Astroglial endfeet exhibit distinct Ca2+ signals during hypoosmotic conditions

Astrocytic endfeet cover the brain surface and form a sheath around the cerebral vasculature. An emerging concept is that endfeet control blood–brain water transport and drainage of interstitial fluid and waste along paravascular pathways. Little is known about the signaling mechanisms that regulate endfoot volume and hence the width of these drainage pathways. Here, we used the genetically encoded fluorescent Ca2+ indicator GCaMP6f to study Ca2+ signaling within astrocytic somata, processes, and endfeet in response to an osmotic challenge known to induce cell swelling. Acute cortical slices were subjected to artificial cerebrospinal fluid with 20% reduction in osmolarity while GCaMP6f fluorescence was imaged with two‐photon microscopy. Ca2+ signals induced by hypoosmotic conditions were observed in all astrocytic compartments except the soma. The Ca2+ response was most prominent in subpial and perivascular endfeet and included spikes with single peaks, plateau‐type elevations, and rapid oscillations, the latter restricted to subpial endfeet. Genetic removal of the type 2 inositol 1,4,5‐triphosphate receptor (IP3R2) severely suppressed the Ca2+ responses in endfeet but failed to affect brain water accumulation in vivo after water intoxication. Furthermore, the increase in endfoot Ca2+ spike rate during hypoosmotic conditions was attenuated in mutant mice lacking the aquaporin‐4 anchoring molecule dystrophin and after blockage of transient receptor potential vanilloid 4 channels. We conclude that the characteristics and underpinning of Ca2+ responses to hypoosmotic stress differ within the astrocytic territory and that IP3R2 is essential for the Ca2+ signals only in subpial and perivascular endfeet.

Recently, it was shown that AQP4 water channels facilitate removal of water (Haj-Yasein et al., 2011) and solutes (Iliff et al., 2012) from the brain parenchyma. A glio-vascularglymphatic pathwaywas proposed to drain excess ISF and waste products, similarly to the lymph system in other organs and tissues. Even though several aspects of the glymphatic hypothesis are controversial (Holter et al., 2017;Jin, Smith, & Verkman, 2016;Smith & Verkman, 2018;Smith, Yao, Dix, Jin, & Verkman, 2017;Tarasoff-Conway et al., 2015), it underscores the potential importance of astrocytic endfeet in brain fluid and volume homeostasis. A deeper understanding of the mechanisms of astroglial volume control may move the field forward.
Until recently, we lacked technology for studying Ca 2+ signals in the distal astrocytic processes and endfeet of adult animals. Studies using synthetic Ca 2+ dye loading mainly reported on Ca 2+ signals in cell bodies and the most proximal processes, leaving~90% of the astrocytic territory unsampled (Tong, Shigetomi, Looger, & Khakh, 2013). Furthermore, dye loading is inefficient in adult animals (ibid.). Here, we took advantage of the ultrasensitive genetically encoded Ca 2+ sensor GCaMP6f (Chen et al., 2013) to assess hypoosmotically induced Ca 2+ signals in astrocytic somata, fine processes, and endfeet in acute cortical slices from adult mice. We report that hypoosmotic stress elicits robust astrocytic Ca 2+ responses, whose signature and underpinning vary within the astrocytic territory.
Experimental groups contained at least four animals. All procedures were approved by the national animal use and care committee (Norwegian Food Safety Authority).

| Plasmid constructs and virus transduction
The pAAV-GFAP-GCaMP6f was constructed as described before Tang et al., 2009). For virus infections, the mice were anesthetized with a mixture of zolazepam (188 mg/kg), tiletamine (188 mg/kg), xylazine (4.5 mg/kg), and fentanyl (26 μg/kg) before viruses were stereotactically injected. The injections were done in somatosensory cortex bilaterally at the following stereotactic coordinates relative to Bregma: anteroposterior −0.8 mm, lateral 2.0 mm. During each injection, 0.28-0.35 μL of purified rAAV was delivered at 0.5 mm depth relative to the cortical surface.

| Slice preparation
Acute cortical slices were prepared from adult mice 2-6 weeks after virus transduction. Mice were euthanized with isoflurane (Baxter).
The temperature in the resting chamber and during recordings was kept at 31 C.

| Two-photon Ca 2+ imaging
Slices were let to rest for 1 hr before they were transferred to a recording chamber (2 mL) where they were held in place by a horseshoeshaped platinum wire and continuously perfused with aCSF. The GCaMP6f fluorescence was imaged by a two-photon laser scanning microscope (model Ultima; Prairie Technologies) with a 25×, 1.05 numerical aperture, water-immersion objective (XLPLN 25×WMP, Olympus), using 900 nm laser pulses for excitation. The laser used was either a Chameleon Vision II (Coherent) or a Mai Tai DeepSee (Spectra Physics). Depending on the expression level of the fluorescent indicator and the tissue depth at which imaging was performed, the laser power applied ranged from 5 to 25 mW (measured by LaserCheck, Coherent). Images were collected from cortical layers I-III at least 40 μm below the section surface. After imaging in normal aCSF, the perfusion solution was switched to hypoosmotic aCSF, in which the concentration of NaCl was reduced from 124 to 93 mM, thus reducing the osmolarity by 20% (verified by freezing-point depression). In a subset of experiments (data in Supporting Information) the NaCl was reduced to give 10 or 5% reduction in aCSF osmolarity. Time series were taken with frame rates ranging from 1 to 3 Hz. In some experiments, 1 μM of the TRPV4 channel antagonist HC067047 (2-methyl-1- [3-(4-morpholinyl) propyl]-5-phenyl-N-[3-(trifluoromethyl)phenyl]-1H-pyrrole-3-carboxamide; Tocris Bioscience, stock solution 10 mM in DMSO), was added to both the normosmotic and the −20% hypoosmotic aCSF.

| Image analysis
Time series of fluorescence images were imported to Fiji ImageJ (Fiji).
Regions of interest (ROIs) were selected manually based on cell morphology from ΔF/F images and SD images (see Section 2.7). Fluorescence was measured in four different astrocytic compartments: somata, fine processes, perivascular endfeet, and subpial endfeet.
ROIs over the fine processes were sampled at least 5 μm, and maximum 30 μm, away from the soma. Larger branches were avoided. For perivascular endfeet, the minimum distance between ROIs was set to 10 μm to avoid that the same endfoot was measured in two different ROIs. The distinction between different subpial endfeet was based on morphology.
2.6 | Induction of brain edema and measurement of brain water content by the wet/dry weight method Wildtype and Itpr2 −/− mice received buprenorphine (2 mg/kg) prior to induction of brain edema/decapitation. Brain edema was induced by intraperitoneal injection of distilled water (10% of body weight) under brief isoflurane anesthesia (2-3%). A subset of mice received intraperitoneal injection of isotonic saline (10% of body weight). The mice were allowed to wake up in separate cages. Forty minutes after water injection the mice were sacrificed by cervical dislocation. The brain was taken out, weighed and put in vacuum oven (80-100 C) overnight. After 24 hr, the dried brain was weighted. The brain water content in percent was calculated as (wet mass − dry weight) × 100/(wet weight). Baseline control mice received only buprenorphine before the brain was taken out and processed as described above.

| Statistics
Mean fluorescence was extracted from each ROI and signals were normalized in consecutive 4-min bins by the following formula: ΔF/ F 0 = (fluorescence − median fluorescence)/median fluorescence. We used spike frequency, spike amplitude, and area under curve (AUC) as measures of Ca 2+ activity. A local maximum was defined to be a spike if the following conditions were met: (i) the local maximum was the largest local maximum in between two subsequent local minima; (ii) the peak-to-peak amplitude of the local maximum and a subsequent local minimum was at least 0.5; (iii) the local maximum was above a threshold of ΔF/F = 0.5. Conditions (i) and (ii) ensured that multiple spikes were possible above the threshold of ΔF/F = 0.5, and condition (iii) ensured that a local maximum below ΔF/F = 0.5 was not defined as spike even though it fulfilled conditions (i) and (ii). The data were divided into two bins of 4 min: one describing baseline activity and the other describing the activity when the slices were bathed in hypoosmotic solution. Mean AUC and mean spike rate for all experiments were plotted as a function of time ( Figure 1d). The spike rate was produced by convolving the spike trains with a normalized Gaussian function of width 10 s. The mean AUC was smoothed by convolving with the same function. Based on these graphs and measured time to fill the recording chamber with hypoosmotic solution (less than 1 min 30 s), the experimental bin was set to 5.5-9.5 min. Generalized linear mixed effect models with nested random effects were used to account for this hierarchical structure in the statistical analysis: ROIs were grouped within brain slices, and the brain slices were grouped within mice. For spike frequency, we assumed an underlying Poisson distribution, AUC was assumed to be normally distributed and spike amplitude to be gamma distributed. The statistical analyses were performed in MATLAB (Version R2014a for Mac OSX), by using the fitlme() function for AUC and fitglme() function for spike frequency and amplitude. For all analysis, we used the nested random effects denoted (1| MouseID) + (1|MouseID:SliceID) + (1|MouseID:SliceID:ROIID) in MATLAB's Wilkinson notation. Here, MouseID is the unique mouse identifier, SliceID is the unique slice identifier, and ROIID is the unique ROI identifier. In Figure 1, the fixed effects were given by two predictors and their interaction term: the categorical sites (soma, processes, perivascular endfeet, and subpial endfeet) and the stimulus (zero or one corresponding to the baseline and experimental bin, respectively). In Figures 2 and 3, the fixed effects were given by three predictors for spikes and AUC: genotype (or channel blocking), sites, and stimulus; with all interaction effects. Spike amplitudes were collected from the experimental bin only, and the genotype and sites were used as predictors. In Figure 4, we used linear regression.
Matlab's fitlm() function was applied with genotype and state (control or hypoosmotic) as predictors. Unless stated otherwise, mean and SEM are shown in the text and figures. traces. Oscillatory Ca 2+ responses in subpial endfeet typically dis-played~10 peaks per min and lasted 1-3 min. Such oscillations were not observed at baseline conditions. Also in fine processes and perivascular endfeet did we observe responses with several or multiple peaks, but these peaks were less distinct and occurred at lower frequency (typically~5 peaks per min).
F I G U R E 1 Legend on next page.
To reveal the kinetics of the Ca 2+ responses, we plotted spike rate and AUC as a function of elapsed time (Figure 1d). Based on these plots and the chamber perfusion time, we defined the bin from 5.5 to 9.5 min (i.e., from 1.5 to 5.5 min after the switch to hypoosmotic aCSF) as the response bin to be compared with the baseline bin (each bin lasting 4 min). The plots revealed that the Ca 2+ responses to hypoosmotic stress lasted only a few minutes and were followed by Ca 2+ signaling patterns resembling those at baseline conditions. Comparison of spike rate in the two bins confirmed that the −20% hypo- When acute brain slices were exposed to aCSF with only 10 or 5% reduction in osmolarity Ca 2+ signaling was unaffected in all four astrocytic compartments ( Figure S1).
3.2 | Osmotically induced Ca 2+ signals in subpial and perivascular astrocytic endfeet rely on IP3R2-mediated Ca 2+ release from the endoplasmic reticulum Next, we assessed the origin of the osmotically evoked Ca 2+ signals in astrocytic processes and endfeet. Itpr2 −/− mice, which lack the IP3R2 Ca 2+ release channel in endoplasmic reticulum (Li et al., 2005), were subjected to the same experimental protocol as wildtype mice ( Figure 2a). In Itpr2 −/− mice, Ca 2+ signals within fine processes increased during hypoosmotic aCSF (p < .001 for comparison with baseline for spike rate) and the signaling response was comparable to that of wildtypes (Figure 2b, p = .46 and p = .54 for comparison for relative change in spike rate and change in AUC, respectively, p = .007 for Ca 2+ spike amplitude during hypoosmotic stress, which was modestly reduced in mutants). In contrast, perivascular and subpial endfeet of Itpr2 −/− mice showed severely suppressed Ca 2+ signaling during hypoosmotic stress (Figure 2a), indicating that Ca 2+ release from the endoplasmic reticulum is essential for the endfoot Ca 2+ response.
Note that subpial endfeet of Itpr2 −/− mice showed a much lower Ca 2+ spike rate than wildtypes both during hypoosmotic and baseline conditions (p < .001 for both conditions, not displayed) and that relative change in spike rate failed to differ between genotypes (p = .19, F I G U R E 3 Impact of dystrophin deficiency and TRPV4 blockage on endfoot Ca 2+ responses to hypoosmotic stress. (a) As in Figure 1c, but traces are from endfeet of mice lacking dystrophin (mdx 3Cv mice). (b) As in Figure 2b, but values are from mdx 3Cv (gray bars) and wildtype (white bars) mice. (c) As in Figure 2c, but values are from mdx 3Cv (gray) and wildtype (white) mice.
(d) As in Figure 2d, but values are from mdx 3Cv (gray) and wildtype (white) mice. Comparison is between mutant and wildtype mice with same statistical tests as in Figure 2. Number of traces, slices, and animals for each compartment were; perivascular endfeet: 31, 8, and 4; subpial endfeet: 43, 10, and 4 (same numbers for all parameters). Number of amplitudes in the hypoosmotic bin was 37 for perivascular endfeet and 82 for subpial endfeet. (e) As in Figure 1c, but traces are from wildtype slices with the TRPV4 antagonist HC067047 added to the bath solutions. (f) As in Figure 2b, but values are for HC067047 treated (gray bars) and nontreated (white bars) wildtype mice. (g) As in Figure 2c, but values are for HC067047 treated (gray) and nontreated (white) wildtype mice. (h) As in Figure 2d, but values are for HC067047 treated (gray) and nontreated (white) wildtype mice.
Comparison is for drug-treated and nontreated wildtypes with same statistical tests as in 3.4 | Impact of IP3R2-mediated Ca 2+ release on brain water accumulation during hypoosmotic stress To investigate whether IP3R2-mediated Ca 2+ signals regulate bloodbrain water transport in vivo Itpr2 −/− and wildtype mice were subjected to water intoxication, a well-established cytotoxic brain edema model (Manley et al., 2000). The mice were shortly anesthetized with isoflurane during intraperitoneal water injection (10% of body weight) but thereafter kept awake since anesthesia suppresses astrocytic Ca 2+ signaling (Thrane et al., 2012). In both genotypes, the brain water content measured by the wet/dry weight method was higher in water-injected than in noninjected mice (p < .001 for both wildtypes and Itpr2 −/− ), but Itpr2 deletion failed to affect the increase in water content (Figure 4; p = .15). Intraperitoneal injection of isotonic saline (10% of body weight) did not increase brain water content in any of the two genotypes (values for noninjected and saline-injected mice were 78.0% ± 0.08% and 78.0% ± 0.20%, p = .76 for comparison, for wildtype mice, and 77.9% ± 0.05% and 77.8% ± 0.08%, p = .67, for Itpr2 −/− mice).

| DISCUSSION
In this study, we characterized astrocytic Ca 2+ signaling in acute cortical slices exposed to aCSF with 20% reduction in osmolarity. We demonstrated that the osmotic challenge elicited Ca 2+ signals that differed in frequency, underpinning, and shape within the astrocytic territory. A striking finding was that astrocytic endfeet under the pia mater and around blood vessels showed the most pronounced Ca 2+ response to hypoosmotic stress, both with respect to increase in relative spike rate and AUC. One could argue that the difference in Ca 2+ response between astrocytic regions reflected unequal osmotic challenge between the superficial and deep brain tissue since the hypoosmotic aCSF was perfused into the slice chamber. However, time plots of Ca 2+ spike rate and AUC of the various astrocytic profiles did not reveal apparent differences in the timing of the responses, and perivascular endfeet were sampled also from deep within the tissue.
Thus, we find it more likely that the diversity of Ca 2+ responses within the astrocytic territory relies on molecular and functional specialization of astrocytic microdomains.
F I G U R E 4 Impact of Itpr2 deletion on brain water content in an in vivo brain edema model. Brain water content was measured in mice 40 min after intraperitoneal injection of water (10% of body weight) and in noninjected (control) mice. Comparison between waterinjected and noninjected mice is indicated (p < .001 for both wildtypes and Itpr2 −/− ). Itpr2 deletion failed to affect the increase in brain water content (p = .15) The distinct Ca 2+ responses of endfeet could rely on faster swelling due to their high density of AQP4 water channels. Notably, AQP4 is reported to speed up swelling and augment intracellular Ca 2+ signaling during hypoosmotic conditions (Benfenati et al., 2011;Mola et al., 2016;Thrane et al., 2011). A role of AQP4 was supported by the finding that mdx 3Cv mice, which have 65% less AQP4 in perivascular endfoot membranes (Enger et al., 2012), exhibited attenuated endfoot Ca 2+ responses. However, other mechanisms might contribute to this finding. Specifically, dystrophin could anchor other membrane proteins involved in volume regulation or transmit mechanical forces to endfoot osmosensors, yet uncharacterized.
The osmotically induced Ca 2+ signals in perivascular and subpial endfeet were severely suppressed in IP3R2-deficient mice indicating that the responses in these domains depend on Ca 2+ release from the endoplasmic reticulum. Despite this fact, Ca 2+ influx across the plasma membrane could also contribute. Notably, TRPV4 nonselective cation channels expressed in glial endfeet (Benfenati et al., 2007;Dunn, Hill-Eubanks, Liedtke, & Nelson, 2013) were found to mediate Ca 2+ signaling responses to hypotonic stimuli (Benfenati et al., 2011;Jo et al., 2015;Ryskamp et al., 2014). While it was proposed that TRPV4-AQP4 interactions "turbocharge" astroglial sensitivity to small osmotic gradients (Iuso & Krizaj, 2016), a recent study concluded that TRPV4 reacted to volume changes rather than osmotic changes (Toft-Bertelsen et al., 2017). Application of a TRPV4 antagonist to our acute brain slices surprisingly increased astrocytic Ca 2+ signaling in baseline, normosmotic conditions, which complicates the assessment of TRPV4's role in hypoosmotically induced Ca 2+ signals. However, TRPV4 channels were not essential for the endfoot Ca 2+ responses, since TRPV4 blockage only attenuated the increase in Ca 2+ spike rate and failed to affect the increase in AUC.
The osmotic Ca 2+ response within fine astrocytic processes was insensitive to Itpr2 deletion, in contrast to the situation in endfeet. This finding was not surprising since the most delicate astrocytic processes are too thin to accommodate endoplasmic reticulum (Patrushev, Gavrilov, Turlapov, & Semyanov, 2013). Thus, it is likely that the intracellular Ca 2+ increase in fine processes during hypoosmotic stress relies on Ca 2+ entry across the plasma membrane. In larger astrocytic processes Ca 2+ release from mitochondria, known to occur during metabolic activity (Agarwal et al., 2017), could also explain the increase in Ca 2+ signaling since volume regulation consumes energy (Olson, Sankar, Holtzman, James, & Fleischhacker, 1986).
Hypoosmotically induced Ca 2+ signals within the astrocytic territory also differed in shape. Whereas the Ca 2+ transients in fine processes typically had a single peak, endfoot Ca 2+ responses also included plateau-type Ca 2+ elevations and rapid Ca 2+ oscillations. The latter were confined to subpial endfeet and there occurred in one third of the profiles. The rapid Ca 2+ oscillations in subpial endfeet were critically dependent on IP3R2 receptors but also showed some dependency on TRPV4 and dystrophin. Thus, the unique oscillatory Ca 2+ signaling pattern in subpial endfeet-which to our knowledge has never been described before-is likely to be triggered by rapid water entry through AQP4 channels, swelling-induced Ca 2+ influx from the extracellular space via TRPV4 channels and Ca 2+ induced Ca 2+ release from the endoplasmic reticulum mediated by IP3R2 receptors.
Surprisingly, the astrocytic cell bodies did not respond to the hypoosmotic challenge in the present experiments. In a previous study using bulk loading of the synthetic Ca 2+ dye Rhod2 AM, we reported that astrocytic cell bodies responded to osmotic swelling with Ca 2+ spikes both in vitro and in vivo (Thrane et al., 2011). However, in that study acute cortical slices were obtained from mouse pups and the experiments performed at room temperature. It is possible that the somatic Ca 2+ response in the immature animals relies on autocrine signaling pathways that are developmentally regulated. Furthermore, the in vivo experiments on adult mice revealed that the osmotically induced Ca 2+ responses in astrocytic cell bodies occurred in the late phase of brain swelling, that is, 30-45 min after intraperitoneal injection of water (Thrane et al., 2011). Furthermore, Ca 2+ spikes occurring during advanced edema could rely on ischemia, which is known to elicit astrocytic Ca 2+ signals (Duffy & Macvicar, 1996).
Brain water entry and exit must occur across subpial and perivascular endfeet which form the outermost layers of neural tissue and are positioned next to the fluid compartments draining the brain. Do endfoot Ca 2+ signals regulate brain water transport across the bloodbrain and brain-CSF interfaces? During acute hypoosmotic conditions the blood-brain water entry is facilitated by endfoot AQP4 water channels (Amiry- Moghaddam et al., 2004;Haj-Yasein et al., 2011;Manley et al., 2000), which most likely are not regulated by channel gating (Assentoft, Larsen, & MacAulay, 2015). However, astrocytic Ca 2+ signals have been implicated in export of osmolytes and water during cell volume regulation (Hoffmann et al., 2009). Given that hypoosmotically induced Ca 2+ signals in endfeet relied on IP3R2 receptors we tested whether Itpr2 −/− and wildtype mice differed in brain water accumulation during water intoxication, an in vivo model of brain edema in which mice are injected with distilled water intraperitoneally. To minimize animal discomfort, we used an osmotic challenge that was lower (~10% reduction in plasma osmolarity) than in our in vitro model (−20% reduction in perfused aCSF osmolarity). We found that IP3R2-deficient mice showed similar increase in brain water content as wildtype animals. Additional in vitro experiments revealed that aCSF with only 10% reduction in osmolarity failed to elicit astrocytic Ca 2+ signals. Thus, it seems that osmotic changes must be of a certain magnitude or occur sufficiently fast to elicit astrocytic Ca 2+ responses. It is likely that our in vivo brain edema model was associated with osmolarity changes that did not meet these criteria.
The osmolarity reduction within the slice that triggered astrocytic Ca 2+ signals must have been substantially less than 20%, since astrocytes responded already when an equal amount of −20% hypoosmotic aCSF and normosmotic aCSF had been mixed in the slice chamber (see Section 3). The precise kinetics of osmolarity changes within the slice under our experimental conditions is hard to estimate, also due to the onset of cell volume regulatory mechanisms with release of osmolytes and the fact that the extracellular space comprises of tunnels and sheets of different size (Kinney et al., 2013).
Whether the osmotic changes in our in vitro model mimic those in vivo remains an open question.
Accumulating evidence suggests that Ca 2+ signals play a role in glial cell volume regulation (Benfenati et al., 2011;Jo et al., 2015).
Inadvertent swelling of astrocytic endfeet must be detrimental for glymphatic function, as it would shrink both gaps between endfeet and the paravascular compartment, both of which are postulated drainage routes (Iliff et al., 2012). At the same time, dynamic volume changes of endfeet could constitute a mechanism for regulating the efficacy of interstitial waste clearance, for example, during the sleepawake cycle. Proper volume control of astrocytic endfeet is also necessary for normal cerebral perfusion. Endfoot swelling may compress capillaries or have indirect effects on vascular tone by releasing vascoactive molecules.
Astrocytic Ca 2+ signals increase not only during osmotic swelling but also following electrical stimulation and application of neurotransmitters (Haustein et al., 2014;Tang et al., 2015). While a plethora of transporters and channels as well as G-protein coupled receptors are considered important for activity-dependent Ca 2+ signals (Shigetomi, Patel, & Khakh, 2016), few investigators have addressed the possibility that swelling could trigger the response. The large surface area to volume ratio of the delicate astrocytic processes in neuropil (Hama, Arii, Katayama, Marton, & Ellisman, 2004) should make them particularly prone to swelling during glutamate uptake, which imposes an osmotic load and cellular water entry. It is tempting to speculate that also spontaneous microdomain Ca 2+ signals, which are largely IP3R2-independent (Srinivasan et al., 2015), reflect signaling during astrocyte volume adjustment.
In conclusion, our study has demonstrated that astrocytic endfeet display osmotically induced Ca 2+ signals that differ quantitatively, qualitatively, and mechanistically from those in other astrocytic microdomains. The endfoot Ca 2+ signals-including the unique rapid Ca 2+ oscillations in subpial endfeet-should be further explored as regulators of water transport and cell volume in health and disease.

ACKNOWLEDGMENTS
We gratefully acknowledge the support by UNINETT Sigma2 AS, for making data storage available through NIRD, project NS9021K. This work was supported by the Research Council of Norway (grants #249988, #240476, and #262552), the South-Eastern Norway Regional

CONFLICTS OF INTEREST
The authors declare no conflicts of interest.