Astrocytes, the most numerous glial cells in the CNS, establish an intricate spatial organization, tiling the parenchyma with distinct yet interacting domains and extending elaborate processes to contact multiple cellular elements including synapses, neuronal somata, other glia, and the vasculature (21–23). This complex spatial arrangement underpins their critical homeostatic functions, but becomes profoundly disrupted during epileptogenesis, positioning astrocyte dysfunction within specific microenvironmental niches as a central element in the disease process (24–26) (Figure 1).

Microglia, the resident myeloid cells of the CNS, are not static sentinels but dynamically survey their allocated parenchymal territories via constant motile processes (65, 66). In the context of epilepsy, their activation state, spatial distribution, and function become profoundly altered, varying significantly depending on proximity to the epileptic focus, the specific brain region, and the stage of epileptogenesis (67–70). This spatial heterogeneity underscores the importance of the local microenvironment in dictating microglial roles, positioning them as critical modulators and potential drivers of pathological processes.

Besides the well-studied astrocytes and microglia, oligodendrocytes are vulnerable components of the spatial immune niche during epilepsy. In lithium-pilocarpine models, hippocampal demyelination and mature oligodendrocyte loss initiate during the acute phase and progressively worsen throughout epileptogenesis. Intriguingly, this degeneration triggers a compensatory, yet transient, regenerative response: oligodendrocyte precursor cells proliferate and upregulate the key differentiation factors Olig1 and Olig2 during the acute and latent windows (93). However, this repair mechanism collapses in the chronic phase as Olig1/2 expression plummets, suggesting that the persistent inflammatory microenvironment eventually overwhelms the lineage’s reparative capacity, cementing white matter injury as a permanent driver of circuit instability.

Among myeloid cells, CCR2⁺ inflammatory monocytes are the most consistently implicated in seizure-linked neuroinflammation. After status epilepticus, CCL2–CCR2 signaling drives robust recruitment of blood-borne monocytes that interdigitate with activated microglia around vessels and within hippocampal laminae; genetic or pharmacologic interruption of this axis mitigates neuronal loss and downstream behavioral sequelae (95–97). Notably, the administration of small-molecule CCR2 antagonists (e.g., RS504393, RS102895) or CCL2 synthesis inhibitors (e.g., bindarit) effectively blocks the ingress of blood-borne monocytes (97, 98). In kainic acid models, these interventions interrupt the downstream Signal Transducer and Activator of Transcription 3 (STAT3)/IL-1β signaling cascade, significantly reducing hippocampal neuronal death and attenuating the severity of spontaneous recurrent seizures (97). Furthermore, recent findings identify a critical therapeutic window where brief CCR2 antagonism post-status epilepticus prevents long-term cognitive comorbidities; this “fleeting” inhibition is sufficient to rescue deficits in working and retention memory and preserve BBB integrity (98, 99). While the impact on seizure arrest can be model-dependent, with studies in pilocarpine rats demonstrating neuroprotection and CA1 volume preservation without a reduction in spontaneous seizure frequency (100), the consensus across models highlights that targeting monocyte infiltration effectively creates a neuroprotective environment and disrupts the inflammatory maintenance of the epileptogenic niche. This vascular interface serves as a critical spatial niche where infiltrating monocytes integrate into a localized neuro-glial signaling loop. Perivascular macrophages at these entry sites not only secrete CCL2 and present antigens but also respond to astrocyte-derived CCL2, collectively enhancing monocyte recruitment and modulating T-cell responses. This activity further aggravates BBB disruption, facilitating increased infiltration of peripheral immune cells. These functions that underscore how “where” these cells act (at the vascular border) is inseparable from “what” they do (amplify leukocyte in-flux and antigen surveillance) (12).

Two mechanistic threads are worth emphasizing for an epileptogenesis framework. First, monocyte-derived macrophages sustain IL-1β/STAT3 signaling cascades that degrade astrocytic K⁺/glutamate buffering and weaken inhibitory drive, embedding hyperexcitability into local circuits (97). Second, border macrophages can loosen BBB properties and remodel the perivascular niche, further lowering the threshold for repeated waves of leukocyte entry during recurrent seizures (12, 101). These data argue for compartment-directed interventions—e.g., CCR2 antagonism or perivascular macrophage modulation—as disease-modifying strategies aimed at the gateways that nucleate pro-convulsant microenvironments.

T cells also accumulate in specific places and windows, and their programs map onto those spaces. In pediatric drug-resistant focal epilepsies, specifically Type II Focal Cortical Dysplasia, resected lesion centers from the frontal and temporal lobes contain abundant memory CD4⁺ and CD8⁺ T cells together with γδ T cells that produce IL-17 and GM-CSF; by contrast, FoxP3⁺ regulatory T cells (Tregs) are present but tend to track inversely with inflammatory burden. Functionally, IL-17 and GM-CSF from γδ cells increase neuronal firing and compromise neuronal viability in slice preparations, while the Treg compartment restrains this inflammatory tone—an arrangement that captures, within a single lesion, how effector and regulatory T cells can pull in opposite directions (102). Human temporal lobe epilepsy associated with glutamic acid decarboxylase (GAD) antibodies adds a temporal dimension: early “encephalitic” phases show dense parenchymal and perivascular lymphocytic cuffs, dominated by granzyme-B⁺ CD8⁺ cytotoxic T cells and plasma cells in hippocampus; later “low-activity” phases exhibit a thinner immune presence despite ongoing clinical disease (103). The causal link from CD8⁺ attack to epilepsy is supported by a neuron-targeted limbic encephalitis model in which antigen expression confined to CA1 leads to priming of CD8⁺ T cells in deep cervical lymph nodes, a wave of hippocampus-restricted CD8⁺ infiltration within the first week, albumin leakage at the BBB, acute seizures and memory impairment, and, ultimately, chronic spontaneous seizures (103). Taken together, these human and mechanistic data converge on the same principle: discrete T-cell populations gather in defined hippocampal and cortical compartments and, there, set the stage for enduring epileptogenic niches (102–104).

CD4⁺ T cells contribute a second, clinically relevant axis. In adults with temporal lobe epilepsy, a peripheral Th1/Th2 skew (increase in IFN-γ, decrease in IL-4) is mirrored in the brain by increased CD4⁺ infiltration with frequent adjacency to Iba1⁺ microglia in hippocampus and neocortex. Mechanistically, Notch1 emerges as a regulatory switch for both Th1 differentiation and CD4⁺ trafficking across a permeable barrier; inhibiting Notch1 dampens microglial activation, reduces seizures, and improves cognition in experimental systems (105). On the regulatory side, Tregs are not merely present—they are prognostically meaningful. In patients and models, higher Treg numbers within epileptogenic zones associate with fewer seizures, whereas experimental Treg depletion worsens gliosis, cytokine production, neuronal loss, and seizure burden; conversely, expanding Tregs in the brain suppresses seizures (16). Together, these strands argue that the composition and positioning of T-cell subsets—CD8⁺ in hippocampal laminae, Th1-skewed CD4⁺ adjacent to microglia, γδ cells and Tregs competing within lesion cores—are central to how immune pressure becomes epileptogenic (16, 102, 105) (Figure 4).

These altered T lymphocyte balances are frequently mirrored in the periphery. Flow cytometric analyses of peripheral blood from patients with intractable epilepsy reveal a significant systemic peripheral T cell imbalance, characterized by an elevated Th17/Treg ratio. This skewing is functionally critical because the excess Th17 cells secrete IL-17A, which disrupts BBB integrity, and simultaneously, the reduction in Tregs diminishes the systemic ‘braking’ of inflammation (106, 107). Notably, this ratio is dynamic; therapeutic interventions such as the ketogenic diet have been shown to restore the Th17/Treg balance towards a healthy baseline, correlating with reduced seizure frequency (107).

At juxtavascular sites, endothelial cells and perivascular macrophages concentrate chemokines and present antigen, drawing T cells and CCR2⁺ monocytes into perivascular sleeves that lie against astrocytic endfeet and microglial processes (12) (Figure 4). Within these sleeves and the adjacent parenchyma, CD8⁺ T cells form synapses with MHC-I–positive neurons and execute cytotoxic programs; in GAD-antibody temporal lobe epilepsy, neuronal death is predominantly T-cell mediated, not complement-driven (103). Th1-skewed CD4⁺ cells, frequently in situ neighbors of microglia, license microglial and monocyte-derived macrophage responses through IFN-γ and TNF, amplifying IL-1 family signaling and pushing inhibitory–excitatory balance toward hypersynchrony. Meanwhile, CCR2-dependent monocyte influx reinforces this state through STAT3-linked IL-1β production that erodes local astrocytic and synaptic homeostasis (96, 97). In short, infiltrating lymphoid and myeloid cells organize pro-convulsant glial states at vascular and synaptic borders. That organization—and its tight spatial footprint—helps explain why small portions of hippocampus or neocortex can become durable seizure generators even as neighboring tissue remains comparatively unaffected.

Additionally, some often-overlooked contributors deserve mention in this spatial frame, which are the B cells, neutrophils, and mast cells. For B-lineage cells, the clearest human signal remains plasma-cell-rich early stages of GAD-antibody temporal lobe epilepsy; although antibodies may be bystanders to CD8-mediated injury, their presence delineates an active, perivascular/parenchymal niche and—importantly—marks a window where immunotherapy can still modify trajectory (103). While B cells have received less attention than their T-cell counterparts, they likely play a multifaceted role in the epileptogenic niche that extends beyond systemic autoantibody production. In addition to differentiating into antibody-secreting plasma cells—which cluster in perivascular spaces during the early active phases of autoimmune epilepsies—B cells can function as potent antigen-presenting cells to CD4+ T cells (108). Furthermore, they serve as a localized source of pro-inflammatory cytokines, including IL-6, GM-CSF, and TNF-α. This allows B cells to sustain local inflammatory loops and promote BBB permeability even in the absence of specific neuronal autoantibodies, acting as ‘architects’ of the chronic immune microenvironment rather than mere transient visitors. Next, neutrophils adhere to cerebral microvessels after seizures through LFA-1/ICAM-1 interactions and prolong postictal hypoperfusion. Blocking neutrophil adhesion normalizes cerebral blood flow, implicating early neutrophil–endothelial dynamics in vascular dysfunction at seizure foci (109). Emerging work on neutrophil extracellular traps suggests they aggravate BBB permeability in CNS disease; experimental NET inhibition protects barrier function and cognition, raising a plausible link to seizure-provoked BBB fragility that merits direct testing in epilepsy (110, 111). Finally, mast cells, which are the early vascular sentinels, can loosen the BBB via histamine and tryptase; while epilepsy-specific evidence is still limited, seizure-linked histamine release and benefit from mast-cell stabilization in experimental status suggest a small but potentially catalytic role at the breach (112, 113). That organization—and its tight spatial footprint—helps explain why small portions of hippocampus or neocortex can become durable seizure generators even as neighboring tissue remains comparatively unaffected (Figure 4, left and middle panels).

A critical question is whether persistent inflammatory niches contribute to epileptogenesis and pharmaco-resistance when immune resolution fails. Evidence suggests indeed they do. Chronically epileptic regions often show enduring, focal “smoldering” inflammation with microglial activation, gliosis, elevated cytokines, and ongoing trafficking of peripheral cells (114). In cohorts with epilepsy of unknown etiology, neural autoantibody prevalence is ~7–8% on average, with some series reporting up to ~15–20% depending on inclusion criteria, supporting an occult immunologic contribution in a subset rather than one-third across the board (115, 116). One consequence is BBB transporter remodeling that undermines antiepileptic drug efficacy. HMGB1, released by neurons and glia under chronic seizure stress—induces BBB P-glycoprotein via TLR4/RAGE-NF-κB signaling, enhancing drug efflux and lowering intraparenchymal drug levels (117, 118). This pathological state is then ‘locked in’ by short-range feedback loops, including the upregulation of multidrug transporters like P-glycoprotein, which physically and functionally reinforce the inflammatory architecture (Figure 4, right panel). Consistently, refractory patients and models show pro-inflammatory cytokine upregulation accompanied by multidrug transporter increases. Notably, COX-2 inhibition can restore phenobarbital anticonvulsant efficacy in a rat model of pharmacoresistant temporal lobe epilepsy, illustrating a direct link between unchecked inflammation and drug resistance (119). Conversely, enhancing immune resolution—e.g., boosting Tregs or recruiting them to foci via CCL20—dampens neuroinflammation and reduces seizures in vivo (16). Taken together, these data support immunomodulatory strategies (targeting CCR2⁺ monocytes, pro-inflammatory T cells, or augmenting Tregs) as disease-modifying adjuncts aimed at dismantling the inflammatory architecture that supports chronic seizures.

Besides, in cohorts with epilepsy of unknown etiology, neural autoantibody prevalence is ~7–8% on average, supporting an occult immunologic contribution in a subset rather than one-third across the board. Within this seropositive population, the pathogenic mechanisms vary significantly. While the frequently identified GAD65 antibodies typically serve as markers for T-cell mediated cytotoxicity, other autoantibodies directly disrupt synaptic function. Specifically, antibodies targeting the glycine receptor have been identified in patients with progressive epilepsy variants that are notably responsive to immunotherapy (120, 121). Mechanistically, these antibodies exert a direct pathogenic effect: in vitro studies using cultured neurons demonstrate that they can directly antagonize glycine receptor currents and induce receptor internalization, thereby disinhibiting neural circuits and promoting hyperexcitability. This highlights a direct functional link where the immune effector (the antibody) physically obstructs the molecular machinery of inhibition (122).

In this review, we advance a spatial thesis of epileptogenesis: seizures emerge and persist because small, anatomically constrained inflammatory niches destabilize circuit homeostasis. We argue that in epilepsy, “where” interactions occur conditions both “what” cells are present and “how” they communicate. Evidence across models and human tissue aligns with a cascade in which focal BBB leakage seeds a perivascular niche; albumin/TGF-β signaling at astrocytic endfeet disrupts Aquaporin-4 (AQP4)/Inwardly rectifying potassium channel 4.1 (Kir4.1) polarity and gap-junction coupling; microglia re-cluster around stressed synapses and altered extracellular matrix; and recruited leukocytes reinforce the same neighborhoods through cytokines and matrix metalloproteinases. These events co-localize in perivascular sleeves, lesion rims, and layer-specific strata, generating self-reinforcing microdomains of local neuroinflammation and neuronal hyperexcitability.

A spatial lens helps reconcile why seemingly “diffuse” neuroinflammation yields focal seizure onset zones. Astrocytes that are well-coupled can dissipate ionic and metabolic burden over distance; when their network fragments, the same burden accumulates locally and tips neurons toward pathological firing. Microglia, guided by ATP and chemokine gradients, concentrate where astrocytic polarity and synaptic support have faltered, shifting from homeostatic surveillance to short-range modulation of synapses and perineuronal nets. Peripheral cells do not simply flood the parenchyma; their ingress is anatomically gated at leaky microvessels, and their effector functions—proteolysis, cytokine release, antigen presentation—remain most active near those gates. The result is not global immune activation but small neighborhoods with the right geometry and cell mix to sustain excitability.

This framework also recontextualizes pharmacoresistance as a spatial mismatch between therapy and pathology. While ASMs effectively reduce neuronal excitability, they typically fail to address the underlying structural deficits—such as BBB leakage, astrocyte uncoupling, and extracellular matrix degradation—that allow hyperexcitable conditions to regenerate once the drug’s effect subsides. Drug exposure itself may be spatially heterogeneous, as lesion rims can express efflux transporters or harbor diffusion barriers, and an edematous or fibrotic matrix can alter local tissue pharmacokinetics. This pathological state is then “locked in” by short-range feedback loops, including dysregulated purinergic signaling, complement-mediated tagging of synapses, and TGF-β autocrine signaling in perivascular astrocytes. These observations strongly suggest that a combination therapy—pairing a conventional ASM with an intervention aimed at niche repair (e.g., stabilizing gliovascular signaling, restoring astrocyte coupling, or modulating microglial checkpoints)—would be more effective than monotherapy, particularly for treating persistent focal epilepsy.

First, modern spatial technologies render the hypothesis that epilepsy is a disease of inflammatory places, beyond diffuse neuroinflammation, directly testable (123). Spatial transcriptomics and multiplex imaging can simultaneously map barrier integrity, astrocyte polarity, microglial states, immune phenotypes, and the cellular interactions within a single tissue section, anchoring these molecular readouts to electrophysiological data and anatomical signatures. By carefully sampling the seizure focus, its surrounding rim, and contralateral tissue, such datasets can generate composite biomarkers—like perivascular disruption indices or astrocyte network integrity scores—that can stage disease and stratify patients. “Window-of-opportunity” clinical trials in surgical candidates are especially promising; administering a short, spatially targeted intervention before resection would allow for direct, on-tissue assessment of pharmacodynamic effects (such as restored AQP4 polarity), providing causal evidence that conventional trials often lack.

Moreover, the spatial heterogeneity of neuroinflammation offers a mechanistic rationale for the efficacy of neuromodulatory therapies such as vagus nerve stimulation, Deep Brain Stimulation, and responsive neurostimulation. Beyond their electrical effects on neuronal synchrony, these interventions may exert anti-inflammatory effects via the cholinergic anti-inflammatory pathway (VNS) or by locally modulating the glial milieu at the stimulation site (DBS). Future trials could leverage spatial biomarkers to identify patients with “inflammatory” focal architectures who might benefit most from targeted neurostimulation that doubles as immunomodulation.

Lastly, the spatial model also illuminates the roles of often-overlooked cell types. Neutrophils, appearing early after a BBB breach, can initiate proteolytic cascades that amplify leakage. Monocytes and perivascular macrophages shape the local chemokine environment and orchestrate antigen presentation at these gateways. The activity of T cell subsets can either propagate or resolve inflammation depending on local cues, while B cells and plasma cells may contribute regionally through antibody and cytokine production. These contributions are unlikely to be uniform; instead, they probably reflect the specific architecture and vascular supply of a given brain region, helping to explain why diverse etiologies like developmental lesions, post-infectious scarring, and autoimmune encephalitis can converge on similar clinical phenotypes.

However, it should be noted that important limitations remain. Spatial assays often trade comprehensive coverage for high resolution and are subject to the inherent biases of RNA and protein capture from a 2D slice of a dynamic 3D process. Tissue handling, fixation, and perioperative timing can introduce artifacts that may be mistaken for biological signals. Therefore, improving reproducibility will require robust cross-platform integration, batch-aware computational analysis, and the transparent preregistration of spatial endpoints. Mechanistically, it remains critical to determine which features are foundational to niche formation versus those that merely maintain it, and how factors such as brain region, patient age, and comorbidities influence these spatial dynamics.