A stage-resolved neuron-glia transcriptional atlas reveals a glial inflammatory pivot in epilepsy Authors: Toni Christoph Berger, Magnus Dehli Vigeland, Hanne Sagsveen Hjorthaug, Cecilie Gjessing Nome, Christoph Patrick Beier, Erik Taubøll, Kaja Kristine Selmer, Kjell Heuser Published: 2026 ABSTRACT BackgroundEpileptogenesis transforms a healthy brain into an epileptic network, yet the temporal and cell-type-specific molecular events driving this transition remain poorly defined. Neuron-glia interactions are essential in this process, but no study has systematically charted their transcriptional dynamics from the acute insult to chronic epilepsy.MethodsUsing the intracortical kainic acid mouse model that recapitulates key hallmarks of mesial temporal lobe epilepsy with hippocampal sclerosis in humans, we performed Fluorescent Activated Nuclear Sorting of NeuN+ (neuronal) and NeuN⁻ (glia) nuclei followed by RNA sequencing at 1 h, 24 h, and 3 months after status epilepticus. Differential expression and integrative GO/KEGG analyses resolved stage-specific molecular programs across cell types.ResultsThe majority of genes differentially expressed in neurons and glia were exclusive to the respective time point investigated. We also identify a sequential reorganization of cellular gene expression changes during epileptogenesis. The acute phase is dominated by a shared stress response and DNA-repair programs in both neurons and glia. At 24 h, glia undergoes a marked transcriptional pivot involving necroptosis-associated, TNFR1/IFN-linked, and COX-2/chemokine pathways, while neurons display immune- and plasticity-related signatures. By 3 months, transcriptional activity is largely confined to glia and enriched for inflammatory, angiogenic, and gliogenic processes, consistent with long-term neurovascular remodeling. Only a few transcripts, including Parp3 (neurons) and Tlr1 (glia), are dysregulated across all stages.ConclusionThese findings reveal an orderly transition from an acute protective-leaning program to a early latent glial inflammatory/regulated-death state, culminating in chronic gliopathy. Our work provides, to our knowledge, the first cell-type–resolved temporal atlas of epileptogenesis and identifies the early latent phase as a mechanistically tractable window for antiepileptogenic intervention.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s40478-026-02323-w. Introduction The concept of epileptogenesis, the process by which a healthy brain transitions into an epileptic state, was first explored in research on focal epilepsies during the 1950s [1]. It describes the metamorphosis of a physiological into an epileptic brain [2]. In humans, this process can be observed in acquired epilepsies, particularly in the context of mesial temporal lobe epilepsy (MTLE), often characterized by progressive, drug-resistant seizures [3–5]. Epileptogenesis typically initiates after an inciting event, e.g. complex febrile seizures, head trauma, hypoxia, stroke or infection [6]. This event triggers a latent phase, during which molecular changes occur without overt clinical symptoms, eventually leading to the chronic phase marked by recurrent seizures [2, 7]. While epileptogenesis in humans can earliest be studied after the occurrence of the first clinical seizure, animal models involving all stages of the process starting from the initial incident, are crucial to a better understanding of epileptogenesis [8] and the development of anti-epileptogenic drugs that prevent the development of epilepsy in the first place [9–12]. Despite substantial progress in our understanding of chronic epileptogenesis [13], one in three patients with epilepsy do not experience sufficient seizure control even when adequately treated [14–16]. Possible shortcomings of previous anti-seizure medication (ASM) development involve the usage of traditional preclinical ictogenesis focused models for assessment of efficacy. Alas these models are often inadequate [15], and lack focus on molecular upstream mechanisms including inflammation [16] and underestimate glial involvement in epileptogenesis [17, 18]. We addressed these issues by using a mouse model for epileptogenesis that recapitulates key hallmarks of human MTLE including neuronal death, reactive gliosis and hippocampal sclerosis [19–21]. In this study, we provide a comprehensive cell- and stage-specific transcriptomic analysis of epileptogenesis, spanning from the first hour post-injury to three months later, mirroring acute, early latent and chronic phase of epileptogenesis. Using advanced cell-sorting techniques, RNA sequencing [22–24], and integrative bioinformatic analyses, we reveal the dynamic and coordinated role of neurons and glia in driving the epileptogenic process. Several transcriptomic studies have examined the molecular landscape of epileptogenesis, providing important context for the present work. Bulk RNA-seq analyses of hippocampal and cortical tissue in rodent status epilepticus (SE) models have demonstrated pronounced stage-dependent transcriptional changes, with acute inflammatory and excitability signatures giving way to subacute immune activation and ultimately chronic remodeling programs [25–27]. However, bulk tissue approaches cannot depict the cellular origin of these transcriptional alterations, which is a fundamental limitation that has motivated the development of cell-separation strategies. Fluorescence-activated nuclear sorting (FANS) combined with RNA-seq has been applied to human temporal lobe epilepsy (TLE) surgical specimens, revealing glial pathology including a pathological hybrid GFAP+/OLIG2+ reactive astrocyte subpopulation absent from control tissue and widespread downregulation of mature homeostatic astrocyte programs in the epileptic neocortex [28]. In parallel, single-nucleus RNA-seq (snRNA-seq) has enabled high-resolution, cell-type-resolved profiling of epileptic tissue. In human TLE surgical specimens, large-scale snRNA-seq has identified specific principal neurons and interneuron subtypes bearing the greatest transcriptomic burden [29], while multi-modal CITE-seq has documented extensive pro-inflammatory microglial activation and direct microglia–T cell crosstalk as a hallmark of human epileptic lesions [30]. In rodent epilepsy models, snRNA-seq and spatial transcriptomics have confirmed pan-glial inflammatory activation signaling networks [31]. However, no study to date has applied cell-type-enriched nuclear sorting coupled with RNA sequencing across all three canonical epileptogenic stages: acute, latent, and chronic, within a single experimental framework. The present study addresses this gap directly, providing a temporally resolved, cell-fraction-specific transcriptional atlas. Methods Animals Adult male C57/BL6N mice (Janvier lab), acquired at an age of 8 weeks were acclimatized for 4 weeks in a controlled environment (21–23° C, 12 h dark/light cycles with dark from 7 pm to 7 am). With water and food available ad libitum, mice were housed 1–4 animals per cage. All animal procedures were approved by the Norwegian Food Safety Authority (FOTS: 14198), the Centre for Comparative Medicine at the Oslo University Hospital and the University of Oslo. Mouse epileptogenesis model for MTLE In brief, animals injected with kainic acid (KA) over the right hippocampus, experience a status epilepticus (SE) as described earlier [19, 32]. Three groups of animals were included in the analysis: (a) those who were sacrificed after 1 h (1 h post injection = 1hpi), (b) those who were sacrificed at 24 h (24 h post injection = 24hpi) and those kept for 3 months (3 months post injection = 3mpi). Except from the 1hpi group sacrificed while having SE, the SE usually lasts approximately 4 h (4.4 +- 2.4 h)several days (5+-2.9 days) [19]. The chronic phase then starts adjunct to the latent phase and is characterized by recurrent, spontaneous seizures [19]. In brief, mice were anesthetized with a mixture of medetomidine (0.3 mg/kg, i.p.) and ketamine (40 mg/kg, i.p.). Kainic acid (70 nl, 20 mM, Tocris) or 0.9% NaCl (control group/sham group = SH) were injected with a Hamilton pipette (Hamilton Company, NV) at a depth of 1.7 mm at anteroposterior − 2 mm lateral + 1.5 mm (right) relative to bregma. Anaesthesia was stopped with atipamezole (300 mg/kg, i.p.) adjunct to the completion of operational procedures. Mice were kept on a heating blanket during and after the operation. Mice were monitored closely for 1 h (for the 1hpi-group) and for 4 h (for the 24hpi and 3mpi groups) after KA injection. Animals in the epilepsy group not displaying convulsive seizures corresponding to Racine grade 5 [33] within this time frame were excluded from further analysis. On the model level, 91% of mice developed chronic epilepsy (See Supplementary Material [19]). In our cohort, mice in the 3mpi group were inspected for > 30 min daily. Mice not displaying spontaneous seizures were excluded from downstream analysis. Chronic epilepsy status and seizure observation The intracortical kainic acid model employed here has been extensively characterized and shown to result in chronic epilepsy in approximately 90% of animals, with progressive spontaneous seizures and hippocampal sclerosis, as described in detail by [19]. In the chronic group (3mpi), animals were not subjected to continuous video-EEG monitoring. Implantation of EEG electrodes was deliberately avoided in the chronic cohort to prevent surgery- and implant-induced transcriptional alterations, which could confound interpretation of gene expression changes. As EEG electrodes were not implanted in the acute (1hpi) or early latent (24hpi) groups, introducing implants exclusively in the chronic group would have introduced a systematic experimental asymmetry across time points. We therefore prioritized transcriptomic comparability across disease stages. However, all animals included in the chronic cohort displayed sporadic spontaneous seizures. To minimize potential confounding effects of acute seizure-induced transcriptional changes, animals were closely observed for at least 1 h immediately prior to sacrifice, and none exhibited behavioral seizures during this period (Fig. 1). Fig. 1Experimental design and analytical workflow. Overview over experimental modalities used. Hippocampi from Kainic acid (KA) and NaCl (SH) injected mice were extracted 1 h (1hpi), 24 h (24hpi) and 3 months (3mpi) past injection. Derived nuclei were sorted via flow cytometry into neuronal and glia enriched (“glia”) fractions. Downstream analyses of differentially gene expression (KA vs. SH) were conducted in R using DESeq2 Experimental design and analytical workflow. Overview over experimental modalities used. Hippocampi from Kainic acid (KA) and NaCl (SH) injected mice were extracted 1 h (1hpi), 24 h (24hpi) and 3 months (3mpi) past injection. Derived nuclei were sorted via flow cytometry into neuronal and glia enriched (“glia”) fractions. Downstream analyses of differentially gene expression (KA vs. SH) were conducted in R using DESeq2 Tissue collection and pooling Hippocampi were collected at 1 hpi, 24 hpi, and 3 mpi for the respective groups, immediately shock frozen in liquid nitrogen and stored at −80 °C. Sample collection was performed between 8 and 10 a.m. during the early light phase to minimize circadian variation in gene expression. Per time point (1 hpi and 3 mpi), and per group (kainic acid group, or sham group), right hippocampi from 8 mice were pooled 2 to 1 prior to further processing (n = 4 per group). In the 24hpi group, 8 mice were pooled 4 to 1 and 2 to 1 resulting in sample size n = 3 per group. To ensure sufficient RNA yield from sorted nuclear fractions, hippocampal samples were pooled prior to FANS. Pooling was performed within each experimental group and time point, thereby reducing technical variability related to low-input nuclei while maintaining biological comparability between groups. The biological replicate in all statistical analyses is accordingly the pool, not the individual animal; conclusions should therefore be interpreted at the pool level. Previous studies using similar intracortical kainate protocols have adopted comparable pooling strategies without compromising differential expression patterns. The selected time points were chosen to capture three biologically distinct phases of epileptogenesis: the immediate acute response to status epilepticus (1hpi), the early latent phase associated with inflammatory and cell-death signaling (24hpi), and the chronic phase characterized by long-term network remodeling (3mpi). While intermediate transitions undoubtedly occur, these time points represent established anchors in experimental epileptogenesis and allowed a focused comparison of early stress responses versus delayed inflammatory and remodeling programs. Fluorescent activated nuclear sorting (FANS) Hippocampal nuclei were sorted into NeuN+ (referred to as neurons) and NeuN- (referred to as glia) fractions using a modified version of the protocol published by Jiang et al. [34]. In brief, 1 mL homogenization buffer was added to each sample, followed by homogenization on a GentleMACS dissociator (Miltenyi), and filtering of the homogenate through a 70 μm strainerDebris was removed by density gradient centrifugation and nuclear pellets were resuspended in 100µL incubation buffer/106 nuclei. Anti-NeuN Alexa Fluor488 (Merck Millipore) was added to each sample to a final concentration of 0.1 µg/mL, and samples were incubated for 1 h on ice. Mouse liver was used as a NeuN- control sample and was processed in parallel with hippocampal samples. A FACSAria (BD Biosciences) was used to sort nuclei into NeuN+ and NeuN- fractions. Propidium iodide (PI) was added prior to sorting, and the following gating strategy used (Supp. Figure 3). A nuclear gate was defined by PI-positive events. 2) Aggregated nuclei were excluded in a dot plot using the pulse width of side scatter (SSC-w) versus the pulse area of forward scatter (FSC-a). 3) A NeuN-negative gate was drawn based on signal from anti-NeuN stained liver sample. NeuN + and NeuN- nuclei were sorted into tubes, and nuclei pelleted by centrifugation. Pellets were resuspended in lysis buffer for downstream RNA isolation. A full description of the FANS procedure is given in Supplementary Document. NeuN+/ NeuN− nuclei fractionation and cellular heterogeneity The NeuN⁻ fraction is acknowledged to comprise a heterogeneous mixture of glial and non-neuronal cell types, including astrocytes, microglia, oligodendrocytes, endothelial cells, and pericytes. This approach has previously been described and validated by our group in the context of epilepsy-related transcriptomic studies [32], where NeuN⁻ nuclei robustly captured glia-enriched gene expression patterns relevant to epileptogenesis. In the present study, enrichment of established marker genes for major non-neuronal populations confirmed successful fractionation and supported the interpretation of NeuN⁻-associated inflammatory and remodeling programs as predominantly glia-related. Isolation of RNA from sorted nuclei Lysates were thawed on ice and total RNA extracted with mirVana miRNA Isolation Kit (Ambion), followed by up-concentration using RNA Clean & Concentrator-5 kit (Zymo Research). RNA concentration and integrity was measured on Bioanalyzer with RNA Pico Kit (Agilent Technologies). For further details see S1 Supporting Information. High throughput mRNAseq SMART-Seqv4 Ultra Low InputRNA Kit for Sequencing (Takara Bio) was used to amplify mRNA from total RNA, and the resulting cDNA was used as input in library preparation with ThruPlex DNAseq Kit (Rubicon Genomics). See S1 Supporting Information for details regarding cDNA synthesis and library preparation. Libraries were combined in one pool and sequenced on two S1 flow cells on NovaSeq (100 bp paired end). Computational methods mRNAseq-post processing All analyses were performed using the mouse genome assembly mm10. Only autosomal genes were included. Genomic coordinates for genes, exons and introns were retrieved from the GENCODE comprehensive annotation set. Raw mRNA-seq reads were first trimmed using Trim Galore! (v0.4.3) and subsequently aligned to the reference genome with Rsubread, the R implementation of the Subread aligner [35]. Quality assessment of the resulting BAM files was carried out with Picard’s CollectRnaSeqMetrics. Gene-level quantification was obtained with the featureCounts function in Rsubread, restricting analyses to uniquely aligned reads. Only reads mapping to annotated mRNA regions were retained for downstream processing. All analyses were performed using the mouse genome assembly mm10. Only autosomal genes were included. Genomic coordinates for genes, exons and introns were retrieved from the GENCODE comprehensive annotation set. www.gencodegenes.org/mouse/release_M16.html). Differential expression analysis Differential expression analysis was performed with the DESeq2 package [36], applying a Wald test on a negative binomial generalized linear model fitted independently to the neuronal and glial datasets, comparing KA samples to their corresponding sham controls. In order to ensure compatibility with Entrez-based GO and KEGG databases used in downstream enrichment, genes lacking an official HGNC symbol were removed. We also removed transcripts with fewer than 10 total reads prior to downstream analysis. After size-factor normalization, a standard DESeq2 pipeline was applied, applying a Wald test for pairwise comparisons of KA versus sham within each time point and cell fraction. Because small group sizes can cause DESeq2 Wald test statistics to be deflated relative to the theoretical standard normal null, producing a right-skewed p-value distribution and an overly conservative Benjamini–Hochberg (BH) correction, we applied a two-step FDR procedure. The Wald test statistics were passed to the fdrtool function [37], which estimates the empirical null distribution directly from the observed statistics and returns recalibrated p-values. These were then subjected to BH multiple testing correction to yield a final false discovery rate (fixedFDR). The DESeq2-native adjusted p-values were retained for inspection but not used as a significance criterion; fixedFDR < 0.05 was the sole threshold applied throughout all downstream analyses. Functional enrichment analysis Functional over-representation analyses for Gene Ontology (GO) categories and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways were performed using the goana and kegga functions from the limma R package, applied separately to the upregulated and downregulated DEG lists at each timepoint and for each cell-type fraction. The background gene universe comprised all genes passing the expression pre-filter, represented as Entrez IDs (species: Mm). A term was retained if it contained more than one DEG and had a nominal p-value below 0.05. Quality control Multidimensional scaling To verify the success of the sorting strategy and to detect potential sample outliers, multidimensional scaling (MDS) plots were generated from the mRNA-seq datasets. Multidimensional scaling plots were generated using the plotMDS function from the limma R package, based on the 100 most variable loci, using log2-counts-per-million values [38]. Final graphics were produced using the ggplot2 package [39]. Expression of neuronal and glial genes in NeuN + and NeuN- fraction Normalised expression values of established marker genes were used to confirm cell-type enrichment. Neuronal (Rbfox3), glial (Aldh1l1, Cx3cr1, Mbp), pericytic (Pdgfrb) and endothelial (Pecam1) markers were plotted to demonstrate expected expression patterns in NeuN+ versus NeuN⁻ nuclei. Integrative bioinformatic analysis with selection of relevant GO and KEGG terms Venn diagrams illustrating overlaps in differentially expressed genes were produced using BioVenn [40]. Epilepsy-relevant GO and KEGG categories were selected through manual curation based on established literature and domain expertise. From the full set of enriched terms (Supplementary Table), we chose categories of particular relevance and extracted the corresponding up- and downregulated genes in neurons and glia for visualisation. Figure 4 summarises this integrative, curated analysis. Results To capture temporally resolved, cell-type-specific transcriptional programs across epileptogenesis, we applied fluorescent activated nuclear sorting (FANS) of NeuN+ (neuronal-enriched) and NeuN⁻ (glial-enriched) hippocampal nuclei at three disease stages: 1 h (acute phase), 24 h (early latent phase), and 3 months (chronic phase) after intracortical kainic acid-induced status epilepticus (SE). The experimental workflow is summarized in Fig. 1. Successful separation of neuronal and non-neuronal nuclei was confirmed by multidimensional scaling (MDS) analysis, which clearly distinguished cell type, treatment (KA vs. sham), and time point (Supp. Figure 1A–F). Cell-type enrichment was further validated by marker gene expression: Rbfox3 (NeuN) was selectively enriched in NeuN+ nuclei, while Aldh1l1, Cx3cr1, Mbp, Pdgfrb, and Pecam1 were preferentially expressed in NeuN⁻ nuclei across all time points (Supp. Figure 2). These quality metrics confirm robust and reproducible cell-type separation throughout the dataset. Two outlier samples (37 and 38) identified by MDS were excluded from all downstream analyses (Supp. Figure 1A–B). Stage- and cell-type resolved differential gene expression Following filtering, 23,369 genes were used for downstream analysis. After processing, alignment and filtering, the mRNASeq samples yielded on average 7.5 giga bases (Gb) data aligned to the mouse genome (median = 9.0 Gb; range = 2.4 Gb − 13.0 Gb). The fraction aligning specifically to mRNA regions varied from 16% to 41.2% (median = 28.5), resulting in an average of 2.2 Gb per sample informative for DGE analysis (median = 2.2 Gb; range = 0.4 Gb − 4.3 Gb). To visualise the magnitude and significance of transcriptional changes at each epileptogenesis stage and in each cell fraction, volcano plots are shown in Fig. 2, with the top differentially expressed genes annotated. These plots reveal pronounced directionality shifts across time points and between cell types, with glia showing proportionally more downregulated genes at 1hpi and a strong upregulatory wave at 24 hpi and 3 mpi.Fig. 2 Stage- and cell-type-resolved volcano plots of differential gene expression. Differential gene expression in the NeuN+ (neuronal-enriched; top row, red) and NeuN⁻ (glial-enriched; bottom row, green) fractions at each epileptogenesis stage (KA vs. SH). (A–C) Neuronal fraction at 1hpi (A), 24hpi (B), and 3mpi (C). D–F Glial fraction at 1hpi (D), 24hpi (E), and 3mpi (F). The x-axis shows log2 fold change; the y-axis shows − log10(FDR-adjusted p-value). Colored dots indicate differentially expressed genes (FDR < 0.05); the dashed horizontal line marks the significance threshold. Total numbers of upregulated (↑) and downregulated (↓) genes are indicated in each panel. The differentially expressed genes with highest logFC and lowest FDR are annotated by name. Gray dots represent non-significant genes Stage- and cell-type-resolved volcano plots of differential gene expression. Differential gene expression in the NeuN+ (neuronal-enriched; top row, red) and NeuN⁻ (glial-enriched; bottom row, green) fractions at each epileptogenesis stage (KA vs. SH). (A–C) Neuronal fraction at 1hpi (A), 24hpi (B), and 3mpi (C). D–F Glial fraction at 1hpi (D), 24hpi (E), and 3mpi (F). The x-axis shows log2 fold change; the y-axis shows − log10(FDR-adjusted p-value). Colored dots indicate differentially expressed genes (FDR < 0.05); the dashed horizontal line marks the significance threshold. Total numbers of upregulated (↑) and downregulated (↓) genes are indicated in each panel. The differentially expressed genes with highest logFC and lowest FDR are annotated by name. Gray dots represent non-significant genes Summarising across all time points and cell fractions (Fig. 3A), the number of differentially expressed genes (DEGs) in the glial fraction increased monotonically, whereas neuronal DEGs peaked at 24hpi and declined by 3mpi. At 1hpi, neurons showed 231 DEGs (192 up, 39 down) and glia 192 (124 up, 68 down). At 24hpi, neurons had 561 DEGs (331 up, 230 down) and glia 340 (172 up, 168 down). By 3mpi, neurons showed only 104 DEGs (101 up, 3 down), whereas glia showed 557 (331 up, 226 down). This divergence with neurons peaking acutely and retreating, glia escalating progressively, is a central organisational feature of the dataset and establishes the framework for interpretation of cell-type-specific pathway analyses below. Detailed lists of up- and downregulated genes and associated GO/KEGG terms are provided in Table 1 (1hpi), 2 (24hpi), and 3 (3mpi) and Supplementary Table. Temporal overlap of differentially expressed genes To assess the stage-specificity of transcriptional responses, we examined overlaps in DEGs across time points within each cell fraction (Fig. 3B–C for neurons and glia, respectively; Fig. 3D–F for overlap by time point). The large majority of DEGs were exclusive to a single time point, indicating that each stage of epileptogenesis is characterised by a distinct transcriptional program rather than a persistent response (Fig. 3B–F). In neurons, 60 genes were differentially expressed both 1hpi and 24hpi, 7 genes at both 24hpi and 3mpi, and 4 genes at both 1hpi and 3mpi. Parp3 was differentially expressed at all time points in neurons. In the glial fraction, 11 genes were differentially expressed both at 1hpi and 24hpi and 27 genes at both 24hpi and 3mpi. 10 genes were differentially expressed at both 1hpi and 3mpi. One gene, Tlr1, was differentially expressed at all time points of epileptogenesis. The greater temporal continuity in glia between the 24hpi and 3mpi stages (27 shared genes) compared with the 1hpi–24hpi transition (11 genes) is consistent with the glial inflammatory program initiated at 24hpi persisting and expanding into the chronic phase. Cross-fraction overlap in the sense of genes differentially expressed in both neurons and glia at the same time point, declined progressively from 46 at 1hpi to 60 at 24hpi and only 5 at 3mpi (Fig. 3D–F). This convergence at 1hpi and progressive divergence thereafter reflects the transition from a shared acute stress response to increasingly cell-type-specific programs, particularly a glial-dominant chronic state.Fig. 3Quantitative overview of differential gene expression during epileptogenesis. A Total number of differentially expressed genes (DEGs; FDR < 0.05) in the NeuN+ (neuronal-enriched; red) and NeuN⁻ (glial-enriched; green) fractions at 1 hpi, 24 hpi, and 3 mpi. Upregulated DEGs are shown above zero; downregulated DEGs (hatched bars) below zero. B–C Venn diagrams depicting temporal overlap of DEGs within the neuronal (B) and glial (C) fractions across the three time points. D–F Venn diagrams showing cross-fraction overlap between neurons and glia at 1hpi (D), 24hpi (E), and 3mpi (F). Neurons in red, glia in green, 1hpi in light, 24hpi in moderate and 3mpi in dark tone Quantitative overview of differential gene expression during epileptogenesis. A Total number of differentially expressed genes (DEGs; FDR < 0.05) in the NeuN+ (neuronal-enriched; red) and NeuN⁻ (glial-enriched; green) fractions at 1 hpi, 24 hpi, and 3 mpi. Upregulated DEGs are shown above zero; downregulated DEGs (hatched bars) below zero. B–C Venn diagrams depicting temporal overlap of DEGs within the neuronal (B) and glial (C) fractions across the three time points. D–F Venn diagrams showing cross-fraction overlap between neurons and glia at 1hpi (D), 24hpi (E), and 3mpi (F). Neurons in red, glia in green, 1hpi in light, 24hpi in moderate and 3mpi in dark tone Integrative bioinformatic analysis with relevant GO/KEGG terms at different stages of epileptogenesis To characterise the biological processes underlying stage- and cell-type-specific transcriptional changes, we performed functional enrichment analysis of GO and KEGG terms and selected epilepsy-relevant categories for detailed examination (Fig. 4; Suppl. Table 4). This approach visualises differentially expressed genes within epilepsy relevant pathways in both neurons and glia throughout the stages of epileptogenesis.Fig. 4Epilepsy-relevant differentially expressed genes in neurons and «glia». Gene-level overview of selected epilepsy-relevant differentially expressed genes across NeuN+ (neuronal-enriched) and NeuN⁻ (glial-enriched) fractions at 1 hpi, 24 hpi, and 3 mpi (KA vs. SH). Genes are grouped by functional (GO/KEGG) categories: IEG/transcription factors, synaptic plasticity, ion channels and receptors, COX-2/oxidative stress, cell death and TNF axis, cytokines and chemokines, JAK-STAT/interferon, reactive glia/astrocyte, glial activation, and glial glutamate and Ca2+ axis. Filled circles indicate upregulated genes; open triangles indicate downregulated genes. Dot size reflects statistical significance (− log10(FDR) thresholds: p < 0.05, p < 0.01, p < 10−5, p < 10−10). Log2 fold change is annotated for genes with |log2FC| ≥ 2.8 Epilepsy-relevant differentially expressed genes in neurons and «glia». Gene-level overview of selected epilepsy-relevant differentially expressed genes across NeuN+ (neuronal-enriched) and NeuN⁻ (glial-enriched) fractions at 1 hpi, 24 hpi, and 3 mpi (KA vs. SH). Genes are grouped by functional (GO/KEGG) categories: IEG/transcription factors, synaptic plasticity, ion channels and receptors, COX-2/oxidative stress, cell death and TNF axis, cytokines and chemokines, JAK-STAT/interferon, reactive glia/astrocyte, glial activation, and glial glutamate and Ca2+ axis. Filled circles indicate upregulated genes; open triangles indicate downregulated genes. Dot size reflects statistical significance (− log10(FDR) thresholds: p < 0.05, p < 0.01, p < 10−5, p < 10−10). Log2 fold change is annotated for genes with |log2FC| ≥ 2.8 At 1hpi, both neurons and glia exhibited enrichment for stress- and damage-response programs (heat shock, MAPK/FoxO, DNA repair), while glia showed relative suppression of pro-inflammatory terms, consistent with an early “protective quiescence”. By 24hpi, the enrichment landscape shifted sharply: neurons displayed peak representation of synaptic and signaling pathways (ion-channel regulation, MAPK/mTOR, PI3K-Akt), whereas glia showed strong induction of inflammatory, cytokine, and regulated-cell-death modules, including TNF-, JAK–STAT-, IL-, and necroptosis-related signatures. In specific, at 24hpi, neuronal enrichment of signaling pathways includes voltage-gated calcium channel subunit Cacna1g (log2FC = + 1.56, p = 3.4 × 10⁻⁸), and adrenergic receptor subunits Adra1a and Adra2c (log2FC = + 1.8 and + 1.4 respectively) involved in neuromodulatory ion-channel regulation. These transcripts are absent from the 1hpi neuronal program and are no longer significantly dysregulated at 3mpi (see Tables 1, 2 and 3 and Supplementary Table), consistent with a transient, 24 h-specific neuronal plasticity response. The number of differentially expressed genes associated with GO/KEGG terms in relation to Glutamate (n = 30; “glutamatergic synapse” GO 24hpi, Supplementary Table) peaked at 24hpi in neurons (1hpi neurons “glutamatergic synapse” GO: n = 11, 3mpi neurons, GO “synaptic transmission glutamatergic” n = 2) . These transitions were compatible with the early latent period as a critical inflection point linking acute stress to chronic gliopathy. At 3mpi, neuronal enrichments waned, while glial terms remained dominated by immune, oxidative stress, and glial remodeling processes (complement, cytokine, BBB, and apoptosis categories). Together, the GO/KEGG profiles showed an orderly progression from early shared stress signaling to glia-driven inflammatory maintenance, corroborating the concept that astrocyte and microglial dysfunction progressively assume dominant roles in sustaining epileptogenic networks. Figure 4 visualizes neuronal and glial contribution to epileptogenesis in terms of epilepsy-relevant GO/KEGG terms. For a detailed list of up- and downregulated genes and associated GO/KEGG terms see Suppl. table. Discussion Using cell-type–resolved RNA-seq across the acute, early latent, and chronic stages of epileptogenesis in the intracortical KA model, we delineate a temporally evolving division of reactivity between neurons and glia. Three consistent reactivity patterns emerge: first an early, broadly shared cellular stress response involving both neurons and glia at 1hpi; then at 24hpi a strong inflammatory and cell-death signaling, which was more pronounced in neurons; and finally, a chronic, predominantly glia-mediated adaptation by 3mpi, consonant with durable neuroinflammation, astrocyte/microglia remodeling, and sustained vulnerability. These trajectories integrate well with histopathological and functional studies in KA-elicited MTLE-and human MTLE tissue, particularly the central role of astrocyte dysfunction and uncoupling as potential upstream drivers of network hyperexcitability [19, 41, 42]. The present study builds on and extends several prior transcriptomic investigations of epileptogenesis. Early bulk RNA-seq studies of hippocampal tissue identified neuroinflammatory and synaptic remodeling signatures but lacked the cell-type resolution to attribute these changes to specific cell populations [25, 43]. More targeted approaches, e.g. laser-capture microdissection of hippocampal subfields and sorted glial populations demonstrated that astrocytic and microglial transcriptomes diverge markedly from neuronal ones in the aftermath of seizure activity [26, 27]. Recent snRNA-seq studies of human mesial temporal lobe epilepsy tissue have revealed disease-associated microglial states, reactive astrocyte subpopulations, and oligodendrocyte lineage alterations, yet are inherently limited to the chronic, post-surgical stage and cannot inform the sequence of events during epileptogenesis [28]. Rodent snRNA-seq studies have begun to bridge this gap but remain largely confined to individual time points [31]. A key advance of the present work is the systematic, within-experiment comparison of neuronal and glial transcriptomes across all three canonical epileptogenic stages, which allows direct identification of the glial inflammatory pivot at 24hpi as a stage-specific, cell-type-specific phenomenon rather than an inference from cross-study comparison. Overview of stage- and cell- specific dynamics The number and fraction of commonly differentially expressed genes in both neurons and “glia” at different time points (Fig. 3D-F) and in the respective cellular fraction throughout different stages of epileptogenesis (Fig. 3B-C) is low, revealing a strongly cell and epileptogenesis stage dependent transcriptional profile. Further, the fraction of differentially expressed genes in both neurons and glia decreases over time (46 shared DEGs at 1hpi → 60 at 24hpi → 5 at 3mpi, from Fig. 3D–F), reflecting the transition from a shared acute stress response to increasingly cell-type-specific programs, particularly a glial-dominant chronic state. Differentially expressed genes throughout all stages Only Parp3 (neurons) and Tlr1 (glia) were differentially expressed across all stages (1 hpi, 24 hpi and 3 mpi). Parp3 encodes enzymes that modify nuclear proteins through poly-ADP-ribosylation, essential for DNA repair, apoptosis regulation, and genomic stability. PARP inhibition has shown neuroprotective effects on hippocampal neurons in epilepsy models [44] and PARP1-3 inhibitors are approved for breast cancer treatment [45]. Although mechanisms of PARP-mediated apoptosis remain unclear, recent studies suggest that PARP inhibition may have therapeutic value in seizure-induced hippocampal damage [44]. Tlr1 encodes Toll-like receptor 1 (TLR1), a key mediator of innate immunity and cytokine production. TLR1 dysregulation has been observed early in epileptogenesis in a time- and location-specific manner, indicating a close link to the process [46]. A recent study further highlights the therapeutic potential of TLR1/2 blockade in preventing posttraumatic epilepsy by reducing seizure susceptibility, preserving neurons, and promoting an anti-inflammatory microglial response [47]. 1 h post-injury (acute phase): broad stress and early immune modulation At 1hpi we observe a pan-cellular stress response (heat-shock and damage-response pathways) in both neurons and “glia” (Table 1, Supplementary Table, Figs. 2 and 4). transcriptional response was dominated by immediate-early gene activation consistent with acute excitatory stress. The most strongly upregulated transcripts included Tnf (Neurons: log2FC 5.23, glia: 2.74), Fos (Neurons: log2FC 4.96, Glia: 1.79), Mir212 (Neurons: log2FC 5.74), Nr4a2 (Neurons: log2FC 4.06, glia: 2.40) and Junb (Neurons: log2FC 4.15, Glia: 1.46) and Egr4 (Neurons: log2FC 4.29, Glia: 2.36, see Table 1). Stress-response and DNA-damage-associated transcripts were also among the differentially expressed genes with highest logFC: Gadd45g (Neurons: log2FC 3.99, glia: 1.55) and Btg2 (Neurons: 4.35, glia: 1.29). Notably, Ptgs2 (cyclooxygenase-2, Neurons: log2FC 3.14, Glia: 1.61), was strongly upregulated, reflecting early prostaglandin-mediated neuroinflammatory signaling. Concurrently, we observe immune pathway dampening in “glia” (e.g., down-tuning of inflammasome/TNF-related components in our data: Glia log2FC: Tnf − 2.23, Traf5 -0.59). Such transient “protective quiescence” has been described in acute CNS insults where early suppression of pro-inflammatory outputs may temporarily limit bystander damage. The prominence of DNA-repair/apoptosis-regulation transcripts in neurons (e.g., Parp3, the most consistently dysregulated neuronal transcript across all stages; Gadd45g, log2FC 3.99; Btg2, log2FC 4.35; Ppp1r15a, log2FC 2.65 at 1hpi, all fixedFDR < 0.001), is consistent with reports that PARP activity modulates neuronal survival after seizures [48]. In parallel, our findings align with the broader concept that astrocyte and microglial activation begin rapidly, yet its initial phenotype can be contextually protective. Early microglial activation states with heightened purinergic signaling after SE were described earlier, illustrating that microglia swiftly sense network distress [49]. In line with this, early-stage (1hpi) expression also points to anti-apoptotic/neuroprotective programs across cell types, including glial downregulation of Nlrp3 (log2FC -0.83, p: 0.005), a sensor mediating caspase-1–dependent release of IL-1β and IL-18 [50]. Both astrocytes and microglia contribute to this pathway. Pharmacological inhibition of NLRP3 has shown protective effects in preclinical models of neuroinflammation [51–53] and epilepsy, reducing seizure susceptibility, neuronal loss, and neurobehavioral impairments [54, 55]. 24 h post-injury (early latent period): induction of necroptosis-linked and cytokine signaling in glia, and immune pathway recruitment in neurons At 24hpi we see a qualitative shift: now neurons display increased numbers of differentially expressed genes overall, including immune-pathway and synaptic-plasticity genes (Table 2, Supplementary Table, Figs. 2 and 4). In specific, in neurons at 24hpi, we 53 genes upregulated that are associated with GO term “immune system process” (e.g. Fos log2FC: 2.18, Csf1 1.15, Cdkn1a 2.87), compared to 29 genes at 1hpi and 15 at 3mpi. Regarding synaptic plasticity, we find 13 genes associated with this GO term upregulated at 24hpi (e.g. mir-132 log2FC: 3.25, Bdnf 1.66) compared to 9 at 1hpi and none at 3mpi. This pattern is compatible with necroptosis-associated signaling, a pathway that can be triggered downstream of TNFR1 and IFN signaling via RIPK1/RIPK3/MLKL, and that directly promotes release of alarmins such as IL-33 [56–58]. In contrast, glia exhibit cohesive inflammatory and regulated-cell-death signatures (e.g., TNFR1-axis components like Tnfrsf1a log2FC: 1.32, Tnfrsf12a 2.66, STAT-ISGF3-related transcripts like Socs3 log2FC: 3.63, Stat2 1.14 and IL-family mediators like Il11 log2FC: 3.94, Il33 2.01, Il4ra 1.95). Notably necroptosis/autophagy involvement in early epileptogenesis (e.g. Camk2d log2FC: 0.70, Stat2 1.14, Tnfrsf1a 1.32 and Il33 2.01) has previously been demonstrated in the KA model used here, with astrocyte loss in CA1 shortly after SE and molecular evidence for necroptosis [59]. Concurrently, COX-2 and chemokine pathways rise to prominence in glia (e.g. Ptgs2 log2FC: 2.53, Sphk1 3.88, Ccl2 3.53, Ccl12 3.11). The COX-2 axis is a well-established early amplifier of post-seizure inflammation and contributes to excitability and neuronal vulnerability; CCL2 has previously been shown to be upregulated in patients with pharmacoresistant epilepsy and postulated to be a mediator of inflammation induced seizure activity [60]. As previously depicted, microglia–astrocyte crosstalk likely repolarizes astrocytes toward a neurotoxic (A1-like) profile during the latent phase. Classically activated microglia (via IL-1α, TNF, and C1q) can induce A1 astrocytes that lose supportive functions and secrete neurotoxic factors, an interaction repeatedly implicated across CNS injuries and relevant to epileptogenesis [49]. These mechanisms dovetail with astrocyte gap-junction uncoupling and impaired glutamate/K+ homeostasis shown to be causal in MTLE, situating our 24hpi expression profile within a coherent pathophysiological chain from glial signaling to network instability [19]. The early latent period appears to be a therapeutic opportunity: targeted inhibition of the NLRP3 inflammasome after SE has earlier been shown to reduce neuronal damage, seizure susceptibility, and chronic inflammation, illustrating that post-insult immunomodulation can alter epileptogenic trajectories [61]. 3 months post-injury (chronic phase): glia-dominated inflammation and remodeling By 3mpi, our dataset shows a marked asymmetry: relatively few neuronal differentially expressed genes persist, whereas glial differentially expressed genes predominantly comprised inflammatory, immune-surveillance, and cell-death GO terms. This mirrors long-term histopathology in MTLE, such as progressive neuronal loss, widespread reactive astrogliosis, microglial activation, and altered astrocytic coupling, and matches observations that astrocyte dysfunction can sustain epileptic networks even after acute neuronal stress has abated. Mechanistically, sustained microglia-to-astrocyte signaling (IL-1α/TNF/C1q), COX-2/chemokine networks (e.g., CXCL10), and necroptosis-linked alarmins (e.g., IL-33) can perpetuate feed-forward inflammatory loops and compromise astrocytic control of extracellular K+ and glutamate, lowering seizure thresholds over months [48]. Together, our chronic-phase profile is consistent with a glia-centric homeostatic failure that maintains epileptogenicity. Therapeutical opportunities Our acute-latent-chronic sequence of changes after SE provides molecular continuity between early microglial activation with purinergic sensitivity, astrocyte death/uncoupling and altered network support in the hippocampus, and long-lasting glial inflammation. Importantly, the specific 24hpi rise of necroptosis- and cytokine-axis associated transcripts in glia positions the early latent phase as a mechanistic bridge from early stress to chronic gliopathy, precisely the window in which several anti-inflammatory or cell death-pathway interventions (e.g., NLRP3 inhibition) have shown disease-modifying promise in preclinical models [55]. The early latent phase (around 24hpi) concentrates actionable biology: IFN/TNFR1-RIPK signaling, COX-2/chemokines, and microglia-astrocyte induction, before rather less reversible remodeling dominates. Intervening here is likely superior to chronically suppressing seizures without altering disease course. Further, our results reinforce glia-targeted therapies (astrocyte coupling support, microglial polarization control, inflammasome blockade) as rational antiepileptogenic strategies, complementary to neuron-centric antiseizure approaches. At last, our data provide changes in transcripts potentially affecting proteins within the TNFR1/IFN-RIPK-MLKL and COX-2/CXCL10 axes meriting evaluation as surrogate readouts of latent-phase engagement and as predictors of long-term outcome in clinical translation. Limitations The following methodological considerations should be considered when interpreting these data Continuous video-EEG monitoring was not performed. Implant-based EEG would not be feasible in the 1hpi cohort and would be of limited interpretability in the 24hpi cohort; moreover, restricting EEG recordings to the chronic (3mpi) group alone would have introduced a systematic confound, as only one disease stage would have been exposed to surgical implantation and a long-term recording device. Given the known effects of surgery and implanted devices on brain inflammatory and transcriptional programs, the experimental design prioritized methodological comparability across stages and avoided implant- or surgery-induced transcriptional confounders. Seizure burden could therefore not be quantified; however, all animals included in this study developed chronic epilepsy in this well-established model. The choice of time points (1hpi, 24hpi, 3mpi) is intended to reflect distinct stages of epileptogenesis (acute, early latent and chronic). Based on the prior, extended characterization of this epilepsy model (see supplementary material [19]), our own observations and the fact of distinct transcriptional profiles at the respective time points (see Figs. 3B-F and 4), these time points represent an adequate approximation of the respective stages of epileptogenesis. However, 24hpi reflects an rather early time point of the latent phase and derived transcriptional data may overrepresent mechanisms priorly shown to be related to the acute period of epileptogenesis as neuroinflammation and blood-brain-barrier dysfunction [62]. To minimize potential acute seizure-related effects, animals were observed prior to tissue collection and showed no overt seizures during this period. Transcriptomic analyses were performed on pooled hippocampal samples following nuclear sorting. Pooling was required to obtain sufficient RNA yield for reliable downstream analyses and represented the only feasible approach for this experimental design. Although pooling precludes assessment of inter-individual variability, quality-control analyses demonstrated robust cell-type separation and consistent stage-dependent transcriptional patterns across samples. The NeuN⁻ nuclear fraction comprises a heterogeneous population of non-neuronal cell types, predominantly astrocytes and oligodendrocyte-lineage cells, with smaller contributions from microglia, endothelial cells, and other vascular-associated populations. Based on prior nuclear-sorting and single-nucleus transcriptomic studies, astrocytes and oligodendrocytes together typically account for the majority of NeuN⁻ nuclei (approximately 70–80%), whereas microglial and vascular-associated cells generally represent smaller fractions (typically < 10–15% each). More refined separation of individual glial and vascular subtypes across multiple stages of epileptogenesis is currently limited by technical constraints. Nevertheless, this approach reliably captures global non-neuronal inflammatory and tissue-remodeling programs that are central to epileptogenic processes. Conclusion Our cell-type-resolved analysis across epileptogenesis reveals an orderly reassignment of molecular responsibility: from a shared, protective-leaning stress program (1hpi), through a glial inflammatory/regulated cell-death pivot (24hpi), to glia-dominated chronic inflammation and remodeling (3mpi). These results support the concept that astrocyte and microglial dysfunction are not mere epiphenomena but active contributors to the progression of MTLE, and they identify the latent window as a treatment-sensitive phase. Interventions that curb necroptosis/inflammasome activity, preserve astrocytic coupling, and temper microglia-to-astrocyte neurotoxic signaling may therefore represent promising antiepileptogenic strategies deserving further experimental validation. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Table 1 Differentially expressed genes in neurons and glia at 1 h post injection (1hpi) Supplementary Material 4 Table 2 Differentially expressed genes in neurons and glia at 24 h post injection (24hpi) Supplementary Material 5 Table 3 Differentially expressed genes in neurons and glia at 3 months post injection (3mpi) Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Table 1 Differentially expressed genes in neurons and glia at 1 h post injection (1hpi) Supplementary Material 4 Table 2 Differentially expressed genes in neurons and glia at 24 h post injection (24hpi) Supplementary Material 5 Table 3 Differentially expressed genes in neurons and glia at 3 months post injection (3mpi)