Hypoosmolar dose-dependent swelling occurs in both pyramidal neurons and astrocytes in acute hippocampal slices

Normal nervous system function is critically dependent on the balance of water and ions in the extracellular space. Pathological reduction in brain interstitial osmolarity results in osmotically-driven flux of water into cells, causing cellular edema which reduces the extracellular space and increases neuronal excitability and risk of seizures. Astrocytes are widely considered to be particularly susceptible to cellular edema due to selective expression of the water channel aquaporin-4 (AQP4). The apparent resistance of pyramidal neurons to osmotic swelling has been attributed to lack of functional water channels. In this study we report rapid volume changes in CA1 pyramidal cells in hypoosmolar ACSF (hACSF) that are equivalent to volume changes in astrocytes across a variety of conditions. Astrocyte and neuronal swelling was significant within 1 minute of exposure to 17 or 40% hACSF, was rapidly reversible upon return to normosmolar ACSF, and repeatable upon re-exposure to hACSF. Neuronal swelling was not an artifact of patch clamp, occurred deep in tissue, was similar at physiological vs. room temperature, and occurred in both juvenile and adult hippocampal slices. Neuronal swelling was neither inhibited by TTX, nor by antagonists of NMDA or AMPA receptors, suggesting that it was not occurring as a result of excitotoxicity. Surprisingly, genetic deletion of AQP4 did not inhibit, but rather augmented, astrocyte swelling in severe hypoosmolar conditions. Taken together, our results indicate that neurons are not osmoresistant as previously reported, and that osmotic swelling is driven by an AQP4-independent mechanism.


Introduction 52
Acute reduction of plasma osmolarity in humans is a medical emergency, often 53 resulting in seizures and sometimes coma or even death (Andrew 1991;Castilla-Guerra 54 et al. 2006). Such deleterious effects on the CNS can be attributed to the sudden 55 change in osmotic pressure within the interstitial space of the brain, which causes cells 56 to take on water leading to "cellular" or "cytotoxic" edema (Kimelberg 1995). Tissue 57 swelling resulting from cellular edema has been observed in many studies examining 58 the effect of hypoosmolar conditions on excitability (Andrew and MacVicar 1994;59 Chebabo et al. 1995b; Kilb et al. 2006). Given that cell swelling shrinks the extracellular 60 space (ECS) and increases tissue resistance, hypoosmolar conditions amplify 61 nonsynaptic excitability in neurons and increase susceptibility to seizure (Lauderdale et 62 al. 2015;Roper et al. 1992; Rosen and Andrew 1990;Schwartzkroin et al. 1998). 63 Most cellular edema in the brain is thought to be driven by water influx through 64 aquaporin-4 (AQP4) channels, which are expressed primarily by astrocytes (Nagelhus 65 et al. 2004). Knockout of AQP4 inhibits the tissue swelling normally associated with 66 seizures (Binder et al. 2004), stroke (Katada et al. 2014), and other models of cytotoxic 67 edema (Papadopoulos and Verkman 2013). In normal physiology, AQP4 is important for 68 water homeostasis in the brain, and appears to be required for the activity-dependent 69 fluxes of water which occur alongside potassium uptake and buffering (Amiry- Astrocytic swelling via AQP4-mediated water uptake is well-established as both a 72 physiological and pathological phenomenon.
Alexa Fluor 594 and SR-101 were excited using a 559 nm semiconductor laser and 234 detected using a 624-724 nm bandpass filter. Pixel dwell times were kept the same as 235 above for consistency. Laser power ≤ 1.5% was sufficient to detect SR-101 labeled 236 astrocytes and Alexa Fluor 594 labeled neurons. To strike an appropriate balance 237 between image resolution and brightness, confocal aperture size was set to 300 µm and 238 PMT voltage ~830 V across all experiments. 239 In one experiment, Thy1-eGFP neurons were examined much deeper within the 240 slice (>60 µm below slice surface) and were in many cases impossible to image using 241 our standard settings. Instead, laser power was increased to 10% and pixel dwell time 242 to 8 µs/pixel, increasing acquisition time per image stack (~15-30 seconds) but 243 significantly boosting cell visibility. We observed no deleterious effects on cell health 244 resulting from the increase in laser power or exposure time. 245 All experiments started in standard ACSF. Where applicable, standard ACSF was 246 replaced by nACSF following patch pipette removal or identification of the cell to be 247 imaged, and allowed to wash in for 10 minutes prior to imaging. Imaging consisted of 248 confocal z-stacks taken through the cell soma, beginning with a single "baseline" stack. 249 As observed by other groups (Hirrlinger et al. 2008;Risher et al. 2009), we found that 250 single images were insufficient for gathering the full extent of the cell body in the x-y 251 plane (which is necessary for soma area measurements, our proxy for cell volume), and 252 were particularly vulnerable to swelling-induced z-shifts across time points. We instead 253 opted for rapid z-stacks through the soma, at 1.0 µm intervals and a zoom level of 3.5x 254 (0.118 µm/pixel). X, Y and Z-shifts were compensated using quick confocal scans to 255 check cell position, and adjusting X-Y position and Z-scan upper/lower limits accordingly before acquiring each stack. To increase scan speed, the "clip scan" function was used 257 to crop the scan window close around the soma, reducing acquisition time to ~15 258 seconds per stack. These settings allowed us to obtain full z-stacks through the cell 259 soma at 1-minute intervals. For simplicity, a z-stack encompassing the cell soma will 260 hereafter be referred to as a "stack". 261 After a baseline stack was obtained, hACSF was applied for 5 minutes and a 262 single stack was acquired at the end of each minute. HACSF was then "washed out" by 263 re-application of normosmolar ACSF (either nACSF or standard ACSF) for 5 minutes, at 264 the end of which an additional stack was acquired. This sequence was repeated an 265 additional 2 times to determine repeatability of effects on cell volume. Each subsequent 266 hACSF application and wash period was lengthened by 1 minute (after which an 267 additional stack was acquired), due to our observations (unpublished) that neuronal 268 slow inward currents (SICs) evoked in these conditions were more "spread out" during 269 subsequent hACSF applications. 270

Volume analysis protocol 271
Analysis of neuron and astrocyte volume was performed using the FIJI distribution 272 of ImageJ as previously described (Lauderdale et al. 2015;Schindelin et al. 2012). Our 273 basic analysis protocol is depicted in Figure 1. First, z-shifts across time points were 274 corrected by choosing a common landmark and matching its slice number across all 275 stacks. Stacks were then concatenated into an x-y-z-t hyperstack ( Figure 1B), and 276 filtered to remove noise (median filter, 2 pixel diameter). A max-intensity z-projection 277 (MIP) was made through this hyperstack, producing a 2D time series ( Figure 1C), in 278 which each frame was a MIP showing the full x-y extent of the soma at a given time point. X-Y shifts over time were corrected using the "Linear Stack Alignment with SIFT" 280 plugin (followed by cropping, to remove the blank areas left by image alignment), and 281 background subtraction was performed using FIJI's background subtraction tool (radius 282 50 pixels, sliding paraboloid method). The resultant time series ( Figure 1D) was then 283 binarized using the "mean" thresholding algorithm ( Figure 1E) and an elliptical ROI was 284 drawn to narrowly encompass the soma across all time points in the series (dashed red 285 circle, Figure 1E). Area above threshold within this ROI (a measure of soma area) was 286 used as a proxy for soma volume. Volume changes at a given time point are reported 287 as percent change from baseline soma volume. Some experiments also include the 288 "average percent change" from baseline over 3 hACSF applications. In these cases, 289 percent change in the 2 nd and 3 rd hACSF periods is calculated based on the preceding 290 nACSF wash instead of the original baseline, providing a more accurate measure of 291 acute, relative volume changes during hACSF application. Representative images, 292 unless otherwise noted, depict the thresholded images used for analysis. 293 In an effort to better analyze neurons from "deep" tissue (which were not amenable 294 to our standard analysis protocol due to poor resolution of cell borders), we employed a 295 simplified version of the microspectrofluorimetric method of measuring volume (Crowe 296 et al. 1995). This method does not rely on cell borders but rather on fluorescence 297 intensity, allowing for estimation of cell volume changes using a region of interest (ROI) 298 placed in the cell center. In brief, this method assumes a fixed number of fluorescent 299 molecules within a cell, and therefore that any changes in cell water content (i.e. cell 300 swelling or shrinking) will alter their concentration. As dye becomes less concentrated, comparisons between time points were adjusted performed using the Holm-Bonferroni 320 method, a stepwise procedure with the same assumptions as the Bonferroni correction 321 but substantially more power for larger numbers of comparisons. Significant between-322 subject effects were further investigated by Student's t-test or (if 3 or more groups) 323 using pairwise comparisons with Bonferroni correction. Significant interactions were investigated in two steps. First, between-group simple effects were tested at each time 325 point using Student's t-test (for 2 groups) or one-way ANOVA followed by Tukey's HSD 326 post-hoc tests (for 3 or more groups). In the case of normality or homoscedasticity 327 violations, appropriate alternatives were chosen (e.g. Welch t-test or Mann-Whitney U). 328 Second, within-subjects simple effects were determined by splitting the file by group and 329 running each as an individual one-way repeated measures ANOVA, with Holm-330 Bonferroni post-hoc tests as above. This "split-file" method has the effect of splitting the 331 error terms by group and was deemed to be more accurate than obtaining simple main 332 effects within the original mixed ANOVA, as the latter uses a pooled error in its 333 calculations. N = 7-10 cells per group (after outlier removal) for all experiments, unless 334 indicated otherwise. Significant differences are reported at the p < 0.05 (*), p < 0.01 (**), 335 and p < 0.001 (***) levels. 336 337 3 Results 338 In a previous study, we found heightened neuronal excitability within one minute of 339 exposing hippocampal slices to 17% or 40% hypoosmolar ACSF (Lauderdale et al.

Neuronal health is unlikely to be a contributing factor to hACSF-induced 392
swelling We began to examine the key differences in experimental protocol between our 394 study and previous work, to determine if an unforeseen factor was responsible for (or 395 contributing to) the neuronal swelling observed in our conditions. 396 Although we had no reason to suspect damage to the cell using whole-cell patch clamp 397 approaches (see Methods), we considered the possibility that the simple act of patch 398 clamping neurons may make them more susceptible to hACSF-induced swelling. We 399 tested this by using hippocampal slices from transgenic Thy1-eGFP mice (Feng et al. In further support of this view, we observed that both shallow ( Figure 4D are inversely proportional to changes in cell volume (Crowe et al. 1995). We reasoned 435 that this method should be much less sensitive to detection problems than our 436 thresholding technique (magenta outlines, Figure 4D however, provide supportive evidence that relative volume changes in neurons exposed 451 to 40% hACSF are independent of depth within the slice. 452 We also compared patch clamp-loaded astrocytes against those loaded with are region-specific, we also measured volume changes in pyramidal neurons from 498 neocortical layer V and passive astrocytes from layer II/III (Figure 6). We found eGFP 499 expression to be particularly low in the cortex of Thy1-eGFP mice, which limited their 500 utility for these experiments. Data were therefore pooled from Thy1-eGFP neurons (n = 501 2) and patch-clamped neurons (n = 8), and compared to pooled hippocampal neurons 502 acquired using both techniques (n = 9 Thy1-eGFP, n = 10 patch clamp). As expected 503 from our observations of hippocampal astrocytes and neurons, cortical layer II/III 504 astrocytes ( Figure 6A) and layer V pyramidal neurons ( Figure 6B) both exhibited rapid 505 volume changes upon exposure to antagonist-free 40% hACSF. Surprisingly, we 506 observed significantly greater swelling in cortical neurons compared to cortical 507 astrocytes ( Figure 6C). Average percent change over 3 hACSF applications ( Figure 6D) 508 in cortical neurons was nearly double that of hippocampal neurons (5 minutes hACSF: 509 CA1 neurons, 8 ± 0.81% increase; Cortical neurons, 14 ± 0.98% increase; p < 0.001). 510 The average percent change did not differ between cortical vs. hippocampal astrocytes 511 (F(1,19) = 0.43, p = 0.52; Figure 6E). It was not immediately apparent whether the 512 increase in cortical neuron osmosensitivity reflected an intrinsic difference in membrane 513 properties, or simply a greater freedom for cell swelling due to the looser packing 514 density of neurons in the cortex. These data suggest that osmosensitivity of pyramidal 515 neurons varies by brain region, and importantly, that significant swelling of pyramidal 516 neurons is not restricted to the hippocampal region. 517

Neuronal swelling is not due to NMDA receptor activation 518
In examining alternate explanations for neuronal swelling in hACSF, we next 519 chose to more closely investigate the possible involvement of NMDA receptor 520 excitotoxicity. Neuronal swelling is a well-established consequence of excitotoxic NMDA 521 receptor activation (Choi 1992 (F(1,19) = 0.006, p = 0.94) between the two age groups. These findings 553 suggest that neuronal swelling is not a developmental phenomenon.

Hypoosmolar swelling of neurons and astrocytes does not require AQP4 555
In astrocytes, water permeability is thought to be regulated by the expression and 556 gating of aquaporin-4 (AQP4) channels (Gunnarson et al. 2008;Nagelhus et al. 2004;557 Song and Gunnarson 2012). Functional water channels have not been found in 558 pyramidal neurons, a proposed reason behind their apparent resistance to osmotic 559 stress (Andrew et al. 2007;Gorelick et al. 2006). Our observation of neuronal swelling in 560 all hACSF conditions, however, was clear evidence that the lack of functional 561 aquaporins was not a limiting factor for cell volume change in neurons, and that water 562 must be entering through an alternate route. Several of the alternate, AQP4-563 independent mechanisms proposed for swelling in astrocytes, including water-564 permeable cotransporters and even water flux directly across the cell membrane 565 (Kimelberg 2005), could also be functional in neurons and contribute to the observed 566 neuronal swelling in our conditions. If these AQP4-independent mechanisms mediate 567 the rapid neuronal swelling in our hACSF, it follows that the rapid astrocyte swelling in 568 our hACSF may also be partially (or fully) independent of AQP4. 569 We therefore examined the acute volume responses of both CA1 pyramidal 570 neurons and s.r. astrocytes to antagonist-free hACSF, using hippocampal slices from 571 AQP4 -/mice ( Figure 9). As foreshadowed by our neuronal data gathered to this point, 572 astrocytes continued to swell even in the absence of AQP4, implying rapid astrocyte 573 volume changes occur independently of AQP4 expression ( Figure 9A). Surprisingly 574 however, AQP4 -/astrocytes swelled significantly more in 40% hACSF compared to 575 wild-type, reaching approximately 1.4 times the volume of wild-type astrocytes by the 576 end of each hACSF application ( Figure 9B; AQP4 -/-, 113.19% ± 0.92%; WT, 109.71% ± 577 0.54%; p = 0.005). By contrast, astrocyte swelling did not differ between AQP4 -/and 578 wild-type astrocytes in 17% hACSF (F(1,17) = 1.000, p = 0.331). We also tested the 579 effects of AQP4 -/on neuronal volume changes in 40% hACSF. Not surprisingly, AQP4 580 deletion had no effect on neuronal swelling (F(1,16) = 0.02, p = 0.891; Figure 9C, D). 581 These data seemed to indicate that AQP4 expression, far from being required for 582 intracellular accumulation of water in astrocytes, may actually play an important role in 583 limiting the degree of swelling during severe hypoosmolar stress. Overall, our results 584 strongly suggest that water influx resulting from hypoosmolar ACSF application is not 585 selective to astrocytes, and is not mediated by AQP4. 586 587 4 Discussion 588 The major finding in this study is that CA1 pyramidal neurons readily swell in 589 hypoosmolar conditions, increasing their volume significantly above baseline even 590 within the first minute of hACSF application. The rate and overall amount of neuronal 591 swelling was found to be almost identical to that observed in stratum radiatum 592 astrocytes, and was observed repeatedly across a variety of experimental conditions. 593 Neuronal swelling was rapidly reversible, returning to baseline after 5-7 minutes in 594 normosmolar ACSF, and readily returned upon re-application of hACSF. While the 595 astrocyte swelling observed is in line with the reportedly high osmosensitivity of 596 astrocytes and astrocyte-like glial cells (Hirrlinger et al. 2008;Risher et al. 2009), the 597 neuronal swelling observed contrasts with previous work concluding that neurons 598 steadfastly retain their volume during osmotic challenge (Andrew et al. 2007;Caspi et 599 al. 2009). Neither glial nor neuronal swelling was found to require AQP4 expression. These data cast doubt over the commonly-accepted roles of neurons and astrocytes in 601 brain water regulation. 602 The possibility that neuronal volume changes are extremely sensitive to subtle 603 differences in methodology was strongly considered and systematically examined in this 604 study. Initial experiments were performed in tissue slices from juvenile mice at room 605 temperature, using whole-cell patch clamp to dialyze individual CA1 hippocampal cells 606 intentionally conservative and likely underestimates actual changes in cell volume. Cell 636 soma area was chosen as a proxy for overall volume changes, since direct measures of 637 cell volume were not feasible for our conditions. Assuming that cell volume is changing 638 equally in all directions, it can be inferred that our measurements reflect only about two-639 thirds of the true change in volume. This inference is partially supported by the analysis 640 of fluorescence intensity as an indication of cell volume changes, which suggested 641 volume increases > 1.5 times those calculated from thresholded images. We also 642 observed that the precise method of image processing had a surprisingly strong impact 643 on measured volume. The overall magnitude of volume change observed could often be 644 increased twofold or more simply through the use of different thresholding methods. 645 These apparent increases in signal-to-noise ratio often came at the expense of "cell" and "background" pixels being incorrectly classified, and the thresholded images 647 inconsistent with what could be observed "by eye" pre-thresholding (unpublished 648 observations). We therefore rejected these methods in favor of the more conservative 649 "mean" thresholding method, which more consistently returned an image in which the 650 cell body could be clearly identified. 651 One of the more exciting findings in the present study was a lack of evidence for 652 AQP4-mediated water influx into astrocytes or neurons exposed to hACSF. While not 653 necessarily surprising for neurons, which do not express AQP4 (Nielsen et al. 1997), 654 lack of inhibition of astrocyte swelling in AQP4 -/mouse tissue was rather unexpected 655 based on a large body of literature on the requirement of AQP4 for facilitated water 656 movement across the astrocyte membrane. Surprisingly, not only was AQP4 not 657 required for astrocyte swelling, but astrocyte volume in severely (40%) hypoosmolar 658 conditions was significantly greater in AQP4 -/mice compared to wild-type animals. 659 There are a couple of potential explanations for this finding. 660 First, AQP4 -/animals have been shown to have a nearly 30% larger resting brain 661 extracellular space compared to wild-type animals (Binder et al. 2004;Yao et al. 2008). 662 It is possible that the enlarged ECS in AQP4 -/mice provides more available space for 663 astrocytes to swell into, leading to increased astrocyte swelling. This explanation, 664 however, fails to account for the lack of increased neuronal swelling in AQP4 -/mice, 665 since neurons would presumably also be adjacent to increased extracellular space. It is 666 also unlikely that astrocyte volume changes in our 40% hACSF were limited by the 667 extracellular space in wild-type animals, as tissue continues to swell in far more severe 668 hypoosmolar conditions (Chebabo et al. 1995b).