Animals were housed under standard conditions in the animal facility of Forschungszentrum Jülich (SV129) or Heinrich Heine University Düsseldorf (Balb/C) according to institutional guidelines under a 12-h light/dark cycle and in small groups with food and water provided ad libitum.

After decapitation under isoflurane anesthesia brains were rapidly removed and placed in oxygenated, ice-cold preparation solution containing (in mM) 125 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 26 NaHCO₃, 0.5 CaCl₂, 5 MgCl₂, and 25 glucose. Sagittal hippocampal or coronal cortical slices (250 μm thickness) were cut with a microtome (Microm HM650V, Thermo Scientific, Walldorf, Germany; frequency 60 Hz, amplitude 1 mm, drive 10) and transferred to a gauze slice holder in oxygenated Ringer’s solution at 37°C (carbogen; 95% O₂/5% CO₂, Untiet et al., 2017). Glial cells were stained with sulforhodamine 101 (SR101, Sigma-Aldrich, St. Louis, MO, United States) in a preparation Ringer’s solution that contained 2 μM SR101 for 20 min at 37°C (Kafitz et al., 2008), followed by a 10 min incubation in standard oxygenated Ringer’s solution with (in mM) 125 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 26 NaHCO₃, 2 CaCl₂, 1 MgCl₂, and 20 glucose, at 37°C. The stained acute tissue slices were kept at room temperature (22–24°C) for at least 30 min before use. Slices were constantly perfused with oxygenated standard Ringer’s solution. All experiments were performed within 5–8 h of brain removal.

Brain slices were incubated in oxygenated standard Ringer’s solution containing 3.5 mM 1-(ethoxycarbonylmethyl)-6-methoxyquinolinium bromide (MQAE, Sigma-Aldrich, Merck, Darmstadt, Germany) (Verkman, 1990) for 30–40 min at room temperature and then transferred to an imaging chamber attached to an upright fluorescence microscope. Experiments were performed using two different imaging systems, since we acquired a new two-photon excitation fluorescence microscope with FLIM modality during the course of this study. Roughly half of the experiments utilized an A1 MP microscope (Nikon, Amsterdam, Netherlands) equipped with a 25× water immersion objective [NA1.1; working distance (WD) 2 mm; XYZ, Nikon] and a mode-locked Titan-Sapphire laser (Mai Tai DeepSee, Newport Spectra Physics; Irvine, CA, United States; output power 2.3 W at 750 nm); the other half of the experiments were performed on a LSM880 microscope (Zeiss, Jena, Germany) equipped with a 20× water immersion objective (NA 1.0, WD 2.1 mm; XYZ, Zeiss) linked to a tunable laser (InSight X3, Newport Spectra Physics, Darmstadt, output power 1.9 W at 750 nm). Two-photon excitation (λ_exc = 750 nm) was carried out at 80 MHz with either 100 fs light pulses or 120 fs light pulses, with identical results. Mean fluorescence lifetimes were measured using multidimensional time-correlated single-photon counting (TCSPC) in a volume of 0.08 μm³ per individual pixel, resulting in a three-dimensional resolution of approximately 1.3 μm in the z-axis and 0.35 μm in the x- and y-axes (Zipfel et al., 2003).

Six to eight images were taken for each measurement, starting at a depth of about 30 μm from the upper cutting edge and reaching not further than approximately 100 μm into the slice. For all brain regions studied, we only included glial cells in the analysis that were clearly stained by SR101, sufficiently loaded with MQAE, and showed region-specific characteristic morphology. Following two-photon (Nikon A1 MP microscope: 900 nm) or one-photon (Zeiss LSM880 confocal microscope: 561 nm) excitation, emitted SR101 fluorescence was filtered using an appropriate bandpass filter (Nikon A1 MP: 595AF60, 565–637 nm; Omega Optical, Brattleboro, VT, United States) or a GaAsP detector (Zeiss, LSM880) with a freely selectable emission detection band (570–700 nm). A single region of interest (ROI) was defined for each glial soma, and the fluorescence lifetime of cells was calculated as the average fluorescence lifetime (τ_ave) of all pixels in the ROI. Fluorescence was filtered by separating MQAE fluorescence (peak emission: 460 nm) from autofluorescence (bandpass 445bp90: 400–490 nm; Omega Optical, Brattleboro, VT, United States) and recorded with a GaAsP hybrid photodetector (HPM-100-40, Becker & Hickl, Berlin, Germany) with non-descanned detection. TCSPC electronics (SPC-152; Becker & Hickl, Berlin, Germany) and acquisition software were used for FLIM as previously described (Kaneko et al., 2004; Untiet et al., 2017).

Fluorescence lifetime images were generated using SPCImage 6.0–8.3 (Becker & Hickl, Berlin, Germany). We usually summed fluorescence intensity decays (FIDs) over nine pixels and assigned the resulting value to the central pixel (bin factor 1). In a few cases, in which the number of photons per pixel was below the critical value of 2000 counts (Figure 1), we used a bin factor of 2 (i.e., we summed photon distributions over the central and the surrounding 24 pixels). Fluorescence decays were fitted with bi-exponential functions (Kaneko et al., 2004; Gilbert et al., 2007; Funk et al., 2008; Untiet et al., 2017), and the average fluorescence lifetime (τ_ave)

was used to calculate [Cl^–]_int values (τ_x = lifetime of the exponential component; a_x = respective amplitude).

Fitting the sum of exponentials to fluorescence decay curves from TCSPC experiments can result in systematic errors for low photon counts (Maus et al., 2001). To define conditions that circumvent this limitation of FLIM-based concentration measurements, we studied the consequences of low photon numbers with the two-photon fluorescence microscopes with FLIM modality used for measuring glial [Cl^–]_int (Figure 1). Using of droplets of aqueous MQAE solutions with [Cl^–] of 15, 40, or 100 mM, we found that at high photon numbers (>1000 photons) the determined fluorescence lifetime is identical to the fluorescence lifetime determined independently with a cuvette-based fluorescence lifetime spectrophotometer (TimeHarp 100, PicoQuant, Berlin, Germany) and is not modified by further increasing the photon count. We then modified the photon number of FIDs by changing the applied binning options. Smaller binning reduced the photon numbers per decay and significantly decreased fluorescence lifetimes (Figure 1A). We combined binning with variation of the excitation power and found that – for less than 1000 photons per FID – the estimated fluorescence lifetime deviates by up to 25% from the correct value (Figure 1B). In similar experiments with two other chloride concentrations (40 mM NaCl and 100 mM NaCl) and shorter fluorescence lifetimes similar, but less pronounced effects of low photon numbers were observed (Figure 1B). A similar effect of low photon counts was observed in experiments with glial cells in the CA1 region of an acute hippocampal slice [Figure 1C (three cells), I: 4.22 ns (296 photons per FID); II: 4.24 ns (269 photons per FID); III: 4.15 ns (473 photons per FID)]. These lifetimes predict incorrectly high chloride concentrations (I: 30.8 mM; II: 30.3 mM; III: 32.1 mM). When the same area was measured with a longer acquisition time, photon numbers per FID were well above 1000 and we obtained the expected fluorescence lifetimes [Figure 1D, I: 4.88 ns (4817 photons per FID); II: 4.96 ns (3941 photons per FID); III: 4.73 ns (6240 photons per FID)], as well as chloride concentrations (I: 19.3 mM; II: 18.1 mM; III: 21.6 mM) near to the mean of [Cl^–]_int estimated for hippocampal (CA1) glia cells (Figure 2F). Therefore, we chose to use a minimum number of 2000 photons per FID in all of our experiments.

For calibration (Figures 2D, 3B) acute tissue slices were incubated in HEPES-buffered solutions containing (in mM) 140 K⁺, 10 Na⁺, 10 HEPES, 10–60 Cl^–, 80–130 gluconate, adjusted to 310 mOsm/L with K-gluconate and to pH 7.4 with KOH, supplemented with 20 μM nigericin (sodium salt; Sigma-Aldrich, Merck, Darmstadt, Germany) and 20 μM tributyltin (chloride salt; Sigma-Aldrich, Merck, Darmstadt, Germany) (Chao et al., 1989; Bevensee et al., 1997; Kaneko et al., 2004; Kovalchuk and Garaschuk, 2012). Since MQAE is quenched much less effectively by HCO₃^– than by Cl^– (Kaneko et al., 2004), intracellular bicarbonate is not expected to contribute to the MQAE lifetime and was therefore ignored in calibration experiments.

DL-threo-β-benzyloxyaspartic acid (DL-TBOA) (100 μM; Tocris Bioscience, Bristol, United Kingdom), R-(+)-butylindazone, (dihydroindenyl)oxy alkanoic acid (R-(+)-DIOA, 100 μM, Sigma-Aldrich, Merck, Darmstadt, Germany) or 3-butylamino-4-phenoxy-5-sulfamoyl benzoic acid (bumetanide; 40 μM, Sigma-Aldrich, Merck, Darmstadt, Germany) were added to both the incubation (20 min incubation) and perfusion solutions. In contrast, 2-amino-5,6,7,8-tetrahydro-4-(4-methoxyphenyl)-7-(naphthalen-1-yl)-5-oxo-4H-chromene-3-carbonitrile (UCPH-101, Abcam, Cambridge, United Kingdom) inhibits excitatory amino acid transporter 1 (EAAT1/GLAST) with very slow unblocking kinetics (Abrahamsen et al., 2013) and is not washed out within the imaging experiment duration (around 30 min). Hence, in relevant experiments, brain slices were incubated with 20 μM UCPH-101 for 20 min and then perfused with standard Ringer’s solution without UCPH-101 for imaging. MQAE fluorescence is quenched not only collisionally by chloride ions (Verkman, 1990), but also by the chemical blockers used in our experiments. Consequently, to account for changes in fluorescence lifetimes, we performed additional calibrations after 20 min incubation with blockers (Figures 4B,D,F,H). The importance of these additional calibration was proven by differences in [Cl^–]_int calculated using the original calibration (without blocker) compared with corrected values after additional calibration. Re-calibration of MQAE lifetimes with three blockers (bumetanide, R-(+)-DIOA, and DL-TBOA) is necessary to accurately assess the roles of specific chloride transporters/channels in setting the [Cl^–]_int. Since MQAE self-quenching reduces the fluorescence lifetime (at fixed [Cl^–]) with increasing [MQAE] (Kaneko et al., 2002; Gensch et al., 2015), we performed calibration experiments using the same MQAE loading protocols as in the actual measurements.

Transient energy restriction in the ischemic penumbra was mimicked by removing glucose from standard Ringer’s solution and adding 5 mM NaN₃ and 2 mM 2-deoxy-D-glucose (Sigma-Aldrich) (Gerkau et al., 2018). Slices were perfused for 2, 5, or 10 min with the ischemia cocktail, and then continuously perfused with oxygenated standard Ringer’s solution during imaging. FLIM measurements under ischemic conditions were always performed as the first experiments of the day, with two to five slices tested for different ischemic periods (2, 5, or 10 min). Fluorescence lifetimes were determined with scanning times of at least 40 or 80 s. Slices with insufficient MQAE loading or strong slice movement during scanning were not included in the analysis. We approximated glial cell volumes from FLIM images by defining a circular ROI surrounding the glial soma and assuming a spherical shape for the glial cell soma (Supplementary Figures 1–3).

P14–P18 mice (both sexes) were killed by CO₂ anesthesia and decapitation. The brains were removed and placed in oxygenated, ice-cold preparation saline (in mM: 125 NaCl, 2.5 KCl, 0.5 CaCl₂, 6 MgCl₂, 1.25 NaH₂PO₄, 26 NaHCO₃, and 20 glucose). Parasagittal brain slices (250 μm) were obtained using a vibratome (Microm HM 650V, Thermo Scientific, Planegg, Germany or 7000smz-2, n.p.i.Tamm, Germany) and incubated for 20 min at 34°C in standard oxygenated Ringer’s solution. During the incubation period, SR101 (0.5–1 μM) was added to selectively stain astrocytes (Kafitz et al., 2008). After staining, brain slices were maintained in standard Ringer’s solution at room temperature until use.

During experiments, brain slices were continuously perfused with standard Ringer’s solution at room temperature. Hypo-osmotic stress was induced by reducing the NaCl concentration of standard Ringer’s solution to 75 mM, thereby decreasing the osmolarity from about 310–220 mOsm/L. Chemical ischemia was induced by perfusing slices with glucose-free standard Ringer’s solution containing 5 mM NaN₃ and 2 mM 2-deoxy-D-glucose (Gerkau et al., 2018). A motorized confocal laser scanning microscope (Nikon Eclipse C1: Fluor 60×/1.00 W, software EZ-C1 3.91 Nikon Instruments, Düsseldorf, Germany) was used to document SR101 fluorescence. Z-stacks were taken at 0.6 μm increments and maximum intensity projections (MIPs) were calculated in ImageJ. Images were deconvolved using Huygens Professional Software (Scientific Volume Imaging, Hilversum, Netherlands).

To simulate ischemia-induced changes in [Cl^–]_int, we modified a recently developed model (Kalia et al., 2021) that describes the temporal evolution of intra- and extracellular Na⁺, K⁺, Cl^–, Ca²⁺ and glutamate concentrations in neurons and astrocytes using the following differential equations:

with Nxi denoting the number of moles of X in compartment i, z_X the valence of ions X, and F Faraday’s constant. The term Ijx,i describes currents mediated by the transporter j with respect to ion X in compartment i. The compartments are allowed to swell and shrink in response to osmotic gradients across the membrane. Volumes W_i of both compartments and molar ionic amounts were used to determine the ion concentrations. The term q denotes several voltage-dependent gates that are used to model some of the ion channels.

Currents Ijx,iⁱ mediated by ion exchangers/cotransporters or channels were modeled as described in Kalia et al. (2021). In the neuron, the following currents were considered:

For astrocytes the model includes the following currents:

Our model differs from the earlier version (Kalia et al., 2021) by the addition of KCC transporters as the Cl^– transport system in the astrocytic plasma membrane, in addition to the NKCC1 and Cl^– leak channels. Similarly, to the neuronal KCC, we assumed that the transport number J_KCC changes proportionally to the driving force of this coupled transporter, as follows:

This gives rise to the additional currents IKCCK+,a and IKCCCl-,a with

We did not explicitly model the glial anion conductances generated by EAAT anion channels; these anion efflux pathways are represented as components of the anion leak conductance of the modeled glial cells.

Our model describes the dynamics of molar amounts NXi of the ions Na⁺, K⁺, Cl^–, Ca²⁺, and glutamate, and compartmental volumes W_i, for i = {n, a}.

We assume total volume and ion amounts to be constant for the entire simulation, by fixing constants C_X such that,

The simulations were performed at room temperature and used initial conditions (baseline resting conditions) and parameters as in Kalia et al. (2021). Under resting conditions, ion concentrations are stable, and all ion fluxes sum up to zero. This restriction was used to compute leak conductances for all ions that were otherwise not fixed.

Figure 2 illustrates fluorescence lifetime imaging of acute MQAE-loaded hippocampal slices following two-photon excitation. CA1 and DG astrocytes and RGL cells were identified based on SR101 fluorescence (Figures 2A,B,C; Kafitz et al., 2008) and their characteristic morphology and localization. CA1 astrocytes were imaged in the stratum radiatum and DG astrocytes in the polymorphic layer at the interface between granule cell layers. The RGL subpopulation of glial cells are easily identified by SR101 staining owing to their characteristic unipolar morphology, with a single main extension projecting through the granule cell layer, and cell body located in the subgranular zone of the DG (Mori et al., 2006; Brunne et al., 2010; Jungblut et al., 2012). They express stem cell markers and can differentiate into either granule neurons or astrocytes (Berg et al., 2018).

MQAE fluorescence is collisionally quenched by Cl^– ions, resulting in an inverse linear relationship between fluorescence lifetime and chloride concentration (Verkman, 1990):

where τ is the MQAE fluorescence lifetime at a given [Cl^–]_int, τ₀ is the MQAE fluorescence lifetime in the absence of chloride, and K_SV is the Stern–Volmer constant. MQAE is not only quenched by Cl^–, so to account for the effects of other quenchers K_SV must be determined for each cell type. We calibrated MQAE lifetimes for various [Cl^–]_int that were preset by permeabilizing the cell membrane with nigericin and tributyltin and then incubating cells in solutions containing 140 mM K⁺ and the chosen [Cl^–] (Chao et al., 1989; Bevensee et al., 1997; Kaneko et al., 2004; Gensch et al., 2015; Untiet et al., 2017). Linear regression analysis of the average fluorescence lifetime at different intracellular chloride concentrations ([Cl^–]_int) derived the K_SV and τ₀ for the three types of hippocampal glial cells (Figure 2D and Table 1). Using the obtained K_SV values, we calculated average [Cl^–]_int values (Figure 2F) of 28.4 ± 3.0 mM in DG astrocytes, 20.6 ± 2.5 mM in CA1 astrocytes, and 20.3 ± 0.7 mM in RGL cells.

We next measured astrocytic [Cl^–]_int in layer II/III of the neocortex (Figure 3). The K_SV for MQAE in neocortical astrocytes (Table 1) is in the same range as the K_SV values obtained for hippocampal CA1 astrocytes and RGL cells. However, the mean [Cl^–]_int is much smaller in neocortical astrocytes (14.0 ± 2.0 mM) than in hippocampal astrocytes.

Taken together, our findings reveal a marked regional heterogeneity in resting [Cl^–]_int in astrocytes. Moreover, average [Cl^–]_int of hippocampal RGL cells is lower than that of cerebellar Bergmann glia, another class of radial glial cells (Untiet et al., 2017).

We recently demonstrated that chloride accumulation via the NKCC1 electroneutral cation and chloride efflux through EAAT anion channels control [Cl^–]_int of Bergmann glia cells (Untiet et al., 2017). EAATs are secondary-active glutamate transporters that can also function as anion channels (Fairman et al., 1995; Wadiche et al., 1995; Machtens et al., 2015; Fahlke et al., 2016). Glial cells additionally express the K⁺-Cl^– cotransporters KCC1 and KCC3 (Mount et al., 1999; Yan et al., 2001; Ringel and Plesnila, 2008) that mediate coupled potassium-chloride efflux. The Cl^–/HCO₃^– exchanger, AE3, is not expressed in glial cells (Kopito et al., 1989; Hentschke et al., 2006), and we therefore did not test whether blocking AE3 or Na⁺-HCO₃^– transporters affect [Cl^–]_int.

We assessed the function of glial anion transporters in hippocampal and neocortical glial cells by measuring [Cl^–]_int in the presence of specific inhibitors: i.e., 40 μM bumetanide (NKCC1 blocker) (Payne et al., 2003), 100 μM R-(+)-DIOA (KCC1–3 blocker) (Mercado et al., 2000), 20 μM UCPH-101 (EAAT1/GLAST blocker) (Abrahamsen et al., 2013), or 100 μM DL-TBOA (non-specific EAAT blocker) (Shimamoto et al., 1998; Figures 4A,C,E,G). This approach is complicated by blocker fluorescence and blocker quenching of MQAE fluorescence that modify MQAE lifetimes and prevent an accurate determination of [Cl^–]_int from standard calibration results. Therefore, we performed MQAE calibration in the presence of bumetanide, DL-TBOA, or R-(+)-DIOA (Figures 4B,D,F,H), and observed substantial variation in characteristic Stern–Volmer parameters for MQAE (Table 1).

These experiments allow for distinction of two classes of glia cells that differ in responses to chloride transport blockers (Figures 4A,C,E,G). Whereas bumetanide decreased the [Cl^–]_int to 10.9 ± 3.1 mM in CA1 (p = 7.91⋅10^–8) and 19.9 ± 1.5 mM in DG (p = 5.51⋅10^–5), it had no significant effect on the [Cl^–]_int of neocortical astrocytes or RGL cells. On the other hand, R-(+)-DIOA increased [Cl^–]_int in neocortical astrocytes to 26.6 ± 8.7 mM (p = 1.37⋅10^–4) and RGL cells to 27.9 ± 4.2 mM (p = 1.15⋅10^–4), but had no effect on the [Cl^–]_int of CA1 or DG astrocytes.

Blocking EAAT1 by UCPH-101 increased [Cl^–]_int most prominently in neocortical astrocytes (21.9 ± 2.3 mM; p = 0.008) and less effectively in CA1 astrocytes (25.0 ± 1.7 mM; p = 0.003) and RGL cells (24.4 ± 2.5 mM; p = 0.013), but had little effect on the [Cl^–]_int of DG astrocytes. DL-TBOA blocks both glial glutamate transporters (EAAT1/GLAST and EAAT2/GLT-1) and increased [Cl^–]_int to a greater extent than UCPH-101 in neocortical astrocytes (25.5 ± 4.1 mM; p = 3.37⋅10^–4) and significantly increased the [Cl^–]_int of RGL cells (32.8 ± 2.7 mM; p = 1.09⋅10^–8). DL-TBOA had no apparent effect on astrocytic [Cl^–]_int in the hippocampus.

We conclude that differences in the resting [Cl^–]_int of different astroglial cell types are associated with differences in the number and/or activity of Cl^– import and export/efflux pathway proteins.

Na⁺-K⁺-ATPases are primary active transporters that generate Na⁺ and K⁺ gradients across glial cells. Since the majority of ATP is produced by oxidative phosphorylation, oxygen restriction reduces cellular ATP levels and Na⁺-K⁺-ATPase activity within short time. Reduced ATP may increase [Na⁺]_int and [K⁺]_out (Hertz et al., 2015; Gerkau et al., 2017) and diminish the membrane potential; it may thus modify all transport processes involved in the chloride homeostasis of glia cells. A major cause of energy restriction is brain ischemia, a neurological complication of vascular disease and a leading cause of disability and death in our aging population. The consequences of ischemia depend on severity of cell damage: extensive cell death occurs in the core ischemic zone, whereas cells in the ischemic penumbra may recover after reperfusion. We mimicked the effects of transient energy restriction in the ischemic penumbra by transiently removing glucose, blocking oxidative phosphorylation with sodium azide, and blocking glycolysis with 2-deoxy-D-glucose, followed by [Cl^–]_int measurements under re-perfusion with oxygenated standard Ringer’s solution. In neocortical tissue slices, application of these components for 2 min induced intracellular ion disturbances (increase in Na⁺, Ca²⁺ oscillations) that were similar to those obtained during spreading depolarization waves in peri-infarct cortical regions in the rodent brain in vivo (Gerkau et al., 2018).

Figure 5 shows changes in the glial [Cl^–]_int after chemical ischemia for 2, 5, and 10 min. Unexpectedly, in CA1 astrocytes and RGL cells, the [Cl^–]_int remained constant at all three time points. In DG astrocytes, we observed a transient increase at 2 min (Figure 5B, p = 0.007) after chemical ischemia that returned to physiological levels at 5 or 10 min. Only in neocortical astrocytes, [Cl^–]_int was increased after 10 min of chemical ischemia (Figure 5D, p = 0.008).

Cl^– is the main physiological anion and a major determinant of cell swelling and regulatory volume changes. Largely unaltered [Cl^–]_int thus predicts the absence of significant volume changes during transient energy restriction. We therefore additionally measured astrocytic volumes in acute neocortical tissue slices stained with SR101 using MIPs from Z-stacks (0.6 μm increments) generated by confocal microscopy (Figures 6A,C). Chemical ischemia for up to 10 min (measured with 5 min delay) neither changed the mean width of astrocyte somata (Figure 6B, control-ischemia: p = 0.275; ischemia-recovery: p = 0.619; control-recovery: p = 0.894) nor caused strong delocalization or deformation of astrocytic extensions (Figure 6A). Notably, we observed increased soma sizes in some cells, but decreased sizes in others (Figure 6B). Since the absence of astrocytic volume changes during energy restrictions was surprising (Kimelberg, 2005), we transiently perfused slices with hypo-osmotic saline (220 mOsm/L) as control. Hypo-osmotic stress significantly increased the width of astrocyte somata from 5.3 ± 0.8 μm to 6.0 ± 1.1 μm (Figure 6D, control-hypo: p = 1.77⋅10^–5; hypo-recovery: p = 2.54⋅10^–9; control-recovery: p = 0.008), which was fully reversible (Figures 6C,D). In addition, primary and secondary astrocyte processes were strongly displaced (Figure 6C).

We conclude that glial [Cl^–]_int and volumes undergo only slight changes under transient chemical ischemia of up to 10 min.

To define the contribution of specific chloride transporters to glial [Cl^–]_int during chemical ischemia, we measured the combined effect of restricted ATP supply and specific chloride transport blockers (Figures 7, 8). To achieve lasting inhibition of chloride transporters and channels, slices were continuously perfused with blocker-containing solutions (bumetanide, R-(+)-DIOA, DL-TBOA) during and after transient chemical ischemia. This was not necessary for UCPH-101 because of its slow unblocking kinetics (Abrahamsen et al., 2013).

Chemical ischemia had no pronounced effect on [Cl^–]_int in CA1 and in DG astrocytes after pre-treatment with specific transport blockers (CA1: Figures 7A,B, 2 min: p = 0.005, Figures 8A,B; DG: Figures 7C,D, 8C,D). In DG astrocytes we observed an increase of the mean [Cl^–]_int at 10 min after ischemic conditions and a pre-treatment with R-(+)-DIOA (Figure 7D, p = 1.94⋅10^–5). In neocortical astrocytes, bumetanide alone had no effect under control conditions. After chemical ischemia, bumetanide decreased [Cl^–]_int (Figure 7G, 2 min: p = 6.53⋅10^–4, 5 min: p = 9.78⋅10^–4), consistent with increased Cl^– efflux/outward transport pathways during energy restriction. R-(+)-DIOA, in contrast, caused a dramatic increase of [Cl^–]_int under ischemic conditions (Figure 7H, 10 min: p = 0.002), but UCPH-101 (Figure 8G, 2 min: p = 0.006), but DL-TBOA did not (Figure 8H). Based on these results, we conclude that ischemia activates both NKCC1 and KCCs in neocortical astrocytes. Since these transporters mediate chloride transport in opposite directions, they partially compensate for each other, resulting in only a limited increase in the [Cl^–]_int under energy restriction. Similar, but less pronounced effects were observed in RGL cells (Figures 7E,F and Figures 8E,F, 5 min: p = 0.005).

In these experiments we approximated glial cell volumes by defining a circular ROI surrounding the glial soma in FLIM images and assuming a spherical shape for the glial cell (Supplementary Figures 2, 3). This approximation did not reveal significant changes in volume under transient energy restriction. However, cell swelling was observed in neocortical astrocytes, when energy restriction in combination with selected anion transport blockers resulted in pronounced increases of [Cl^–]_int. Under similar conditions, we observed increased cell soma sizes, i.e., in the DG region after 10 min transient energy restriction combined with R-(+)-DIOA. Blocking EAAT1 and EAAT2 anion channels, respectively, lead only to slight volume changes in both directions. Cell swelling was not observed in other brain regions, probably due to only small changes in the [Cl^–]_int.

To understand how changes in anion transporter activity shape the response to transient energy restriction, we used a recently developed quantitative model of a glutamatergic synapse to identify key determinants of synaptic failure during energy deprivation (Kalia et al., 2021). To account for the effects of R-(+)-DIOA-dependent KCC block on [Cl^–]_int, the model was expanded by inserting KCC cotransporters as an additional Cl^– transport system into astrocytes. Anion conductances generated by EAAT anion channels are represented as components of the glial leak anion conductance. Since the Cl^–/HCO₃^– exchanger, AE3, is not expressed in glial cells (Kopito et al., 1989; Hentschke et al., 2006), no HCO₃^– transporter was integrated into the model. Neocortical and in DG astrocytes represent two extremes in both resting [Cl^–]_int and changes in [Cl^–]_int following the selective inhibition of anion transporters (Figures 2–5, 7, 8). As NKCC1 is the dominant Cl^– influx and KCC the dominant Cl^– efflux pathway in our model, we manipulated resting [Cl^–]_int and changes of [Cl^–]_int in response to energy restriction by varying the rates of these two transporters. With the exception of Cl^–, Na⁺ and K⁺ leak conductances, all other parameters remained unchanged from the original model (Kalia et al., 2021).

Supplementary Figure 4 illustrates the iterative approach we used in optimizing these two parameters for the example of neocortical astrocytes. NKCC1 and KCC flux rates were varied around baseline values, which were either obtained from Kalia et al. (2021) for NKCC1 or set to PKCCa=1.3 ⋅ 10^–6 fmolmsmV for K⁺-Cl^– exchange. We modified normalized NKCC1 flux rates (P^a_NKCC) between 1 and 100% and normalized KCC flux rates between 1 and 500% of these baseline values. For each set of P^a_KCC and P^a_NKCC, changes in neocortical [Cl^–]_int were calculated under resting conditions, as well as during and after transient energy restriction.

Variation of secondary active transporters requires adjustment of the leak conductances of the transported substrates to ensure that the sum of ion fluxes remains zero under resting conditions. Higher KCC and lower NKCC1 transport rates are thus associated with lower Cl^– leak conductances. At KCC rates exceeding the starting values, as well as for reduced KCC rates at NKCC1 rates below baseline, this limiting condition results in physiologically unreasonable negative values for astrocytic Cl^– conductances (Supplementary Figure 4A). For P^a_NKCC between 0.01 and 0.1 of the starting values, tenfold or hundredfold decreased KCC rates require implausibly low Cl^– leak conductances that result in continuously increasing [Cl^–]_int after energy restriction (Supplementary Figure 4A). These predictions were in disagreement with experimental results, and these parameter values were thus discarded. The initial test thus demonstrates that P^a_NKCC must be around the starting values. The response of our model to energy restriction alone, however, did not suffice to restrain P^a_KCC. We thus tested the effects of varying P^a_KCC at P^a_NKCC fixed to the baseline value on the consequences of pharmacological block of glial anion transporters (Supplementary Figure 4B). 3.5-fold increased P^a_KCC flux rates predicted best the effect of the KCC blocker DIOA on astrocytic [Cl^–]_int in neocortical astrocytes and was thus used for further analyses.

Using this approach, we found two distinct parameter regions corresponding to [Cl^–]_int measured in cortical and DG astrocytes (Figures 9B,C). They are shown in a log–log plot of the NKCC1 and KCC flux rates in Figure 9A: cortical results were well described by higher numbers in functional KCCs (approximately 10 times) in astrocytes than in neurons; and results for DG astrocytes were well described by lower numbers in both functional NKCC1 (about 10 times lower than in the cortex) and KCCs (about 100 times lower than neuronal KCC) (Figure 9A). Within the tested parameter ranges, no bifurcations were observed in the two-parameter space. Astrocytic [Cl^–]_int smoothly changed upon parameter variation, and simulation results were consistent also for large perturbations of the parameters. The effect of DL-TBOA was simulated by blocking both neuronal and astrocyte glutamate transport and partially blocking astrocytic Cl^– leak currents (Figure 9C). In the cortex, DL-TBOA caused a sharp increase in [Cl^–]_int, followed by a plateau (at +200% of baseline). In contrast, in the DG DL-TBOA caused a slow rise to +5% of baseline at the end of the ischemic block. The EAAT anion conductance is modeled as part of a glial resting anion conductance (leak conductance). Since the exact contribution of EAAT anion channels to the total resting anion conductance is unknown, we varied the degree of resting conductance blockade by DL-TBOA, but found no discernible differences. These findings suggest that DL-TBOA mostly affects neuronal and glial glutamate transport, whereas changes in the glial resting conductance have only minor effects on [Cl^–]_int.

NKCCs and KCCs are both electroneutral transporters, so that only Na⁺ and K⁺ gradients across the membrane act as driving forces. Differences in NKCC and KCC flux rates may also be represented by variation in [Na⁺] and [K⁺], and one might thus be able to model the separate [Cl^–]_int in DG and neocortical astrocytes by modifying cation transport rates. However, this possibility is excluded by experimental data. Intracellular [Na⁺] has been compared for both types of glia, with virtually identical results (Langer and Rose, 2009; Ziemens et al., 2019), and external [K⁺] does not vary under resting conditions. Thus, hippocampal and neocortical astrocytes are not expected to differ in cation transporter expression, and differences in [Cl^–]_int are not due to separate driving forces for NKCCs and KCCs.

Moderate transient ischemia was initially simulated by blocking the Na⁺-K⁺-ATPase in both neurons and astrocytes to 50% of baseline activity (Kalia et al., 2021). In these simulations, astrocytes were subjected to energy deprivation for 5 or 10 min and then allowed to recover for 15 or 10 min, respectively (Figure 9B; gray shading). Additionally, astrocytes were treated with bumetanide (to block NKCC1), R-(+)-DIOA (to block astrocytic KCCs), or DL-TBOA (to block neuronal and astrocyte EAAT) for 20 min (light gray region), followed by energy deprivation for 10 min (dark gray block), and then another 30 min of transport block after energy restoration (Figure 9C). In response to transient ischemia for 5 min, [Cl^–]_int in DG hippocampal astrocytes changed only slightly, whereas cortical astrocytes underwent a transient increase in [Cl^–]_int that was fully reversible by restoring primary-active Na⁺-K⁺-transport after 5 min of energy restriction (Figure 9B). In response to 10 min energy restrictions, simulated [Cl^–]_int rose to a larger extent than observed in experiments, for cortical as well as for DG astrocytes. However, experimentally observed differences between the two classes of astrocytes were reproduced; cortical astrocytes reacted with larger chloride accumulation to ATP restriction than DG astrocytes (Figures 5, 9B).

In DG astrocytes, bumetanide triggered Cl^– efflux under control conditions, and additional chemical ischemia changes the effects only slightly (Figure 9C, lower panel). R-(+)-DIOA caused small increases in [Cl^–]_int after chemical ischemia in DG astrocytes. In cortical astrocytes, the model predicted a bumetanide-mediated reduction of [Cl^–]_int that was partially reversed by energy restriction. This prediction differs from experimental observations, in which bumetanide had only minor effects under control conditions (Figure 4), but decreased [Cl^–]_int in combination with chemical ischemia (Figure 7G). In simulations, R-(+)-DIOA caused massive Cl^– influx into the cortex (+300% of baseline), which was further enhanced by ischemic conditions and not reversed by energy restoration. This result is qualitatively, but not quantitatively similar to our experimental results (Figure 7H).

A possible explanation for the deviation between simulated and experimentally observed [Cl^–]_int is provided in Figure 10, in which changes in [Cl^–]_int are compared for blocking Na⁺-K⁺-ATPase to 60% baseline pumping capacity with the results upon 50% reduction from Figure 9. At higher remaining pump activity, the model predicts [Cl^–]_int in DG astrocytes almost perfectly (Figure 10B). With less pronounced block of the Na⁺-K⁺-ATPase, the model predicts fully reversible changes in [Cl^–]_int for cortical astrocytes after 10 min of chemical ischemia. In cortical astrocytes (Figure 10A), the combination of chemical ischemia for 10 min with bumetanide reduced [Cl^–]_int to 6 mM in model astrocytes, closely similar to experimental results (Figure 7G). The effects of R-(+)-DIOA was comparable at 50 and 60% percentage remaining pump activity, and the TBOA effect was fully reversible at 60% rest activity.

Pronounced differences in [Cl^–]_int changes upon small variations in Na⁺-K⁺-ATPase activity (Figure 10) are expected when simulating ion concentrations under ischemic conditions. Mathematical modeling demonstrated the existence of a tipping point of Na⁺-K⁺-ATPase pump activity that results in cell swelling and the establishment of a Gibbs–Donnan-like equilibrium state (Dijkstra et al., 2016; Kalia et al., 2021). After reaching this state, the re-establishment of the Na⁺-K⁺-ATPase does not bring the system back to its physiological state. Such a saddle-node bifurcation is characterized by large changes of the state variables, e.g., the ion concentrations, upon small variations of a model parameter close to the bifurcation point. It predicts that cells that differ only slightly in remaining ATP levels may exhibit pronounced differences in ion transport activities and in intracellular ion concentrations. Saddle-node bifurcation behavior predicts large inter-cell variability that makes mathematical modeling extremely complicated.

We conclude that our model qualitatively describes our experimental results under control conditions as well as under chemical ischemia. It shows that differences in the expression/activity of specific inward and outward chloride transporters can explain the experimentally observed region-specific differences in glial chloride homeostasis.