K+ efflux through postsynaptic NMDA receptors suppresses local astrocytic glutamate uptake

Abstract Glutamatergic transmission prompts K+ efflux through postsynaptic NMDA receptors. The ensuing hotspot of extracellular K+ elevation depolarizes presynaptic terminal, boosting glutamate release, but whether this also affects glutamate uptake in local astroglia has remained an intriguing question. Here, we find that the pharmacological blockade, or conditional knockout, of postsynaptic NMDA receptors suppresses use‐dependent increase in the amplitude and duration of the astrocytic glutamate transporter current (IGluT), whereas blocking astrocytic K+ channels prevents the duration increase only. Glutamate spot‐uncaging reveals that astrocyte depolarization, rather than extracellular K+ rises per se, is required to reduce the amplitude and duration of IGluT. Biophysical simulations confirm that local transient elevations of extracellular K+ can inhibit local glutamate uptake in fine astrocytic processes. Optical glutamate sensor imaging and a two‐pathway test relate postsynaptic K+ efflux to enhanced extrasynaptic glutamate signaling. Thus, repetitive glutamatergic transmission triggers a feedback loop in which postsynaptic K+ efflux can transiently facilitate presynaptic release while reducing local glutamate uptake.

). Patch-clamp experiments combined with biophysical simulations have suggested that, during synaptic transmission, the bulk of intra-cleft K + arrive through postsynaptic NMDARs that remain activated for 100-200 ms, rather than through short-lived K + efflux due to AMPAR activation or during presynaptic spike repolarization . This retrograde signaling also depends on postsynaptic depolarization which is required to remove the Mg 2+ block of NMDAR channels (Nowak et al., 1984). The extracellular K + elevation generated during synaptic transmission is cleared by diffusion, neuronal and astrocytic Na + /K + ATPase, astrocytic inwardrectifying K + channels (K ir ), and various other channels and transporters. K + current through astrocytic K ir channels could last for hundreds of milliseconds in response to a single synaptic stimulus, suggesting a relatively long dwell time of extracellular K + transients (Cheung et al., 2015;Lebedeva et al., 2018;Meeks & Mennerick, 2007). Because the membrane potential of astrocytes is predominantly determined by their K + conductance, increases in extracellular K + could be sufficient to depolarize local astrocytic membranes. Although the latter should, in theory, reduce voltage-dependent glutamate uptake through the main glial glutamate transporter GLT-1 (Grewer et al., 2008;Grewer & Rauen, 2005;Mennerick et al., 1999), whether this sequence of events takes place in reality remains largely unknown. Intriguingly, induction of synaptic long-term potentiation has been shown to boost extrasynaptic glutamate escape, by driving withdrawal of local astrocytic processes (Henneberger et al., 2020). However, it has remained unknown if any similar effect occurs in the course of regular excitatory activity. Because a better understanding of such phenomena should shed light on the fundamental rules of local signal integration in excitatory brain circuits, we have embarked on a multi-disciplinary study to explore them.

| Hippocampal slice preparation
Animal procedures were carried out under the oversight of the UK Home Office (as per the European Commission Directive 86/609/ EEC and the UK Animals [Scientific Procedures] Act, 1986) and by institutional guidelines. Young C57BL/six mice (3-4 weeks of age) male were anesthetized using isoflurane and decapitated (at the time of virus injection, we could not identify the animal's sex with certainty, and later we were compelled to use all injected animals, in accord with the 3Rs principles, the bulk of which were males). The brain was exposed, chilled with an ice-cold solution containing (in mM): sucrose 75, NaCl 87, KCl 2.5, CaCl 2 0.5, NaH 2 PO 4 1.25, MgCl 2 7, NaHCO 3 25, and D-glucose 25. Hippocampi from both hemispheres were isolated and placed in an agar block. Transverse slices (350 μm) were cut with a vibrating microtome (LeicaVT1200S) and left to recover for 20 min in the same solution at 34 C. Then slices were incubated at 34 C in a solution containing (in mM): NaCl 119, KCl 2.5, NaH 2 PO 4 1.25, MgSO 4 1.3, CaCl 2 2.5, NaHCO 3 25, and D-glucose 11. For experiments with intracellular blockade of K ir , the solution was supplemented with 100 μM BaCl 2 . After that, slices were transferred to the recording chamber and continuously perfused at 34 C with the same solution. All solutions were saturated with 95% O2 and 5% CO2. Osmolarity was adjusted to 298 ± 3 mOsM.

| AAV transduction
For viral gene delivery of AAV2/1 h.Synap.SF-iGluSnFR-A184V (Penn Vector Core, PA, USA), pups, male and female (P0-P1), were prepared for aseptic surgery. To ensure proper delivery, intracerebroventricular (ICV) injections were carried out after an adequate visualization of the targeted area (Kim et al., 2014), as described previously . Viral particles were injected in a volume 2.5 μL/hemisphere (totaling 5 Â 10 9 genomic copies), using a glass Hamilton microsyringe at a rate not exceeding of 0.2 μL/s, 2 mm deep, perpendicular to the skull surface, guided to a location approximately 0.25 mm lateral to the sagittal suture and 0.50-0.75 mm rostral to the neonatal coronary suture. Once delivery was completed, the microsyringe was left in place for 20-30 s before being retracted. Pups (while away from mothers) were continuously maintained in a warm environment to eliminate the risk of hypothermia in neonates. After animals received AAV injections, they were returned to the mother in their home cage. Pups were systematically kept as a group of litters. Every animal was closely monitored for signs of hypothermia following the procedure and for days thereafter to ensure that no detrimental side effects appear. For the transduction of glutamate sensors in vivo, there were three-to four-weeks to suffice.

| iGluSnFR imaging
Femtonics Femto2D-FLIM imaging system was used for two-photon imaging, integrated with two femtosecond pulse lasers MaiTai (SpectraPhysics-Newport) with independent shutter and intensity control and patch-clamp electrophysiology system (Femtonics, Budapest). Patch pipettes were pulled from borosilicate-standard wall filament glass (G150F-4; Warner Instruments, CT, USA) with 4-5 MΩ resistance. CA1 pyramidal neurons expressing iGluSnFR sensor were patch-clamped with either KMS-or NMDG-based internal solution, supplemented with the morphological tracer dye Alexa Fluor 594 (50 μM). The Alexa Fluor 594 channel was used to identify a region of interest and recorded along with the iGluSnFR signal. After at least 45 min required for the dye diffusion and equilibration in the dendritic arbor, glutamate imaging from individual spines was carried out using an adaptation of the previously described method (Henneberger et al., 2020;Jensen et al., 2017). For the fast imaging of AP-mediated glutamate transients point scans were performed over the dendritic spines and scanned at a sampling frequency of 500 Hz.

| Astrocyte simulations
Simulations were carried out using a detailed 3D biophysical model of a (statistically) reconstructed CA1 astrocyte using a NEURON-compatible model builder ASTRO (www.neuroalgebra.com and https://github.com/ F I G U R E 1 NMDAR-mediated K + efflux induces an activity-dependent increase of I GluT amplitude and decay time. (a) A schematic showing stimulating (SC stim) and recording (Reg) electrode positions for the recording of synaptically-induced currents in CA1 stratum radiatum astrocyte. (b) Current-voltage relationship of passive astrocyte. Inset: Astrocytic current (ΔI) was measured in response to voltage steps (ΔV m ). (c) Astrocytic currents induced by single and 5 Â 50 Hz stimulation of Schaffer collaterals. DL-TBOA application was used to isolate I K (blue), which was then subtracted from combined current (I K + I GluT , gray) to obtain I GluT (green). (d) Sample traces of I GluT in response to single and 5 Â 50 Hz stimulation in control (gray), in the presence of D-APV (green), in the presence of D-APV (orange) and recorded from CA1-GluN1-KO mice (purple). (e) Amplitudes of I GluT peaks during 5 Â 50 Hz stimulation normalized to the amplitude of I GluT in response to a single stimulus. Activitydependent facilitation in control (gray) was reduced by the application of D-APV (green) or D-APV + NBQX (orange) or in CA1-GluN1-KO mice (purple). (f) τ decay of I GluT in response to single and 5 Â 50 Hz stimulation. There was no significant difference in τ decay of I GluT in response to a single stimulus between control (gray), D-APV (green), D-APV + NBQX (orange) or in CA1-GluN1-KO mice (purple). τ decay of I GluT in response 5 Â 50 Hz stimulation was significantly larger in control than in D-APV, D-APV + NBQX, or in CA1-GluN1-KO mice. The data are presented as mean ± SEM. ns p > .05, **p < .01 and ***p < .001, two-sample t test LeonidSavtchenko/Astro), as outlined in detail earlier (Savtchenko et al., 2018). In brief, 'average' astrocyte morphology was obtained from a sample of protoplasmic astrocyte in hippocampal stratum radiatum ($4-week-old rats) using the systematic procedures as follows. First, generating soma and primary (optically discernible) branches to match sample-average numbers and dimensions; second, adding nanoscopic branches that have statistically generated dimensions matching experimental EM data (size distributions); third, adjusting biophysical properties of nano-branches to match biophysics of reconstructed 3D EM processes; fourth, adjusting the numbers and density of nano-branches to match the statistics of tissue volume fraction and surface-tovolume ratios obtained empirically from 3D EM data; fifth, distributing biophysical membrane mechanisms across the model cell membrane, including K ir 4.1 channels and GLT-1 transporters in accord with the experimental recordings. A full description of the model and its implementation, for either desktop-or cloud-computing, are available from www.neuralgebra.com and links therein.
The model was populated with K ir 4.1 channel, with the kinetics in accord with (Jérémie Sibille et al., 2015), and membrane unit conductance of GKir =0.1 mS cm À2 .
The kinetic is described by the equation: where V A1 = À14.83 mV, V A2 = À105.82 mV, V A3 = 19.23 mV, NK = 0.81, E k is the Nernst astrocyte K + potential, V is the astrocyte membrane potential, [K + ] 0 is the extracellular K + concentration and V A1 (an equilibrium parameter, which sets I Kir current to 0 at À80 mV), NK,V A2 and V A3 are constant parameters calibrated by the I-V curve.
The leak passive current I pas = G pas (V À E pas ) was added to stabilize the astrocyte membrane potential at E pas = À85 mV, G pas = 0.001 mS/cm 2 .
The kinetics of glutamate transporters was determined by a simplified scheme of 6 states: And function u(V,P) = exp(P V/53.4), where V is astrocyte membrane voltage, and P is a charge translation between states C i -> C j .
The glutamate transporter current (Z. Zhang et al., 2007) was calculated according to the following equation: where e is charge on an electron charge 1.6 Â 10 À19 (coulombs) and den is a density of transporters Initial ion concentrations: Diffusion of intracellular potassium is described by the equation built into the Neuron: To fit the experimental observation, we fit the kinetic scheme and

| Drugs and chemicals
All drugs were made from stock solutions kept frozen at À20 C in 100-200 ml 1000 Â aliquots. DL-2-amino-5-phosphonovaleric acid  test; Figure 1f). The τ decay of I GluT recorded in response to a single stimulus in CA1-GluN1-KO mice was also not significantly different from τ decay in control mice (KO mice: 3.82 ± 0.69 ms, n = 4; p = .07 for the difference with control mice; two-sample t-test; Figure 1f).
The values of τ decay became significantly larger in response to 5 Â 50 Hz stimulation (23.65 ± 3.22 ms, n = 12; p < .001 for the difference with single stimulus; paired-sample t-test; Figure 1f). This increase was significantly reduced by bath application of D-APV (τ decay = 12.71 ± 1.91 ms, n = 6, p = .01 for the difference with control, two-sample t-test; Figure 1f). Adding NBQX to D-APV produced a further reduction in τ decay , albeit statistically insignificant (τ decay = 7.91 ± 1.35 ms, n = 6, p < .001 compared to control, p = 0.13 compared to D-APV, two-sample t-test; Figure 1e). These results are consistent with an earlier suggestion that synaptically released glutamate does not overwhelm glutamate transporters upon the blockade of AMPARs and NMDARs (Diamond & Jahr, 2000).
τ decay in CA1-GluN1-KO mice was not significantly different from τ decay in D-APV (KO mice: 9.42 ± 2.82 ms, n = 5; p = 0.21 for the difference with D-APV, two-sample t-test; Figure 1f). These findings suggest that postsynaptic NMDARs are required both for the activitydependent facilitation of glutamate release and for the activitydependent reduction of glutamate uptake.

| Activity-dependent increase of I GluT decay time does not depend on afferent recruitment
NMDARs activation requires voltage-dependent unblocking of the receptor channel (Nowak et al., 1984). Therefore, greater postsynaptic depolarization should, in theory, recruit more NMDARs and thus produce larger K + efflux. To test this hypothesis, we recorded I GluT s at two different stimulation strengths ( Figure 2a). Initially, the stimulation strength was adjusted to achieve I GluT $ 10 pA (weak stimulus), then the strength was doubled (strong stimulus). Strong stimulation increased the I GluT to 17.07 ± 1.8 pA (n = 8; p = .03, paired sample ttest; Figure 2a). Surprisingly, we did not observe a significant enhancement in the activity-dependent facilitation of I GluT upon stronger stimulation (p strengh = 0.41; p stimulus < .001; interaction: F [4, 60] = 0.01, p = 0.99; n = 7; two-way RM ANOVA; Figure 2b). Nor did we observe a significant difference in τ decay between weak and strong stimulation groups, either with one or with five stimuli tests (single stimulus: 5.57 ± 1.3 ms for weak stimulation, 5.89 ± 1.34 ms for strong stimulation, p = 0.87, paired sample t-test; 5 stims: 27.24 ± 5.29 ms for weak stimulation, 25.29 ± 3.77 ms for strong stimulation, p = 0.78, paired sample t-test; n = 6; Figure 2c). Thus, although stronger stimulation releases more glutamate per tissue volume, hence generates larger uptake currents, the current kinetics appears unaffected. This is likely because glutamate transporter binding and uptake occur only inside the microscopic vicinity of individual synapses, within 5-10 ms postrelease. Unless such 'uptake hotspots' substantially overlap in the tissue volume, engaging additional synapses would not be expected to affect glutamate uptake kinetics.
To provide further controls, we have carried out experiments in which astrocytic K + currents (I K ) were compared directly under strong and weak stimuli, in conditions of single and burst stimuli. As expected, I K increased with the increased stimulus strength or number. Importantly, in response to a stimulus burst, the use-dependent increase in I K was similar under either weak or strong stimuli ( Figure S1), thus arguing that it scales linearly independently of the number of synapses/afferent fibers involved.

| Activity-dependent increase of I GluT decay time is abolished by the blockade of astrocytic K ir
Next, we asked if the activity-dependent increase in τ decay involves the K ir -dependent depolarization of astrocytic leaflets. Earlier work has shown that dialising astrocytes with 100 μM BaCl 2 blocks K ir channels responsible for I K in astrocytes (Afzalov et al., 2013).
Although the genetic deletion or downregulation of the main astrocytic K ir channel, Kir4.1, could provide useful insights in their functional roles (Sibille et al., 2014), it could also cause sustained membrane depolarization (Djukic et al., 2007), which would in turn suppress the other powerful astrocytic K + clearance mechanism, the Na/K-ATPase (Larsen et al., 2014). This pump is also sensitive to internal astrocytic Na + which homeostasis might be affected in the Kir4.1-downregulated cells (Kirischuk et al., 2012). We therefore opted for pharmacological manipulation that would permit real-time comparison between control and affected cells.

| Astrocyte depolarization but not extracellular K + affects I GluT
Our findings suggest that K + efflux through postsynaptic NMDARs decreases glutamate uptake. The two candidate underlying mechanisms are (1) depolarization of the astrocytic membrane and (2) a decrease in the astrocytic transmembrane K + gradient. Synaptically released glutamate rapidly binds to astrocytic transporters (Diamond & Jahr, 1997). From the bound state, glutamate À can be translocated into the astrocyte cytoplasm, along with 3 Na + and 1 H + , in exchange for 1 K + (Grewer & Rauen, 2005). Therefore, the glutamate translocation step is both K + À and voltage-dependent (Grewer et al., 2008;Mennerick et al., 1999). To assess the relative contributions of K + and voltage, we recorded I GluT induced by the single-pulse spot uncaging of glutamate that generates a typical single-synapse (unitary) EPSC (uI GluT ), as described previously (Henneberger et al., 2020), near the astrocyte soma, to minimize voltage-clamp error (Figure 3a), under the two sets of conditions as follows. Firstly, we obtained the relationship between the extracellular K + concentration and astrocyte membrane potential (Figure 3b). An increase in extracellular K + produced similar astrocyte depolarization as previously reported (Ge & Duan, 2007). Second, we mimicked the K + -induced depolarization by holding the cell in voltage-clamp mode at the membrane potentials seen at high K + . The membrane depolarization alone reduced uI GluT to the same degree as did the corresponding K + concentration uI GluT (p K+ = 0.18; p depolarization < .001; interaction: F F I G U R E 3 Depolarization of astrocyte rather than a decrease in K + gradient suppresses glutamate uptake. [2, 36] = 0.5, p = 0.59; n = 7; two-way RM ANOVA; Figure 3c). These results suggest that depolarization alone can attenuate glutamate uptake, thus occluding the effects of K + elevations. We also observed a depolarization-dependent increase in τ decay of uI GluT (F[2, 21] = 6.2, p = .007, n = 8, one-way RM ANOVA; Figure 3d). Correspondingly, when different extracellular K + concentrations were applied under a constant membrane potential of À85 mV, no significant change in the amplitude of uI GluT (F[2, 13] = 2.2, p = 0.14, n = 4; 8; 4, one-way RM ANOVA) or in τ decay (F[2, 19] = 7.6, p = .003, n = 11; 7; 4 for each condition one-way RM ANOVA) was observed (Figure 3e, f). These findings suggest that physiologically relevant elevations in extracellular K + affect I GluT through astrocyte depolarization rather than by reducing the driving force for K + .

| A biophysical underpinning of the relationship between extracellular K + and glutamate uptake
To understand through which biophysical mechanism an extracellular K + rise can affect the kinetics of astrocytic glutamate uptake, we employed a realistic, multi-compartmental 3D model of a stratum radiatum astrocyte, which was developed and validated earlier using the NEURON-compatible model builder ASTRO (Savtchenko et al., 2018). The model features known membrane astrocytic mechanisms, including GLT-1 transporters and K ir 4.1 K + channels, distributed in the model membrane to match multiple experimental observations (Materials and Methods).
First, we employed the model to simulate extracellular K + elevation (from 2.5 to 5 mM over a 20 μm spherical area, 1 s duration) within the 3D astrocyte territory (Figure 4a). This generated local K + influx through K ir 4.1 channels, triggering diffuse equilibration of intracellular K + across the tortuous cell lumen (Figure 4b). The evolving dynamics of extra-and intracellular K + was paralleled by local membrane depolarization (Figure 4c). Aiming to mimic synchronous multisynaptic glutamate release, we also simulated a brief extracellular glutamate rise (0.1 mM for 1 ms, at 0.9 s postonset) within a 3 μm spherical area (Figure 4d) placed inside the area of extracellular K + elevation. To understand the effect of extracellular K + rises on glutamate uptake, we, therefore, ran the glutamate release test in two conditions, one at the baseline extracellular K + and one during K + elevation, as above. In these tests, cloud-computing modeling could However, during repetitive stimuli, slowing down or reducing I GluT would reflect a greater fraction of transporters bound by glutamate (before the transmembrane transfer step) on the astrocytic surface. As the fraction of free transporters on the astrocytic surface decreases, the probability of glutamate molecules traveling further from the release sites increases. In this case, I GluT would reflect the dynamics of glutamate escape and removal to a much greater degree. When the glutamate translocation rate is reduced by astrocytic depolarization, more transporters can buffer extracellular glutamate (Diamond & Jahr, 1997). Such increased glutamate buffering (binding-unbinding) by transporter molecules could increase the dwell time of glutamate near the synaptic cleft (Lehre & Rusakov, 2002;Zheng et al., 2008).
Therefore, we attempted to assess the dynamics of extracellular glutamate concentration using the tests as follows. First, we recorded excitatory postsynaptic potentials (EPSPs) in CA1 pyramidal neurons in response to a single stimulus and to 5 Â 50 Hz stimulation, in the F I G U R E 5 Synaptic K + efflux increases glutamate dwell time in the synaptic cleft and enhances its spillover. (a) Sample traces of EPSPs recorded in CA1 pyramidal neurons filled with either KMS-or NMDG-based internal solutions in response to single-pulse or 5 Â 50 Hz stimulation, as indicated. (b) and (c) The summary plot of τ decay of EPSPs in response to a single-pulse (b) and 5 Â 50 Hz (C) stimulation. Replacement of intracellular K + (green) for NMDG (orange) significantly increased τ decay in both cases. (d) The summary plot showing the ratio of τ decay of EPSP in response to 5 Â 50 Hz stimulation to τ decay of EPSP in response to single-pulse stimulation. The activity-dependent prolongation of EPSP was observed when the cell was filled with KMS-but not with an NMDG-based solution. (e) CA1 pyramidal neuron expressing iGluSnFR loaded with fluorescent dye Alexa-594 via patch pipette, shown in two emission channels, as indicated. (f) Left, zoomed-in area boxed in E; arrows, analyzed dendritic spines. Right, glutamate traces recorded at the corresponding spines. Gray bar -stimulation. (g) Averaged traces and a summary plot showing the τ decay of glutamate transients recorded from 53 dendritic spines (five cells) and 20 dendritic spines (three cells) with KMS and NMDG-based solutions, respectively. The substitution of K + with NMDG shortened the glutamate dwell time around the synapses. The data are presented as mean ± SEM. ns p > .05, *p < .05, **p < .01, and ***p < .001, two-sample (b,c,g) and one-sample (d) t test presence of 100 μM picrotoxin, a GABA A receptor antagonist ( Figure 5a). The cut was made between CA1 and CA3 regions to prevent epileptiform activity. One of the two intracellular solutions was used: potassium methanesulfonate (KMS)-based or N-methyl-Dglucamine (NMDG)-based. Having NMDG in the postsynaptic cell reduced the use-dependent facilitation of EPSPs (Figure 5a), lending support to the earlier finding that K + efflux boosts presynaptic release efficacy . In contrast to NMDG-based solution, KMS-based solution allowed K + efflux through postsynaptic receptors, which curtailed EPSP. Therefore, the decay time constant (τ decay ) of EPSP was smaller in KMS than in NMDG recordings (KMS: 91.6 ± 20.63 ms, n = 7; NMDG: 210.8 ± 24.5 ms, n = 6; p = .003, two-sample t-test; Figure 5b). The τ decay of burst EPSP in response to 5 Â 50 Hz stimulation was increased in KMS but not in NMDG (KMS: τ decay = 136.52 ± 17.02 ms, τ burst/single = 1.4 ± 0.14, n = 6, p = .03, single-sample t-test for the ratio; NMDG: τ decay = 221.9 ± 23.5, τ burst/ single = 1.06 ± 0.05, n = 6, p = 0.34, single-sample t-test for the ratio; p < .001, two-sample t-test between ratios; Figure 5c, d).
This finding suggests that glutamate spillover is regulated by K + efflux through postsynaptic receptors. Alternatively, this result may reflect the voltage and activity-dependent regulation of K + current that curtails EPSPs (Ichinose et al., 2003). Therefore, we next attempted to directly evaluate extracellular glutamate transient with genetically encoded glutamate sensor iGluSnFR (Figure 5e). CA1 pyramidal neurons expressing the sensor (methods as described earlier [Jensen et al., 2019;Jensen et al., 2017]) were loaded through the patch pipette with either KMS-or NMDG-based intracellular solution. We documented synaptically evoked extracellular glutamate transients using 500 Hz line-scans placed across visually identified dendritic spines in the second-order apical dendrite branches (Figure 5f). Glutamate responses to burst stimulation (5 Â 50 Hz) were thus recorded. The value of τ decay for glutamate transients was significantly larger if the cell was loaded with the KMS-based intracellular solution compared to the NMDG-based solution (KMS: 31.56 ± 2.04 ms, n = 53 spines from five cells; NMDG: 21.8 ± 1.65 ms, n = 20 spines from 3 cells; p < .001, two-sample t test; Figure 5g). This finding supports the suggestion that activity-dependent prolongation of EPSPs is due to glutamate spillover boosted by K + efflux from the postsynaptic terminal. ) and unconditioned EPSP in response to stim 1 only. Right, a ratio of conditioned EPSP (stim 1 after stim 2) versus nonconditioned EPSP (stim 1 only) in pathway 1 (path 1) and a ratio of conditioned EPSP (stim 2 after stim 1) versus nonconditioned EPSP (stim 2 only) in pathway 2 (path 2). (c) Examples of two pathway recordings with KMS-(left) and NMDG-based (right) internal solutions. Gray bar -application of MK-801, activity-dependent channel blocker of NMDARs. Replacement of intracellular K + for NMDG reduced the blockade of the silent pathway during stimulation of the active pathway. This points to reduced glutamate spillover. (d) The summary plot showing EPSP reduction in the silent pathway. The EPSPs were normalized to their baseline amplitude. The reduction was significantly larger when the postsynaptic cell was filled with K + . The data are presented as mean ± SEM. ns p > .05, *p < .05, twosample t test crosstalk, we performed a modified two-pathway experiment (Henneberger et al., 2020;Scimemi et al., 2004). Briefly, NMDARsmediated EPSPs in CA1 pyramidal neurons were pharmacologically isolated with NBQX and picrotoxin, AMPAR and GABA A receptor blockers, and recorded during depolarizing voltage steps to À40 mV.

| Synaptic K + efflux boosts intersynaptic crosstalk
Two bipolar electrodes were placed in CA1 stratum radiatum on the opposite sides of the slice to recruit independent afferent pathways of Schaffer collaterals (Figure 6a). The lack of cross-facilitation of EPSPs in response to paired stimulation (first-second and second-first pathway) confirmed pathways independence (Scimemi et al., 2004) (

| DISCUSSION
In the course of excitatory synaptic transmission, astrocytes clear released glutamate and excess of K + from the extracellular space (Verkhratsky & Nedergaard, 2018). We observed an activitydependent increase in the amplitude and τ decay of synapticallyinduced I GluT in hippocampal astrocytes. The increase of the amplitude of I GluT is consistent with the reported facilitation of release probability mediated by K + accumulation in the synaptic cleft (Contini et al., 2017;Shih et al., 2013). Indeed, this increase was suppressed by the blockade or genetic deletion of the postsynaptic of NMDARs, a major source of K + efflux during synaptic transmission. Similarly, the increase of τ decay was abolished in these tests. The result is consistent with the early report suggesting that glutamate transporters are not overwhelmed during HFS upon blockade of ionotropic postsynaptic receptors (Diamond & Jahr, 2000). Thus, glutamate transporters can effectively clear glutamate in hippocampal CA1 unless local glutamate translocation is partly suppressed by K + efflux through postsynaptic NMDARs. Depolarization of perisynaptic astrocytic leaflets or a decrease of the transmembrane K + gradient, or both, can potentially underpin reduced glutamate uptake (Grewer et al., 2008). However, we found that K + elevation alone does not affect I GluT , while astrocyte depolarization increases its τ decay . How much K + accumulation during synaptic transmission can depolarize leaflets cannot be directly measured because these processes are beyond optical resolution for light-microscopy voltage imaging, nor can it be accessed with electrodebased techniques. Previous simulation studies suggest that K + can rise up to 5 mM in the synaptic cleft during a single EPSC . However, the K + concentration drops rapidly outside of synaptic cleft, and its effect on the astrocyte may be strongly attenu-  . Hence, a significant contribution of Ca 2+ -dependent K + channels to K + efflux during synaptic transmission is unlikely. Another possibility is the activation of voltage-gated K + channels during EPSPs. However, blockade of NMDARs reduces field EPSPs to a much lesser extent than it reduces I K . Thus, no major contribution of voltage-gated K + channels to K + efflux during synaptic transmission has been observed.
Another question is to what extent K + released at an individual synapse affects glutamate uptake in perisynaptic astrocytic leaflets.
Biophysical models predict that K + concentration rapidly drops with distance from the cleft, with no effect on astrocytic processes at nearest-neighbor synapses . The low input resistance combined with a large membrane area severely limits the spread of current-triggered membrane depolarization in astrocytes. Therefore, increasing the number of activated synapses should not affect the kinetics of glutamate uptake unless their K + hotspots become overlapped in space. The latter scenario could occur during synchronous activation of multiple neighboring synapses or during epileptic bursts.
Such events could produce wide-spread elevations of extracellular K + affecting large astrocyte territories.
In summary, we conclude that glutamate spillover is prevented by efficient glutamate uptake unless astroglial transporters are downregulated or withdrawn from the immediate synaptic environment.
Recent reports suggest that a decrease in glutamate uptake can shift the sign of synaptic plasticity, reducing long-term potentiation (LTP) and promote long-term depression (LTD) (Valtcheva & Venance, 2019). Rate-based LTP induction in one set of synapses requires NMDARs activation and thus should lead to K + hotspots in the perisynaptic space. This K + should, in turn, depolarize astrocytic leaflets and downregulate glutamate uptake. However, leaflets within brain active milieu are 'shared' between neighboring synapses (Gavrilov et al., 2018;A. Semyanov & Verkhratsky, 2021). We thus speculate that LTP in one set of synapses could suppress LTP and facilitate LTD in their neighbors if they are activated immediately after while astrocyte is depolarized.
LTP induction per se could change the synapse's ability to release K + into the extrasynaptic space (Ge & Duan, 2007). The number of synaptic AMPARs is thought to increase during LTP, and, although they should not contribute significantly to K + efflux due to their fast inactivation, they can facilitate activation of NMDARs by removing their voltage-dependent Mg 2+ block . Therefore, LTP not only increases the quantal efficiency of the synapse but promotes K + -dependent facilitation of glutamate release and spillover at this synapse, potentially complementing the effect of the (possibly transient) perisynaptic astrocytic leaflet withdrawal after LTP induction (Henneberger et al., 2020). K + -dependent facilitation of presynaptic glutamate release after LTP could be also linked to the putative perisynaptic mechanism of LTP (Kullmann, 2012).
Our observations emphasize the physiological importance of changes in ionic concentrations in the synaptic cleft and perisynaptic space. Since the volumes of these spaces are tiny, the concentration changes could be significant. Accumulation of K + in the synaptic cleft is paralleled by local Ca 2+ depletion (Rusakov & Fine, 2003), which is also sensed by astrocytes (Torres et al., 2012) and might affect release efficacy in the opposite direction to that of excess K + . How the astrocyte mediated K + buffering affects the time-course of perisynaptic K + elevation and how far K + can diffuse in the extracellular space remains to be established.