Factors determining the density of AQP4 water channel molecules at the brain–blood interface

We used adult male and newborn C57BL/6 mice (Jackson Laboratories, Boulder, CO) and two previously described transgenic mouse strains. Aqp4−/− animals were generated as described (Thrane et al. 2011) and backcrossed 10 times on a C57BL/6 background; α-syntrophin−/− (α-Syn−/−) animals were generated as described (Adams et al. 2000). The mice were allowed ad libitum access to food and drinking water. Animal experiments were performed according to the European Council law on protection of laboratory animals, with the approval of the University of Oslo’s Animal Care and Use Committee. Every effort was made to minimise the number of animals.

A list of all antibodies used in this study is shown in Table 1.

Adult animals were anaesthetised with a single intraperitoneal injection of Equithesin (150 μL) and transcardially perfused using 4 % formaldehyde in 0.1 M phosphate buffer (PB) for 15 min with an initial 20 s perfusion with ice cold 2 % dextran in 0.1 M PB. Prompt neck stiffness was used as an indication of satisfactory perfusion. Brains were removed, left in the fixation solution overnight and stored in a 1:10 dilution of the same solution in 0.1 M PB. For newborn (P0) animals, both perfusion velocity and anaesthesia dosage were titrated according to weight and animals were initially sedated using chloroform gas. We employed a modified version of the pH-shift perfusion technique (Eilert-Olsen et al. 2012), initially defined by Berod et al. (1981), for those animals processed further for high resolution electron microscopy to obtain better ultrastructure and minimise epitope loss. This allowed for one half of the brain being used for light and immunofluorescence microscopy while the other was used for embedding in Lowicryl resin.

Brains (n = 4 per group, n = 2 for each specificity control) were cut into 0.5–1.0 mm slices, regions were dissected, cryoprotected, quick-frozen in liquid propane (−170 °C), and subjected to freeze substitution. Specimens were embedded in methacrylate resin (Lowicryl HM20) and polymerised by UV light below 0 °C. Ultrathin sections (70–100 nm) were cut using an Ultratome (Reichert Ultracut S, Leica) and placed on 300 mesh grids. Immunogold labelling was carried out as previously described (Promeneur et al. 2013). Briefly, sections were rinsed in Tris-buffered saline with Triton X-100 (TBS-T; 5 mM Tris–HCl, 0.3 % NaCl, 0.1 % Triton X-100), incubated in 2 % human serum albumin (HSA), followed by primary antibody (anti-AQP4 or anti-α-syntrophin) overnight, secondary antibody (15 nm gold) for 90 min, and contrasted with 2 % uranyl acetate for 90 s and 0.3 % lead citrate for 90 s. The sections were examined using a Tecnai 12 electron microscope at 80 kV. The examiner was blinded for animal genotype. Appropriate knockout tissue was used as negative control (Aqp4−/− and α-Syn−/− mice) for the two immunogold experiments.

A list of the regions dissected and analysed in this study is shown in Table 2.

Quantitative analysis was performed as previously described (Amiry-Moghaddam et al. 2004; Lunde et al. 2015). Briefly, images of 20–30 capillaries were acquired from each subregion present on each section. Care was taken to acquire images so that a similar distance of astrocyte membrane abutting on endothelium and pericytes was shown on each picture. We defined inclusion and exclusion criteria for capillaries, astrocyte endfeet and pericytes prior to image collection. In particular, a pericyte was defined as a perivascular cell surrounded by a clearly defined ad- and abluminal basal lamina. Linear densities of gold particles over astrocyte membranes were determined by an extension of analysis [Soft Imaging Systems (SIS), Münster, Germany]. Linear densities were determined semi-automatically and transferred to SPSS Version 22 (SPSS, Chicago, IL) for statistical analysis.

Mouse brains (n = 3 per group) were cryoprotected and sectioned as described above. Sections were washed with PBS, exposed to 2 % H₂O₂ for 10 min to deplete endogenous peroxidase activity, and washed three times with PBS. Blocking was done using 2 % BSA in PBS-T (PBS, 0.1 % Triton X-100). Incubation with primary antibody diluted in blocking buffer was done overnight in cold room (4 °C), followed by three washes with PBS-T. A 1 h incubation with secondary antibody in blocking buffer ensued, followed by three washes with PBS-T, incubation for 1 h with biotinylated-streptavidin horseradish peroxidase complex (GE healthcare UK limited, Rpn 1051 V) diluted 1:100 in PBS-T, three washes with PBS-T, then three washes with PBS. Sections were then incubated for 5 min with 0.5 mg/mL diaminobenzidine (3,3-diaminobenzidine tetrahydrochloride 10 mg/tab, Sigma D5905, DAB) dissolved in 0.1 M PB and exposed to DAB solution containing 0.01 % H₂O₂ (90 s). The DAB/HRP reaction was stopped by three washes with 0.1 M PB. Sections were mounted using glycerine gelatine, scanned using a Mirax scanner (Zeiss) and examined using a light microscope.

WT and α-Syn−/− mice were anaesthetised with isofluorane and decapitated. Brains were dissected out and split in two with a midline, sagittal cut. Regional dissections (n = 4) and whole brain samples (n = 4) were processed for quantitative RT-PCR analysis by overnight incubation in RNAlater (Ambion) and storage at −80 °C until further processing. Regional dissections (n = 4) from parietal cortex and colliculus inferior were obtained and processed for western blotting. Samples were put into 2 mL Lysing matrix D (MP Biomedicals) Eppendorf tubes containing RIPA (radio-immunoprecipitation assay) homogenisation buffer with protease inhibitor cocktail (Roche) added immediately before use, samples were homogenised in a FastPrep FP120 Cell Disrupter (MP Biomedicals), shaken at intervals and intermittently cooled on ice. Supernatant of a 9300 × g spin for 30 min at 4 °C was used for immunoblotting; pellets were frozen at −80 °C. Protein concentrations were determined with a DC protein assay (Bio-Rad) according to the manufacturer’s instructions. Equal numbers of right and left hemispheres were included in each analysis.

Gels were cast using the Mini-Protean III cell from Bio-Rad as previously described (Sorbo et al. 2007). Both stacking and resolving gels were cast with added 3 M urea. 12 % SDS-PAGE gels were run at 150 V for 60 min, blotted with the Criterion cell (Bio-Rad), equilibrated with Towbin buffer (25 mM Tris, 192 mM glycine) supplemented with 20 % methanol and blotted onto 0.2 μm pore size PVDF membranes (Bio-Rad) at 100 V for 40 min. Blots were rinsed briefly in TBS-T (20 mM Tris; 137 mM NaCl; 0.05 % Tween-20 (Sigma), 0.05 % NaN₃; pH = 7.6) prior to incubation for 30 min with 5 % non-fat dried milk powder (Applichem) dissolved in TBS-T (blocking buffer). Blots were incubated with rabbit anti-AQP4 (Sigma) diluted in blocking buffer overnight at 4 °C, then washed in TBS-T, incubated in a 1:5000 dilution in TBS-T of alkaline phosphatase conjugated to mouse monoclonal anti-Rabbit IgG (A2556 Sigma) for 1 h at room temperature and subsequently washed in TBS-T for 3 h. Finally, blots were incubated for 15 min with the ECF substrate (Amersham, ECF Kit), rinsed briefly in TBS-T, dried and scanned with a Kodak desktop scanner (Kodak Image Station 4000 MM PRO). Certain blots were re-probed with an anti-actin antibody (Sigma). Briefly, blots were immersed in methanol, incubated with blocking buffer for 30 min, then incubated with anti-actin diluted in blocking buffer at 4 °C overnight, rinsed and scanned as described above for the initial primary antibody. All samples were heat treated for 10 min at 37 °C, spun at 13,000 rpm for 10 min, resuspended and loaded onto the gel. All samples were assayed for total protein content using the DC-kit (Bio-Rad) with BSA diluted in H₂O as standard. 5 μg of protein was loaded in each well, except for control tissue (Aqp4−/−) where 10 μg was added. Precision plus protein standards Dual colour (Bio-rad) was used as Mw ladder.

Quantitative RT-PCR was carried out as described previously (Lunde et al. 2015). In brief, total RNA was extracted using RNeasy Lipid Tissue kit (Qiagen). 2 μg of RNA from whole brain and 100–1000 ng of RNA from regions were reverse-transcribed using oligo d(T)18 primer and RevertAid H Minus First Strand cDNA transcription reagents (Fermentas). The real-time PCR was set up using 2 μL template, corresponding to 5 ng cDNA, in a reaction volume of 20 μL containing AB Power SYBR Green PCR Master Mix (Applied Biosystems) and primers (200 nM) for detection of mouse Aqp4: TTTGGACCCGCAGTTATCAT (forward) and GTTGTCCTCCACCTCCATGT (reverse). Expression was assessed by absolute quantification using the StepOnePlus real-time PCR system (Applied Biosystems). Tbp (ACGGACAACTGCGTTGATTT (forward); CAAGGCCTTCCAGCCTTATAG (reverse)) was used as endogenous control for normalisation of gene expression. SPSS was used for statistical analysis. T test was used for comparing Aqp4 expression in WT and α-Syn−/− mice. For comparing expression among regions, one-way ANOVA Bonferroni post hoc was used.

Light microscopic immunohistochemistry shows distinct caudo-rostral and ventro-dorsal gradients in AQP4 immunolabelling. Thus, labelling is notably stronger in cerebellum, olfactory bulb, spinal cord and brainstem than in cerebral cortex and striatum (Fig. 1). In the adult brain, AQP4 labelling is concentrated around blood vessels and subjacent to pia. A regional heterogeneity in AQP4 distribution is evident in the adult brain (Fig. 1a) as well as postnatally (Fig. 1b; also see Lunde et al. 2015). However, at this early stage the AQP4 immunolabelling is not concentrated around vessels or subpially, but rather evenly distributed in the neuropil. Absence of labelling in Aqp4−/− mice attests to the selectivity of the immunolabelling (Fig. 1b inset).

Quantitative, high-resolution AQP4 immunogold histochemistry on ultrathin sections confirms and expands observations from light microscopy. AQP4 is found to be constitutively expressed in pericapillary endfeet and enriched in the endfoot membranes where levels display significant regional variation (Fig. 2a). Cortex (CX) and molecular layer of cerebellum (CB-m) display sparse labelling in contrast to the intense labelling of regions such as inferior colliculus (IC) and cerebellar granular layer (CB-gran). Linear density of gold particles along perivascular endfoot membranes was determined for capillaries of cortex, hippocampus, cerebellum, inferior colliculus and spinal cord and provides evidence of significant region-specific variations. Heterogeneity is also present within subregions of the same anatomical region, shown here for hippocampus and cerebellum (Fig. 2b). No statistically significant difference could be found between grey and white matter regions of the spinal cord. In cerebellum, AQP4 density in the molecular layer is considerably lower than that in the granular layer and white matter. Within the hippocampal region, perivascular immunogold density is significantly lower in dentate gyrus than in stratum pyramidale and stratum radiatum. When the immunogold data is grouped according to region, the spinal cord and neocortex show the highest and lowest mean linear densities, respectively (14.80 vs. 9.82 particles per um; Fig. 4a, left, b, top).

Due to prior evidence implicating α-syntrophin as an anchoring molecule of endfoot AQP4 (Neely et al. 2001), we assessed AQP4 localisation in α-Syn−/− mice. Genetic deletion of α-syntrophin leads to a pronounced reduction in perivascular AQP4 labelling in all regions and subregions analysed, as exemplified by images from cerebellum (Fig. 3). The representative image from an α-Syn−/− mouse (Fig. 3b) suggests a redistribution of AQP4 whereby perivascular and subpial AQP4 immunostaining in α-Syn−/− animals is not detectable, but staining of astrocyte processes surrounding Purkinje cells appears more intense than in wild-type mice (Fig. 3a). Immunogold detection of AQP4 is more sensitive than immunofluorescence, and—albeit for the notable reduction (79–94 %)—demonstrates residual pericapillary AQP4 immunogold labelling in α-Syn−/− animals (Fig. 4a). Immunogold labelling was abolished in control sections from Aqp4−/− mice (specificity control, not shown).

The present analysis is restricted to the perivascular labelling. It should be noted, however, that the granular layer of the cerebellum contains a sizeable non-endfoot AQP4 pool that is largely independent of α-syntrophin and dystrophin (Fig. 3, see Amiry-Moghaddam et al. 2004; Nicchia et al. 2008b).

Residual AQP4 labelling around capillaries was quantified by immunogold histochemistry (Fig. 4a–c). Strikingly, in α-Syn−/− mice, perivascular AQP4 immunogold linear density in spinal cord is more than fourfold higher than in cortex (Fig. 4c). Following α-syntrophin gene deletion, cortex remains the region of lowest AQP4 linear density (mean 0.58 particles/um) while spinal cord still displays the highest perivascular labelling (3.16 particles/um). There is thus compelling evidence for retained, heterogeneous expression of AQP4 in α-Syn−/− mice.

The relative AQP4 loss following α-syntrophin gene deletion is most pronounced in CX (94 %) followed by IC (89 %), CB (87 %), HC (83 %) and SC (79 %). The nominal loss in AQP4 linear density following α-syntrophin deletion, calculated as difference between means of linear density of AQP4 in knockout vs. wildtype (Fig. 4b), is greatest for IC and SC (11.71 and 11.64 particles/um, respectively), followed by CB (10.96), HC (9.83) and CX (9.25).

Perivascular α-syntrophin immunogold labelling (Fig. 4d) in WT mice shows how α-syntrophin levels correlate positively with that of AQP4 density (Fig. 4b) in endfoot membranes. Capillaries of cortical regions display the lowest density of α-syntrophin followed by hippocampus, cerebellum, spinal cord and inferior colliculus. As illustrated (Fig. 4b, d), the ratio of AQP4 to α-syntrophin varies little between regions. The calculated nominal ratio between AQP4 and α-syntrophin (in terms of immunogold particles) is 5.82 for cortex, 5.89 for hippocampus, 4.84 for cerebellum, 4.54 for inferior colliculus and 5.87 for spinal cord. Labelling was abolished in control sections from α-Syn−/− mice (specificity control, not shown).

Shown in Fig. 5a is quantitative evidence of regional differences in Aqp4 mRNA for both WT and α-Syn−/− genotypes. For WT animals, the mean copy number of Aqp4 mRNA is highest in cerebellum, followed by inferior colliculus, whole brain, then hippocampus and cortex. Similarly, for α-Syn−/− animals, the mean copy number of Aqp4 mRNA is highest in cerebellum followed by inferior colliculus, whole brain, hippocampus and cortex. For WT animals, the number of mRNA copies detected in cerebellum is fourfold that detected in cortex (p < 0.001), and the amount of mRNA copies detected in inferior colliculus is twice as high as in cortex (p < 0.001). We find no statistical difference between hippocampus and cortex in regard to total copy numbers of Aqp4 mRNA. This is true for WT as well as α-Syn−/− mice. Notably, there is no difference between Aqp4 expression in whole brain samples (p = 0.188) when comparing WT and α-Syn−/−. The same holds true for all regions (from CB, IC, HC and CX).

Immunoblots from inferior colliculus (IC) and cortex demonstrate two major bands at the expected molecular weight, corresponding to M1 and M23 isoforms of AQP4, respectively (Fig. 5b, c). Both bands appear stronger in the immunoblot containing homogenates from inferior colliculus (Fig. 5b) compared to those from cortex (Fig. 5c) and provide qualitative data in accordance with the quantitative real-time PCR analysis. Anti-actin immunoblot on cortical samples is included to demonstrate presence of protein in Aqp4−/− mice. Importantly, no difference in AQP4 labelling patterns or intensity can be seen between WT and α-Syn−/− mice, as assessed for cortex and inferior colliculus (Fig. 5b, c). This is in line with earlier studies, indicating that deletion of α-syntrophin causes a mislocalisation rather than a loss of AQP4 from brain tissue (Neely et al. 2001).

As measured by quantitative immunogold labelling in WT mice, the overall AQP4 linear density is 15 % higher for endfoot membrane domains abutting pericytes than for endfoot membrane domains abutting endothelium (Fig. 6b, p < 0.001). This enrichment is statistically significant and present in all examined regions (p < 0.05 for CX, HC, SC and p < 0.001 for CB and IC), ranging from a factor of 1.26 higher in cortex to 1.08 in spinal cord. Nominally, the difference in linear density ranges from 2.60 (IC) to 1.01 (HC) gold particles per μm. We next investigated to what extent the regional, non-uniform expression of AQP4 in pericapillary endfeet was dependent on α-syntrophin. The difference between adjacent membrane domains remains following α-syntrophin gene deletion. Overall AQP4 linear density is 16 % higher in endfeet abutting pericytes compared to adjacent membranes abutting endothelium (Fig. 6c, p < 0.05) in α-Syn−/− animals. The nominal difference of linear density between these two membrane domains is decreased in α-Syn−/− mice, however, and a statistically significant difference (p < 0.05) was only observed in CX (Fig. 6c). As is the case for AQP4, α-syntrophin is also enriched in endfoot domains abutting pericytes compared to those abutting endothelium, a difference which reaches statistical significance for all examined regions (Fig. 6d, p < 0.001). The relative difference in α-syntrophin linear density between pericyte and endothelium abutting endfoot membrane domains is constant across examined regions, by a factor ranging from 1.27 (CB) to 1.38 (HC).

Based on our quantitative immunogold data, we conclude that perivascular endfoot membranes differ considerably with respect to the amount of AQP4 they contain. Differences occur between regions and subregions, as well as on the level of individual capillaries (with endfoot membranes facing pericytes containing more AQP4 than endfoot membranes facing endothelia). This suggests that there is a corresponding non-uniformity in the capacity of water transport. AQP4 is the only aquaporin that is known to be concentrated in endfoot membranes. However, there is not necessarily a strict correlation between pericapillary AQP4 expression and water transport capacity. Thus, we cannot rule out the possibility that AQP4 is differentially regulated, depending on region, subregion or membrane microdomain, Several factors have been identified that affect water flux through AQP4 (Yukutake et al. 2008; Yukutake and Yasui 2010; Zelenina 2010) although direct evidence for bona fide gating is still missing (Nagelhus and Ottersen 2013). Available evidence suggests that the capacity for water transport at the brain–blood interface determines the propensity for oedema formation as well as the capability for oedema resolution (Amiry-Moghaddam et al. 2003a; Papadopoulos et al. 2004; Zeynalov et al. 2008).

By use of quantitative immunogold analyses in combination with genetic manipulation we showed that a major fraction of the AQP4 pool in endfoot membranes is dependent on the presence of α-syntrophin. While it has long been known that α-syntrophin serves as a molecular anchor of endfoot AQP4 (Neely et al. 2001), this study is the first to show that the densities of α-syntrophin and AQP4 appear to be stoichiometrically coupled. Thus, the nominal ratio between the two molecules in terms of their immunogold labelling intensities varies little among regions (4.54 for inferior colliculus, 4.84 for cerebellum, 5.82 for cortex, 5.87 for spinal cord and 5.89 for hippocampus). Obviously, this ratio does not necessarily reflect the actual molecular ratios, as AQP4 and α-syntrophin may differ in regard to their labelling efficiency (defined as number of gold particle per target molecule). A more direct assessment of the actual molecular ratio (through biochemical approaches) must await the development of techniques for selective isolation of endfoot membranes. However, the uniformity in terms of AQP4/α-syntrophin gold particle ratios is consistent with the idea that the complement of AQP4 in the endfoot membrane is governed by the number of α-syntrophin molecules that are targeted to this membrane.

Our finding of a fairly uniform AQP4/α-syntrophin ratio does not necessarily imply that α-syntrophin is the only anchoring molecule for AQP4. Indeed, confirming previous observations (Amiry-Moghaddam et al. 2004), we show that there is a residual pool of the astrocyte endfoot AQP4 following targeted deletion of α-syntrophin. Here we extend these observations by demonstrating that the size of the residual pool varies in proportion to the amount of AQP4 in WT endfeet. In other words, the percentage of perivascular AQP4 removed by deletion of α-syntrophin does not differ significantly between regions. The most probable explanation is that the residual AQP4 pool is anchored by a molecule whose expression is proportional to that of α-syntrophin. This unknown anchoring protein may well be another syntrophin, such as β-syntrophin (Puwarawuttipanit et al. 2006). This hypothesis will be explored once β-syntrophin knockout animals are made available.

α-Syntrophin and β-syntrophin are parts of a multimember molecular assembly—the dystrophin associated protein complex (DAPC; Frigeri et al. 2001; Neely et al. 2001). Mdx mice that lack the brain dystrophin DP71 also lack a sizeable endfoot pool of AQP4 (Frigeri et al. 2001; Vajda et al. 2004). The DAP complex is held in place through its attachment to laminin (Culligan et al. 2001; Neely et al. 2001). This explains why there is such a sudden drop in AQP4 as soon as the endfoot plasma membrane is reflected away from the capillary.

In line with the important role of DAPC in AQP4 anchoring, deficient O-glycosylation of α-dystroglycan (a member of DAPC) causes a reduction in perivascular AQP4 (Michele et al. 2002; Rurak et al. 2007). Further, agrin (which interacts with α-dystroglycan) is required for normal expression of AQP4 in brain endfeet. It is important to note, however, that the brain also contains DAPC-independent pools of AQP4. One such pool is found in cerebellar astrocytes Nicchia et al. 2008b). This complexity notwithstanding, the present data show that α-syntrophin is a major player in controlling the expression of AQP4 at the brain–blood interface.

Our quantitative RT-PCR analysis revealed pronounced regional differences in the amount of Aqp4 mRNA. The Aqp4 transcript levels were high in those regions that displayed high levels of AQP4 in perivascular endfeet and were not sensitive to α-syntrophin deletion.

Superimposed on the regional differences in the size of the endfoot AQP4 pool there is a heterogeneity in AQP4 distribution within individual endfoot membrane domains. Specifically, the densities of AQP4 and α-syntrophin immunogold particles are higher where endfoot membranes abut pericytes than where they abut endothelium. This is true not only for cortical astrocytes (Gundersen et al. 2013) but for all regions included in the present analysis. The local increase in the amount of AQP4 and α-syntrophin could be a direct effect of pericytes—through secretion of soluble factors or extracellular matrix molecules. As pericytes are contractile, they could also affect contiguous membranes mechanically. This is of relevance, as a recent cell-culture study from our laboratory showed increased AQP4 levels in astrocyte processes subjected to mechanical stress (Camassa et al. 2015).

Taken together, our data suggest that α-syntrophin is a key factor in regulating the AQP4 pool of perivascular endfoot membranes. Drawing on quantitative, high resolution immunogold analyses, we provide evidence that the number of AQP4 molecules is stoichiometrically coupled to that of α-syntrophin. A definition of the exact molecular ratio between these molecules must await the development of appropriate techniques for isolating endfoot membranes. That AQP4 and α-syntrophin covary is not a coincidental correlation but reflects a causality: removal of α-syntrophin via genetic deletion removes a substantial fraction of endfoot AQP4 that varies little across regions. What determines the size of the AQP4 pool and hence the endfoot’s water fluxing capacity, therefore, is the number of α-syntrophin molecules that are targeted to the endfoot membrane. The alternative explanation—that α-syntrophin regulates Aqp4 expression through regulation at the transcriptional level—is ruled out by our finding that the number of Aqp4 mRNA copies remains constant in α-syntrophin knockout animals.

Our experiments show that the level of AQP4 in endfeet is correlated to the level of Aqp4 mRNA in the corresponding region. This observation does not necessarily imply that the level of endfoot AQP4 is controlled at the transcriptional level. Thus, the transcriptional regulation of Aqp4 is complex (Pisani et al. 2011; Rossi et al. 2010), requiring further studies.