Introduction
The anatomical pathways and physiological mechanisms for the production, circulation and absorption of cerebrospinal fluid (CSF) and brain interstitial fluid (ISF) are currently an area of intense research focus. Historically, ISF is considered to be produced at the blood–brain barrier and to drain out of the brain towards the CSF [
1,
14,
24,
53]. Pathways for a bulk flow of ISF from the brain parenchyma were found to be along peri-(or para-) vascular spaces (PVS) to reach the subarachnoid space (SAS) [
13,
24,
53,
57] or along white matter fiber tracts to reach the ventricles [
13,
46,
48]. However, several reports were published that ran counter to this view and suggested that a flow of CSF could occur from the SAS into the brain [
8,
11,
40,
45,
52]. Of special note, Rennels et al. proposed a rapid CSF microcirculation to the brain with paravascular influx around arteries and efflux along veins. Although this concept was not fully accepted at the time [
27], this work was recently cited as supporting evidence for a proposed glymphatic system of flow by Iliff et al. [
29]. In experimental studies, tracers injected into the cisterna magna in mice were found with 2-photon imaging through a cranial window to enter the PVS around arteries penetrating into the cortex. At later time points tracers were also present in the PVS around veins. It was proposed that an astrocyte-facilitated convective flow existed through the parenchyma with influx along arteries and efflux along veins. Interestingly, this system was found to be more active during sleeping or anesthetized conditions compared to awake conditions [
54]. However, at this time, these studies remain highly controversial and several groups have challenged various aspects of the concept [
1,
3,
5,
19,
21,
25,
26,
50,
51].
While it is traditionally understood that CSF is absorbed mostly through arachnoid projections into the venous blood, we have recently demonstrated, using in vivo dynamic fluorescence imaging, that bulk outflow of CSF occurs predominantly through the lymphatic system in mice [
37]. The major CSF outflow pathways were found along cranial nerves (e.g., olfactory and optic) to reach lymphatic vessels outside the skull, consistent with many previous reports [
10,
32,
33,
39]. We have also demonstrated that lymphatic transport of interstitial fluid from skin and the peritoneal cavity is vastly increased during awake conditions compared to anesthetized conditions [
44]. Therefore, it is of interest to determine how CSF outflow differs between awake and anesthetized mice and how these potential differences could influence the spread of CSF to the PVS of the brain. Thus, the first aim of the current study was to assess how lymphatic outflow from the CSF to the systemic blood is altered during awake conditions compared to anesthetized conditions. We next aimed to determine whether the outflow of CSF was correlated with the spread of tracers from the CSF into the PVS of the brain surface and brain parenchyma. Finally, we attempted to confirm whether an influx of tracers through the arterial PVS could be demonstrated under in vivo anesthetized conditions using through-skull near-infrared (NIR) imaging or by magnetic resonance imaging (MRI).
Materials and methods
Mice
Female C57BL/6J-
Tyrc-J albino or C57BL/6J wild type mice (Jackson Laboratories, Bar Harbor, ME) and Prox1-GFP [
12] and SMMHC-GFP [
55] reporter mice on C57BL/6J backgrounds were kept under specific pathogen–free conditions and used for experimental studies at the age of 2–3 months. All mouse experiments were approved by Kantonales Veterinaramt Zurich (license numbers 185/13, 196/13 and 161/16) and by the Landesamt für Gesundheit und Verbraucherschutz, Saarbruecken, Germany (license number: 15/2017), and performed following the regulations of the Swiss Federal Welfare Act (TSchG) and the European legislation on the protection of animals (Directive 2010/63/EU).
Studies of CSF outflow during awake or anesthesia conditions
At 20 min before intracerebroventricular (i.c.v.) infusion, 0.1 mg/kg buprenorphine was injected subcutaneously. Mice were anesthetized by intraperitoneal injection of 80 mg/kg ketamine and 0.2 mg/kg medetomidine or inhalation of 2% isoflurane and fixed in a stereotaxic frame (RWD, San Diego, CA). For the awake group, lateral ventricle tracer infusion (as described below) was performed under 2% isoflurane and the mice were allowed to recover (within 5–10 min after infusion) and were awake for 60 min before imaging. Mice were observed to be active and behaving normally. For the isoflurane group, tracer infusion was performed under 2% isoflurane and the mice were kept anesthetized under 2% isoflurane for 60 min on a heating pad (37 °C) before imaging. For the ket/med group, 80 mg/kg ketamine and 0.2 mg/kg medetomidine were injected intraperitoneally before tracer infusion and the mice were kept anesthetized for 60 min on a heating pad (37 °C) before imaging. About 2 min before imaging at the saphenous vein, mice from the awake and isoflurane groups were given 80 mg/kg ketamine and 0.2 mg/kg medetomidine intraperitoneally. Mice were first imaged for the blood signal at the saphenous vein, then overdosed by intraperitoneal injection of 400 mg/kg ketamine and 1 mg/kg medetomidine for post-mortem imaging.
Infusion of tracers into the lateral ventricle
The skull was thinned with a dental drill (RWD) at a location 0.95 mm lateral and 0.22 mm caudal to the bregma. A 34 G steel needle was inserted into the right lateral ventricle 2.35 mm ventral to the skull surface. For the standard protocol, 2.5 µL of 200 μM P40D680 or P40D800 [
43] tracer at the speed of 1 µL/min was then infused with a syringe pump (Stoelting, Wood Dale, IL). For determination of the outflow dynamics in response to different volume infusions, alternate protocols involving 1.0 µL of 500 μM P40D680 at 0.4 µL/min and 5.0 µL of 100 μM P40D680 at 2 µL/min were used. The needle was left in place for 2.5 min and then slowly removed while observing if any significant backflow occurred. After tracer infusion, the injection hole in the skull was filled with bone wax (Ethicon, Somerville, NJ) and the scalp was sutured, except when in vivo through-skull imaging was subsequently carried out. Animals were excluded if significant backflow occurred or ex vivo analysis of brain slices indicated that the injection of tracer into the ventricle was not successful. For experiments where ventricular infusion was followed by dynamic contrast-enhanced MRI imaging, an identical protocol was used with the following modifications: 2.5 µL of a Gadospin D solution at 25 mM gadolinium (nanoPET Pharma GmbH, Germany) was infused instead of P40D680, a 33 G steel needle was used in combination with a NanoJet syringe pump (Chemyx, Stafford, CT) and the needle was left in place after infusion for 5 min.
Infusion of tracers into the cisterna magna
Mice were anesthetized by intraperitoneal injection of 80 mg/kg ketamine and 0.2 mg/kg medetomidine, fixed in a stereotaxic frame (Kopf, Tujunga, CA) and the body temperature was maintained at 37 °C using a heating pad. A surgical procedure to access the cisterna magna was performed [
30]. After a small skin incision over the occipital bone/cervical spinal cord was made, the three covering muscle layers were carefully dissected under a stereomicroscope using fine forceps and scissors. A beveled glass capillary micropipette (Sutter instruments, Novato, CA, USA) with a diameter of < 60 μm was made using a Sutter P97 Pipette puller (Sutter instruments) and was positioned perpendicular to the ear bars and advanced to penetrate the dura until resistance was overcome, indicating entry into the cisterna magna as previously described [
2]. Overall, 5 µL of a Gadospin D solution at 25 mM gadolinium (nanoPET Pharma GmbH) was infused at the speed of 1 µL/min with a NanoJet syringe pump (Chemyx). After the infusion, the micropipette was left in place for 10 min to avoid reflux. Following its withdrawal, the wound was closed.
NIR imaging of CSF lymphatic transport to systemic blood
For noninvasive imaging of tracer signals in blood [
37], fur above the saphenous vein region was removed with a razor and depilation cream before the i.c.v infusion. At the desired time point after i.c.v. infusion, mice were anesthetized with ket/met as described above. Mice were then positioned under a Zeiss StereoLumar.V12 stereomicroscope with AxioVision software (Carl Zeiss, Feldbach, Switzerland) and a Photometrics Evolve 512 camera (Photometrics, Tucson, AZ) in a supine position on a heating pad (37 °C) for imaging. The autofluorescence signal on the GFP channel was used to position the saphenous blood vessels at 64 × zoom. An image under the Cy5 filter was acquired with exposure time and camera gain settings of 200 ms and 200, respectively.
For dynamic imaging of tracer outflow to blood with different infusion rates, mice were positioned within 5 min after the completion of infusion under the stereomicroscope as above. Dynamic imaging was initiated 5 min after the completion of the ventricle infusion by acquisition of a sequence of images (1 image every 15 s for 55 min) with a Cy5 filter set to monitor the NIR signal of the saphenous vein. Exposure time and camera gain settings were 200 ms and 200, respectively.
Assessment of lymphatic transport to blood
Using AxioVision software, a circular region of interest (ROI) of radius 100 μm was placed over the saphenous vein on the acquired images. The mean signal intensity was then recorded within this region. For quantification of signal enhancement, tissue background signals were subtracted using the mean values from three uninjected mice with the same image acquisition settings.
For the dynamic imaging, a table of fluorescence intensity in counts versus time was exported into Microsoft Excel using the measure profile function. Since there was a slight loss of signal at the beginning of the scans due to photo bleaching of tissue autofluorescence, baseline intensity in counts was calculated as an average signal of the lowest ten consecutive imaging frames. This baseline intensity was then subtracted from the fluorescence intensity values in order to plot fluorescent signal enhancement versus time in min. The transport time to blood was determined as the point at which signal enhancement value was 100 counts above baseline levels.
Analysis of tracer distribution on the brain surface and in CNS-draining lymph nodes
Images of P40D680 tracer spread on the surfaces of the brain and within the deep cervical and mandibular lymph nodes were acquired with a Zeiss AxioZoom V16 microscope and a QImaging OptiMOS sCMOS camera (QImaging, Surrey, Canada) combined with a light-emitting diode illumination system pE-4000 (CoolLED Ltd, Andover, UK) and ZEN 2 software (Carl Zeiss, Feldbach, Switzerland). In excised brains, images were acquired over the contralateral dorsal hemisphere (20 × , 20 ms exposure) and of the entire ventral side of the brain (11.2 × , 200 ms exposure). Images of lymph nodes were acquired in situ at 25 × and 200 ms exposure time. Since there were no apparent differences in signal in the lymph nodes on the injected and contralateral sides, the average value of the nodes from each side was used. For quantification of signal enhancement, tissue background signals were subtracted using the mean value from three uninjected mice with the same image acquisition settings.
In vivo imaging through the skull
After i.c.v. tracer infusion, the mouse was transferred to the microscope and the head fixed in a stereotactic frame on a heating pad (37 °C). The skull was kept hydrated with warmed PBS under a glass coverslip, and dynamic imaging of the contralateral half of the brain was carried out for a maximum of 60 min after tracer infusion through the skull of the anesthetized mouse using a Zeiss StereoLumar.V12 stereomicroscope (as described above). Dynamic imaging through the skull was carried out with a gain of 300 and an exposure time of 50 ms (Cy5 filter for P40D680 tracer) or 200 ms (ICG filter for P40D800 tracer). In some cases, imaging was continued during administration of an overdose of 400 mg/kg ketamine and 1 mg/kg medetomidine (i.p.), or after an overdose with ketamine/medetomidine followed by transcardiac perfusion with ice-cold PBS.
For determination of CSF dynamics, circular ROIs of 1 mm diameter were placed over the location of the contralateral lateral ventricle and over the location of the quadrigeminal cistern in the acquired video using ImageJ/FIJI. Signal intensity within these ROIs was determined over time.
Magnetic resonance imaging
Animals were examined in a horizontal-bore 9.4 T animal scanner (BioSpec Avance III 94/20; Bruker Biospin GmbH, Ettlingen, Germany) with a BGA12S gradient system with ParaVision 6.0.1 (Bruker Biospin GmbH) and a linearly polarized coil with an inner diameter of 40 mm (Bruker Biospin GmbH). Imaging was performed with a three-dimensional time of flight gradient recalled echo sequence (3D-TOF-GRE) originally adapted for imaging of peripheral lymph vessels [
41] with recovery time of 12.0 ms, echo time of 2.5 ms, flip angle 25
o, matrix 600 × 400 × 180, field of view 36.00 mm × 25.92 mm × 18.00 mm, zero fill 2, 1 average and a scan time of 4 min 19 s 200 ms. Signal intensity (SI) reduction in blood vessels was achieved by placement of a saturation slice over the mouse heart.
Three-dimensional maximum intensity projection (MIP) reconstructions, SI and noise measurements were performed with ParaVision 6.0.1 (Bruker Biospin GmbH). For SI analysis, ROIs were created manually in a single slice per measurement, 1 mm dorsal to the circle of Willis. For SI measurements in cortical parenchyma and noise determination from background SI, identically sized circular ROIs were employed, while cortical tissue containing vessels was segmented manually (see Supplemental Fig. 4).
For animals in which injection of MRI contrast medium was not successful (wrong location of the injection or significant backflow from the injection site), MR imaging experiments were discontinued.
Brain sections
Mouse brains were dissected and fixed in 4% PFA at 4 °C for 48 h. A section of the brain between 1 mm rostral and 1 mm caudal to the needle insertion site was cut out by razor blade. Coronal sections were made from dorsal to ventral side at a thickness of 100 µm with a vibratome (Leica VT1000 S). Images were acquired with a Zeiss AxioZoom V16 microscope and ZEN 2 software.
Analysis of tracer influx into the brain parenchyma
The percentage area of tracer coverage in brain sections was determined using ImageJ/FIJI as follows. The cerebral cortex was selected as an ROI on images of coronal brain sections. A uniform threshold was applied to all images, and background signal manually erased. The percentage area coverage of the signal within the ROI was determined. Analysis of all images was independently carried out in a blinded manner by two investigators, and an average was taken of six coronal sections per mouse.
The deepest tracer penetration within the dorsal cerebral cortex was determined using lines drawn perpendicular to the brain surface with ImageJ/FIJI. The two largest values were taken for the injected and contralateral sides, and an average of these four values was taken for six sections per mouse by two independent investigators. Due to clear differences between images from perfused and non-perfused brains, blinding was not feasible for this analysis.
Statistics
Mice were randomly allocated to different experimental groups. Group sizes were estimated based on pilot studies to determine the success rate and reproducibility of the intraventricular infusions. All data are presented as mean ± SD. Means of two groups were compared using an unpaired two-tailed Student’s t test. Means of three groups were compared with one-way ANOVA with the Tukey’s multiple comparison post hoc test. Correlation analysis was done using Pearson’s correlation. All analyses were performed using GraphPad Prism V5.0 (GraphPad Software, San Diego, CA) and p < 0.05 was accepted as statistically significant.
Discussion
In this study, we demonstrated using intraventricular infusions of a bulk flow tracer and fluorescence stereomicroscopy that CSF outflow was increased in awake mice compared to anesthetized mice. Evaluation of the dynamics of this outflow supports the conclusion that CSF clearance occurs through the lymphatic system in this species. An inverse relationship was found between the efflux of CSF to the systemic blood and the spread of CSF tracers to the paravascular spaces as assessed ex vivo. In vivo imaging of PVS spread of tracers at the brain surface demonstrated that this spread was limited to the PVS of larger caliber arteries and veins. Finally, we found that a rapid influx of tracers to the PVS of the cortical parenchyma occurred just after the death of the animal.
A substantial increase in the outflow of CSF tracers to the systemic blood in awake mice was found compared to mice that were anesthetized. There is very little similar data to compare to regarding this point, although one previous report demonstrated increased CSF clearance of radiolabeled tracer from the spine in active compared to resting subjects [
17]. It is likely that physical activity affects CSF pressure and increases flow towards efflux pathways. Respiration and arterial pulse are driving forces for CSF flow and mixing [
16,
20] and the rates of both are naturally higher in awake versus anesthetized animals. These results also lend further support for a CSF outflow pathway that is predominantly lymphatic in nature [
9,
37]. Lymphatic clearance is highly dependent on muscular activity and transport through lymphatic vessels is dramatically increased when the subject is awake [
22,
44,
49]. The concept of lymphatic rather than venous outflow of CSF still remains to be rigorously evaluated in humans, however, some evidence does exist to support this pathway in non-human primates and cadavers [
31,
38].
An increased CSF outflow while awake is not fully consistent with the concept of increased CSF flushing through the brain during sleep or deeply anesthetized conditions [
54]. In the glymphatic model, CSF within the arterial PVS serves as an input fluid to the brain. If the CSF turnover is rapid during awake conditions, this would limit access of CSF-infused tracers to the PVS, a factor that was not considered during the original study [
25,
54]. Another recent MRI study has challenged the concept that increased glymphatic flow occurs in anesthetized mice compared to awake mice [
21]. However, this study concluded that CSF influx was actually increased during awake conditions, which we have not found to be the case. It is unclear whether the tracer that was measured in the brain by MRI in vivo was intravascular after effluxing the CSF to reach the systemic circulation rather than a direct influx into the parenchyma from CSF as proposed by the authors. In addition, the low-molecular weight tracer used in this study as well as other recent MRI studies [
7,
34,
47] would exhibit significant concentration gradient-dependent diffusion into the brain parenchyma and thus would not accurately demonstrate the bulk flow pathways of CSF.
It is important to note that it was not our aim to assess sleeping, unanesthetized mice with our methods, nor have we monitored clearance of tracer injected into the parenchyma. Therefore, theoretically, an increased clearance of metabolites from the brain might indeed occur under sleeping conditions. The destination for the efflux fluid in the glymphatic model is not exactly clear; one version of the model shows fluid exiting through the PVS around veins to reach the CSF while another proposes that there may be connections to lymphatic vessels within the dura mater [
4,
28,
35,
36]. However, we were unable to demonstrate any paravascular influx of CSF into the brain during anesthetized in vivo conditions and we have found significantly increased clearance of CSF under waking conditions. Therefore, at this point, it is difficult to imagine how an increased brain flushing during sleep via a bulk CSF influx through the glymphatic system could be occurring as originally envisaged.
The close correlation between signals in the basal cisterns and the PVS of the brain surface highlights the circle of Willis as an important potential access point for CSF and its components into the PVS. The circle of Willis is the main source of the arteries that feed into the brain. A recent study in rats has confirmed an earlier observation that stomata exist in the pia mater covering these arteries which may provide anatomical pathways for CSF and macromolecules from the basal cisterns to the PVS [
42,
56]. The major CSF outflow routes along several cranial nerves [
37] (e.g., olfactory, optic, trigeminal) are located immediately rostral to the circle of Willis, indicating that CSF flow could be directed either out of the skull or into the arterial PVS depending on the physiological conditions.
Our data support the concept that some portion of CSF can spread to the brain surface PVS in vivo, at least under anesthetized conditions where there is reduced CSF outflow. We have also observed a spread of tracer from the PVS of arteries to the PVS of veins on the brain surface, consistent with the recently published concept of low resistance pathways for CSF flow at this location [
6]. It will be interesting to examine at an anatomical level whether connections are present between the leptomeningeal sheaths of the PVS of arteries and veins. CSF may also spread along the large vessels that penetrate from the ventral aspect into the brain, such as into the midbrain as shown in MRI studies [
15]. A recent MRI study in humans has shown that the diffusion of a low molecular weight contrast agent into the brain parenchyma correlated with the availability of the contrast agent in the surrounding CSF [
47]. Thus, a dynamic balance between CSF turnover and CSF solute distribution to the brain may exist that could be altered under different physiological or pathological conditions.
The most surprising finding of this study was a major influx of tracer into the PVS that occurred immediately after the death of the animal. This observation was predicted by the prominent neuroscientist Lewis Weed who wrote in 1914 that “ordinarily, after death, cerebrospinal fluid is aspirated by the brain” [
53]. One might expect that the substantial loss of blood pressure that occurs within the brain at death would establish a gradient for flow from the CSF to the brain parenchyma [
53]. Thus, quantifications of tracer that are performed within the PVS or parenchyma of the brain ex vivo are likely overestimates of the levels that are present in vivo. This phenomenon may also account for some of the previous findings based on ex vivo assessments that have concluded that CSF influx was occurring into the brain within minutes of tracer administration [
2,
8,
45,
52].
In sum, our data support the conclusion that a bulk flow of CSF into the brain along the PVS likely does not occur under physiological conditions and that the rate of outflow of CSF is a major factor determining whether substances within CSF can spread to the brain. This work may have important implications in drug delivery to the CNS after intrathecal administration and in several conditions that may lead to altered CSF flow dynamics such as traumatic brain injury and stroke.