Abstract

Background:

Dravet syndrome (DS) is a severe developmental and epileptic encephalopathy, mainly caused by SCN1A gene mutations. Its core characteristics are heat sensitivity and refractoriness, and immunoinflammatory factors can participate in the occurrence and development of the disease. At present, the regulatory role of immune inflammation activation in DS has been confirmed, but the specific molecular core connecting systemic inflammation and central nervous system signals and its translational relevance to broader pediatric drug-resistant epilepsy (DRE) remains unclear.

Methods:

We conducted a multi-level integrative analysis combining transcriptomic mining of the GEO database to identify the Stat3 hub, with clinical validation in a real-world pediatric cohort from our hospital, comparing DRE (including DS) and self-limited epilepsy with centrotemporal spikes (SeLECTS), to assess the clinical relevance of systemic inflammatory indices (NLR, SII, CRP). Findings were mechanistically verified in Scn1a+/− mice via qRT-PCR, Western blotting, and immunofluorescence, using robust linear models to confirm central-peripheral inflammation correlations.

Results:

Transcriptomic profiling of Scn1a+/− mice revealed a distinct inflammatory landscape (PC1 = 86%) dominated by JAK-STAT signaling, with Stat3 identified as a consensus hub. Clinically, this systemic inflammatory signature was observed in our pediatric cohort (n = 140). Baseline inflammatory indices (NLR, SII, CRP) were significantly elevated in patients with drug-resistant epilepsy compared to those with SeLECTS (p < 0.001). Multivariable analysis further identified CRP as an independent factor closely associated with progression to drug resistance (OR = 2.79, p = 0.025). In vivo validation confirmed p-STAT3 hyperactivation in hippocampal gliosis (p < 0.0001), which exhibited robust linear correlations with peripheral markers (r ≥ 0.94, p < 0.001).

Conclusion:

This study identifies systemic and neuroinflammatory changes in DS associated with increased STAT3 signaling and this inflammatory signature is also observed in the broader pediatric DRE population. By bridging verified molecular mechanisms with real-world clinical data from our pediatric cohort, we suggest peripheral indices (NLR, SII, CRP) that may serve as accessible clinical indicators of disease severity in pediatric DRE. Pending functional validation, these findings identify STAT3 as a pathway of interest and a potential therapeutic candidate, supporting the development of adjunctive anti-inflammatory therapies targeting neuroimmune cascades for DS and broader refractory epilepsies.

1 Introduction

Epilepsy affects approximately 1% of the global population, with SCN1A mutations identified as a predominant genetic driver (Xie et al., 2020; GBD Epilepsy Collaborators, 2025), accounting for 38% of monogenic cases in our prior cohort (ChiCTR1900022164) (Wang et al., 2025). The most severe manifestation of these mutations is Dravet syndrome (DS), a catastrophic infantile-onset developmental and epileptic encephalopathy (Peycheva et al., 2020). While SCN1A haploinsufficiency and subsequent NaV1.1-dependent interneuron failure are primary etiologies, genotypic uniformity often fails to predict the substantial clinical heterogeneity in DS (Zhang et al., 2025). This divergence implicates pathogenic drivers beyond the ion channel itself, redirecting focus toward malleable pathways that intersect with network excitability.

Emerging frameworks position neuroinflammation not merely as a reactive consequence, but as an active propellant of epileptogenesis. In Dravet models, seizure recurrence correlates with aberrant hippocampal inflammation and blood–brain barrier compromise, creating a permissive environment for immune cell infiltration (Zhao et al., 2022). These inflammatory cascades likely sustain a pro-excitatory milieu rather than simply accompanying neural activity (Brenet et al., 2024). Within this landscape, the JAK2/STAT3 signaling axis emerges as a critical integration node, coupling cytokine stimulation with transcriptional programs that perpetuate glial activation and circuit vulnerability.

However, the precise architecture of STAT3 signaling within the SCN1A-deficient, fever-sensitive brain remains opaque. Although STAT3 is an established immunomodulator, its specific function as a network hub bridging neuroimmune activation and seizure susceptibility in DS has not been systematically delineated (Zimmerman et al., 2024). It remains unclear whether these accessible systemic markers can serve as reliable mirrors of the central STAT3-driven pathology. This necessitates a rigorous interrogation of the STAT3 pathway to determine its role as both a mechanistic key regulatory node and a translational target.

Here, we interrogate the STAT3-centered regulatory landscape in DS through a convergent bioinformatics and experimental framework. By integrating transcriptomic data from Gene Expression Omnibus (GEO) transcriptomes with clinical validation from a real-world pediatric cohort of drug-resistant epilepsy (DRE) and self-limited epilepsy with centrotemporal spikes (SeLECTS, a common age-dependent focal epilepsy with favorable outcomes serving as the clinical baseline control), we identify STAT3 as a signaling component associated with secondary neuroinflammation and evaluate its association with accessible systemic surrogates. Crucially, we functionally verify these findings in Scn1a+/− mice. This integrative strategy provides biological context for the observed correlations, decoding the intersection of STAT3-mediated secondary inflammation and SCN1A deficiency to evaluate these systemic markers as reliable mirrors of central pathology.

2 Materials and methods2.1 Study design and workflow

This study was conducted as a multi-level integrative analysis to investigate the inflammatory and immune mechanisms underlying DS. Initially, a transcriptomic dataset from the Gene Expression Omnibus (GEO) was integrated to identify inflammation-related differentially expressed genes (DEGs) and screen for core upstream transcription factors (TFs). To establish the clinical and epidemiological relevance of these molecular findings, systemic inflammatory markers including the neutrophil-to-lymphocyte ratio (NLR), systemic immune-inflammation index (SII), and C-reactive protein (CRP) were evaluated for their association with epilepsy risk using a clinically matched pediatric cohort from our institution. Furthermore, the molecular and clinical evidence was verified through in vivo experiments using a Scn1a+/− mouse model, which involved characterizing neuroinflammatory pathology in the hippocampus and cortex via quantitative PCR, Western blotting, and immunofluorescence, as well as assessing the correlation between central neuroinflammation and peripheral inflammatory profiles. The overall workflow, encompassing data integration, population-level validation, and experimental verification, is illustrated in Figure 1.

Flowchart of the study design. The workflow integrates transcriptomic discovery of inflammatory drivers, clinical validation of systemic markers (NLR, SII, CRP) in a pediatric clinical cohort, and in vivo verification of the JAK-STAT3 axis and glial activation in Scn1a+/− mice.

2.2 Data acquisition and processing

Mining public transcriptomic datasets from the GEO database has proven to be a highly effective and reliable strategy for elucidating core molecular mechanisms and identifying novel therapeutic targets across various central nervous system (CNS) disorders (Zha, 2025). Raw RNA-sequencing data were retrieved from the GEO database under accession number GSE112627. Concurrently, an independent transcriptomic dataset (GSE289689) was incorporated for supplementary support, with its detailed bioinformatic workflow available in the Supplementary Methods. The GSE112627 dataset comprised six samples (Supplementary Table S1), with gene-level quantification performed using HTSeq-count. Raw count data were assembled into a global expression matrix using R software (v4.4.1). Prior to differential expression analysis, low-abundance genes (defined as having a count <10 in more than 50% of samples) were filtered out to reduce noise and improve statistical power.

2.3 Differential expression analysis

Differential gene expression analysis between Scn1a+/− and wild-type (WT) mice was performed using the DESeq2 package (v1.44.0). To ensure statistical robustness, p-values were adjusted using the Benjamini–Hochberg procedure to control the false discovery rate (FDR). Significant differentially expressed genes (DEGs) were defined by an adjusted p-value <0.05 and an absolute log2 fold change (|log2FC|) >0.585 (corresponding to a 1.5-fold change).

For dimensionality reduction and quality assessment, raw counts were normalized using the variance stabilizing transformation (VST). Principal component analysis (PCA) and uniform manifold approximation and projection (UMAP, n_neighbors = 5) were utilized to visualize global sample clustering and distinct group separation, with 95% confidence ellipses added via the ggforce package to delineate group boundaries. To visualize the overall distribution of DEGs, volcano plots were generated using ggplot2. Additionally, hierarchical clustering analysis of the top 100 upregulated genes was visualized using the pheatmap package (scaled by row Z-score) to highlight distinct expression signatures.

2.4 Functional enrichment and transcription factor analysis

Gene Ontology (GO), biological process and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed using the clusterProfiler package (v4.12.6). For significantly upregulated genes, gene symbols were converted to Entrez IDs using org.Mm.eg.db (v3.21.0). An FDR threshold of <0.05 was applied to identify significantly enriched terms and pathways.

To identify potential upstream regulators, all significantly upregulated genes were subjected to TF enrichment analysis using the Enrichr web tool. Predictions were sourced from three curated libraries: “ENCODE and ChEA Consensus TFs from ChIP-X,” “TRRUST Transcription Factors 2019,” and “Transcription Factor PPIs.” For each library, candidates were ranked by p-value, and the top 20 TFs from each were intersected to identify robust consensus regulators.

2.5 Construction of transcriptomic surrogates for CNS immune activation

To conceptualize the central inflammatory microenvironment, gene-set-based transcriptomic surrogates were constructed using variance stabilizing transformation (VST) normalized values. These indices are designed to reflect broader CNS myeloid and lymphoid activation states. Canonical marker genes were aggregated to represent the transcriptomic signatures of myeloid/neutrophil-like activation (Sneu: S100a9, Fcgr3, Csf3r), lymphoid activation (Slym: Lck, Zap70), and platelet-related pathways (Splt: Itga2b, Gng11). The central transcriptomic surrogates (transcriptomic NLR and transcriptomic SII) were calculated mathematically homologous to peripheral indices, serving as proxies for the central neuroimmune burden.

The expression levels of Stat3 and the CRP-surrogate Ptx3 were normalized using the same VST scale.

2.6 Pediatric clinical cohort validation

To validate the clinical translatability of the identified inflammatory signatures specifically in the context of severe pediatric epilepsy, a retrospective clinical cohort analysis was conducted at our center. The analytical cohort comprised 140 pediatric patients hospitalized between 2010 and 2025, divided into a DRE (n = 69... and a SeLECTS control group (BPE, n = 71).The systematic screening and enrollment workflow for this pediatric cohort is delineated in the flowchart provided in Supplementary Figure S1.

To delineate the clinical cohort and isolate disease-associated immune signatures, we employed a retrospective longitudinal design with stringent gating criteria. Inclusion necessitated: (1) a definitive clinical diagnosis of DRE or SeLECTS adjudicated by pediatric epileptologists; (2) age ≤14 years; and (3) the availability of complete serological profiles (inclusive of total white blood cell, neutrophil, lymphocyte, and platelet counts, alongside C-reactive protein) acquired at the initial hospital presentation, prior to the initiation of long-term poly-pharmacotherapy. To mitigate confounding systemic variables, patients were excluded if they presented with: (1) total leukocyte counts exceeding age-calibrated physiological limits, or concurrent clinical evidence of acute systemic infections; (2) preexisting primary hematological or systemic autoimmune disorders; or (3) documented exposure to systemic glucocorticoids or adrenocorticotropic hormone (ACTH) within 4 weeks preceding serological sampling.

Systemic inflammatory indices (NLR, SII, and CRP) were derived from serological profiles obtained at the initial clinical presentation. This approach was utilized to evaluate the association between the early post-seizure peripheral inflammatory burden and the subsequent progression to drug resistance.

2.7 Animals and hyperthermia-induced seizure model

The animal study was reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of Beijing BioWork Technology Co., Ltd. (Approval No. BW-IACUC-2025-172). The Scn1atm1Kea mouse line (generously provided by Prof. Long-Jun Wu, Rutgers University) was maintained on a 129S6/SvEvTac background. To generate the experimental F1 hybrid cohorts, heterozygous (Scn1a+/−) mice were crossed with C57BL/6J wild-type mice. Genotyping was performed via PCR to identify Scn1a+/− mutants and wild-type (WT) littermates (Supplementary Figure S2C). The amplification utilized a three-primer system, including a common forward primer, a wild-type reverse primer, and a mutant-specific reverse primer, to distinguish the genotypes based on DNA fragment size. Detailed sequences for these genotyping primers are provided in Supplementary Table S2.

A total of 48 male F1 mice aged 3 weeks were assigned to the DS model group (Scn1a+/−, n = 24) and the control group (WT, n = 24). To assess seizure susceptibility, a hyperthermia-induced seizure protocol was employed. Briefly, core body temperature was elevated by 0.5 °C every 2 min using a controlled infrared heating lamp until a generalized seizure occurred or a threshold of 43 °C was reached. Seizure threshold temperature, duration, and severity (Racine scale) were recorded. Detailed housing conditions and induction procedures are provided in the Supplementary Methods.

2.8 Anesthesia, euthanasia, and tissue collection

To ensure animal welfare and experimental consistency, anesthesia and euthanasia were performed following a standardized protocol.

For terminal tissue harvesting, mice were deeply anesthetized via an intraperitoneal (i.p.) injection of the 1.25% (w/v) tribromoethanol solution at a dose of 250 mg/kg. The depth of anesthesia was confirmed by the loss of the pedal withdrawal reflex. Subsequently, mice were euthanized via transcardial perfusion with ice-cold phosphate-buffered saline (PBS) to remove systemic blood. The hippocampus and cortex were rapidly dissected on ice, snap-frozen in liquid nitrogen, and stored at −80 °C for molecular analysis. All brain tissues were harvested exactly 24 h after the hyperthermia-induced seizure to evaluate the acute-on-chronic neuroinflammatory response.

2.9 Peripheral inflammation assessment

Peripheral blood was collected from mice via retro-orbital puncture 3 days prior to euthanasia (which corresponds to 2 days prior to the hyperthermia-induced seizure protocol). This timeline was strictly established to capture the primary, genotype-driven baseline systemic inflammatory tone, excluding the acute physiological stress induced by hyperthermia. The NLR was determined from an automated complete blood count (CBC). Serum was separated by centrifugation (3,000 × g, 15 min, 4 °C) and stored at −80 °C for subsequent analysis of IL-6 and CRP levels using mouse-specific ELISA kits (ZC-137519, ZC-137382) per the manufacturer’s instructions.

2.10 Quantitative real-time PCR (qRT-PCR)

Total RNA was extracted from the hippocampus and cerebral cortex of Scn1a+/− and WT mice using AG RNAex Pro RNA (Accurate Biology, AG21102). RNA (1 μg) was reverse transcribed to cDNA in a 20 μL reaction mixture using the qRT-PCR was performed with SYBR Green Pro Taq HS qPCR Kit III (Accurate Biology, AG11739) using QuantStudio 7 Flex (ABI). Primers for qRT-PCR are listed in Supplementary Table S2. Meltcurve analysis confirmed transcript-specific amplification. Gene expression was normalized to GAPDH as an internal control. Relative mRNA expression was calculated using the comparative cycle threshold method (CT), also known as 2−ΔΔCT method.

2.11 Western blotting

Total protein was extracted from the hippocampus and cerebral cortex of Scn1a+/− and WT mice using RIPA buffer (Cell Signaling Technology, 9806) supplemented with protease inhibitors and phosphatase inhibitors. Protein yield was determined with a BCA kit (Cell Signaling Technology, 7780). Equal lysate aliquots containing 20 μg of total protein per lane were separated by 10% SDS-PAGE and electrotransferred to PVDF membranes. Blocking procedures were optimized based on target characteristics. Membranes dedicated to phosphorylated targets were blocked with 5% bovine serum albumin (BSA) in TBST for 1 h at room temperature, whereas 5% non-fat milk was applied for non-phosphorylated proteins. The membranes were then incubated with primary antibodies overnight at 4 °C. The following antibodies were used: p-STAT3 (1:500, Affinity, AF3293), STAT3 (1:1000, Affinity, AF6294), IL-6 (1:1000, Affinity, DF6087), p-JAK1 (1:500, Affinity, AF2012), JAK1 (1:1000, Affinity, AF5012), Cyclin D1 (1:1000, Affinity, AF0931), GFAP (1:1000, Affinity, DF6040), IBA1 (1:1000, Affinity, DF6442), and GAPDH (1:5000, ZenBio, 200306-7E4). After washing with TBST, the membranes were incubated with HRP-conjugated species-specific secondary antibodies for 1 h at room temperature. To ensure rigorous quantitative normalization, membranes initially probed for phosphorylated kinases (p-STAT3 and p-JAK1) were stripped using a commercial stripping buffer, thoroughly washed, re-blocked, and reprobed for their respective total proteins and the loading control on the exact same membrane. Protein bands were visualized using the Tanon 4800 imaging system, and band intensity was quantified using ImageJ software.

2.12 Double immunofluorescence staining and Nissl staining

The hippocampus and cerebral cortex of Scn1a+/− and WT mice were immersed in 10% neutral buffered formalin for 6 h, dehydrated in a gradient of ethanol, embedded in paraffin, and sectioned at a thickness of 5 μm. For double immunofluorescence staining, paraffin-embedded tissue sections were successively subjected to deparaffinization, initial antigen retrieval, peroxidase inactivation in regions of interest and permeabilization. Non-specific binding was blocked using 3% BSA at room temperature, followed by incubation with primary antibodies overnight at 4 °C. The primary antibodies used include NeuN (1:80, Affinity, DF6145), GFAP (1:500, Oasis Biofarm, OB-PGPO55-02), p-STAT3 (1:80, Affinity, AF3293), IBA1 (1:500, Oasis Biofarm, OB-PGPO49-02). Secondary antibodies include Goat Anti-Rabbit IgG H&L (AF488, ZenBio, 550037) and Donkey-anti-Guinea pig IgG (AF555, Oasis Biofarm, D-GP555). Following DAPI counterstaining and mounting, the sections were scanned and digitized using a ZEISS AxioScan7 digital slide scanner equipped with a 20× objective.

Paraffin sections were subjected to deparaffinization, and stained using the Nissl Stain Kit (Solarbio, G1430) per the manufacturer’s instructions. Slides were scanned on a digital pathology scanner (KF-PRO-120, KFbio).

2.13 Statistical analysis

Statistical analyses were performed using R software (version 4.4.1) for bioinformatics and clinical data validation, and GraphPad Prism (version 10.0) for experimental data visualization.

For comparisons between two groups, statistical significance was determined using the unpaired Student’s t-test for normally distributed data or the Mann–Whitney U test for non-normally distributed data.

To ensure statistical rigor, correlation analyses were tailored to the specific nature of the datasets. For the bioinformatics analysis of transcriptomic data, correlations were conducted using robust linear models (RLM) to mitigate the influence of potential biological outliers and background noise. Results for these transcriptomic analyses are reported as weighted correlation coefficients (weighted r) and p-values derived from robust t-tests. Conversely, for the in vivo experimental validation data (e.g., correlations between biochemical and histological quantitative assays), standard Pearson correlation analysis was utilized following the confirmation of normal data distribution.

For the pediatric clinical cohort, the normality of continuous variables was assessed using the Shapiro–Wilk test. Normally distributed data are presented as mean (standard deviation) and were compared using Student’s t-test. Skewed continuous variables, including hematological counts and systemic inflammatory indices (NLR, SII, CRP), are summarized as median (interquartile range, IQR) and were compared using the non-parametric Mann–Whitney U test. Categorical data are expressed as counts (percentages) and were analyzed via the chi-square test. To mitigate dimensional bias and facilitate direct effect size comparisons, continuous systemic inflammatory indices (NLR, SII, and CRP) underwent Z-score standardization prior to regression analysis. Multivariable logistic regression models were constructed to determine independent factors associated with the progression to drug resistance. To prevent multicollinearity, each standardized inflammatory marker was modeled separately, with all models adjusting for predefined covariates including age at onset, sex, and body mass index (BMI). Restricted cubic splines (RCS) coupled with logistic regression were applied to model the non-linear dose–response probabilities between inflammatory burden and the risk of refractory progression. Statistical significance was established at a two-sided p < 0.05.

3 Results3.1 Transcriptomic profiling identifies Stat3 as a hub regulator of the pro-inflammatory phenotype

To comprehensively delineate the molecular landscape associated with the post-seizure state in SCN1A deficiency, we analyzed bulk RNA sequencing data (GEO accession: GSE112627) derived from brain tissues of Scn1a+/− (KO) mice and wild-type (WT) littermates on a [129S6 × B6] F1 background. Dimensionality reduction analyses confirmed distinct transcriptomic profiles driven by genotype. Principal component analysis (PCA) revealed a clear segregation of samples, with the first principal component (PC1) capturing 86% of the total variance, indicating that genotype is the primary driver of transcriptional heterogeneity (Figure 2C). This separation was further corroborated by the uniform manifold approximation and projection (UMAP) plot, which displayed two discrete clusters corresponding to the WT and KO groups (Figure 2D). Differential expression analysis identified a widespread dysregulation of the transcriptome in Scn1a+/− mice. The volcano plot visualized the distribution of differentially expressed genes (DEGs), highlighting a significant number of up-regulated (red) and down-regulated (blue) genes based on stringent statistical thresholds (Figure 2B). To characterize the most prominent molecular shifts, we performed hierarchical clustering on the top 100 significantly up-regulated genes. The resulting heatmap demonstrated a robust and consistent elevation of these genes across all KO samples compared to controls (Figure 2A, top panel), suggesting a uniform pathological response to sodium channel haploinsufficiency.

Transcriptomic landscape and consensus hub identification. (A) Hierarchical clustering heatmap of DEGs. (B) Volcano plot of significant DEGs. (C,D) Dimensionality reduction visualizing sample distribution via PCA (C) and UMAP (D) plots. (E,F) Functional enrichment analyses showing top KEGG pathways (E) and GO terms (F). (G) Venn diagram of consensus TF predictions across three databases. (H) Box plot of brain Stat3 expression (VST).

To decode the biological significance of these transcriptional alterations, we conducted functional enrichment analyses. KEGG pathway analysis revealed a striking enrichment of inflammation- and immunity-related signaling cascades. The most significantly enriched pathways included the “TNF signaling pathway” (padj = 1.64 × 10−6) and “NF-kappa B signaling pathway” (padj = 3.38 × 10−4), indicating a severe inflammatory burden. Notably, pathways directly linked to STAT3 activation were also prominent, including “Cytokine-cytokine receptor interaction” (padj = 0.003) and the “JAK-STAT signaling pathway” (padj = 0.018) (Figure 2E).

Consistent with these findings, GO biological process analysis indicated a substantial activation of immune responses alongside structural remodeling. The top-enriched terms were dominated by “extracellular matrix organization” (padj = 2.02 × 10−15) and “acute inflammatory response” (padj = 2.69 × 10−9) (Figure 2F). Significantly the analysis identified the “receptor signaling pathway via JAK-STAT” (padj = 0.006) as a significantly enriched process. Detailed inspection of the gene lists within these inflammatory and signaling terms revealed the recurrence of Stat3 alongside other key mediators, collectively pointing to the JAK-STAT signaling axis as a potential key regulatory node of the neuroinflammatory phenotype in the Scn1a+/− brain.

To identify the master regulators orchestrating this inflammatory program, we performed an upstream TF enrichment analysis using three independent databases: ENCODE, TRRUST, and transcription factor PPIs. Intersection analysis of predicted TFs from these sources identified STAT3 as the sole robust consensus hub regulator shared across all three independent datasets (Figure 2G). Furthermore, to validate this prediction at the transcriptional level, we quantified the specific expression of Stat3 in our RNA-seq data. As shown in the box plot, the normalized Stat3 expression was significantly elevated, rising from a median of approximately 10.25 (VST) in the WT controls to 11.21 (VST) in the DS group (Figure 2H). The centrality of this mechanism was further supported by an independent external supportive dataset (GSE289689), which demonstrated robust up-regulation of STAT3 expression and significant positive enrichment of the JAK-STAT signaling cascade in severe epilepsy phenotypes (Supplementary Figures S2A,B). Collectively, these data support an association between Stat3 and the observed inflammatory profile observed in Scn1a-deficient brains.

3.2 Central transcriptomic immune surrogates correlate with STAT3 expression

To investigate the link between central pathology and systemic inflammation, we calculated composite inflammatory signatures using the specific marker gene summation method based on the DS transcriptomic datasets. Transcriptomic NLR, transcriptomic SII, and brain Ptx3 levels were significantly elevated in the DS group compared to controls (Figure 3A). Robust correlation analysis revealed that the expression of the upstream hub STAT3 was strongly and positively correlated with the transcriptomic NLR level (weighted r = 0.91, p = 0.0034; Figure 3B), the transcriptomic SII level (weighted r =0.81, p = 0.042; Figure 3C), and brain Ptx3 levels (weighted r = 0.98, p < 0.001; Figure 3D). Collectively, these data suggest STAT3 as a signaling node associated with the observed neuroimmune changes.

Systemic inflammatory profiles and clinical validation. (A) Box plots of transcriptomic markers (brain Ptx3, NLR, and SII) in WT and DS groups. (B–D) Correlation analyses between Stat3 expression (VST) and transcriptomic NLR (B), SII (C), and brain Ptx3(D). (E) Violin plots of inflammatory markers in the SeLECTS and DRE groups. (F) Forest plot of multivariable logistic regression for DRE risk factors. (G) Dose–response curves of NLR, SII, and CRP for DRE probability. (H) Receiver operating characteristic (ROC) curves evaluating discriminative capacity.

3.3 Systemic inflammatory indices are elevated in DRE and the DS subgroup

To rigorously validate our findings in a clinically relevant context, we evaluated a pediatric cohort consisting of patients with DRE (n = 69, including 9 DS cases) and SeLECTS (n = 71) (Table 1). Baseline levels of NLR, SII, and serum CRP were significantly elevated in the DRE group compared to SeLECTS controls (all p < 0.001; Figure 3E). Multivariable logistic regression with Z-score standardization and adjustment for predefined covariates identified CRP as an independent factor associated with progression to drug resistance (OR = 2.79, 95% CI: 1.23–7.43, p = 0.025). Positive trends were observed for NLR (OR = 1.61, 95% CI: 1.00–2.80, p = 0.064) and SII (OR = 1.72, 95% CI: 1.01–3.27, p = 0.068) (Table 2 and Figure 3F). Restricted cubic spline modeling demonstrated a non-linear dose response trajectory where the probability of DRE escalated sharply as standardized inflammatory indices exceeded their respective thresholds (Figure 3G). Receiver operating characteristic (ROC) analysis evaluated the discriminative capacity of these markers, revealing moderate effect sizes. The combined inflammatory panel yielded the highest discriminative capacity with an area under the curve of 0.81, compared to single indicators including CRP (AUC = 0.78), SII (AUC = 0.66), and NLR (AUC = 0.66) (Figure 3H).

VariableSeLECTS group (n = 71)DRE group (n = 69)DS subgroup (n = 9)p-valueAge at onset, median [IQR], y7.00 [5.00, 9.00]1.00 [0.40, 3.00]0.80 [0.40, 4.00]<0.001Sex, n (%)0.283Male46 (64.8)38 (55.1)4 (44.4)Female25 (35.2)31 (44.9)5 (55.6)BMI, median [IQR], kg/m217.12 [14.93, 21.30]15.98 [14.82, 17.12]16.74 [16.08, 17.49]0.020Concurrent ASMs, median [IQR]1.00 [0.00, 1.50]3.00 [2.00, 4.00]3.00 [2.00, 3.00]<0.001WBC, median [IQR], ×109/L6.27 [5.76, 7.08]6.46 [5.93, 7.73]6.05 [5.24, 6.62]0.163Neutrophil ratio, mean (SD), %44.34 (7.85)49.47 (8.13)48.97 (5.60)<0.001Lymphocyte ratio, mean (SD), %45.66 (8.60)40.22 (7.63)39.82 (4.59)<0.001Platelets, median [IQR], ×109/L228.00 [215.50, 248.00]243.00 [201.00, 293.00]218.00 [204.00, 286.00]0.139CRP, median [IQR], mg/dL0.06 [0.05, 0.08]0.10 [0.08, 0.13]0.16 [0.10, 0.18]<0.001NLR, median [IQR]0.99 [0.78, 1.18]1.21 [1.00, 1.55]1.20 [1.00, 1.41]<0.001SII, median [IQR]219.47 [177.56, 293.35]298.50 [218.28, 382.31]321.05 [218.54, 395.09]<0.001

Baseline demographic and laboratory characteristics of the study population, stratified by epilepsy status.

Data are presented as median [interquartile range, IQR] for skewed continuous variables, mean (standard deviation) for normally distributed variables, or n (%). p-values represent comparisons between the SeLECTS and DRE groups via the Mann–Whitney U test, Student’s t-test, or chi-square test as appropriate. To eliminate potential hematological confounding from long-term polytherapy, inflammatory indices (WBC, CRP, NLR, SII, and cell ratios) were derived from peripheral blood samples collected at the initial hospital visit prior to extensive drug exposure. Conversely, the number of concurrent ASMs reflects the maximum drug burden recorded at the final follow-up, serving as an indicator of clinical refractoriness. SeLECTS, self-limited epilepsy with centrotemporal spikes; DRE, drug-resistant epilepsy; DS, Dravet syndrome; ASMs, anti-seizure medications.

VariableOR (95% CI)p-valueAge at onset (per year)0.55 (0.44–0.67)<0.001Male (vs. female)0.56 (0.18–1.64)0.299BMI0.95 (0.82–1.11)0.498NLR (per 1-SD increase)1.61 (1.00–2.80)0.064SII (per 1-SD increase)1.72 (1.01–3.27)0.068CRP (per 1-SD increase)2.79 (1.23–7.43)0.025

Multivariable logistic regression analysis for predictors of refractory epilepsy.

Multivariable logistic regression models were utilized to evaluate the independent association between inflammatory markers and the risk of refractory epilepsy. The odds ratios (ORs) for NLR, SII, and CRP correspond to a 1-standard deviation (SD) increase in these continuous variables. Each inflammatory marker was modeled separately to prevent multicollinearity, with all models adjusting for age at onset, sex, and body mass index (BMI).

3.4 Hyperactivation of the JAK-STAT3 pathway in Scn1a+/− mice

To validate the phenotypic susceptibility of our model, the F1 cohorts were subjected to a hyperthermia-induced seizure protocol. Kaplan–Meier survival analysis demonstrated that all Scn1a+/− mice exhibited generalized tonic–clonic seizures (equivalent to Racine score 5) at a significantly lower temperature threshold (mean = 41.68 °C; range = 40.5–42.5 °C). Notably, none of the WT littermates exhibited any seizure activity up to the 43.0 °C cutoff limit (Supplementary Figure S2D). Following phenotypic validation, mechanistic verification in the hippocampus and cortex of Scn1a+/− mice confirmed the pathway’s involvement. Quantitative PCR demonstrated robust transcriptional upregulation of pathway components (n = 6/group). IL-6 mRNA transcripts surged three fold in both brain regions compared to WT littermates (p < 0.0001), accompanied by significant elevations in STAT3 and NLRP3 (Figures 4A,B). Western blotting revealed a significant elevation in the ratio of phosphorylated STAT3 (Tyr705) to total STAT3 (p-STAT3/Total STAT3) in Scn1a+/− mice (p < 0.0001), while total STAT3 expression remained constant. Additionally, IL-6 protein levels were significantly higher in the mutant group, as were the levels of the glial markers IBA1 and GFAP (Figures 4C,D).

Activation of the JAK-STAT3 signaling axis and neuroinflammatory response in Scn1a+/− mice. (A,B) Quantitative PCR analysis of Stat3, Nlrp3, and Il6 mRNA levels in the hippocampus (A) and cortex (B) of WT and Scn1a+/− mice. (C) Representative western blot images showing expression levels of p-STAT3, STAT3, p-JAK1, JAK1, Cyclin D1, IL-6, IBA1, and GFAP in brain tissues. GAPDH served as the loading control. (D) Densitometric quantification of relative protein abundance for p-STAT3/STAT3, p-JAK1/JAK1, Cyclin D1, IL-6, GFAP, and IBA1. Data are presented as mean ± SD (****p < 0.0001).

To provide a more complete characterization of the canonical signaling pathway and determine whether the increased p-STAT3 translates into functional transcriptional activation, we further evaluated the upstream kinase and downstream targets. Western blot analysis revealed that the ratio of phosphorylated JAK1 to total JAK1 (p-JAK1/Total JAK1) was significantly elevated in the hippocampus of Scn1a+/− mice, confirming upstream activation. Additionally, the expression of Cyclin D1, a canonical downstream transcriptional target of STAT3, was also markedly up-regulated compared to WT littermates (Figures 4C,D).

3.5 Neuroinflammatory pathology and central-peripheral correlation

Double immunofluorescence labeling in the brain parenchyma showed that p-STAT3-positive signals were predominantly localized within GFAP-