Preliminary investigations into human neurofluid transport using multiple novel non-contrast MRI methods

Abstract

We discuss two potential non-invasive MRI methods to study phenomena related to subarachnoid cerebrospinal fluid (CSF) motion and perivascular fluid transport, and their association with sleep and aging. We apply diffusion-based intravoxel incoherent motion (IVIM) imaging to evaluate pseudodiffusion coefficient, D*, or CSF movement across large spaces like the subarachnoid space (SAS). We also performed perfusion-based multi-echo, Hadamard encoded arterial spin labeling (ASL) to evaluate whole brain cortical cerebral blood flow (CBF) and trans-endothelial exchange (T_ex) of water from the vasculature into the perivascular space and parenchyma. Both methods were used in young adults (N = 9, 6 F, 23 ± 3 years old) in the setting of sleep and sleep deprivation. To study aging, 10 older adults (6 F, 67 ± 3 years old) were imaged after a night of normal sleep and compared with the young adults. D* in SAS was significantly (p < 0.05) reduced with sleep deprivation (0.016 ± 0.001 mm²/s) compared to normal sleep (0.018 ± 0.001 mm²/s) and marginally reduced with aging (0.017 ± 0.001 mm²/s, p = 0.029). Cortical CBF and T_ex were unchanged with sleep deprivation but significantly lower in older adults (37 ± 3 ml/100 g/min, 578 ± 61 ms) than in young adults (42 ± 2 ml/100 g/min, 696 ± 62 ms). IVIM was sensitive to sleep physiology and aging, and multi-echo, multi-delay ASL was sensitive to aging.

Introduction

The glymphatic system is a brain-wide network of perivascular pathways along which subarachnoid cerebrospinal fluid (CSF) and brain interstitial fluid exchange, supporting solute distribution and clearance.^{1 –3} Implicated in the clearance of amyloid-β,^{2 –4} tau,^5,6 and α-synuclein,^7,8 impairment of glymphatic function is proposed to contribute to the development of neurodegenerative disorders such as Alzheimer’s disease. Initial studies in mice characterized perivascular fluid exchange using dynamic 2-photon microscopy following intrathecal fluorescent CSF tracer injection, ² an approach with high spatial and temporal resolution but with a limited field of view. Brain-wide fluid exchange, organized around the macroscopic cerebral vasculature, was subsequently described in rodents using dynamic contrast-enhanced magnetic resonance imaging (MRI) following intrathecal gadolinium-based contrast agent (GBCA) administration. ¹

Studies by Ringstad, Eide, and colleagues have used a similar intrathecal approach to confirm key features of the glymphatic transport model initially described in rodents and the human brain. In reference participants being evaluated for neurosurgical intervention, these authors report that serial MRI scanning following intrathecal GBCA injection results in initial cisternal and subarachnoid space (SAS) enhancement, followed by a centripetal enhancement of the gray and white matter tissue. ⁹ Using serial T1 mapping over three days, Watts et al. showed that signal enhancement peaked in the CSF and gray matter (GM) at approximately 10 hours and in white matter (WM) at approximately 40 hours. ¹⁰ A recent study also confirmed sleep-active solute clearance from the human brain, comparing the effects of overnight sleep deprivation on the clearance of intrathecally injected GBCA. ¹¹ Since intrathecal injections are an off-label use for GBCA and challenging to implement in a research setting, several recent studies have sought to extract similar information from MRI following intravenous GBCA injection. In one study, heavily T2-weighted fluid-attenuated inversion recovery (FLAIR) showed GBCA transport from the vasculature into the CSF via the choroid plexus along perivascular spaces and perineural sheaths of cranial nerves. ¹² In a recent study, we further characterized CSF, GM, and WM enhancement that occurs between 0–6 hrs following IV GBCA administration. ¹³ While these contrast-based methods for assessing solute transport through the CSF and brain interstitium remain a gold standard, these approaches remain technically challenging and suffer from poor temporal resolution. Functional MRI studies revealed that coordinated ventricular CSF pulsations increase during sleep, perhaps reflecting increased glymphatic exchange. ¹⁴ These coordinated pulsations appear to arise from slow wave-associated low-frequency vasomotor fluctuations that are hypothesized to be one of the drivers of perivascular glymphatic convective transport of solutes.^15,16

Here we present two modified diffusion and perfusion MRI methods that reflect neurofluid transport and may be related to downstream glymphatic exchange: (1) Intravoxel incoherent motion (IVIM) MRI, a diffusion-based approach to measure CSF flow in the subarachnoid space and (2) multi-echo Hadamard encoded arterial spin labeling (ASL) to study water exchange from the blood compartment into the brain interstitium.

IVIM conceptually detects a diffusion regime associated with pseudorandom motion in tissue at lower b-values that corresponds to a water movement over long distances (reflected in a faster pseudodiffusion coefficient, D*), which was initially proposed to measure blood perfusion. ¹⁷ Such pseudorandom motion has been estimated using D* in the CSF spaces like the ventricles.^{18 –20} This motion contrasts the typical slow Gaussian diffusion coefficient, D, measured with diffusion imaging, corresponding to faster water movement over very small distances.

The subarachnoid space (SAS) has not yet been studied using IVIM. Its anatomy is not well documented in humans. However, one study by Benko et al. measured the trabeculae in the SAS and showed that overall, in 8 individuals, they occupied almost 25% of the volume, with the frontal and superior SAS containing upwards of 30% trabeculae by volume. These trabeculae are collagenous fibers that are cylindrical, tree-like, and fan-like in shape and create a maze within the SAS. In histological studies of the rodent SAS, Saboori et al. characterized the anatomy of this compartment and its associated trabeculae. They further developed a finite element model suggesting that CSF motion within the SAS followed Darcy’s law of permeability. It is similar to fluid flow through a porous medium and is pseudorandom in nature.^21,22 This represents therefore the fluid flow that IVIM aims to measure. In the brain gray and white matter, IVIM measurements are mainly influenced by blood perfusion, making it more challenging to characterize ISF-mobility as ISF-mobility needs to be discriminated from blood perfusion. In this work, we use IVIM-derived D* as a marker of CSF motion in the SAS.

We applied an advanced ASL sequence to calculate blood flow and ensure that the D* signals are not entirely driven by blood perfusion. This approach allows us to quantify the trans-vascular exchange of water, which may be related to the glymphatic transport of solutes. While this measure is influenced by blood-brain barrier permeability, it may also represent changes in glial water transport through aquaporin-4 (AQP4) channels. A study in Aqp4 knockout mice showed that trans-vascular exchange increased in the knockout mice by almost 30% compared to wild-type mice and decreased significantly in aged mice. ²³ Only a few studies have explored this advanced ASL approach in humans and in the context of sleep modulation.

In the present study, we used a within-participant paired sleep/sleep deprivation design to test the ability of IVIM within the SAS and multi-echo ASL to detect sleep-related changes in SAS and perivascular fluid transport, as well as an across-participant design to test the sensitivity of these techniques to detect aging-related changes in SAS and perivascular fluid transport.

Methods

Participant characteristics and study procedures

This study was approved by the Institutional Review Board of the University of Washington (UW) School of Medicine, and all participants provided written informed consent before any study procedures were conducted in accordance with the Declaration of Helsinki. Ten normal healthy older adults (6F/4M, age = 67 ± 3 years old) and fifteen normal healthy young adults (6F/7M, age = 23 ± 3 years old) were enrolled. Demographic information is detailed in Table 1. Cognitively normal older adults were identified and referred by the UW Alzheimer’s Disease Research Center (ADRC). For this study, they underwent a brief cognitive performance evaluation and a single MRI scan after a night of normal sleep. Cognitive testing included the General Practitioner’s Cognitive Assessment (GPCOG) ²⁴ and the Memory Impairment Screen (MIS). ²⁵ Participants were asked to perform their regular activities before coming in for their study visit.

Table 1.

Parameter values (Average ± Std. Deviation) for IVIM.

Parameter/region	Old	Young
		Normal sleep	Sleep deprivation
D* mm2/s
SAS	0.017 ± 0.001	0.018 ± 0.001	0.016 ± 0.001
CP	0.019 ± 0.001	0.019 ± 0.001	0.017 ± 0.001
GM	0.013 ± 0.001	0.014 ± 0.001	0.014 ± 0.001
WM	0.013 ± 0.001	0.014 ± 0.001	0.013 ± 0.001
CSF	0.019 ± 0.002	0.020 ± 0.002	0.019 ± 0.002
D mm2/s
SAS	0.0014 ± 0.0002	0.0013 ± 0.0002	0.0012 ± 0.0003
CP	0.0024 ± 0.0003	0.0019 ± 0.0006	0.0019 ± 0.0006
GM	0.0008 ± 0.0001	0.0008 ± 0.0004	0.0008 ± 0.0005
WM	0.0007 ± 0.0005	0.0007 ± 0.0003	0.0007 ± 0.0003
CSF	0.00025 ± 0.0001	0.00025 ± 0.0001	0.00025 ± 0.0001
f
SAS	0.17 ± 0.02	0.17 ± 0.02	0.17 ± 0.02
CP	0.19 ± 0.04	0.21 ± 0.02	0.21 ± 0.03
GM	0.10 ± 0.01	0.10 ± 0.01	0.10 ± 0.01
WM	0.09 ± 0.01	0.08 ± 0.01	0.09 ± 0.01
CSF	0.15 ± 0.01	0.19 ± 0.02	0.19 ± 0.03

IVIM: intravoxel incoherent motion; SAS: subarachnoid space; CP: choroid plexus; GM: gray matter; WM: white matter; CSF: cerebrospinal fluid.

Young, healthy adults were recruited using flyers in and around the institution. They underwent two scans, one after a night of normal sleep and one after 24 hours of sleep deprivation. They were also instructed to follow their regular and identical routines for both visits except for the sleep deprivation. They completed the GPCOG and MIS at each MRI visit. For the sleep deprivation visit, participants were asked to stay awake for 24 hours before their MRI scan. Starting the morning before, they were instructed to complete their everyday routine and not nap. From 8 pm onwards, at every hour, they called a designated phone line to let the study team know that they were awake. If they missed a check-in, their scan was canceled. The order of sleep and sleep deprivation visits were randomized. All scans were conducted between 0800–1000 hours for all participants. For the young adults, both scans were scheduled at the same time in the morning.

Imaging

All imaging was performed on a 3 T Philips Ingenia Elition scanner with a 32-channel head coil. Identical MRI scans were performed on all participants. An eye tracker (Eyelink 1000) was used to ensure the participant was awake. The technologists or study team spoke to the participants between scan sequences to ensure they did not sleep. The MRI protocol consisted of (1) 3D T1 acquisition with resolution = 1 × 1 × 1 mm³, matrix size 256 × 256 × 176, repetition time (TR)/echo time (TE) = 9.2/3.5 ms, (2) Intravoxel Incoherent Motion (IVIM) protocol using 11 b-values (10, 40, 80, 100, 150, 200, 300, 500, 700, 900, 1000 s/mm²) and 6 directions and one b = 0 s/mm², resolution = 1 × 1 ×5 mm³, slices = 30, TR/TE = 3000/62 ms. ¹⁷ To reduce motion-related distortion, a two-echo SPLICE acquisition was used. ²⁶ SPLICE, or Split acquisition of fast spin echo approach modifies the readout to collect two echoes, which are combined to reduce distortion and increase signal-to-noise ratio or SNR. (3) T2-prepared Hadamard encoded arterial spin labeling (ASL) with background suppression and resolution = 3.75 × 3.75 ×5 mm³, slices = 18, label duration = 3400 ms, effective post-labeling delays = 650, 1210 and 2083 ms, and echo time TE = 0, 40, 80, 160 ms.^27,28 A second reference, M0 scans at all four TE values, was acquired with identical parameters but no labeling or background suppression. All MRI data will be available upon request.

Analyses

The IVIM MRI was processed to calculate perfusion or fluid fraction, f, and pseudodiffusion coefficient, D*, to measure CSF mobility in the SAS. For this purpose, IVIM data were motion-corrected and aligned with the B0 image and skull-stripped using FSL v6.0. Due to SPLICE acquisition, no distortion correction was necessary. Next, the diffusion data was skull-stripped in FSL ²⁹ and fitted to a two-compartment model in MATLAB (v2021a)

$$S=S0\left[f{e}^{-bD*}+\left(1-f\right)e^{-bD}\right]$$

where S0 = signal intensity at b = 0 image. D, D*, and f are calculated using Bayesian model fitting based on Gustafsson et al. ³⁰ Figure 1 shows the IVIM images and the derived parameters (D, D*, and f ) for one participant over the entire imaging volume.

Figure 1.

(a) IVIM acquisition involves multi-b value acquisition in at least 3 directions. Our acquisition used 6 directions. 12 b-values (10, 20, 40, 80, 100, 150, 200, 300, 500, 700, 900, 1000 s/mm²) were used. The two exponentials corresponding to the 2 diffusion regimes are shown in blue and green. Blue corresponds to D*, with faster diffusion reflecting water movement over long distances, and green corresponds to D, with slow diffusion reflecting water movement over short distances. Parametric maps for D*, D, and f, tin the whole brain of one participant, are shown in b, c, and d, respectively.

The D, D*, and f were then registered to their T1 image in a 2 mm MNI space atlas. The b = 0 image was used for registration to T1 and finally to MNI space and the transformation matrix was then applied to the D, D*, and f parameter maps. Using FSL FAST, the T1 image was segmented into GM, WM, and CSF. GM and WM masks were obtained using a probability threshold of 80%. A dilated brain mask was used to eliminate the brain stem regions, and the resultant CSF mask was thresholded to include only >95% CSF. This stringent threshold was used to ensure no partial volume with cortical GM. Ventricular, MNI-space ROI was subtracted, and the extra-axial CSF was identified as the SAS. All ROI generation was automated (Supplementary Figure 1) but verified manually for each participant. If excess CSF from the ventricular horns in the frontal lobe were included in the ROI, it was manually removed voxel by voxel. GM ROI is shown in magenta and overlaid on the participants’ T1 image in MNI space. White matter ROI is shown in white. CP is shown in yellow, and SAS in green. Our main variables of interest were D* in the SAS using IVIM and T_ex in the cortical GM using ASL. Other ROIs were included for the sake of completeness. No other region-specific hypothesis was investigated in this work. The ventricular CSF, cortical GM, and subcortical WM were used as ROIs for comparison while being aware that the IVIM effect could be due to CSF and ISF mobility, perfusion, or a combination of both. Additionally, an ROI was manually drawn in the choroid plexus (CP), representing CSF production and an area showing both perfusion and CSF-related IVIM effects.

The Hadamard-encoded ASL data were rearranged to calculate perfusion-weighted difference maps at the post-labeling delays of 650, 1210, and 2083 ms for further analysis. The difference maps were registered to the M0. First, the 4-echo M0 data were fit to an exponential decay to estimate extravascular T2 (assuming that most signals were extravascular). The fitting was performed in MATLAB v2021a using the fminsearch function, a search algorithm proposed by Lagarias et al. ³¹ Based on the work of Ohene et al., we used a constrained fitting for a two-compartment (intravascular and extravascular) model to the difference image at a post-labeling delay of 1210 to obtain T2 of the intravascular compartment. ²³ Next, the multi-TI data was used to estimate cerebral blood flow (CBF) and arterial arrival time (δ_a) using FSL’s BASIL Toolbox by an estimate of the tissue transit time (d). To calculate the tissue transit times, we used the 2-compartment kinetic model proposed by Wang et al. ³² The difference in the two transit times, T_ex, was used to measure the trans-vascular water permeability. Since ASL measures perfusion and is typically unsuitable for white matter tissue, measurement was made only in cortical GM. Figure 2 shows the raw difference images at each post-labeling delay time and each echo time.

Figure 2.

The top row shows the M0 reference image in grayscale. The bottom 3 rows show ASL difference images for each of the post-labeling delay times 650, 1218, and 2089 ms. Each echo time is shown along the columns.

The same ROIs as for the IVIM analysis were used. All data is available upon request.

Statistical considerations

A two-way, paired t-test was used to compare the MRI measurements after a night of normal sleep and sleep deprivation in young adults. A t-test assuming homoscedastic distributions (Kolmogorov-Smirnov test for normality, p = 0.01) was used to compare MRI measurements between young and old adults. Associations with GPCOG and MIS were tested using a linear regression model of all MRI measures. Primary comparisons included only D* in the SAS and cortical T_ex between young and old adults and between young adults after normal sleep and a night of sleep deprivation. The two measurements are independent, and comparisons were corrected using a Bonferroni correction at p = 0.025, accounting for the investigation of aging and sleep. We expect a global decrease in CBF and T_ex with aging and sleep; there is yet no compelling evidence for region-based analysis for perfusion-related changes with sleep. Third, a small sample size would have limited meaningful observations in this cohort due to multiple comparisons.

Results

All 10 older adults completed cognitive assessment and MRI sessions. Of the thirteen young adults, nine (6F/3M) completed both visits for sleep and sleep deprivation. Of the 4 participants who did not complete the study, three could not comply with the sleep deprivation protocol, and one was removed from the study due to incidental findings affecting study interpretations. As a result, their sleep deprivation visit was canceled.

The GPCOG and MIS measures (averages ± standard deviation) in young adults after a night of normal sleep were 9 ± 1 and 8 ± 0, respectively. Their measures after a night of sleep deprivation were 8 ± 1 and 8 ± 0, respectively. In older adults, the corresponding scores were 9 ± 1 and 8 ± 0. Cognitive test scores were not significantly different between groups or conditions.

The results from IVIM studies are shown in Figure 3, with representative images provided for IVIM parameters D*, D, and f (Figure 3(a)) from one participant. Overall data for D*, D, and f following overnight sleep and overnight sleep deprivation are shown in Table 1 and graphically in Figure 3(b), (d) and (e). Using a paired t-test, the D* values in the SAS were significantly lower in young adults after sleep deprivation compared to the values after a night of normal sleep (Figure 3(c), p = 0.006). D* was marginally significantly lower in older compared to young adults (p = 0.029, not significant after Bonferroni correction at p = 0.025). No differences were observed in any other IVIM parameters, D (Figure 3(d)), f (Figure 3(e)), or region.

Figure 3.

(a) IVIM parameters, D*, D, and f in older adults (first column), in young adults with normal sleep (middle column), and the same young adults with sleep deprivation (third column) in the SAS. (b) D* values in all three groups: young adults after a night of normal sleep (blue), young adults after 24 hours of sleep deprivation (yellow), and older adults after normal sleep (red). (c) Only D* was significantly different (p = 0.006) after 24 hours of sleep deprivation compared to a night of normal sleep and (d, e) D, and f values in all three groups: young adults after a night of normal sleep (blue), young adults after 24 hours of sleep deprivation (yellow), and older adults after normal sleep (red). No difference was observed with aging. Supplementary Figure 1 shows the ROIs generated for this study.

The results from the multi-echo ASL studies are shown in Figure 4(a), with representative images provided for ASL parameters, T_ex, and CBF. Overall ASL measures for each group are provided in Table 2 and visualized in Figure 4(b) and (c). Figures 4(b) and (c) show a comparison in young adults after normal sleep and sleep deprivation and older adults. No significant difference was observed in these measurements. However, there is significantly reduced CBF and T_ex in older adults compared to young adults after normal sleep (p = 0.005 and p = 0.0001, respectively). The intravascular T2 was calculated to be 152.00 ± 6.91 ms), similar to prior studies. ³³ Furthermore, the intravascular fraction was 0.59 ± 0.02, 0.36 ± 0.02, and 0.24 ± 0.02 at post-labeling delays = 650, 1210 and 2083 ms respectively. The arterial transit times in young adults after normal sleep, after sleep deprivation, and in older adults after normal sleep were 647 ± 136 ms, 600 ± 97 ms, and 815 ± 112 ms, respectively. It significantly differed between older and young adults (p = 0.007) but not between the two sleep conditions in the young adults.

Figure 4.

(a) ASL parameters, Whole-brain cortical CBF, and T_ex in older adults (first column), in young adults with normal sleep (middle column), and the same young adults with sleep deprivation (third column). CBF was not significantly different in young adults after sleep deprivation (yellow) compared to those after a night of normal sleep (blue). Detailed paired comparison is shown in (b) CBF was significantly different in older adults (red) compared to young adults (p = 0.005) and (c) T_ex was also not significantly different in young adults after sleep deprivation (yellow) compared to those after a night of normal sleep (blue). Like CBF, T_ex was significantly shorter in older adults (red) than in young adults (p = 0.0001).

Table 2.

Parameter values (Average ± Std. Deviation) for ASL.

Region/parameter	Old	Young
Gray matter (GM)		Normal sleep	Sleep deprivation
CBF ml/100 g/min	37 ± 3	42 ± 2	42 ± 4
Tex ms	578 ± 61	695 ± 62	697 ± 63

ASL: arterial spin labeling; CBF: cerebral blood flow; T_ex: exchange time.

No significant associations were detected with GPCOG or MIS. However, a lower T_ex after sleep deprivation was marginally significantly associated with better GPCOG performance in young adults (r = −0.70, p = 0.04). No other association was significant.

Discussion

This work explored two novel MRI approaches and their ability to detect aging and short-term sleep-related changes in factors reflecting neurofluid transport, which may be related to downstream glymphatic exchange. Using IVIM, we showed D*, reflecting CSF mobility in the SAS, was significantly reduced after sleep deprivation compared to following a night of normal sleep. This parameter was also reduced in healthy older adults compared to healthy young adults. Using multi-echo ASL, we observed that perfusion and trans-vascular water exchange time were reduced in older adults compared to young adults. At the same time, this parameter was not sensitive to overnight sleep deprivation.

A key component of glymphatic transport is the movement of CSF within the SAS space and along perivascular spaces into the brain interstitium. Current approaches using intrathecal gadolinium are challenging to implement for research purposes, and unsuitable for repeated measurements to assess changes in glymphatic transport across different physiological or pathological states. Here, we used IVIM as an MRI marker of CSF mobility within the SAS as an upstream indicator of glymphatic transport. However, it is important to realize that SAS CSF flow does not represent glymphatic exchange between the CSF and brain interstitial compartments. Still, based on the Gadolinium tracer uptake in the SAS followed by the centripetal movement along perivascular spaces into the brain parenchyma, the SAS has been identified as an important proximal segment of this process, linking to CSF flow along the perivascular spaces of penetrating pial arteries and hence exchange with the interstitial fluid.

Le Bihan proposed IVIM as an approach to discriminate microvascular blood flow from water diffusion in the brain. He further also showed its utility in measuring D* of the CSF in the ventricles. ¹⁹ Hare et al., conducted a comparative study of IVIM with and without CSF nulling (in a single participant) and found that IVIM signal in the brain was primarily driven by CSF. ³⁴ In this study, we used IVIM D* to measure CSF-mobility within the large SAS space as it flows through and around the trabeculae in a pseudorandom manner.^21,35 IVIM utilizes a broad range of b-values with the low b-values to sensitize itself to CSF-mobility. Low b-values have also been used by other studies to assess CSF flow in different ways prior to our study. For instance, Wells et al. used b = 100 s/mm² to determine the diffusion and direction of CSF flow along the SAS of large arteries in rodents, which are, however, still far from the site of glymphatic exchange and may represent a different magnitude of CSF motion compared to spaces around smaller pial arteries or the SAS. ³⁶ They also measured CSF diffusion in the nearby SAS, with values similar to those we found in our study. Le Bihan et al. attempted to use low b values (b = 250 s/mm²) to determine the signature-index (similarity of the tissue under observation with a library of healthy and disease tissue signals) to evaluate changes in tissue microstructure and indirectly in brain fluid transport, assuming that all perfusion-related IVIM effects are minimized at this value.^17,37,38 Bito et al. used a b-value of 100 and 1000 s/mm² and showed that diffusion at the lower b-value (which would be analogous to our D* parameter) was greater in the CSF spaces of large vessels and in the extra-axial CSF, while the diffusion at b = 1000 s/mm² was much lower. ³⁹ They established the pseudorandom nature of CSF flow and further noted that their approach could not capture all types of pseudorandom motion, so IVIM or multi-shell diffusion tensor imaging would be appropriate. CSF mobility and thus the underlying flow velocities are unknown and likely to vary in the SAS, in the ventricles, in the MRI visible perivascular spaces, and around large blood vessels, and they could span a range of velocities. We did not assume a single fluid flow velocity cut-off and used the entire b-values range from 0 to 1000 s/mm² to account for all flows. Indeed, when considering a large CSF ROI like the ventricles (without the choroid plexus), D* was significantly (p ≪ 0.001) greater in the ventricles than in the SAS. D showed a trend towards a higher value in the ventricles than in the SAS (p = 0.035). No difference was detected with aging or sleep deprivation. If CSF velocity was uniform and known across the brain, IVIM could be replaced by a single low b-value diffusion tensor imaging approach. The added advantage would be that the directionality of fluid flow could then also be estimated from the tensors. However, Bito et al., found that fractional anisotropy in the SAS was low ∼0.30, which likely suggests that the SAS flow, which is in line with our hypothesis, would be a pseudorandom mobility, where directionality cannot be accurately determined. Other CSF spaces like the Aqueduct of Sylvius had higher anisotropy of 0.48 in the superior inferior direction representing thus more laminar than random mobility.

Additionally, they estimated both types of CSF diffusion in every voxel but could not derive the relative contribution of each process on a per voxel basis. The ‘f’ parameter in IVIM allowed us to quantify the contribution of each diffusion process.

In our study, as mentioned earlier, we consider D* to represent the CSF mobility within the SAS. We believe that D represents the movement of water within the collagenous fibers of the trabeculae, but this is only speculation. However, in agreement with Bito et al.’s work, values of D are much smaller than D* as well as those typically observed in brain tissue.

IVIM is sensitive to perfusion-related effects, and the SAS includes penetrating vasculature. We minimized hemodynamic contributions by considering an ROI containing voxels comprised of >95% CSF. We further tested in 4 individuals an identical IVIM protocol (with 6 slices) as above, in which signals from all tissue types, including GM, WM, CSF, and blood, will contribute to the total signal, and compared it with a similar IVIM protocol albeit with prolonged TE (340 ms) to minimize contributions of GM, WM, and blood, i.e., only CSF signals remain. Differences in D, D*, and f were minimal between the two protocols (−3 ± 10%, 2 ± 10%, and 3 ± 4%, respectively). As a validation, we found that in the WM, D reduced by 65 ± 10%, D* increased by 33 ± 17%, and f reduced by 22 ± 9%. As expected, the decreases in D and f are substantial in the WM. Increases in D* are unlikely, and we attribute these increases to fitting inaccuracies in the presence of a very low signal in the WM. Interestingly, the IVIM parameters are not substantially changed in the SAS, indicating that D, D*, and f in the SAS are subject to only minimal contamination by perfusion effects. Supplementary Figure 2 shows the diffusion-weighted images in one individual at both TE values, the IVIM parameters map, and the exponential decay signals. As with TE = 62 ms, D and D* in the ventricular CSF ROIs are greater than in the SAS, as evidenced by the faster initial exponential decay. Our decay curves suggest the optimum b-value to detect SAS CSF diffusion may be around 150–200 s/mm². This is based on a 37% signal drop from the signal at b = 0 on an exponential curve or a log ratio value of −0.99 i.e., ln(0.37). Further investigation in a larger group of participants is needed to confirm this observation.

IVIM provides three parameters: diffusion, D, pseudodiffusion D*, and fluid volume f. Due to the high variability and noise in the IVIM signal, the product fD* is often used as a measure of fluid movement instead of D* alone. We chose to consider the two separately and have observed that with sleep deprivation, D* is significantly reduced, while f is unchanged. Although not detected under the present conditions, fluid volume changes may occur in the SAS across these conditions. Studies to date observed an increase in interstitial fluid volume with sleep, which in this case would be defined by f. ³ However, the subjects were not scanned during sleep but after sleep deprivation, which might explain the absence of a clear change in f. For an independent verification, we performed volumetric quantification of CSF from the T1 data for all individuals. The average CSF volume in older adults (174533 ± 44119 mm³) was significantly greater than in young adults (132843 ± 18730 mm³, p = 0.009). This is in accordance with the existing literature, where aging increases atrophy and CSF volume. ⁴⁰ No significant difference was observed between sleep and sleep deprivation conditions (120712 ± 14944 mm³). One drawback of our approach that remains is that in GM and WM, we cannot separate the contribution of the perfusion and the CSF compartments.

Many studies have shown that perfusion is reduced with age^41,42 and in older adults with poor sleep (duration and quality). ⁴³ However, the acute effects of sleep deprivation are still unknown. A recent study with electroencephalogram (EEG) and ASL showed that non-REM sleep and greater slow wave activity were associated with reduced prefrontal CBF. ⁴⁴ In our study, we observed aging-related changes in CBF as expected, but we did not observe any change in CBF upon sleep deprivation. These results may conflict with prior studies. Zhou et al. recently showed that with 36 hours of sleep deprivation, hippocampal and prefrontal CBF reduced compared to a baseline scan after normal sleep. Another study by Poudel et al. showed that CBF was reduced after 24 hours of sleep restriction and an added restriction of sleep to only 4 hours prior to the 24 hour sleep-restriction. One possible explanation may be that a single night of sleep deprivation (∼24 hours) may not be sufficient to evoke such changes in resting CBF in young, healthy adults. A second possibility is that regional effects on CBF occur in response to sleep deprivation but that these effects are obscured by the present approaches that assess global changes in CBF.

The advantage of our ASL approach was that we could measure trans-vascular exchange time or the permeability of the cerebral vasculature to water. Prior studies have shown this measure to be sensitive to aging and Aqp4 gene deletion in humans and mice.^23,45 No previous studies have been conducted in humans or mice to examine the relationship between trans-vascular water permeability and sleep. Our study found that the permeability, measured in terms of exchange time (T_ex), was significantly lower with increasing age. The exchange time was unchanged after sleep deprivation, and sleep deprivation effects were not significant. This may be because of the relatively small sample size in the present study, or it may be that regional differences in T_ex following sleep or sleep deprivation may be obscured using the current global measures.

T_ex is dependent on multiple processes, i.e., aging, BBB permeability, changes in the volume of fluid-filled perivascular spaces, and AQP4 expression and localization. While BBB breakdown and greater MRI-visible perivascular space burden will reduce T_ex, reduced perivascular AQP4 localization might be expected to increase T_ex in older adults. Studies into the AQP4-dependence of changes in T_ex have been conducted in mice. However, perivascular AQP4 localization observed in human astrocytes is starkly different from that observed in mice, with non-perivascular astroglial processes exhibiting greater AQP4 localization than in mice. How these differences in AQP4 localization, and the differences in AQP4 expression and localization in the white matter far removed from the endothelial-gray matter boundary affect T_ex is unknown. Hence it is currently difficult to determine yet how closely T_ex is associated with perivascular glymphatic exchange in humans compared to changes in BBB permeability; further investigations into these interactions are warranted.

To keep acquisition times low, we used only three delay times in our current implementation which could limit the ability to detect subtle differences in T_ex. Using more post-labeling delays would improve the accuracy of the arterial arrival times and tissue transit times needed for calculation of T_ex. A different approach to measuring trans-vascular permeability is the use of diffusion-weighted ASL. Gold et al. showed that with this approach, the water exchange rate was faster with aging in humans.^46,47 Our observation of a lower T_ex aligns with this previous observation. Diffusion-weighted ASL has not been used to study acute sleep deprivation, but a faster exchange rate was observed in individuals with obstructive sleep apnea. ⁴⁸ Taken together, we believe perfusion is likely sensitive to sleep deprivation but may require a more intense sleep restriction protocol or greater regional resolution. Both the multi-echo ASL and diffusion-weighted ASL sequences have low SNR (SNR being slightly higher in the diffusion-weighted ASL), but are non-standard sequences, and availability is vendor and scanner dependent.

Finally, many MRI methods have been proposed to study perivascular glymphatic transport. Diffusion tensor image analysis along the perivascular space (DTI-ALPS) is a diffusion-weighted MRI approach that proposes to evaluate the diffusion of CSF along perivascular spaces and is sensitive to aging effects and sleep physiology. The ALPS index is significantly lower in individuals with obstructive sleep apnea and is associated with N2 sleep stage duration in healthy older adults.^{49 –51} DTI-ALPS relies heavily on correctly identifying the orientation of perivascular spaces surrounding periventricular medullary veins, which is often tricky. Secondly, the sequence has a lower resolution on the order of 2 mm³ isotropic, while the signal in a typical DTI sequence is not specific to CSF or ISF. Perivascular spaces are much smaller than these dimensions, and partial voluming is a major challenge in this approach. It is difficult to determine whether changes in the ALPS index are due to changes in fluid and/or fluid motion in perivascular compartments, differences in blood signal, or frank axonal injury or degeneration in the vicinity of these vessels. IVIM also suffers from poorer resolution, but the multiple b-values allow discrimination between water pools with different mobility. Moreover, our ROIs in SAS can be placed more efficiently and reliably than identifying PVSs in a diffusion-weighted imaging sequence.

Another approach is to capture fluctuations in the BOLD fMRI signal within the ventricular CSF compartment or the use of ultrafast fMRI. Fultz et al. reported that the BOLD fMRI signal in the 4th ventricle was closely coupled with sleep stages. They and Helakari et al. ⁵² further show that CSF fluctuations are associated with changes in CBF and cerebral blood volume (CBV) due to cardiac and respiratory fluctuations during sleep. These findings suggest that CSF dynamics within the ventricles are driven at least in part by cerebral hemodynamics. While the advantage of ASL is that it provides quantitative perfusion maps and, in our study, also a measure of trans-vascular permeability, the BOLD fMRI signal is faster, allowing better tracking of hemodynamics dependent on neuronal activity. Therefore, a fast fMRI scan would provide complementary information to ASL and should be considered in future multi-modal MRI studies.

There are several additional considerations for future studies. First, our measurements were made during awake states; thus, our findings represent the result of sleep and sleep deprivation. To define the dynamics of fluid transport through the course of physiological sleep, these methods should be evaluated during sleep and wake with in-magnet sleep-EEG studies. This would also improve our understanding of how the IVIM and ASL parameters are related to sleep. Finally, cognitive tests in this study were added to determine normal cognition but were not sensitive to sleep-related attention deficits. The inclusion of sleep-sensitive cognitive measures, such as a psychomotor vigilance task, could therefore aid in defining the relationship among sleep, cognitive performance, and MRI measures of glymphatic function in future studies.

Conclusion

In this work, we showcase two brain imaging methods, IVIM and multi-echo, multi-delay ASL, to test their sensitivity to CSF flow and trans-endothelial permeability changes upon sleep modulation and aging, seeking to define their potential utility as correlates of glymphatic transport in the human brain. We observed that the diffusion-based IVIM approach was sensitive to both sleep physiology and aging, while perfusion-based ASL approach was sensitive to aging but not sleep physiology. Both approaches are attractive to better understand the interaction of vascular and CSF circulation measured with our methods, with ISF solute transport during sleep and to assess the risk of dementia. Even though we show a strong low p-value (high statistical significance) for our within-subject design for decreases in SAS D* with sleep, the study sample size is low due to challenges in recruiting. For this particular study, participants completing both sleep and sleep deprivation arms were 60% of those completing only one arm and 45% of those who agreed to be in the study. Therefore, observations in our work should be considered preliminary. Future studies could include measures of sleep architecture, markers of dementia pathogenesis, and sleep-sensitive cognitive measures in addition to these MRI measures.