Circadian control of brain glymphatic and lymphatic fluid flow

After establishing a day/night difference of influx in vivo, we generated a full time course of glymphatic influx under KX anesthesia (Fig. 1a, d). CM injections of CSF tracer were performed every 4 hours across a standard 24 h light cycle, for a total of six time points. Using ex vivo tissue processing, we prepared 100 µm thick coronal sections through the brains and quantified mean pixel fluorescence intensity across six slices collected at 600 µm intervals starting at 1.2 mm anterior to bregma, to calculate the mean total tracer influx per entire brain (n = 10–12 mice per time point; Supplementary Fig. 1b, Supplementary Fig. 2). Cosinor analysis of mean pixel intensity of brains across a 24 h cycle revealed a significant rhythm in glymphatic influx with a peak at ZT6.2, around mid-day (cosinor analysis: F(2, 67) = 5.949, p = 0.004; R² = 0.151). The difference in influx between the day and night was ~22% (average day influx: 7.91 ± 0.4 MPI, average night influx: 6.20 ± 0.3 MPI). Because these experiments were all performed under KX anesthesia, and some anesthetics have potential to shift circadian timing¹⁸, we generated full time courses under two additional anesthetics, pentobarbital and 2,2,2-tribromoethanol (also known as avertin) (Fig. 1d; n = 7–11 mice per time point for each anesthetic). Anesthesia impacted overall tracer influx, as reported previously¹¹, where KX had significantly more influx (average influx under KX: 7.08 ± 0.3 MPI) when compared with pentobarbital (4.42 ± 0.1 MPI) or tribromoethanol (3.47 ± 0.1 MPI; Kruskal–Wallis H(3) = 105.2, p < 0.0001). More importantly, we show significant rhythms under both pentobarbital (cosinor analysis: F(2,52) = 5.527, p = 0.0069; R² = 0.175) and 2,2,2-tribromoethanol (cosinor analysis: F(2,49) = 6.479, p = 0.0035; R² = 0.209) with acrophases around mid-day (ZT7.8 and ZT6.6, respectively) and similar amplitudes (Fig. 1e, f).

In addition to using a cosinor analysis to determine rhythmicity and circadian parameters of glymphatic influx, we also analyzed the data using CircWave^19–21. This re-analysis confirmed our previous findings, where the simplest fit of the data were a single-harmonic sinusoid curve that accounted for ~20% of the overall variance. We also compared the Center of Gravity (CoG), a measure of phase and mesor of the data sets, and found similar results to our 95% confidence intervals above, where anesthetic changed the mesor but each phase estimate was approximately mid-day (Supplementary Fig. 1c–e). These experiments demonstrate that regardless of anesthesia, glymphatic influx of CSF tracer into the brain exhibits diurnal variation with peak influx around mid-day.

Although glymphatic influx is significantly higher in the sleeping¹ or anesthetized¹¹ brain, we tested whether awake mice showed day/night variation of CSF influx. We found no significant difference of CSF tracer between the day (ZT4–8, n = 7 mice) and night (ZT16–20, n = 9 mice) in total influx, anterior to posterior distribution, or in distinct subregions (Supplementary Fig. 1f–h). This was unsurprising as the awake brain has limited glymphatic function¹.

AQP4 localization to the vascular endfeet of astrocytes is dependent upon the dystrophin-associated complex (DAC)^27–29, components of which exhibit highly enriched gene expression in astrocytes (Dtna, Dmd, Dag1, Aqp4) according to a cell-specific transcriptome data base³⁰. We hypothesized that day/night changes in gene expression for components of this complex may support rhythmic localization of AQP4 to perivascular endfeet of astrocytes. Brains were harvested at ZT6 (n = 5 mice) and ZT18 (n = 5 mice), then reverse transcription PCR (RT-qPCR) was used to measure gene expression. First, we measured the clock gene Arntl1 and the non-clock gene homolog Arnt³¹. As expected, there was a significant day/night difference in Arntl1 (Welch’s t test, t(7.912) = 2.504, p = 0.037; Fig. 3h), and no difference in Arnt. Next, we measured gene expression of key components of the DAC: Dag1, Dmd, Dtna, Snta1, and Aqp4 (Fig. 3g, h). We found significant increased expression of Dag, Dtna, and Aqp4 during the day compared to the night (Welch’s t tests, Dag: t(7.874) = 2.335, p = 0.0483; Dtna: t(6.272) = 2.837, p = 0.0283; Aqp4: t(7.973) = 0.038). Both Dmd and Snta1 had similar trends of expression, but the day/night differences were not significant (Fig. 3g). In addition to AQP4, several genes encoding gap junctions (Gja1, Gjb6), cation channels (Mlc1), and bicarbonate transporters (Slc4a4) are upregulated in astrocytes³⁰, and putatively associated with the DAC²⁷. We found only Gja1 exhibited day/night differences in gene expression, with a reduction during the night (Welch’s t test, t(7.201) = 2.43, p = 0.0445). This may indicate that temporal localization of connexin-43 to astrocytic endfeet is regulated by the DAC. Increased gene expression of key components of the DAC, including AQP4, match the increased localization of AQP4 and increased glymphatic influx of CSF into the brain during the day.

Next, we tested the hypothesis that diurnal changes in glymphatic CSF tracer influx are not driven by cycles of light and dark (LD). We chose constant light (LL) over constant dark (DD) to avoid effects of light pulses that are inevitable during live-animal surgery. The surgery takes ~10 min, more than enough to phase shift mice in constant darkness^32,33. Mice were housed in constant light (LL) for 10 days with continuous activity monitoring to ensure chi-squared periodogram estimates of free-running period were accurate³⁴. Because long-term (>30 days) LL induces behavioral arrhythmicity and altered SCN organization^35,36, we wanted to ensure that general cage activity was unchanged in LL (n = 6 cages of 2 mice) compared with DD (n = 10 cages of two mice). Using chi-squared periodogram analysis we found animals in DD and LL had similar variability in periodicity, and every cage had significant Qp values from a chi-squared periodogram analysis (Supplementary Fig. 5a–c), though the amplitude of the Qp was reduced in the constant light group (DD: 1016.0 ± 61 Qp, LL: 757.7 ± 85 Qp; Welch’s t test: t(10.13) = 2.477, p = 0.0325). Similar to previous reports, mice in LL had an average period of 25.35 ± 0.1 h. Using average activity profiles for 10 days in LL and DD, we found no evidence of splitting in LL (Supplementary Fig. 5d). Although there was a trend toward decreased activity, specifically in the active phase in LL (t test: t(14) = 2.089, p = 0.0555; Supplementary Fig. 5e, f), there was no significant difference in the average amount of activity between light cycles (t test: t(14) = 1.727, p = 0.1062). Thus, 10 days in LL does not drastically alter circadian behavior, indicating the circadian system is intact.

Lymphatic vessels contract to draw fluid into the lymph nodes³⁷. We next tested whether circadian variation in mandibular lymph node filling was dependent on circadian differences in lymph vessel contraction rate (Fig. 5k, l). KX anesthetized mice were injected subcutaneously in the cheek with fluorescein isothiocyanate (FITC)-conjugated 3 kD dextran (0.25% in normal saline, 20 µL per cheek), then placed on their backs with the mandibular lymph nodes exposed via skin resection. Imaging acquisition began following lymph vessel filling (<2 min post cheek injection). In both a standard LD cycle (day: n = 7 mice; night n = 5 mice) and in 10 days LL (rest: n = 7 mice; active: n = 7 mice) we found no significant differences in contractions per minute between the rest and active phases (two-way analysis of variance (ANOVA), main effect of phase: F(1, 22) = 0.603, p = 0.4457; interaction light*phase: F(1, 22) = 0.173, p = 0.6815). We conclude that intrinsic contraction rate of the mandibular lymph node vessels is not responsible for circadian drainage of CSF to the lymph nodes.

The reduction in day-time influx in AQP4 KO animals compared with littermate (LM) controls has already been published^7,8. Next, we tested whether this persists during the night, and whether there are LM/KO day/night differences in drainage to the lymph nodes. Using the ex vivo tissue processing as described above, we found no significant difference in nighttime influx (t test: t(14) = 0.2208, p = 0.8284) between LM (n = 3 mice) and KO animals (n = 13 mice; three mice in addition to those shown in Fig. 6b) (Supplementary Fig. 6a). Previous work has shown that LM controls for AQP4 KO animals have similar day-time influx compared to C57BL6 (C57) animals⁸. Because our animals are on a C57 background, we grouped the littermate controls with the nighttime C57 under KX from Fig. 1. All three LM fit within the median quartiles of the C57 data, indicating they are comparable. Though trending toward decreased influx in KOs, there was no significant difference between WT (n = 37 mice) and KO influx in the brain (t test: t(48) = 1.962, p = 0.0556).

To measure lymph node drainage, we compared fluorescence in the mandibular lymph nodes after 30 min of CSF circulation during the day and night in LM (n = 3 mice per time point) and KO (day: n = 9 mice, night: n = 10 mice; three additional mice per group pooled with 30 min from pump start in Fig. 6) (Supplementary Fig. 6b). With the small sample size, we found no significant differences between genotypes (two-way ANOVA: F(1,21) = 0.184, p = 0.672), time (F(1,21) = 0.191, p = 0.667), or the interaction between the two factors (F(1,21) = 0.902, p = 0.353). Because our AQP4 KO line is backcrossed to C57, and the majority of LM fell within the two median quartiles of drainage distribution, we next pooled LM with C57 values at 30 min post pump start (Fig. 5) (WT day: n = 15 mice, WT night: n = 18 mice). Overall, drainage to the mandibular lymph nodes was different between groups (Kruskal–Wallis test: H(3) = 8.464, p = 0.037; Fig. 6b). Prior to Bonferroni correction, WT day vs. night, and WT night vs. KO night comparisons were significantly different (p = 0.020 and p = 0.011, respectively) supporting increased drainage only during the night in WT mice, though this effect was gone after correction.

Overall, our observations support the hypothesis that daily changes in glymphatic function and drainage of CSF to the lymph nodes is supported by daily rhythms in polarization of AQP4 to the vascular endfeet of astrocytes.

Male and female C57BL/6 mice (aged 3–5 months, weight between 25 g and 30 g) were acquired from Charles River Laboratories (Wilmington, MA) in equal numbers for each experimental group to control for any potential sex differences. Male and female AQP4 KO mice⁶² were bred in the University of Rochester vivarium, and backcrossed to C57BL/6 mice for 20+ generations before use. A minimum of five mice were used in each group, with exact animal numbers stated in the results and figure legends. Mice were group-housed either in a 12:12 light/dark cycle or under constant light with ad libitum access to food and water. All experiments were approved by the University of Rochester Medical Center Committee on Animal Resources. All of the University of Rochester’s animal holding rooms are maintained within temperature (18–26 degrees Celsius) and humidity ranges (30–70%) described in the ILAR Guide for the Care and Use of Laboratory Animals (1996). All efforts were made to keep animal usage to a minimum.

For constant light experiments, animals were housed two per cage to reduce risk of hypothermia⁶³. Activity was monitored continuously in 5 min bins via the Comprehensive Lab Animal Monitoring System (Columbus Instruments). Circadian behavioral analysis was completed with ActogramJ. Free-running period of each cage was determined using at least 10 days of activity and a χ² periodogram, which was used to confirm behavioral rhythmicity and estimate experimental times in conjunction with activity onset of the cage.

Anesthesia was administered as follows: 100 mg kg⁻¹ racemic ketamine and 20 mg kg⁻¹ xylazine kg i.p. (KX); pentobarbital 60 mg kg⁻¹ i.p.; 2,2,2-tribromoethanol (also known as Avertin) 120 mg kg⁻¹ i.p.. Depth of anesthesia was determined by the pedal reflex test; if the mouse responded to toe pinch, an additional one-tenth the initial dosage was given and the tracer experiment delayed until full unconsciousness was obtained. Directly prior to CM infusion, the animal received an additional one-tenth the initial dosage, and the pedal reflex was tested every 5–10 min during the tracer circulation time to ensure proper anesthesia throughout the study. Animals under avertin anesthesia were most likely to need supplemental dosing.

Fluorescent CSF tracer (bovine serum albumin, Alexa Fluor^TM 647 conjugate; 66 kDa; Invitrogen, Life Technologies, Eugene, OR) was formulated in artificial CSF at a concentration of 0.5% weight by volume. Anesthetized mice were fixed in a stereotaxic frame, the CM surgically exposed, and a 30 gauge needle connected to PE10 tubing filled with the tracer was inserted into the CM. Ten microliters of CSF tracer was infused at a rate of 2 μl min⁻¹ for 5 min with a syringe pump (Harvard Apparatus)¹⁰. For the awake animal cohort: mice were anesthetized with 1–2% isoflurane, and 0.25% bupivicane was applied topically to the surgery site at the beginning of and after recovery from the procedure. Mice were allowed to resurface from anesthesia before pump start. Total volume of CSF tracer was increased to 12 µL to compensate for decreased glymphatic function, pump speed remained the same (2 μl min⁻¹).

To visualize tracer movement from the cisternal compartments into the brain parenchyma, the animals were killed by decapitation and the brain removed 30 min after the start of intracisternal infusion (note that the needle was left in place after the infusion to prevent backflow of CSF). The brain was fixed overnight by immersion in 4% paraformaldehyde in phosphate-buffered saline (PBS). Coronal vibratome slices (100 μm) were cut and mounted. Tracer influx into the brain was imaged ex vivo by macroscopic whole-brain and whole-slice conventional fluorescence microscopy (Olympus; Stereo Investigator Software).

Tracer influx was quantified by a blinded investigator using ImageJ software¹¹. The cerebral cortex in each slice was manually outlined, and the mean fluorescence intensity within the cortical ROIs was measured. An average of fluorescence intensity was calculated between six slices for a single animal, resulting in a single biological replicate (Supplementary Fig. 2). Equivalent slices were used for all biological replicates.

Animals were anesthetized with KX at ZT6 or ZT18 and were prepared for transcranial imaging^64,65. In brief, animals were placed in a stereotaxic frame and the skin was removed from the top of the head to expose the skull. The animal was then placed under the macroscope (MVX10 Research Macro Zoom Microscope, Olympus). A CM injection was then performed as described above, with imaging beginning at pump start. Image acquisition was once per minute on the Cy5 channel, 30 min total (31 images per animal).

All experiments occurred between CT 2–7 or CT 14–19 depending on if it was the rest or active phase of the animal, respectively. Cannulas (guide cannula: 26 G, C315G SPC, 4.5 mm bellow pedestal, dummy: 33 G, C315DC/SP, 0.1 mm projection; PlasticsOne, Roanoke, VA) were implanted into mice 24–36 h prior to experimentation into striatum (AP: +0.6 from Bregma, LM: −2 and DV: −3.25 mm). At the appropriate experimental time points, mice in either LD or 10 days LL were anesthetized with KX and fitted with a microdialysis probe (inner cannula: 33 G, C315I/SP, 0.1 mm projection), had skin over the femoral vein resected to better visualize EB in the blood stream, and EB was injected into the brain (4% in ACSF, 1 μL, 0.2 μL min⁻¹). Images of the femoral vein were taken under a macroscope (MVX10 Research Macro Zoom Microscope, Olympus) using the Cy5 filter set, for EB visualization, and GFP, for venous structure visualization, once every 15 min for 90 min (14 images per animal) beginning at the start of EB injection. Brains were collected upon experiment end, drop-fixed overnight in paraformaldehyde (PFA) and sliced the next day in the same manner as the CSF influx experiments. Slices were used to ensure parenchymal injections did not hit the ventricle and volume of injection, as measured by % area, was similar between animals. All images were analyzed in ImageJ. Mean pixel intensity in the femoral vein was quantified on the Cy5 channel, with the GFP channel used to ensure placement of the intra-venous region of interest.

Equivalent 100 µm thick brain sections between −1.2 mm to −1.8 mm from Bregma were selected for staining for each mouse. Slices were permeablilized with 0.1% Triton-X-100 in PBS, blocked with 7% normal donkey serum (Jackson Immunoresearch) in PBS with 0.03% Triton-X-100 and incubated with primary antibody overnight, followed by three washes in PBS and incubation with the fluorophore-linked secondary antibodies (Invitrogen) for 2 hours. Stained slices were mounted with Fluoromount G (Thermofisher Scientific). Primary antibody used was rabbit anti-AQP4 (AB3594, Millipore), secondary antibody used was Alexa 594 donkey anti-rabbit (A21207, Invitrogen, 1:500 dilution), cell nuclei were identified using DAPI (D1306, Invitrogen). A subset of animals were cardiac-perfused with wheat germ agglutinin conjugated with Alexa Fluor 488 (Thermofisher Scientific) at a working concentration of 15 µg mL⁻¹ in PBS, followed by 4% PFA to enable better visualization of the vasculature. The brains were harvested and processed for immunohistochemistry as described above.

All analyses were done with investigators blinded to time-of-day, as images were taken by a separate investigator who randomized file names. Equivalent images of dorsal cortex, lateral cortex, ventral cortex, were imaged for each slice (three images per slice). For a subset of animals, additional images were taken of CA1 in hippocampus and hypothalamus. The mean immunofluorescence intensity was quantified using ImageJ. For quantification of AQP4 polarization, 50 μm segments centered on blood vessels (as identified by vascular-shaped AQP4 localization) were analyzed using the line-plot tool in ImageJ. Vessels chosen were ~6 µm wide (range of 5–8 µm) and 32 µm long (range of 20–55 µm). Three vessels that matched this criteria were chosen per image.

For polarity calculation, the baseline was defined as the average intensity over 10 µm, −20 µm to −10 µm from peak fluorescence. Baseline fluorescence was subtracted from the peak intensity of the segment. This subtraction method is ideal because it looks at changes beyond background fluorescence localized to vascular endfeet. For each dataset (LD and LL), the adjusted peak fluorescence values were normalized to the highest value out of all groups compared in the statistical analysis, one value per graph for ease of comparison and visualization, giving a range from 0 to 1.

Mice were killed at ZT6 and ZT18, mice at ZT18 were killed in a dark room using red light lamps to avoid light pulses. Brains were extracted and cut in half along the sagittal plane with the cerebellum removed, one half was used for protein extraction. Brain tissue was homogenized in lysis buffer (25 mm Tris•HCl pH 7.6, 150 mm NaCl, 1% NP-40, 1% sodium deoxycholate, 0.1% sodium dodecyl sulphate (SDS)) with a protease inhibitor cocktail (1:100 ratio; Sigma Aldrich) using a Fisherbrand 150 homogenizer. Samples were kept at 4 °C or on ice, to avoid freeze thaw cycles. Samples were sonicated 3 × 10 s (Branson Sonifier 150D) and centrifuged at 13,000 × g for 5 min at 4 °C. Supernatant was transferred to a new tube and diluted 1:10 in lysis buffer. Protein concentration was determined by a Pierce BCA assay (Thermofisher). Samples were prepared for loading in sample buffer (100 mm β-mercaptoethanol and Laemmli Buffer, Bio-rad) at 20 µg protein per well. Dual Color Ladder (Bio-rad) was loaded to identify molecular weights. The gel was 4–15% Bis-acrylamide (Bio-rad). Gels were transferred to PVDF membranes (Bio-rad), then membranes were blocked in: Tris-buffered saline (Bio-rad), 0.1% Tween-20 (Sigma), and 5% fat free milk (Bio-rad), for 1 h on an orbital rocker. Membranes were incubated with primary Rabbit anti-AQP4 antibody (1:1,000 dilution; Chemicon, AB3594, Lot #: 3213917) in TBST 5% milk overnight at 4 °C. Then, incubated 1 h at room temperature with Donkey anti-Rabbit-HRP secondary antibody (1:10,000 dilution; Jackson Immunoresearch, #711-035-152, Lot#59390). Membranes were developed using Femto kit (1:20 dilution; Thermofisher) and imaged on a Chemidoc MP imaging system (Bio-rad). Membranes were stripped using stripping buffer (2% SDS, 62.5 mm Tris, pH 6.7, 100 mm β-mercaptoethanol) at 50 °C for 10 min. Membranes were blocked again, then incubated with mouse anti-βactin primary antibody (1:5000 dilution; Cell Signaling Technology, #12262 s, Lot #: 2) in TBST 5% milk at 4 °C overnight. Donkey anti-Mouse-HRP secondary antibody (1:10,000 dilution; Jackson Immunoresearch #715-035-150, Lot #: 143140) was applied for 1 h and the membranes developed and imaged.

Brains were collected at ZT6 or ZT18, cut in half along the midline with the cerebellum removed, and flash-frozen on dry ice then stored in a −80 °C freezer until RNA isolation (n = 5 per group). RNA was isolated using Trizol®, cleaned via RNeasy Mini kit (Qiagen, Germany) following manufacturer’s instructions and quantified on a nanodrop spectrophotometer. First-strand cDNA was synthetized using TaqMan Reverse Transcription Reagents (Applied Biosystems, USA). Real-time PCR samples were prepared in triplicate with 5 ng of RNA in FastStart Universal SybrGreen Mastermix (Roche Diagnostics, Germany) and amplified on a CFX Connect Real-Time System Thermocycler (Bio-Rad, USA). Primer sequences are listed in Supplementary Table 1. Melting-curve analysis was performed following each PCR to confirm reaction specificity. Results were normalized within samples to 18s gene expression. Fold changes were calculated using the ΔΔCt method⁶⁶.

Mice were anesthetized with KX between CT 4–8 or CT 16–20 and received a CM injection as described above. Upon completion of the tracer infusion (5 min) the mice were placed supine under the macroscope (MVX10 Research Macro Zoom Microscope, Olympus) with all skin over the neck resected to reveal the superficial lymph nodes. Images were taken on the Cy5 channel for 50 minutes at a rate of one image per minute, 51 images total (ORCA Flash 4.0 CMOS Camera, Hammamatsu). The speed of the lymph node filling (fluorescent intensity per min) was evaluated by the slope of a linear regression on the first 20 minutes of each recording (20 images).

Mice were anesthetized with KX between CT 4–8 or CT 16–20. Mice had superficial lymph nodes exposed and were placed under the macroscope, as described above. FITC-conjugated 3 kD dextran (Invitrogen, 0.25% in normal saline, 20 µL per cheek) was injected into each cheek of the mouse. Imaging acquisition began following lymph vessel filling (<2 min post cheek injection) at a rate of 2.5 hz for 5 min (751 images per animal). Images were analyzed in ImageJ. A horizontal line segment centered on the lymph vessel was drawn and fluorescence over time was calculated using this region of interest. Contractions per minute (the number of oscillations of fluorescence, similar to ref. ³⁷) was calculated from each lymph vessel, and averaged for each animal.

For visual evaluation of the differences in tracer intensity and dynamics of the day and night animals, population-based averages were created. The dorsal view time series (Fig. 1b) were collapsed into videos, motion corrected and spatially aligned using ITK-SNAP⁶⁷. At each time frame, the mean of each group was computed and displayed (Supplementary Video 1). For the coronal brain slices (Fig. 6c) an initial template was computed as the mean of the unaligned images. All slices were then registered to the template and a new mean computed. This iterative process was repeated eight times with increasingly refined registration methods (four iterations of rigid registrations, two of affine registrations, and two of non-linear registrations). Image registration was performed using Advanced Normalization Tools (ANTs) 2.1.0 and scripted with Python 3.6.

All statistical analysis was performed in Graphpad Prism version 7.0, with the exception of the cosinor analyses. Cosinor analyses^68,69 were performed using PASW Statistics 18. For comparisons of means in samples with normal distributions and homogeneous variances (as indicated by a Levene’s test), an unpaired t test was used for two groups. For more than two groups an ANOVA was used for comparisons between means, followed by Bonferroni p value correction for multiple comparisons. In cases of a non-normal distribution (as indicated by a Shapiro–Wilk test) or unequal variances (Levene’s test), a nonparametric Mann–Whitney test or a Welch’s t test was used for comparisons between two means, and a nonparametric Kruskal–Wallis test was used for comparisons between more than two means, followed by Bonferroni p value correction for multiple comparisons. Significance was ascribed at p < 0.05.

Further information on research design is available in the Nature Research Reporting Summary linked to this article.