Brain water dynamics across sleep stages measured by near-infrared spectroscopy: Implications for glymphatic function

The study addresses how brain fluid dynamics vary across sleep states and whether these variations reflect glymphatic activity, a clearance pathway involving CSF-ISF exchange thought to be enhanced during NREM sleep and reduced during REM sleep. Prior animal and human work indicates increased CSF and ISF volumes during sleep, with glymphatic function linked to EEG slow wave activity and vasomotor pulsations. However, invasive methods (intrathecal contrast MRI) and MRI’s incompatibility with natural sleep hinder temporal resolution across sleep transitions. The authors hypothesize that brain water content measured non-invasively with NIRS increases during transitions into NREM sleep, decreases during transitions into REM sleep and upon awakening from NREM, and is larger in the first NREM cycle than the last, reflecting state-dependent glymphatic-related dynamics.

Background work has characterized glymphatic pathways via tracer-based imaging (two-photon microscopy, dynamic contrast MRI) and demonstrated increased solute transport under sleep or dexmedetomidine anesthesia, with human studies showing brain-wide tracer movement and links to sleep deprivation/quality. Glymphatic activity is associated with expansion of CSF and ISF volumes without total brain volume change and correlates with EEG slow wave activity and vasomotor pulsations during NREM. There is ongoing controversy regarding the mechanisms (bulk flow vs diffusion) and the extent of CSF-ISF exchange. Diffusion MRI and impedance spectroscopy suggest increased extracellular water during sleep, while some recent work proposes reduced brain clearance during sleep/anesthesia, complicating interpretation. NIRS has been proposed to detect water content changes and potentially reflect glymphatic-related CSF dynamics but requires validation and careful separation from hemodynamic confounds such as CBV.

Design and participants: A prospective overnight study at Seoul National University Bundang Hospital recruited 65 healthy volunteers (19–49 years), excluding for neurological disease, major medical comorbidities, smoking/heavy alcohol use, drug abuse, major sleep problems, or NIRS contraindications. Screening used validated questionnaires (Berlin Questionnaire, Insomnia Severity Index, CH-RLSq, Epworth Sleepiness Scale, Morningness-Eveningness Questionnaire, Pittsburgh Sleep Quality Index). After exclusions (poor sleep quality, OSA, PLMs, excessive PSG/NIRS artifacts), 41 participants (men=24; mean age=26.9±7.7) were analyzed. Ethics approval was obtained; informed consent was given.

Polysomnography (PSG): Full overnight PSG (Grass Comet-PLUS; electrodes F3-A2, F4-A1, C3-A2, C4-A1, O1-A2, O2-A1; 200 Hz sampling) with EOG, ECG, submental and tibialis EMG, respiratory parameters, oximetry, snoring, and body position. Sleep stages were scored in 30 s epochs (AASM criteria). Standard indices computed included TST, sleep efficiency, stage percentages/durations, AHI, oxygen desaturation index, arousal index, PLM index.

NIRS device and acquisition: A custom wireless NIRS operated at 8 Hz for up to 10 h, using three wavelengths (780, 850, 925 nm) to measure oxyhemoglobin, deoxyhemoglobin, and water content. Two detectors and one laser source per forehead side, detector-source separation 3 cm, enabling ~1.5 cm penetration to sample cortical/CSF compartments. Measurements were in arbitrary units using the Modified Beer-Lambert Law with published extinction coefficients and differential pathlength factors. Left/right measurements were averaged to mitigate postural effects.

Preprocessing and artifact handling: Physiologic noise (heartbeat, respiration, neuronal activity, vasomotion) reduced with a low-frequency filter (0.01 Hz cutoff). Motion artifacts detected via gyroscope and corrected using spline interpolation. Channels with baseline SNR <30 dB were excluded; backup channels were used when available.

CBV (plasma water) filtering: To reduce crosstalk between water and hemoglobin (CBV-related plasma fluid), a static linear minimum mean square estimator (LMMSE) was applied: ΔH2O_filtered(t) = ΔH2O_meas(t) − αΔC_HbT(t), estimating and removing the plasma-correlated component to isolate tissue water changes more reflective of ISF/CSF. Algorithm performance was validated via simulation (Supplementary).

Transition analyses: Measurements began 30 min before sleep and continued through the night and for 60 min post-awakening. Transitions analyzed included WAKE→NREM, NREM→WAKE, REM→NREM, and NREM→REM. The final 5-min segment of the source stage served as baseline; destination stages had to be stable (≥60 min for WAKE→NREM, NREM→WAKE, REM→NREM; ≥30 min for NREM→REM). Water signals were segmented into 5-min epochs and block-averaged; changes ΔH2O(X) computed relative to baseline. Epochs with SNR <30 dB at any point were excluded. When multiple transitions of the same type occurred within a subject, values were averaged. Sensitivity analysis of baseline bin length (5, 10, 15, 20 min) for WAKE→NREM showed consistent trends.

First vs last NREM comparison: Because SWA is greater in early NREM, transitions into the first and last NREM periods were compared. For these analyses, the first 5-min destination segment (WD,1) served as baseline to minimize heterogeneity of source stages. Paired t-tests evaluated ΔH2O differences across 5-min segments.

Statistics: Linear mixed-effects models with Time as fixed effect and Subject as random intercept estimated transition-related changes at 5-min intervals. Cohen’s d effect sizes calculated. Analyses performed in SPSS v22 and MATLAB.

• Sample characteristics (n=41): Mean TST 6.4±1.1 h; sleep efficiency 79.4±12.3%; stage composition N1 17.3%, N2 60.3%, N3 2.4%, REM 20.1%; AHI 1.2±1.7 events/h; minimum nocturnal SpO2 92.2±2.7%.
• Transition outcomes (linear mixed-effects model; ΔH2O relative to source-stage baseline):
  • WAKE→NREM (n=28): Significant increase in water content, peaking ~40 min; overall ΔH2O=0.57 A.U. (95% CI 0.33–0.81), p<0.001, d=0.77. Segment-wise significant increases from 20–25 min onward (e.g., 40–45 min: 0.52 [0.29–0.76], p<0.001, d=0.62).
  • NREM→WAKE (n=11): Significant decrease; overall ΔH2O=−0.93 A.U. (95% CI −1.25 to −0.60), p<0.001, d=−1.25. Progressive declines with stabilization ~50 min (e.g., 45–50 min: −0.98 [−1.30 to −0.65], p<0.001, d=−1.31).
  • REM→NREM (n=34): Significant increase; overall ΔH2O=0.62 A.U. (95% CI 0.46–0.78), p<0.001, d=1.10; continued rise beyond 60 min (e.g., 55–60 min: 0.64 [0.47–0.81], p<0.001, d=1.02).
  • NREM→REM (n=17): Significant decrease; overall ΔH2O=−0.40 A.U. (95% CI −0.70 to −0.09), p<0.05, d=−0.53; smaller magnitude than other transitions.
• First vs last NREM periods: The increase in brain water during the first NREM period was faster and larger than during the last, with significant differences emerging at 15 min and persisting ~30 min; largest difference at 25–30 min: Δ=0.70 A.U., d=0.86, p<0.01.
• Robustness: Similar transition-dependent patterns observed in uncorrected (raw) data, supporting robustness of findings despite CBV correction.

Findings demonstrate pronounced, state-dependent fluctuations in brain water content across sleep transitions: increases during NREM onset and after REM→NREM, and decreases during NREM→WAKE and NREM→REM. These patterns parallel established physiology in which glymphatic-related processes (ISF/CSF expansion, enhanced vasomotor pulsations, and elevated SWA) are prominent during NREM, particularly early in the night, and reduced during REM. The larger and faster water increases in the first NREM cycle align with higher SWA and homeostatic sleep drive. By applying CBV filtering to the NIRS signal, the analysis emphasizes water shifts more attributable to ISF and extracerebral CSF rather than plasma-related changes, strengthening the interpretation of sleep-stage dependence. However, because NIRS measures bulk tissue water without compartmental specificity and glymphatic mechanisms remain debated (bulk flow vs diffusion; variability in CSF-ISF exchange), the results should be interpreted as reflecting broader brain fluid dynamics rather than direct glymphatic flow. The technique’s non-invasive, continuous nature suggests utility for tracking sleep-related brain fluid changes and potentially complementing MRI or biomarker approaches to clarify glymphatic function in humans.

Water-sensitive NIRS enables non-invasive, continuous monitoring of brain water content throughout natural sleep. The observed increases during NREM and decreases during REM and wakefulness are consistent with, but do not prove, glymphatic-related activity, with stronger effects in early-night NREM. The method does not directly measure glymphatic flow or resolve fluid compartments. Future studies should validate specificity by combining NIRS with advanced MRI, tracer-based methods, or CSF biomarker assessments across sleep stages, refine signal processing (e.g., adaptive CBV filtering, short-separation regression), and expand cohorts to improve statistical power and balance. Establishing reliable measures of sleep-related brain fluid dynamics could advance understanding of neurodegenerative risk mechanisms, inform interventions, and facilitate monitoring of therapeutic responses.

• No a priori statistical power analysis due to unknown expected effect sizes for NIRS-measured water changes; future studies should use current effect estimates for planning.
• Unbalanced datasets and missing observations across transitions because participants maintained destination stages for varying durations; linear mixed-effects modeling mitigated this but precision may be affected.
• First-night effect not controlled, potentially influencing sleep architecture and quality.
• Use of published, non-individualized optical properties (scattering/absorption), increasing measurement error and vulnerability to crosstalk.
• Motion artifacts can degrade signal; analysis excluded artifact-affected data, limiting exploration of motion dependencies.
• NIRS measures bulk water without compartment specificity (ISF, ICF, CSF, plasma), making inference to glymphatic activity indirect even with CBV correction.
• Static LMMSE assumes stable ratio between plasma fluid and total hemoglobin; hematocrit variations during sleep may necessitate adaptive filtering for non-stationary signals.
• Potential long-term signal drift (optode repositioning, temperature changes, systemic hemodynamics); although mitigated by preprocessing and transition-focused analyses, comparisons between early and late NREM may be more susceptible.
• Lack of phantom validation; constructing dynamic, multi-layered phantoms with water-specific wavelengths is technically challenging.
• Limited spatial coverage and depth, susceptibility to motion, and absence of structural information compared to MRI; extracerebral contributions may persist without short-separation regression.
• Other physiological factors (CSF production/absorption, antidiuretic hormone secretion, vasomotion, respiration) could influence measured water dynamics.