The physiological conditions that swell mammalian neurons are clinically important but contentious. Distinguishing the neuronal component of brain swelling requires viewing intact neuronal cell bodies, dendrites, and axons and measuring their changing volume in real time. Cultured or dissociated neuronal somata swell within minutes under acutely overhydrated conditions and shrink when strongly dehydrated. But paradoxically, most central nervous system (CNS) neurons do not express aquaporins, the membrane channels that conduct osmotically driven water. Using 2-photon laser scanning microscopy (2PLSM), we monitored neuronal volume under osmotic stress in real time. Specifically, the volume of pyramidal neurons in cerebral cortex and axon terminals comprising cerebellar mossy fibers was measured deep within live brain slices. The expected swelling or shrinking of the gray matter was confirmed by recording altered light transmittance and by indirectly measuring extracellular resistance over a wide osmotic range of −80 to +80 milliOsmoles (mOsm). Neurons expressing green fluorescent protein were then imaged with 2PLSM between −40 and +80 mOsm over 20 min. Surprisingly, pyramidal somata, dendrites, and spines steadfastly maintained their volume, as did the cerebellar axon terminals. This precluded a need for the neurons to acutely regulate volume, preserved their intrinsic electrophysiological stability, and confirmed that these CNS nerve cells lack functional aquaporins. Thus, whereas water easily permeates the aquaporin-rich endothelia and glia driving osmotic brain swelling, neurons tenatiously maintain their volume. However, these same neurons then swell dramatically upon oxygen/glucose deprivation or [K+]0 elevation, so prolonged depolarization (as during stroke or seizure) apparently swells neurons by opening nonaquaporin channels to water.
During acute overhydration, the brain quickly absorbs water and may swell enough to herniate through the base of the cranium or around the rigid meninges (Grande and others 1997; Law 1998; Somjen 2004). With acute dehydration, the brain gives up water and shrinks losing buoyancy in the surrounding cerebrospinal fluid that can tear blood vessels (Adrogue and Madias 2000). Sudden and deadly shifts in plasma osmolality between −20 and +30 milliOsmoles (mOsm) are usually man-made (Andrew 1991; Paczynski and others 1997; Arieff and Kronlund 1999; Finberg 2000; Rapoport 2000) as during osmotic opening of the blood-brain barrier for drug delivery (Rapoport 2000), osmotherapy for diabetics (Finberg 2000), accidental water intoxication, or overaggressive clinical rehydration (Andrew 1991). Brain cells compensate for gradual change in plasma osmolality over hours and days (Chan and Fishman 1979; Pollock and Arieff 1980; Gullans and Verbalis 1993; Finberg 2000), and the brain can even adjust its volume over 90 min to hypernatremia kept within physiological limits (Cserr and others 1991).
Cultured or dissociated nerve cell bodies swell, change their intrinsic electrophysiological properties, and may actively regulate their volume upon abrupt (but unphysiologically low) osmotic shifts (Horie and others 1989; Pasantes-Morales and others 1993; Leaney and others 1997; Aitken and others 1998; Patel and others 1998; Inoue and others 2005) as do pyramidal cell bodies imaged near the cut surface of hippocampal slices (Zhang and Bourque 2003). Yet, dissociated hippocampal neurons can resist volume change under acute osmotic stress (Somjen and others 1993; Aitken and others 1998). A slowly ramping osmotic shift does not alter neuronal soma volume, interpreted as an “isovolumetric” regulation (Tuz and others 2001), but could also indicate resistance to volume change. The relevance of large osmotic shifts to intact brain has been questioned (Andrew and others 1997; Law 1998). Encased in a rigid support structure, the mammalian brain cannot survive the abrupt 60- to 160-mOsm increases or decreases required to observe changing neuronal volume in vitro. Where then is the evolutionary pressure for central nervous system (CNS) neurons to develop volume regulatory mechanisms described in isolated neurons?
Most intact CNS neurons lack functional identified water channels (aquaporins) that mediate transmembrane osmotic water movement (Agre and others 2002). Moreover, the intrinsic properties of pyramidal neurons in osmotically stressed brain slices remain surprisingly stable (Rosen and Andrew 1990, 1991; Ballyk and others 1991; Saly and Andrew 1993) even as the slice abruptly swells or shrinks without displaying volume regulation (Andrew and MacVicar 1994; Krizaj and others 1996; Andrew and others 1997). So, as the principle output neurons of the mammalian cerebral cortex, do pyramidal neurons actually change their volume under osmotic stress?
Brain slices contain fully arborized neurons that can be imaged deep to the cut surface. Here, we first confirm changing tissue volume under osmotic stress by imaging altered light transmittance (LT) and indirectly measuring extracellular resistance. Then using 2-photon laser scanning microscopy (2PLSM), we monitor real-time volume of pyramidal neurons and cerebellar axons during acute osmotic stress. We propose that these neurons resist osmotic volume change as a benefit of their low water permeability, thereby stabilizing neuronal structure and function even as the surrounding glial network swells or shrinks. Yet, these same neurons clearly swell under conditions evoking maintained depolarization.
Materials and Methods
Brain Slice Preparation
Male C57 black mice (30–80 days old) or Sprague-Dawley rats (22–32 days old) were decapitated using a guillotine. All procedures follow National Institutes of Health guidelines for the humane care and use of laboratory animals and undergo yearly review by the Animal Care and Use Committees at the Queen's University and the Medical College of Georgia. The brain was quickly removed and placed in cold, oxygenated (95% O2/5% CO2) artificial cerebrospinal fluid (aCSF) of the composition described below but with equiosmolar sucrose replacing NaCl (Kirov and others 2004). Using a Leica 1000-T vibratome, coronal slices were cut at 400 μm. Each slice was then dissected along the midline and equilibrated in aCSF at room temperature for at least 1 h prior to data acquisition. The aCSF was prepared by dissolving in double-distilled water: 120 mM NaCl, 3.3 mM KCl, 26 mM NaHCO3, 1.3 mM MgSO4·7H2O, 1.23 mM NaH2PO4, 11 mM D-glucose, and 1.8 mM CaCl2. The aCSF osmolality was 287–289 mOsm and pH was 7.3–7.4. In oxygen/glucose deprivation (OGD) studies, N2 replaced O2, and the glucose concentration was reduced from 11 to 1 mM. NaCl was added to the aCSF to osmotically balance the removed glucose.
Imaging Changes in LT
Individual slices were transferred from a submersion type storage chamber (30 °C) to a recording chamber with aCSF flowing at 3 mL/min where the temperature was slowly raised to 33–34 °C. Each slice was held down at its edges using small pieces of silver wire and viewed with the 1.6× objective lens of an inverted light microscope. A halogen lamp with a voltage-regulated direct current power supply transilluminated the slice. Video images were obtained using a charged coupled device (CCD) camera set at maximum gain and medium black level. With the CCD gamma set at 1.0, output was linear with respect to change in light intensity. Images were obtained using a frame grabber (DT 2867, Data Translation, Marlboro, MA) in a pentium computer controlled with axon imaging workbench 2.2 software (Axon Instruments). An experiment required a series of images, each image consisting of 128 averaged frames acquired at 30 Hz. The first image of the series was the control transmittance (Tcont) that was subtracted from each of the subsequent images (Texp) in the series. Each subtracted image demonstrated any change in LT over time was divided by the gain set by adjusting the software. The difference signal was normalized by dividing by Tcont, which varies across the slice depending on the zone sampled. For example, Tcont was lower in white matter than gray matter. Note that the plotted data are normalized, but the images are not. This value was then presented as a percentage of the digital intensity of the control image of that series. That is,
To measure the evoked field potential or the spontaneous negative shift that signifies anoxic depolarization, a micropipette (5–10 MΩ) was pulled from thin-walled capillary glass, filled with 2 M NaCl, and mounted on a 3-dimensional micromanipulator. It was connected by an AgCl-coated silver wire to an amplifier probe and the output monitored using an online oscilloscope. The amplified signals were digitized (Neurodata Instruments) and stored on videocassette tape. The extracellular micropipette was placed in CA1 pyramidale, and a concentric bipolar electrode (Rhodes Electronics, Houma, LA) was placed in CA3 stratum radiatum. In neocortical slices, the recording electrode was placed in layers II/III and layer VI stimulated immediately below. A current pulse (0.1–1.5 mA; 0.1-ms duration; 0.25 Hz) was applied to produce a population spike of near-maximal amplitude. Digitized data were signal averaged (6–12 sweeps/trace), displayed, and plotted using pCLAMP software (Axon Instruments).
The 2PLSM Imaging
Neocortical and hippocampal slices (400 μm) were taken from 30- to 88-day-old C57 black mice of the B6.Cg-TgN(thy1-GFP)MJrc strain. These mice have a small proportion of pyramidal neurons (neocortex/hippocampus) and mossy fiber (MF) axon terminals (cerebellum) that express green fluorescent protein (GFP) (Feng and others 2000). Brain slices from these mice respond like Sprague-Dawley rats to osmotic and metabolic stress (Joshi and Andrew 2001). About half of the pyramidal neurons were sampled from neocortical layer V and the other half from the hippocampal CA1 region. An imaging chamber was mounted on a fixed stage of an upright Axioscope II FS microscope (Carl Zeiss, Jena, Germany). All osmotic experiments started and ended in control (287–289 mOsm) aCSF. Osmotic stress entailed a 15- to 20-min exposure to −40 or +40 mOsm aCSF followed by return to normosmotic aCSF. Alternately, −40 mOsm aCSF was followed by +40 (or +80) mOsm and then returned to normosmotic aCSF. Imaging was at an excitation wavelength of 870 nm using the Zeiss LSM 510 NLO meta multiphoton system directly coupled with the Spectra-Physics (Mountain View, CA) Mai-Tai Ti:Sapphire laser. Three-dimensional time-lapse images were taken at 0.5- to 1.0-μm increments with a 63×/0.9 numerical aperature water-immersion objective (Carl Zeiss) using 4× optical zoom, resulting in a nominal spatial resolution of 28 pixels/μm (12 bits per pixel, 2.24 μs pixel time). Emitted light was detected by an LSM 510 NLO scan module with the pinhole entirely opened. Data acquisition was controlled by Zeiss LSM 510 software. Utilizing the full width at half maximum of a point spread function measured with subresolution beads, the microscope resolution was calculated as 0.38 μm in the lateral dimension and 1.95 μm in the axial dimension.
Volume Analysis with 2PLSM
Early in the study, it became clear that simply measuring cell body and dendrite diameter did not reveal osmotic effects, so we devised 4 more sensitive techniques in case we were missing volume changes. 1) The gray value distribution of pixels comprising an image stack was compared across the treatment conditions. For each treatment, a stack was recentered relative to a common landmark and then similarly cropped. Using Matlab software, pixels were sorted by their 12-bit value (0–4095) into 8 bins. The gray scale was applied to the images where 0 = black and 4095 = white. An increase in the proportion of white pixels (representing GFP staining) signified an expansion of the imaged structure into adjacent unstained space, assuming overall brightness remained constant. In this respect, OGD swelling was often accompanied by a general fading of GFP, so this technique was only applied to nonfading OGD fields (the fading involves change in the optical properties of the swelling tissue in complex ways that are probably physiologically significant but poorly understood). Pixel distributions obtained for images in control aCSF were subtracted from those obtained in experimental conditions and then divided by the total pixel number in the cropped image, yielding a percent change from control for each grayscale value bin. Finally, these values were subjected to a repeated-measures analysis of variance (ANOVA). 2) Control profiles of somata in stacked images were digitally traced by hand and filled to create a mask from behind by which swelling could be discerned. 3) A control and experimental image stacks were pseudocolored green and red, respectively, rendered transparent, and overlaid. Overlapping regions projected as yellow, whereas disparate regions remained red or green. 4) Several neuronal landmarks (spines, branch points) were selected in a control image and linked in the x–y axes to form a polygon. This polygonal area was measured in square microns using the LSM 510 software during each experimental treatment. An expanding polygonal area indicated that neuronal components were moving apart in space. These values were analyzed using a repeated-measures ANOVA. Differences between control and treatment conditions measured in the same slices were determined by simple-effects analysis. Note that “glial” swelling alone will not spread apart neuronal components. The osmotic analogy is that neurons are like the fingers of a hand that dips a dry sponge in water: the fingers maintain their volume, whereas the sponge (the glial cells) swells.
Spine head analysis was performed blind to experimental conditions. Initially, we examined 104 spine heads on 36 different dendritic branches in 11 slices from 9 animals. Of these, 77 were exposed to the complete paradigm (control–hypo–hyper–wash aCSF) and so comprised the final data set. Only bright spines extending in the x–y axes were measured. Unambiguously, thin and mushroom spines were marked and then compared during control and osmotic stress (Kirov and Harris 1999). To assess changes in spines between different conditions, individual sections through a z-series of the dendrite were examined to ensure that they had not rotated out of view or had become obscured by their dendritic shaft or by one another. From individual sections of the image stack, the maximal width of the spine head was determined. The brightness of the spine (measured as GFP fluorescence) is proportional to its volume (Svoboda 2004), so relative volume could be estimated as the spine fluorescence divided by the dendritic fluorescence (Svoboda and others 1996). To correct for potential general changes in fluorescence intensity during different osmotic conditions caused by, for example, inconsistencies in excitation, the average pixel value at the brightest point of spine image was measured, background fluorescence was subtracted, and the data were normalized to the average peak fluorescence of the largest dendrite in the imaged field.
There were technical challenges involved with 2PLSM imaging of cell volume changes because the entire brain slice dramatically altered its volume during the initial minutes of osmotic challenge, OGD, and high-K+ treatment. Even though neurons did not obviously change volume during osmotic stress, changing glial volume invariably shifted the imaged neuronal components out of the original focal plane during the first 2 min of osmotic change. Note that this did not occur with glycerol exposure that, unlike mannitol, does not shrink slices (Andrew and others 1997). This movement necessitated recentering and adjusting the field of focus on the fly just prior to acquiring a new image stack. The imaged neuron could experience any combination of roll, pitch, or yaw that shifted neuronal components out of register in 2-dimensional space. This was primarily a problem when overlaying images. A positive aspect of the movement was that it provided an internal indicator that the focal plane was indeed shifting at the expected time (i.e., at the onset of osmotic change, OGD, or elevated [K+]0) when glial volume was changing.
To monitor changes in astrocytic volume, hippocampal slices (400 μm) were taken from 37- to 40-day-old albino mice of the FVB/N-Tg(GFAPGFP)14Mes/J strain. These mice display GFP fluorescence in astrocytes dispersed throughout the CNS (Zhuo and others 1997). Single astrocytic cell bodies, their processes, and their capillary end-feet are clearly visualized.
Imaging Changes in LT
To confirm that the tissue volume was altered as expected during osmotic challenge, LT changes were measured before, during, and after an acute 15-min shift in osmolality. Figure 1A shows a schematic of the passive volume responses expected. Actual effects of hypo-osmotic stress (−40 mOsm aCSF) on the neocortical slice are shown in Figure 1B. LT across gray matter began to increase within 1 min and plateaued in all layers within 12 min. The increase, indicative of cell swelling and thus greater tissue transparency, was maintained for the duration of the exposure. Layers II–III displayed the greatest sensitivity with a mean normalized peak LT change of 51%, whereas layer I showed the lowest response of 25%. On return to normosmotic aCSF, LT immediately decreased to baseline levels in all layers. There was neither a delayed response back toward baseline during the hypo-osmotic stress nor a baseline undershoot upon return to control aCSF, thereby indicating lack of a regulatory volume decrease (RVD). Figure 1C shows the effect of an acute hyperosmotic stress (+40 mOsm aCSF) using mannitol. Being membrane impermeable, it quickly draws water out of cells possessing plasma membrane water channels. LT was reduced maximally by 15 min. As with acute hypo-osmotic stress, layers II–III showed the greatest sensitivity with a mean peak LT change of −37% from control and layer I responded least at −23%. In all 9 slices, there was no evidence of a regulatory volume increase (RVI) as shown in Figure 1A. Unlike mannitol, glycerol permeates glial plasma membranes and so does not create an osmotic gradient between extracellular aCSF and intracellular cytoplasm. As an osmotically “ineffective” agent, +40 mOsm glycerol evoked a mean LT change of less than ±3.2% in any neocortical layer (Fig. 1D). Upon return to control aCSF (17:00–32:00), there was no significant change in LT across the slice, confirming that the LT decrease evoked by mannitol was osmotically driven.
The effects of −40 and +40 mOsm aCSF represented responses observed across a range of osmotic levels as demonstrated in Figure 1E–H. The normalized mean peak LT change (±standard deviation) in response to osmolality was plotted over time. Each layer responded proportionally to increasing degrees of both hypo- and hyperosmotic stress. The average peak LT response by each neocortical layer to the different levels of osmotic change was plotted in Figure 1I. In all layers, no sign of volume regulation was observed.
The addition of taurine to aCSF superfusing brain slices has been reported to facilitate RVD, so we repeated some of the previous experiments using slices that were prepared, incubated, and osmotically stressed in the presence of 1 mM taurine. As tested in 5 slices, RVD was still not observed at −50 or +50 mOsm by examining ΔLT changes in neocortex and hippocampus as above (not shown). In a second series of experiments, the CA1 field potential was orthodromically evoked during −50- or +50-mOsm exposure in 1-mM taurine slices (Fig. 2A). This technique indirectly measures altered cell volume by monitoring the voltage amplitude of the evoked population spike that is proportional to extracellular tissue resistance. Thus, voltage amplitude increases in response to decreasing osmolality and vice versa (Andrew and others 1989; Dudek and others 1990). The population spike amplitude increased within 2–3 min of exposure to −50 mOsm aCSF with no gradual recovery in the face of hypo-osmotic stress that would indicate RVD. Amplitude returned to normal in control aCSF and then reversibly decreased in hyperosmotic aCSF (Fig. 2A,B). The tissue shrunk as expected but with no indication of RVI.
The 2PLSM of Pyramidal Neurons and Their Dendrites
Having confirmed appropriate volume responses by the cortical gray matter, we searched for evidence that pyramidal neurons contribute to these volume changes using 2PLSM. Near the surface of hippocampal or neocortical slices, swollen pyramidal somata and grossly varicose dendritic arborizations were imaged (not shown). Regions deeper than 50 μM appeared healthy with occasional beaded dendrites originating from somata near the sliced surface. At this depth or deeper, a total of 21 pyramidal cell bodies in 12 fields and 50 dendrites in 14 fields were imaged after osmotic stress, OGD, or high-[K+]o exposure. All experiments started and ended in control (287–289 mOsm) aCSF. A field was first imaged for 15–20 min in control aCSF to confirm a stable signal before changing to hypo-osmotic aCSF (−40 mOsm, 15–20 min). A subset of slices was then exposed to hyperosmotic aCSF using mannitol (+40 or +80 mOsm). This order was occasionally reversed. Each osmotic treatment was 20–25 min in duration. In some cases, a field was exposed only to OGD.
The focal plane of a given optical section slowly shifted during the first 3 min of hypo-osmotic challenge, confirming that slice volume was indeed changing. However, neither cell bodies (Figs 3B2 and 4A2,B2,C2) nor dendrites (Figs 3C2 and 5A) detectably increased in size in −40 mOsm aCSF as sampled at 7–9 and 15–20 min. (The only exception is shown in Fig. 4C3 where neuronal swelling was irreversible, indicating damage.) Upon introduction of hyperosmotic aCSF of +40 or +80 mOsm, a shift in the focal plane again occurred over 2–3 min using mannitol, but not using equiosmolar glycerol. This is consistent with our LT data showing that mannitol, but not glycerol, shrinks slice volume (Fig. 1C,D). Despite the reduction in tissue volume, in no case was shrinking of pyramidal somata (Fig. 3B3) or dendrites (Figs 3C3 and 5A–D) detected. The relation between LT change and neuronal volume is shown in Figure 3. The gray matter swells (Fig. 3A2) and shrinks (Fig. 3A3) in response to hypo- and hyperosmotic conditions, respectively, and then returns to near baseline (Fig. 3A4). However, the soma (Fig. 3B1–B4) and dendrites (Fig. 3C1–C4) do not change volume. Most telling was the consistent shape and number of dendritic spines, indicating that their membrane was not recruited to accommodate cell expansion (Fig. 5E). However, in response to OGD, the same dendrites swell and bead (Figs 3C5, 4D2,E2, and 5C,D) and the same cell bodies swell (Figs 3B5,F and 4A6,B6,D,E).
Increased volume can be shown quantitatively as a transition from dark to light pixels if the GFP-filled neuronal compartment (light pixels) expands to comprise a larger proportion of the background field (dark pixels). Thus in Figure 4F1, there is no significant change from the control pixel distribution when cell body fields are imaged in −40 mOsm aCSF (F8,40 < 1, not significant [NS]). Pixel distributions from 12 dendritic fields did not change significantly when exposed to either hypo- or hyperosmotic treatment (Fig. 6A1). Both conditions showed a slight increase in light pixels upon osmotic stress (<5%) (F8,88 = 6.19, P < 0.01). Indeed, opposite to that expected if cells were shrinking in response to hyperosmotic treatment, image brightening was stronger than for hypo-osmotic treatment as suggested by a significant interaction between condition and binned pixel value (F8,88 = 5.03, P < 0.01). In contrast, OGD exposure results in a much more dramatic recruitment (∼10%) of pixels from dark to light in fields of cell bodies (Fig. 4F2, F3,8 = 7.88, P < 0.01) and dendrites (Fig. 6A2). Note that this measure is ineffective if swelling is accompanied by fading of the fluorophore as was commonly observed with OGD (Figs 4A6 and 6B2). Volume expansion was imaged (Fig. 5C) and plotted (Fig. 6A2) in the only dendritic field that did not fade following OGD.
There was a need to further quantify neuronal swelling (or lack of it) across the population, so we measured whether distances between neuronal landmarks within an imaged field were changed by osmotic challenge or OGD. Distinct points were joined to delineate a polygon (e.g., Fig. 5D), and the area in square microns was calculated using the LSM510 software. A swelling neuron will display greater point separation and so an increased polygonal area. Shrinking should elicit converging points and a reduced polygonal area. Polygon areas from somatic, dendritic, and axonal regions exposed to hypo- and hyperosmotic conditions (n = 21) were pooled and entered into a repeated-measures ANOVA. As shown in Figure 6D1, this revealed no difference in areas measured among the control, hypo-, and hyperosmotic treatments (F2,40 < 1). We wanted to confirm that these neurons were healthy and volume responsive, so a subset of these polygons were measured again following exposure to OGD for 10 min while 2PLSM sampling continued (Fig. 6D2). These data confirmed that neuronal swelling occurred during OGD but not during hypo-osmotic stress. Furthermore, no neuronal shrinking was detected during mannitol exposure. Change in polygonal area showed a main effect of condition (F3,36 = 5.06, P = 0.02), which was due to neuronal expansion during OGD (F1,36 = 12.07, P < 0.01) but not during osmotic stress (all F values < 1). Finally, we examined all fields exposed to OGD (n = 22), some without previous osmotic change (Fig. 6D3). The volume increase was again significant (F1,21 = 9.7, P < 0.01) and was approximately 12% greater than the control condition.
Figure 5E1 shows the mean spine head diameter (n = 77) in control aCSF, following a 15-min exposure to 40 mOsm aCSF, then following a 15-min exposure to +40/+80 mOsm aCSF, and finally following a wash in control aCSF. There was a significant main effect of condition (control, hypo-osmotic, hyperosmotic, and wash) (F3,228 = 6.74, P < 0.01). Simple-effects analysis showed that this was due to a diameter increase under hyperosmotic conditions (F1,228 = 7.02, P < 0.01) as there was no change from control upon exposure to hypo-osmotic aCSF (F1,228 = 2.96, NS), and control did not differ from wash (F1,228 < 1). Figure 5E2 shows the mean spine head brightness for the same 77 spines. There was also a significant main effect of condition (F3,228 = 45.14, P < 0.001), again resulting from an increase in fluorescence upon exposure to hyperosmotic (F1,228 = 82.0, P < 0.001) media. There was no change from control upon exposure to hypo-osmotic aCSF (F1,228 < 1), and control did not differ from wash (F1,228 = 1.1, NS). As an internal control, this brightening in the hyperosmotic condition was not observed in measures of larger fluorescent volumes contained within parent dendrites (F2,20 < 1), suggesting that hyperosmotic media was not concentrating the dye in the dendrites. In summary, hypo-osmolality did not significantly affect spine head diameter or brightness, but hyperosmolality reversibly increased both values, indicating a small increase in spine head volume. This is the opposite expected for an osmotic effect, and the cause is unclear.
The 2PLSM of Cerebellar Axon Terminals
The effect of acute osmotic challenge on mammalian axons and their terminals was directly observed using 2PLSM of the cerebellar slice preparation, again from transgenic mice of the B6.Cg-TgN(thy1-GFP)MJrc strain. Cerebellar MFs originate from cell bodies located in several nuclei of the spinal cord and brain stem. Each MF is an axon that enters the granular layer of the cerebellar cortex and forms a chain of intermittent terminal clusters (rosettes) up to 15 μm in diameter (Fig. 7A). Single terminals may sprout from a rosette (Fig. 7B,D). MFs containing GFP (n = 12 rosettes from 4 fields) were imaged 60–150 μm deep in the slice during 20 min of overhydration (−40 mOsm) or dehydration (+40 or +80 mOsm). As with cortical slices described above, 2 min of slice expansion (or shrinkage) initially altered the focal plane when osmolality was changed. However, again osmotic swelling or shrinking of terminals was not detected when corresponding hypo- and hyperosmotic image stacks were overlaid whether osmotic stress was for 3.5 (Fig. 7A3) or 18 min (Fig. 7A4). The 3–5 min immediately following osmotic challenge would sample well before purported volume regulation could compensate for induced swelling or shrinkage. When these same MF terminals were then imaged during OGD for 10 min, many axons became beaded (asterisks in Fig. 7B4,C2,D2), and their terminals dramatically swelled post-OGD (arrows in Fig. 7B4,C2,D2). As seen with pyramidal somata and dendrites, pixel distribution in field images was not significantly altered in hypo-osmotic media (Fig. 7E1, F3,8 < 1). In contrast, axon terminal volume did increase post-OGD as shown by the significant recruitment of dark to light pixels (Fig. 7E2, F3,8 = 3.91, P < 0.05).
The 2PLSM during High-K+ aCSF
Exposure to aCSF with 26 mM K+ immediately depolarizes neurons and glia and probably saturates the astrocytes' capability to reduce extracellular [K+]; it can evoke neuronal cell body swelling within 2 min (Figures 4A4,A5; 4B4,B5 and 8A) that is reversible. Likewise, dendrites reversibly swell in response to elevated K+ and return to control volume by 20 min (Fig. 8B,C). Figure 8D shows a histogram of the mean change in pixel gray value relative to control for a pool of 4 somatic fields and 3 dendritic fields exposed to 26 mM K+ aCSF for 2–5 min. There is a general increase in cell volume as signified by a recruitment of dark to light pixels. Only the experiments showing consistent GFP fluorescence were analyzed because fading itself interferes with the pixel distribution.
Neuronal resistance to volume change under acute osmotic challenge has 3 possible explanations. The neurons might regulate volume second by second without cell volume detectably increasing or decreasing, which would be unprecedented for any cell type under acute stress. Second, some unidentified water channel might close or be rapidly retrieved from the plasmalemma upon hypo- and hyperosmotic stress; no known channel displays such behavior. A third explanation is that resting neuronal membrane is largely water impermeable, similar to the plasmalemma of oocytes (Preston and others 1992; Agre and others 2002), distal renal tubule cells (Somjen 2004; Tian and others 2004), and the epithelial lining of stomach and bladder (Krylov and others 2001).
It is widely thought that mammalian neurons in intact animals initially swell passively in response to acute hypo-osmotic stress and then regulate their volume back toward baseline over a matter of minutes, as observed in cultured and dissociated nerve cell bodies (Gullans and Verbalis 1993; Pasantes-Morales and others 1993; Leaney and others 1997; Aitken and others 1998; Patel and others 1998; Inoue and others 2005). Our report challenges the assumption that CNS neurons regulate their volume during acute osmotic stress (or even need to) because their plasma membrane so poorly conducts water. Although water will partition into phospholipid bilayers and diffuse across plasma membrane in response to osmotic pressure, some cell types noted above resist high osmotic gradients with specialized membranes that impede the equilibration of water between compartments. These “barrier membranes” not only lack aquaporins but also have outer leaflets whose lipid components confer both low fluidity and high resistance to water permeation (Krylov and others 2001). High microviscosity and low transmembrane water conduction will each resist cell volume perturbation. Like fingers holding a dry sponge and submerging it in water, we propose that CNS neurons maintain their volume, whereas the surrounding glial network (the sponge) osmotically swells.
Brain Slices and Osmotic Challenge
Imaging LT through submerged tissue slices has proven useful for measuring brain slice swelling/shrinkage during acute osmotic change (Andrew and MacVicar 1994; Andrew and others 1997; Aitken and others 1998), excitotoxicity (Polischuk and others 1998), and spreading depression (Somjen 2001; Anderson and Andrew 2002). With hypo-osmotic swelling, tissue LT increases proportionally with extracellular tissue resistance as the extracellular space decreases with glial swelling. Conversely, LT is reduced in hyperosmotic media coinciding with decreased extracellular resistance as cells shrink (Andrew and others 1997). Several processes likely underlie the increased transmittance displayed by swollen glial cells. These include organelle swelling, unfolding of the plasma membrane into a more planar configuration, and reduction of the cytoplasmic refractive index (Jarvis and others 1999). On the other hand, within several minutes of OGD, healthy pyramidal cell dendrites lose spines and morph into strings of beads. Each bead is 2–5 μm in diameter, an ideal range for scattering near-infrared light (Malm 2000), so dendritic conformation is another major parameter that influences LT. In this case, beading reduces LT, providing an indicator of neuronal damage.
Neocortical slices exposed to hypo-osmotic stress swelled but displayed no evidence of an RVD. All LT responses reached a steady maximum level and then returned to baseline in control aCSF. Preincubation in 1 mM taurine did not reveal RVD in our hands as has been previously reported (Kreisman and Olson 2003). All slices exposed to hyperosmotic stress displayed reduced LT in proportion to the level of elevated osmolality with no evidence of an RVI. These findings are similar to those reported under periods of up to 40 min of osmotic stress in hippocampal slices (Andrew and MacVicar 1994; Andrew and others 1997). The LT changes are strongly correlated with osmotic swelling and shrinking by astrocytes (Kirov and others 2006) but not by neurons in our cortical slices.
Neocortical slices were also exposed to hyperosmotic stress using the osmotically ineffective substance glycerol, which raises osmolality but does not alter glial volume likely because glycerol permeates specific aquaporin channels (aquaglyceroporins) in glia (Amiry-Moghaddam and Ottersen 2003; Badaut and others 2004). Accordingly, glycerol changes LT in brain slices by less than ±3.2% in any neocortical layer, an insignificant effect. We have found with 2PLSM that +40 mOsm glycerol has no effect on neuronal volume (data not shown). So lacking aquaglyceroporins, the neurons probably exclude glycerol (as they do mannitol) but do not shrink because (as with mannitol) their plasma membrane impedes water efflux. Note that glycerol administration is useful clinically to dehydrate the brain because it evokes diuresis by the kidney.
Neurons and Osmotic Challenge
Using 2PLSM, we imaged GFP-containing pyramidal cell bodies, their dendrites and spines in cerebral cortex, as well as large axon terminal clusters, the cerebellar MF that originate from somata in brain stem and spinal cord. Under osmotic challenge ranging from −40 to +80 mOsm, all 4 compartments resisted any volume change but subsequently swelled during maintained depolarization induced with OGD or 26 mM K+. A reasonable explanation is that neuronal membrane at normal resting potential poorly conducts water. Under physiological conditions, aquaporins are the family of water-selective channels that conduct osmotically driven water across cell membranes (Agre and others 2002; Amiry-Moghaddam and Ottersen 2003). A single human aquaporin-1 channel conducts ∼3 billion water molecules per second, the direction dependent on the prevailing osmotic gradient (Zeidel and others 1992). Much less water follows the active transport of ions or small organic molecules (Amiry-Moghaddam and Ottersen 2003). Glial cells and brain capillaries highly express specific aquaporins (Rash and others 1998; Amiry-Moghaddam and Ottersen 2003; Badaut and others 2004; Solenov and others 2004). Using in situ hybridization, one study has suggested that CNS neurons express low levels of aquaporin4 (AQP4) mRNA (Venero and others 1999), but several immunocytochemical studies have not detected AQP4 in neurons (e.g., Rash and others 1998). AQP9 is detected specifically in catecholaminergic neurons in noncortical regions of rat (Badaut and others 2004), so these neurons may prove to be osmoresponsive.
Why do neurons lacking aquaporins suddenly swell when strongly depolarized, as during seizure (Somjen 2004) and spreading depression (Somjen 2001; Anderson and Andrew 2002)? Some water enters as hydration shells of influxing Na+, Cl−, and Ca2+, and not all water molecules are stripped from each sodium ion as it passes through its voltage-sensitive channel (Kandel and others 2000). Maintained depolarization also swells neurons during Na+/K+ pump failure arising from OGD (Obeidat and others 2000) or from cooling (Volgushev and others 2000; Kirov and others 2004). Energy deprivation alters the conformation and distribution of cytoskeletal actin (Gisselsson and others 2005), and water enters by opening of Na+ channels and probably as yet unidentified large-pore channels that maintain the anoxic depolarization induced by OGD (Anderson and others 2005). Cotransport of water linked to high extracellular glutamate and K+ arising from depolarization may move water into cells against the resultant osmotic gradient (MacAulay and others 2004). Note that the conduits of water influx to strongly depolarized neurons are still poorly characterized.
Intracellular recordings from hippocampal slices (Ballyk and others 1991; Saly and Andrew 1993) and neocortical slices (Rosen and Andrew 1990, 1991) have shown that varying the extracellular osmolality between −60 and +60 mOsm did not change the resting potential, action potential threshold, or total cell input resistance of pyramidal neurons. The resiliency of pyramidal cell volume under osmotic challenge helps explain this stability. This is not at odds with data showing that overhydration increases cortical excitability, whereas dehydration dampens it because intrinsic neuronal changes are not involved. Rather, hypo-osmolality (−40 mOsm) increases the number of spontaneous excitatory post-synaptic potentials (EPSPs) and amplitude of evoked EPSPs as recorded intracellularly (Ballyk and others 1991; Saly and Andrew 1993). Together with increased field effects (Dudek and others 1990; Roper and others 1992), the enhanced EPSP input promotes seizure activity. Such excitability promotes acute hypo-osmolar syndromes (Andrew and others 1989; Andrew 1991; Finberg 2000). Hyperosmolality induced with mannitol (but not glycerol) has the opposing effect of decreasing EPSP activity and ephaptic effects, thereby reducing population excitability without altering the endogenous electrophysiological properties of the pyramidal neurons (Rosen and Andrew 1990, 1991; Ballyk and others 1991; Saly and Andrew 1993).
Neuronal volume stability without volume regulation during acute osmotic stress was also observed when imaging the axon terminals of neurons originating in brain stem and spinal cord. This implies that low water permeability is a general feature of resting CNS neurons, although additional neuronal types need to be tested. Again, without aquaporins in axonal membrane, little transmembrane flux occurs regardless of the imposed extracellular osmolality. Whereas this concept may be novel to neuroscientists, it is not to nephrologists (Tian and others 2004). Cells lining distal kidney tubules lack aquaporins and are therefore impermeable to water despite intratubule osmolalities ranging from 100 (dilute urine) to >600 mOsm (concentrated urine). Interestingly, this nephron segment is also the location of osmoresponsive channels (TRPV4) that may detect transmembrane osmotic change (Tian and others 2004) exactly where it is needed, that is, where a minimal transmembrane osmotic gradient exists. These channels appear restricted within CNS to certain osmosensitive and mechanosensory neurons (Liedtke and Friedman 2003). Finally, a very low but consistent leak of water into neurons during several hours of hypo-osmolality would explain a loss of intracellular osmolytes, such as taurine observed, for example, in Purkinje neurons (Nagelhus and others 1993). The accompanying water loss would counteract seepage across an otherwise poorly permeable membrane.
Our study provides physiological confirmation of the hypothesis that if neurons lack functional aquaporins, they can resist volume change in the face of acute osmotic stress. Low water permeability provides the brain's neuronal compartment with protection from abrupt osmotic shifts not afforded by the highly water-permeable (i.e., aquaporin rich) brain capillaries (Grande and others 1997) and glia (Badaut and others 2004; Solenov and others 2004). However, conditions evoking maintained depolarization (ischemia, head trauma, and seizure) cause prolonged opening of channels that mediate significant water influx to cell bodies, dendrites, and axons. Our findings question the generalization made in current neurology textbooks (e.g., Ferrante 2003) that all brain cell membranes are normally water permeable and that the different fluid compartments of the brain are therefore in osmotic equilibrium.
We would like to thank Ms Christine Molnar for technical help. The founding mice of our GFP transgenic colony were a generous gift from Dr Joshua Sanes, Harvard University. This work was funded by the Heart and Stroke Foundation of Ontario #T-4478 (RDA), the Canadian Institutes of Health Research #MOP 69044 (RDA), and the National Institutes of Health #KO1MH02000 (SAK). Conflict of Interest: None declared.