Direct imaging of glymphatic transport using H217O MRI

The recently proposed glymphatic pathway for solute transport and waste clearance from the brain has been the focus of intense debate. By exploiting an isotopically enriched MRI tracer, H217O, we directly imaged glymphatic water transport in the rat brain in vivo. Our results reveal glymphatic transport that is dramatically faster and more extensive than previously thought and unlikely to be explained by diffusion alone. Moreover, we confirm the critical role of aquaporin-4 channels in glymphatic transport.


Introduction
The transport of solutes within the brain parenchyma is of fundamental importance to nutrient delivery and the clearance of metabolites, neurotransmitters, and toxic macromolecules (e.g., β-amyloid).
Conventionally, interstitial solutes were considered to be transported via diffusion, but recent evidence suggests an additional bulk flow of the interstitial fluid (ISF). The glymphatic (glial-lymphatic) hypothesis proposes that subarachnoid cerebrospinal fluid (CSF) is driven by arterial pulsation along the perivascular space surrounding penetrating arteries, with influx into the brain interstitium mediated by the astroglial water channel aquaporin-4 (AQP4) (1). It is proposed that this influx results in the bulk flow of ISF, which then exits along perivenous spaces, providing an efficient clearance mechanism for waste products from the parenchyma. Glymphatic transport has been shown to increase during sleep, increasing the clearance of β-amyloid in mice, potentially explaining the decades old mystery of why sleep is restorative (2). Glymphatic transport has been further shown to increase in ischemic stroke and play an important role in post-stroke edema (3). Conversely, decreased glymphatic transport has been shown in animal models of Alzheimer's disease (4), vascular dementia (5) and traumatic brain injury (6).
Moreover, evidence that the glymphatic pathway is present in humans has been obtained via the intrathecal injection of MRI tracers, with delayed tracer clearance in a cohort of patients with dementia (7).
However, the glymphatic hypothesis has proven to be highly controversial (8,9), particularly regarding the proposed bulk flow of ISF (10)(11)(12) and the mediating role of AQP4 (8,13,14). A key limitation of previous studies has been the use of tracer molecules that are much larger than water (18 Da), for example, in ex vivo fluorescence imaging (700-3000 Da) (1) and in vivo MRI (Gd-DTPA, 938 Da) (15). The fact that these tracers cannot be transported by AQP4 channels most likely means that they underestimate the true magnitude of ISF flow, which will depend on the rate at which CSF water 4 molecules enter the interstitium. We reasoned that using water molecules as a tracer would yield novel insights into glymphatic transport and the role of AQP4 channels.
Water molecules have previously been used as tracers by exploiting various isotopes. For example, tritiated water ( 3 H2O) has been used to investigate CSF transport by measuring the radiation levels of blood samples using a liquid scintillation counter (16). H2 (20) and H2 17 O has also been administered as a tracer to image cerebral blood flow (21). In a series of studies, Nakada et al. employed indirect 1 H detection of H2 17 O to investigate the interstitial circulation. Using an intravenous bolus injection of H 2 17 O (20% 17 O-enriched) they showed that exchange of blood water with the interstitial fluid and CSF was dependent on AQP4 and not AQP1 (22,23). They also used the same method to demonstrate that water influx into CSF was significantly impaired in SP-bearing transgenic mice (24).
In order to investigate the glymphatic system more directly, we decided to infuse H2 17

Results
To directly image glymphatic water transport within the rat brain, the highly enriched H2 17 O tracer (90% 17 O-enriched) was infused into the CSF of anesthetized rats at the cisterna magna. Serial MRI revealed glymphatic transport that was strikingly more rapid and extensive than that previously observed using conventional Gd-DTPA tracers ( Figure 1B) (15). Movement of the Gd-DTPA tracer through the subarachnoid space was slightly retarded compared with that of H2 17 O. Moreover, as previously observed (15), the Gd-DTPA tracer was slow to penetrate the parenchyma, resulting in a buildup of Gd-DTPA concentration in the subarachnoid space and ventricles.
By contrast to the Gd-DTPA tracer, H2 17 O rapidly penetrates the parenchyma in all brain regions ( Figure   1C). Thus, unlike Gd-DTPA, the concentration of H2 17 O does not build up in the subarachnoid space and ventricles ( Figure 1C). This is best appreciated by viewing the supplemental video which shows, side-byside, the temporal evolution of both tracers. For example, H2 17 O fully penetrates the rostral cortex within 10 minutes ( Figure 1C), whereas, as previously observed (15), Gd-DTPA does not penetrate even after 85 minutes. There is much debate in the literature regarding the existence (9), or not (8), of a bulk convective flow of ISF. From previous measurements, we estimate, using equation 7, the root mean squared (rms) displacement of water molecules in the brain to be 1.05 mm in 15 minutes, whereas for Gd-DTPA molecules it is 0.68 mm. This reflects the larger molecular weight of Gd-DTPA (938 Da) compared with H2 17 O (19Da). In addition, Gd-DTPA molecules are restricted to the extracellular compartment whereas H2 17 O is able to more freely diffuse into cells. However, comparing the estimated water displacements due to only diffusion, with the actual H2 17 O images ( Figure 2D), it appears unlikely 6 that diffusion alone explains the rapid brain-wide distribution of H2 17 O after 15 minutes, thus, supporting the argument for a bulk convective flow of the ISF.
We next exploited the H2 17 O tracer to investigate the importance of AQP4 channels to glymphatic transport. The group pretreated with an AQP4 inhibitor (TGN-020, i.p., IC50 = 3.1 μM) (26) experienced an 80 ± 10% reduction in H2 17  The distribution of both tracers within the cerebellum is particularly interesting. In the control group, the Gd-DTPA tracer was slow to penetrate the cerebellum, but by 85 minutes post-injection it did penetrate the outer regions ( Figure 1B). However, H2 17 O quickly penetrated the whole cerebellum, but was more rapidly removed from the white matter (figure 1B). In the group pretreated with an AQP4 inhibitor (TGN-020), the penetration of H2 17 O into the grey matter of the cerebellum was considerably slowed but was strong by 85 minutes, whereas there was still no penetration of the white matter even after 85 minutes. This is presumably related to the high concentration of AQP4 in the cerebellum (27).

Discussion
A serious limitation of previous in vivo glymphatic experiments has been their use of large tracer molecules (MRI tracer Gd-DTPA, MW 938 Da) that cannot be transported by AQP4 water channels in the brain. We hypothesized that the use of these large tracer molecules would result in the underestimation of the rate that CSF water molecules enter the brain parenchyma, thereby underestimating the true flow of interstitial water. By labelling actual water molecules, via highly 17 O enriched water, we were able to detect glymphatic water transport in vivo using high SNR T2-weighted MRI. The ability of our H2 17 O MRI method to directly image water transport in vivo provides an accurate means of studying glymphatic transport.
Reasons for the discrepancy between the transport of H2 17 O and Gd-DTPA tracers include the large difference in molecular weight (19 Da for H2 17 O and 938Da for Gd-DTPA) and the presence of the astrocytic water channel AQP4. Molecular tracers such as Gd-DTPA that lack a specific transport pathway (such as ion transporters or channels) are able to reach the parenchyma only through the ~20 nm clefts between overlapping astrocytic end feet. Water, however, is also able to travel through the intracellular space mediated by highly selected AQP4 water channels which are expressed in astrocytic end feet that cover the entire vasculature of the CNS. For this reason, it is likely that previous experimental studies based on large tracer molecules may have systematically underestimated subarachnoid CSF water flow into the brain and, thus, underestimated the convective bulk flow of ISF.
Asgari et al. (28) modelled the astrocytic syncytium between CSF and the compartment of the brain interstitium by including ,in their model, AQP4 on the plasma membranes, an abundance of AQP4 on the perivascular surfaces, and 20 nm inter-end feet gaps. They demonstrated that the resistance to water flow through ECS is two orders of magnitude larger than through the intracellular space (ICS) of astrocytes. This appears to be a likely explanation for the very rapid penetration of the parenchymal seen using the H2 17 O tracer compared with Gd-DTPA.
An alternative hypothesis of CSF physiology has been proposed by Klarica et al. in which CSF production and absorption occurs at the level of the capillaries and depends on hydrostatic and osmotic forces (29).
That hypothesis was not supported by our experiments using an AQP4 inhibitor (TGN-020), which clearly demonstrated that the rapid penetration of H2 17 O into the parenchyma (Figure 2) is strongly dependent on AQP4, which is absent from the endothelium of brain capillaries. However, it does support the glymphatic hypothesis of Nedergaard et al. (15). 8 Klarica et al. have also suggested a continuous bidirectional mixing of water between the blood, ISF and the CSF compartments, with no unidirectional net flow of CSF along the CSF spaces (30). Again, this is not supported by our H2 17  Similar to natural sleep, general anaesthesia has been shown to enhance the transport of CSF tracers e.g. Gd-DTPA. Studies by the Nedergaard group reported CSF tracer transport was highest under ketamine/xylazine (K/X) anaesthesia and lower with α-chloralose, Avertin, or isoflurane (31). At least one of the effects of anaesthesia appears to be increased extracellular volume fraction (32), which may explain the enhanced transport of extracellular tracers. However, as the H2 17 O tracer can more freely travel through intracellular and extracellular space, we would hypothesize that sleep or general anaesthesia may have less influence on its rate of transport.
In summary, we have conclusively demonstrated that glymphatic flow imaged using our H2 17 O tracer is much more rapid and extensive than when imaged using the Gd-DTPA tracer (see supplemental movie). This is strong evidence that the interstitial fluid experiences a substantial bulk flow which can more rapidly clear waste molecules from the parenchyma than by diffusion alone. Further, we were able to conclusively demonstrate that these glymphatic flows are strongly mediated by AQP4. We believe these advances will not only answer much of the controversy surrounding the glymphatic hypothesis, but also provide a valuable new tool for future investigations into associated neurological disorders. 9

Study design
Male Wistar rats (280 to 300 g, 20 to 24 weeks old) were randomly assigned to five experimental groups.
The first study was designed to test our hypothesis that the use of H2 17 O tracer would provide a more accurate measurement of glymphatic transport than Gd-DTPA. Two groups were used (H2 17 O (n=7), Gd-DTPA (n=7)). The second study was designed to test the hypothesis that AQP4 channels are critical to glymphatic transport using our new H2 17 O method (AQP4 inhibitor (n=6), vehicle (n=6), aCSF (n=3)).
After the surgical implantation of an intracisternal cannula, each rat was placed inside the MRI scanner and MRI was continued for a total of 85 minutes (for an overview, see Figure 1A). In the drug-treated group, TGN-020 was administered intraperitoneally 15 min prior to the MRI study. Respiration, blood pressure (BP) and heart rate (HR) were continuously monitored during MRI measurements and body temperature was maintained at 37.0 ± 0.5°C (see Supplemental Table 1).

Supplemental movie
The video displays differences in the brain-wide distribution of H2 17  Respiration, BP and HR were continuously monitored during MRI measurements and body temperature was maintained at 37.0 ± 0.8°C (see Supplemental Table 1). At the end of the experiment, the animal was euthanized.

Magnetic Resonance Imaging
MRI data were acquired using a Bruker PharmaScan 7T/16 cm system controlled by Paravision 5.1 software (Bruker BioSpin, Ettlingen, Germany) with a gradient coil insert (internal diameter = 90 mm, 300 mT/m) and a four-channel phased-array surface receive coil used for rat brain imaging. Two Furthermore, aCSF was used as a negative control. The scanning protocol for all studies consisted of 3 baseline scans followed by the intrathecal infusion of tracer via the CM catheter (50 μl at 1.8 μl/min; total time, 28 min). MRI images were continually acquired over a period of 85 min.

Gd-DTPA tracer imaged with T1-weighted imaging
To visualize the glymphatic pathways using gadolinium (T1W shortening effects), 3D T1-weighted FLASH aCSF was used as a negative control.

H2 17 O tracer imaged with high-SNR T2-weighted imaging
The quadrupolar moment of the 17

Equation 3
Where P is the fraction of water molecules containing 17

MRI data analysis
MATLAB R2018b (MathWorks, UK) code was developed in-house for post-processing MRI images. The general postprocessing procedure consisted of brain extraction, head motion correction, and voxel-by voxel conversion to a percentage signal change. Briefly, a brain mask was created for the removal of non-brain tissue, improving the performance of the following coregistration steps: using rigid body alignment of each scan to the mean pre-contrast image, scan-to-scan misregistration caused by head movement was corrected. The resulting registrations were visually inspected to ensure adequate alignment. To ensure that voxel intensity represented a percentage change relative to the average 13 baseline images, all time-series images were subtracted and divided by the baseline average image using the following expression: For the Gd T1W image study: For the H2 17 O/aCSF T2W image study: Where D is the diffusion coefficient and x is the particle displacement from its initial position x0, during an observation time, t. For this Gaussian function the root mean square displacement, xrms, of an ensemble of particles is given by: From LG contributed to the design of the experiments, performing surgical procedures on rats and acquiring the MRI data.
LG also contributed to writing the manuscript.
LW was the secondary PhD supervisor of MSA and contributed to the design of the experiments, to discussion and interpretation of results and to the writing of the final manuscript.  The theoretically calculated displacements of H2 17 O due to diffusion only, are 0.53mm, 0.75mm and 0.91mm, at times 231s, 459s and 665s, respectively. It is clear that the rapid H2 17 O penetration of the parenchyma cannot be explained by diffusion alone, indicating the presence of convective ISF flow. Significant differences denoted by asterisks: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Data presented as mean ± SEM.