Mitochondrial Glutamine Metabolism Drives Epileptogenesis in Primary Hippocampal Neurons

Abstract

All available antiseizure medications aim at symptomatic control of epilepsy, but there is no strategy to stop the development of the disease. The main reason is the lack of understanding of the epileptogenic mechanisms. Closing this knowledge gap is an essential prerequisite for developing disease-modifying therapies that can prevent the onset of epilepsy. Using primary cocultures of hippocampal neurons and glial cells derived from rat pups of either sex, we show that epileptiform paroxysmal depolarization shifts (PDS) induce neuronal glucose hypometabolism which is compensated for by increased glutaminolysis. Glutaminolysis not only provides sufficient ATP to support electrical activity but also leads to decreased vesicular glutamate release, thereby promoting neuronal hypersynchrony. Moreover, prolonged promotion of PDS increased neuronal arborization and synaptic density, which in combination with spontaneous recovery of neuronal glucose metabolism led to seizure-like discharge activity. Since inhibition of glutaminolysis did not prevent the PDS-induced morphogenesis but eliminated seizure-like activity, we propose that glutaminolysis is a causative process linking neuronal metabolism with electrical activity thereby driving epileptogenesis.

Significance Statement

The available pharmacotherapy for epilepsy provides symptomatic control of seizures by interfering with ictogenesis. However, understanding the preceding epileptogenic processes would offer an opportunity to intervene in the development of the disease. The electrical activity and glucose metabolism of the brain regions corresponding to the epileptic foci are disturbed long before the first seizures occur. The significance of the altered neuronal activity and metabolism is not well understood. We show that abnormal neuronal electrical activity called paroxysmal depolarization shifts increase neuronal arborization and lead to metabolic shifts making neurons transiently rely on glutamine. Interplay of these processes induces glucose hypometabolism and hypersynchronization and ultimately leads to seizure-like discharge activity, thus replicating several key features of epilepsy.

Introduction

Epilepsies are brain diseases resulting in the occurrence of seizures, and the latter are characterized by signs and/or symptoms caused by abnormal excessive or synchronous neuronal activity (Fisher et al., 2014). Mechanisms behind seizure onset (ictogenesis) are poorly understood but appear to comprise fast electrical and chemical events. By targeting ictogenesis, antiseizure medications (ASMs) provide symptomatic seizure control in a majority of patients but do not cure the disease as seizures often reappear after drug withdrawal (Devinsky et al., 2018; Löscher and Klein, 2021). Epileptogenesis, on the other hand, comprises all processes that lead within days, weeks, or even months to a chronic condition (i.e., epilepsy), in which repeated ictogenesis becomes more likely (Weaver, 2003). Despite promising reports from animal models of epilepsy, to date there is no clinically proven antiepileptogenic drug available (Löscher, 2024). Hence, there is a need for drugs interfering with the pathogenetic processes underlying epilepsies. However, as most seizures occur without identifiable temporal relation to underlying causal preconditions, epileptogenesis is difficult to assess. In this respect, acquired epilepsies, specifically those that develop after brain injuries, offer opportunities to unveil underlying mechanisms (Lucke-Wold et al., 2015). Spontaneous seizures develop within months to years after a brain insult (Löscher et al., 2010), and this period of symptomatic quiescence, known as latent period, can be employed to interfere with the processes that render the brain prone to generate epileptic seizures.

The latent period of epileptogenesis is characterized by abnormal synchronized electrical activity in the brain (Staley and Dudek, 2006) which appears in EEG recordings as interictal spikes (IIS). Initially, IIS have been observed in epilepsy patients during symptom-free intervals between seizures (hence the name “interictal”; Staley et al., 2011). In animal models of epilepsy, however, they were found to precede the first spontaneous seizure (Dyhrfjeld-Johnsen et al., 2010; Staley et al., 2011), and in humans who acquired epilepsy as a result of war-related brain insults, IIS also preceded the first seizures (Roseman and Woodhall, 1946; Staley et al., 2005). At the cellular level, IIS manifest as synchronous neuronal discharges known as paroxysmal depolarization shifts (PDS; Matsumoto and Marsan, 1964; Hotka and Kubista, 2019). Whether PDS may contribute to the processes of epileptogenesis, which is accompanied by electrophysiological, morphological, and metabolic changes (Buckmaster et al., 2002; Kovac et al., 2017), has remained unknown.

Although being a hallmark of the same disease, PDS are functionally distinct from seizures: underlying ionic conductances are distinct (Kubista et al., 2019), and antiepileptic drugs suppress seizures, but not IIS/PDS (Dyhrfjeld-Johnsen et al., 2010). In addition, IIS and seizures differ in their metabolic background: while glucose hypometabolism is observed during preictal/interictal periods in both animal models (Guo et al., 2009) and human patients (Vielhaber et al., 2003), the rate of glucose and oxygen utilization increases during seizures (Witte et al., 1994; Federico et al., 2005; Yang et al., 2013; Bazzigaluppi et al., 2017). Although detection of hypometabolism by fluorodeoxyglucose-PET is used to localize epileptic foci (Burneo et al., 2015; Zhu et al., 2017), little is known about the underlying mechanism.

To test for a direct link between PDS and glucose hypometabolism and for a role of both in epileptogenesis, we established an in vitro model based on dissociated hippocampi in primary cell culture; therein, PDS can be evoked by pharmacological means. Exposure of the cultures to PDS-promoting conditions led to a metabolic switch from glucose utilization to neuronal glutaminolysis. The switch from glycolysis toward glutaminolysis decreased glutamatergic transmission which was sufficient to synchronize neuronal activity. At the same time, neurons increased their dendritic arborization and formed new excitatory synapses. After 3 d in PDS-promoting conditions, a subsequent, spontaneous switch back from glutaminolysis to glucose oxidation occurred. This led to restoration of glutamatergic transmission and to a conversion of synchronized neuronal activity into seizure-like events. The results reveal IIS/PDS and neuronal glutaminolysis as key processes driving epileptogenesis which may represent potential targets for future antiepileptogenic drugs.

Material and Methods

Primary cell culture of hippocampal neurons

Pregnant Sprague Dawley rats were obtained from Charles River Laboratories. Neonatal animals of either sex were killed by decapitation in full accordance with all rules of the Austrian animal protection law (for details, see Hotka et al., 2020). Brains were removed to dissect the hippocampi in ice-cold buffer. Primary cocultures of hippocampal neurons and glial cells were prepared after enzymatic digestion of the tissue with papain and mechanical dissociation with Pasteur pipettes (trituration) as described previously (Hotka et al., 2020). Neurons were cultured for at least 21 d at 37°C and 5% CO₂ in Dulbecco's modified Eagle's medium—high glucose (20 mM glucose; D5796, Sigma-Aldrich) supplemented with 10% γ-irradiated fetal bovine serum (S 0415, Biochrom). For the subset of experiments shown in Figures 1G and 2, neurons were cultured for 21 d at 37°C in a Neurobasal A medium (Invitrogen) supplemented by B27 (Invitrogen) and GlutaMAX (Invitrogen).

Figure 1.

Monitoring of the neuronal metabolic response to PDS firing. A, Sequence of events leading to the development of acquired epilepsy indicated together with the characteristic electrical activity. B, Example traces of neuronal firing patterns induced by Bic + BayK (top panel) and by Bic + Isra (bottom panel). Individual PDS events induced by Bic + BayK are depicted in a detailed section on the right side in comparison with a typical Bic + Isra-induced discharge pattern (waves of discharges with bottom amplitudes and shortened duration). C, Acute response of neuronal cytosolic ATP/ADP ratio to application of PDS-inducing solution (Bic + BayK, indicated by horizontal bar) in the presence of glucose (20 mM, blue trace) and with coapplication of glucose and pyruvate (1 mM, red trace), lactate (1 mM, orange trace), and glutamine (100 µM, black trace). The baseline ATP/ADP ratio before drug application, 10.57 ± 2.94 (mean ± SD; n = 14 neurons). D, Statistical comparison of cytosolic ATP/ADP ratios obtained from the end of each experiment illustrated in panel C. n = 9–26 neurons. E, Effect of local superfusion of neurons with PDS-inducing solution (perfusion on with Bic + BayK) on the neuronal cytosolic ATP/ADP ratio of cells bathed in a solution containing Bic + BayK (perfusion off in Bic + BayK). n = 5 neurons. F,G, Statistical comparison of the effect of glutamine supplementation on the cytosolic ATP/ADP ratio in PDS-firing neurons cultured in the presence (F, n = 22 neurons) or absence (G, n = 55 neurons) of glia. H, Effect of acute application of inhibitors of glutaminolysis (200 µM SP, 50 µM R162) on the cytosolic ATP/ADP ratio. All inhibitors were coapplied to neurons via local superfusion in the continuous presence of Bic + BayK+100 µM glutamine. Neurons were in the presence of Bic + BayK + glutamine for 20 min prior to drug application. n = 5–11 neurons. I, Effect of acute application of aminoaspartate aminotransferase inhibitor AOA (1 mM) on cytosolic ATP/ADP ratio in control neurons in external buffer (black trace) and in neurons firing PDS in Bic + BayK buffer (orange trace). Neurons were in the presence of Bic + BayK + glutamine for 20 min prior to drug application. n = 11–20 neurons. J,M, Acute response and statistical evaluation of neuronal cytosolic ATP/ADP ratio to application of Bic + Isra (J,K) or 0-Mg²⁺ (L,M) calculated from the end of each experiment illustrated in panel J or L, in the presence or absence of exogenous glutamine. n = 6–12. ***p < 0.001 and ****p < 0.001.

Figure 2.

Effect of PDS firing on neuronal glucose and glutamine utilization. A, Experimental outline. Neurons cultured in the absence of glia were exposed to Bic + Isra or Bic + BayK, in a solution containing 2 mM glucose and 100 µM glutamine for 30 min and for 2 h in the presence of their respective isotopically labeled analogs [¹³C₅ glutamine (100 µM) and U¹³C₆ glucose (2 mM)]. B, Molar percent enrichment (M.P.E.) of U¹³C₆ glucose in downstream glycolytic metabolites (1-f6bp, fructose 1,6-bisphosphate; g6p, glucose 6-phosphate; pg, phosphoglycerate; pep, phosphoenolpyruvate; pyruvate; lactate) in neurons exposed to Bic + Isra and Bic + BayK for 2 h. n = 4. C,D, Monitoring of the TCA cycle activity by glucose-derived labeled carbons incorporated into malate (C) and citrate (D). n = 4. E, Incorporation of the U¹³C₆ glucose label into glutamate in Bic + Isra- and Bic + BayK-treated neurons. n = 4. F,G, Monitoring of the TCA cycle activity by glutamine-derived labeled carbons incorporated into malate (F) and citrate (G). n = 9. H, Incorporation of the ¹³C₅ glutamine label into glutamate in Bic + Isra- and Bic + BayK-treated neurons. n = 9. I, Incorporation of the ¹³C₅ glutamine label into the total neuronal glutamine pool in Bic + Isra- and Bic + BayK-treated neurons. n = 9; *p < 0.05, **p < 0.01, and ***p < 0.001. The total levels of individual metabolites are shown in Extended Data Figure 2-1.

Figure 2-1

Total metabolite levels and example electrical activity from experiments on figure 2. (A-D) Total metabolite levels expressed as total ion counts corresponding to the experiments using 13C glucose shown in figure 2 (B-E). (E-H) Total metabolite levels expressed as total ion counts corresponding to the experiments using 13C glutamine shown in figure 2 (F-I). (I) Example traces from neurons co-cultured with glial cells in Dulbecco's modified Eagle's medium (DMEM) of PDS induced by acute application of Bic + BayK in the presence of high (20  mM, HG, left trace) or low (2  mM, LG, middle trace) glucose. Similar PDS were also elicited in the neurons by Bic + BayK in neurons in absence of glial cells (Neurobasal A medium, right trace). n = 4-9, **p < 0.01. Download Figure 2-1, TIF file ^{(1.3MB, tif)} .

Drugs and chemicals

Final concentrations used in experiments are indicated in parentheses preceding the compounds, ordering numbers following them: (0.03%) dimethyl sulfoxide (DMSO; D2650), (100 µM) glutamine (49419), (3 µM) isradipine (Isra; I6658), (3 µM) Bay K8644 (BayK; B112), (50 µM) R162 (5380980001), (200 µM) succinyl phosphonate (SP; HY-12688), (10 µM) bicuculline (Bic; 2503), (10 µM) carbonylcyanid-4-(trifluormethoxy)phenylhydrazon (FCCP; C2920), (10 µM) 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX; C127), (2 mM) sodium pyruvate (P8574), (10 mM) sodium l-lactate (71718), (0.5 µM) tetrodotoxin (TTX; A1076), (200 µM) valproic acid, (VPA; P4543-10G), (75 µM) vigabatrin (Vig; 5059890001), and (1 µM) D(−)-2-amino-5-phosphonopentanoic acid (APV; A8054-1MG). Bulk chemicals were purchased from Sigma-Aldrich. Isra and BayK were dissolved from 10 mM stock solutions in DMSO to a final concentration of 3 µM in aqueous buffer. Hence, 0.03% DMSO was also added to the respective control solutions. In experiments with R162, additional 0.1% DMSO was added to the control buffer.

Transfections

Transfections were performed using Lipofectamine 2000 reagent (Invitrogen, Thermo Fisher Scientific) according to the manufacturer's instructions with some modifications. In brief, neuronal cultures older than 21 d in vitro (DIV) were incubated in 500 µl of antibiotic-free medium containing 3 µl of Lipofectamine and 1 µg of plasmid DNA for 3 h. Then, the transfection medium was replaced with the original medium, and cultures were kept for 3 more days in the incubator. Experiments were performed on Day 3 after transfection.

Viral transductions

All viral transductions were performed using adeno-associated viruses (AAV) in vitro by direct addition of viral particles into culture dishes. For details on the virus production, see Mosshammer et al. (2022). The optimal amount of virus was estimated by titration, and typically 1 µl of AAV particles was used per 2.7 ml of culture media. Neurons were used for the experiment 9 d after transduction.

Imaging

Hippocampal cells were cultured in glass bottom dishes (P35GC-1.5-14-C, MatTek). Before each experiment, the culturing medium was replaced by external solution composed of the following (in mM): 140 NaCl, 3 KCl, 2 CaCl₂, 2 MgCl₂, 10 HEPES, and 20 glucose; pH was adjusted to 7.4 by NaOH. Experiments were performed at room temperature, and cells were superfused continuously using an eight-reservoir drug application system (Octaflow II) with eight-channel micromanifold converging into a 100-µM-diameter quartz outlet. Neurons were imaged using a Nikon A1R confocal microscope equipped with a focus clamp.

The ATP/ADP ratio was determined in neurons expressing the genetically encoded fluorescence sensor PercevalHR (referred to as “Perceval”) incorporated into an AAV vector, with excitation at 403 and 488 nm wavelengths, while the emission was detected at 525 nm. The ratio F488/F403 reflects the intraneuronal ATP/ADP ratio. The expression plasmid for PercevalHR (GW1-PercevalHR) was a gift from Gary Yellen (Addgene plasmid #49082).

Intracellular Ca²⁺ was measured using Fluo-4 AM (F14201, Invitrogen, Thermo Fisher Scientific). Neurons were incubated for 15 min at 37°C in 1 µM Fluo-4 AM-containing buffer solution followed by a washout with extracellular solution. Neurons were excited by 488 nm, and emission was detected at 525 nm.

Mitochondrial membrane potential (Ψ_mt) was measured with tetramethylrhodamine methyl ester (TMRM, T668, Thermo Fisher Scientific). Cultures were equilibrated for 45 min in 1 nM TMRM at 37°C. The incubation period was not followed by a washout, and the dye was present in all perfusing solutions. Neurons were excited at 562 nm, and emission was detected at 595 nm. The terms hyperpolarization and depolarization of Ψ_mt were defined as any increase or decrease of the TMRM fluorescence relative to the value prior to the application of test solutions or compounds (Connolly et al., 2018).

To monitor neuronal morphology, we excited neurons transfected with GFP by 488 nm, and emission was detected at 525 nm. In order to capture all neuronal extensions, Z-stack images of individual GFP-expressing neurons were obtained using a 63× oil immersion objective.

Metabolic tracing experiments

Neuronal cultures were removed from the incubator and equilibrated at room temperature. The culturing medium was replaced by external solution composed of the following (in mM): 140 NaCl, 3 KCl, 2 CaCl₂, 2 MgCl₂, 10 HEPES, 2 glucose, and 0.1 glutamine; pH was adjusted to 7.4 by NaOH for 30 min. External solutions containing unlabeled glucose or glutamine were then exchanged for solutions containing equimolar concentrations of the isotopically labeled analogs 13C5 glutamine (CLM-1822-H-0.1, Cambridge Isotope Laboratories) and U13C6 glucose (CLM-1396-1, Cambridge Isotope Laboratories). Neurons were incubated for further 2 h under otherwise identical conditions at room temperature. Sample harvesting: after 2 h, incubating solutions were removed, and cultures were washed with ice-cold PBS and immediately frozen with −20°C mixture of HPLC grade MeOH:ACN:H₂O (2:2:1, quenching solution) and put on dry ice for 15 min. Afterward, the samples were thawed on ice, and the material was collected by scraping and was transferred along with the quenching solution into separate tubes. Samples were exposed to three rounds of freeze–thaw cycles in liquid nitrogen interleaved with 10-s-long vortexing. After this procedure, the samples were left for 1 h at −20°C and were then centrifuged at 13,000 rpm. The resulting supernatant containing metabolite extracts was collected and stored at −80°C for the analysis. The analysis of metabolite tracing experiments was performed by the Metabolomics Facility at Vienna BioCenter Core Facilities, member of the Vienna BioCenter.

Metabolite extracts were analyzed by hydrophilic interaction liquid chromatography (HILIC), directly coupled to tandem mass spectrometry (LC-MS/MS) via electrospray ionization. Metabolites were analyzed with HILIC, by injecting 1 µl of each sample onto a polymeric iHILIC-(P) Classic HPLC column (HILICON, 100 × 2.1 mm; 5 µm) connected with a guard column. A flow rate of 100 µl/min was used and a linear gradient (A, acetonitrile; B, 10 mM aqueous ammonium bicarbonate, supplemented with 0.1 µg/ml medronic acid) was applied, starting with 25% B and ramping up to 70% B in 8 min. Using a TSQ Quantiva mass spectrometer (Thermo Fisher Scientific), employing selected reaction monitoring. The following transitions were used for quantitation in the negative ion mode: fructose 1,6-bisphophate m/z 339 → m/z 97, glucose-6-phosphate m/z 259 → m/z 97, dihydroxyacetone phosphate m/z 169m/z → m/z 97m/z, phosphoglycerate m/z 185 → m/z 79, 2-phosphoenolpyruvate (PEP) m/z 167 → m/z 79, lactate m/z 89 → m/z 43, pyruvate m/z 87 → m/z 43, citrate m/z 191 → m/z 111, and malate m/z 133 → m/z 115. Glutamine m/z 147 → m/z 130 and glutamate m/z 148 → m/z 130 were measured in the positive ion mode. Retention times were validated using authentic standards. In addition to the unlabeled metabolites (all ¹²C) containing additional ¹³C atoms were analyzed, e.g., glucose-6-phosphate (+6 ¹³C) m/z 265 → m/z 97, 2-phosphoenolpyruvate (+3 ¹³C) m/z 170 → m/z 79, or malate (+2 ¹³C) m/z 135 → m/z 117, malate (+3 ¹³C) m/z 136 → m/z 118, and malate (+4 ¹³C) m/z 137 → m/z 119. Untreated samples were used to determine the native ¹³C incorporation in the respective metabolites, and these percentages were subtracted accordingly. From the refined data, relative ¹³C incorporation percentages were calculated.

Molar percent enrichment of isotopes was calculated as the percentage of all labeled atoms within the metabolite as follows:

$$\sum _{i=1}^n{\displaystyle \frac{M_{\mathrm{i}}\times i}{n},}$$

where n represents the number of carbon atoms in the particular metabolite and M_i denotes the relative abundance of the ith mass isotopomer.

Electrophysiology experiments

Current-clamp and voltage-clamp measurements were performed using a MultiClamp 700B amplifier (Axon Instruments) and the Clampex 10.5 software, which is part of the pCLAMP 10 electrophysiology data acquisition and analysis software package (Molecular Devices). Signals were low-pass filtered at 10 kHz and were digitized with a Digidata 1440A digitizer (Molecular Devices) at a sampling rate of 20 kHz. Patch pipettes were made of borosilicate capillaries (GB150-8P, Science Products) with a Sutter P97 horizontal puller (Sutter Instrument). Tip resistances were between 2 and 4 MΩ. The pipette solution was composed of the following (in mM): 120 potassium gluconate, 1.5 sodium gluconate, 3.5 NaCl, 1.5 CaCl₂, 0.25 MgCl₂, 10 HEPES, and 5 EGTA; pH was adjusted to 7.3 by KOH. All recordings were made in a perforated-patch mode using back filling of the pipettes with amphotericin B (500 µg/ml), which was added to the pipette solution just before experiments. The pipette tip had been filled by capillary force with amphotericin B-free solution. Experiments were started only when the series resistance had decreased to the lowest achievable level (between 20 and 30 MΩ), which usually required ≥15 min. Experiments were performed at room temperature (22–24°C), and cells were superfused continuously using a DAD-12 drug application system (Adams & List) with a 12-channel micromanifold converging into a 100-µM-diameter quartz outlet. The tip of the outlet was positioned in close proximity (∼250 µm) to the patch-clamped cell. The external solution was composed of the following (in mM): 140 NaCl, 3 KCl, 2 CaCl₂, 2 MgCl₂, 10 HEPES, and 20 glucose; pH was adjusted to 7.4 by NaOH. To isolate miniature excitatory postsynaptic potentials (mEPSPs), we maintained the neurons at −70 mV by means of slow current injections, and we added 10 µM Bic with 500 nM TTX to the external solution. To increase signal-to-noise ratios, we filtered the recorded signals with a 1 kHz low-pass Bessel filter. The recorded signals were corrected for the baseline, and individual events were analyzed using the threshold event detection algorithm in pCLAMP 10 (Molecular Devices). During recording of glutamatergic, GABAergic, and sucrose-induced currents, 500 nM TTX was present in all solutions. Test compounds in external solution were applied from reservoirs connected to 1 of the 12 micromanifold channels.

CRISPR/CAS9-mediated Glud1 knock-out

To achieve deletion of glutamate dehydrogenase (GDH) 1, we have chosen a two-vector strategy. pAAV-EFS-SpCas9 carrying Cas9 was a gift from Ryohei Yasuda (Addgene plasmid #104588; http://n2t.net/addgene:104588; RRID: Addgene_104588) and pX552 carrying sgRNA, and KASH-GFP under hSyn promoter was a gift from Feng Zhang (Addgene plasmid #60958; http://n2t.net/addgene:60958; RRID: Addgene_60958).

In pX552 the segment corresponding to sgRNA was exchanged using site-directed mutagenesis for a sgRNA targeted to the Glud1 gene (TGCAAGGCTTGATGATCCGC). Viral particles were produced from both plasmids by cotransfecting them along with helper plasmid pHelper (NovoPro) and pAAV-DJ packaging vector (Grimm et al., 2008) into tsa-201 cells. Viral particles were harvested after 3 d of incubation as previously described (Ben-Simon et al., 2022). Neurons were transduced at 14 DIV either with the complete set of AAVs or alternatively only by a virus carrying sgRNA. Cells were used for the experiments 9 d after transduction.

Immunofluorescence

Cultures were fixed in 5% paraformaldehyde for 7 min at room temperature, washed, blocked with 1% goat serum in PBS for 2 h, permeabilized with 0.1 Triton X-100 for 15 min, and incubated in rabbit anti-Glud1 (1:500, Proteintech, 14299-1-AP) overnight at 4°C. Following a triple washout with PBS, cells were incubated for 30 min at room temperature with the fluorescent secondary antibody goat anti-rabbit Alexa Fluor 594 (1:1,000, Thermo Fisher Scientific, A11012). After washing with PBS, cells were stained with 4′,6-diamino-2-phenylindole, dihydrochloride (DAPI; 300 nM) for 15 min at room temperature. After washing, the cells were visualized using confocal microscopy.

Data analysis

In all imaging experiments, fluorescence intensity was calculated as mean intensity from a defined region of interest positioned at neuronal somata. Neurons were taken into analysis irrespective of their neurotransmitter phenotype. All fluorescence data measured over time were normalized to unity by dividing the whole trace by a representative baseline value. Averaged normalized data traces are displayed in the figures together with standard errors of the mean (SEM).

The synchronicity of neuronal calcium responses shown in Figures 3 and 4 was analyzed in MATLAB (MathWorks) using a custom script. Briefly, Fluo-4 signals obtained from the somata of individual neurons were normalized to the corresponding baseline values. Cross-correlation coefficient (Corr coeff) was calculated in MATLAB by the “corrcoef” function applied to individual normalized fluorescent traces, and the average value of Corr coeff was computed for each experiment. To identify event positions within each Fluo-4 time series, we calculated the first derivative together with the root mean square (RMS) value. Only peaks larger than 1.5× RMS were identified as events. To characterize the size of events, we estimated the area under the curve by integrating the fluorescence values of individual events from traces whose baselines were set to 0. Sizes of events obtained from all traces were then averaged, and a mean event size was plotted. One to two fields of view were analyzed per culture dish.

Figure 3.

PDS, LTCC-mediated Ca²⁺ influx and glutaminolysis are required for the generation of synchronized, seizure-like neuronal activity. Neurons were incubated in PDS-inducing solution (Bic + BayK) or solution that induces rapid neuronal firing without PDS (Bic + Isra) for a total of three days. Neuronal discharge patterns were monitored at the time points of 1 h, 24 h and 72 h. A, Example traces representing events recorded by perforated-patch current–clamp electrophysiology after 1 h, 24 h and 72 h of exposure to Bic + BayK (top traces) or Bic + Isra (bottom traces). Note the change of timescale between traces corresponding to 1/24 and 72 h. B, Statistical comparison of the mean event size under the two conditions shown in A over time. n = 12–15 events. C, Example traces illustrating the effects of prolonged exposure to Bic + BayK in the absence (top traces) or presence of SP (bottom traces) on spontaneously occurring Ca²⁺ signals obtained from six neurons within one field of view. Fluo-4 traces of individual neurons are stacked for better visibility. Sensitivity of Bic + BayK-induced discharge events to ASMs is shown in Extended Data Figure 3-1. D,E, Evaluation of synchronicity (D) and size of individual Ca²⁺ signals (E) in the course of 72 h of incubation in the presence of Bic + BayK, Bic + Isra, and Bic + BayK + SP. n = 9–16 culture dishes. F, Example traces from calcium imaging experiments on neurons exposed to an alternative model of epileptogenesis utilizing 10-min-long application of 10 µM glutamate followed by 10 DIV of recovery in the presence or absence of GDH inhibitor R162 or the LTCC antagonist Isra. Fluo-4 traces of six individual neurons are stacked for better visibility. G,H, Statistical evaluation of synchronicity (G, by means of the correlation coefficient) and of mean size (H) of the Ca²⁺ signals from experiments shown in F. n = 12 fields of view. *p < 0.05, **p < 0.01, and ***p < 0.001.

Figure 4.

Effect of genetic deletion of GDH1 on mitochondrial membrane voltage and spontaneously occurring Ca²⁺ signals in neurons exposed to Bic + BayK for 72 h. A, Example fluorescence micrographs illustrating the immunofluorescent evaluation of Crispr/Cas9-mediated deletion of GDH1 in primary hippocampal neurons. Transduced neurons were labeled with GFP (green), neuronal nuclei were stained with DAPI (blue), and GDH1 was detected using a GDH1-directed primary and a fluorescently labeled secondary antibody (red). Scale bar, 10 µM. B, Fluorescence signals of neurons transduced with Crispr/Cas9 are compared with neurons transduced with sgRNA only. Scale bar, 7 µm. n = 16–26 neurons. C, Fluorescence micrographs of GDH1 KO neurons and of neurons transduced with sgRNA only loaded with TMRM to illustrate the effect of GDH1 deletion on V_mito in the presence of external buffer, Bic + BayK + glutamine and Bic + BayK + glutamine + FCCP. D, V_mito monitored with TMRM was evaluated in GDH1 KO neurons and control neurons (sgRNA) over time. The effect of PDS induction can be seen as a decrease of normalized TMRM fluorescence, which is indicative of mitochondrial depolarization. At the end of each experiment, the protonophore FCCP (1 µM) was applied to induce full depolarization of the mitochondria. Comparison of baseline TMRM levels in sgRNA and GDH1 KO neurons indicated a trend toward increased TMRM fluorescence in GDH1 KO neurons, but this difference was not statistically significant. (sgRNA = 431.61 ± 387.5; n = 26 neurons; GDH1 KO = 593.2 ± 366.4; n = 24 neurons). E, TMRM fluorescence data determined before (external buffer) and after addition of Bic + BayK for 10 min are compared for GDH1 KO neurons and neurons transduced with sgRNA only. n = 24–26 neurons. F, Example Fluo-4 fluorescence traces obtained from six neurons within a field of view illustrating the effect of GDH1 deletion on Ca²⁺ signals occurring spontaneously in presence of Bic + BayK in control neurons (sgRNA) but not in GDH1 KO neurons. G,H, Statistical evaluation of neuronal synchronicity (by means of the correlation coefficient of the Ca²⁺ signals) and of the mean size of individual Ca²⁺ signals in GDH1 KO neurons and neurons expressing only sgRNA. n = 7 culture dishes. *p < 0.05 and ****p < 0.001. I, Example traces representing events recorded by perforated-patch current–clamp electrophysiology after 72 h of exposure to Bic + BayK in control neurons (sgRNA) and in neurons with deleted GDH1 (GDH1 KO). J, Statistical evaluation of average event sizes quantified as the area under the curve obtained from control neurons (sgRNA, black) and GDH KO neurons (red). n = 39–60 events from 13–20 neurons.

Figure 3-1

Effect of ASMs on seizures and epileptogenesis. (A) Experimental outline: Neurons were incubated in PDS-inducing solution for three days to develop seizure-like events. During the imaging antiseizure drug valproic acid (200  µM, VPA) was applied and its effect on calcium transients was monitored. (B, C) Effect of VPA on neuronal synchronicity evaluated by correlation coefficient (Corr coeff) calculated in between neurons within the same field of view (B) and mean area under the curve of discharge events (D). n = 5 dishes. (D) Experimental outline: To monitor the effect of ASMs on seizure-development, neurons were incubated in PDS-inducing solution in presence of either valproate (200  µM) or vigabatrin (75  µM) for three days. The drugs were removed only six hours before imaging and the resulting discharge activity was monitored in their absence. (E,F) Statistical comparison of the effect of continues presence of antiseizure medications on neuronal synchronicity and size of individual discharge events. n = 6-7 dishes. Download Figure 3-1, TIF file ^{(1.3MB, tif)} .

The neuronal complexity of GFP-expressing neurons was assessed by Sholl analysis in Fiji (Schindelin et al., 2012). Maximal intensity projections from z-stack images of neurons were computed and were subjected to thresholding and tracing. Traced binary images were analyzed. Spine density was estimated from a distal dendrite of GFP-expressing neurons. A short linear stretch of a dendrite was chosen, and spines were identified by their morphology and manually counted. To calculate the spine density, we divided the total number of spines by the length of a dendrite on which spines were located.

Areas under the curve of individual miniature EPSPs were determined from 3-min-long recordings using event detection algorithm of Clampex 10.5 (Molecular Devices).

Statistical analysis

GraphPad Prism version 10.3.1 was used to prepare graphs and to perform statistical analyses, subsequent to testing for normal distribution of the data. Unless stated otherwise, the n numbers represent individual neurons in imaging as well as electrophysiological experiments and independent culture dishes in metabolite tracing experiments.

The data used in this study were all obtained from at least two independent cell culture preparations. Each preparation contained neurons derived from two female and two male rat pups. Unpaired t test was used for comparison of two conditions, while one-way ANOVA with Tukey's multiple comparisons was used whenever three or more experimental conditions with normally distributed samples were compared. One, two, three, and four asterisks indicate p < 0.05, p < 0.01, p < 0.001, and p < 0.0001, respectively. Statistically nonsignificant differences (p ≥ 0.05) are not labeled.

Results

Glutamine supports neuronal bioenergetics during PDS firing

IIS/PDS as observed during the latent phase of epileptogenesis arise in response to brain insults or status epilepticus (SE) and are distinct from seizures seen during the chronic phase of epilepsies (Fig. 1A). Experimentally, PDS can be triggered in primary hippocampal cocultures of neurons and glial cells by blockage of GABA_A receptors by Bic (10 µM) and simultaneous enhancement of L-type calcium currents by BayK (3 µM; Fig. 1B; Stiglbauer et al., 2017; Meyer et al., 2021). Since we used the N-methylated derivative of Bic (Bic methiodide), which was reported to inhibit calcium-activated potassium channels of the SK family (Johannson et al., 2001), reduced activity of these channels may also contribute to PDS formation. Bic-induced hyperexcitability mimics one aspect of epileptogenesis caused by the loss of inhibitory interneurons (Toth et al., 1997; Staley and Dudek, 2006; Huusko et al., 2015; Parga Becerra et al., 2021). BayK increases the activity of L-type calcium channels (LTCC) which has been shown to occur pathophysiologically due to release of ROS (Akaishi et al., 2004; Jarrett et al., 2008) or various stress hormones such as glucocorticoids, IGF-1, or GDF-15 (Bence-Hanulec et al., 2000; Gao et al., 2006; Chameau et al., 2007; Johnstone and Hool, 2014; Lu et al., 2016). PDS evoked by Bic + BayK in vitro (Stiglbauer et al., 2017) are indistinguishable from those observed in brain slices (Schiller, 2002) or in animal models of epilepsy (Matsumoto and Marsan, 1964). For control purposes, PDS waveforms evoked by Bic + BayK can be compared with electrical events induced by Bic plus the L-type channel blocker Isra (3 µM; Fig. 1B); this latter combination evoked enhanced neuronal firing but without PDS formation (Stiglbauer et al., 2017).

To study neuronal metabolism during PDS discharge, we used the genetically encoded fluorescent indicator of the cytosolic ATP/ADP ratio Perceval HR (Tantama et al., 2013) as reported previously (Hotka et al., 2020; Dhoundiyal et al., 2022). Induction of PDS firing by Bic + BayK led to an immediate decline in the ATP/ADP ratio even though sufficient glucose was available throughout the experiment (Fig. 1C, blue trace). This decrease was insensitive to the addition of pyruvate (1 mM) or lactate (1 mM) to the glucose-containing solution (Fig. 1C,D, red and orange traces). However, when glucose was supplemented by a physiologically relevant concentrations of glutamine (100 µM; Dolgodilina et al., 2016), the decrease was attenuated, and within 5 min, the neuronal ATP/ADP ratio reached a steady state (Fig. 1C,D, black trace). This effect was not seen when adding an equal concentration of proline (Fig. 1C,D, green trace). Within the glutamate–glutamine cycle, glial cells supply neurons with glutamine (Bak et al., 2006). In neurons, glutamine-derived glutamate is not only stored in vesicles but can be fed into the TCA cycle after conversion to alpha-ketoglutarate in order to support cellular bioenergetics. To test for a potential metabolic role of glia-derived glutamine during PDS firing, we exposed the cultures to Bic + BayK either under stationary conditions (with Bic + BayK being present in the bath) or with continuous superfusion (cells under investigation being superfused with Bic + BayK). The rationale behind this experiment is that, in contrast to stationary conditions, glutamine derived from glia is washed out by cell superfusion. Consistent with this, in stationary buffer, the ATP/ADP ratio remained stable even in glutamine-free Bic + BayK solution (Fig. 1E, first 2 min of the trace) but started to decline as soon as superfusion began (Fig. 1E). To confirm a role of glia, values of neuronal ATP/ADP ratios in the presence of Bic + BayK were compared between neuron/glia cocultures and pure neuronal cultures: in the presence of glia cells, neurons were able to maintain ATP/ADP ratio stable during Bic + BayK-induced PDS firing (Fig. 1F) but failed to do so in the absence of glia (Fig. 1G). Hence, during PDS firing, neurons utilize glutamine to synthesize ATP, whether derived from glia or added exogenously.

To assess the metabolic fate of glutamine further, we prevented glutamate from entering the TCA cycle by two different strategies: a GDH inhibitor (R162, 50 µM; Fig. 1H, red trace) was used to inhibit the conversion of glutamate to alpha-ketoglutarate, and SP (200 µM; Fig. 1H, blue trace) was employed to inhibit the downstream enzyme alpha-ketoglutarate dehydrogenase (aKGDH): both inhibitors let the ATP/ADP ratio in PDS-firing neurons collapse even in the presence of glucose and glutamine (Fig. 1H). An alternative pathway of glutamate oxidation involves its conversion to alpha-ketoglutarate by the malate–aspartate shuttle enzyme aspartate aminotransferase (Daikhin and Yudkoff, 2000). Indeed, in neurons incubated in external buffer (control, black trace), inhibition of this enzyme by aminooxyacetic acid (AOA; 1 mM) led to a gradual decline of the cytosolic ATP/ADP ratio. However, in PDS-firing neurons (Bic + BayK, orange) AOA application did not affect the ATP/ADP ratio to a detectable level, indicating that the activity of the malate–aspartate shuttle was low. Therefore, during PDS firing, glutamate entry to the TCA cycle appears to be predominantly controlled by GDH.

To learn whether the neuronal utilization of glutamine was specific for PDS, we exposed the neurons to 10 µM Bic in the presence of the LTCC antagonist Isra instead of BayK: with LTCC being blocked, glutamine was not able to support neuronal ATP synthesis (Isra, 3 µM; Fig. 1J,K). To find out whether neurons might utilize glutamine during convulsions, we omitted the Mg²⁺ ions from the external solution (0-Mg²⁺ solution; Fig. 1L,M) to trigger seizure-like activity (a detailed electrophysiological example of a 0-Mg²⁺-induced seizure–like discharge is shown in Figure 1 in Meyer et al., 2021). Under such seizure-promoting conditions, glutamine was not able to support neuronal ATP production (Fig. 1J–M, black traces). Thus, neuronal glutamine metabolism provides the bioenergetic basis for ATP synthesis during PDS generation, but not during high-frequency action potential firing.

Increased neuronal glutaminolysis substitutes for a decrease in glucose metabolism

Interictal glucose hypometabolism is observed in animal models (Dubé et al., 2001) as well as human patients (Rausch et al., 1994). As neurons can use glutamine to generate ATP during PDS firing, we assumed that this increase in glutaminolysis would be accompanied by a decrease in glucose utilization.

Metabolism of glucose and glutamine was monitored in parallel by using the stable isotope tracers [U-¹³C₆] glucose (CLM-1396) and [U-¹³C₅] glutamine (CLM-1822) to quantify label incorporation into glycolytic and TCA cycle metabolites. Neurons were exposed to either PDS-inducing conditions (Bic + BayK) or Bic + Isra as control, both applied to the cultures for 2 h in the presence of the tracers (Fig. 2A). After Bic + BayK exposure, glycolytic intermediates (Fig. 2B, blue bars) and the TCA cycle intermediates malate and citrate (Fig. 2C,D) contained less labeling derived from the glucose tracer than after exposure to Bic + Isra (Fig. 2B–D, red bars). Since glutamate synthesis is tightly linked to glucose oxidation (Rothman et al., 2024), [U-¹³C₆] glucose-derived label incorporation into glutamate was reduced after Bic + BayK as compared with Bic + Isra exposure (Fig. 2E). Vice versa, malate, citrate, glutamate, and glutamine contained more [U-¹³C₅] glutamine-derived labeling after Bic + BayK than after Bic + Isra exposure (Fig. 2F–I). These data indicate that glucose hypometabolism during PDS discharge is compensated for by an increased feeding of glutamine/glutamate into the TCA cycle. Total levels of analyzed metabolites as well as example electrical activity induced by Bic + BayK incubation of neurons in the absence of glia are shown in Extended Data Figure 2-1. The example electrical activity shown in Extended Data Figure 2-1I illustrates that neither a reduction in the glucose level nor a change in culture medium to Neurobasal A medium, which was used to isolate neurons in the tracing experiments, altered the inducibility or gross appearance of the PDS in the electrophysiological recordings.

Glutaminolysis is required for neuronal synchronization and development of seizure-like activity

To study long-term consequences of PDS firing and concomitant glucose hypometabolism, neuronal cultures were incubated in Bic + BayK or Bic + Isra for up to 3 d. Neuronal activity was monitored by patch-clamp recordings in a current-clamp mode after 1, 24, and 72 h (Fig. 3A). Within the first hour of incubation, Bic + BayK triggered PDS, whereas in Bic + Isra-treated cells, only single action potential firing at short intervals was observed (Fig. 3A, left traces). After 24 h of exposure to Bic + BayK, depolarizing events increased about threefold in duration (duration of Bic + BayK 1 h = 0.7 ± 0.45 s; duration of Bic + BayK 24 h = 2.41 ± 01.2 s) and about ninefold after 72 h (duration Bic + BayK 72 h = 6.01 ± 1.7 s). This prolongation translated into a sevenfold increase of the mean event size within 72 h (Fig. 3A, middle traces; Fig. 3B). Such long-lasting events with durations of up to 10 s (Fig. 3A, top right trace) no longer fulfill the criteria to be classified as PDS; they rather resemble seizure-like events in terms of their duration (Dulla et al., 2018). In Bic + Isra-treated neurons, there was no change in the event size after 24 h, and the increase after 72 h was less pronounced than in Bic + BayK-treated neurons (2.8-fold increase within 72 h; Fig. 3A, bottom traces; Fig. 3B). Moreover, events observed after 72 h of Bic + Isra remained short and did not exceed 3 s (Fig. 3A, bottom right trace). Thus, after 72 h in Bic + BayK, PDS convert into seizure-like activity, but no such transformation is seen in Bis + Isra.

A fundamental characteristic of epilepsies is highly synchronous firing within neuronal ensembles. To evaluate synchronicity of the neuronal activities in our cultures, we employed calcium imaging using Fluo-4 AM (Fig. 3C–E) as a proxy. Synchronicity was determined by means of the correlation coefficient of the recorded spontaneously occurring Ca²⁺ signals. When cultures were exposed to Bic for up to 1 h, whether in combination with BayK (Fig. 3C, top traces) or Isra (Fig. 3C, bottom traces), Ca²⁺ transients were the same with respect to synchronicity and size, and this was also the case when the aKGDH inhibitor SP (200 µM) was coadministered (Fig. 3D,E). However, after 24 h in Bic + BayK, these Ca²⁺ transients were completely synchronized as indicated by a correlation coefficient close to 1, and this remained unchanged for up to 72 h (Fig. 3D). Neurons incubated in either Bic + BayK + SP or Bic + Isra never reached comparable levels of synchronicity (Fig. 3D). In Bic + BayK, the size of Ca²⁺ transients displayed a marked increase after 48–72 h (Fig. 3E). A seizure-like nature of the neuronal activity underlying these long-lasting Ca²⁺ elevations was supported by their sensitivity to acute application of the broad-spectrum ASM VPA (200 µM; Extended Data Fig. 3-1). However, the development of this seizure activity was unaffected by either VPA or another ASM, Vig (Extended Data Fig. 3-1), when these drugs were continuously coadministered with Bic + BayK for 72 h and then removed 6 h before imaging. Vig was chosen because, in addition to its main mode of action on GABA transaminase, it is reported to act as a weak inhibitor of the mTOR complex (Zhang et al., 2013), which has been implicated in epileptogenesis (Guo et al., 2024).

In contrast to Bic + BayK, the size of the Ca²⁺ transients in neurons exposed to Bic + Isra or Bic + BayK + SP did not change throughout 72-h-long monitoring (Fig. 3E).

Notably, fluorescence measurements using Perceval revealed that the ratio of cytosolic ATP/ADP ratio was the same whether neurons had been incubated for 72 h in Bic + BayK [5.32 ± 3.7 (mean cytosolic ATP/ADP ratio ± SD; n = 46)] or Bic + BayK + SP [5.05 ± 3.5 (mean cytosolic ATP/ADP ratio ± SD; n = 52)]. Hence, the failure of neurons exposed to the aKGDH inhibitor to display highly synchronous and huge Ca²⁺ transients is not caused by energetic failure. Thus, the development of abnormally synchronized and exaggerated neuronal activity under PDS-promoting conditions is dependent on functional aKGDH.

The substrate of aKGDH is alpha-ketoglutarate which can be derived from glutamate by GDH. To confirm that the key role of aKGDH in the development of seizure-like activity is related to the feeding of glutamate into the TCA cycle, we blocked GDH by R162, and we subjected the neuronal/glial cocultures to an alternative in vitro epileptogenesis model: short-term exposure to increased glutamate concentrations followed by 10-day recovery periods (Sun et al., 2001; Jablonski et al., 2021). After this period, neurons displayed synchronous huge Ca²⁺ transients which, however, did not evolve when GDH was blocked by R162 for the entire 10 d recovery period. To confirm that LTCCs are crucially involved in the development of these synchronous huge Ca²⁺ transients, we alternatively exposed cultures to Isra (instead of R162), which also prevented the occurrence of these correlates of seizure-like activity (Fig. 3F–H).

To further test a potential key role of GDH in the PDS-induced metabolic change, we removed GDH1 (gene, Glud1) by Crispr/Cas9-mediated knock-out. In order to direct the deletion of GDH1 to neurons, the guide RNA was expressed under the control of the synapsin (hSyn) promoter, and the expression of KASH-GFP was driven by the very same vector. Accordingly, successfully transduced neurons were identified by KASH-GFP fluorescence, and GDH1 expression was quantified over neuronal somata as mean fluorescence provided by anti-GDH1 antibody staining. Example micrographs of this series of experiments are shown in Figure 4A. Fluorescence data were compared for knock-out neurons (GDH1 KO) versus neurons expressing sgRNA and KASH-GFP only (sgRNA; Fig. 4B). Within the latter, GDH1 appeared in a punctuate pattern distributed over entire neuronal somata that also contained GFP. In knock-out neurons, somata were marked by KASH-GFP but showed no GDH1-dependent fluorescence signal (Fig. 4A,B).

Successful knock-out was also confirmed by monitoring functional consequences of GDH1 deletion. Therefore, mitochondrial membrane potential (V_mito) was assessed by TMRM (Fig. 4C–E) fluorescence. In neurons transduced with sgRNA only, exposure to Bic + BayK + glutamine left the TMRM signal unaltered as compared with baseline levels, but the latter were reduced by a subsequent addition of R162 to inhibit GDH1. After deletion of the enzyme, neurons were unable to maintain stable V_mito when exposed to Bic + BayK + glutamine, and R162 failed to exert additional effects (Fig. 4D,E, red trace). These results confirm that GHD1 is crucial for energy production in neurons exposed to Bic + BayK.

When firing patterns within neuronal ensembles were tested after 72 h incubation in Bic + BayK, neuronal GDH1 deletion was found to reduce the correlation coefficient and the mean event size of Ca²⁺ signals (Fig. 4F–H) and to lead to a reduction in the size of individual electrical discharges as measured by patch clamp (Fig. 4I,J). Thus, knock-out of GDH1 not only compromised mitochondrial energy supply during PDS firing but also prevented the neurons from developing seizure-like discharge activity.

A decline in glutamatergic transmission links glutaminolysis to neuronal synchronization

The next question we addressed was if and how the synchronization of neuronal activity is causally linked to the switch in glutaminolysis. Mechanisms of neuronal synchronization are multiple and include synaptic plasticity (Gansel, 2022). We considered that alterations of synaptic transmission, which is under tight control of metabolism (Myeong et al., 2024), may contribute to the synchronization process in our experiments. To evaluate synaptic properties, mEPSPs (Fig. 5A–C) were measured in mixed cultures of hippocampal cells. In neurons exposed to Bic + BayK for 24 h, mEPSP amplitudes were smaller than under control conditions, but after 72 h, they were increased (Fig. 5B). The frequency of mEPSP remained unaltered after 24 h but was increased after 72 h in Bic + BayK (Fig. 5C). Hence, the emergence of synchronization apparently correlated with a reduction of excitatory synaptic neurotransmission.

Figure 5.

Mechanism of neuronal synchronicity induced by glutaminolysis. Postsynaptic and presynaptic function was monitored in current-clamp or voltage-clamp measurements or in Fluo-4 fluorescence imaging experiments. Neurons were incubated up to 72 h (A–C,I) or for 24 h (D,E) in either PDS-inducing solution (Bic + BayK), Bic + Isra, or the corresponding solvent control (DMSO). In panels F–H, acute responses were monitored for 4 min of application of Bic + BayK together with the indicated receptor blockers. A–C, Example voltage traces (with inserts depicting individual events), cumulative probability of mEPSP amplitudes with their corresponding statistical evaluation and statistical evaluation of mEPSP frequencies recorded in the presence of Bic + BayK together with TTX in the course of 72 h. n = 6–10 neurons. *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.001. D, Average postsynaptic glutamatergic current evoked by vesicle release resulting from the application of hyperosmotic (500 mM) sucrose solution and statistical comparison of peak current density determined from Bic + BayK-, Bic + Isra-, and DMSO (control)-treated neurons. n = 7–10. E, Peak current density of receptor agonist evoked currents induced by application of 100 µM glutamate, 100 µM NMDA, 100 µM AMPA, or 100 µM GABA determined from Bic + BayK-, Bic + Isra-, and DMSO (control)-treated neurons. F, Example Fluo-4 fluorescence traces obtained from six neurons within a field of view stacked to illustrate the immediate synchronization of neuronal Ca²⁺ responses to acute application of Bic + BayK by partial postsynaptic inhibition with the AMPA receptor blocker CNQX (125 nM) but not the NMDA receptor blocker APV (1 µM). G,H, Statistical comparison of effects of postsynaptic inhibition illustrated in panel F on neuronal synchronicity and size of individual discharge. n = 6 culture dishes. I, Currents as in panel D induced by 10-s-long application of hyperosmotic sucrose solution to hippocampal neurons measured by patch clamp over the course of 3 d of treatment with Bic + BayK. The graph on the right side summarizes peak current density data and indicates statistical evaluation. *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.001.

As glutamate packaging into synaptic vesicles is tightly bound to glucose oxidation, a PDS-induced reduction in the rate of glucose metabolism must be expected to lead to diminished glutamate synthesis (Hertz and Chen, 2017). In addition, glutamate flux into the TCA cycle during glutaminolysis may further deplete the presynaptic glutamate pool that is needed for vesicle refilling (Ikemoto et al., 2003; Ishida et al., 2009). Therefore, we hypothesized that PDS-induced glutaminolysis might lead to a decrease in vesicular glutamate. To test this hypothesis, we discharged the readily releasable pool of vesicles by superfusion of our cultures with hypertonic (500 mM) sucrose, and we monitored the postsynaptic response in patch-clamp recordings (Rosenmund and Stevens, 1996). Neurons incubated in Bic + BayK for 24 h showed significantly smaller sucrose-evoked currents than control neurons (incubated only in solvent) or neurons incubated in Bic + Isra (Fig. 5D). When postsynaptic currents were induced through direct receptor activation by either glutamate, NMDA, AMPA, or GABA, no differences were observed (Fig. 5E). Thus, the decrease of the postsynaptic response in the hypertonic sucrose experiments appeared to be due to a presynaptic deficit, and we conclude that the same holds true for the mEPSP amplitude reduction seen after 24 h of exposure to Bic + BayK.

Our finding that a reduction in glycolysis appears to be linked with increased glutaminolysis and compromised glutamate release raises the question as to whether a resulting decrease in glutamatergic transmission might contribute to the synchronized neuronal activity that was observed on Days 1 and 2 in mixed neuronal/glial cultures exposed to PDS-inducing solution (Fig. 3C,D). To resolve this issue by an approach that does not rely on manipulating vesicular glutamate, neurons in the presence of glia cells were exposed to Bic + BayK in the absence or presence of antagonists of ionotropic glutamate receptors. In Bic + BayK, neurons demonstrated Ca²⁺ transients at high frequencies that were not synchronized (corr coeff = 0.5; Fig. 5F,G). Inhibition of NMDA receptors by 1 µM APV led to an immediate increase in synchronicity (corr coeff = 0.75) but no change in the size of the Ca²⁺ signals (Fig. 5G,H). Additional partial block of AMPA receptors by 125 nM CNQX enhanced synchronicity even further and enlarged the event size (Fig. 5G,H). In fact, Ca²⁺ transients in neurons acutely exposed to Bic + BayK + APV + CNQX were similar to those after 24 h exposure to Bic + BayK (compare the respective traces of Fig. 5F and Fig. 3C). All effects of these antagonists were reversible upon washout (Fig. 5F; Fig. 5G,H, rightmost bars). These results indicate that a reduction in glutamatergic transmission is sufficient to synchronize neuronal activity such as the one observed in neurons exposed to PDS-inducing solution.

To investigate the recovery of mEPSP amplitude seen after 72 h (Fig. 5A,B), we monitored sucrose-induced vesicle release after 24, 48, and 72 h from PDS induction (Fig. 5I). We found that a 48-h-long decrease in sucrose-evoked release of vesicular glutamate observed in Bic + BayK was followed by its complete recovery after 72 h (Fig. 5I). The restored release of glutamate therefore helps to explain the recovery of the mEPSP amplitude. However, based on the observed increase in mEPSP frequency (Fig. 5A,C), it is likely that further changes occur at the release sites.

Glutaminolysis rather than morphological alteration is linked to development of seizure-like activity

To investigate further presynaptic changes that may be responsible for the increase in mEPSP frequency observed 72 h after PDS induction (Fig. 5C), we chose to examine neuronal dendritic arborization and spine density, which can give a measure of the number of excitatory synapses (McKinney, 2010).

To reveal whether induction of PDS might affect neuronal morphology in the long-term, we exposed the GFP-expressing neurons to Bic + BayK for 72 h and then subjected them to Sholl analysis to asses dendritic branching. After 3 d in Bic + BayK, the complexity of dendritic arborization was increased in comparison with the respective controls (Fig. 6A,B). When neurons were incubated in Bic + BayK + SP to inhibit aKGDH, morphological alterations were the same as in Bic + BayK only and again significantly different from controls (Fig. 6A, right panel, B). In addition, neurons exposed to both, Bic + BayK and Bic + BayK + SP, had increased densities of dendritic spines as compared with neurons incubated in solvent (Fig. 6C,D). Bic + BayK + SP represents a condition in which the development of seizure-like activity is prevented (Fig. 3). Thus, the increase in dendritic arborization shown here does not suffice to allow for the development of epileptiform activity. Hence, we conclude that generation of energy by glutaminolysis, which is blocked by SP as glutamate cannot enter the TCA cycle, is required to initiate the changes that establish a condition in which seizure-like activity may occur. However, this idea is at odds with the observation that glucose utilization increases during seizures (Bazzigaluppi et al., 2017; Yang et al., 2013, p. 7967; Zilberter and Zilberter, 2017). Therefore, we focused on changes in glucose metabolism during the exposure of neurons to Bic + BayK for 72 h. Acute exposure of neurons to Bic + BayK did not raise cellular uptake of the fluorescent glucose analog 2NBDG (applied for 30 min) when compared with resting, i.e., pharmacologically unstimulated, neurons; in Bic + Isra, in contrast, 2NBDG uptake was accelerated (Fig. 6E,F). Likewise, after 48 h in Bic + BayK, the rate of 2NBDG uptake was still the same as in resting neurons, but after 72 h, glucose uptake increased to a level indistinguishable from that of neurons experiencing acute seizure-like activity due to the removal of Mg²⁺ (Fig. 6G,H). This increase in glucose uptake was accompanied by a decrease in neuronal glutaminolysis as indicated by a lack of effect of the GDH inhibitor R162 on the cytosolic ATP/ADP ratio after 72 h of incubation in Bic + BayK (Fig. 6I, blue trace and bar), which was otherwise seen after 24 and 48 h of incubation (Fig. 6I, black and red trace and bar; Fig. 1H,I, red trace and bar). This suggests a transient nature of the elevation of GDH-mediated glutaminolysis. Collectively, these observations indicate a reestablishment of glucose metabolism that goes hand in hand with the observed replenishment of the pool or the content of glutamatergic synaptic vesicles (the current experiments cannot distinguish between these two scenarios) as indicated by the increase in sucrose-evoked postsynaptic currents after 72 h in Bic + BayK (Fig. 5I). Hence, our data demonstrate that the increased neuronal glucose uptake and restored glutamate release occurred in parallel to the increase in dendritic arborization and in synaptic density (Fig. 6A–D). However, the transient switch to glutaminolysis seems to represent a condition sine qua non (i.e., being essential) for the emergence of seizure-like activity in our model, presumably due to its synchronizing effect on neuronal activity.

Figure 6.

Seizure generation is accompanied by increase in dendritic complexity, increased spine number, and reintroduction of glucose metabolism. A, Micrographs of GFP-expressing neurons cultured for 72 h in the presence of solvent only (DMSO), Bic + BayK or Bic + BayK + SP. Scale bar, 25 µm. B, Sholl analysis of neurons treated as indicated in panel A. n = 8–17. C, Example micrographs of spine containing dendrites from neurons treated as indicated in panel A. Scale bar, 10 µM. D, Statistical evaluation of the spine number per micrometer of dendritic length. Scale bar, 5 µm. n = 20–34 dendrites. E, Representative micrographs illustrating the uptake of 2NBDG (green) into the neuronal cytosol in the course of 30 min. Scale bar, 15 µM. F–H, Uptake of 2NBDG shown as an increase of mean fluorescence over time calculated from a region of interest placed at the neuronal body from neurons treated acutely with Bic + BayK (F, blue trace), Bic + Isra (F, red trace), or under control condition (F, gray trace). G, Same as in F for neurons incubated in Bic + BayK for 48 h (G, black trace) and 72 h (G, orange trace) or for neurons experiencing acute seizure-like activity induced by application of 0-Mg^2 + external buffer (G, pink trace). n = 10–14. H, Statistical evaluation of 2NBDG uptake by comparison of the slope of the traces shown in panels F and G. I, The presence or absence of a contribution of glutaminolysis to ATP production over the 3 d of exposure to Bic + BayK + glutamine revealed by an acute effect of R162 on cytosolic ATP/ADP ratio in neurons after 1 (black trace and bar), 2 (red trace and bar), and 3 d (blue trace and bar). n = 7–12. *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.001. For comparison, the dashed black line indicates ATP/ADP levels after acute application of R162 in neurons incubated in Bic + BayK for 20 min shown in Figure 1H (red trace).

Discussion

PDS as cellular correlates of preictal and/or IIS have been known for four decades (Gorji and Speckmann, 2009). Nevertheless, their pathophysiological relevance has remained enigmatic. Preictal/interictal periods in both animal models (Guo et al., 2009) and human patients (Vielhaber et al., 2003) are characterized by peculiar glucose hypometabolism. Again, the pathological relevance of the latter is unknown (Dienel et al., 2023). The results reported here were obtained in primary cultures of hippocampal cells and provide evidence for a link between PDS, glucose hypometabolism, and glutaminolysis and their role in the emergence of seizure-like activity.

From glutaminolysis to increased neuronal synchronicity

In neurons that were tuned to fire PDS by a block of GABA_A receptors and a concomitant elevation of Ca²⁺ influx via LTCCs (Meyer et al., 2021), energy supply was provided by the use of glutamine while glucose metabolism was slightly reduced. This metabolic shift appeared to be maintained during prolonged exposure to PDS-promoting conditions for at least 48 h and resulted in a decrease of the amount of glutamate that can be liberated either from readily releasable vesicles by osmotic shock as demonstrated by diminished sucrose-evoked currents or from single vesicles by spontaneous fusion as evidenced by a reduction of mEPSP amplitudes. This impairment in glutamatergic transmission, according to our findings, promoted synchronized neuronal activity which is a key feature of epilepsy (Fisher et al., 2014). This seemingly contradicts the common paradigm that strong excitation or reduced inhibition is required for neuronal synchronization (Shao et al., 2019; van Hugte et al., 2023) but is consistent with recent in silico and in vitro data that shed light on how a reduction in synaptic coupling may foster neuronal synchronization (van Drongelen et al., 2005; Jacob et al., 2019; Wright et al., 2021): a decline in synaptic fidelity can be expected to decrease the likelihood of smaller events to be transferred which, in turn, will favor the spreading of more pronounced and/or prolonged events within a neuronal network. In the end this raises synchronicity as observed in the present experiments. We do not intend to claim that impaired neuronal metabolism might be the only mechanism by which neurons synchronize and by which prolongation of the discharge patterns is achieved. Multiple additive mechanisms are likely to occur in parallel and to contribute to the phenotype observed. Nevertheless, metabolic shifts appear to represent key processes. However, to obtain a more complete picture, changes in ionic conductances that may arise at various stages of epileptogenesis in the present model should be considered in the future.

Glucose reutilization as a potential contributor to ictogenesis

The loss of glutamatergic tone and the resulting rise in synchronicity were accompanied by morphological adaptations such as an increased dendritic arborization and an expansion of dendritic spines, two alterations that were not dependent on glutaminolysis. As a functional correlate of these changes, the frequency of mEPSPs was heightened. This increase in excitatory transmission was complemented by restored release of vesicular glutamate which occurred concomitantly with an increase in glucose utilization. Functionally, this was reflected by an increase in mEPSP amplitudes. Ultimately, all these processes resulted in the emergence of seizure-like activity which was prevented when the interim energy supply through glutaminolysis was impeded.

Accordingly, the observed sequence of neuronal metabolic changes closely recapitulates what has been observed in animal models and human patients: during preictal/interictal periods, glucose hypometabolism is prevalent (Vielhaber et al., 2003; Guo et al., 2009), whereas glucose utilization rises within epileptic foci during seizures (Witte et al., 1994; Yang et al., 2013; Bazzigaluppi et al., 2017). The above results show that conditions that favor PDS firing transiently impede glucose utilization and promote glutaminolysis. After 72 h glucose consumption steps in again and neuronal firing converges into hypersynchronous seizure-like activity. The accompanying increase in dendritic arborization and synapse density alone was not sufficient to provoke seizure-like activity but can be expected to facilitate the switch to seizure-like activity which otherwise relies on the dynamic metabolic alterations.

Firing of PDS enables quasi neurodevelopmental processes in mature neurons

PDS represent depolarizations associated with neuronal disorders, with durations of up to several hundred milliseconds and amplitudes that allow for significant Ca²⁺ influx via high-voltage–activated calcium channels. In this respect, PDS are similar to giant depolarizing potentials (GDPs) which occur in a synchronized fashion during neuronal development, are initiated by depolarizing GABA_A receptor responses (rather than ionotropic glutamate receptor responses in PDS), and govern the differentiation of neurons (Kubista et al., 2019). PDS similarly to GDPs affect neuronal morphology by increasing neuronal arborization and spine formation.

Previously, we found that PDS induce reverse mode operation of ATP synthase which helped to maintain the mitochondrial membrane potential (Hotka et al., 2020). The present data indicate that ATP supply during reverse mode operation of ATP synthase triggered by PDS relies on glutaminolysis. Adult neurons typically meet their energy demand by glucose oxidation, whereas dependence of neurons on glutamine as energy source is rather a characteristic of immature neurons undergoing growth and differentiation (Agostini et al., 2016). From this perspective, PDS-induced reliance on glutamine with an accompanying increase in morphogenesis resembles a phenotype typical of immature neurons. Possibly, a quasiregression toward a less mature phenotype may aid brain repair following an insult. Neuronal repair mechanisms and differentiation share several key processes that contribute to neuronal plasticity and include increased transcription and translation, protein synthesis, and morphogenesis (Liu and Jan, 2020; Joy and Carmichael, 2021). Glutamine may support those processes in multiple ways. Glutamine represents a readily available source of nitrogen and carbon, which are required for nucleotide and amino acid synthesis or for lipid production (Yoo et al., 2020). In addition, glutamine-derived glutamate is used for glutathione synthesis and for ATP production within the TCA cycle (Rae et al., 2024). In this way, increased uptake of glutamine helps to maintain the redox balance and at the same time provides fast bioenergetic support for the abovementioned processes.

PDS, a double-edged sword in epileptogenesis

Whether PDS play a pro- or rather antiepileptic role remains an unresolved question (Kubista et al., 2019). IIS/PDS occur early after a precipitating injury (Staley and Dudek, 2006; Dyhrfjeld-Johnsen et al., 2010; Puttachary et al., 2016), and we could demonstrate that acute induction of PDS firing led to reverse mode operation of mitochondrial ATP synthase (Hotka et al., 2020) which was associated with neuroprotective actions. Concomitant increase of neuronal glutaminolysis can be expected to reinforce this mechanism as follows. Neurons utilizing glutamate as fuel have decreased evoked release of glutamatergic vesicles, and the resulting loss of neuronal excitability has been shown to protect neurons from excitotoxic injury (Divakaruni et al., 2017). However, our data indicate that such neuroprotective mechanism may be transient only, as prolonged neuronal glutaminolysis ultimately led to hyperexcitability and seizure generation. Thus, our results from experiments on cultured hippocampal cells collectively argue in favor of a neuroprotective role of PDS despite their potentially epileptogenic activity. Notably, the concurrent neuroprotective and epileptogenic potential of PDS limit their value as target for intervention with epileptogenesis. PDS interruption may be well suited to halt the development of seizures after an insult; however, neurodegeneration might be exacerbated when PDS are suppressed too early. In vivo investigations will be necessary to corroborate this claim.

ASMs do not inhibit PDS-induced epileptogenesis

Although ASMs may provide neuroprotection following SE, evidence supporting their potential to prevent the onset of spontaneous seizures is scarce (Miziak et al., 2020) (D’Antuono et al., 2010). When VPA was administered continuously for 4 weeks to animals immediately after a 4 h SE, the resulting loss of hippocampal neurons was mitigated, but seizure frequency or severity after discontinuation of VPA was not (Brandt et al., 2006). One reason for this lack of effect might be the inability of ASMs to suppress IIS: with epileptiform activity in brain slices triggered by 4-aminopyridine, VPA turned out to suppress ictal discharges and to leave interictal discharges unaffected (D’Antuono et al., 2010). In line with this observation, acute application of VPA onto neurons incubated in Bic + BayK for 72 h prevented the occurrence of seizure-like events but left shorter synchronous spikes unaffected (Extended Data Fig. 3-1). Moreover, incubation of neurons in Bic + BayK plus VPA or VIG for 3 d failed to suppress seizure generation (Extended Data Fig. 3-1). The primary target of ASMs is high-frequency action potential firing during epileptic seizures. In contrast to seizures, IIS/PDS represent relatively short events consisting of a depolarized plateau with only minor contribution of action potential discharge. Since LTCC-mediated Ca²⁺ influx, in particular the one via Ca_v1.3 channels (Stiglbauer et al., 2017), provides an important contribution to the depolarized plateau, LTCC inhibitors might be used to halt PDS. Given the key role of LTCC isoform Ca_v1.2 in various brain functions (Berger and Bartsch, 2014), a subtype-selective Ca_v1.3 inhibitor would be required to avoid major side effects. However, such a drug is not available at present (Filippini et al., 2023); hence the easy way to handle in vitro model described here may prove useful to identify additional targets in PDS-induced epileptogenesis.