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

Abstract

This study investigates brain water dynamics across the sleep-wake cycle using near-infrared spectroscopy (NIRS) and linear mixed-effects modeling, motivated by prior observations that glymphatic activity increases during non-rapid eye movement (NREM) and decreases during REM sleep. Forty-one healthy volunteers underwent polysomnography with concurrent cerebral NIRS, with measurements taken 30 minutes before sleep, throughout the night, and for 60 minutes after waking. Brain water content (arbitrary unit, A.U.) was block-averaged for 5-minute epochs and analyzed across WAKE→NREM, NREM→WAKE, NREM→REM, and REM→NREM transitions. Water content significantly increased during WAKE→NREM (0.57 A.U., d = 0.77, p < 0.001) and decreased during NREM→WAKE (−0.93 A.U., d = −1.25, p < 0.001). Decreases during NREM→REM (−0.40 A.U., d = −0.53, p < 0.05) were followed by increases during REM→NREM (0.62 A.U., d = 1.10, p < 0.001). Brain water accumulation was significantly greater during the first compared to the last NREM cycle (0.70 A.U., d = 0.86, p < 0.01). These findings reveal robust, state-dependent fluctuations in brain water content that parallel established glymphatic physiology. Water-sensitive NIRS may offer a promising non-invasive approach to monitoring sleep-related brain fluid dynamics in humans, though further multimodal studies are needed to determine its specificity for glymphatic activity.

Introduction

Sleep is a complex and vital physiological process that is essential for maintaining overall health and brain function.^1–4 An emerging area of sleep research is the study of the glymphatic system, a network that activates during sleep and facilitates the clearance of waste products from the brain.^5–8 The glymphatic system involves periarterial passage of cerebrospinal fluid (CSF), the aquaporin-4 dependent influx of CSF into the brain interstitial space, the mixing of CSF with interstitial fluid (ISF), and the drainage of these fluids via perivenous pathway to meningeal lymphatics or systemic circulations.^5,8–10

Although the glymphatic system is still subject to debate,^11–14 evidence has increasingly supported its critical role in maintaining brain homeostasis. It regulates ISF and facilitates the removal of metabolic waste, including amyloid-beta, which is implicated in neurodegenerative diseases such as Alzheimer disease.^{5,6,8,10,15–17}

In animal studies, the glymphatic system has been documented, traced, and evaluated using intracisternal injection of fluorescent dyes or gadolinium-based MRI contrast, followed by imaging techniques such as two-photon microscopy or dynamic contrast MRI scanning in awake, sleeping, and anesthetized states.^5,6,9,18,19 While two-photon microscopy provides excellent spatial resolution, ²⁰ it is unsuitable for humans due to its invasiveness and potential adverse effects of fluorescent tracers. Dynamic contrast MRI with intrathecal gadobutrol injection has been used to study glymphatic activity in humans. ²¹ This technique has demonstrated brain-wide glymphatic enhancement and clearance through the parasagittal dura and cervical lymph nodes.^21–23 It has also shown altered clearance patterns due to sleep deprivation and poor sleep quality.^24,25 However, this technique's invasiveness and unapproved status limit its use, and MRI's incompatibility with natural sleep complicates its ability to observe sleep-wake transitions, including non-rapid eye movement (NREM) and REM sleep cycle.

Activation of the glymphatic system is associated with CSF inflow into the interstitial space, resulting in an approximately 40–60% increase in ISF volume within 30 to 60 minutes after sleep onset.^6,18 This increase is explained by the expansion of the interstitial space and the simultaneous contraction of intracellular volume. ⁶ The glymphatic activity is also associated with CSF volume. During sleep or under dexmedetomidine anesthesia, CSF volume increases without altering the total brain volume.^19,26 Since CSF and ISF consist primarily of water, measuring brain tissue water content could provide a means to evaluate glymphatic activity. Specifically, sleep-wake state-dependent changes in brain CSF and ISF content are potential targets for glymphatic measurements.

Near-infrared spectroscopy (NIRS) is a non-invasive imaging technique widely used in neuroscience research. It employs lasers in the near-infrared spectrum, typically between 700 and 1000 nm, where light is specifically absorbed by oxyhemoglobin and deoxyhemoglobin. NIRS monitors cerebral hemodynamics by detecting the differential absorption and scattering of near-infrared light.^27,28 By leveraging the water absorption spectrum, NIRS can also measure tissue water content. ²⁹ Recent study suggests that NIRS has the potential to monitor glymphatic activity by detecting dynamic variations in CSF water content in the occipital area while the subjects perform visual task in an awake state. These variations are presumed to correlate with dynamics of glymphatic system. ³⁰

In this study, we designed a wireless, portable NIRS system to measure water content and hemodynamic status. We applied the device to the forehead throughout the entire duration of overnight sleep to assess sleep-wake-dependent changes in water content in the frontal area, a key region for glymphatic activity.^8,21 The brain water content measured by NIRS originates from extracerebral CSF as well as intracellular fluid (ICF), ISF, and plasma in brain parenchyma. ³⁰ When the glymphatic system becomes active during sleep or under the anesthesia, both ISF and CSF volume increase while ICF volume decreases to facilitate CSF inflow.^6,19,26 Plasma fluid directly correlates with cerebral blood volume (CBV), and total hemoglobin, defined as the sum of oxyhemoglobin and deoxyhemoglobin, represents CBV. For the analysis, we implemented an algorithm to filter out the water fraction associated with CBV in our measured signal. ³¹ Because CBV fluctuates dynamically during sleep—typically decreasing in NREM and varying widely in REM,^32–34 removing this component enhances the detection of water shifts that are most likely attributable to ISF and extracerebral CSF expansion. Given that ICF volume contracts when the glymphatic system becomes active, any net increase in the filtered water content during sleep transitions (e.g., WAKE→NREM, REM→NREM) may reflect glymphatic activity. Understanding these dynamics provides additional insight into how the glymphatic system operates across different sleep stages, although these signals likely reflect the combined dynamics of ISF, extracerebral CSF, and ICF, rather than direct glymphatic flow. Previous research suggests that the glymphatic system is active during NREM sleep, relative to wakefulness, but not during REM sleep.^6,35–37

This study aims to measure brain water dynamics across the sleep-wake cycle using NIRS and to test the hypothesis that variations in brain water content reflect glymphatic activity. Specifically, we hypothesize that brain water content increases during transitions to NREM sleep but decreases during transitions to REM sleep and upon awakening from NREM sleep. Furthermore, we propose that the increase in water content during the transition to NREM sleep will be greater in the first NREM cycle than in the last. To test this hypothesis, we conducted overnight polysomnography (PSG) alongside concurrent NIRS measurements in young, healthy volunteers.

Materials and methods

Study subjects

Sixty-five healthy volunteers aged between 19 and 49 years (men = 37, mean age = 25.5 ± 8.0 years) participated in this study at the Center for Sleep Medicine, Seoul National University Bundang Hospital between August 2018 and February 2021. All participants underwent a restrictive screening protocol using structured questionnaires about sociodemographic characteristics, comorbid medical conditions, and sleep problems to exclude subjects with confounding factors that could potentially influence the glymphatic activity. The exclusion criteria were: (1) history of stroke, epilepsy, brain trauma, or other central nervous system diseases; (2) presence of chronic disabling conditions (hypertension, heart disease, diabetes, and respiratory diseases); (3) history of smoking or heavy alcohol use; (4) history of drug abuse or dependence; (5) presence of major sleep problems; and (6) the presence of contraindications for wearing the NIRS equipment. This study was approved by the Seoul National University Hospital (B-1710/427-006, B-2011/586-304). All subjects were informed about the experimental protocol prior to participation and gave their written informed consent to participate. The study was conducted in accordance with the ethical principles of the Declaration of Helsinki.

Major sleep problems were screened by validated questionnaires and confirmed by a sleep clinician. The questionnaires included the Berlin Questionnaire, Insomnia Severity Index, Cambridge–Hopkins diagnostic questionnaire for restless leg syndrome, Epworth Sleepiness Scale, Morningness-Eveningness Questionnaire, and Pittsburgh Sleep Quality Index.^38–43 We excluded subjects with high risk for obstructive sleep apnea (OSA), clinical insomnia, restless legs syndrome, excessive daytime sleepiness, poor sleep quality, or extreme chronotype. Three participants were excluded due to poor sleep quality, resulting in a total of 62 participants. Additional exclusions were made for participants with OSA (3 participants), periodic limb movements (PLMs) during sleep (2 participants), or excessive PSG or NIRS artifacts (16 participants).

Experimental procedures

The participants were fitted with PSG and NIRS probes between 21:00 and 22:00, after which they were instructed to stay in bed (Figure 1(a)). They were required to have habitual bed (22:00–00:00) and wake-up (07:00–09:00) times broadly consistent with the self-reported sleep timings, and to obtain ≥7 h of sleep. The measurements began 30 minutes before falling asleep and were continued for 60 minutes after waking up to compare water content changes during wakefulness with those during sleep states.

Figure 1.

The measurement setup using NIRS and PSG. (a) Schematic representation and the photograph of the NIRS device and PSG attachment. (b) The customized wireless NIRS device. The system consists of the control board and two optode patches simultaneously measuring the left and right frontal areas and (c) illustration of the light propagation during NIRS measurement, penetrating the brain.

Height (cm) and weight (kg) were measured in the morning fast state after the PSG and NIRS evaluations.

Polysomnographic recording

The participants underwent a full overnight PSG between February 2018 and June 2019. PSG was performed using Grass® Comet-PLUS®encephalography (EEG)/PSG equipment (Natus Medical Incorporated, San Carlos, CA, USA) with the following parameters: F3-A2, F4-A1, C3-A2, C4-A1, O1-A2, and O2-A1; 200 Hz sampling rate with electrooculography; electrocardiography; and submental and tibialis anterior electromyography (EMG). The EEG signal was bandpass filtered at 0.3–35 Hz. Thoracic and abdominal efforts from plethysmography belts, airflow, oxygen saturation, snoring, and body position were also recorded. All overnight sleep stages in 30 s epochs, including awake, N1, N2, N3, and REM sleep, were scored by a well-trained sleep technician according to the American Academy of Sleep Medicine standard criteria. ⁴⁴ Apnea was defined as a ≥90% drop in airflow from baseline for ≥10 s with ongoing respiratory efforts. Hypopnea was defined as airflow reduced ≥30% from baseline for ≥10 s with either ≥3% oxygen desaturation or EEG arousal. The apnea-hypopnea index (AHI) was used to assess the presence of OSA (AHI ≥5). Limb movements were scored according to standard criteria, in which individual limb movements with a clear EMG amplitude of ≥8 µV above the resting EMG were scored if the duration was between 0.5–10 s. PLMs were defined as four or more consecutive limb movements with individual movements separated by an interval of 5–90 s. The PLM index was the total number of PLM per h of sleep, and the PLM index ≥15 was the threshold for PLM during sleep. Standard metrics were generated, including total sleep time (TST), sleep efficiency, sleep onset latency, wake after sleep onset, NREM and REM stages, PLM index, and AHI.

NIRS protocol

A custom-designed wireless NIRS device was used to continuously monitor cortical water content during sleep (Figure 1(b)). The device was designed to operate with a sampling rate at 8 Hz for 10 hours while wirelessly transmitting data to a receiver. ⁴⁵ To measure oxyhemoglobin, deoxyhemoglobin, and water content, we included three laser wavelengths: 780, 850, and 925 nm. Although the 925 nm laser is less absorbed by water than the 980 nm laser used in a previous study, ³⁰ it is sufficiently sensitive to water compared to the 780 and 850 nm lasers used to measure hemoglobin. ⁴⁶ This provides sufficient signal quality for both water and hemoglobin measurements and allows the device to operate at low power (<5 mW), making it safe for long-term use. In the NIRS device, we set two detectors and one laser source on each side (right and left) of the forehead. The two detectors were positioned at geometrically symmetric locations, each 3 cm from the corresponding laser source. This configuration allowed incident photons to penetrate 1.5 cm into the brain tissue, enabling the measurement of CSF and cortical water, as well as hemodynamic changes (Figure 1(c)).^30,47 Each detector serves as a backup for the other on the same side of the forehead in case of malfunction or poor signal quality.

The content of brain water, oxyhemoglobin and deoxyhemoglobin were measured in an arbitrary unit (A.U.) using the Modified Beer-Lambert Law: the extinction coefficients and differential partial length factors of each wavelength were retrieved from previous works. ⁴⁸ Water and hemodynamic contents estimated from the left and right forehead were averaged to adjust possible postural influence. ⁴⁹

The analysis was implemented with the following tools to discriminate the signal of interests, remove physiologic noises and artifacts, and secure the signal quality. Physiologic noises caused by heartbeat, respiration, neuronal activity, and vasomotion were removed using a low-frequency filter with a 0.01-Hz cut-off frequency. ⁵⁰ A gyroscope was used to detect motion, and motion artifacts were removed using spline interpolation. ⁵¹ Signal-to-noise ratio (SNR) was measured while participants were kept awake before the start of the experiment. If a channel had a SNR below 30 dB, it was excluded in the analysis and the back-up channel in the corresponding side was included in the analysis.

The measured water signal included contributions from plasma fluid water that correlate with CBV, leading to crosstalk between the water signal and the hemoglobin signal, which serves as a representative of CBV. To address this issue, we employed a static linear minimum mean square estimator (LMMSE). ³¹ This estimator is designed to filter out the portion of the water signal associated with the total hemoglobin signal, thereby reducing crosstalk and ensuring that the measured signal accurately reflects brain water levels without the confounding effects of CBV. This is crucial for documenting the changes in CSF and ISF content, independent of CBV variations, as indicators of glymphatic activation. The method is applied as follows:

$${\Delta H_{2}O}_{\mathit{filtered}}\left(t\right)={\Delta H_{2}O}_{\mathit{meas}}\left(t\right)-{\Delta H_{2}O}_{\mathit{plasma}}\left(t\right)={\Delta H_{2}O}_{\mathit{meas}}\left(t\right)-\mathrm{\alpha }\Delta C_{\mathit{HbT}}\left(t\right)$$ (1)

where ${\Delta H_{2}O}_{\mathit{meas}}\left(t\right)$ represents the change in the water content signal extracted using the Modified Beer-Lambert Law, ${\Delta H_{2}O}_{\mathit{plasma}}\left(t\right)$ is the fraction of plasma fluid within the total change, and ${\Delta H_{2}O}_{\mathit{filtered}}\left(t\right)$ denotes the remaining change in water content after excluding the influence of plasma fluid. The change in plasma fluid ${\Delta H_{2}O}_{\mathit{plasma}}\left(t\right)$ is proportional to the change in hemoglobin concentration $\Delta C_{\mathit{HbT}}\left(t\right)$ , with $\mathrm{\alpha }$ representing this proportion. Since ${\Delta H_{2}O}_{\mathit{filtered}}\left(t\right)$ represents the water signal uncorrelated with plasma fluid,

$$\mathrm{\alpha }=\frac{{\Delta H_{2}O}_{\mathit{meas}}\left(t\right)·\Delta C_{\mathit{HbT}}\left(t\right)}{\Delta C_{\mathit{HbT}}\left(t\right)·\Delta C_{\mathit{HbT}}\left(t\right)}$$ (2)

To isolate changes in water content influenced by the glymphatic system within the tissue, the subsequent analysis used ${\Delta H_{2}O}_{\mathit{filtered}}\left(t\right)$ obtained from equations (1) and (2). To validate the performance of this algorithm in reducing crosstalk caused by plasma fluid, we conducted a simulation study. By modeling potential plasma fluid changes in the measurement signal and applying our algorithm, we confirmed that this estimator effectively reduces the influence of plasma fluid (see Supplementary Figure 1 and 2).

Statistical analysis

Transition analysis between different sleep stages

Forty-one participants (men = 24, mean age = 26.9 ± 7.7) with appropriate data were included in this analysis. Changes in water content were monitored during transitions between one source and one destination state to investigate the relationship between brain activity and water content (Figure 2(a)). ⁵⁰ The transition analysis was performed on four state transitions (WAKE →NREM, NREM→WAKE, REM→NREM, and NREM→REM).

Figure 2.

Transition analysis from the NIRS measurement. (a) Relationship between sleep-wake state and brain water content across the night. The water content increase during NREM sleep, while it decreases during REM sleep and WAKE period and (b) change in the brain water content during transitions between source and destination stage. The dotted box (from panel (a)) captures an example of REM-to-NREM sleep transition. The water content was block-averaged in each 5-minute segment. ${\mathit{W}}_{\mathit{S}}$ indicates the averaged water content of the source stage, and ${\mathit{W}}_{\mathit{D},\mathit{X}}$ illustrates the average water content of the destination stage’s X^th segment.

In each transition analysis, we used the final five-minute segment of the preceding source stage as a reference to examine changes in water content (Figure 2(b)). The destination stage was maintained until there was no disturbance from other sleep stages for at least two minutes, consistent with the procedures outlined in previous studies.^52–55 To ensure data reliability and stage stability, we excluded any 5-minute bin in which the SNR dropped below 30 dB at any time point, typically due to transient disruptions such as motion artifacts or changes in forehead contact. Transitions from WAKE→NREM, NREM→WAKE, and REM→NREM were included only when the destination stage was continuously maintained for at least 60 minutes. For NREM→REM transitions, where REM duration is typically shorter, segments with at least 30 minutes of the destination stage were included in the analysis. When multiple transitions of the same type occurred within a subject, water content values were averaged.

The extracted water fraction data were segmented into 5-minute epochs and block-averaged as follows:

$$W_S,\mathrm{ }{W}_{D,1},\mathrm{ }{W}_{D,2},\dots ,\mathrm{ }{W}_{D,X}$$ (3)

Changes in water content at each time point (X) were expressed as $\Delta H_{2}O\left(X\right)$ , representing the difference between the baseline ( $W_S$ ) and the X^th 5-minute segment in the destination stage ( $W_{D,X}$ ).

To assess these changes, we applied linear mixed-effects models with Time as a fixed effect and Subject as random intercept. In the primary analysis, we used a 5-minute reference bin for the source stage, based on initial data availability and pilot testing. To evaluate the robustness of this choice, we conducted a sensitivity analysis for the WAKE→NREM transition, testing 5-, 10-, 15-, and 20-minute reference bins (see Supplementary Figure 3). While longer reference bins reduced the number of participants included, the overall trends in water content changes remained consistent, confirming the validity of the 5-minute bin for the primary analysis.

Effect sizes were calculated using Cohen’s d, where d = 0.2 indicates a small effect, d = 0.5 a medium effect, and d = 0.8 a large effect. ⁵⁶ All statistical analyses were performed using SPSS version 22 (IBM, Armonk, NY, USA) and MATLAB (MathWorks, Natick, MA, USA).

Comparison of fluid dynamics between the first and last NREM stages

Glymphatic activity correlates with EEG slow wave activity (SWA).^6,35,36 Since the first NREM sleep period contains more SWA than the last NREM, we anticipated and compared the differences in brain water content changes during transitions to NREM sleep between the first and last NREM period. To measure these changes, we used the water content in the first 5-minute segment ( $W_{D,1}$ ) of the destination stage as a baseline for each transition, rather than the water content in the last 5-minute of the source stage ( $W_S$ ). This approach minimizes any bias caused by the heterogeneity of the source states, such as differences between the wake state preceding the first NREM period and the REM sleep preceding the last NREM period. Furthermore, the 5-minute interval after transitioning into NREM sleep is too short to result in significant changes in brain water content, making it a suitable reference for each transition analysis. A paired t-test was performed to compare the changes in water content:

$$\Delta H_{2}O\left(X\right)=W_{D,X}-W_{D,1}$$ (4)

for each X^th segment of the destination phase.

All statistical analyses were performed using SPSS software version 22 (IBM, Armonk, NY, USA) and MATLAB (MathWorks, Natick, MA, USA).

Results

Study population

Table 1 shows the participants’ demographic and sleep characteristics in addition to PSG measures. The mean TST was 6.4 h (range, 3.2–7.7 h), with a mean sleep efficiency of 79.4% (range, 43.0–96.7%). Although the TST was shorter than the recommended sleep duration of ≥7 h, ⁵⁷ it was similar to the TST of the general Korean population.^58,59 The low sleep efficiency and N3 sleep (%) observed in our study can be attributed to the first-night effect and racial differences. In the Korean adult population, the TST consists of <5% slow wave sleep. ⁵⁹

Table 1.

Demographic and sleep characteristics of the participants.

Variables	All participants (n = 41)
Demographics and sleep characteristics
Age (years)	26.9 ± 7.7
Male, n (%)	24 (58.5)
Body mass index (kg/m2)	22.0 ± 2.3
Insomnia Severity Index	3.6 ± 3.6
Morningness-Eveningness Questionnaire	50.6 ± 8.1
Epworth Sleepiness Scale	5.5 ± 3.5
Pittsburgh Sleep Quality Index	3.6 ± 1.6
Polysomnography measures
Total sleep time (h)	6.4 ± 1.1
Sleep efficiency (%)	79.4 ± 12.3
Sleep onset latency (min)	19.8 ± 24.0
Wake after sleep onset (min)	74.4 ± 50.1
N1 (%)	17.3 ± 8.1
N2 (%)	60.3 ± 7.8
N3 (%)	2.4 ± 3.5
REM (%)	20.1 ± 6.7
N1 (min)	63.1 ± 25.1
N2 (min)	232.1 ± 53.1
N3 (min)	9.0 ± 13.6
REM (min)	78.0 ± 33.8
Apnea-hypopnea index (events/h)	1.2 ± 1.7
Oxygen desaturation index (events/h)	1.2 ± 1.3
Arousal index (events/h)	21.6 ± 10.2
Minimum oxygen saturation in sleep	92.2 ± 2.7

Data are presented as means ± standard deviations or numbers and percentages.

Transition analysis

Figure 3 illustrates the estimated changes in brain water content (ΔH₂O) over time for each sleep-wake state transition, based on linear mixed-effects modeling. Brain water content was estimated for each 5-minute interval relative to the baseline (time 0) of the destination stage. Table 2 summarizes the estimated change, along with their statistical significance and effect sizes. Due to incomplete maintenance of the destination stage or poor SNR during certain epochs, the final number of transitions included in the analysis was 28 for WAKE→NREM, 11 for NREM→WAKE, 34 for REM→NREM, and 17 for NREM→REM.

Figure 3.

Estimated changes in brain water content during sleep-wake state transitions based on linear mixed-effects modeling. Changes in brain water content relative to baseline (time 0) were estimated at 5-minute intervals during the destination stage using a linear mixed-effects model. Error bars indicate 95% confidence intervals. Significant differences from baseline are marked as *p < 0.05 and **p < 0.01. Transitions include WAKE→NREM (n = 28), NREM→WAKE (n = 11), REM→NREM (n = 34), and NREM→REM (n = 17).

Table 2.

Changes in the brain water contents during the state transitions.

Tdestination (min)	WAKE → NREM (n = 28)			NREM → WAKE (n = 11)
	ΔH2O (A.U.) [95% CI]	p-value	Cohen d	ΔH2O (A.U.) [95% CI]	p-value	Cohen d
0–5	−0.22 [−0.45–0.02]	ns	−0.26	−0.35 [−0.67–−0.02]	p < 0.05	−0.47
5–10	−0.16 [−0.40–0.07]	ns	−0.19	−0.42 [−0.75–−0.10]	p < 0.05	−0.57
10–15	−0.05 [−0.28–0.19]	ns	−0.06	−0.58 [−0.91–−0.26]	p < 0.001	−0.78
15–20	0.12 [−0.12–0.35]	ns	0.14	−0.57 [−0.90–−0.25]	p < 0.001	−0.77
20–25	0.27 [0.03–0.50]	p < 0.05	0.32	−0.58 [−0.90–−0.25]	p < 0.001	−0.78
25–30	0.42 [0.18–0.65]	p < 0.001	0.49	−0.56 [−0.88–−0.23]	p < 0.001	−0.75
30–35	0.47 [0.23–0.70]	p < 0.001	0.55	−0.61 [−0.94–−0.29]	p < 0.001	−0.83
35–40	0.48 [0.24–0.71]	p < 0.001	0.57	−0.64 [−0.97–−0.32]	p < 0.001	−0.87
40–45	0.52 [0.29–0.76]	p < 0.001	0.62	−0.75 [−1.08–−0.43]	p < 0.001	−1.01
45–50	0.50 [0.26–0.73]	p < 0.001	0.59	−0.98 [−1.30–−0.65]	p < 0.001	−1.31
50–55	0.45 [0.21–0.68]	p < 0.001	0.53	−0.94 [−1.27–−0.62]	p < 0.001	−1.27
55–60	0.34 [0.11–0.58]	p < 0.001	0.41	−0.93 [−1.25–−0.60]	p < 0.001	−1.25
	REM → NREM (n = 34)			NREM → REM (n = 17)
Tdestination (min)	ΔH2O (A.U.) [95% CI]	p-value	Cohen d	ΔH2O (A.U.) [95% CI]	p-value	Cohen d
0–5	−0.03 [−0.19–0.14]	ns	−0.04	−0.35 [−0.66–−0.04]	p < 0.05	−0.47
5–10	−0.03 [−0.20–0.14]	ns	−0.05	−0.39 [−0.69–−0.08]	p < 0.05	−0.52
10–15	0.04 [−0.13–0.20]	ns	0.06	−0.37 [−0.68–−0.06]	p < 0.05	−0.50
15–20	0.11 [−0.06–0.28]	ns	0.18	−0.37 [−0.67–−0.06]	p < 0.05	−0.49
20–25	0.20 [0.03–0.36]	p < 0.05	0.31	−0.36 [−0.67–−0.05]	p < 0.05	−0.48
25–30	0.34 [0.17–0.50]	p < 0.001	0.53	−0.40 [−0.70–−0.09]	p < 0.05	−0.53
30–35	0.44 [0.28–0.61]	p < 0.001	0.70	•	•	•
35–40	0.53 [0.36–0.70]	p < 0.001	0.84	•	•	•
40–45	0.56 [0.39–0.73]	p < 0.001	0.89	•	•	•
45–50	0.60 [0.43–0.76]	p < 0.001	0.95	•	•	•
50–55	0.60 [0.43–0.76]	p < 0.001	0.95	•	•	•
55–60	0.64 [0.47–0.81]	p < 0.001	1.02	•	•	•

During the WAKE→NREM transition, brain water content increased significantly (ΔH₂O = 0.57 A.U., 95% CI: 0.33–0.81; p < 0.001; Cohen’s d = 0.77), peaking after approximately 40 minutes of NREM sleep. In the NREM→WAKE transition, a significant decrease was observed (ΔH₂O = −0.93 A.U., 95% CI: −1.25 to −0.60; p < 0.001; Cohen’s d = −1.25), with values stabilizing after about 50 minutes.​ The REM→NREM transition also showed a significant increase in brain water content (ΔH₂O = 0.62 A.U., 95% CI: 0.46–0.78; p < 0.001; Cohen’s d = 1.10), with levels continuing to rise beyond 60 minutes without a clear plateau. Conversely, the NREM→REM transition exhibited a modest but significant decrease (ΔH₂O = −0.40 A.U., 95% CI: −0.70 to −0.09; p < 0.05; Cohen’s d = −0.53), with a smaller magnitude of change compared to the other transitions.

Transitional changes in brain water content differed significantly between the first and last NREM periods (Figure 4). The increase in water content during the first NREM period was both faster and greater than during the last period. The differences in the Δ water between the first and last NREM periods became significant 15 minutes after the transition into NREM sleep and persisted for about 30 minutes, with the largest difference observed at the 25–30 minutes segment (0.70 A.U., d = 0.86; p < 0.01).

Figure 4.

Comparison of brain water changes between the first and last NREM sleep periods. Boxplots display ΔH₂O for the first and last NREM periods, with red lines indicating the median, blue boxes representing the interquartile range, and red circles marking outliers. The timing, the number of subjects, and the differences in ΔH₂O (in A.U. and Cohen’s d) between the first and the last NREM periods are noted for each segment below the plot. Brain water content increases significantly faster and greater during the transition to the first NREM sleep compared to the last NREM sleep (* for p < 0.05, ** for p < 0.01).

Discussion

This study proposes a non-invasive, continuous, and real-time method for monitoring brain water content using NIRS, emphasizing the potential of water-specific NIRS in evaluating the glymphatic system. During overnight PSG, a NIRS device utilizing water-sensitive 925-nm wavelength light was applied to the forehead to monitor changes in water content during sleep-wake transitions in the frontal area, a region potentially associated with prominent glymphatic activity.^8,21 We filtered out the contribution of plasma fluid from the measured water signal to focus on brain water changes driven by extracerebral CSF, as well as cortical ISF and ICF. When the glymphatic system is activated, both ISF and CSF volumes increase, while ICF volume decreases to facilitate CSF inflow.^6,19,26 Our analysis showed that brain water content is highly dependent on the sleep-wake cycle: it increased during NREM sleep and decreased during REM sleep and upon awakening. Additionally, the increase in brain water content during the first NREM cycle was significantly higher than during the last cycle. These findings are consistent with, but not conclusive evidence for, state-dependent glymphatic activity.^6,35–37 Assuming the measured brain water content accurately reflects glymphatic activity, this study is the first to demonstrate overnight changes in glymphatic activity across multiple cycles of NREM transitions and its reduction during REM sleep in humans.

Plausible mechanism of brain water dynamics in relation to glymphatic system

The glymphatic system is crucial for clearing metabolic waste from the brain, primarily through the movement of CSF and ISF.^5–8 The glymphatic activity is associated with expansion of ISF and extracerebral CSF,^6,19,26,35 which is the focus of water-specific NIRS measurements.

This system is particularly active during sleep, with NREM sleep being a key period for its enhanced function. ⁶ The glymphatic system is influenced not only by circadian rhythm ⁶⁰ but also by homeostatic drives. EEG SWA activity, a marker of homeostatic drive, closely correlates with glymphatic activity,^6,35,36 and is most pronounced during NREM sleep, especially in the early part of night, while being reduced during REM sleep.^59,61 Arterial pulsation is the main driver of CSF influx along the perivascular spaces,^62,63 and NREM sleep promotes brain-wide vasomotor pulsation, ⁶⁴ further explaining the enhanced glymphatic activity during NREM sleep.

The brain water dynamics observed in our study appear to be closely tied to these glymphatic activities. Brain water increases during the transition to NREM sleep and decreases during the transition to REM sleep and upon awakening (Figure 3). Furthermore, the amplitude and the rate of water increase are significantly enhanced during the first NREM period compared to the last (Figure 4), which is consistent with the greater presence of SWA in the earlier part of the night. During REM sleep, the reduction in SWA, coupled with widespread brain activation, is associated with decreased glymphatic activity, as observed in pigeons. ³⁷

However, we cannot confirm that the measured brain water changes directly indicate glymphatic activity, given the ongoing controversies surrounding the glymphatic system. These debates particularly focus on the mechanism of CSF flow in the interstitial space—whether it occurs through bulk flow or diffusion—and whether CSF-ISF exchanges take place.^11–14 Recent findings further complicate the interpretation of our results. One diffusion MRI study suggests that bulk flow, in addition to diffusion, serves as a transport mechanism, supporting the glymphatic model. ⁶⁵ Conversely, another study indicates that brain clearance by CSF inflow is less active during sleep and under anesthesia, and may not necessarily require the ISF volume expansion associated with CSF inflow into interstitial space. ⁶⁶ Instead, clearance could be driven by an increased rate of CSF inflow and drainage.^66,67

Our primary assumption was that brain tissue water changes are associated with glymphatic activity and that these changes can be documented using the water-specific NIRS system. However, delineating the specific mechanisms or resolving the controversies surrounding the glymphatic system was not the main goal of our study. Further research is needed to establish a direct correlation between brain water changes and other direct or indirect measures of glymphatic activity throughout the night, such as dynamic contrast MRI or analysis of overnight changes in CSF biomarkers for protein clearance in the brain

Potential utility of Water-Specific NIRS and other glymphatic system measures

The vital importance of the glymphatic system is underscored by the rapid increase in the aging population and the corresponding rise in neurodegenerative diseases such as Alzheimer and Parkinson disease.^7,8 Developing and applying non-invasive techniques to monitor glymphatic activity overnight across multiple NREM-REM sleep cycles is crucial for advancing our understanding of the role of glymphatic system in brain health.

Water-specific NIRS, which measures the absorption and scattering of near-infrared light by water in brain tissues, offers a promising approach to tracking brain water dynamics. ³⁰ These dynamics are believed to reflect glymphatic function, particularly during sleep.^6,19,26,35 Unlike more invasive methods, such as intrathecal injections used in dynamic contrast MRI,^21–23 NIRS provides a safer and more practical solution for continuous, real-time monitoring in both research and clinical settings. This could significantly enhance our ability to study the glymphatic system’s significance, mechanisms, determinants and modifiers, as well as to monitor therapeutic effects.

While NIRS provides a practical and non-invasive way to track brain water dynamics, it is important to emphasize that it measures bulk water content in cortical tissue without differentiating among individual compartments such as ISF, ICF, CSF, or plasma. Although we applied a plasma-filtering algorithm (LMMSE) to reduce CBV influence, the inference that brain water changes reflect glymphatic activity remains indirect. These signals may arise from overlapping processes including intracellular shifts, vascular factors, or general CSF circulation, rather than exclusively from perivascular flow. Our primary findings demonstrating state-dependent changes in brain water content were based on this CBV-corrected analysis. Importantly, a complementary analysis using uncorrected (raw) data revealed similar transition-dependent patterns (Supplementary Figure 4), lending support to the robustness of these observations. Nonetheless, given the indirect nature of the signal and unresolved contributions from other compartments, NIRS-derived brain water content should be interpreted cautiously and regarded as a proxy for broader brain fluid dynamics rather than a specific marker of glymphatic function.

Previous studies have shown that the glymphatic activity is associated with brain water change, as measured by diffusion MRI and recently developed electrical impedance spectroscopy analysis.^26,65 During sleep, water diffusivity, especially in extracellular compartment, increases during sleep, along with CSF volume expansion without changing total brain volume, and is negatively correlated with sleep fragmentation. ²⁶ Another study applied electrical impedance spectroscopy to measure brain parenchymal electrical resistance, suggesting that the electrical resistance decreases during sleep due to increase of extracellular water associated with glymphatic activity. ⁶⁸ This study found that brain parenchymal resistance decreased during overnight sleep, correlating with glymphatic function measured by dynamic contrast MRI and EEG spectral power. These findings collectively suggest that brain water content increases during sleep. Therefore, we suggest that glymphatic activity may partially account for this pattern but acknowledge the limitations in directly attributing bulk water changes to perivascular glymphatic flow.

While NIRS holds great potential, it is important to acknowledge its limitations. Unlike MRI, which offers high spatial resolution and broad coverage, NIRS has a limited field of view in both spatial extent and depth, lacks structural information, and is susceptible to motion artifacts. Additionally, interpreting NIRS data can be complicated by factors such as CBV, which may influence the measured water signal. To address this, we used the static LMMSE in this study for filtering out CBV contributions. ³¹ However, this analysis assumes a stable ratio of plasma fluid to total hemoglobin concentration. Given that hematocrit levels can vary during sleep, future studies might consider using adaptive filtering suitable for non-stationary signals to more accurately eliminate the impact of plasma fluid. ⁶⁹

A key limitation of prolonged NIRS monitoring is the potential for signal drift, which can arise from various factors, including optode repositioning, temperature fluctuations, or gradual changes in systemic hemodynamics. ⁵⁰ To mitigate these effects, we implemented several preventative measures, such as pre-measurement stabilization, filtering, signal quality checks, optode fixation, temperature control, and motion artifact detection. ²⁷ Despite these efforts, the possibility of signal drift over the course of a six-hour recording cannot be fully excluded. While transition-based analyses are less susceptible to signal drift due to their short temporal intervals and cyclical distribution throughout the night, comparisons between early and late NREM periods involve a longer monitoring duration and may be more vulnerable to gradual shifts in the signal. Given this potential source of bias, future studies should consider additional correction methods, such as short-separation channel regression to remove extracerebral contributions and drift correction algorithms to improve long-term signal stability.

Recent advancements underscore the feasibility of conducting all-night MRI studies to capture the full spectrum of sleep stages with minimal sleep deprivation. ⁷⁰ This methodology can provide detailed data on CSF dynamics associated with glymphatic activity across different sleep stages, including NREM and REM cycle.^35,70 However, even with such advancements, the constraints of MRI—such as accessibility, cost, and the challenging sleep environment within the scanner—still present limitations. NIRS, despite its own challenges, may complement MRI by offering a more naturalistic setting for monitoring sleep-related brain activity.

Study strengths and limitations

This study has several notable strengths. First, the use of a customized wireless NIRS device enabled continuous, non-disruptive monitoring of brain water content during natural overnight sleep. Second, the integration of simultaneous PSG recordings allowed for precise alignment of brain water dynamics with specific sleep stages and transitions. This multi-modal approach revealed consistent patterns of increased water content during NREM sleep and decreases during REM sleep and wakefulness. Third, a key methodological strength lies in the application of the LMMSE to minimize the confounding influence of CBV-related plasma water signals. Our primary findings—demonstrating state-dependent changes in brain water content across the sleep-wake cycle—were derived using this correction. We note that the consistency of findings across both corrected and uncorrected datasets (see Supplementary Figure 4) strengthens confidence in the robustness of our results, particularly in relation to potential CBV-related confounds.

Our study had several limitations. First, we did not perform a priori statistical power analysis because the effect size—specifically the ‘expected change in brain water during state transitions measured by the current NIRS system'—is largely unknown.^71,72 However, our findings could serve as a valuable guide for future studies aiming to measure and track changes in brain water content during state transitions. Second, our analysis was challenged by the varying durations for which participants maintained each sleep stage, resulting in an unbalanced dataset with missing observations. To address this limitation, we implemented linear mixed-effects models with Time as a fixed effect and Subject as a random intercept. This approach allowed us to include all collected measurements in the statistical model while accounting for within-subject correlations and missing data points. Nevertheless, the heterogeneity in duration for which participants maintained each sleep stage resulted in unbalanced analytical samples across transitions, potentially affecting the precision of our estimates. Future studies with larger and more balanced datasets would enable more comprehensive modeling approaches to further refine our understanding of brain water dynamics during sleep transitions. Third, the first-night effect was not controlled for, which could have influenced sleep architectures and quality, potentially impacting our results. Fourth, individual optical coefficients, such as scattering and absorption coefficients, were not used; instead, we relied on previously published statistical means of human optical properties. This lack of personalized optical properties may have introduced measurement errors and increased vulnerability to crosstalk artifacts. Fifth, intermittent head and body movement can degrade signal integrity and make the data inappropriate for analysis. This study used a dataset unaffected by motion artifacts, which limits our ability to investigate the dependency of water content on motion. Sixth, other factors related to glymphatic flow, such as the dynamics of CSF production and absorption, secretion of antidiuretic hormone during sleep, vasomotion, and respiration, may have influenced our findings. Finally, we did not perform phantom validation to empirically confirm the accuracy of our water measurements. Constructing a dynamic phantom that accurately replicates physiological changes in water concentration, however, presents considerable technical challenges. Because NIRS measures light scattered and transmitted through multiple layers of tissue, recreating these scattering properties and varying water content in a controlled setup is exceedingly complex. High absorption of water-specific wavelengths (e.g., 925 nm or 980 nm) can overshadow subtle fluctuations in water concentration, making it difficult to detect small changes reliably. Designing a phantom that simulates these time-varying conditions—including different tissue layers and blood perfusion effects—would require advanced, multi-layered constructs and precise control of water levels, complicating both construction and calibration. We acknowledge this as a limitation of our study and plan to explore suitable phantom-based approaches in future work.

Conclusion

This study underscores the potential of using water-specific NIRS as a non-invasive, continuous, and real-time tool for monitoring brain water content across sleep-wake cycles. While the observed state-dependent changes in brain water content are consistent with previously reported glymphatic activity patterns, the current methodology does not directly measure glymphatic flow or differentiate among specific fluid compartments such as ISF, ICF, or CSF. Our findings indicate that brain water content increases during NREM sleep and decreases during REM sleep and wakefulness, which may reflect, but do not confirm, underlying glymphatic mechanisms. Further studies are needed to validate these findings by simultaneously measuring brain water content and glymphatic activity using other modalities, such as advanced MRI techniques or biomarker analysis.

Establishing reliable methods for measuring and monitoring glymphatic activity is fundamental to understanding the role of brain clearance mechanisms in the development of neurodegenerative diseases. Such methods could also help identify modifiable risk factors, develop prevention or intervention strategies, and monitor therapeutic responses and disease progression.