampal atrophy and cortical thinning, has primarily cortical manifestations, and leads to episodic memory and learning deficits [12, 13]; whereas Parkinson’s disease (PD) and Lewy body diseases start subcortically, with early degeneration in the basal ganglia and brainstem, leading to slowed processing speed and executive dysfunction [12]. These differences are reflected in sleep patterns. In AD, EEG microstructure disturbances appear earlier, including reduced NREM slow oscillations and sleep spindles [14, 15], followed by macro-level changes in sleep architecture. Conversely, PD and Lewy body diseases often have rapid eye movement (REM) sleep abnormalities and sleep fragmentation due to brainstem degeneration [16], impacting sleep macro architecture first, and later EEG microstructure as cortical structures are affected. In the same line, using our previously developed sleep EEG-based brain age index [17], which mainly includes age-dependent patterns with cortical origin, we found older brain age index in people with AD [18], but not in PD. Therefore, sleep EEG microstructures help differentiate neurodegenerative diseases with cortical versus subcortical origins. Moreover, sleep can measure treatment effectiveness, such as treating AD using lecanemab [19], where reduced AD pathology might show improvement in EEG [8, 20]. At its root, the currently limited clinical utility of these findings reveals a gap in bridging the neurodegeneration-sleep gap.

Unlike AD’s cortical or PD’s subcortical manifestations, multiple sclerosis (MS) involves both components. MS is a chronic autoimmune disease characterized by demyelination, inflammation, and neurodegeneration, involving impaired WM integrity and cortical thinning [21, 22]. However, there are no studies of sleep EEG microstructures in MS in the literature, indicating a critical gap to address. Due to its mixed pathology, MS could be a model disease for understanding structure-related issues, tracking disease progression, and guiding treatment.

Obstructive sleep apnea (OSA) involves multiple structural pathologies in the WM that link the limbic system, pons, cortices, and cerebellum [23], and GM related to the motor regulation of the upper airway and frontal and temporal cortices supporting cognition [24, 25]. These structural pathologies can manifest as EEG abnormalities, including reduced slow-wave activity [3, 26] and impaired spindle-breathing coupling [27]. These patterns may help identify patients at risk for cognitive decline, dementia, or mood disorders secondary to OSA. An observational case-control study of 18 participants with high-density EEG revealed reduced powers in multiple bands in the parietal region during sleep in participants with OSA [28]. In an observational study in 36 children between 5 and 9 years old, OSA participants had lower sigma band power in N2 than primary snoring and healthy controls, which modestly correlated with memory consolidation in both sleep and wake conditions. Despite these observational studies, the Apnea Positive Pressure Long-term Efficacy Study (APPLES) [29, 30] is a large-scale randomized controlled trial (RCT) showing no long-term cognitive benefit of continuous positive airway pressure (CPAP) at a population level. Meanwhile, an RCT [31] of 148 participants with normal cognition and apnea–hypopnea index ≥15/hour showed a beneficial effect of CPAP at 6 months on cortical thickness and functional connectivity of the default mode network, but no improvement in cognitive score. Meanwhile, some small-scale RCTs of CPAP show beneficial effects on declarative memory [32, 33]. The differential response to CPAP may indicate brain resilience to OSA at the individual level [34–36]. However, in-depth analyses of physiological responses to CPAP are needed. Specifically, NREM sleep consists of stable and unstable phases, as indicated by the cyclic alternating pattern [37] from the cortical perspective and cardiopulmonary coupling (CPC) [38] from the respiratory-autonomic perspective. In a healthy brain, CPAP induces REM rebound [34] and stable sleep rebound (non-cyclic alternating pattern or CPC high-frequency coupling), reflecting sleep quality otherwise masked by OSA.

In line with the above clinical implications for cortical and subcortical diseases, Kozhemiako et al. [1] have a few limitations and future directions below that can further enhance their clinical value.

The multi-organ interactions affecting sleep EEG, which originate mainly subcortically, should be considered. Sleep is holistic and involves complex network physiology across organ systems. Acknowledging these interactions provides a clearer understanding of physiology to inform clinical decisions. Notable examples include infraslow oscillations correlating with norepinephrine release in the locus coeruleus [39, 40], spindle-breathing coupling [27, 41] reflecting respiratory nuclei, REM atonia, and phasic/tonic REM [42, 43] related to brain-muscle coordination, and heart rate changes post-arousal [44, 45] indicating brain-autonomic interaction.

Kozhemiako et al. found that the sleep EEG markers of aging were independent of MRI metrics that go beyond gross anatomy—a result possibly driven by non-neuronal factors that are sensitive to aging and not captured by the MRI metrics. As our understanding of sleep physiology advances, non-neuronal factors, such as glial cells [46] and the glymphatic system [40, 47], are becoming increasingly important future directions for both clinical and research studies. Microglia actively regulate neuronal excitability [48, 49], influencing EEG spectral dynamics; oligodendrocytes produce myelin essential for WM integrity; astrocytes maintain the blood-brain barrier and influence glymphatic clearance, directly impacting brain health [50, 51]; and the glymphatic system and meningeal lymphatic vessels (MLV) [52] regulate synaptic physiology. Future brain structure quantifications incorporating advanced imaging techniques sensitive to glial cells and the glymphatic system (e.g. myelin water imaging [53], PET markers of microglial activation [54], and diffusion tensor imaging along perivascular spaces [55]) could potentially yield more comprehensive insights into how non-neuronal cells and systems shape sleep EEG.

Functional outcomes such as cognition and non-refreshing sleep, or incident disease outcomes [56], should be considered. For cognitive functional outcome, it is well known that the spindle-SO coupling supported by the thalamocortical circuits is important for memory consolidation [57], especially episodic memory. On the other hand, procedural memory and mood are more associated with REM and amygdala activation [58, 59], whereas the associated REM EEG patterns are less studied. Another important cognitive functional outcome is forgetting, which could be related to the SO duration and fast-to-slow switching SO ratio. There are studies [60] demonstrating that spindles coupled to “short SOs,” i.e. delta oscillation (1–4 Hz), are responsible for forgetting. Together with spindles coupled to “long SOs,” i.e. slow oscillations (<1Hz) that facilitate remembering, they form a dynamic homeostasis to maintain the memory system. For spindles, this type of study is also an important venue for comparing slow vs. fast spindles, which is under debate as to whether they have the same structural origin and functional outcomes [14]. For non-refreshing sleep, the alpha–delta pattern (alpha intrusion) in NREM is also not studied here, which may reflect impairments in thalamocortical gating [61].

In summary, based on Kozhemiako et al. [1], we advocate that cortical and subcortical sleep profiles contain clinical information about disease pathology, progression, and recovery, which serves as a window into brain health.

Disclosure Statement

Financial disclosure: HS and WG have nothing to disclose. MBW is a co-founder, scientific advisor, consultant, and has a personal equity interest in Beacon Biosignal. MBW received support from the NIH (RF1NS120947, R01HL161253) and NSF (2014431). RJT was supported by an AASM Foundation Strategic Research Award. Nonfinancial disclosure: HS, WG, and MBW have nothing to disclose. RJT discloses (1) patent and license/royalties from MyCardio, LLC, for the ECG-spectrogram; (2) patent and license/royalties from DeVilbiss-Drive for an auto-CPAP algorithm; (3) consulting for Jazz Pharmaceuticals, Guidepoint Global, and GLG Councils.