1. Neuroscience
Download icon

Aquaporin-4-dependent glymphatic solute transport in the rodent brain

  1. Humberto Mestre
  2. Lauren M Hablitz
  3. Anna LR Xavier
  4. Weixi Feng
  5. Wenyan Zou
  6. Tinglin Pu
  7. Hiromu Monai
  8. Giridhar Murlidharan
  9. Ruth M Castellanos Rivera
  10. Matthew J Simon
  11. Martin M Pike
  12. Virginia Plá
  13. Ting Du
  14. Benjamin T Kress
  15. Xiaowen Wang
  16. Benjamin A Plog
  17. Alexander S Thrane
  18. Iben Lundgaard
  19. Yoichiro Abe
  20. Masato Yasui
  21. John H Thomas
  22. Ming Xiao  Is a corresponding author
  23. Hajime Hirase  Is a corresponding author
  24. Aravind Asokan  Is a corresponding author
  25. Jeffrey J Iliff  Is a corresponding author
  26. Maiken Nedergaard  Is a corresponding author
  1. University of Rochester Medical Center, United States
  2. University of Copenhagen, Denmark
  3. Nanjing Medical University, China
  4. RIKEN Center for Brain Science, Japan
  5. Ochanomizu University, Japan
  6. The University of North Carolina, United States
  7. Oregon Health and Science University, United States
  8. Haukeland University Hospital, Norway
  9. Lund University, Sweden
  10. Keio University, Japan
  11. University of Rochester, United States
  12. Saitama University, Japan
  13. Duke University School of Medicine, United States
Research Article
  • Cited 23
  • Views 3,106
  • Annotations
Cite this article as: eLife 2018;7:e40070 doi: 10.7554/eLife.40070

Abstract

The glymphatic system is a brain-wide clearance pathway; its impairment contributes to the accumulation of amyloid-β. Influx of cerebrospinal fluid (CSF) depends upon the expression and perivascular localization of the astroglial water channel aquaporin-4 (AQP4). Prompted by a recent failure to find an effect of Aqp4 knock-out (KO) on CSF and interstitial fluid (ISF) tracer transport, five groups re-examined the importance of AQP4 in glymphatic transport. We concur that CSF influx is higher in wild-type mice than in four different Aqp4 KO lines and in one line that lacks perivascular AQP4 (Snta1 KO). Meta-analysis of all studies demonstrated a significant decrease in tracer transport in KO mice and rats compared to controls. Meta-regression indicated that anesthesia, age, and tracer delivery explain the opposing results. We also report that intrastriatal injections suppress glymphatic function. This validates the role of AQP4 and shows that glymphatic studies must avoid the use of invasive procedures.

https://doi.org/10.7554/eLife.40070.001

Introduction

A brain-wide fluid transport pathway, known as the glymphatic system, supports the rapid exchange of cerebrospinal fluid (CSF) and interstitial fluid (ISF) along perivascular pathways (Iliff et al., 2012). The glymphatic system consists of three principal sequential anatomic segments: (i) CSF inflow along the perivascular spaces surrounding penetrating arteries, (ii) dispersion of CSF through the wider interstitium, and (iii) efflux of ISF along the large-caliber draining veins to re-enter the CSF within the ventricular and cisternal compartments (Jessen et al., 2015). Ultimately, interstitial solutes cleared to the CSF exit the brain through meningeal lymphatic vessels flanking the venous sinuses, along cranial and spinal nerve sheathes, and across the cribriform plate (Louveau et al., 2015; Aspelund et al., 2015). Astrocytic endfeet ensheath the cerebral vasculature, and the abundantly expressed astroglial water channel aquaporin-4 (AQP4) localizes primarily to the perivascular endfoot membrane domain abutting the basal lamina. As this anatomic arrangement provides a route for rapid water movement between the perivascular space and the glial syncytium, AQP4 has been proposed to support perivascular fluid and solute movement along the glymphatic system (Nedergaard, 2013).

Several groups have independently shown that the astrocytic AQP4 is essential for fast glymphatic transport. Iliff et al. demonstrated a significant suppression of both perivascular CSF tracer influx and interstitial mannitol and amyloid-β (Aβ) clearance in Aqp4 knockout (KO) mice (Iliff et al., 2012). Subsequent work demonstrated that Aqp4 gene deletion exacerbated glymphatic pathway dysfunction after traumatic brain injury (TBI) and promoted the development of neurofibrillary pathology and neurodegeneration in the post-traumatic brain (Iliff et al., 2014). Plog et al. similarly found that Aqp4 KO mice exhibit slowed transport of interstitial solutes to the blood after TBI, reflected by a significant reduction in plasma biomarkers of TBI, including GFAP, neuron-specific enolase, and S100β (Plog et al., 2015). Xu et al. reported that deletion of Aqp4 exacerbated Aβ plaque accumulation and cerebral amyloid angiopathy in the APP/PS1 murine model of Alzheimer’s disease (Xu et al., 2015). Achariyar et al. found significantly reduced distribution of FITC-ApoE3, 125I-apoE2, 125I-apoE3, 125I-apoE4, as well as 14C-inulin in Aqp4 KO mice following tracer injection to the CSF (Achariyar et al., 2016). Lundgaard et al. reported that glymphatic clearance of lactate was reduced in Aqp4 KO mice (Lundgaard, 2016). Finally, Murlidharan et al. demonstrated that Aqp4 KO mice exhibit significantly impaired clearance of adeno-associated viruses (AAV) infused into the ventricles, and concluded that glymphatic transport profoundly affects various aspects of AAV gene transfer in the CNS (Murlidharan et al., 2016). A recent MRI study showed the AQP4 facilitator, TGN-073, potentiated the transport of interstitial fluid from the glia limitans externa to pericapillary Virchow-Robin space (Huber et al., 2018).

The critical role of AQP4 in supporting perivascular CSF-ISF exchange was recently questioned in a report by Smith et al. (2017), in which the authors failed to detect any reduction in CSF tracer influx into the brain parenchyma of Aqp4 KO mice compared to wild-type (WT) controls. Because a key element of the glymphatic hypothesis is the role of astroglial water transport in supporting perivascular CSF-ISF exchange, we consider it critical to re-examine the role of AQP4 in this process, with an aim to resolve the discrepant reports. Using data generated from five independent laboratories, we have undertaken such a re-evaluation of the effects of Aqp4 deletion on perivascular glymphatic exchange. Results of this analysis consistently confirm that Aqp4 deletion impaired perivascular glymphatic flow relative to that in wild-type mice. Our conclusion is strengthened by the use of four independently generated Aqp4 KO lines, including the line used by Smith et al. (2017), as well as the α-syntrophin (Snta1) KO line, which lacks AQP4 perivascular localization despite normal expression levels (Amiry-Moghaddam et al., 2003).

Smith et al. (2017) also questioned the existence of tissue bulk flow based injecting tracers of varying molecular sizes into cortex or striatum. We questioned this approach based on the prior finding that traumatic brain injury is linked to an immediated and sustained reduction in CSF influx (Iliff et al., 2014). As expected, our analysis showed that insertion of glass pipettes into the brain markedly reduced brain-wide CSF tracer influx. Thus, interstitial fluid transport should not be studied following invasive procedures in the brain.

Results

The study included data from five laboratories using four independently generated Aqp4 KO lines and one Snta1 KO mouse line (Figure 1a–e) (Fan et al., 2005; Ikeshima-Kataoka et al., 2013; Ma et al., 1997; Thrane et al., 2011; Adams et al., 2000). Immunohistochemistry done in parallel with the glymphatic experiments verified that AQP4 was indeed deleted in all the Aqp4 KO mouse lines (Figure 1a–c and e). In the Snta1 KO mice, immunofluorescence demonstrated that perivascular AQP4 polarization was absent in this line (Figure 1d).

Strategy used for generation of KO mice and the experimental design of the study.

Top row of each panel represents the institution where the line originated from and the strategy used to generate the four global Aqp4 KO mice and the Snta1 KO mice. Second row of each panel shows the five labs that collected data on glymphatic system function in the five transgenic mouse lines: (a) Nanjing Medical University (NMU), (b) RIKEN Center for Brain Science (RIKEN), (c) University of North Carolina (UNC), (d) Oregon Health and Science University (OHSU), (e) University of Rochester Medical Center (URMC). (a-e) Immunohistochemical analysis showing the lack of AQP4 expression in the global KOs and (d) lack of perivascular AQP4 localization in Snta1 KO compared to WT mice. Scale bar: 50 µm. (a-e) Third row displays the volume and rate used for the intracisterna magna (CM) injections for each experiment, the tracer used, and the experiment duration. Note that URMC collected two full data sets using an injection rate of either 1 or 2 µl/min. (a-e) The last row display the analyses strategy employed by each of the five research group in Figures 26. (f) Additional experiments performed in Figure 7 tested the effect of intrastriatal (IS) injection on global glymphatic function. TxRd, Texas Red; BDA, biotinylated dextran amine; BSA-647, bovine serum albumin-Alexa Fluor 647; DCE-MRI, dynamic contrast-enhanced magnetic resonance imaging.

https://doi.org/10.7554/eLife.40070.002
NMU: Aqp4 gene deletion reduced the penetration of intracisternally injected tracer into the brain parenchyma.

Texas Red-conjugated dextran (TRd3, 3kD) was injected intracisternally into WT and Aqp4 KO mice. Thirty minutes after injection, the anesthetized animals were perfusion fixed, and the fluorescence was evaluated ex vivo. (a) Representative near infrared (NIR) fluorescence images of the dorsal and ventral whole-brains of four mice per genotype. (b-c) Quantification of the mean integrated optical density (MIOD) of TRd3 on the dorsal (b) and ventral (c) brain surface of WT (grey) and Aqp4 KO (purple) mice from 4.0 mm anterior to 8.0 mm posterior to bregma. (d) Representative images of coronal brain sections at +0.5 mm from bregma from five pairs of WT and Aqp4 KO mice showing TRd3 distribution within the brain. (e) High magnification micrographs of the hypothalamus (lined area in d) showing the fluorescence intensity of TRd3 within the perivascular space (star) and adjacent brain parenchyma (dotted line) of WT mice and Aqp4 KO mice, respectively. (f) Quantification of the percentage area of whole-slice fluorescence of the both genotypes for 6–8 forebrain sections (+1.7 to −0.7 mm from bregma) of each mouse. (g) Diagram showing the subregional analysis of brain sections at the level of 0.5 mm anterior to bregma. (h) Quantification of the mean fluorescence intensity (AU, arbitrary units) of TRd3 of the dorsal, ventral and lateral brain regions, respectively. (i-j) Quantification of the mean fluorescence intensity of TRd3 along the perivascular space and the interstitium adjacent to the vessels under the ventral surface of the hypothalamus of the both genotypes. Shades and error bars represent standard deviation. Data in Figure 2b and c were analyzed by repeated-measures ANOVA, N = 4 per group. Data in Figure 2f,h,i and j were analyzed by Student’s t-test. p-Values shown are comparisons between WT and Aqp4-/-. ns: not significant (Figure 2—source data 1).

https://doi.org/10.7554/eLife.40070.003
RIKEN: Aqp4 -/- mice display compromised CSF tracer infiltration under ketamine-xylazine anesthesia.

(a) Schematic diagram for CM injection of BDA tracer (left) and experiment schedule (right). (b) Examples of SA-enhanced BDA distribution 30 min after CM injection. Slices at an anterior-posterior position of bregma are presented for WT (upper) and Aqp4-/- (lower) mice. Depth profile is calculated for the cortical position 3 mm lateral to the midline. (c) Mean depth profiles of SA-enhanced BDA signals for WT (black, N = 4) and Aqp4-/- (green, N = 9) mice. (d) Mean SA-enhanced BDA signal intensities (3 mm lateral to the midline, depths 0–800 µm) along anterior-posterior positions for WT (black) and Aqp4-/- (green) mice. Shades and error bars represent SEM. *p < 0.05, **p < 0.01, t-test (Figure 3—source data 1).

https://doi.org/10.7554/eLife.40070.005
UNC: CSF tracer influx is decreased in Aqp4 KO mice.

(a) Coronal sections from a C57BL/6 wild-type mouse (WT), CD1 background strain control (Aqp4+/+), and Aqp4 KO mice (Aqp4-/-) showing a fluorescent CSF tracer, BSA-647 and co-labeling with DAPI. Scale bar: 1 mm (b) Mean pixel intensity in arbitrary units (A.U.) for six brain sections of each mouse for all three groups. n = 3 (WT), 7 (Aqp4+/+), 6 (Aqp4-/-). One-way ANOVA Tukey’s multiple comparisons test, Interaction term: p = 0.0110, F = 6.512, ns: not significant. (c) Diagram showing the anterior-posterior range of the quantified coronal sections relative to bregma from (b). (d) Quantification of the slices shown in (c) + 1.2 to −1.8 mm from bregma. Repeated measures two-way ANOVA with Tukey’s multiple comparisons test, Interaction term: p = 0.038, F = 2.085, p values shown are comparisons of WT and Aqp4+/+ vs. Aqp4-/-. (e) Diagram depicting the ROIs included in the regional analysis of brain slices at +0.6 mm from bregma. CPu: caudoputamen; HT: hypothalamus; BF: basal forebrain; DC: dorsal cortex; LC: lateral cortex; VC: ventral cortex. (f) Mean pixel intensity of brain regions shown in (e) for coronal sections + 0.6 mm from bregma. Repeated measures two-way ANOVA Tukey’s multiple comparisons test, Interaction term: p < 0.0001, F = 8.109. Data is presented as mean ±SEM (Figure 4—source data 1).

https://doi.org/10.7554/eLife.40070.007
URMC: Glymphatic influx of CSF tracer is facilitated by AQP4.

(a) Representative images from an in vivo transcranial optical imaging experiment of a WT control mice (Aqp4+/+) and Aqp4 KO (Aqp4-/-) mice starting at 15 min after intracisternal delivery of 10 µl (2 µl/min) of a 66 kDa fluorescent tracer, BSA-647. Scale bar: 2 mm. (b) Mean pixel intensity (MPI) of BSA-647 over a 30 min experiment, imaging was started at the beginning of the injection. Two-way repeated measures ANOVA with Sidak’s multiple comparisons test, overall model *p = 0.0329, multiple comparisons *p < 0.05; n = 6–7/group. (c) Coronal sections collected 30 min after intracisternal injection of two varying molecular size tracers, 3 kDa Texas Red dextran and BSA-647. Mean pixel intensity for a (d) 3 kDa dextran and (e) BSA from six coronal sections between +1.2 and −1.8 mm from bregma for each animal. Two-tailed unpaired t-test, **p = 0.0097, ****p < 0.0001; n = 6–8/group. (f) Coronal section from an Aqp4+/+ mouse stained and imaged for AQP4 (magenta) and BSA-647 tracer (cyan). (g) Mean pixel intensity from a replicate experiment quantified the same as (e). Two-tailed unpaired t-test, *p = 0.0335, n = 4–6/group. (h) Mean tracer penetration depth profiles normalized to the fluorescence at the pial surface of a coronal section at bregma from the set of experiments displayed in (g). The line was placed orthogonal to the cortical surface at the most dorsal position where tracer could be found at the pial surface (f). Tracer depth in the WT mice was measured at the same position as the KO mouse. Two-way repeated measures ANOVA with Sidak’s multiple comparisons test, overall model ****p < 0.0001, multiple comparisons ****p < 0.0001, n = 4–6/group. (i) Cortical depth at which the fluorescence at the surface decreases by half for the profiles in (h). Two-tailed unpaired t-test, *p = 0.0161, n = 4–6/group. (j) Relative copy numbers (RCN) of the Aqp4 gene locus and the excision sequence (k) showing successful deletion of Aqp4 exon 1–3. RCN was quantified by qPCR for mice used in the experiment in (g). Data expressed as mean ±SEM (Figure 5—source data 1).

https://doi.org/10.7554/eLife.40070.009
Deletion of the adapter protein α-syntrophin impairs AQP4 perivascular localization, and CSF influx into the brain parenchyma.

Dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) was acquired on a 11.75 preclinical MRI scanner, and was used to characterize the effect of α-syntrophin deletion on gaditeridol influx into the brain. Representative images of AQP4 perivascular localization in wild-type mice (a), and the loss of perivascular localization of AQP4 seen in the Snta1-/- mice (b). Scale bar: 50 µm, inset scale bar: 10 µm. (c-h) Coronal slice of T1-weighted images acquired by DCE-MRI demonstrate the reduced influx of gaditeridol contrast agent into the parenchyma in Snta1-/- mice relative to wild-type mice at 30 and 60 min. Scale bar: 1 mm. (i-n) Quantification of T1 weighted signal in various brain subregions normalized to baseline at each time point. Traces for each individual animal are presented (lines) along with the summary statistics (mean ±SEM, two-way ANOVA). WT n = 5, ASYNKO n = 7. CTx = cortex (p = 0.0035) Hip = hippocampus (p = 0.0003) Subcortical = subcortical regions (p = 0.0185) 3V = 3rd Ventricle (p = 0.0284) Total (p = 0.0085) (Figure 6—source data 1).

https://doi.org/10.7554/eLife.40070.011
A unilateral intrastriatal injection reduces global glymphatic function.

(a) KX-anesthesized mice received an injection into striatum. After injection, the glass capillary was slowly removed and the skull sealed with silicone elastomer. One hour later, mice received an intracisternal injection of a 66 kDa BSA-647 tracer and brains were removed and drop fixed 60 min later. (b) Control mice only received the intracisternal injection but not the intrastriatal injection. Mice that received an intrastriatal injection were injected with either HiLyte 488-amyloid-β1-40 or an aCSF sham. (c) Global glymphatic tracer influx was quantified from a total of six coronal sections between +1.2 and −1.8 mm from bregma for each animal. Ordinary one-way ANOVA posthoc Tukey’s multiple comparisons test, Control vs. Aβ-injected: ***p = 0.0006; Control vs. Sham: ***p = 0.0001; n = 6–8/group (Figure 7—source data 1).

https://doi.org/10.7554/eLife.40070.013

NMU: reduced influx of a fluorescent CSF tracer in Aqp4 KO mice

We injected the fluorescent tracer Texas Red-dextran (3 kD, TRd3) intracisternally, as previously described (Iliff et al., 2012). We first compared overall penetration of TRd3 into the dorsal and ventral surfaces of the whole-brain between WT and Aqp4 KO (Aqp4-/-) mice using ex vivo near infrared fluorescence imaging. Quantification of the mean integrated optical density (MIOD) of TRd3 on the dorsal and ventral brain surface along anterior-posterior position of bregma showed a significant reduction in Aqp4 KO mice compared with WT mice (Figure 2a–c). Interestingly, CSF tracer entry into the ventral brain was higher than into the dorsal brain. We also compared penetration of TRd3 into the brain on the serial coronal forebrain slides (+1.7 to −0.7 mm from bregma) between the two genotypes. The percentage area of whole-slice fluorescence was significantly reduced in Aqp4 KO mice compared to WT controls (Figure 2d,f). Subregional quantification on the coronal section at the level of 0.5 mm anterior to the bregma showed that penetration of CSF tracer was high in the ventral and lateral brain surface of WT mice, but was comparable in the dorsal surface of the brain with Aqp4 KO mice (Figure 2g–h), further supporting difference in brain regions of CSF tracer influx. Specifically, Aqp4 KO markedly impaired the influx of TR-d3 into both the perivascular space and the brain parenchyma in the hypothalamus, one of the brain regions with the highest expression of AQP4 (Fan et al., 2005; Nielsen et al., 1997) (Figure 2e). Quantification of the intensity of TR-d3 as a function of the distance from the brain surface showed rapid decay of tracer with increasing distance from the ventral surface of the hypothalamus in Aqp4 KO mice. The tracer was almost undetectable at 500 µm below the brain surface in both the perivascular space and in the brain parenchyma in Aqp4 KO mice. By contrast, the TR-d3 fluorescence intensity remained at 90% of the pial surface intensity to a depth of 500 µm in the perivascular space as well as in the adjacent parenchyma in WT mice (Figure 2i–j). Taken together, these results replicate the previous finding that AQP4 facilitates the transfer of intracisternally injected TR-d3 from the CSF into the parenchyma (Iliff et al., 2012).

RIKEN: histological enhancement of CSF tracer reveals compromised parenchymal tracer infiltration in Aqp4 KO mice

The RIKEN and Keio group used another strain of Aqp4 KO mice, in which the exon 1 of Aqp4 is replaced with eGFP. In this experiment, biotinylated dextran amine (BDA, 0.5%, 70 kDa) was injected under deep ketamine-xylazine (KX) anesthesia (Figure 3a). Because Aqp4 KO mice express eGFP, we used Alexa 594-conjugated streptavidin (SA) to visualize BDA. This approach has an advantage that BDA signals are amplified by the histological processing and clearly distinguished from intrinsic fluorescence (i.e. before histological processing). As a result, BDA distribution appeared less extensive in the cortex of Aqp4 KO mice (Figure 3b). Quantification of SA-enhanced BDA signals in the cortical parenchyma showed depth-dependent profiles that decayed with cortical depth after the initial peak for both wild type and Aqp4 KO mice. Notably, BDA signals in the cortex of Aqp4 KO mice are of lower intensity and penetrated less into the parenchyma at an anterior-posterior position of bregma (Figure 3c). Next, we examined if the compromised BDA penetration in Aqp4 KO mice is generally observed across the anterior-posterior extent of the cortex. Indeed, Figure 3d shows that BDA distribution in Aqp4 KO cortex is broadly compromised.

UNC: CSF tracer influx is decreased in Aqp4 KO mice compared to background strain controls and wild-type mice

We assessed the reproducibility of the experiments performed in Smith et al. (2017) by using the same Aqp4 KO mouse line but employing the original methodology reported by Iliff et al. (2012). In order to determine if the background strains of the different Aqp4 KO lines have an effect on CSF inflow, we included both CD1 and C57BL/6 controls. We compared the intraparenchymal entry of Alexa Fluor 647-conjugated bovine serum albumin (66 kDa, BSA-647) delivered into cisterna magna of ketamine/xylazine-anesthetized mice. Thirty minutes after the injection, the brains were harvested, and tracer distribution evaluated in coronal sections as described previously (Iliff et al., 2012). Parenchymal distribution of tracer showed a significant reduction in Aqp4 KO mice compared with both CD1 mice (Aqp4+/+) and wild-type C57BL/6 (WT) mice (Figure 4a–b). No significant difference in global tracer influx was seen between the two control groups. To evaluate if CSF influx was globally decreased or only limited to a particular brain region, we analyzed sections at different anterior-posterior coordinates as analyzed in Smith et al. (2017) (Figure 4c). Differences in CSF influx between Aqp4 KO and controls was most notable in coronal sections anterior to bregma (Figure 4d), perhaps reflecting that these sections include the cortical segment of the middle cerebral artery, a main site of CSF influx. Regional quantification showed that this difference was the result of decreased tracer penetration in the hypothalamus and areas of the basal forebrain (Figure 4e–f). Interestingly, CSF tracer entry along the ventral and lateral cortex was highest in the WT compared to the Aqp4+/+ control (Figure 4f) suggesting possible strain-dependent differences. We conclude that it is possible to replicate the original findings presented in Iliff et al. in the mouse line reported by Smith et al. when using similar methodology.

URMC: cerebrospinal fluid entry to brain occurs along the glymphatic pathway and is facilitated by the presence of AQP4 water channels

To evaluate the entry pathways of CSF to the brain we infused Alexa 647-conjugated bovine serum albumin (BSA) and a Texas Red 3 kDa dextran into the cisterna magna of anesthetized WT and Aqp4 KO mice. In vivo transcranial optical imaging (Plog et al., 2018) of the far-red BSA-647 showed more CSF influx in the WT compared to the KO (Figure 5a and b). Tracer evidently entered the brain parenchyma through a network of perivascular spaces of the large cerebral arteries on the pial surface. More tracer could be found on the dorsal cortical surface of WT mice compared to Aqp4 KO mice. After 30 min, brains were extracted and immersion-fixed. The tracer distribution was quantified in coronal sections of both WT and Aqp4 null mice. The KO mice had less influx for both the 3 kDa dextran and the 66 kDa BSA compared to the WT (~32.7% and ~47.06%, respectively), despite the 22x difference in molecular weight (Figure 5c–e). In a replicate experiment where the injection volume and injection rate was half of the previous, Aqp4 deletion also reduced CSF entry through the glymphatic pathway by ~42.4% (Figure 5f and g). The relative suppression of CSF tracer influx observed is comparable across previously reported findings, despite use of different tracer molecules and injection paradigms (Iliff et al., 2012). The depth of tracer penetration from the cortical surface was quantified using a similar approach to that reported in Smith et al. (2017) (Figure 5f and h). Tracer penetration into brain was found to be ~2 fold higher in the WT than in the KO mice (Figure 5i). To ensure that the mice were in fact knockouts, mice were genotyped using qPCR for both the Aqp4 locus (Figure 5j) and the excision sequence (Figure 5k). The results confirmed that all mice were either homozygous for the Aqp4 locus (Aqp4+/+) or were homozygous knockouts (Aqp4-/-) possessing no copies of exon 1–3 of the Aqp4 gene.

OHSU: reduced glymphatic CSF influx in Snta1 KO mice with loss of polarized expression of AQP4 in vascular endfeet of astrocytes

To assess the role of perivascular astrocytic localization of AQP4 in mediating CSF flux into the brain parenchyma, we utilized the Snta1 KO mouse line. These mice lack expression of the adaptor protein α-syntrophin, which links AQP4 to the dystrophin associated complex and is critical for maintenance of the perivascular localization of AQP4 (Jessen et al., 2015). Immunolabeling of AQP4 illustrates the loss of AQP4 perivascular localization in the Snta1 KO mice (Figure 6a–b). Higher magnification imaging reveals that, while perivascular localization is lost, widespread AQP4 expression is still detectable by immunofluorescence, consistent with a previous characterization of the mouse line (Figure 6a–b insets (Neely et al., 2001)). We next sought to determine if parenchymal CSF influx kinetics are altered when perivascular AQP4 localization is lost. Here, we used dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) to measure the influx of the contrast agent gadoteridol (Gad) into brain parenchyma after intracisternal injection. We performed serial T1-weighted imaging at 10 min intervals following administration of the contrast agent (Figure 6c–h). At 30 min after the start of the injection, elevated levels of Gad were detected in both WT mice and Snta1 KO mice, particularly along the ventral surface of the brain (Figure 6e–f), but by 60 min Gad signal within the parenchyma of Snta1 KO mice was significantly reduced compared to wild type mice (Figure 6g,h,n). Regional assessment revealed decreased signal in Snta1 KO mice across brain regions, which was most pronounced in cortex and hippocampus compared to subcortical brain regions and within the ventricles (Figure 6j–m). Taken together, these data illustrate a role for α-syntrophin in facilitating CSF-ISF exchange and, by extension, support the reported role of AQP4 in this process. Furthermore, these results suggest that the perivascular localization of AQP4 contributes to the kinetics of CSF influx into the brain parenchyma.

Insertion of an injection pipette in striatum is linked to a global suppression of CSF influx

Smith et al. (2017) questioned the existence of bulk flow within the brain parenchyma based on imaging tracer dispersion following local injections in cortex or striatum. The tracers were delivered by glass pipettes inserted through a cranial burr hole. The surgical procedure associated with insertion of glass pipettes is obviously traumatic. Since acute and chronic traumatic injury markedly suppress CSF influx (Iliff et al., 2014), we questioned the validity of studying interstitial tracer dispersion using an invasive approach. To this end, we compared CSF tracer (BSA-647, 66 kDa) influx in three groups of mice anesthetized with KX: (i) controls with no surgery beyond CM injection, (ii) insertion of a glass pipette (12 µm diameter tip) and injection of 20 nl (4 nl/min) HiLyte 488-amyloid-β1-40 in aCSF, or (iii) same as group two but HiLyte 488-amyloid-β1-40 was not added to the 20 nl aCSF (Figure 1f). One hour later, the BSA-647 tracer was injected into cisterna magna and allowed to circulate for 1 hr before the brain was harvested, immersion-fixed, and the distribution of BSA-647 analyzed in coronal brain sections (Figure 7a–b). The analysis showed that the two groups exposed to invasive surgery exhibited markedly suppressed CSF tracer influx compared with the control group (Figure 7c). The suppression of CSF tracer influx in animals with insertion of the glass pipettes was symmetric with no difference in tracer influx between the ipsilateral and the contralateral hemisphere (p > 0.05, 3–5 slices per animal, n = 7, Wilcoxon rank-sum test). These observations show that invasive procedures acutely suppresses brain-wide CSF influx and that tracer injection should not be used in the study of the glymphatic system.​

Meta-analysis

A potential cause of conflicting results could be due to varying methodology used in each study. We used meta-analysis to evaluate this heterogeneity and better understand the relationship between AQP4 and CSF-ISF transport using all of the available literature that evaluated tracer injections in global Aqp4 KO and wild-type mice. Studies were largely divided into two types: 1) intracisternal and 2) intracerebral tracer injections (Figure 8a–b), a separate meta-analysis was conducted for each of these two main experimental approaches.

Evidence evaluating the role of AQP4 in CSF influx and ISF efflux.

(a) Meta-analysis from experiments that delivered either fluorescence- or radio-labeled tracers into the cisterna magna of both Aqp4 KO and wild-type rodents. (b) Meta-analysis from studies that delivered intracerebral tracers to evaluate clearance or transport of tracers out of the brain. p-Value is from the overall random effects model. Data in forest plots presented as standardized mean difference (SMD) with a 95% confidence interval (CI). Th: thalamus; Cn: caudate nucleus.

https://doi.org/10.7554/eLife.40070.015

CSF influx

The search for studies that delivered tracers into the cisterna magna to evaluate CSF entry yielded a total of 13 eligible independent datasets from six different laboratories (Supplementary file 1). Experiments were done in a total of 144 mice (92.4%) and rats (7.6%) from five different Aqp4 KO lines (UIO: 50.7%; UCSF mouse: 25.7%; RIKEN: 9.03%, UCSF rat: 7.64%, and NMU: 6.94%) with a wide age range (6–24 weeks), the age of the rat line was not reported. Results were obtained using three different anesthetic protocols (ketamine/xylazine: 68.8%; tribromoethanol: 24.3%; chloral hydrate: 6.94%) and it was not possible to determine sex due to variable reporting. Experiments consisted of the delivery of fluorescently- (86.1%) and radioisotope-labeled (13.9%) tracers that ranged in size (4.5–70 kDa) and concentration. Tracer delivery varied in total volume (5–10 µl) and the injection rate (1–2 µl/min) while experiment duration ranged from 15 to 30 min. Standardized mean differences (SMD) were used in the meta-analysis to account for different outcome measures (e.g. fluorescence intensity, thresholded area, % injected radiation), and demonstrated a significant overall decrease in tracer entry in Aqp4 KO mice compared to controls (SMD with 95% confidence interval: −1.70 [−2.49; −0.90]; p<0.0001) with a high degree of interstudy heterogeneity (I2 = 73.1%; p<0.0001; Figure 8a). Leave-one-out and meta-regression sensitivity analyses were used to explore the sources of heterogeneity. As expected based on the direction and magnitude of the effect estimates reported in the studies, the high degree of heterogeneity was primarily caused by three datasets extracted from Smith et al. We could not confidently determine from the text of these studies whether these experimental figures were based on separate, independent experiments, or whether they were data collected in the same set of mice. Due to this uncertainty, we conducted leave-one-out analyses that would recapitulate all possible scenarios. The analysis showed that when these data sets were excluded, there was a reduction in heterogeneity (I2 = 65.7%, 70.3%, and 71.4%; p = 0.0007, p = 0.0001 and p < 0.0001). When all three data sets from this publication were excluded, the heterogeneity becomes non-significant (I2 = 0.0%, p = 0.4622). Due to the relevance of this study, we included all data sets in the final results reported (Supplementary file 1). Using meta-regression to further explore sources of heterogeneity, we tested whether the: KO line, age, anesthesia type, tracer properties, injection paradigm, experiment duration, and detection method were significant covariates (Supplementary file 1). The KO line used (p = 0.0001, Test of Moderators), the age of the mice (p = 0.0006), the anesthetic protocol (p < 0.0001), and the injection rate (p < 0.0001) explained a significant proportion of the observed variance between the datasets. The variables that explained the largest proportion of heterogeneity were anesthesia type and injection rate. The three data sets that reported null results used tribromoethanol (Avertin), an anesthetic that was not used in any other studies (all of which found a significant association). The same three data sets were also the only ones to report variable injection rates (10 µl delivered over 5–7 min) in comparison with the remaining studies that used exact injection rates. Thus, it is possible that the choice of anesthesia type and injection rate could explain the heterogeneity. Animal age was also a significant source of heterogeneity, accounting for 85.28% of the variance between studies. This may be in part explained by the fact that two out of the three data sets that demonstrated no association were completed in mice that were older than the mice used in the majority of other studies and reported the largest age range (12–24 weeks). Although it is not possible to determine the exact role that these covariates played in the observed results, both anesthesia and aging have been shown to modify the degree of intracisternal tracer entry, providing another potential explanation for the observed heterogeneity (Benveniste et al., 2017; Kress et al., 2014). Similarly, the heterogeneity stemming from the KO line may also be explained by the fact that the majority of UCSF KO mice (72.9%) were in the null studies. Interestingly, the fourth study (UNC, this report) that utilized the UCSF KO mouse line, but an alternate anesthetic (ketamine/xylazine) observed a significant effect. Unfortunately, due to low study number we were unable to include two covariates (e.g. anesthesia type and KO line) in the meta-regression to test this interaction. Additional covariates including tracer properties, the injection paradigm, and the detection method were not significant sources of heterogeneity suggesting that multiple experimental methodologies should be able to observe an effect between KO and control.

ISF efflux

There were nine eligible studies produced by four independent research groups that injected tracers or contrast agents into the brain of Aqp4 KO or wild-type rodents in order to evaluate brain clearance or ISF efflux (Supplementary file 2). These experiments were conducted in a total of 87 rodents from both mouse (72.4%) and rat (27.6%) KO lines. There were two murine lines (UIO: 57.5%; UCSF: 14.9%) and one rat line (Beijing: 27.6%) generated by three separate groups. Age of the mouse lines ranged broadly (0–24 weeks), while the age of the rat line could not be confirmed but a weight range of 250–300 g was reported. Four anesthetic protocols were used for these experiments (ketamine/xylazine: 57.5%; hypothermia: 6.9%; tribromoethanol: 8.0%; sodium pentobarbital: 27.6%). Fluorescence-conjugated tracers (14.9%), radio-labeled (57.5%) tracers, and gadolinium-based contrast agents (27.6%) that ranged in size (0.18–45 kDa), concentration, and where they were delivered within the brain (striatum: 40.2%; cortex: 39.1%; thalamus: 13.8%; lateral ventricle: 6.9%). Tracer delivery varied in total volume (0.02–3 µl) and injection rate (0.017–3 µl/min), differing based on the injection site or whether the experiment was done in mice or rats. Experiment durations ranged between 45 and 240 min and one of the studies (Smith et al.) did not report this. Results from the overall meta-analysis showed a significant decrease in the transport of the injected substance in Aqp4 KO mice relative to controls (−1.88 [-2.88; −0.87]; p = 0.0003) with high heterogeneity (I2 = 70.0%, p = 0.0008; Figure 8b). Similar to the previous model, leave-one-out analysis of the two null datasets reduced the heterogeneity (I2 = 60.1%; p = 0.0143 and I2 = 65.3%; p = 0.0053, respectively) and removing both datasets made the heterogeneity non-significant (I2 = 17.2%; p = 0.299). The same covariates used in the first model were used for meta-regression analysis, in addition to an injection site variable. Meta-regression was unable to identify significant sources of heterogeneity; most likely because of two datasets from Teng et al. that found a difference in thalamus, but not in striatum, in the same rat line using identical methodology (Teng et al., 2018). Despite the potential biological significance of this, the study was the only one to use rats, MRI as the detection method, and had an experiment duration that was 4-fold longer than the other studies. Therefore, we decided to run a reduced model on all available mouse experiments using fluorescence or radioactivity detection methods. The meta-analysis was still highly significant and showed decreased clearance in the Aqp4 KO mouse lines (−1.92 [-3.02; −0.82]; p = 0.0006) with a lower degree of heterogeneity (I2 = 64.1%, p = 0.0103). The KO line used (p = 0.0205), age (p = 0.0075), the anesthetic protocol (p = 0.0075), the injection rate (p = 0.0056), the experiment duration (p = 0.0075), and the detection method (p = 0.0205) explained a significant proportion of the observed variance between the datasets. Similar to previous results, tribromoethanol again seemed to function as a proxy for the null study from Smith et al. since none of the others used this anesthetic. The mouse KO line from UCSF explained a significant proportion of heterogeneity compared to the other lines, likely because the null dataset accounted for half of the studies that used this line. Age of the experimental animals was also a significant source of heterogeneity and can be explained by the wide age range (0–24 weeks) found between the studies but the lack of specific age reporting could also potentially explain this. Injection rate accounted for the largest proportion of heterogeneity in this analysis. The injection rate used by Smith et al. (5 × 100 ms pulses at 1 Hz and 10 psi), was highly significant in the meta-regression model compared to the reference injection rate (0.017 µl/min, p = 0.0007), whereas all other injection paradigms were not significantly different. The Smith paradigm was therefore the driver of the explanatory value of the injection rate covariate in the meta-regression model. Similar to anesthesia type, this variable also functioned as a proxy for the null study since all other datasets reported an exact volumetric injection rate. The reported injection protocol was used for experiments performed in both mice and rats despite different injection volumes (~20 nl vs. 30,000 nl) suggesting that the rate varied significantly between experiments. The time the tracer was allowed to circulate after the injection (experiment duration) was also a highly significant source of variation between studies. The experiment duration that accounted for all the heterogeneity was the unreported time from the null study. Only one separate study evaluated a time point that did not report a significant difference between KO and control (30 min; Iliff et al., 2012). However, this study did report a difference at 60 min after injection, so it is possible that if Smith et al. used shorter experimental durations this could account for the observed heterogeneity (Supplementary file 2) (Iliff et al., 2012). The detection method (i.e. fluorescence and radioactivity) used was also found to be a significant source of heterogeneity. The largest contributor to heterogeneity were fluorescent detection methods, mostly likely due to the fact that the study that obtained a null result used this approach despite most studies using radiolabeled tracers. The remaining covariates: tracer properties, the injection volume, and the injection site were not significant sources of heterogeneity in this analysis.

Discussion

This study evaluated the role of the astrocytic water channel AQP4 in the dispersion of fluorescent tracers or contrast agents by five independent groups. Four of the groups (NMU, RIKEN, UNC, URMC) injected CSF tracers in cisterna magna and compared the influx of the tracer in coronal sections prepared from Aqp4 KO and WT mice, whereas the OHSU group injected contrast agent in cisterna magna and then compared CSF transport by DCE-MRI in Snta1 KO and WT mice. Compilation of data from the six independently-conducted experiments documented that deletion of Aqp4 (Aqp4 KO mice) or mis-localization of AQP4 (through Snta1 gene deletion) suppressed influx of CSF tracers or contrast agents. In fact, a meta analysis that included all reports published to date came to the same conclusion (Figure 8). Thus, our observations provide strong and concordant support for the glymphatic model in which AQP4 supports the perivascular influx of CSF and efflux of ISF (Iliff et al., 2012). Supporting evidence of this has also been shown in humans; two independent single nucleotide polymorphism (SNP) variant association studies identified multiple SNPs in the Aqp4 gene that were associated with cognitive decline in Alzheimer’s disease (AD) and increased Aβ burden in humans with reduced sleep quality (Burfeind, 2017; AIBL Research Group et al., 2018). A separate study by the OHSU group also found that reduced perivascular polarization of AQP4 was a significant predictor of AD status in postmortem human brains (Zeppenfeld et al., 2017).

Independent replication studies of biological phenomena are essential, both to confirm novel experimental findings, as well as to extend and refine initial observations and interpretations. The original report from the URMC group on the glymphatic system showed that deletion of astrocytic water channels suppressed both the influx of CSF tracers injected into the cisterna magna and the efflux of tracers infused directly into striatum (Iliff et al., 2012). Follow-up of that finding by the NMU group documented that Aβ accumulation and cerebral amyloid angiopathy were significantly increased in a double-hit murine model with Aqp4 deletion and Aβ over-expression (Xu et al., 2015). The URMC/OHSU groups subsequently reported that deletion of Aqp4 aggravated neurofibrillary pathology following TBI (Iliff et al., 2014), whereas the UNC group showed that deletion of Aqp4 significantly reduced clearance of AAV (Murlidharan et al., 2016). Thus, these reports conceptually address the same idea, although the replications entailed different approaches and methodologies. The dissenting report from Dr. Verkman’s laboratory belongs to the rarer category of replication studies, which (as their title states) seeks to test directly the glymphatic hypothesis and the AQP4-dependence of solute transport in brain. Since their study was intended to gauge the veracity of the original findings, it should be expected that it follow the experimental design and data analysis of the original study (Picho et al., 2016), and their methodology should be presented in sufficient detail to allow the reader to assess the fidelity of their replication (Brandt et al., 2014). However, Smith et al. (2017) failed to meet these requirements in several key respects. First, their methodologies deviate from the original report in critical ways. In particular, Iliff et al. used ketamine/xylazine (KX) anesthesia rather than tribromoethanol (Avertin) (Iliff et al., 2012). Xylazine is an α2 adrenergic agonist that blocks release of norepinephrine from the locus coeruleus projections across the neuraxis. This is an important point, because norepinephrine has been identified as the key humoral regulator of glymphatic solute transport (Xie et al., 2013), and norepinephrine attenuates glymphatic function. However, with the exception of the photo-bleaching experiments carried out only in WT mice, Smith et al. (2017) anesthetized their mice with Avertin, an injectable anesthetic with an unknown mechanism of action. Avertin is not approved for use in several European countries and is restricted to terminal procedures in the US (Arras et al., 2001; Meyer and Fish, 2005). The importance of the anesthetic agent in the present context was first highlighted by Groothuis et al. (2007), who reported that parenchymal solute efflux was 100-fold slower in rats anesthetized with pentobarbital than those that received ketamine/xylazine. Anesthesia-dependent effects were also noted by Benveniste et al., that reported a marked difference in influx of contrast agents in isoflurane-anesthetized animals versus low-dose isoflurane plus dexmedetomidine (an α2 agonist) (Benveniste et al., 2017). The importance of anesthesia is supported by the experiments reported here: The UNC group used ketamine/xylazine anesthesia and observed AQP4-dependent glymphatic transport (Figure 4), despite using the same Aqp4 KO mouse line as in the Smith et al. (2017). In addition, the methodology of Smith et al. was not presented in sufficient detail to assess the fidelity of the replication. Key methodological details, such as the injection approach and rate,anesthesia, or the age of the animals were not provided for data collected in rats. Neither did the investigators assess whether Aqp4 was deleted in CNS albeit this was the first report utilizing Aqp4 KO rats. In fact, no such data have been published so far to our knowledge (Smith et al., 2017). For these reasons, the present study is not a replication study of Smith et al. (2017). Instead, we used the methodology reported in Iliff et al. (2012) and further optimized it in subsequent studies (Kress et al., 2014; Iliff et al., 2013a).

It is also important to note that the recovery period that follows general anesthesia differs from the awake state. A recent study reported that increasing anesthesia concentration decreased glymphatic function (Gakuba et al., 2018), as opposed to the original finding that the glymphatic system is maximally active during natural sleep or anesthesia (Xie et al., 2013). However, the recent study injected CSF tracers under inhalation anesthesia followed by a short recovery period (Gakuba et al., 2018). This is not a valid approach, because awakening from general anesthesia is associated with cognitive disturbances (Su et al., 2016; Paredes et al., 2016; Numan et al., 2017) and inhaled anesthetics in particular are linked to a high rate of delirium upon emergence from anesthesia (Brioni et al., 2017). Close evaluation of the imaging in Gakuba et al. shows a failed intracisternal injection, as enhancement of contrast agent is evident in the paraspinal muscles and extracranial spaces (Gakuba et al., 2018). In our experience, experimenter error during cisterna magna injections is a significant source of variability (Xavier et al., 2018). Thus, careful consideration of the anesthesic regimen and long periods of post-anesthesia recoveries are essential when studying the glymphatic system.

The initial study articulating the glymphatic model confirmed seminal work by Cserr and colleagues that interstitial tracers are cleared from brain tissue at rates that are independent of molecular weight (Iliff et al., 2012; Cserr et al., 1981). While these early studies support the role of bulk flow in the clearance of interstitial solutes from the brain interstitium, they lacked the spatial resolution necessary to resolve whether bulk flow was occurring throughout the entire brain interstitium or was restricted to a subset of compartments permissive to flow. A large body of indirect experimental evidence supports the occurrence of bulk flow in the periarterial space in the brain (see reviews by Hladky and Barrand, 2014 and Nicholson and Hrabětová, 2017). Smith et al. (2017) themselves provide such evidence, stating that 'occasionally, when the injected dextran entered the perivascular space of vessels near the injection site, both large and small dextrans traveled substantial distances away from the injection site’ (Smith et al., 2017). There is also indirect evidence that this flow is peristaltic, propelled by pulsations of the arterial walls (e.g. Hadaczek et al. (2006)). This mechanism has recently been confirmed by direct observations of pulsatile flow in periarterial spaces (Bedussi et al., 2018; Mestre et al., 2018). Using two-photon microscopy and particle-tracking velocimetry, Mestre et al. gathered hundreds of thousands of velocity measurements at sampling rates up to 30 Hz over half hour experiments, mapping the perivascular flow in space and time in great detail: their results show that the flow pulses in synchrony with the heartbeat, is driven by pulsations of the arterial wall, and produces a net (average) flow in the same direction as the blood flow. This flow is in accordance with several earlier fluid-dynamic models and experimental studies (Bilston et al., 2003; Schley et al., 2006; Wang and Olbricht, 2011; Rennels et al., 1985; Iliff et al., 2013b), and with in vivo dynamic imaging studies, including MRI experiments in human subjects, demonstrating the rapid distribution of CSF tracers into and through the brain interstitium along peri-arterial routes (Iliff et al., 2013a; Eide and Ringstad, 2015; Ringstad et al., 2017). In addition to perivascular spaces, white matter tracts also serve as permissive routes for interstitial solute and fluid movement (Ohata and Marmarou, 1992; Rosenberg and Kyner, 1980).

As a consequence of the now well-established bulk flow along periarterial spaces, there must be an outlet for this directional flow: conservation of mass - as expressed by the continuity equation - demands that this volumetric flow must continue through other parts of the glymphatic system – that is, in the neuropil and the perivenous spaces. Although the full volumetric net flow must pass through the neuropil, flow speeds will be lower there because of the much larger total cross-sectional area of the available extracellular flow channels, and due to the availability of the gap junction-coupled astroglial syncytium as an intracellular route for water and solute movement (Asgari et al., 2016). In fact, a meta-analysis that include the more than 200 Aqp4 KO mice and rats studied so far added further support to the finding that glymphatic influx and efflux are both highly dependent upon AQP4 expression (Figure 8 , Supplementary file 1 and 2). The meta-analysis identified the anesthesia protocol, wide age range of the experimental animals and the injection paradigm are potential causes for the conflicting data reported by Smith et al. However, it is important to note that glymphatic transport is affected by multiple pathways other than AQP4. For example, the sleep-wake cycle (Xie et al., 2013), brain injury in the setting of trauma or ischemia (Iliff et al., 2014; Wang et al., 2017), exercise (He et al., 2017; von Holstein-Rathlou et al., 2018; Yin et al., 2018), amyloid-β accumulation and acute amyloid-β toxicity (Xu et al., 2015; Peng et al., 2016), omega-3 fatty acids (Ren et al., 2017), plasma osmolarity (Plog et al., 2018), and aging (Kress et al., 2014) are important regulators of glymphatic transport. How AQP4 on a cellular level facilitates CSF-ISF exchange remains to be firmly established. The primary role of AQP channels is to lower the driving forces for fluid transport across the plasma membranes (Verkman et al., 2017). AQP4 is highly enriched in astrocytic endfeet facing the perivascular space and thus strategically positioned to facilitate CSF exchange with interstitial fluid. Accordingly, loss of the polarized expression of AQP4 in reactive astrocytes lead to a reduction in parenchymal CSF tracer influx (Iliff et al., 2014; Kress et al., 2014). Moreover, deletion of AQP4 potentiated the increases in intracranial pressure in a murine model of hydrocephalus consistent with the notion that AQP4 supports and facilitates intraparenchymal fluid flow (Bloch et al., 2006).

The claim by Smith and Verkman (2018) that solute transport in the parenchyma is primarily driven by diffusion rather than convection (bulk flow) is based on their finding of a dependence of this transport on the size of the solute particles. They claim that diffusive transport depends strongly on particle size, whereas convective transport does not, so long as the ratio of the particle size to the pore or channel size is less than about 0.5. In support of this claim, they cite a single reference (Dechadilok and Deen, 2009), which does not actually support their claim, but instead shows a significant dependence of flow resistance on particle size for much smaller particles. The influence of solutes or suspended particles on flow through narrow channels is a complicated problem in hydrodynamics, entailing many factors, including solute concentration, electrical charge, shape of the particle, channel geometry, tortuosity, and possible obstructions (Nicholson and Hrabětová, 2017). Hence, it would require a very sophisticated theoretical model to evaluate the role of bulk flow in the transport observed experimentally by Smith et al. (2017). Their findings are in general agreement with recent modeling studies suggesting that at the smallest microscopic scales, encompassing small numbers of vessels and the interposed neuropil, diffusion is sufficient to account for observed solute movement (Asgari et al., 2016; Jin et al., 2016; Holter et al., 2017). However, it is important to note that these findings do not exclude a contribution of bulk flow within the entire interstitium, and do not determine whether the experimental approaches used to detect flow on the microscopic scale are sufficiently sensitive to detect bulk flow occurring at very low flow rates. Also, the majority of the studies contradicting the glymphatic model are based on theoretical modeling and not experimental evidence (Asgari et al., 2016; Jin et al., 2016; Faghih and Sharp, 2018). We suspected that the surgical procedures required for intraparenchymal tracer injections utilized in Smith et al. (2017) might suppress influx of CSF tracers. To formally test this potential caveat, CSF tracer influx was analyzed in mice in which a glass pipette was inserted in striatum 1 hr earlier (Figure 7). The analysis showed that insertion of a glass pipette suppressed CSF tracer influx significantly and independently of intrastriatal injection of aCSF with or without HiLyte 488-amyloid-β1-40 compared to non-surgical controls. We conclude that dispersion of tracers in CNS should not be studied using invasive procedures. Thus, as articulated by Holter et al. in their recent modeling study (Holter et al., 2017), there is a need to develop more appropriate technical approaches for evaluating slow interstitial bulk flow over long distances (Ratner et al., 2017), such as DCE-MRI in nonhuman primates or human subjects. Until then, the glymphatic model of CNS fluid and solute flow should properly include the rapid exchange of CSF with the brain ISF along perivascular spaces, with CSF principally entering along peri-arterial spaces and ISF draining towards the ventricular and CSF compartments along peri-venous pathways and along cranial and spinal nerves, supported by AQP4-dependent astroglial water transport. On this last element, the present findings from five independent groups do not concur with the negative findings reported by Smith et al. (2017). Instead, our compiled data provides strong and consistent support for our claim that solute transport in the rodent brain is facilitated by the polarized expression of AQP4 in astrocytic endfeet.

Materials and methods

Key resources table
Reagent type
(species)
or resource
DesignationSource or
reference
IdentifiersAdditional
information
Gene
(Mus musculus)
Aqp4MGI:107387
Gene
(Mus musculus)
Snta1NCBI_gene:
20648,
MGI:101772
Strain,
strain background
(Mus musculus)
Aqp4 null/eGFP
knock-in mice
RIKEN BRCRBRC06500;
SJL.Cg-Aqp4<tm1.1(GFP)
Mysi>/MysiRbrc;
CDB0758K-1
Strain,
strain background
(Mus musculus)
Aqp4 knockout
mice
PNAS 108,
846–851
(2011).
PMID: 21187412
OIU
Strain, strain
background
(Mus musculus)
C57BL/6NBeijing Vital River
Laboratory
Animal
Technology
Co., Ltd.
213
Strain, strain
background (Mus
musculus, C57Bl/6J)
male wild type (WT)Jackson
Laboratories,
000664
RRID:IMSR_JAX:000664
Strain, strain
background
(Mus musculus,
C57BL/6NCrl)
male wild type (WT)Charles River,
027
RRID:IMSR_CRL:27
Strain, strain
background
(Mus musculus,
C57Bl/6J)
male Snta1-/-Jackson
Laboratories,
012940,
PMID:10995443
RRID:MGI:2181419
Cell line
(Mus musculus)
E14K ESJ Neurosci Res.
2005 Nov 15;82
(4):
458–64.
PMID: 16237719
129P2/OlaHsd
Transfected
construct
(Mus musculus)
AQP4 replacement
targeting vector
J Neurosci Res.
2005
Nov 15;82(4):
458–64.

PMID: 16237719
Constructed
using
the positive–
negative
selection
cassettes
derived

from the
vectors
pPolII long neo bpA
and pXhoMC1TK,
containing the neoR
and HSVtk
genes, respectively.
Biological
sample
(Mus musculus)
CD1 blastocystsBeijing Vital River
Laboratory Animal
Technology Co., Ltd.
201
Antibodyanti-Aqp4
(rabbit polyclonal)
SigmaA5971,
RRID:AB_258270
(1:1000)
Antibodybiotinylated-
conjugated
goat anti-rabbit
IgG
(polyclonal)
Vector
Laboratories
BA-1000,
RRID:AB_2313606
(1:200)
AntibodyAQP4 (rabbit
primary Ab)
(polyclonal)
MilliporeAB3594,
RRID:AB_91530
(1:400–500)
AntibodyAlexa Fluor 594
(donkey anti-rabbit
secondary Ab)
(polyclonal)
ThermoFisherA21207,
RRID:AB_141637
(1:500)
AntibodyCy3-conjugated
donkey
anti-rabbit
(polyclonal)
Jackson Immuno
Research
711-165-152,
RRID:AB_2307443
(1:500)
Sequence-
based reagent
oligonucleotide
primers
of Aqp4 and
excision
sequence
TransnetyxProbe ID: Aqp4-2
WT and Aqp4-2 KO
Sequence-
based reagent
oligonucleotide
primers
of Aqp4 and Neo
J Neurosci Res.
2005
Nov 15;82
(4)
:458–64.
PMID: 16237719
AQP4 forward
primer, ACC
ATA AAC TGG
GGT GGC TCA
G; WT AQP4
reverse primer,
TAG AGG ATG
CCG GCT CCA
ATG A; and Neo,
CAC CGC TGA
ATA TGC ATA
AGG CA.
Peptide,
recombinant
protein
Beta - Amyloid
(1 - 40), HiLyte
 Fluor 488 - labeled,
Human
AnaspecAS-60491–010.5% in aCSF
Commercial
assay or kit
Kwik-CastWorld Precision
Instruments
KWIK-CAST
Commercial
assay or kit
Elite ABC KitVector
Laboratories
PK-7200
Chemical
compound,
drug
BDAMolecular
Probes/Thermo
Fisher
D1957; BDA-70,000
Chemical
compound,
drug
Alexa-594-conjugated
streptavidin (SA)
Molecular
Probes/Thermo
Fisher
S11227(1:1000)
Chemical
compound,
drug
Alexa Fluor
647 conjugated
bovine serum albumin
Molecular
Probes/Thermo
Fisher
A347850.5–1% in aCSF
Chemical
compound,
drug
Texas Red conjugated
dextran, 3000 MW,
paraformaldehyde,
chloral hydrate
Invitrogen,
Sigma-Aldrich,
Sigma-Aldrich
D3328, 158127, 47335 U
Software,
algorithm
ImageJ softwarehttps://imagej.nih.gov/ij/index.htmlRRID:SCR_003070
Software,
algorithm
FIJIRRID:SCR_002285
Software,
algorithm
MetaMorph BasicMolecular
Devices
RRID:SCR_002368
Software,
algorithm
MatlabMathworksRRID:SCR_001622
Software,
algorithm
cellSensOlympusRRID:SCR_016238
Otherartificial CSFSigma-AldrichAll componets
in
ACSF (concentrations
in mM):
126.0 NaCl,
3 KCl, 2 MgSO4,
10.0 dextrose,
26.0 NaHCO3,
1.25 NaH2PO4,
2 CaCl2 are
purchased
from Sigma-
Aldrich

Generation of transgenic mouse lines

NMU

Request a detailed protocol

Aqp4 KO mice were generated by Dr. Fan Yan by targeted gene disruption as described previously (Fan et al., 2005). In brief, an AQP4 replacement targeting vector was constructed using the positive-negative selection cassettes derived from the vectors pPolII long neo bpA and pXhoMC1TK, containing the neoR and HSVtk genes, respectively. The targeting construct was linearized with NotI and introduced into E14K ES cells by electroporation. ES cell clones that were G418/gancyclovir-resistant were isolated, amplified, and screened for targeting fidelity using Southern blot analysis. Cells from two targeted clones were microinjected into CD1 blastocysts and implanted into pseudopregnant recipients (Fan et al., 2005). Five-month old Aqp4 KO and WT mice were used in the present study.

RIKEN

Request a detailed protocol

Aqp4 null/eGFP knock-in mice (SJL.Cg-Aqp4<tm1.1(GFP)Mysi>/MysiRbrc) were generated as described previously (Ikeshima-Kataoka et al., 2013). Briefly, knockout of AQP4 was accomplished by replacing the 250 nucleotides in exon 1 (corresponding to the region from Ser 18 to Thr 101) with eGFP cDNA and a PGK-neomycin cassette flanked by flippase recombination target sequences. The targeting vector was electroporated into TT2 ES cells derived from a C57BL/6 female and a CBA male. The PGK-neomycin cassette was removed by crossing with B6-Tg (CAG-FLPe) 37 (Kanki et al., 2006). The resultant mice were backcrossed with the SJL/J strain (Charles River, Japan) for more than 12 times (RIKEN CDB accession no. CDB0758K-1; http://www.clst.riken.jp/arg/mutant%20mice%20list.html). Both B6-Tg (CAG-FLPe) 37 and SJL.Cg-Aqp4<tm1.1(GFP)Mysi>/MysiRbrc >mice are provided by the RIKEN BRC through the National Bio-Resource Project of MEXT, Japan (accession no. RBRC01835 and RBRC06500, respectively).

UNC

Request a detailed protocol

The constitutive Aqp4 KO mouse model was generated by Dr. Alan Verkman (Ma et al., 1997). In brief, Aqp4 KO mice were generated on a CD1 background, following the construction of a targeting vector for homologous recombination of a 7 kb SacI AQP4 genomic fragment, in which part of the exon one coding sequence was deleted. All animal experiments reported in this study were conducted on Aqp4 KO mice (B6/129), B6/129 controls, and wild-type C57BL/6 mice.

URMC

Request a detailed protocol

Aqp4 KO mice were generated by Dr. Ole Petter Ottersen using GenOway technique, as described previously (Thrane et al., 2011). Their strategy involved cloning and sequencing of a targeted region of the murine Aqp4 gene in a 129/Sv genetic background. Identification of a targeted locus of the Aqp4 gene permitted the delete exons 1–3 to avoid any expression of putative splice variants. Hence, a flippase recognition target (FRT)-neomycin-FRT-LoxP–validated cassette was inserted downstream of exon 3, and a LoxP site was inserted upstream of exon 1 (Thrane et al., 2011). The mice were backcrossed for 20 + generations with C57BL/6N mice prior to experimentation.

OHSU

Request a detailed protocol

The Snta1 KO mouse line was generated by Dr. Stanley Froehner (Adams et al., 2000), and was purchased through The Jackson Laboratory (Snta1tm1Scf/J, #012940). In brief, exon 1 of the mouse Snta1 gene was replaced with a neomycin resistance cassette. The resulting construct was electroporated into 129P2/OlaHsd-derived E14 embryonic, and these cells were then injected into C57BL/6J blastocysts. The resulting mice were bred with C57BL/6J mice for at least 11 generations before the colony was established. Adult male and female mice (WT 15.4 ± 1.6 weeks, Snta1 KO 14.7 ± 1.6 weeks) were used in these studies.

Injection, imaging and analysis of CSF tracers or contrast agents

NMU

Request a detailed protocol
Intracisternal tracer injections

Intracisternal injections of Texas Red conjugated dextran (TRd3; 3 kDa, Invitrogen) were adapted from a previous report (Iliff et al., 2012). In brief, the mice were anesthetized by a 4% chloral hydrate solution. A ten-microliter volume of 0.5% TRd3 dissolved in artificial CSF was infused into the cisterna magna through a 50 μL syringe mounted with a 27-gauge needle (Hamilton, Reno, NV, USA), connected to a constant current syringe pump (TJ-2A/L07-2A, Suzhou Wen Hao Chip Technology Co. Ltd., Jiangsu, China). The intracisternal injection was carried out at a rate of 2 µL/min. Thirty min after the start of injection, the still deeply anesthetized animals were perfused trans-cardially with 4% paraformaldehyde (PFA), and the brains removed and post-fixed in the same fixative for 24 hr.

Tissue processing and image analysis of fluorescent tracer

A subset of WT and Aqp4 KO brains (N = 4 per genotype) were scanned by an ex vivo near infrared (NIR) fluorescence imaging system (Azure biosystems c600, CA, USA). Briefly, the ventral and dorsal brains were imaged (excitation length 660 nm, exposure time: 50 ms), as well as the corresponding bright-field images (Supplementary file 3). The fluorescence and bright-field images were superimposed using Adobe Photoshop 8.0. Based on the Mouse Brain Atlas, fluorescence was measured along anterior-posterior axis based on the localization of bregma using NIH Image J software (Figure 2a). The mean integrated optical density (MIOD) of TRd3 at a total of 80 sites were sequentially measured from 4 mm anterior to 8 mm posterior to the bregma. In addition, the telencephalon from five mice per genotype was cut into 100 µm coronal sections using a vibratome (Leica, Wetzlar, Germany). Sections were mounted on gelatin-coated slides, washed, and cover-slipped with the slices coated in PBS/glycerol. Images of 6–8 coronal sections collected sequentially from 1.7 mm anterior to 0.7 mm posterior of the bregma were imaged by a Leica DM4000B digital microscope (Leica Microsystems, Wetzlar, Germany) under a 1.25x objective, with uniform exposure time, offset, and gain. The distribution of the tracer (TRd3) was quantified using NIH Image J software as described previously (Iliff et al., 2012). In brief, the fluorescent tracer coverage within the whole section was detected using the same threshold and constant settings for minimum and maximum intensities. Subregional analysis of CSF tracer distribution was analyzed in coronal section at the level of 0.5 mm anterior to the bregma. Tracer penetration along perivascular spaces and into brain parenchyma was quantified on high-magnification images of the hypothalamus, which were captured with 10x objective. The fluorescence intensity within the perivascular space was measured along large vessels (diameter >20 µm) extending 500 µm below the brain surface. Two or three separate vessels from each image were analyzed and the data averaged. The fluorescence intensity in the brain parenchyma was also determined along a corresponding linear region adjacent to the vessels (Figure 2e). The imaging and subsequent analysis was performed by an investigator who was blind to animal genotype.

RIKEN

Request a detailed protocol

Adult mice (postnatal 10 to 23 weeks old, mixed gender) were anesthetized by a ketamine-xylazine cocktail (70 mg/kg of ketamine and 10 mg/kg of xylazine) and fixed to a stereotaxic frame. The membrane of the cisterna magna was exposed by surgical procedures. A glass micropipette (tip diameter ~50 µm, preloaded with 0.5% biotinylated dextran amine in saline (BDA, 70 kDa, D1957, Molecular Probes), was carefully pierced through the cisterna magna membrane using a manipulator. A total of 10 µL of the BDA containing solution was infused at a rate of 1 µL/min using a syringe driver (KDS Legato Series, KD Scientific). Thereafter, two supplementary ketamine-xylazine administrations were made (30 min apart, 50% of the initial dose). Five minutes after the second supplementary anesthesia, Mice were transcardially perfused with fixative containing 4% paraformaldehyde in 0.1 M phosphate buffer. After one day of post-fixation in the same fixative, 60 µm slices were incubated in 0.1% triton-X TBS containing Alexa-594-conjugated SA (1:1000) for one hour. Images of SA-processed brain sections were acquired using a standard fluorescent microscope (Olympus BX51 equipped with a DP74 digital camera, Supplementary file 3). An offset, defined as the median of the top 1% of pixel intensities of an unprocessed brain section, is subtracted from the raw image of the SA-processed brain section. For depth profiles were calculated from the pial position of 3 mm from the midline using ImageJ.

UNC

Request a detailed protocol
Intracisternal tracer injections

The experiments were performed as described previously (Iliff et al., 2012). In brief, mice (15–18 weeks old) were anesthetized with ketamine/xylazine (100/10 mg/kg) and then a cannula was surgically implanted into the cisterna magna (Xavier et al., 2018). Afterwards, an Alexa Fluor 647-conjugated bovine serum albumin tracer was infused into the cistern (0.5% in aCSF; 10 µl at 1 µl/min for 10 min) and allowed to circulate for 30 min. Body temperature was maintained throughout the experiment. Perfusion-fixation is expected to alter intracranial pulsatility, a process that has been shown to drive CSF flow (Iliff et al., 2013b); therefore, 30 min post-injection the mice were decapitated, and the brain drop fixed in 4% PFA.

Tissue processing and microscopy

The mouse brains were post-fixed prior to sectioning. In brief, 100 μm thick coronal plane vibratome sections were obtained using a Leica VT 1200S apparatus (Leica Biosystems, IL). Free-floating mouse brain sections were mounted with ProLongTM Gold Antifade Mounting Media with DAPI (Life Technologies, CA). We used an epifluorescent macroscope for imaging mouse brain sections (Olympus, Supplementary file 3). The images were pseudocolored and analyzed on FIJI software. The sections were analyzed following methodology from both Iliff et al. (2012) and Smith et al. (2017). Mean pixel intensity was calculated for six coronal sections covering +1.2 to −1.8 mm from anterior to posterior relative to bregma. Subregional analysis of the coronal section at position +0.6 mm was conducted to identify which regions accounted for the differences in CSF influx (Kress et al., 2014).

URMC

Request a detailed protocol
Intracisternal tracer injections

Mice (8- to 10-week-old males) were anesthetized with ketamine/xylazine (100/10 mg/kg) and fixed in a stereotaxic frame whereupon the posterior atlanto-occipital membrane overlying the cisterna magna was surgically exposed. For all CSF tracer experiments, Alexa 647-conjugated bovine serum albumin (BSA-647; 66 kDa; 0.5%) was injected into the subarachnoid CSF via cisterna magna puncture, at a rate of 1 or 2 µL/min for a period of five min (5 µL or 10 µL total volume) through a 30-gauge syringe pump (Harvard Apparatus). To visualize penetration of fluorescent CSF tracers into the brain parenchyma ex vivo, anesthetized animals were decapitated at 30 min after the start of the injection, a time point previously identified to correspond to robust tracer penetration of similar molecular weight compounds in young male C57BL/6 mice (Iliff et al., 2012). Brains were then removed and post-fixed in 4% PFA for 24 hr before being sliced with a vibratome into 100 µm coronal sections which were slide-mounted using PROLONG anti-fade gold with DAPI (Invitrogen).

In vivo transcranial optical imaging

Transcranial optical imaging was performed in KX-anesthetized mice as recently described (Plog et al., 2018). To prepare the mouse, the skin overlying the dorsal convexity was reflected and afterwards the cisterna magna was cannulated as described above. Fluorescent CSF tracers Texas Red 3 kDa dextran and BSA-647 (10 µl at 2 µl/ min, 0.5% in aCSF) were injected into the cisterna magna and imaged using a fluorescent macroscope (MVX10, Olympus) using a PRIOR Lumen 1600-LED light source and ORCA Flash 4.0 digital camera (Hamamatsu). The mouse’s head was stabilized with ear bars using a customized stereotaxic frame and placed under the microscope stage. Imaging was done at a 20x magnification and acquired in the far-red emission channel (647 nm). Due to high scattering at the excitation/emission wavelength of the Texas Red 3 kDa dextran through the intact skull this channel was not acquired. Images were collected every minute for 30 min following the start of the injection. MetaMorph Basic imaging software was used for image acquisition (Molecular Devices). The same exposure time (100 ms) was used throughout the experiment and between the WT and KO groups. After 30 min, the brain was extracted and fixed as described below.

Intrastriatal tracer injections

KX-anesthetized mice were placed in a stereotaxic apparatus and the skin was opened to expose the skull. Carefully, periosteum was removed and bregma was identified with a surgical microscope. A burr hole was done with a hand-held drill on the coordinates AP:+0.5 DV:+3.0 LM:+2.0 mm from bregma, using a 27G needle to break the meningeal layer. A pulled glass capillary (TW100-3, World Precision Instruments) with a 12 µm tip diameter was introduced in the brain parenchyma. Once in place, 20 nL of 0.5% Amyloid-β1-40 in aCSF (as a sham) was delivered in a perfusion rate of 4 nL/second by a micropump (UMP3 UltraMicroPump with Micro4 controller, World Precision Instruments) as done by Smith et al. Control mice did not receive an injection. The micropipette remained in place for 5 min and then was very slowly withdrew, to avoid any possible backflow. If backflow was observed, the animal was excluded from the experiment. Immediately after the capillary was removed, a small amount of Kwik-Cast Sealant (KWIK-CAST, World Precision Instruments) was applied to seal the site of injection. After 60 min, mice received an injection in cisterna magna with BSA-647 (1% in aCSF) as descried above to evaluate glymphatic function. ​

Ex-vivo imaging of fluorescent CSF tracers

BSA-647 and TxRd 3 kDa dextran circulation along perivascular pathways and penetration into the brain parenchyma was visualized by conventional fluorescence microscopy of 100 µm vibratome coronal brain sections, as described previously (Iliff et al., 2012). For coronal section quantification, six brain sections (per animal were imaged and analyzed by a blinded investigator using an Olympus fluorescence macroscope (MVX10) under 20x magnification to generate whole-slice images (using the MetaMorph software, Molecular Devices, Supp. File 3). High-magnification montage images of coronal sections were acquired using an Olympus BX51 fluorescent microscope at 4x objective magnification using cellSens software (Olympus). Tracer penetration was quantified using NIH Image J software as described previously (Iliff et al., 2012). The first slice was collected at the beginning at the anterior aspect of the corpus callosum, one section was collected every 500 µm apart until a total of six sections had been collected for each animal. Fluorescence of the CSF tracer BSA-647 was measured in each slice as mean pixel intensity (MPI). The MPI from the six brain slices from each animal were averaged to define CSF penetration within a single biological replicate. For the ipsilateral to contralateral comparison, 3 to 5 slices from the injection site where selected from each mouse, and their tracer penetration was quantified, averaged and analyzed by Wilcoxon rank-sum test using different ROIs for each hemisphere. A subset of brain slices were immunostained for AQP4 as previously described (Kress et al., 2014). Tracer penetration depth was evaluated using Fiji and Matlab. A coronal section obtained from bregma from each mouse was used for the analysis. In the KO mice, a 500 µm line was placed orthogonal to the cortical surface at the most dorsal position where tracer could be found at the pial surface (Figure 5f). Tracer depth in the WT mice was measured at the same position as the KO mouse despite there being more tracer spread over the dorsal convexity. The line profile was normalized to the maximum fluorescence at the cortical surface (ΔF/Fmax). Tracer penetration depth was quantified as the distance at which Fmax was reduced by half (d1/2) similar to Smith et al. (2017).

OHSU

Request a detailed protocol
Dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) of glymphatic transport

Three- to 6-month-old mice were anesthetized with isoflurane, with induction at 3–5%. The posterior atlanto-occipital membrane was exposed surgically. To enable the contrast injection during MRI scanning, we used a pulled glass micropipette with trimmed end (external diameter of approximately. 016 mm) to perform the injection. The micropipette was fixed in place for the duration of the imaging session with superglue. Gadoteridol (68 mM, osmotically adjusted), a contrast medium was infused by syringe pump (Harvard Apparatus) at a rate of 500 nl/min for 20 min (10 µl total volume), with a 2 µl chase of saline.

MR imaging

CSF circulation was quantified by dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). All imaging was performed using a Bruker-Biospec 11.75 T preclinical scanner equipped with a 20 mm I.D. quadrature RF volume-coil with a specially designed head holder (Supplementary file 3). Heart rate, oxygen saturation and respiratory rate were monitored, and core temperature was maintained at 37°C using a warm air temperature control system (SA Instruments). Upon placement of the glass micropipette, isoflurane anesthesia was switched to ketamine-xylazine (100–10 mg/kg) for the duration of the experiment. Consecutive T1 weighted 3D FLASH images were obtained at 10 min intervals (TR/TE 16/2.8 ms, flip angle 15°, matrix 256 × 192×192, 100 × 100×100 μm resolution), for a total of 90 min. Injection was initiated after acquiring the first image. If no elevation in signal was detectable in the cortical or subcortical brain regions or the basal cistern within the first 30 min, the imaging session was aborted, and the animal was excluded from further analysis.

MR data analysis

Due to frequent occurrence of motion artifact during the last 30 min of acquisition (70–90 min), these three time points were excluded from analysis, and those animals showing movement artifacts during the first 60 min of acquisition were also excluded. Linear rigid body registration was used to spatially align the image series to the baseline image (FSL). Regions of interest (ROIs) were drawn on a coronal image slice best matching that depicted in Figure 6I, and mean intensity values were determined with FIJI software. Repeated measures two-way ANOVA with Sidak multiple comparisons correction was used to analyze the data. Multiple comparisons were used to determine the significance of genotype differences at each time point.

Meta-analysis

Literature search

Request a detailed protocol

A review of the literature on PubMed and Google Scholar was conducted for all studies published to date (November 2018) on global Aqp4 KO models in rodents.

Inclusion criteria:

Request a detailed protocol

Out of 6527 studies that were retrieved from the online search, only tracer studies that evaluated fluid flow in the brain in Aqp4 KO versus wild-type controls were included.

Exclusion criteria

Request a detailed protocol

Studies were excluded if either mean, standard deviation, or sample size could not be extracted. Studies that evaluated tracer transport in a disease model were also eliminated. Non-English studies were not excluded. Only one study Binder et al. 2004 was excluded due to the inability to estimate sample size (Binder et al., 2004). After exclusion, a total of 11 studies satisfied the selection criteria. Five data sets from the present study (NMU, RIKEN, UNC, URMC) were also included.

Data extraction

Request a detailed protocol

Means, standard deviations, and sample sizes for each experiment were extracted from the text where possible. If exact numbers were not reported, we used graphical extraction using FIJI. When standard error of the mean was reported, standard deviation was calculated using the square root of the sample size. In cases where sample size was reported as a range, the lowest number was chosen for analysis. In one study, it was impossible to determine if two separate analyses were derived from the same group of mice. In order to avoid inadvertently excluding data (e.g. If the study was in fact conducted in two separate sets of mice) we included both experimental data sets in the meta-analysis (Smith et al., 2017).

Data analysis

Request a detailed protocol

Studies addressed mainly two parts of intracranial fluid flow: (1) CSF influx to brain after delivery of tracers to the cisterna magna and (2) tracer transport after being injected into the interstitial fluid in the brain parenchyma. To address these two types of experiments, we decided to run two separate models using the inverse variance method for meta-analysis. Since these were evaluated using different outcome measures between fluorescence (fluorescence intensity and thresholding approaches), radioisotope, and MRI experiments we opted for a standardized mean difference (SMD) using the Cohen’s D method for the final analysis. After data was extracted, analysis was done in R (R-Project). Due to high heterogeneity (I2 >70%), a random effects model was selected after attempting to reduce heterogeneity by implementing alternate standardized mean difference formulas (e.g. Glass’s delta, Hedge’s G); however, these attempts did not eliminate heterogeneity. Sources of heterogeneity were then explored using mixed-effects model meta-regression. Candidate predictors of heterogeneity included: anesthesia type, species, strain, age, tracer properties, injection properties, injection location, and experimental method. Inclusion of more than one covariate in the meta-regression resulted in model non-convergence due to the relatively small number of studies and the distribution of these factors across studies so covariates were included one at a time. Data in forest plots is shown as SMD with 95% confidence intervals (CI) and p values less than 0.05 were considered significant.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
    Optimization of intraperitoneal injection anesthesia in mice: drugs, dosages, adverse effects, and anesthesia depth
    1. M Arras
    2. P Autenried
    3. A Rettich
    4. D Spaeni
    5. T Rülicke
    (2001)
    Comparative Medicine 51:443–456.
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
    MRI with intrathecal MRI gadolinium contrast medium administration: a possible method to assess glymphatic function in human brain
    1. PK Eide
    2. G Ringstad
    (2015)
    Acta Radiologica Open, 4, 10.1177/2058460115609635, 26634147.
  19. 19
    Is bulk flow plausible in perivascular
    1. MM Faghih
    2. MK Sharp
    (2018)
    Paravascular and Paravenous Channels? Fluids and Barriers of the CNS, 15, 10.1186/s12987-018-0103-8.
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
    Glymphatic clearance controls state-dependent changes in brain lactate concentration
    1. I Lundgaard
    (2016)
    Journal of Cerebral Blood Flow and Metabolism: Official Journal of the International Society of Cerebral Blood Flow and Metabolism 36:2112–2124.
    https://doi.org/10.1177/0271678X16661202
  39. 39
  40. 40
    Flow of cerebrospinal fluid is driven by arterial pulsations and is reduced in hypertension
    1. H Mestre
    2. J Tithof
    3. T Du
    4. W Song
    5. W Peng
    6. AM Sweeney
    7. G Olveda
    8. JH Thomas
    9. M Nedergaard
    10. DH Kelley
    (2018)
    Nature Communications, 9, 10.1038/s41467-018-07318-3, 30451853.
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
    Transcranial optical imaging reveals a pathway for optimizing the delivery of immunotherapeutics to the brain
    1. BA Plog
    2. H Mestre
    3. GE Olveda
    4. AM Sweeney
    5. HM Kenney
    6. A Cove
    7. KY Dholakia
    8. J Tithof
    9. TD Nevins
    10. I Lundgaard
    11. T Du
    12. DH Kelley
    13. M Nedergaard
    (2018)
    JCI Insight, 3, 10.1172/jci.insight.126138, 30333324.
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
  66. 66
  67. 67
  68. 68
  69. 69
    Cannula implantation into the cisterna magna of rodents
    1. ALR Xavier
    2. NL Hauglund
    3. S von Holstein-Rathlou
    4. Q Li
    5. S Sanggaard
    6. N Lou
    7. I Lundgaard
    8. M Nedergaard
    (2018)
    Journal of Visualized Experiments, 10.3791/57378, 29889209.
  70. 70
  71. 71
  72. 72
  73. 73

Decision letter

  1. David Kleinfeld
    Reviewing Editor; University of California, San Diego, United States
  2. Sean J Morrison
    Senior Editor; Howard Hughes Medical Institute, University of Texas Southwestern Medical Center, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Aquaporin-4-dependent glymphatic solute transport in the rodent brain" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Sean Morrison as the Senior Editor. The reviewers have opted to remain anonymous.

This is a well appreciated consortium effort to address the issue of rebalance of water through the Aquaporin-4 channels. The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

A number of critical issues must be resolved before a final decision can be made regarding publication.

Data issues

The images of the wild type and APQ4-/- muse shown in Figure 2A, B must be from the same brain areas and must be in good focus and have high dynamic range (you may consider taking the logarithm of the image or using γ ~ 0.5). I would argue that you need to show three to four such pairs as these constitute primary data.

The issue of image quality holds for Figures 4A and 6C-H as well.

In keeping with the current push (and requirement in some journals), please show all data points in Figure 2-5 and 7 as a scatter plot with Tukey boxes, or equivalent rather than means and SDMs; this is similar to what you do in Figure 5.

Please indicate the brain area(s) that are used in each analysis; this could be the range of transverse coordinates relative to bregma.

As the data was taken across many laboratories, please make a table of parameters (brain regions, analysis method, camera, etc.) for each analysis.

A potential physical mechanism should be presented, in the Discussion, for driving CSF from the perivascular to the interstitial and then periventricular spaces, rather than have solution slosh to and fro.

Vocabulary issue:

To avoid overselling, please change title to "Glymphatic solute movement in the rodent brain has an Aquaporin-4-dependent component".

To emphasize the mass action nature of water transfer, please use "flux" rather than "transport", which implies direct energy usage (as in an ATP driven ion exchange).

State explicitly that in this report, all of the data is on influx of CSF as part of the proposed glymphatic system.

Please answer all questions from the specific comments of Reviewer #2.

Reviewer #1:

This manuscript reports evidence from 5 laboratories that deletion of Aquaporin 4 (Aqp4) partially attenuates the movement of labelled solutes injected into the cisterna magna from CSF in the subarachnoid space into the ISF in the peri-vascular space along penetrating arterioles of the mouse cerebral cortex. The main purpose of this particular report is to provide evidence that counters a previous report in eLife in which deletion of Aqp4 was found to have no effect on such solute movement. The reason that this matters is that the present authors use this evidence to support a major role for Aqp4 and astroglial endfeet in their hypothetical model of a 'glymphatic' system of active, astrocyte-assisted convection 'transport' of molecules along perivascular spaces for clearance. The authors have proposed (elsewhere) a critical role for this proposed system in clearance of β-amyloid.

In principle, I find the current work sufficiently well done, and the topic of sufficient importance and interest, that the results presented here warrant publication. Nevertheless, I also find that the authors markedly over-state certain aspects of their findings and use terms that are not supported by the findings. For this reason, there are a number of major specific concerns that need to be addressed before moving forward.

1) It is not appropriate to use the term "Aqp4-dependent" in the title, Abstract (second sentence), or anywhere else in the text. The results presented do not demonstrate a "dependence" of solute movement on Aqp4. The most direct and appropriate quantitative evaluation of the effects of Aqp4 deletion in this context appears in Figure 5 and shows an approximate reduction of only 25%. Although this reduction is significant, it is very far from absolute. There is still a majority of detectable solute movement in the absence of Aqp4. Based on this and the other results, the authors can at best say that Aqp4 "facilitates" solute transport, which is the term they use in the final sentence of their paper, or "supports", which is the term they use at the end of the Abstract. The title should not say more than "Aqp4-facilitated", or "Aqp4-supported".

2) As a biologist, I am also not comfortable with the use of the word 'transport' in the manner being done throughout this paper. To me, 'transport' is a term used in biology for something that requires energy and occurs across a barrier, for example transport of molecules across a membrane assisted by a transporter molecule. Here, there is no obvious barrier to the movement of solutes, and there is no obvious requirement for energy to assist what is essentially the passive movement of solutes along a convection flow of fluid current or along a concentration gradient. I would argue that the term 'transport' should also not be used in the title or anywhere in the paper and should be substituted with 'movement' or flux. To me there is a clear distinction between the transport of water across the membrane via Aqp4 and the flow of ISF and the movement of solutes, which to me does not constitute 'transport'. It is noteworthy that the present authors generally use the term 'movement' of solutes throughout the main text. I would be interested to hear other reviewers' comments about this, and of course the authors can provide arguments that they think justify the use of the term 'transport'.

3) NMU data, Figure 2. The results of this experiment seem selectively quantified and over generalized. I am not convinced that the two survey images in 2A show equivalent regions of the brain in the WT and Aqp4-/- mice. To me it seems likely that the two images shown may be misleading. Both survey images show that the most intense labeling occurs in the subarachnoid space and that this is far more intense along the ventral brain surface. The WT image (left) is clearly taken from a frontal region where there is intensely labeled subarachnoid space between the two hemispheres. In contrast, the Aqp4-/- image appears to be taken from a region that is relatively similar, but far enough posterior that the hemispheres have already joined and there is no subarachnoid space between the hemispheres. If this is the case, then the comparison is misleading and inappropriately exaggerates the visual appearance of fluorescence intensity in favor of the WT mouse. The images presented are pixelated and blurry, so that the precise location is not readily detectable in the absence of a counterstain such as DAPI. This comparison matters because if the author used these sections for their quantification, that too will have been skewed in favor of the WT mice. The authors need to present images where it is clear to see that they have been take from precisely the same region and where the selective presence of midline subarachnoid tissue does not bias the interpretation in favor of one mouse versus another. In addition, the authors should ensure that this type of bias did not occur in the quantification. The main text states that Figure 2b is taken from hypothalamus, whereas the legend says it is taken from hippocampus. Clearly there is a brain surface involved and so hippocampus seems unlikely. Ideally it would be preferable for the authors to show high magnification images taken from the same sections that are shown in the survey images, and to indicate by using boxes on the survey images where the high magnification images were taken from. Regarding the quantification, the individual data points for each animal should be shown in all of the graphs to represent the scatter. The authors should also indicate which cortical region was used for the perivascular space and parenchymal measurements. Ideally this would correspond to the areas shown in the images. From the results it appears that the measurements were taken only from a ventral brain region. This is somewhat misleading. The survey images that are provided suggest that the lateral and dorsal surfaces of the brain exhibit almost no differences in the penetration of fluorescence from the subarachnoid space. I think other cortical regions should be evaluated from other regions where on the survey images there appears to be no difference in the degree of fluorescence penetration in the WT and Aqp4-/- mice. If this is not done, then the authors should discuss why there are such large differences in cortical areas and why they feel it is appropriate to select only the region with the largest difference and then generalize their conclusions from this region to the whole brain. Lastly, the authors should make an explicit statement that they ensured that all images used for comparison were collected and evaluated using the same settings for fluorescence intensity and contrast etc.

4) Individual data points must be added to all graphs in all figures. At present it is not possible to judge the scatter and variation of most of the data. This applies to Figures 2, 3, 4 and 6. This information was provided for the present Figure 5 and was also provided for all data in the Smith et al. paper whose work is being challenged. It is a requirement.

Reviewer #2:

General comments:

1) In this group of studies, the authors address intracranial fluid flow including CSF influx to the brain after delivery of tracers to the cisterna magna and tracer transport after injection into the interstitial fluid in the brain parenchyma in four independently generated Aqp4 KO lines as well as an α-syntrophin KO line.

2) Compilation of data from the five independently-conducted experiments showed that deletion or mislocalization of AQP4 suppressed the influx of CSF tracers or contrast agents.

3) The authors interpret these data as supportive evidence for the glymphatic model in which astroglial water transport via AQP4 supports the perivascular influx of cerebral spinal fluid (CSF) and the efflux of interstitial space fluid (ISF).

4) The heterogeneity of the individual studies has to be taken into consideration. The authors do address the concern of conflicting results due to varying methodology used in each study. The paper would be improved if the same methodologies were performed for the independently generated Aqp4 KO lines.

5) Most critically: (1) the authors' hypothesis that anesthetic effects (tribromoethanol vs. ketamine/xylazine) account for the difference between Smith et al. (Smith et al., 2017) results and the current findings is not tested; and (2) no further data are provided in support of the AQP4 dependence of parenchymal clearance of solutes (brain to CSF efflux), as all of the data in the current manuscript relate to CSF to brain penetration (influx).

6) Please see specific comments below.

Specific comments:

1) NMU group injected fluorescent tracer Texas Red-dextran intracisternally which displayed reduced influx in the Aqp4 KO mice as shown in Figure 2 following a thirty-minute infusion. However, this group did not detail location of individual slices collected (6-8 per mouse) as described by the RIKEN group in Figure 3D and UNC group in Figure 4D. These methodological details should be clarified.

2) The authors claim that "High magnification imaging of the hypothalamus revealed that Aqp4 KO impaired the influx…(Figure 2B)" but Figure 2B is neither high magnification nor does it clearly show the hypothalamus. More data would be required to assess this claim.

3) RIKEN and Keio group claim to show that BDA distribution was compromised in Aqp4 KO mice following infusion in Figure 3. However, details of infusion (they say "infusion" in Figure 3A but "injection" in the figure legend) are not clear. Also, was only 1 time point (30 min) performed?

4) Figure 3B is not convincing for a difference between WT and Aqp4 KO.

5) Figure 3C could have been formatted more appropriately to give more clarity (configured similar to Figure 2D). Depth at half-maximum intensity is shown in Figure 3E whereas fluorescence intensity at multiple distances from the brain surface could have provided more information if demonstrated better in Figure 3C. Thus, measurement criteria for penetration depth are heterogeneous across Figures 3C-E.

6) UNC group used the same Aqp4 KO line used in Smith et al. (Smith et al., 2017) in order to assess the reproducibility of the experiments by using methodology reported by Iliff et al. (Iliff et al., 2012). A main justification for disputing the dissimilarity of results is the anesthesia used in the Smith et al. study. The authors argue that the use of tribromoethanol (Avertin) could have effected parenchymal solute efflux yet they didn't test this themselves. It would have been valuable for the UNC group to perform the same experiments using both ketamine/xylazine and tribromoethanol to further support their conclusions. This should have been performed to directly test the idea that the disparity in results is anesthetic state dependent.

7) URMC group claims to show (Figure 5) that Aqp4 deletion reduces entry of Alexa Fluor 647-conjugated BSA from CSF to brain. However, comparison of Figures 5E and 5G and 5F and 5H does not show a convincing difference.

8) In addition, they only looked at mean pixel intensity (MPI) relative to bregma (Figure 5I) and depth was not taken into consideration, unlike the data in other figures in this manuscript. It would have been beneficial to look at MPI respective to definitive depths over a relative distance. This would be a more direct measurement of extent or depth of penetration.

9) OHSU used the Snta1 KO mouse line to access the role of perivascular astrocytic localization of Aqp4 in mediating CSF flux into the parenchyma using DCE-MRI following intracisternal injection of gadoteridol. Figure 6 successfully demonstrates reduced accumulation of Gad in Snta1 KO mice across multiple brain regions.

10) However, Gad is a contrast agent and is not a direct measure of CSF flow itself, thus the conclusion that α-syntrophin facilitates CSF-ISF exchange is not clearly supported.

11) Also, again there is potential confusion created by using both the terms "injection" and "infusion" to describe the experiments (subsection “OHSU: Reduced glymphatic CSF influx in Snta1 KO mice with loss of polarized expression of AQP4 in vascular endfeet of astrocytes”), this should be corrected.

12) All of the experiments done in this paper involve penetration of tracers from CSF injection/infusion into brain parenchyma. None of the experiments directly tests the clearance of intraparenchymally infused agents from brain to CSF. This was a prominent finding of Smith et al. (Smith et al., 2017), that there was no significant difference in clearance of solutes in AQP4 KO mice following intraparenchymal infusion. In other words all experiments in this manuscript are CSF to brain, but the claims are bidirectional (Figure 7) that there is also decreased ISF clearance from brain to CSF in the absence of AQP4. Why did none of the current experiments investigate or substantiate this claim?

13) The Discussion reiterates the fact that the UNC group were unable to replicate Smith's findings however they did not follow Smith's methodologies so reproducibility is debatable since methods deviated. This omission reduces the claim of the current manuscript to be a "refutation" of Smith et al.'s findings.

14) An obvious experiment that would eliminate the effects of anesthesia would be to implant an infusion catheter in the cisterna magna and allow animals to recover for several days, and then carry out infusion of tracers in the complete absence of any anesthesia whatsoever and then rapidly perfuse at different time points and assess parenchymal penetration depth criteria. While this may be technically difficult in mice, stereotactically targeted infusions have been performed in awake behaving mice in many locations, and in any event this would be feasible in rats (and AQP4 KO rats have been described). This experiment would clearly move the field forward beyond disparate data relative to anesthetic usage.

https://doi.org/10.7554/eLife.40070.021

Author response

A number of critical issues must be resolved before a final decision can be made regarding publication

Data issues

The images of the wild type and APQ4-/- muse shown in Figure 2A, B must be from the same brain areas and must be in good focus and have high dynamic range (you may consider taking the logarithm of the image or using γ ~ 0.5). I would argue that you need to show three to four such pairs as these constitute primary data.

Following the editor’s suggestions, we have provided in Figure 2, five pairs of the low magnification images of WT and Aqp4 KO forebrain sections that are located at 0.5 mm anterior to the bregma. In addition, we added a new set of analyses based on ex vivo near infrared fluorescence imaging to compare overall tracer (TRd3) distribution along the ventral and dorsal brain regions comparing WT and Aqp4 KO mice. We are now also providing confocal images to obtain higher resolution and eliminate out of focus signal of the coronal brain sections as requested by the reviewer.

The issue of image quality holds for Figures 4A and 6C-H as well.

We have updated Figure 4A by applying a γ of 0.5 per the editor’s suggestion. The images presented in Figure 6C-H are not confocal micrographs, but rather were generated using a 11.75 T preclinical MRI scanner, which is limited in resolution to 100µm3 voxels. As such, the images are currently presented at their highest resolution. To clarify this for readers, additional description has been added to the legend of Figure 6.

“Figure 6. Deletion of the adapter protein α-syntrophin impairs AQP4 perivascular localization, and CSF influx into the brain parenchyma. […]Traces for each individual animal are presented (lines) along with the summary statistics (mean ± SEM, 2-way ANOVA). WT n=5, ASYNKO n=7. CTx = cortex (P = 0.0035) Hip = hippocampus (P = 0.0003) Subcortical = subcortical regions (P = 0.0185) 3V = 3rd Ventricle (P = 0.0284) Total (P = 0.0085)”.

In keeping with the current push (and requirement in some journals), please show all data points in Figure 2-5 and 7 as a scatter plot with Tukey boxes, or equivalent rather than means and SDMs; this is similar to what you do in Figure 5.

All data points are now shown for Figures 2-7.

Please indicate the brain area(s) that are used in each analysis; this could be the range of transverse coordinates relative to bregma.

As the data was taken across many laboratories, please make a table of parameters (brain regions, analysis method, camera, etc.) for each analysis.

We have included this information in the requested table ‘Imaging Parameters Table’. In addition, we have expanded Figure 1 to include the experimental details from each group.

A potential physical mechanism should be presented, in the Discussion, for driving CSF from the perivascular to the interstitial and then periventricular spaces, rather than have solution slosh to and fro.

We have on request of the editor included a discussion:

“How AQP4 on a cellular level facilitates CSF/ISF exchange remains to be firmly established. […] Moreover, deletion of AQP4 potentiated the increases in intracranial pressure in a murine model of hydrocephalus consistent with the notion that AQP4 supports and facilitates intraparenchymal fluid flow (Ren et al., 2017).”

Vocabulary issue:

To avoid overselling, please change title to "Glymphatic solute movement in the rodent brain has an Aquaporin-4-dependent component".

We are puzzled by the objection to our use of the term "Aqp4-dependent." The critique rests on the assertion that we can only use the term if the solute movement depends only on Aqp4, which we do not claim. The term “dependent” is commonly used when the dependence is on more than one variable. For example, in the thermodynamics of simple substances, any one thermodynamic variable can be expressed as a function of any two of the other thermodynamic variables, and it is perfectly correct to say, for example, that the density of a gas is “temperature-dependent” when we know that it also depends on the pressure. In the more complicated case of glymphatic flow, the flow no doubt depends on several variables, and we have shown that Aqp4 expression is certainly one of them. The primary purpose of our paper is to refute the claims in the Smith et al., 2017 paper, which claims that solute transport is “aquaporin-4-independent” in its title: therefore, it is highly appropriate that we use the term "Aqp4-dependent" in our title. If we were to use a different term, it would be confusing to the readers.

For clarity, we have in response to the reviewer’s comment added the following sentence to the Discussion:

“However, it is important to note that glymphatic transport is affected by multiple pathways other than AQP4. For example, the sleep-wake cycle (Xie et al., 2013), brain injury in the setting of trauma or ischemia (Iliff et al., 2014; Wang et al., 2017), exercise (He et al., 2017; von Holstein-Rathlou, Petersen and Nedergaard, 2018; Yin et al., 2018), amyloid-β accumulation and acute amyloid-β toxicity (Xu et al., 2015; Peng et al., 2016), omega-3 fatty acids (Ren et al., 2017), plasma osmolarity (Plog et al., 2018), and aging (Kress et al., 2014) are important regulators of glymphatic transport.”

To emphasize the mass action nature of water transfer, please use "flux" rather than "transport", which implies direct energy usage (as in an ATP driven ion exchange).

We respectfully argue that it is preferable to retain the term “transport” in the title of our paper and in the text, with the following justification. We are dealing with bulk flow in an organ (brain) and not carrier or pump mediated transport across a plasma membrane. We use the term precisely as it is used in most papers dealing with solute transport in the brain. Indeed, the primary purpose of our paper is to refute the claims in the Smith et al., 2017 paper, which itself uses the term "transport" in its title ("Test of the 'glymphatic' hypothesis demonstrates diffusive and aquaporin-4-independent solute transport in rodent brain parenchyma") and in the first sentence of its Abstract ("Transport of solutes through brain involves diffusion and convection"). It would be inconsistent and confusing if we were to replace "transport" with a different term in our title. There is a long list of papers that use the term transport in the same context that we have: for example, it is used in the titles of several papers that we cite in our paper (the papers by Asgari et al., Benveniste et al., Holter et al., Murlidharan et al., Plog et al., Ratner et al., and Schley et al.). Also, we point out that the reviewer is incorrect in stating that there is no energy requirement to assist in the movement of solutes along a convective flow: the flow itself requires energy to drive it, in the form of the work done by pressure forces to overcome the resistance due to viscosity.

State explicitly that in this report, all of the data is on influx of CSF as part of the proposed glymphatic system.

We have now included the sentence: “This study focused on re-evaluating the role AQP4 in glymphatic influx of CSF tracers”. In addition, we assessed how intraparenchymal tracer injection affected CSF tracer influx. The latter analysis was included, because the Smith et al., publication refuted the existence of bulk flow based on dispersion of tracers delivered by intraparenchymal injection. We replicated the procedures used by Smith et al. but analyzed its effect of CSF tracer influx. Our analysis showed that the procedures associated with intraparenchymal injection – including preparation of a cranial burr hole, penetration of dura with the glass pipette and tissue injury when the pipette is inserted – consistently suppressed CSF tracer inflow. Based on these observations, we conclude that it is not possible to establish whether convective flow exists in the tissue based on the acute invasive procedures utilized in the Smith et al., 2017 publication.

Please answer all questions from the specific comments of Reviewer #2.

Reviewer #1:

[…] In principle, I find the current work sufficiently well done, and the topic of sufficient importance and interest, that the results presented here warrant publication. Nevertheless, I also find that the authors markedly over-state certain aspects of their findings and use terms that are not supported by the findings. For this reason, there are a number of major specific concerns that need to be addressed before moving forward.

1) It is not appropriate to use the term "Aqp4-dependent" in the title, Abstract (second sentence), or anywhere else in the text. The results presented do not demonstrate a "dependence" of solute movement on Aqp4. The most direct and appropriate quantitative evaluation of the effects of Aqp4 deletion in this context appears in Figure 5 and shows an approximate reduction of only 25%. Although this reduction is significant, it is very far from absolute. There is still a majority of detectable solute movement in the absence of Aqp4. Based on this and the other results, the authors can at best say that Aqp4 "facilitates" solute transport, which is the term they use in the final sentence of their paper, or "supports", which is the term they use at the end of the Abstract. The title should not say more than "Aqp4-facilitated", or "Aqp4-supported".

We are puzzled by the objection to our use of the term "Aqp4-dependent." The critique rests on the assertion that we can only use the term if the solute movement depends only on Aqp4, which we do not claim. The term “dependent” is commonly used when the dependence is on more than one variable. For example, in the thermodynamics of simple substances, any one thermodynamic variable can be expressed as a function of any two of the other thermodynamic variables, and it is perfectly correct to say, for example, that the density of a gas is “temperature-dependent” when we know that it also depends on the pressure. In the more complicated case of glymphatic flow, the flow no doubt depends on several variables, and we have shown that Aqp4 concentration is certainly one of them. The primary purpose of our paper is to refute the claims in the Smith et al., 2017 paper, which claims that solute transport is “aquaporin-4-independent” in its title: therefore, it is highly appropriate that we use the term "Aqp4-dependent" in our title. If we were to use a different term, it would be confusing to the readers.

For clarity, we have in response to the reviewer’s comment added the following sentence to the Discussion:

“However, it is important to note that glymphatic transport is affected by multiple pathways other than AQP4. For example, the sleep-wake cycle (Xie et al., 2013), brain injury in the setting of trauma or ischemia (Iliff et al., 2014; Wang et al., 2017), exercise (He et al., 2017; von Holstein-Rathlou, Petersen and Nedergaard, 2018; Yin et al., 2018), amyloid-β accumulation and acute amyloid-β toxicity (Xu et al., 2015; Peng et al., 2016), omega-3 fatty acids (Ren et al., 2017), plasma osmolarity (Plog et al., 2018), and aging (Kress et al., 2014) are important regulators of glymphatic transport.”

2) As a biologist, I am also not comfortable with the use of the word 'transport' in the manner being done throughout this paper. To me, 'transport' is a term used in biology for something that requires energy and occurs across a barrier, for example transport of molecules across a membrane assisted by a transporter molecule. Here, there is no obvious barrier to the movement of solutes, and there is no obvious requirement for energy to assist what is essentially the passive movement of solutes along a convection flow of fluid current or along a concentration gradient. I would argue that the term 'transport' should also not be used in the title or anywhere in the paper and should be substituted with 'movement' or flux. To me there is a clear distinction between the transport of water across the membrane via Aqp4 and the flow of ISF and the movement of solutes, which to me does not constitute 'transport'. It is noteworthy that the present authors generally use the term 'movement' of solutes throughout the main text. I would be interested to hear other reviewers' comments about this, and of course the authors can provide arguments that they think justify the use of the term 'transport'.

We respectfully argue that it is preferable to retain the term “transport” in the title of our paper and in the text, with the following justification. We are dealing with bulk flow in an organ (brain) and not carrier or pump mediated transport across a plasma membrane. We use the term precisely as it is used in most papers dealing with solute transport in the brain. Indeed, the primary purpose of our paper is to refute the claims in the Smith et al., 2017 paper, which itself uses the term "transport" in its title ("Test of the 'glymphatic' hypothesis demonstrates diffusive and aquaporin-4-independent solute transport in rodent brain parenchyma") and in the first sentence of its Abstract ("Transport of solutes through brain involves diffusion and convection"). It would be inconsistent and confusing if we were to replace "transport" with a different term in our title. There is a long list of papers that use the term transport in the same context that we have: for example, it is used in the titles of several papers that we cite in our paper (the papers by Asgari et al., Benveniste et al., Holter et al., Murlidharan et al., Plog et al., Ratner et al., and Schley et al.). Also, we point out that the reviewer is incorrect in stating that there is no energy requirement to assist in the movement of solutes along a convective flow: the flow itself requires energy to drive it, in the form of the work done by pressure forces to overcome the resistance due to viscosity.

3) NMU data, Figure 2. The results of this experiment seem selectively quantified and over generalized. I am not convinced that the two survey images in 2A show equivalent regions of the brain in the WT and Aqp4-/- mice. To me it seems likely that the two images shown may be misleading. Both survey images show that the most intense labeling occurs in the subarachnoid space and that this is far more intense along the ventral brain surface. The WT image (left) is clearly taken from a frontal region where there is intensely labeled subarachnoid space between the two hemispheres. In contrast, the Aqp4-/- image appears to be taken from a region that is relatively similar, but far enough posterior that the hemispheres have already joined and there is no subarachnoid space between the hemispheres. If this is the case, then the comparison is misleading and inappropriately exaggerates the visual appearance of fluorescence intensity in favor of the WT mouse. The images presented are pixelated and blurry, so that the precise location is not readily detectable in the absence of a counterstain such as DAPI. This comparison matters because if the author used these sections for their quantification, that too will have been skewed in favor of the WT mice. The authors need to present images where it is clear to see that they have been take from precisely the same region and where the selective presence of midline subarachnoid tissue does not bias the interpretation in favor of one mouse versus another. In addition, the authors should ensure that this type of bias did not occur in the quantification. The main text states that Figure 2b is taken from hypothalamus, whereas the legend says it is taken from hippocampus. Clearly there is a brain surface involved and so hippocampus seems unlikely. Ideally it would be preferable for the authors to show high magnification images taken from the same sections that are shown in the survey images, and to indicate by using boxes on the survey images where the high magnification images were taken from. Regarding the quantification, the individual data points for each animal should be shown in all of the graphs to represent the scatter. The authors should also indicate which cortical region was used for the perivascular space and parenchymal measurements. Ideally this would correspond to the areas shown in the images. From the results it appears that the measurements were taken only from a ventral brain region. This is somewhat misleading. The survey images that are provided suggest that the lateral and dorsal surfaces of the brain exhibit almost no differences in the penetration of fluorescence from the subarachnoid space. I think other cortical regions should be evaluated from other regions where on the survey images there appears to be no difference in the degree of fluorescence penetration in the WT and Aqp4-/- mice. If this is not done, then the authors should discuss why there are such large differences in cortical areas and why they feel it is appropriate to select only the region with the largest difference and then generalize their conclusions from this region to the whole brain. Lastly, the authors should make an explicit statement that they ensured that all images used for comparison were collected and evaluated using the same settings for fluorescence intensity and contrast etc.

We thank the reviewer very much for raising this critique. We have addressed all issues in the revised manuscript. We are providing the requested detailed information with regard to slice collection and imaging of WT and Aqp4 KO mice in the Materials and methods section. The high magnification images of the hypothalamus have been marked in the low magnification images. We also performed a subregional analysis of CSF tracer in the ventral, lateral and dorsal parts of the forebrain. All data are now displayed as scatter plots. Furthermore, an additional data set has been included. The distribution of CSF tracer (TRd3) at the dorsal and ventral surfaces of the whole-brain is now compared in WT and Aqp4 KO mice using ex vivo near infrared fluorescence imaging. The corresponding images, quantitative data, and methodological description have been added. To facilitate the review, the new Materials and methods section is as follows:

“Tissue processing and image analysis of fluorescent tracer. A subset of WT and Aqp4KO brains (N = 4 per genotype) were scanned by an ex vivo near infrared (NIR) fluorescence imaging system (Azure biosystems c600, CA, USA). […] The imaging and subsequent analysis was performed by an investigator who was blind to animal genotype.”

4) Individual data points must be added to all graphs in all figures. At present it is not possible to judge the scatter and variation of most of the data. This applies to Figures 2, 3, 4 and 6. This information was provided for the present Figure 5 and was also provided for all data in the Smith et al. paper whose work is being challenged. It is a requirement.

Thanks, individual data points are now displayed in all figures.

Reviewer #2:

[…] Specific comments:

1) NMU group injected fluorescent tracer Texas Red-dextran intracisternally which displayed reduced influx in the Aqp4 KO mice as shown in Figure 2 following a thirty-minute infusion. However, this group did not detail location of individual slices collected (6-8 per mouse) as described by the RIKEN group in Figure 3D and UNC group in Figure 4D. These methodological details should be clarified.

The methodological details of section collection and imaging analysis have been added to the revised manuscript. The requested data has been added to Figure 2.

2) The authors claim that "High magnification imaging of the hypothalamus revealed that Aqp4 KO impaired the influx…(Figure 2B)" but Figure 2B is neither high magnification nor does it clearly show the hypothalamus. More data would be required to assess this claim.

We have expanded the data analysis presented in Figure 2.

3) RIKEN and Keio group claim to show that BDA distribution was compromised in Aqp4 KO mice following infusion in Figure 3. However, details of infusion (they say "infusion" in Figure 3A but "injection" in the figure legend) are not clear. Also, was only 1 time point (30 min) performed?

We corrected the figure legend to indicate that tracer is introduced by continuous “injection”. We thank the reviewers for this suggestion. In Figure 3A, we now added a blue horizontal bar to indicate the timing and duration of tracer injection. In the respective Materials and methods section, as well as in Figure 1, and in ‘Supplementary file 1’, we have indicated that the CM injection lasted 10 minutes. Yes, the analysis was performed at a single 30 min time point, in line with all the other groups in this submission. This is also the same time point used by both Iliff et al., 2012 and Smith et al., 2017.

4) Figure 3B is not convincing for a difference between WT and Aqp4 KO.

We have changed the dynamic range of the micrographs in Figure 3B to visualize tracer penetration in the parenchyma. The same dynamic range was applied for both WT and Aqp4-/- images.

5) Figure 3C could have been formatted more appropriately to give more clarity (configured similar to Figure 2D). Depth at half-maximum intensity is shown in Figure 3E whereas fluorescence intensity at multiple distances from the brain surface could have provided more information if demonstrated better in Figure 3C. Thus, measurement criteria for penetration depth are heterogeneous across Figures 3C-E.

We have now reanalyzed and presented our data in a similar style as Figure 2D/E to indicate fluorescence intensity from the brain surface at multiple distances. Indeed, the half distance measure is rather confusing, and we agree that analyzing from the depth 100 µm gives a better insight of tracer penetration difference between WT and Aqp4-/- mice. For this very reason, we now drop Figures3D and E. A new plot, featuring parenchymal fluorescence signal intensity of layer 2 and below, is displayed as Figure 3D.

6) UNC group used the same Aqp4 KO line used in Smith et al. (Smith et al., 2017) in order to assess the reproducibility of the experiments by using methodology reported by Iliff et al. (Iliff et al., 2012). A main justification for disputing the dissimilarity of results is the anesthesia used in the Smith et al. study. The authors argue that the use of tribromoethanol (Avertin) could have effected parenchymal solute efflux yet they didn't test this themselves. It would have been valuable for the UNC group to perform the same experiments using both ketamine/xylazine and tribromoethanol to further support their conclusions. This should have been performed to directly test the idea that the disparity in results is anesthetic state dependent.

The reviewer is correct in stating that anesthesia is a big factor contributing to the variability. In fact, we have studied in detail the effect of anesthesia on the glymphatic system in a separate submission to another journal. We uploaded the manuscript (Hablitz et al.) as a Related Manuscript File when the manuscript was originally submitted. This manuscript compares glymphatic influx in mice anesthetized with 6 different anesthetic agents. The study included more than 100 mice and documents that ketamine/xylazine is the preferable anesthetics in the study of the glymphatic system. CSF tracer influx under tribromoethanol anesthesia (the one used in Smith et al.) is consistently lower than in ketamine/xylazine anesthesia. We have uploaded the submitted manuscript (Hablitz et al.) again, with the intent that the reviewers have access to it.

7) URMC group claims to show (Figure 5) that Aqp4 deletion reduces entry of Alexa Fluor 647-conjugated BSA from CSF to brain. However, comparison of Figures 5E and 5G and 5F and 5H does not show a convincing difference.

In response to the reviewer’s comments, Figure 5 has been updated. We have included an independent replication of the original experiment. In this experiment, we used a novel in vivo transcranial imaging approach to evaluate CSF tracer influx between wildtypes and Aqp4 KO mice in Figure 5A and B. We also included a Texas Red 3 kDa dextran tracer in addition to BSA-647 in Figure 5C-E.

8) In addition, they only looked at mean pixel intensity (MPI) relative to bregma (Figure 5I) and depth was not taken into consideration, unlike the data in other figures in this manuscript. It would have been beneficial to look at MPI respective to definitive depths over a relative distance. This would be a more direct measurement of extent or depth of penetration.

We thank the reviewer for this suggestion, we have now included a tracer penetration depth analysis in Figure 5H and I.

9) OHSU used the Snta1 KO mouse line to access the role of perivascular astrocytic localization of Aqp4 in mediating CSF flux into the parenchyma using DCE-MRI following intracisternal injection of gadoteridol. Figure 6 successfully demonstrates reduced accumulation of Gad in Snta1 KO mice across multiple brain regions.

10) However, Gad is a contrast agent and is not a direct measure of CSF flow itself, thus the conclusion that α-syntrophin facilitates CSF-ISF exchange is not clearly supported.

Gadoteridol is similar to the fluorescent tracers used in optical imaging studies here employed as a CSF tracer. It is not possible to image water itself and tritiated water moves too quickly across the BBB (seconds) for data collection. Thus, all prior studies are similar to this study based on quantification of tracer dispersion.

11) Also, again there is potential confusion created by using both the terms "injection" and "infusion" to describe the experiments (subsection “OHSU: Reduced glymphatic CSF influx in Snta1 KO mice with loss of polarized expression of AQP4 in vascular endfeet of astrocytes”), this should be corrected.

Thanks, we have in the revised manuscript restricted our use to the term “injection”.

12) All of the experiments done in this paper involve penetration of tracers from CSF injection/infusion into brain parenchyma. None of the experiments directly tests the clearance of intraparenchymally infused agents from brain to CSF. This was a prominent finding of Smith et al. (Smith et al., 2017), that there was no significant difference in clearance of solutes in AQP4 KO mice following intraparenchymal infusion. In other words all experiments in this manuscript are CSF to brain, but the claims are bidirectional (Figure 7) that there is also decreased ISF clearance from brain to CSF in the absence of AQP4. Why did none of the current experiments investigate or substantiate this claim?

We thank the reviewer for this question. We have in the past avoided the use of intraparenchymal injections because the injections (which require opening of the skull and tissue damage) affects CSF tracer influx. Prompted by the reviewer’s question, we have now included a new Figure 7 that formally documents that the invasive methodology utilized by Smith et al., profoundly alters CSF influx and therefore cannot be utilized to study bulk flow clearance in brain.

13) The Discussion reiterates the fact that the UNC group were unable to replicate Smith's findings however they did not follow Smith's methodologies so reproducibility is debatable since methods deviated.

The Smith at al. publication was presented as a replication study: i.e. the title ("Test of the 'glymphatic' hypothesis demonstrates diffusive and aquaporin-4-independent solute transport in rodent brain parenchyma". Please refer to the Discussion of the revised manuscript for a discussion on what a proper replication study is. Our study did not intend to replicate Smith et al., who used (i) an inappropriate anesthetic regiment for study of the glymphatics, (ii) used a novel and untested approach for the cisterna magna injections based on variable injection rate, and (iii) prepared cranial burr holes and introduced glass pipettes to study the glymphatic system. We have, in response to the reviewer, clearly stated that the aim of our study was not to replicate the Smith et al. study. The strength of our study is that it shows that AQP4 is an essential component of the glymphatic system, since 5 independent labs in 5 different transgenic mouse lines and six independent experiments found that deletion or mislocation of AQP4 reduced CSF tracer influx. The data presented in this manuscript is further supported by meta-analysis (Figure 8).

Discussion: “For these reasons, the present study is not a replication study of Smith et al. (Smith et al., 2017). Instead we used the methodology reported in Iliff et al., 2012 (Iliff et al., 2012) and further optimized it in subsequent studies (Neely et al., 2001; Brandt et al., 2014).”

14) An obvious experiment that would eliminate the effects of anesthesia would be to implant an infusion catheter in the cisterna magna and allow animals to recover for several days, and then carry out infusion of tracers in the complete absence of any anesthesia whatsoever and then rapidly perfuse at different time points and assess parenchymal penetration depth criteria. While this may be technically difficult in mice, stereotactically targeted infusions have been performed in awake behaving mice in many locations, and in any event this would be feasible in rats (and AQP4 KO rats have been described). This experiment would clearly move the field forward beyond disparate data relative to anesthetic usage.

We thank the reviewer for this suggestion. The proposed studies are not included here because our aim was to establish the importance of AQP4 in glymphatic activity under anesthesia, not during wakefulness when CSF inflow is very low (Xie et al., 2013). Unfortunately, Dr. Verkman has to our knowledge not yet validated that AQP4 was deleted in the rats utilized in Smith et al., 2017.

https://doi.org/10.7554/eLife.40070.022

Article and author information

Author details

  1. Humberto Mestre

    Center for Translational Neuromedicine, University of Rochester Medical Center, Rochester, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing, Performed meta-analysis
    Contributed equally with
    Lauren M Hablitz, Anna LR Xavier, Weixi Feng, Wenyan Zou, Tinglin Pu, Hiromu Monai, Giridhar Murlidharan, Ruth M Castellanos Rivera, Matthew J Simon, Martin M Pike, Virginia Plá, Ting Du and Benjamin T Kress
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5876-5397
  2. Lauren M Hablitz

    Center for Translational Neuromedicine, University of Rochester Medical Center, Rochester, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Humberto Mestre, Anna LR Xavier, Weixi Feng, Wenyan Zou, Tinglin Pu, Hiromu Monai, Giridhar Murlidharan, Ruth M Castellanos Rivera, Matthew J Simon, Martin M Pike, Virginia Plá, Ting Du and Benjamin T Kress
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6159-7742
  3. Anna LR Xavier

    Center for Translational Neuromedicine, Faculty of Medical and Health Sciences, University of Copenhagen, Copenhagen, Denmark
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Humberto Mestre, Lauren M Hablitz, Weixi Feng, Wenyan Zou, Tinglin Pu, Hiromu Monai, Giridhar Murlidharan, Ruth M Castellanos Rivera, Matthew J Simon, Martin M Pike, Virginia Plá, Ting Du and Benjamin T Kress
    Competing interests
    No competing interests declared
  4. Weixi Feng

    Jiangsu Province Key Laboratory of Neurodegeneration, Center for Global Health, Nanjing Medical University, Nanjing, China
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Humberto Mestre, Lauren M Hablitz, Anna LR Xavier, Wenyan Zou, Tinglin Pu, Hiromu Monai, Giridhar Murlidharan, Ruth M Castellanos Rivera, Matthew J Simon, Martin M Pike, Virginia Plá, Ting Du and Benjamin T Kress
    Competing interests
    No competing interests declared
  5. Wenyan Zou

    Jiangsu Province Key Laboratory of Neurodegeneration, Center for Global Health, Nanjing Medical University, Nanjing, China
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Humberto Mestre, Lauren M Hablitz, Anna LR Xavier, Weixi Feng, Tinglin Pu, Hiromu Monai, Giridhar Murlidharan, Ruth M Castellanos Rivera, Matthew J Simon, Martin M Pike, Virginia Plá, Ting Du and Benjamin T Kress
    Competing interests
    No competing interests declared
  6. Tinglin Pu

    Jiangsu Province Key Laboratory of Neurodegeneration, Center for Global Health, Nanjing Medical University, Nanjing, China
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Humberto Mestre, Lauren M Hablitz, Anna LR Xavier, Weixi Feng, Wenyan Zou, Hiromu Monai, Giridhar Murlidharan, Ruth M Castellanos Rivera, Matthew J Simon, Martin M Pike, Virginia Plá, Ting Du and Benjamin T Kress
    Competing interests
    No competing interests declared
  7. Hiromu Monai

    1. RIKEN Center for Brain Science, Wako, Japan
    2. Ochanomizu University, Tokyo, Japan
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Humberto Mestre, Lauren M Hablitz, Anna LR Xavier, Weixi Feng, Wenyan Zou, Tinglin Pu, Giridhar Murlidharan, Ruth M Castellanos Rivera, Matthew J Simon, Martin M Pike, Virginia Plá, Ting Du and Benjamin T Kress
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6975-7218
  8. Giridhar Murlidharan

    Gene Therapy Center, The University of North Carolina, Chapel Hill, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Humberto Mestre, Lauren M Hablitz, Anna LR Xavier, Weixi Feng, Wenyan Zou, Tinglin Pu, Hiromu Monai, Ruth M Castellanos Rivera, Matthew J Simon, Martin M Pike, Virginia Plá, Ting Du and Benjamin T Kress
    Competing interests
    No competing interests declared
  9. Ruth M Castellanos Rivera

    Gene Therapy Center, The University of North Carolina, Chapel Hill, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Humberto Mestre, Lauren M Hablitz, Anna LR Xavier, Weixi Feng, Wenyan Zou, Tinglin Pu, Hiromu Monai, Giridhar Murlidharan, Matthew J Simon, Martin M Pike, Virginia Plá, Ting Du and Benjamin T Kress
    Competing interests
    No competing interests declared
  10. Matthew J Simon

    Department of Anesthesiology and Perioperative Medicine, Oregon Health and Science University, Portland, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Humberto Mestre, Lauren M Hablitz, Anna LR Xavier, Weixi Feng, Wenyan Zou, Tinglin Pu, Hiromu Monai, Giridhar Murlidharan, Ruth M Castellanos Rivera, Martin M Pike, Virginia Plá, Ting Du and Benjamin T Kress
    Competing interests
    No competing interests declared
  11. Martin M Pike

    Advanced Imaging Research Center, Oregon Health and Science University, Portland, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Humberto Mestre, Lauren M Hablitz, Anna LR Xavier, Weixi Feng, Wenyan Zou, Tinglin Pu, Hiromu Monai, Giridhar Murlidharan, Ruth M Castellanos Rivera, Matthew J Simon, Virginia Plá, Ting Du and Benjamin T Kress
    Competing interests
    No competing interests declared
  12. Virginia Plá

    Center for Translational Neuromedicine, University of Rochester Medical Center, Rochester, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Humberto Mestre, Lauren M Hablitz, Anna LR Xavier, Weixi Feng, Wenyan Zou, Tinglin Pu, Hiromu Monai, Giridhar Murlidharan, Ruth M Castellanos Rivera, Matthew J Simon, Martin M Pike, Ting Du and Benjamin T Kress
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-5501-0789
  13. Ting Du

    Center for Translational Neuromedicine, University of Rochester Medical Center, Rochester, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Humberto Mestre, Lauren M Hablitz, Anna LR Xavier, Weixi Feng, Wenyan Zou, Tinglin Pu, Hiromu Monai, Giridhar Murlidharan, Ruth M Castellanos Rivera, Matthew J Simon, Martin M Pike, Virginia Plá and Benjamin T Kress
    Competing interests
    No competing interests declared
  14. Benjamin T Kress

    Center for Translational Neuromedicine, University of Rochester Medical Center, Rochester, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Contributed equally with
    Humberto Mestre, Lauren M Hablitz, Anna LR Xavier, Weixi Feng, Wenyan Zou, Tinglin Pu, Hiromu Monai, Giridhar Murlidharan, Ruth M Castellanos Rivera, Matthew J Simon, Martin M Pike, Virginia Plá and Ting Du
    Competing interests
    No competing interests declared
  15. Xiaowen Wang

    RIKEN Center for Brain Science, Wako, Japan
    Contribution
    Data curation, Investigation, Methodology
    Competing interests
    No competing interests declared
  16. Benjamin A Plog

    Center for Translational Neuromedicine, University of Rochester Medical Center, Rochester, United States
    Contribution
    Methodology
    Competing interests
    No competing interests declared
  17. Alexander S Thrane

    1. Center for Translational Neuromedicine, Faculty of Medical and Health Sciences, University of Copenhagen, Copenhagen, Denmark
    2. Department of Ophthalmology, Haukeland University Hospital, Bergen, Norway
    Contribution
    Conceptualization, Visualization, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
  18. Iben Lundgaard

    1. Center for Translational Neuromedicine, University of Rochester Medical Center, Rochester, United States
    2. Department of Experimental Medical Science, Lund University, Lund, Sweden
    3. Wallenberg Center for Molecular Medicine, Lund University, Lund, Sweden
    Contribution
    Conceptualization, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
  19. Yoichiro Abe

    Department of Pharmacology,School of Medicine, Keio University, Tokyo, Japan
    Contribution
    Resources, Funding acquisition, Validation, Investigation, Methodology, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6163-8794
  20. Masato Yasui

    Department of Pharmacology,School of Medicine, Keio University, Tokyo, Japan
    Contribution
    Resources, Funding acquisition, Validation, Investigation, Methodology, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
  21. John H Thomas

    1. Department of Mechanical Engineering, University of Rochester, Rochester, United States
    2. Department of Physics and Astronomy, University of Rochester, Rochester, United States
    Contribution
    Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
  22. Ming Xiao

    Jiangsu Province Key Laboratory of Neurodegeneration, Center for Global Health, Nanjing Medical University, Nanjing, China
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    mingx@njmu.edu.cn
    Competing interests
    No competing interests declared
  23. Hajime Hirase

    1. RIKEN Center for Brain Science, Wako, Japan
    2. Brain and Body System Science Institute, Saitama University, Saitama, Japan
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    hajime.hirase@riken.jp
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3806-6905
  24. Aravind Asokan

    1. Gene Therapy Center, The University of North Carolina, Chapel Hill, United States
    2. Department of Molecular Genetics and Microbiology, Duke University School of Medicine, North Carolina, United States
    3. Department of Surgery, Duke University School of Medicine, Durham, United States
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    aravind.asokan@duke.edu
    Competing interests
    No competing interests declared
  25. Jeffrey J Iliff

    1. Department of Anesthesiology and Perioperative Medicine, Oregon Health and Science University, Portland, United States
    2. Knight Cardiovascular Institute, Oregon Health and Science University, Portland, United States
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    iliffj@ohsu.edu
    Competing interests
    No competing interests declared
  26. Maiken Nedergaard

    1. Center for Translational Neuromedicine, University of Rochester Medical Center, Rochester, United States
    2. Center for Translational Neuromedicine, Faculty of Medical and Health Sciences, University of Copenhagen, Copenhagen, Denmark
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Visualization, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    maiken_nedergaard@urmc.rochester.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6502-6031

Funding

Japan Society for the Promotion of Science (18K14859)

  • Hiromu Monai

Knut och Alice Wallenbergs Stiftelse (Helse Vet)

  • Alexander S Thrane

Japan Society for the Promotion of Science (17K19637)

  • Yoichiro Abe

Japan Society for the Promotion of Science (16H05134)

  • Yoichiro Abe

Japan Society for the Promotion of Science (18H02606)

  • Masato Yasui

National Institute on Aging (RF1 AG057575-01)

  • John H Thomas
  • Maiken Nedergaard

Japan Society for the Promotion of Science (16H01888)

  • Hajime Hirase

Japan Society for the Promotion of Science (18H05150)

  • Hajime Hirase

Human Frontier Science Program (RGP0036/2014)

  • Hajime Hirase

Japan Society for the Promotion of Science (Core-to-Core Program)

  • Hajime Hirase

Lundbeckfonden (Visiting Professorship)

  • Hajime Hirase

National Institutes of Health (NS061800)

  • Aravind Asokan

National Institutes of Health (NS099371)

  • Aravind Asokan

National Institutes of Health (HL089221)

  • Aravind Asokan

National Institutes of Health (NS089709)

  • Jeffrey J Iliff

National Institutes of Health (AG054456)

  • Jeffrey J. Iliff

National Institutes of Health (NS100366)

  • Maiken Nedergaard

National Institutes of Health (NS078394)

  • Maiken Nedergaard

National Institutes of Health (AG048769)

  • Maiken Nedergaard

Dr. Miriam and Sheldon G. Adelson Medical Research Foundation

  • Maiken Nedergaard

EU Joint Programme – Neurodegenerative Disease Research (643417/DACAPO-AD)

  • Maiken Nedergaard

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

This study was funded by the NIH (NS100366, NS078394, NS089709, NS061800, AG048769, AG054456, AG054093, NS099371 and HL089221), NNSF81671070, the Knut and Alice Wallenberg Foundation, Western Norway Regional Health Authority (Helse Vest). RIKEN Center for Brain Science (formerly, RIKEN Brain Science Institute), KAKENHI grants (18K14859 to HM, 16H01888 and 18H05150 to HH, 17K19637 and 16H05134 to YA, 18H02606 to MY), HFSP (RGP0036/2014 to HH), JSPS Core-to-Core Program: A Advanced Research Networks, HH is a recipient of the Lundbeck Foundation Visiting Professorship. We thank Dan Xue for expert graphic illustrations, Hayley Martin for assistance with meta-analysis, Amanda M Sweeney and Genaro E Olveda for expert technical assistance, and Paul Cumming for comments on the manuscript.

Ethics

Animal experimentation: All experiments were approved by the Institutional Animal Care and Use Committee of Nanjing Medical University (IACUC-1601106), Wako Animal Experiment Committee, RIKEN (Recombinant DNA experimentation protocol: 2016-038; Animal experimentation protocol: H29-2-227), The University of North Carolina at Chapel Hill Institutional Animal Care and Use Committee (protocol 15-109), the University Committee on Animal Resources of the University of Rochester (protocol 2011-023), and the IACUC of Oregon Health & Science University (protocol IP00000394). All experiments were performed in accordance with the approved guidelines and regulations. All efforts were made to minimize animal suffering and to reduce the number of animals used for the experiments.

Senior Editor

  1. Sean J Morrison, Howard Hughes Medical Institute, University of Texas Southwestern Medical Center, United States

Reviewing Editor

  1. David Kleinfeld, University of California, San Diego, United States

Publication history

  1. Received: July 20, 2018
  2. Accepted: December 17, 2018
  3. Accepted Manuscript published: December 18, 2018 (version 1)
  4. Accepted Manuscript updated: December 19, 2018 (version 2)
  5. Version of Record published: December 27, 2018 (version 3)
  6. Version of Record updated: December 28, 2018 (version 4)

Copyright

© 2018, Mestre et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 3,106
    Page views
  • 712
    Downloads
  • 23
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

  1. Further reading