Inhibition of endocannabinoid degradation in astrocytes reprograms glial reactivity and prevents seizure sequelae Authors: Chudai Zeng, Fei Gao, Mei Hu, Jian Zhang, Dexiao Zhu, Li Sun, Jianlu Lyu, Mingzhe Pan, Chu Chen Published: 2026 ABSTRACT AbstractBackgroundTemporal lobe epilepsy (TLE) is the most common form of focal epilepsy and is characterized by a pathological cascade of excitotoxicity that leads to neuroinflammation, progressive neuronal loss, and subsequent cognitive decline. Despite its prevalence, effective therapies remain lacking. Previous studies have demonstrated that the dysregulation of the endocannabinoid system contributes to epileptic activity. In particular, inactivation of monoacylglycerol lipase (MAGL), the key rate-limiting enzyme responsible for the degradation of the endocannabinoid 2-arachidonoylglycerol (2-AG), an endogenous lipid mediator with anti-inflammatory and neuroprotective properties, suppresses seizures and reduces neuroinflammation. However, the cellular and molecular mechanisms underlying these protective effects remain unclear.MethodsTo dissect the cellular mechanisms underlying MAGL-mediated neuroprotection, we employed a kainic acid (KA)-induced status epilepticus model in mice with global, astrocyte-specific (aKO), and neuron-specific (nKO) deletion of mgll. We combined single-nucleus RNA sequencing (snRNA-seq) to map the transcriptomic landscape of glial responses with pharmacological interventions and AAV-mediated gene manipulation to validate key signaling pathways, as well as behavioral assays to assess functional recovery.ResultsWe demonstrated that astrocyte-specific mgll deletion attenuated seizure susceptibility and hippocampal neuroinflammation, whereas neuron-specific, mgll deletion did not reproduce this broader protective phenotype. Transcriptomic profiling revealed that astrocytic MAGL deficiency fundamentally reshaped the glial response to injury by preventing the transition to pro-inflammatory reactive astrocyte states and suppressing the activation of disease-associated microglia (DAM). Mechanistically, we identified a signaling pathway in which the neuroprotective effects of MAGL inhibition depend on cannabinoid receptor 1 (CB1) activation and are mediated by downstream peroxisome proliferator-activated receptor γ (PPARγ) signaling. Genetic deletion of CB1 abolished the protective effects, whereas pharmacological blockade or AAV-mediated knockdown of PPARγ attenuated these effects. Furthermore, aKO mice exhibited reduced neuronal loss, preserved synaptic structural integrity and protection against post-seizure cognitive deficits.ConclusionThese findings reveal astrocytic MAGL as a crucial regulatory node after status epilepticus and support a model in which CB1-dependent mechanisms and astrocytic PPARγ-dependent regulation jointly contribute to 2-AG-mediated neuroprotection, attenuating neuroinflammation, preserving synaptic integrity, and reducing post-seizure behavioral deficits in this KA-induced SE model.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12974-026-03883-3.Keywords: Epilepsy, 2-arachidonylglycerol, Monoacylglycerol lipase, Cannabinoid receptor, Astrocyte, Neuroinflammation Introduction Temporal lobe epilepsy (TLE), particularly acquired forms, is characterized not only by recurrent spontaneous seizures but also by a devastating sequence of events often initiated by a precipitating insult, such as status epilepticus (SE) [1]. This initial insult triggers acute excitotoxicity, leading to a cascade of neuroinflammation and progressive neuronal loss [2]. Although many antiseizure medications are available, nearly 30–40% of patients do not respond to common drugs, leading to drug-resistant epilepsy. While current therapies often target neuronal excitability, they fail to address the underlying mechanisms that drive seizure recurrence and disease progression [3]. In particular, inflammatory responses orchestrated by non-neuronal cells are increasingly viewed as a pivotal determinant of the long-term fate of the injured brain. Understanding how to modulate these glial-driven inflammatory cascades represents a promising avenue for neuroprotective intervention during the post-SE injury. The endocannabinoid (eCB) system has emerged as a potent brake on this pathological progression, integrating synaptic activity with inflammatory and metabolic signaling. This neuromodulatory system comprises endogenous cannabinoids (primarily 2-arachidonoylglycerol [2-AG] and anandamide [AEA]), cannabinoid receptors (CB1 and CB2), and the enzymes responsible for endocannabinoid synthesis and degradation [4, 5]. Upon neuronal depolarization and calcium influx, 2-AG, the most abundant endocannabinoid in the brain [5], is synthesized on-demand from membrane phospholipids by diacylglycerol lipase-α (DAGLα), released into the synaptic cleft, and travels retrogradely to activate presynaptic cannabinoid type 1 receptors (CB1R), thereby reducing glutamate release and providing a critical negative feedback mechanism against hyperexcitability [6]. In experimental epilepsy models, eCB signaling has been implicated in limiting excitotoxicity and dampening network hyperactivity [7, 8]. Moreover, clinical success of cannabidiol in treating drug-resistant epilepsies has renewed interest in targeting the eCB system for epilepsy treatment [9]. Though the mechanism of cannabidiol in epilepsy is multifactorial, involving additional targets such as TRPV1, GPR55, T-type calcium channels, and adenosine receptors [10]. Critically, the eCB system serves as a key interface between neuronal activity and inflammatory signaling. While increased 2-AG availability can enhance cannabinoid receptor-dependent protective signaling, including CB1-mediated suppression of excitatory neurotransmission and CB2-related modulation of immune responses, its hydrolysis releases arachidonic acid (AA), a precursor for cyclooxygenase (COX)-mediated production of proinflammatory prostaglandins (particularly PGE2) [5, 11–13]. This dual nature of 2-AG metabolism underscores its complex involvement in both seizure control and neuroinflammation. Monoacylglycerol lipase (MAGL; gene name: Mgll) is the primary enzyme responsible for degrading 2-arachidonoylglycerol (2-AG) in the brain. In various epilepsy models, global inhibition of Mgll, which enhances 2-AG signaling while reducing its downstream metabolites, has been shown to suppress seizure activity [7, 14–16] and reduce neuroinflammation [16]. These findings collectively establish the therapeutic potential of targeting MAGL to enhance endocannabinoid tone. However, the key unresolved question is not simply whether MAGL inhibition is beneficial, but which cellular compartment mediates this protection. Global MAGL inhibition affects neurons, astrocytes, and other brain cell types simultaneously, making it difficult to determine whether its beneficial effects arise primarily from neuronal control of synaptic transmission, astrocytic regulation of inflammatory lipid metabolism, or coordinated interactions between these compartments. This distinction is particularly important because neurons and astrocytes play anatomically and functionally distinct roles in hippocampal circuits. Neuronal 2-AG signaling selectively targets CB1R-positive presynaptic terminals but not astrocytic CB1R [17], whereas astrocytic MAGL has been implicated in the metabolism of 2-AG-derived lipid mediators and the regulation of inflammatory responses. Prior work has demonstrated cell type-specific roles of MAGL in other brain regions and physiological contexts [18, 19], but it remains unclear how this division of 2-AG metabolism operates in the hippocampus after SE, where excitotoxic stress, astrocyte reactivity, microglial activation, and selective neuronal vulnerability converge. Defining the cell type-specific contribution of MAGL is therefore essential for interpreting previous pharmacological studies and for developing more targeted strategies that preserve beneficial endocannabinoid signaling while minimizing the limitations of broad, cell-indiscriminate pathway manipulation. In this study, we dissected the cell-specific role of 2-AG signaling in altering the trajectory of KA-induced post-seizure injury. We report that astrocytic MAGL deletion suppresses neuroinflammation in the subacute phase and provides sustained protection into the chronic phase, whereas neuronal deletion offers no such benefit. Mechanistically, these protective effects are mediated via CB1 and PPARγ signaling pathways, establishing astrocytic metabolic reprogramming as a key strategy to suppress post-seizure injury. Materials and methods Animals Mgllflox/flox animals were generated by the Texas A&M Institute for Genomic Medicine (TIGM), using a targeting strategy, as described previously [20]. Briefly, the targeting vector was introduced into the C57BL/6N ES cell line (JM8) via homologous recombination to flank Exon 2 with LoxP sites. To validate the conditional strategy, global Mgll knockout (MAGL KO) mice were generated by crossing Mgllflox/flox mice with the Sox2-Cre line (Tg(Sox2-cre)1Amc/J, JAX #004783). Correct targeting and recombination were verified using long-distance PCR and sequencing. Cell-type-specific knockouts were generated as follows: Neuronal (nKO): Mgllflox/flox crossed with Syn1-Cre (JAX #003966). Astrocytic (aKO): mgllflox/flox crossed with GFAP-Cre (JAX #024098). For mechanistic validation, global Cannabinoid Receptor 1 knockout mice (Cnr1−/−, CB1-KO) were also utilized. Experimental cohorts included both male and female mice (10-12 weeks of age). Animals from each genotype were randomly assigned to either vehicle or kainic acid (KA) treatment groups. All mice were housed in a temperature-controlled facility on a 12-hour light/dark cycle with ad libitum access to food and water. All animal studies were performed in compliance with the US Department of Health and Human Services Guide for the Care and Use of Laboratory Animals, and the care and use of the animals reported in this study were approved by the Institutional Animal Care and Use Committee of University of Texas Health San Antonio. Kainic acid-induced status epilepticus model To induce status epilepticus (SE), mice received a single intraperitoneal (i.p.) injection of 17.5 mg/kg kainic acid (KA; HB0355, Hellobio) dissolved in sterile saline. Control mice received an equivalent volume of saline vehicle. Following injection, mice were monitored by a blinded observer continuously for 2 h to evaluate seizure severity using the modified Racine scale: Stage 1, immobility and staring; Stage 2, head nodding; Stage 3, unilateral forelimb clonus; Stage 4, bilateral forelimb clonus and rearing; Stage 5, rearing and falling; Stage 6, generalized tonic-clonic seizures with falling; Stage 7, seizures leading to death. During this monitoring period, seizure incidence was defined as the occurrence of seizure of Stage 4 or higher more than twice. The seizure assessment was based on behavioral severity and does not reflect the full electrographic seizure burden. Pharmacological interventions For pharmacological inhibition of MAGL, CB1R knockout mice or nKO mice were intraperitoneally injected with the selective inhibitor 4-Nitrophenyl-4-[bis (1,3-benzodioxol-5-yl)(hydroxy)methyl]piperidine-1-carboxylate (JZL-184, 10 mg/kg) or vehicle (DMSO) every other day for 3 injections. JZL-184 was prepared and dissolved in a vehicle containing Tween 80 (10%), DMSO (10%) and saline (80%) as described previously [21]. To investigate the role of PPARγ, the antagonist GW9662 (5 mg/kg, i.p.; Cayman Chemical) was administered 30 min prior to JZL-184 treatment [22]. For delayed JZL-184 treatment, WT mice were randomly assigned to JZL-184 or DMSO groups and administered 17.5 mg/kg KA. After a 2-hour post-injection observation period, mice received 10 mg/kg JZL-184 or DMSO. Additional doses were administered on Days 1, 3, and 5, and tissues were collected on Day 6. Mice were randomly assigned to treatment groups and monitored for seizure susceptibility as described above. In all experiments, sex was balanced across groups. When group sizes were even, equal numbers of male and female mice were used; when odd, sex distribution was matched across groups within each experiment. Behavioral assaysrains were post-fixed, cryoprotected in Open Field Test (OFT) Hyperactive and anxiety-like behavior was assessed in an open field apparatus (30 × 30 cm) as described previously [20, 23]. Mice were placed in the center of the arena and allowed to explore freely for 30 min. Movement was tracked using an automated video tracking system (EthoVision, Noldus, version 17). Total distance traveled, mean activity, and time spent freezing were recorded. Novel Object Recognition (NOR) The NOR test was conducted in a 30 × 30 × 30 cm square open field as described previously [20, 24]. Briefly, the NOR test consisted of three phases: habituation, training, and testing. Mice first underwent a 10-minute habituation session in the empty arena. The NOR assay then comprised two 10-minute stages separated by a 4-hour intertrial interval. In the training stage, animals were presented with two identical objects. In the test stage, animals were presented with one familiar object (from training) and one novel object. Exploration was defined as the frequency with the head oriented toward and within 2 cm of an object. Object exploration during the test stage was quantified using the EthoVision video-tracking system (Noldus, version 17). The recognition index (RI) was calculated based on the following equation: RI = FN/(FN + FF), where FN is the frequency devoted to the novel object and FF is the frequency for the familiar object. Tissue processing and immunofluorescence Animals were anesthetized with ketamine/xylazine (200/10 mg/kg) and transcardially perfused with ice-cold PBS followed by 4% paraformaldehyde (PFA), as described previously [20, 25, 26]. Brains were post-fixed, cryoprotected in 20% sucrose, and sectioned at 25 μm using a cryostat. Sections were washed in PBS and blocked in 5% normal goat serum in PBS containing 0.3% Triton and 1% BSA for 2 h at room temperature. For PSD-95 staining, antigen retrieval was performed in citrate buffer (pH 6.0) at 95 °C for 10 min prior to blocking. For PPARγ staining, sections were permeabilized with 0.5% Triton X-100 for 40 min before blocking.). Sections were then incubated overnight at 4 °C with primary antibodies, including anti-GFAP (1:1000, Cat# G3893, MilliporeSigma; 1:10000, Cat# AB4674, Abcam), anti-Iba1 (1:1000, Cat# MABN92, MilliporeSigma), anti-NeuN (1:1000, Cat# Ab177487, Abcam), anti-PSD-95 (1:500, Cat# 124011, SYSY synaptic systems), anti-CB1 (1:1000, Cat# 10006590, Cayman), anti-PPARγ (1:300, Cat# 2435, Cell Signaling) and anti-GFP (1:1000, Cat# ab290, Abcam). After washing, sections were incubated with appropriate fluorophore-conjugated secondary antibodies (1:1000) for 1 h at room temperature, followed by nuclear counterstaining with 4-6-Diamidino-2-phenylindole (DAPI). Images were acquired using a Zeiss Imager II deconvolution microscope with SlideBook 2024 software. To ensure representative sampling, 3–4 sections per mouse were selected at evenly spaced rostrocaudal levels spanning the hippocampus. For each animal, values from all analyzed sections were averaged to generate a single data point. Images were acquired as z-stacks (1 μm step size for 20× imaging and 0.5 μm step size for 63× imaging), and maximum intensity projections were generated prior to analysis. Quantification was performed using SlideBook 2024 software. GFAP staining was imaged using a 20× objective in the DG, CA1 and CA3. Mean fluorescence intensity was quantified within defined regions of interest. Iba1 staining was imaged using a 20× objective in the DG, CA1 and CA3. Iba1-positive cells were identified using adaptive threshold-based masking followed by automated segmentation, and cell counts were normalized to the analyzed area. PSD-95 puncta were imaged using a 63× objective in the stratum radiatum of the CA1 region. For each section, three fields of view were acquired, and three sections per mouse were analyzed. Puncta were detected using adaptive thresholding and particle segmentation with defined size constraints (0.05–1.0 μm²), and results were expressed as puncta density and normalized to vehicle group in each genotype. CB1 staining was imaged using a 20× objective in the CA1 region, and mean fluorescence intensity was quantified. NeuN staining was imaged using a 20× objective for the four-genotype comparison at Day 6 and a 63× objective for WT versus aKO comparisons. At Day 6, mean fluorescence intensity was quantified in the CA1 pyramidal cell layer. At the chronic timepoint, NeuN stained area was measured. For PPARγ analysis, images were acquired using a 63× objective. All GFP-positive astrocytes from each image were identified, and their nuclei were manually outlined. At least 30 nuclei per mouse were analyzed. Nuclear PPARγ signal was quantified as integrated fluorescence intensity per nucleus. Due to potential variability in staining and imaging conditions across independent experiments, values were normalized to the means of the corresponding control group within each experiment to minimize inter-experimental variability. Histochemistry Degenerated neurons were detected using Fluoro-Jade C (FJC), which is an anionic dye that specifically stains the soma and neurites of degenerating neurons and thus is unique as a neurodegenerative marker. As described previously [20, 26, 27], coronal brain Sect. (25 μm thickness) were prepared using a cryostat and mounted onto gelatin-coated slides. Sections were incubated in the solution with FJC (0.0001% solution, Cat# TR-160-FJC, Biosensis) for 10 min, followed by 3 × 1-min wash with distilled water. Slices were dried naturally at room temperature without light. To ensure representative sampling of the entire hippocampus, four sections per mouse were selected at evenly spaced rostrocaudal intervals spanning the dorsal to ventral hippocampus. For each section, images of the hippocampus were acquired using a SlideBook 2024-controlled Zeiss deconvolution microscope under identical acquisition settings across all groups. Quantification of FJC puncta was performed using SlideBook 2024 software. Briefly, a fluorescence intensity threshold was applied to generate a mask of FJC-positive signal, followed by automated object segmentation to identify discrete puncta/cell bodies. The number of FJC-positive cells was quantified within the hippocampus and normalized per section. For each animal, values from the four sections were averaged to obtain a single representative value. Quantitative real-time PCR (qPCR) Total RNA was extracted from harvested tissues with the RNeasy Mini Kit (Qiagen) and treated with RNase-free DNase (Qiagen), following the manufacturer’s instructions. The RNA concentration was measured by a Microvolume UV-Vis Spectrophotometer (NanoDrop One, ThermoFisher Scientific), and RNA integrity was verified by electrophoresis in a 1% agarose gel. Reverse transcription was performed using the iScript cDNA synthesis kit (BioRad) with 1 µg of total RNA, with 4 µl of 5× iscript reaction mix, and 1 µl of iscript reverse transcriptase in a final volume of 20 µl. Samples were incubated at 25 °C for 5 min, followed by 42 °C for 30 min, and reactions were stopped by heating to 85 °C for 5 min. Real-time RT-PCR primers specific for IL-1β, IL-6, TNFα, and GAPDH were designed using Beacon Designer Software (BioRad) and synthesized by IDT (Coralville, IA). Primer sequences are listed in Table S1. Reactions were performed in duplicate in a 25 µL total volume containing 12.5 µL of 2× iQ SYBR Green Supermix (Bio-Rad), 5 µL of 1:10 diluted cDNA template, and 400 nM of each primer. The PCR cycling protocol was as follows: 95 °C for 3 min, followed by 45 cycles of 95 °C for 30 s, 58 °C for 45 s, and 95 °C for 1 min. A melt-curve analysis was conducted at the end of each run to confirm amplification specificity. Amplicon sizes were further verified by electrophoresis on a 3% agarose gel. Amplification and analysis were performed using the iCycler iQ Multicolor Real-Time PCR Detection System (Bio-Rad). Relative gene expression was quantified using the 2-ΔΔCT method, with β-actin as the internal control as its expression was not affected by our experimental treatments. ΔCT was calculated as (CT of gene of interest) - (CT of β-actin), and ΔΔCT as (ΔCT of treated sample) - (ΔCT of control), as previously described [20, 25, 26, 28, 29]. Technical replicates were performed for each detected gene to ensure measurement accuracy and reproducibility. Public bulk RNA sequencing dataset analysis To evaluate the endogenous transcriptional dynamics of Mgll relative to neuroinflammation, we re-analyzed publicly available bulk RNA-sequencing datasets obtained from the NCBI Gene Expression Omnibus (GEO). We utilized dataset GSE213393, comprising transcriptomic profiles from the mouse hippocampal dentate gyrus (DG) following kainic acid (KA) or saline injection for 3, 7, and 14 days. Additionally, dataset GSE99577 was employed to examine temporal gene expression changes in the CA1 region across a time course of 6 h to 12 days post-injection. Normalized expression data were retrieved and processed using R. For robust comparison across samples, raw expression values for each gene were Z-score normalized. To quantify neuroinflammation, a composite neuroinflammation score was derived by calculating the mean of the Z-normalized expression values for a defined panel of canonical astrocytic markers, microglial markers, cytokines, and complement system. Temporal dynamics and correlations between mgll and this inflammation score were visualized using time-course line plots and heatmaps. Single-nucleus RNA sequencing Sample preparation, library construction, and RNA sequencing Hippocampal tissues from WT and aKO mice treated with Veh or KA for single-nucleus RNA sequencing were harvested and stored in liquid nitrogen as previously described [30–32]. Single nucleus sample preparation, library construction, and RNA sequencing were outsourced to Novogene Corporation Inc. (Sacramento, CA). Single nucleus RNA sequencing libraries were prepared using the Chromium Next GEM Single Cell 3′ Gel Bead v4 Kit (10x Genomics) and sequencing was performed on the Illumina NovaSeq6000 system at Novogene. RNA-seq data processing and analysis Sequencing data were processed using CellBender (v4.0) to remove ambient RNA profiles. Initial filtering was performed using emptyDrops to retain droplets with an FDR < 0.01. Strict quality control was applied to remove low-quality nuclei: cells were excluded if they exhibited > 15% mitochondrial reads or > 1% hemoglobin reads. To reduce contamination from ambient myelin debris, cells with > 3% myelin-associated reads were also excluded. Nuclei were retained if they expressed between 400 and 12,000 genes and had > 700 unique molecular identifiers (UMIs). For doublet identification, scDblFinder was utilized to detect and remove artificial doublets [33]; clusters exhibiting high mitochondrial content (> 95th percentile within the cluster) were further filtered to ensure dataset purity. Normalization and variance stabilization were performed using SCTransform (v2), regressing out the percentage of mitochondrial reads. 3,000 highly variable genes were selected for downstream analysis. Dimensionality reduction was performed using t-distributed Stochastic Neighbor Embedding (tSNE) on the first 30 principal components. Graph-based clustering was performed at a resolution of 1.0. Integration of samples was achieved using Seurat’s SCT-based integration workflow to correct for batch effects and align shared biological states across conditions. Cell-type annotation Mouse hippocampal cell types were annotated manually using established gene markers as listed in Fig. 3B. The following markers were used to define major cell classes: Excitatory neurons: Camk2a, Slc17a7, Tbr1; Inhibitory neurons: Gad1, Gad2, Sst, Vip, Pvalb; Immature neurons: Dcx, Neurod1, Tubb3; Astrocytes: Gfap, Aqp4, Aldh1l1, Slc1a3; Oligodendrocytes: Mog, Mbp, Plp1; Oligodendrocyte precursor cells (OPCs): Pdgfra, Cspg4, Sox10; Microglia: Cx3cr1, P2ry12, Tmem119, Aif1; Endothelial cells: Pecam1, Cldn5, Flt1; Pericytes: Pdgfrb, Rgs5, Vtn; Fibroblasts: Col1a1, Col1a2, Dcn; Ependymal cells: Foxj1, Tmem212; Choroid plexus epithelial cells: Ttr, Aqp1. Excitatory neuron subtypes (e.g., CA1, CA3, DG granular) were further resolved based on cluster-specific enrichment and spatial identity inferred from cluster topology. Single-nucleus RNA sequencing reveals distinct cell type-specific transcriptional responses to KA in aKO mice. (A) t-Distributed Stochastic Neighbor Embedding (t-SNE) visualization of nuclei isolated from the hippocampus, clustered into 12 distinct cell types. (B) Dot plot showing the expression of canonical marker genes used to identify major cell types. Dot size represents the percentage of cells expressing the gene, and color intensity represents the average expression level. (C) Feature plots displaying the expression of specific marker genes used to define neuronal subpopulations. (D) t-SNE plot showing the sub-clustering of neuronal populations into Dentate Gyrus (DG) granular cells, CA1 pyramidal neurons, CA3 pyramidal neurons, Subiculum, Inhibitory neurons, and Cajal-Retzius cells. (E) Heatmaps showing the relative expression of differentially expressed genes (DEGs) across experimental groups (WT-Veh, WT-KA, aKO-Veh, aKO-KA) in DG granular cells, CA1 pyramidal neurons, CA3 pyramidal neurons. (F) Scatter plots comparing the transcriptional response to KA (log2 Fold Change) between WT (x-axis) and aKO (y-axis) mice in DG granular cells, CA1 pyramidal neurons, astrocytes, and microglia. Points are colored based on their response pattern. Pearson correlation coefficients (R) are indicated for each cell type. (G) Gene Ontology (GO) enrichment analysis of pathways significantly upregulated in DG in both WT and aKO mice (Shared pathways Up) following KA treatment. Dot size indicates gene counts; Pathways represent adjusted p-value.(H) Heatmap displaying the expression of key transcription factors across the four experimental groups. (I) Bar charts showing significant GO terms for pathways downregulated in WT-KA mice and pathways upregulated in aKO-KA mice. The x-axis represents -log10 (adjusted p-value) Differential gene expression and functional analysis Differentially expressed genes (DEGs) were identified using the FindMarkers function in Seurat via the Wilcoxon rank-sum test, as described previously [20, 26, 34]. Significant DEGs were defined by an adjusted p-value < 0.05 and an absolute log_2 fold-change > 0.5. Functional enrichment analyses, including Gene Ontology (GO), were performed using the clusterProfiler R package (version 4.10.1) and the org.Mm.eg.db database (version 3.18.0) with default parameters. Additionally, Gene Set Enrichment Analysis (GSEA) was conducted utilizing the Molecular Signatures Database (MSigDB) collection (curated gene sets). Enrichment scores were calculated within individual categories (e.g., KEGG, GO), and the results were integrated for comprehensive visualization. Trajectory inference To model astrocyte state transitions, trajectory analysis was performed using Monocle3. The Seurat object containing astrocyte clusters was converted to a subset. The trajectory graph was learned using learn_graph, and cells were ordered in pseudotime using order_cells, with the root node defined by the homeostatic cluster 0. Cell-cell communication analysis Intercellular communication networks were inferred using the CellChat R package (version 2.1). Interaction probabilities were computed based on the expression of known ligand-receptor pairs in the CellChatDB mouse database [35]. Significant interactions were identified using a permutation test, and communication strength was visualized using heatmaps to compare signaling pathways between WT and aKO conditions. Co-expression network analysis (hdWGCNA) To identify gene co-expression modules associated with specific cell states, high-dimensional Weighted Gene Co-expression Network Analysis (hdWGCNA) was applied. Metacells were constructed to reduce sparsity, and a soft power threshold was selected to achieve a scale-free topology. Co-expression modules were constructed, and module eigengenes were correlated with experimental conditions (Genotype and Treatment) to identify disease-associated modules. Gene regulatory network analysis To infer transcription factor (TF) activity from single-nucleus gene expression data, we utilized the VIPER (Virtual Inference of Protein-activity by Enriched Regulon analysis) method [36] implemented in the decoupleR R package. A comprehensive gene regulatory network (regulon) containing signed TF-target interactions was obtained from the CollecTRI resource (mouse organism) via OmnipathR [37]. CollecTRI integrates multiple resources of curated regulons to provide broad coverage of transcriptional interactions [38]. Using this prior knowledge network, the gene expression matrix (log-normalized counts) was transformed into a protein activity matrix. The run viper function estimates the activity of each TF based on the enrichment of its downstream target genes within the gene expression signature of each individual nucleus. The resulting Normalized Enrichment Scores (NES) were used to quantify relative TF activity, where positive scores indicate activation and negative scores indicate repression. Adeno-associated viruses (AAV) injection WT or aKO mice at 8–10 weeks of age were anesthetized with ketamine/Xylazine (200/10 mg/kg) and placed in a stereotaxic frame. AAV5-GFAP-eGFP-PPARγ-shRNAmir vectors or AAV5-GFAP-eGFP control vectors were provided by Vector Biolabs (Malvern, Pennsylvania). AAV Vectors were stereotaxically injected into the bilateral side of the hippocampus in aKO mice at the coordinate: AP, -2.3, ML, 2, and DV, -2 (3.0 µl for each side at 0.2 µl /min), as described previously [23]. SE was induced 4 weeks following AAV injections. The knockdown efficiency is quantified by immunointensity. Statistical analysis Statistical analyses were mainly performed using OriginPro software (OriginPro 2020). Data are mainly presented as mean ± standard error (SEM). Non-parametric data are presented as median (interquartile range, IQR). Two-group comparisons were analyzed using the t-testor Mann-Whitney U test (for non-parametric data). Multiple group comparisons were analyzed using Two-way Analysis of Variance (ANOVA) followed by Tukey’s post-hoc test. For non-parametric multiple group comparisons, Kruskal-Wallis ANOVA is used. Time-course data (Racine scores) were analyzed using Two-way repeated measures ANOVA and Fisher LSD for multiple comparison. Survival curves (seizure incidence) were analyzed using the Kaplan-Meier method with the Log-rank (Mantel-Cox) test or Holm-Sidak multiple comparisons. Proportions were compared using the Chi-square test and Bonferroni method was used for multiple comparison. A p-value < 0.05 was considered statistically significant. Results Astrocytic Mgll deletion confers robust neuroprotection against kainic acid-induced status epilepticus To determine the cellular source of MAGL-mediated neuroprotection, we compared wild-type (WT) mice with astrocyte-specific (aKO), neuron-specific (nKO), and total (tKO) Mgll knockout mice to define the relative contributions of 2-AG signaling in astrocytes versus neurons to seizure activity (Fig. 1A). Upon administration with KA, both tKO and aKO mice exhibited marked resistance to seizure induction, characterized by a significantly prolonged latency to the first generalized seizure (Racine stages 4-5) (Fig. 1B). Sex was not a significant predictor of seizure latency, and no significant interaction between sex and genotype was observed (Table S2), indicating that the observed genotype effects are unlikely to be confounded by sex differences. Furthermore, the maximum Racine score and the percentage of seizure onset within one hour were significantly reduced in aKO and tKO mice compared to WT controls (Fig. 1C, S1A). In sharp contrast, neuron-specific deletion alone was insufficient to confer comparable resistance; nKO mice displayed seizure severity and latency profiles statistically similar to WT controls (Fig. 1B&C). These findings suggest that astrocytic Mgll deletion makes a major contribution to the anti-seizure phenotype observed in global knockout mice, whereas neuronal Mgll deletion alone may contribute to baseline synaptic buffering and does not reproduce the broader anti-seizure and anti-inflammatory protection. Because KA-induced SE can produce excitotoxic hippocampal injury and subsequent alterations in exploratory activity and anxiety-like behavior, we next assessed locomotor and anxiety-related behavioral changes using the open field test [39]. Consistent with KA-induced excitotoxicity, WT and nKO mice displayed marked hyperactivity, traveling significantly longer distances and spending less freezing time compared to vehicle controls. However, aKO and tKO mice exhibited locomotor profiles indistinguishable from baseline, indicating preservation of normal network function (Fig. 1D; Fig. S1B). We next assessed neuronal injury using Fluoro-Jade C (FJC) staining. Consistent with excitotoxic damage, WT mice displayed FJC-positive signals in the hippocampus following KA administration (Fig. 1E). In contrast, FJC-positive puncta were not readily detected in either tKO or nKO mice under the same conditions (Fig. S1C). While the absence of detectable FJC labeling in aKO mice is consistent with their reduced seizure, the lack of overt FJC signal in nKO mice despite comparable seizure severity suggests that neuronal injury may be differentially regulated across genotypes. To further examine neuronal integrity, we quantified NeuN immunoreactivity and observed no significant differences between KA and vehicle groups across all genotypes at this time point (Fig. S2A). This indicates that overt neuronal loss is limited under these conditions and that NeuN may not sensitively capture early or subtle injury. We also evaluated synaptic alterations by quantifying PSD95 puncta in the stratum radiatum of the CA1 region. Following KA treatment, WT mice exhibited a reduction in PSD95 levels, indicative of synaptic disruption. In contrast, this effect was attenuated in nKO mice, where PSD95 density showed higher variability and failed to reach statistical significance (Fig. S2B). Together, these data suggest that conventional markers of neuronal and synaptic integrity may show differential sensitivity across genotypes and injury states. Finally, to determine whether the seizure susceptibility observed in nKO mice could be rescued by inhibiting the remaining non-neuronal MAGL pool we treated nKO mice with JZL-184, a potent and selective MAGL inhibitor [21]. Pharmacological inhibition of the remaining non-neuronal MAGL in nKO mice effectively rescued the phenotype, significantly prolonging the latency to seizure onset (Fig. 1F). These findings support the conclusion that the protective effects of global MAGL knockout against kainic acid-induced seizures arise primarily from inactivation of MAGL in astrocytes. Astrocyte-specific mgll deletion attenuates kainic acid (KA)-induced seizure susceptibility and hippocampal neurodegeneration (A) Schematic representation of the genetic strategy used to generate total mgll knockout (tKO) mice (using Sox2-Cre), astrocyte-specific mgll knockout (aKO) mice (using Gfap-Cre), and neuron-specific mgll knockout (nKO) mice (using Syn-Cre). (B) Kaplan-Meier curves showing the cumulative seizure incidence rate following KA administration in WT, tKO, aKO, and nKO mice. Statistical significance was determined using the Kaplan-Meier estimator with Holm-Sidak’s multiple comparison test. (C) Time course of seizure severity quantified by the Racine scale over 2 h post-KA injection. Data were analyzed using two-way repeated measures ANOVA followed by Tukey multiple comparison test. (D) Representative heatmaps illustrating the activity tracks of mice in the Open Field test. Warmer colors indicate strictly higher time spent in that location. Quantification of locomotor and anxiety-like behaviors in the Open Field test, showing normalized total distance traveled and normalized freezing time. Data are means ± SEM. *p < 0.05, **p < 0.01, ***p < 0.001 (n = 11 ~ 13/group, ANOVA with post hoc Tukey test). (E) Representative immunofluorescence images of Fluoro-Jade C (FJC) staining in the hippocampus of WT and aKO mice following KA treatment. High-magnification insets (colored boxes) display the CA1 and CA3 regions. Scale bars: 500 μm (overview) and 100 μm (insets). Quantification of FJC-positive puncta density (puncta/mm²) in the hippocampus, showing significantly reduced neurodegeneration in aKO mice compared to WT Data are means ± SEM. **P < 0.01 (Kruskal-Wallis ANOVA, n = 8/group). (F) Latency to first seizure of nKO mice with JZL-184 or DMSO. Data are median (IQR). *P < 0.05 (n = 6–7/group, unpaired Student’s t-test) Astrocytic MAGL serves as a critical regulator of KA-induced hippocampal neuroinflammation To contextualize our findings within the broader biological landscape of epilepsy, we re-analyzed public bulk RNA-sequencing datasets from the hippocampal dentate gyrus (GSE213393) [40]. We observed distinct temporal dynamics in the endocannabinoid (eCB) system following KA-induced seizures (Fig. 2A). Interestingly, while Mgll expression positively correlated with neuroinflammation scores in control and contralateral tissues, this relationship was reversed in the ipsilateral injured hippocampus (Fig. 2B). Several previous studies have shown that seizure activity enhances the production of 2-AG as a self-protective mechanism [41, 42]. This suggests that MAGL in the brain plays an important role as a compensatory adaptive response to acute neuroinflammation in maintaining brain homeostasis [43]. Further analysis of CA1 tissue (GSE99577) [44] showed a transient, biphasic response of Mgll expression, which was downregulated first and then upregulated, along with a sustained increase of neuroinflammation score, particularly for the markers of astrocytes and microglia (Fig. S3A, B). This transient engagement suggests that the brain’s endogenous protective mechanism is insufficient to arrest the progression of gliosis. Comparison of seizure susceptibility, neuroinflammation, and behavioral deficits in global (tKO), astrocyte-specific (aKO), and neuron-specific (nKO) mgll knockout mice. (A) Heatmap illustrates the temporal mRNA expression profiles of the endocannabinoid system (eCB) in the hippocampus of mice treated with saline or KA at days 3, 7, and 14 post-injections. The color scale indicates relative expression levels. (B) Pearson correlation analysis between mgll expression levels and a composite neuroinflammation score (derived from the expression of Aif1, Cx3cr1, Cd68, Il1b, Ccl2, Cxcl10, Ptgs2, Nos2, C1qa, C3, Itgam, and Serping1) across different time points (Day 3, 7, 14) in ipsilateral saline-injected DG, contralateral KA-injected DG, and ipsilateral KA-injected DG. The linear regression significance was determined by ANOVA (p-values indicated on plots). (C) Relative mRNA expression levels of proinflammatory cytokines Il1b, Il6, and Tnfa in the hippocampus. Data are means ± SEM. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.001 (two-way ANOVA with post hoc Tukey test, n = 5/group). (D) Representative immunofluorescence images of GFAP staining (green) in the DG across different genotypes and treatments. Nuclei are stained with DAPI (blue). Scale bar: 100 μm. GFAP immunofluorescence intensity in the DG regions was quantified and normalized to the mean of the corresponding vehicle group for each genotype and represent individual mice. Data are means ± SEM. *p < 0.05, **p < 0.01, ****p < 0.001, compared with Veh in each genotype; #p < 0.05, ##p < 0.01, ####p < 0.001 compared with WT-KA (two-way ANOVA with post hoc Tukey test, 6 mice for vehicle groups and 8 mice for KA groups). (E) Representative immunofluorescence images of Iba1 staining (green) in the DG across different genotypes and treatments. Scale bar: 100 μm. Iba1+ cells were counted in the DG, and the density was normalized to normalized to the mean of the corresponding vehicle group for each genotype. Data represent individual mice (6 mice for vehicle groups and 8 mice for KA groups). Data are means ± SEM. *p < 0.05, **p < 0.01, ****p < 0.001 compared with Veh in each genotype, #p < 0.01, ##p < 0.01, ####p < 0.001 compared with WT-KA (two-way ANOVA with post hoc Tukey test) Thus, we examined whether this astrocyte-specific protection extended to the neuroinflammatory sequelae of SE. qPCR analysis demonstrated robust induction of proinflammatory cytokines, specifically Il1b, Il6, and Tnfa, in the hippocampi of WT and nKO mice (Fig. 2C). In contrast, this cytokine storm was significantly attenuated in aKO and tKO mice. Immunohistological analysis at Day 6 revealed that nKO mice, similar to WT controls, exhibited profound reactive gliosis, characterized by widespread upregulation of GFAP and Iba1. Astrogliosis was pervasive throughout the hippocampus, including DG, CA1 and CA3. In contrast, aKO and tKO mice maintained glial markers at near-basal levels (Fig. 2D, S3C&E). Iba1 was used to detect microgliosis. Activated microglia, characterized by hypertrophic cell bodies and thickened processes, were recruited to the DG granule cell layer, showing an increased immunointensity in WT and nKO mice (Fig. 2E). The results of CA1 and CA3 are similar (Fig. S3D&F). These histological observations were confirmed by transcriptional profiling. Collectively, these data identify astrocytic 2-AG signaling as the dominant cellular mediator of MAGL-dependent protection against excitotoxic injury and inflammation. To further determine whether the attenuation of neuroinflammation is solely a secondary consequence of reduced seizure severity or reflects a direct modulation of post-seizure pathological processes, we next examined the effects of pharmacological MAGL inhibition when administered after the establishment of SE (Fig. S4A). JZL-184 was delivered following the onset of generalized seizures, thereby minimizing its influence on seizure initiation. Despite the delayed administration, JZL-184 treatment resulted in a noticeable reduction in reactive gliosis at Day 6. Consistent with the regional heterogeneity of hippocampal vulnerability, the effects of delayed MAGL inhibition appeared to be subfield-dependent. In the DG, JZL-184 treatment led to a significant reduction in reactive gliosis, as evidenced by decreased GFAP and Iba1 immunoreactivity compared to vehicle-treated controls (Fig. S4B&C). Meanwhile, microglial activation, as assessed by Iba1 staining, was attenuated broadly across the DG, CA1, and CA3 subfields (Fig. S4B&C). Together, these results suggest that enhancement of 2-AG signaling is not limited to suppressing seizure initiation, but can also attenuate the downstream neuroinflammatory cascade following established SE. This supports a model in which astrocytic MAGL regulates both the acute excitotoxic response and the subsequent propagation of glia-mediated pathology. Astrocytic Mgll deletion confers region- and cell type-specific resilience by preserving metabolic homeostasis and neuronal Identity To dissect the molecular mechanisms and cellular heterogeneity underlying aKO-mediated neuroprotection, we performed single-nucleus RNA sequencing (snRNA-seq) on pooled bilateral hippocampal tissue from five mice per group (WT-Veh, WT-KA, aKO-Veh, aKO-KA). Only mice with Racine score reaching 4 and above after KA injection were selected. Following rigorous quality control, 25,141 high-quality nuclei were retained (Fig. 3A). Quality assessments are provided in Figures S5A&B. Unsupervised clustering and manifold learning resolved the cells into 12 major cell types: Slc17a7 (excitatory neurons), Gad1/Gad2 (inhibitory neurons), Sox11(immature neurons), Slc1a3 (astrocytes), P2ry12/Cx3cr1 (microglia), oligodendrocytes, oligodendrocyte precursor cells (OPC), endothelial cells, pericytes, ependymal cells, fibroblasts and choroid plexus epithelial cells (Fig. 3A). Canonical marker expression confirmed distinct molecular identities for all clusters (Fig. 3B). To resolve the diversity of excitatory neurons, we further identified Dentate Gyrus (DG) granule cells (Prox1+, Fam163b+), CA1 (Sv2b+, Mpped1+, Prkcb+) and CA3 (Sv2b+, Grik4+) pyramidal cells, Subiculum neurons (Fn1+), and Cajal-Retzius cells (Reln+) (Fig. 3C&D, Fig. S5C). Given their central role in hippocampal excitability, we prioritized the analysis of principal excitatory types. We first assessed the global impact of KA on transcriptional states. Interestingly, the response to KA was cell-type dependent. In DG granule cells, the response to seizure activity appeared highly conserved. Heatmap analysis revealed a striking concordance in expression profiles, with WT and aKO mice displaying nearly identical patterns of up- and downregulated gene clusters (Fig. 3E). This similarity was statistically quantified by scatter plot analysis of Log2FC values, where genes aligned tightly along the diagonal, yielding a high Pearson correlation coefficient (Pearson’s R = 0.80) (Fig. 3F). In sharp contrast, the response in CA1/CA3 pyramidal cells was markedly divergent. In the vulnerable CA1 and CA3 pyramidal populations, aKO mice exhibited a blunted expression of WT-KA induced DEGs (Fig. 3E&F; Fig. S5D). This suggests that aKO confers a capacity for active transcriptional remodeling in pyramidal cells that is otherwise compromised in the wild-type condition. In inhibitory neurons, aKO showed resilience to KA-induced gene upregulation (Fig. S5E&F). This pattern of genotype-dependent resistance extended to the glial compartment; for example, the reactive astrocyte marker Gfap was significantly attenuated in aKO mice, and microglia also displayed a dynamic and distinct trajectory shift. These global data suggested that aKO does not simply suppress all signaling but specifically protects vulnerable populations (CA1) while preserving the baseline response in resistant populations (DG). Finally, we assessed the subfield-specific neuronal responses. In DG granule cells, we identified a core set of pathways shared between aKO-KA and WT-KA (vs. their respective vehicles). These shared signatures confirmed that both genotypes responded to the excitatory impairment, mounting a common program of endocytosis and regulation of cell matrix (Fig. 3G). While global gene expression analysis suggested a conserved response to KA in the DG, regulatory network analysis based on transcription factors revealed a critical divergence in cellular resilience (Fig. 3H). In WT-KA mice, we observed a collapse of key granule cell identity and maintenance factors. Specifically, Prox1, the master regulator of DG neuronal identity, was markedly downregulated in WT-KA neurons but preserved in aKO-KA mice. Furthermore, aKO-KA mice exhibited sustained activity of neurogenic and stem-cell associated factors (Sox9, Id3, Nanog) and metabolic regulators (Tfam, Bmal1) that were otherwise lost in the WT condition. Conversely, the inflammatory driver Irf4 was activated in WT-KA neurons but blunted in the aKO. These data indicate that while aKO DG neurons sense the seizure stimulus, they are protected from the subsequent loss of cellular identity and metabolic failure. CA1 is a region classically vulnerable to metabolic and excitotoxic stress. In WT mice, KA triggered a bioenergetic crisis in CA1 neurons, defined by the extensive downregulation of oxidative phosphorylation and electron transport chain genes (Fig. 3I). In contrast, aKO mice exhibited a distinct, pro-survival response to KA. CA1 neurons in aKO mice exhibited a robust upregulation of pathways associated with structural plasticity, such as actin cytoskeleton organization (Fig. 3I). Analysis of upregulated pathways in WT-KA and downregulated pathways in aKO-KA revealed no significant GO enrichment (p.adjust > 0.05, Table S2), further suggesting that neurons in aKO mice maintain the capacity to remodel synaptic connections and reinforce synapse integrity in response to the seizure insult. Astrocytic Mgll deletion alters astrocyte functional states and suppresses disease-associated microglia (DAM) activation We then focused our analysis on astrocytes. Following additional quality control aimed at removing oligodendrocyte and microglial contamination, 2,671 astrocytes were retained and re-clustered into six transcriptionally distinct subclusters (Fig. 4A). Notably, aKO mice contained a higher number of astrocytes than WT mice, in both Veh- and KA-treated conditions (Fig. 4B& C). Upon KA administration, WT mice exhibited a marked shift in astrocyte composition, characterized by the emergence of a KA-induced Cluster 5 and the reduction of Cluster 0. In contrast, aKO mice primarily expanded Cluster 2 (Fig. 4B&C). To resolve the heterogeneity of the astrocyte population, we first assessed the expression of canonical marker genes across the identified subclusters (Fig. 4D). It revealed distinct molecular identities corresponding to different functional states. Cluster 0 and Cluster 1 displayed high expression of homeostatic markers (Slc1a2, Glul, Ndrg2), suggesting a baseline or quiescent phenotype. In contrast, Cluster 5 exhibited a strong reactive and inflammatory profile, defined by the upregulation of Gfap, Vim, and immune-associated genes (Stat3, Il1r1). Cluster 4 was clearly identifiable as a proliferative population, exclusively expressing cell cycle markers such as Mki67, Top2a, and progenitor markers (Prox1, Fabp7). The Gene Set Enrichment Analysis (GSEA) corroborated our marker-based classification while revealing a deeper metabolic divergence between the subsets (Fig. S6A). Cluster 2 presented a unique bioenergetic profile. Unlike the glycolytic Cluster 1, Cluster 2 showed robust enrichment for oxidative phosphorylation and ATP biosynthetic processes, alongside immunomodulatory pathways like interleukins 4 and 13 signaling. The proliferative identity of Cluster 4 was functionally confirmed by the massive upregulation of cell cycle pathways. Cluster 5, aligning with its inflammatory marker profile, showed enrichment for cellular response to hypoxia and axon guidance, suggesting these astrocytes were responding to acute tissue injury and environmental stress. Mgll deletion alters astrocyte functional states and suppresses disease-associated microglia (DAM) activation. (A) UMAP visualization of astrocytes clustered into 6 distinct subclusters (Clusters 0 ~ 5). (B) UMAP plots split by experimental group (WT-Veh, WT-KA, aKO-Veh, and aKO-KA), visualizing the distribution of astrocyte subclusters across conditions. (C) Stacked bar chart showing the percentage composition of each astrocyte subcluster within the four experimental groups. (D) Dot plot of canonical marker genes used to annotate astrocyte states: Homeostatic, Reactive, Inflammatory, Progenitor, and Proliferative. Dot size represents the percentage of expressing cells, and color intensity indicates average expression levels. (E) Pseudotime trajectory analysis of astrocyte subpopulations, colored by pseudotime values. (F) UMAP showing the bifurcation of astrocyte trajectories into two distinct lineages (Lineage 1 in yellow and Lineage 2 in green). (G) Quantification of the proportion of cells belonging to Lineage 1 versus Lineage 2 across experimental groups. Statistical significance was determined using the Chi-square test. (H) Heatmap displaying the activity of key transcription factor regulons across astrocyte clusters (0 ~ 5). (I) Cell-cell communication analysis (CellChat) illustrating the communication probability between astrocyte subclusters (rows) and neuronal populations (columns). (J) UMAP visualization of microglia clustered into distinct subpopulations. (K ~ M) UMAP plots highlight a specific disease-associated microglial module identified by hdWGCNA. The red dashed line indicates a population of activated microglia present in WT-KA (K) but largely absent in aKO-KA (L), corresponding to the hdWGCNA module (M). (N) Dot plot showing the expression of Homeostatic microglial markers and Disease-associated microglial (DAM) markers across microglial clusters To determine the dynamic trajectories underlying these responses, we performed pseudotime analysis (Fig. 4E). Astrocytes displayed multipotent reactivity (Fig. 4F). Both WT and aKO astrocytes switched into lineage 1, which switched into inflammatory Cluster 5 (Fig. S6B), while aKO astrocytes preferentially accumulated in lineage 2, leading to Cluster 2 (Fig. 4F&G). Density plot along pseudotime suggested that aKO astrocytes positioned within Cluster 2 (Fig. S6C). To understand this bifurcation, direct comparison of transcription factors between Cluster 2 and Cluster 5 demonstrated that aKO cells activate a completely different set of regulators focused on survival (e.g., Foxo3, Tox) and lipid metabolism (e.g., Ppara, Aebp1), suggesting a retained capacity to support neuronal homeostasis (Fig. 4H). We further inferred intercellular communication networks using CellChat. We specifically examined the endocannabinoid signaling pathway (2-AG–Cnr1). Remarkably, among all astrocyte subpopulations, only Cluster 2 exhibited significant communication probabilities with neuronal targets (Fig. 4I), which can be contributed to the enrichment of Dagla (diacylglycerol lipase-α, the synthetase of 2-AG) expressing astrocytes (Fig. S6D). Given the intimate functional coupling between astrocytes and microglia in neuroinflammation, we next interrogated the microglial landscape. Re-clustering of microglia yielded four distinct subclusters (Fig. 4J). In WT mice, KA injection triggered a profound emergence of a specific disease-associated population, Cluster 2, a shift that was significantly attenuated in aKO mice (Fig. 4K&L, S7A&B). WGCNA identified a specific gene module predominantly expressed in Cluster 2 (Fig. 4M, S7C), which was significantly enriched for pathways involving apoptosis and cytokine signaling (Fig. S7D). The result was confirmed by canonical markers. Cluster 0, 1, and 3 maintained canonical surveillance markers (e.g., P2ry12, Cx3cr1), consistent with a homeostatic phenotype. In contrast, the WT-dominant Cluster 2 displayed a pro-inflammatory signature characterized by high expression of Spp1 and Cd63, a profile aligning with Disease-Associated Microglia (DAM) previously implicated in aberrant synaptic pruning (Fig. 4N). Pseudobulk analysis confirmed this genotype-specific divergence: aKO-KA microglia exhibited a blunted induction of the DAM signature while sustaining high expression of the homeostatic program, suggesting that aKO prevents the polarization of microglia into an inflammatory state (Fig. S7E). Further validation of reactive astrocyte and disease-associated microglial markers was performed at the mRNA level (Fig. S6F). Following KA-induced SE, WT mice exhibited significant upregulation of reactive astrocyte-associated genes, including Vim, Stat3, Cd44, and C3, together with increased expression of disease-associated microglial markers such as Spp1, Apoe, and Ctsb. In contrast, these injury-associated transcriptional responses were markedly attenuated in aKO mice, suggesting that astrocytic Mgll deletion mitigates excitotoxicity-induced glial reactivity and the transition toward neuroinflammatory states. The attenuation of seizure susceptibility and hippocampal neuroinflammation is mediated via CB1 and PPARγ signaling pathways To determine the signaling pathways downstream of MAGL inhibition, we first investigated the requirement of the CB1 receptor. We administered the specific MAGL inhibitor JZL-184 to Cnr1 knockout (Cnr1−/−) mice prior to KA induction (Fig. S8A, B). After JZL-184 administration, CB1R expression was measured and no significant reduction was observed (Fig. S8C), consistent with previous studies [45]. In the absence of CB1 receptor, the neuroprotective effects of JZL-184 were completely abolished; we observed no significant extension of seizure latency (Fig. 5A&B) and no reduction in seizure severity (Fig. 5C) compared to vehicle-treated controls. Furthermore, JZL-184 failed to attenuate reactive astrogliosis (GFAP) or microglial activation (Iba1) in the DG (Fig. 5D&F) or the CA1 and CA3 subfields (Fig. S8D&E). This widespread lack of rescue confirms that CB1 signaling is a prerequisite for 2-AG-mediated neuroprotection. Transcriptomic profiling indicated robust metabolic reprogramming in aKO astrocytes was represented by PPAR signaling pathways. Although our analysis specifically highlighted PPARα activity, the endocannabinoid 2-AG has been demonstrated to resolve neuroinflammation by activating PPARγ [20, 23, 46]. To determine if this specific pathway drives the protective phenotype, we treated wild-type mice with JZL-184 alongside the selective PPARγ antagonist GW9662, administered every other day for three doses with the final dose given 30 min before KA injection (Fig. 5G). While GW-9662 alone did not worsen seizure (Fig. 5H, Table S4), the concurrent blockade of PPARγ effectively reversed the beneficial effects of JZL-184 on seizure severity(Fig. 5I). Similarly, GW9662 treatment negated the anti-inflammatory capability of JZL-184, restoring high levels of GFAP and Iba1 expression in the hippocampus (Fig. 5J, K). However, comparing these two interventions revealed a notable divergence in seizure onset, GW9662 did not fully abolish the JZL-184-induced extension of seizure latency to the same degree as global Cnr1 deletion (Fig. 5H), suggesting that other CB1-dependent mechanisms contribute independently to the delay in seizure onset, likely involving acute synaptic modulation. The neuroprotective effects of MAGL inhibition are mediated via CB1 receptors and PPARγ signaling pathways. (A) Assessment of seizure susceptibility in CB1 receptor knockout mice treated with the MAGL inhibitor JZL-184 compared to vehicle controls. Data are means ± SEM. (n = 12 ~ 13 mice/group). (B) Cumulative seizure incidence rate (Log-rank test). (C) Time course of Racine scores (Two-way repeated measures ANOVA). (D) Representative immunofluorescence images of GFAP (green) and Iba1 (red) staining in the hippocampus of DMSO- and JZL-treated mice following KA or saline (Veh) administration. Scale bar: 100 μm. (E&F) GFAP immunofluorescence intensity and Iba1+ cells in the DG regions were quantified and was normalized to the means of the corresponding vehicle group and represent individual mice. (two-way ANOVA with post hoc Tukey test, n = 4 mice for vehicle group and 6 mice for KA group). (G) Schematic timeline of the rescue experiment using GW-9662 (GW). (H) Kaplan-Meier curves showing cumulative seizure incidence (n = 8). Statistical significance was determined by the Log-rank test with Holm-Sidak’s multiple comparison test. (I) Time course of seizure severity (Racine score). **p < 0.01, ***p < 0.001 (two-way repeated measures ANOVA with post hoc Tukey test, n = 8 animals/group). (J) Representative immunofluorescence images of GFAP (green) and Iba1 (red) in the DG, CA1, and CA3 regions across the four treatment groups (Veh, GW, JZL, and JZL + GW). Scale bar: 100 μm. (K) GFAP immunofluorescence intensity and Iba1 + cells were quantified and was normalized to the mean of the corresponding vehicle group and represent individual mice. Data are means ± SEM. *p < 0.05, **p < 0.01, ****p < 0.005 ,****p < 0.001 compared with JZL (two-way ANOVA with post hoc Tukey test, n = 6/group) To directly test the astrocyte-specific contribution of PPARγ, we performed stereotaxic bilateral hippocampal injection of AAV expressing a GFAP promoter-driven PPARγ-targeting shRNA (AAV-GFAP-shPPARγ) or control shRNA (AAV-GFAP-Control) in aKO mice prior to KA induction (Fig. S9A). Successful transduction and selective PPARγ knockdown in GFAP-positive astrocytes were confirmed by immunostaining (Fig. S9D). Astrocyte-specific PPARγ knockdown did not significantly alter seizure latency or Racine score severity compared to control AAV-injected aKO mice (Fig. S9B&C). This stands in contrast to the latency effect observed with systemic GW9662, which suppresses PPARγ across all cell types over several days prior to KA injection and therefore likely reflects cumulative transcriptional changes in non-transduced astrocytes or non-astrocytic cell types that alter the pre-seizure brain state. The selective absence of a seizure latency effect with astrocyte-specific knockdown thus indicates that astrocytic PPARγ may not govern acute seizure threshold. In contrast, AAV-GFAP-shPPARγ mice exhibited significantly greater reactive astrogliosis and microglial activation than controls, as reflected by elevated GFAP and Iba1 immunoreactivity in the hippocampus (Fig. S9E). This dissociation identifies astrocytic PPARγ as a transcriptional node specifically governing the post-seizure neuroinflammatory trajectory rather than acute neuronal excitability. Together with the pharmacological data, these findings support a model in which MAGL-dependent 2-AG elevation engages CB1-dependent signaling that bifurcates into an acute synaptic component modulating seizure threshold and a sustained astrocytic PPARγ-dependent transcriptional component that limits post-seizure glial activation and secondary neuroinflammation. Astrocytic Mgll deletion confers sustained protection against chronic behavioral deficits and neuronal loss To evaluate whether the neuroprotective effects of astrocytic Mgll deletion translate into long-term functional recovery, we assessed behavioral and histological outcomes in the chronic phase (Day 21). We first examined locomotor activity and anxiety-like behavior using the Open Field Test. WT-KA mice exhibited significant hyperactivity, characterized by increased total distance moved and mean activity, alongside a marked reduction in freezing time compared to vehicle controls (Fig. 6A&B). Strikingly, this behavioral phenotype was completely abrogated in aKO-KA mice, which displayed locomotor profiles indistinguishable from baseline vehicle controls. We next assessed cognitive function using the Novel Object Recognition (NOR) task. While WT-KA mice displayed a severe deficit in recognition memory, indicated by a significantly reduced recognition index, aKO-KA mice retained intact cognitive performance (Fig. 6C), suggesting the successful preservation of hippocampal memory circuits. Consistent with our snRNA-seq findings linking aKO astrocytes to synaptic assembly pathways, we evaluated both neuronal survival and synaptic integrity. Immunofluorescence staining for NeuN revealed that WT-KA mice suffered extensive neuronal loss in the CA1 region, whereas aKO-KA mice displayed robust preservation of neuronal density (Fig. 6D). Furthermore, we quantified the density of the postsynaptic marker PSD-95. WT-KA mice suffered a profound loss of synaptic puncta in the hippocampus; however, this synaptic erosion was effectively prevented in aKO-KA mice, which maintained PSD-95 density at levels comparable to vehicle controls (Fig. 6E). Astrocyte-specific Mgll deletion prevents KA-induced cognitive deficits and synaptic loss. A Representative heatmaps illustrating the activity tracks of mice in the Open Field test. Warmer colors indicate strictly more time spent in that location. B Assessment of hyperactive activity and anxiety-like behavior in the Open Field test. Quantification of total distance moved, mean activity, and freezing time in WT and aKO mice treated with Vehicle or KA. Data are mean ± SEM. ****p < 0.001 compared with Veh, ####p < 0.001 compared with WT-KA (ANOVA with post hoc Tukey test, n = 9 ~ 10 animals/group). C Representative heatmaps of the Novel Object Recognition (NOR) test, illustrating the exploration time spent around the objects. Quantification of the Recognition Index is shown as mean ± SEM. ***p < 0.001 compared with Veh in each genotype, p < 0.001 compared with WT-KA (ANOVA with post hoc Tukey test, n = 11 ~ 12 animals/group). D Representative images of NeuN (red) staining in the CA1 region with quantification. Nuclei are stained with DAPI (blue). Scale bars: 50 μm. NeuN stained area is measured and data represent individual mice. Data are means ± SEM. *p < 0.05 (ANOVA with post hoc Tukey test, n = 6 mice for vehicle group and 8 mice for KA group). E Representative immunofluorescence images of the postsynaptic marker PSD-95 (red) in the hippocampus (CA1 region). Numbered boxes (1, 2) correspond to the enlargement insets on the right. Scale bars: 10 μm. PSD-95 puncta were counted, and the density was normalized to the mean of the corresponding vehicle group for each genotype. Data points in the quantification of PSD-95 positive puncta represent individual mouse. Data are means ± SEM. *p < 0.05 *p < 0.05 (ANOVA with post hoc Tukey test, n = 6 mice for vehicle group and 8 mice for KA group). F-G Representative immunofluorescence images of GFAP (green) and Iba1 (red) in the DG, CA1, and CA3 regions across the four groups. Scale bar: 100 μm. GFAP immunofluorescence intensity and Iba1 + cells were quantified and was normalized to the mean of the corresponding vehicle group and represent individual mice. Data are means ± SEM. *p < 0.05, **p < 0.01, ****p < 0.005 ,****p < 0.001 compared with corresponding Veh group in each genotype and #p < 0.05, ###p < 0.005 compared with WT-KA group (two-way ANOVA with post hoc Tukey test, n = 6/group) To determine if this functional rescue corresponded to a suppression of chronic pathology, we evaluated neuroinflammation and synaptic integrity. Immunofluorescence analysis revealed that WT-KA mice sustained widespread reactive astrogliosis and microglial activation throughout the DG, CA1, and CA3 subfields at Day 21. In contrast, aKO-KA mice exhibited significantly attenuated glial activation across all hippocampal regions (Fig. 6F&G), confirming that the early engagement of the astrocytic protective axis prevents the establishment of chronic neuroinflammation. Collectively, these data demonstrate that the astrocyte-specific modulation of 2-AG signaling provides durable neuroprotection, preserving synaptic structure and cognitive function long after the initial insult. Discussion In the present study, we identify astrocytic MAGL as a critical metabolic regulator that influences the transition from adaptive neural responses to catastrophic excitotoxicity and neuroinflammation following SE. By combining cell type-specific genetic knockouts with single-nucleus transcriptomics, we demonstrate that the neuroprotective efficacy of MAGL inactivation is driven primarily by astrocytic 2-AG signaling. This protection is not merely a consequence of reduced seizure susceptibility but appears to stem from a fundamental reprogramming of the glial inflammatory response. While previous studies have established that systemic MAGL inhibition suppresses seizures through enhanced endocannabinoid (eCB) signaling, it has generally been assumed that this protection arises from direct excitatory neuron communication [7, 14], which can thereby suppress glutamatergic transmission at the synapse [47]. Our data suggest a more complex picture. We observed that neuron-specific Mgll deletion failed to replicate the protective phenotype of global or astrocyte-specific Mgll knockout, though it did confer a partial and more limited degree of protection. Neurons express high levels of presynaptic MAGL that terminates retrograde endocannabinoid signaling and modulates synaptic transmission and plasticity. However, recent evidence indicates that neuron-derived 2-AG selectively signals to presynaptic CB1R within a restricted synaptic microdomain, rather than diffusing to activate astrocytic CB1R [17]. While neurons are the primary source of synaptic 2-AG, astrocytic MAGL is principally responsible for generating 2-AG–derived arachidonic acid (AA), which is a precursor of proinflammatory prostaglandins like PGE2, to regulate inflammatory processes [19]. Consistent with this spatial compartmentalization, prior studies have demonstrated that deletion of neuronal MAGL does not alter seizure susceptibility [48]. This dissociation suggests that the therapeutic potential of 2-AG mobilization lies not merely in dampening synaptic release, but in its capacity to resolve the profound neuroinflammatory response that accompanies SE. These findings parallel recent observations in traumatic brain injury, where astrocytic, but not neuronal, 2-AG signaling was found to be critical for reducing secondary injury cascades [20]. Specifically, SE typically triggers a rapid induction of proinflammatory cytokines, lowering seizure thresholds and promoting secondary post-seizure injury [2, 49]. The neuroinflammatory cascade following SE is further amplified by reactive oxygen species generated through cell-type-specific activation of NADPH oxidase, particularly NOX2, which has been shown to contribute to seizure susceptibility, neuroinflammation, and cognitive impairment in epilepsy models [50–53]. We show that astrocytic Mgll deletion (aKO) was associated with attenuation of the cytokine storm and reactive gliosis typically triggered by SE, thereby preserving hippocampal structural integrity, synaptic density, and long-term cognitive function. In contrast, WT and nKO mice exhibited severe upregulation of these cytokines and extensive reactive gliosis. These results align with emerging evidence that astrocytes act as the primary immune sensors mediated by 2-AG signaling in the epileptic brain, capable of orchestrating both the initiation and resolution of neuroinflammatory responses through their extensive spatial coverage and metabolic coupling with neurons [54–57]. The significant phenotypic differences observed between nKO and aKO mice suggest that astrocytic MAGL acts as an important endogenous mediator in shaping neuroinflammatory responses to excitotoxic insults. However, we note that this interpretation is derived from comparisons of constitutive knockout lines, and additional mechanisms may contribute to the observed effects. By deleting Mgll in astrocytes, we likely increase the local availability of 2-AG, which potentially engages a feed-forward protective mechanism that arrests the polarization of astrocytes into neurotoxic states. This is consistent with previous reports indicating that eCB signaling can suppress astrocyte activation in models of traumatic encephalopathy [27] and Alzheimer’s disease [58], and our findings extend this protective role to the acute excitotoxic environment of SE. Moreover, the spatial compartmentalization of 2-AG signaling may be particularly important in this context. Unlike systemically administered cannabinoids that activate CB1 receptors throughout the brain, astrocytic MAGL deletion may create microdomains of elevated 2-AG specifically in perisynaptic astrocytic processes, allowing for precise spatiotemporal control of both neuronal excitability and glial reactivity [19]. Our single-nucleus RNA sequencing (snRNA-seq) analysis provides a high-resolution mechanistic explanation for this anti-inflammatory phenotype. In WT mice, KA induction drove astrocytes toward a reactive and inflammatory trajectory (Cluster 5) and microglia toward a Disease-Associated Microglia (DAM) phenotype (Cluster 2) implicated in aberrant synaptic pruning [59]. Strikingly, aKO astrocytes largely bypassed this inflammatory fate, instead populating a distinct metabolic state (Cluster 2) characterized by enhanced oxidative phosphorylation and lipid metabolism pathways. This metabolic reprogramming resembles the neuroprotective astrocyte phenotype, where astrocytes upregulate bioenergetic support to sustain neurons during stress [60, 61]. This metabolic reprogramming may intersect with endogenous antioxidant programs, as astrocytic Nrf2 signaling has been shown to play a distinct and complementary neuroprotective role in both acute and chronic epilepsy models, and has been proposed as a disease-modifying therapeutic target [62–65]. This suggests that Mgll deletion does not simply silence astrocytes but actively repurposes them to support neuronal bioenergetics during stress, potentially through enhanced lactate shuttle activity, provision of metabolic substrates, and engagement of antioxidant transcriptional programs [66]. Furthermore, the blunted DAM signature observed in microglia of aKO mice was accompanied by this altered astrocyte state. We interpret this as a correlative finding consistent with the emerging concept that astrocyte-microglia crosstalk shapes neuroinflammatory outcomes in various disease contexts [67, 68]. Whether this microglial shift reflects direct astrocyte-driven reprogramming, a secondary response to reduced neuronal injury, or both, cannot be resolved from the current data and warrants direct investigation in future studies. Mechanistically, our pharmacological rescue experiments clarify the signaling pathways underlying this protection. The complete loss of efficacy in CB1 knockout mice confirms that CB1 receptor activation is the prerequisite for 2-AG-mediated protection in this model, consistent with extensive literature demonstrating CB1 as the primary mediator of endocannabinoid effects in the brain [7, 14, 15, 69]. However, as CB1R knockout mice lack CB1 in all cell types, the specific cellular locus of this CB1-dependent effect, whether neuronal, astrocytic, or both, remains to be defined, and cell-type specific CB1 deletion will be required to resolve this question in future work. A previous study has shown that CB2 receptors, which are primarily expressed in microglia, can modulate neuroinflammatory responses following seizures and may contribute to the protective effects of 2-AG beyond CB1 signaling [7]. In our study, however, JZL-184 did not provide detectable protection in CB1 knockout mice, suggesting that CB1-dependent mechanisms predominate in this model. The partial reduction of JZL-184 protection with PPARγ antagonism and the reduction of aKO protection by astrocyte-specific PPARγ knockdown using an AAV approach, together implicate astrocytic PPARγ as a functionally relevant downstream node in this pathway. The partial rather than complete rescue by both approaches suggests that PPARγ represents one of likely several parallel effector mechanisms rather than a single obligatory downstream mediator. These findings support a hypothesis of dual-mechanism of endocannabinoid signaling: an immediate, membrane-mediated modulation of excitability, followed by a sustained, anti-inflammatory genomic effect mediated by the intracellular PPARγ signaling pathway. The rapid, membrane-proximal effects likely involve CB1-mediated inhibition of adenylyl cyclase, reduction of cAMP levels, and suppression of presynaptic calcium influx [70, 71], while the slower PPARγ-dependent effects involve transcriptional reprogramming [72]. It is this secondary axis that appears crucial for preventing the cascade of reactive gliosis and secondary neurodegeneration. Previous studies also showed that the anti-neuroinflammation effect of astrocyte-specific MGL-deficiency cannot be fully abrogated by the inverse CB1 receptor agonist [73]. By engaging nuclear receptors, accumulated endocannabinoids likely repress NFκB and other inflammatory transcription factors [22, 74], preventing the transformation of the hippocampus into a pro-inflammatory environment that facilitates secondary neurodegeneration. In parallel, inhibition of MAGL is expected to limit the production of arachidonic acid and its downstream pro-inflammatory mediator PGE2, thereby further attenuating inflammatory signaling [75]. Indeed, PPARγ activation has been shown to suppress the expression of COX-2, iNOS, and pro-inflammatory cytokines in epilepsy, providing a plausible molecular mechanism for the anti-inflammatory phenotype we observed [76]. The downstream consequences of this immunometabolic rescue are evident in the preservation of neuronal viability and identity. The CA1 subfield, typically vulnerable to metabolic collapse during SE, due to its high energy demands and susceptibility to excitotoxicity [77], exhibited preserved mitochondrial gene expression and structural integrity in aKO mice. Similarly, the maintenance of neuronal markers and synaptic proteins in the chronic phase (Day 21) suggests that by limiting the acute inflammatory insult, aKO prevents the maladaptive synapse loss and circuit remodeling associated with chronic post-SE pathology [78, 79]. The functional preservation of memory and prevention of hyperactive behavior in the chronic phase further validates that the initial suppression of neuroinflammation translates into durable circuit integrity and behavioral outcomes. These findings align with complementary disease-modifying strategies targeting oxidative stress, protein homeostasis, and immune-mediated neuronal injury in epilepsy [80–82]. Finally, we acknowledge important limitations of the current translational framing. The cell-type specific genetic knockout models used here represent constitutive preconditioning paradigms rather than clinically actionable interventions. To address this, we demonstrate that post-SE systemic administration of JZL-184 also reduces glial activation, providing proof-of-concept that pharmacological MAGL inhibition retains therapeutic relevance when applied after the initial insult. Nevertheless, the potential liabilities of chronic MAGL inhibition warrant careful consideration. Prolonged MAGL inhibition is associated with CB1 receptor desensitization and downregulation following sustained 2-AG elevation and functional tolerance as previously reported with chronic JZL-184 administration [83]. These considerations suggest that intermittent or targeted dosing strategies may be necessary to preserve therapeutic efficacy while minimizing adverse adaptations, and dose-optimization studies will be an important direction for future work. Summary Our study redefines the cellular logic of endocannabinoid neuroprotection. We propose that astrocytes serve as the critical reservoir for 2-AG-mediated resilience. By deleting Mgll in astrocytes, we effectively uncouple the seizure insult from its devastating inflammatory sequelae. This blockade of glial reactivity preserves the metabolic and structural support systems required for neuronal survival. Consequently, targeting astrocytic endocannabinoid metabolism represents a promising therapeutic strategy to mitigate seizure-induced neuroinflammatory injury and preserve hippocampal integrity after status epilepticus. Data availability Source data are provided with this paper. Additional data supporting the findings of this study are available from the corresponding authors upon reasonable request. The single-nucleus RNA sequencing (snRNA-seq) data have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE318570 and are freely available to the public. Supplementary Information Supplementary Material 1. Supplementary Material 2. Acknowledgements This work was supported by startup and endowment funds from Joe R. & Teresa Lozano Long School of Medicine at UT Health San Antonio (to C.C.). The authors thank the NIMH Chemical Synthesis and Drug Supply Program for providing JZL184. Authors’ contributions C.Z. and C.C. conceived the project and designed the experiments. C.Z., L.S., M.H., J.L., D.Z., F.G., M.P., J.Z. and C.C. performed the experiments and/or analyzed the data. C.Z. and C.C. wrote the manuscript. Data availability Source data are provided with this paper. Additional data supporting the findings of this study are available from the corresponding authors upon request. The single-nucleus RNA sequencing (snRNA-seq) data have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE318570 and are freely available to the public. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Data Availability Statement Source data are provided with this paper. Additional data supporting the findings of this study are available from the corresponding authors upon reasonable request. The single-nucleus RNA sequencing (snRNA-seq) data have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE318570 and are freely available to the public. Source data are provided with this paper. Additional data supporting the findings of this study are available from the corresponding authors upon request. The single-nucleus RNA sequencing (snRNA-seq) data have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE318570 and are freely available to the public.