Periodicity of cerebral flow velocity during sleep and its association with white-matter hyperintensity volume

Impaired sleep-related activation of the cerebral waste-clearance system might be related with the brain aging process. We hypothesized that cerebral blood-flow pattern changes during sleep might reflect the activation of the cerebral waste-clearance system and investigated its association with the cerebral white-matter hyperintensity (WMH) volume. Fifty healthy volunteers were prospectively recruited. In addition to the baseline transcranial Doppler parameters, the mean flow velocity (MFV) of the middle cerebral artery was monitored during waking and short-term non-REM sleep. Spectral density analysis was performed to analyze the periodic MFV variation patterns. For the aged subgroup (>50 years, n = 25), the WMH volumes in the total, subcortical, and periventricular regions were measured. The MFV periodic pattern during sleep was substantially augmented over that in the waking status. Spectral density analysis of MFV showed a noticeable peak in the very-low–frequency (VLF) band during sleep status (sleep/waking ratio 2.87 ± 2.71, P < 0.001). In linear regression analysis in the aged subgroup, the sleep/waking ratio of the VLF peak was inversely associated with total (P = 0.013) and subcortical (P = 0.020) WMH volumes. Sleep-related amplification of the cerebral flow-velocity periodicity might reflect the activation of cerebral waste clearance system during sleep, and be related to the pathogenesis of cerebral WMH.

and the slow vasomotor oscillations in very-low and low frequencies (VLF and LF) which have distinct spatiotemporal patterns 10 . Taken together, there might be a sleep-associated cerebral hemodynamic regulation which involves in the facilitation of those mechanisms during sleep.
In this study, we hypothesized that changes in the cerebral blood-flow parameters in the early sleep phase might reflect the activation of the waste-clearance system during sleep and may therefore contribute to the pathogenesis of cerebral WMH. In this regard, we investigated the changes in blood-flow parameters during short-term non-rapid eye movement (non-REM) sleep in young and aged populations with no reported neurological diseases using transcranial Doppler (TCD) monitoring, and its association with WMH volume on brain MRI.

Results
Fifty participants, including 37 males (74.0%) and 13 females (26.0%), with a mean age of 47.3 ± 12.6 years (range, 24-69 years) were included. The average duration of sleep during the three consecutive nights prior to the evaluation was 6.4 ± 1.2 h (range, 4.0-10.5 h). In the sleep monitoring, the mean duration of waking status was 299.4 ± 95.6 s (range, 134-729 s), and the mean duration of non-REM sleep status was 877.1 ± 321.3 s (range, 603-1683 s). Compared to the group aged ≤50 years, the group aged >50 years showed a higher frequency of hypertension and diabetes mellitus, a higher risk of obstructive sleep apnea (OSA), lower mean flow velocity (MFV), higher Epworth sleepiness scale (ESS) scores, and higher pulsatility index (PI) values during both waking and sleep statuses ( Table 1).
Compared to the findings in the waking status, the MFV modestly increased and PI decreased during the sleep status in both age groups (Fig. 1A,B). However, a substantial increment in MFV variation was observed in the sleep status (Fig. 1C). When graphing the MFV change over time, a periodic pattern in MFV variation with a period of approximately 50 s appeared within a few minutes after the onset of sleep in most patients (Fig. 2). When a spectral density analysis was applied to the MFV-time data, a peak in the VLF band during the sleep status appeared in the majority of subjects, which was not evident during the waking status (Fig. 3). Accordingly, in both age subgroups, the absolute power of this peak in the VLF band and the relative power of the VLF band significantly increased during the sleep status compared to the waking status ( Fig. 1D-F and Table 1). The mean sleep/waking ratio of the VLF peak was 2.9 ± 2.7, indicating that the periodic component of the MFV relevant to the VLF peak frequency increased by approximately three times during non-REM sleep.
Correlation analyses revealed that the sleep/waking ratio of the VLF peak was highly correlated with the sleep/waking ratio of MFV variation (r = 0.724, P < 0.001) and showed inverse correlation with the average sleep duration (r = −0.318, P = 0.041), whereas other variables such as age, blood pressure parameters, TCD parameters, and sleep monitoring parameters did not show any significant correlation (Supplemental Table S1). Among categorical parameters, none of the sex, vascular risk factors, medications, ESS score of >10, or a high risk of OSA were significantly associated with higher sleep/waking ratio of MFV variation (Supplemental Table S2). The sleep/ waking ratios of MFV, PI, MFV variation, relative power of each frequency band, and the relative power of the peak in the VLF band were comparable between the two age groups (Table 1).
In the WMH volume analysis in the subgroup aged >50 years, the mean total WMH volume was 2.8 ± 2.3 mL, subcortical WMH volume was 1.3 ± 1.1 mL, and periventricular WMH volume was 1.6 ± 1.7 mL. Correlation analyses revealed that total WMH volume was associated with increasing age (r = 0.453, P = 0.026) and lower sleep/waking ratio of MFV variation (r = −0.434, P = 0.034); the subcortical WMH volume was associated with lower sleep/waking ratio of MFV variation (r = −0.499, P = 0.013) and the lower sleep/waking ratio of the VLF peak (r = −0.406, P = 0.049); and the periventricular WMH volume was associated with increasing age (r = 0.463, P = 0.023) ( Table 2). Among categorical parameters, none of the sex, vascular risk factors, medications, or a high risk of OSA were significantly associated with higher values for the three WMH volume categories (Table 3). Due to high co-linearity, the sleep/waking ratios of MFV variation were discarded and only the sleep/waking ratios of the VLF peak were included in the multivariate analysis.

Discussion
In the present study, we observed a marked amplification of the periodic pattern of middle cerebral artery (MCA) MFV variation with a VLF occurring shortly after the onset of sleep, which resulted in a high peak power density in the VLF band in spectral density analysis during non-REM sleep. Notably, the sleep/waking ratio of the VLF peak was associated with the total and subcortical WMH volumes, after adjusting age and sex.
Numerous evidences have indicated that sleep disturbances such as decreased sleep quality 11 , short sleep duration 12 , and sleep-disordered breathing 13 are highly associated with cognitive impairment in aged population. Increased concentration of β-amyloid in the brain after an inadequate sleep has been recognized 8 , and it is proposed that sleep deprivation increases the production of β-amyloid 14 . Association between sleep apnea and increased WMH volume has been reported and its pathophysiologic basis was presumed to be chronic inflammation and activation of stress hormones in the brain 15 . Regarding OSA, a negative correlation between slow-wave EEG activities in polysomnography and CSF β-amyloid levels in the following morning was found in healthy patients but not in OSA patients, indicating that OSA might interfere with the clearance of cerebral metabolites 16 . www.nature.com/scientificreports www.nature.com/scientificreports/  Table 1. Clinical, sleep questionnaire, transcranial Doppler, and white matter hyperintensity profiles of the study population. Data are reported as a number (percentage), mean ± standard deviation, or median (interquartile range, IQR). ACEi/ARB, angiotensin converting enzyme inhibitor/angiotensin receptor blocker, ESS, Epworth sleepiness scale, OSA, obstructive sleep apnea, MFV, mean flow velocity, PI, pulsatility index, ULF, ultra-low frequency band, VLF, very low frequency band, LF, low frequency band, HF, high frequency band, and WMH white matter hyperintensity. † Average sleep duration of the last 3 nights prior to the evaluation, ‡ Determined per the criteria of the Berlin questionnaire, * P < 0.05, ** P < 0.01.
The pathophysiologic link between the sleep-related amplification of MFV periodicity and the volume of WMH might be postulated as follows: First, sleep-related amplification of MFV periodicity might reflect the cerebral microvacsular compliance, of which reduction is the main mechanism underlying the progression of WMH. Since sleep regulates the autonomic influence toward a reduction in the sympathetic tone, sleep would be able to increase arterial compliance and amplify the variabilities in cerebral blood flow. However, the capacity of amplifying the blood flow variability might be largely depend on the structural properties of cerebral microvessels 6,17 . Second, intracranial pressure (ICP) also oscillates synchronously with the MFV oscillation, known as the B-wave of ICP [18][19][20] . Therefore, amplification of the MFV oscillation might induce a periodic fluctuation in the hydrostatic pressure of the CSF 18 , facilitate the convection of CSF to ISF, and consequently enhance cerebral waste clearance during sleep [4][5][6]9 . Third, since the regulation of B-waves is independent from the pCO 2 level, blood pressure, respiration, central venous or airway pressure 21 , the MFV periodicity might represent a distinct motive of glymphatic system regulation and its sleep-related amplification might reflect the enhanced glymphatic function during sleep. A recent study using ultra-fast magnetic resonance encephalography revealed a vasomotor oscillation in VLF band that have distinct spatiotemporal patterns from those generated from cardiac or respiratory activities. In that study, the authors argued that this VLF oscillation might represent a distinct motive of regulating glyamphtic system function 10 . Taken together, sleep-induced amplification of MFV oscillation might be a distinctly regulated physiologic process that aims to enhance the efficacy of cerebral waste clearance and reflect the cerebral microvascular compliance.
The sleep/waking ratio of the VLF peak power was associated with the subcortical and total WMH volumes, but not with the periventricular WMH volume. This sight-specificity might be explained by that the periventricular white matter is mainly supplied by the penetrating arterioles from the MCA whereas the subcortical region is mainly supplied by the leptomeningeal branches 17,22 . Compared to the leptomeningeal branches, the MCA-penetrating arterioles have higher resistance and the CSF-ISF exchange via the paravascular space of the MCA-penetrating arterioles might be determined by intrinsic arteriolar compliance rather than dynamic changes in cerebral blood flow 17 . However, arteriolar compliance marker such as MCA PI was not significantly associated with the WMH volume. This might be due to the relatively low mean age of the current study population, as PI reflects the long-term process of vascular aging 23 .
The present study has some limitations to be addressed. First, due to that study was performed during daytime and the short duration of monitoring, the findings of this cannot be directly applied to the relationship between the cerebral WMH and the hemodynamic changes during nocturnal long-term sleep. In this study, the examiner had to hold the probe to maintain a fixed angle and insonation depth that best evaluates the MCA flow velocities throughout the monitoring. In this regard, it was impossible to perform long-term sleep monitoring. Advanced analyses such as correlating the VLF periodicity changes to EEG spectral plots or separately evaluating the VLF periodicity in different sleep stages might help more directly elucidate the mechanism and the physiologic role of VLF oscillation of cerebral blood flow during sleep. However, those evaluations were unavailable in this study setting, as they required long-term monitoring data. Second, some factors such as presence of OSA, use of ACEi/ ARB or calcium-channel blockers, or whether a participant is better in taking daytime sleep, might possibly affect the VLF periodicity changes during short-term sleep, but their effect were not fully evaluated in this study. Third, due to the small sample size, only a few variables were included in the multivariate analyses. Fourth, due to the cross-sectional design, this study did not establish a causative relationship between a diminished sleep/waking ratio of the VLF peak power and the progression of WMH. Fifth, this study did not include patients with higher WMH volumes. Future studies with larger study populations, including a healthy geriatric population as well as dementia patients, follow-up MRI analyses, and whole-night monitoring of MFV are warranted to clarify the clinical significance of sleep-related amplification of MFV oscillation in the pathomechanism underlying cerebral WMH.
In conclusion, sleep-related amplification of the cerebral flow oscillation, measured as the sleep/waking ratio of VLF peak power in the spectral analysis of the flow velocity, was associated with total and subcortical WMH volumes.

Materials and Methods
Study subjects. Fifty volunteers were prospectively recruited from the local community. Participants were cognitively normal, had no history of any neurological disease, had normal sleep quality defined as a Pittsburgh Sleep Quality Index of 4 or less, had no significant intracranial arterial stenosis evaluated by TCD or MR angiography nor a preexisting ischemic/hemorrhagic lesions in MRI 24,25 , and had no medications that regulate alertness, affect sleep structure, or alter the vascular tone. However, patients who were taking angiotensin converting enzyme inhibitors/angiotensin receptor blockers (ACEis/ARBs) or calcium-channel blockers to treat hypertension were included in this study. Participants were categorized according to their ages into those aged ≤50 years (young subgroup, n = 25) and those aged >50 years (aged subgroup, n = 25) 17,26 . This study was approved by the institutional review board of Seoul National University Hospital (SNUH). All methods were performed in accordance with the institutional review board of SNUH regulations and STROBE guidelines for observational studies. Written informed consent was obtained from all participants.
Acquisition of clinical data. Clinical profiles that included demographic data and information regarding body mass index (BMI, kg/m 2 ) and the presence of hypertension, diabetes mellitus, hyperlipidemia, and a smoking history in the past five years were obtained 17,26 . Systolic, diastolic, and mean blood pressure (SBP, DBP, and MBP, all mmHg), and heart rate (HR,/min) were obtained using an electronic manometer after more than 5 min of rest in the sitting position.
Sleep questionnaire. Subjects were encouraged to sleep regularly for more than three consecutive nights prior to the day of TCD monitoring. Before the evaluation, data for the following sleep questionnaires were collected: Berlin questionnaire (BQ, for screening a high risk of obstructive sleep apnea) 27 , ESS (a rating scale of 0 to 24 evaluating the likelihood of dozing or falling asleep during routine daily activities. ESS scores of >10 indicate significant sleepiness) 28 , and self-reported latency (min) and average sleep duration (h) in the latest three days.
Transcranial doppler evaluation. Intracranial arteries were sonographically evaluated using a 2-MHz pulsed-wave and range-gated TCD probe (ST3 Digital PMD 150; Spencer Technologies; Redmond, WA, USA) with a transmission power level of 100 mW/cm 2 , pulse repetition frequency of 8000 Hz, filter frequency of 200 Hz, gain of 4 decibels, and range of 200 decibels. The evaluations were initiated between 2-4 PM. Throughout the evaluation, the laboratory room was maintained in a light-off noiseless state. The TCD evaluation protocol was standardized for every patient and was conducted by a TCD sonographer with 15 years of experience.
Before the monitoring, the flow in the MCA, the anterior cerebral artery, and the internal carotid artery (ICA) was evaluated to exclude stenosis of the anterior cerebral arteries. Then, TCD parameters, including peak systolic, minimal diastolic, and mean flow velocities (PSV/MDV/MFV, cm/s) and the pulsatility index (PI) were obtained along the MCA with insonation depths of 64, 60, 56, 50, and 46 mm 25,29 . Sleep monitoring. For sleep monitoring, two TCD probes were positioned on both temporal windows using a marc 600 headframe (Spencer Technologies), and the positions of the probes were adjusted until a maximal signal with an insonation depth of 56 mm was obtained. The side of a higher flow velocity was selected for monitoring and PSV, MDV, and MFV were recorded every second throughout the monitoring 30 . To assess the sleep-waking state, electroencephalography (EEG) using six electrodes (F3, F4, C3, C4, O1, and O2) and www.nature.com/scientificreports www.nature.com/scientificreports/ electrooculogrpahy (EOG) were monitored during the examination 28,30 . The sleep onset was defined according to the American Academy of Sleep Medicine scoring manual 31 . When the patients re-entered to wake stage after the onset of sleep, the monitoring was terminated. The short duration of monitoring is due to the technical difficulty that the examiner had to hold the probe to maintain a fixed angle and insonation depth that best evaluates the MCA flow velocities throughout the study. In this regard, it was impossible to perform TCD monitoring during long-term sleep. As REM sleep stage is related with cerebral blood flow parameters profiles comparable to those of waking status 30,32 , only the initial non-REM sleep period from the sleep onset were included in the analyses.

Analysis of blood-flow parameters. For both sleep and waking statuses, the mean values of MFV and
PI and the variations in MFV were measured. To evaluate the periodic component of MFV variation, a fast Fourier transformation was applied to the MFV data for each status 29 . A spectrum density analysis was then performed to express the power ([cm/s] 2 ) as a function of frequency (Hz). The frequency domain was divided into four frequency bands: the ultra-low-frequency (ULF, <0.0003 Hz), very-low-frequency (VLF, 0.003-0.04 Hz), low-frequency (LF, 0.04-0.15 Hz), and high-frequency (HF, >0.15 Hz) 33,34 . The relative power of VLF, LF, and HF bands (%) was calculated by numerical integration of powers in the domain of each frequency band, divided by the numerical integration of powers in the total frequency domain. Since a long recording time (>12 h) is a prerequisite for reliable assessments of the ULF domain parameters, the ULF data was discarded 34 .
To evaluate sleep-related changes in these flow parameters, the sleep/waking ratios of the MFV, PI, MFV variations, and the relative power of each frequency band were calculated. Since a noticeable peak appeared at the VLF band during the sleep status, we also evaluated the sleep/waking ratio of the normalized power of the VLF peak. First, in the spectral density graph of the sleep status, the normalized absolute power of the VLF peak was calculated by dividing the absolute power of the VLF peak by the numerical integration of power throughout the total frequency domain. Second, in the spectral density graph of the waking status, the normalized power of the VLF peak frequency of the waking status was calculated in the same way. Third, the sleep/waking ratio of the VLF peak was calculated by the normalized power of the VLF peak in the sleep status divided by that in the waking status. www.nature.com/scientificreports www.nature.com/scientificreports/ Analysis of WMH volume. In the group of patients aged >50 years, MRI was performed using a 1.5-T unit (Ingenia; Philips, Best, Netherlands) according to protocols that included T1/T2-weighted imaging, the fluid-attenuated inversion recovery (FLAIR) and gradient echo (GRE) sequences, and time-of-flight magnetic resonance angiography (MRA). FLAIR sequences were obtained with the following parameters: slice thickness/ gap of 4.0/0.0 mm, 24-27 slices covering the entire brain, repetition time/echo time (TR/TE) = 9000-9900/97-163 ms, a field-of-view (FOV) = 240 × 240 mm, and matrix = 220 × 220. FLAIR and GRE images were reviewed to exclude preexisting ischemic or hemorrhagic lesions, and MRA was performed to exclude significant stenosis in cerebral arteries.
For volumetric analysis of WMH, FLAIR sequences were registered into an offline workstation. Using a semi-automated freeware NeuRoi (Nottingham university, Nottingham, UK) 17,26 , the WMH lesion was identified and classified as periventricular or deep subcortical and its boundaries were drawn and volume was calculated by a neurologist (WJL, with 7 years of experience and blinded to other data) according to previously described protocols of which the reproducibility was established 17,26 . Statistical analysis. For all statistical analyses, SPSS 21.0 (SPSS Inc., Chicago, IL, USA) was used. Data are reported as numbers (percentage), mean ± standard deviations, or as medians (interquartile range, IQR). Pearson     Table 4. Linear regression analyses for factors associated with white matter hyperintensity severity. a R 2 = 0.497 and P = 0.003, b R 2 = 0.391 and P = 0.017, and c R 2 = 0.369 and P = 0.024 for the linear regression equations. † The variables were log transformed to obtain a normal distribution. B, unstandardized coefficient and β, standardized coefficient. * P < 0.05, ** P < 0.01. www.nature.com/scientificreports www.nature.com/scientificreports/ coefficients or Spearman's test were used to measure correlations between continuous or ordinal variables and WMH volume. Paired t test was performed to compare the blood flow parameters between sleep and waking status. Student's t tests or Mann-Whitney U tests were used to compare the sleep/waking ratio of the VLF peak power and the WMH volume among dichotomized subgroups. Variables with P values < 0.15 in univariate analyses were included in a multivariate linear regression analysis using an enter method. Age and sex were adjusted in the multivariate regression analysis regardless of the significance of the association in univariate analyses. In the multivariate analysis, WMH volume parameters and parameters from the spectral density analysis were log-transformed to obtain a normal distribution. The variance inflation factor (VIF) was measured to verify the multi-collinearity between variables. A scatterplot of the standardized predicted values and the standardized residuals was made to check the assumption for linearity of the regression model. For all analyses, P values < 0.05 were considered to be statistically significant.

Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on request.