Cell-type-resolved transcriptomic landscape of human focal cortical dysplasia

Significance

Focal cortical dysplasia (FCD) is a major cause of drug-resistant epilepsy, yet its biological heterogeneity remains poorly defined. We profiled 487,286 nuclei from paired lesional and perilesional cortices across FCD I-III to generate a cell-type-resolved transcriptomic atlas. Distinct and shared transcriptional programs were identified across neuronal, glial, and vascular compartments, with prominent dysregulation in inhibitory interneurons, deep-layer projection neurons, astrocytes, and endothelial cells. Coordinated inflammatory, metabolic, and hypoxia-responsive pathways highlight nonneuronal contributions to epileptogenesis. These results define molecular correlates of histopathological heterogeneity, nominate candidate tissue-detectable markers, and provide a framework for mechanism-based diagnosis and targeted therapies in FCD.

Abstract

Focal cortical dysplasia (FCD) is a major cause of drug-resistant epilepsy and displays substantial clinical and histopathological heterogeneity, yet the cellular and molecular bases underlying this diversity remain poorly defined. Here, we performed single-nucleus RNA sequencing of 487,286 nuclei from 34 paired lesional cores and perilesional cortices across FCD I-III, generating a cell-type-resolved transcriptomic atlas of human FCD. Comparative analyses identified both shared and subtype-specific transcriptional alterations across neuronal, glial, and vascular compartments. Inhibitory interneurons and deep-layer projection neurons exhibited prominent dysregulation, whereas astrocytes and vascular cells showed coordinated activation of inflammatory, metabolic, and hypoxia-responsive pathways. Several genes displayed consistent subtype-associated expression patterns, including ZNF254, DRG1, and ABHD17A in astrocytes, and ATF4 in endothelial cells. These results link cell-type-specific transcriptional programs to histopathological heterogeneity across FCD subtypes, identify candidate tissue-detectable markers, and provide insight into nonneuronal contributions to epileptogenic cortical malformations.

Focal cortical dysplasia (FCD) is among the most common structural substrates underlying drug-resistant epilepsy in children and young adults (1). The International League Against Epilepsy (ILAE) classifies FCDs into three major subtypes, type I-III, based on histopathological criteria, including abnormalities in cortical lamination, dysmorphic neurons, balloon cells, and associated principal lesions (2, 3). While this framework provides essential diagnostic guidance, substantial overlap in clinical presentation, seizure semiology, neuroimaging features, and postsurgical outcomes across FCD subtypes suggests that shared cellular and molecular programs may underlie phenotypically distinct lesions, underscoring the limitations of morphology-based classification alone (3–6). These observations point to a need for molecularly informed frameworks that can better capture the biological heterogeneity of FCDs.

At the molecular level, somatic mutations affecting the PI3K-AKT-mTOR signaling pathway have been identified in a subset of FCD II lesions, providing important mechanistic insights into abnormal cortical development and epileptogenesis (3, 7–12). However, such mutations account for fewer than half of FCD II cases, and most FCD I/III lesions lack consistently identifiable genetic drivers (8, 13, 14). Moreover, the mosaic distribution of pathogenic variants across cortical layers and cell types further complicates genotype–phenotype correlations (13, 15–22). These findings suggest that additional, nongenetic mechanisms (potentially involving glial dysfunction, vascular remodeling, and altered cellular microenvironments) substantially contribute to FCD pathogenesis, yet remain insufficiently characterized (22–28).

Here, we apply large-scale single-nucleus RNA sequencing (snRNA-seq) to 34 paired lesional cores and perilesional cortical samples from pediatric patients with FCD I/II/III to build a comprehensive molecular atlas of human FCDs. By integrating cell composition, gene expression, and pathway-level analyses, we define shared and subtype-specific cellular programs that clarify the organizational principles of FCD pathology and nominate candidate cell-type-specific mechanisms underlying disease heterogeneity.

1. Results

1.1. A Cell-Type-Resolved Transcriptomic Atlas of Human FCDs.

To define cellular composition and transcriptional programs underlying FCD subtypes, we performed snRNA-seq on paired lesional core and perilesional cortical samples from 34 pediatric patients, including 10 FCD I, 15 FCD II, and 9 FCD III (Fig. 1A and Dataset S1). Following quality control and correction for ambient RNA contamination (SI Appendix, Fig. S1 A–C), 487,286 high-quality nuclei were retained for downstream analyses (Fig. 1B and SI Appendix, Fig. S1D and Dataset S2), enabling systematic comparisons across FCD subtypes and tissue regions.

Fig. 1.

A cell-type-resolved transcriptomic atlas of FCDs. (A) Schematic of the droplet-based snRNA-seq workflow applied to 34 paired neocortical lesion and perilesional samples from FCD patients. High-quality nuclei were profiled to define cell-type-specific transcriptional signatures. Pathogenesis and transcriptional heterogeneity were assessed across FCD subtypes using pseudoblank matrices at gene, pathway, and cell levels. WGCNA, weighted gene coexpression network analysis; DEGs/DEPs, differentially expressed genes/pathways. (B) UMAP visualization of all sequenced nuclei and stratified by FCD subtypes. Colors denote 17 annotated cell types and 4 major lineages: glial cells (GC), vascular cells (VC), interneurons (IN), and excitatory neurons (ExN). AST, astrocyte; EC, endothelial cells; OPC, oligodendrocyte progenitor cell; Oligo, oligodendrocyte; SMC, smooth muscle cell. (C) Dot plot showing expression of the canonical marker gene used for cell-type annotation. Dot size indicates the fraction of cells expressing the gene; color intensity indicates normalized expression. (D) Distribution of nuclei across annotated cell types. The cellular composition of the FCD neocortex is consistent with expected cortical architecture (25).

Reference-guided annotation classified nuclei into 17 transcriptionally distinct populations spanning four major cellular lineages: excitatory neurons, inhibitory interneurons, glial cells, and vascular cells (Fig. 1B and SI Appendix, Fig. S1D and Dataset S2). Neuronal populations comprised laminar-resolved excitatory neurons across cortical layers 2 to 6 and four major interneuron subclasses. Nonneuronal populations included astrocytes, oligodendrocyte lineage cells, microglia, endothelial cells, and smooth muscle cells (SMCs) (Fig. 1B and SI Appendix, Fig. S1D). Canonical marker genes exhibited consistent and lineage-specific expression patterns across clusters and FCD subtypes, supporting robust and reliable cell-type annotation (Fig. 1C and SI Appendix, Fig. S2 A and B and Dataset S3). Concordance with previously reported human cortical single-cell RNA sequencing (scRNA-seq) datasets further validated the annotation framework (25, 29) (SI Appendix, Fig. S2C).

At the global level, overall cell-type proportions were broadly comparable between lesional and perilesional samples, with glial cells representing the largest fraction (58.9%), followed by neurons (37.0%), consistent with prior studies of the human neocortex (Fig. 1D and SI Appendix, Fig. S1D) (30). Collectively, these data establish a high-resolution, cell-type-resolved transcriptomic atlas of human FCDs.

1.2. Concomitant Cellular and Transcriptional Dysregulation Underlie FCD Lesions.

To characterize cellular and molecular alterations associated with FCD lesions, we integrated multiple analytical approaches, including compositional proportion analysis (CPA), transcriptional prioritization, pseudobulk differential expression, weighted gene coexpression network analysis (WGCNA), and disease ontology (DO) enrichment analysis of the concerned genes (Fig. 2A and SI Appendix, Fig. S3A).

Fig. 2.

Coordinated cellular and transcriptional alterations in FCD lesions. (A) Analytical workflow for within-group analysis to define FCD pathogenesis. Pseudobulk RNA-seq matrices per sample and cell type were used to assess differential expression, pathway enrichment, and WGCNA modules. Cell-type proportion changes were assessed using scProportionTest, and transcriptional prioritization was evaluated using Augur. DEG/DEPs, differentially expressed genes/pathways; AUC, area under the receiver operating characteristic curve. (B) Sample-wise CPA of major cell types comparing FCD lesions with the paired perilesional cortex. Endothelial cells (EC), SMC, astrocytes (AST), and deep-layer excitatory neurons show significant alterations. The red dashed line indicates the cutoff of │log₂FC│ > 0.58 and false discovery rate (FDR) < 0.05. Alt., alterations in CPA. (C) Group-wise CPA across FCD subtypes comparing lesion and perilesional tissues. Fold change (FC) and FDR are shown, and the label in red indicates the significant cell types. (D) Augur-based transcriptional prioritization scores for each cell type in lesion vs. perilesional comparisons. (E) Number of DEGs identified per cell type from paired differential expression analyses. Heatmaps summarize enriched FCD-associated DO terms derived from GeDiPNet, DisGeNET, and Jensen disease databases. (F and G) Enrichment of DO terms associated with DEP (F) and WGCNA modules (G). The significance was defined by a cutoff of P < 0.01 in multiple-group DEP analysis. Trait-associated modules were defined by │ρ│ > 0.45. H. Synaptic gene ontology (SynGO) enrichment of significant WGCNA modules (highlighted in G) across indicated cell types. Cor, correlation of SynGO terms was calculated using the simpleEnrichment R package (with setting of “measure =” Sim_XGraSM_2013).

Sample-wise CPA revealed that endothelial cells, SMCs, astrocytes, deep-layer excitatory neurons (L5/6-FEZF2), and IN-SV2C interneurons frequently exhibited significant abundance differences between lesional and perilesional regions (Fig. 2B). Similarly, glial and vascular populations showed the most pronounced overall abundance changes across FCD lesions in group-wise CPA (Fig. 2C). Augur-based prioritization highlighted substantial variability in microglial populations, suggesting context-dependent transcriptional heterogeneity within lesions (Fig. 2D).

Pseudobulk differential expression analyses identified prominent transcriptional alterations in IN-SV2C interneurons and vascular cell populations, particularly endothelial cells and SMCs (Fig. 2E and SI Appendix, Fig. S3A and Dataset S4). Notably, DO enrichment analysis of differentially expressed genes (DEGs) in vascular cells revealed strong associations with seizures, epilepsy, and related neurological conditions, exceeding those associations observed in neuronal populations such as IN-VIP and L2/3-CUX2 (Fig. 2E and Dataset S5). Notably, astrocytes and endothelial cells showed enrichment for pathways related to neuronal development, including axon guidance and central nervous system myelination (SI Appendix, Figs. S3 B and S4 A and B and Dataset S6).

Gene set variation analysis (GSVA)-based differential expression protein (DEP) analysis further uncovered disease-associated alterations predominantly affecting interneurons, astrocytes, and SMCs (Fig. 2F and Dataset S7). Several metabolic pathways, including carbohydrate and amino acid metabolism, were differentially regulated across specific cell populations (SI Appendix, Figs. S3C and S5A and Dataset S8). Furthermore, WGCNA identified disease-associated modules enriched for synaptic and neuronal signaling processes, particularly within deep-layer excitatory neurons (│ρ│ > 0.45, P < 0.05) (Fig. 2 G and H and SI Appendix, Figs. S3D and S6A and Dataset S9).

Together, these analyses indicate that FCD lesions are characterized by coordinated transcriptional alterations across both neuronal and nonneuronal compartments, prominently involving inhibitory interneurons, astrocytes, and vascular cells.

1.3. Subtype-Specific Distinction Among FCD Subtypes.

To resolve subtype-specific molecular features, we performed stratified analyses of FCD I/II/III (Fig. 3A). Sample-wise CPA revealed the most pronounced subtype-dependent differences in cellular composition within L2/3-CUX2 excitatory neurons. Specifically, L2/3-CUX2 neurons were reduced in FCD I/III, but showed a relative increase in FCD II (Fig. 3B). In addition, astrocytes, endothelial cells, and SMCs exhibited marked alterations predominantly in FCD II, whereas these populations remained comparatively stable in FCD I/III (Fig. 3B).

Fig. 3.

Subtype-specific cellular programs distinguish FCD subtypes. (A) Workflow for within-subtype analysis using pseudobulk matrices to identify DEGs, DEPs, and WGCNA modules. Analyses (similarly with Fig. 2) were performed separately for FCD I, II, and III. (B) Sample-wise CPA for all annotated cell types within FCD I, II, and III, comparing lesions with the paired perilesional cortex. Significance was defined as FDR < 0.05 and │log₂FC│ > 0.58. (C) Number of DEGs per cell type across FCD subtypes. DEG analysis was performed in a paired manner (lesion vs. perilesion) with│log₂FC│ > 1 and adjusted P_adj < 0.05. (D) Augur-based transcriptional prioritization for each cell type in lesion vs. perilesional comparisons. The red dashed line indicates a high-priority AUC threshold (AUC > 0.75). (E) Differentially expressed pathway analysis for neuronal populations using DisGeNET gene sets. Heatmap shows the total number of DEPs per subtype (│log₂FC│ > 0.3, P < 0.05). (F) Number of enriched DO terms for genes within subtype-associated WGCNA modules based on DisGeNET enrichment (P < 0.05; odds ratio > 5). (G) Representative histological features of cortical tissues from different FCD subtypes (Bottom), as revealed by hematoxylin-eosin staining, along with corresponding quantitative analyses (Top). Distinct characteristic features are indicated by symbols in the figures; CV, cortical vessels. (Scale bar, 100 μm.) (H) Representative neuroimaging findings (H) and quantitative analyses (I) in FCD patients. MRI, magnetic resonance imaging; ASL, arterial spin labeling MRI; PET, positron emission tomography MRI. Both the lesion and its symmetrical normal side (served as control) in a scan of a patient were quantified for a paired comparison, ****P < 0.0001 by paired Student’s t test.

Augur-based prioritization analyses demonstrated higher cell-type-specific separability within individual FCD lesions relative to the global dataset, most notably in neuronal populations (Figs. 2D and 3C). Across FCD subtypes, cell clusters in FCD III exhibited larger area under the curve (AUC) values than FCD I/II (Fig. 3C). Deep-layer neurons displayed greater prioritization in FCD I, whereas upper-layer neurons demonstrated greater prominence in FCD II/III (Fig. 3C). Differential expression analyses further uncovered marked gene-level heterogeneity across FCD subtypes. FCD II exhibited the greatest number of DEGs, followed by FCD III, whereas FCD I showed comparatively few transcriptional alterations (Fig. 3D and Dataset S10). Notably, glial and vascular populations were major contributors to transcriptional changes in FCD II, whereas neuronal populations were disproportionately affected in other subtypes (Fig. 3D). Despite this heterogeneity, FCD I/III exhibited overlapping sets of DEGs, notably enriched in IN-PV (GRID2, RELN, NELL1, MAN1A1, etc.), IN-SV2C (IQGAP2, RYR3, COL11A1), and L4-ARHGAP populations (SMOC2, CUX2, DPF3, etc.) (SI Appendix, Fig. S7 A and C). Functional enrichment analyses of these overlapping DEGs revealed that shared transcriptional perturbations were strongly linked to epilepsy-related processes, synaptic signaling, and associated biological ontologies (SI Appendix, Fig. S7 B and C).

GSVA-based DEP analyses demonstrated that disease-associated terms, including neurological and psychiatric disorders, were predominantly enriched in neuronal populations (particularly interneurons) across FCD subtypes (Fig. 3E and Dataset S11). Vascular populations also showed associations with brain-related disease terms, especially in FCD II/III (Fig. 3E). Consistent with the observations in DEGs, DEP analyses also revealed extensive pathway alterations in FCD II/III (SI Appendix, Fig. S8A and Dataset S11), including processes related to neuronal migration and synaptic signaling (SI Appendix, Fig. S8B). Furthermore, DEPs in these populations were also enriched in distinct pathways. For example, in FCD II, astrocytes exhibited significant alterations in mitochondrial functions—particularly mitochondrial rRNA/tRNA processing and ATP formation via chemiosmotic coupling (Dataset S12). In contrast, astrocytes from FCD III showed marked dysregulation in RNA processing pathways, including 7-methylguanosine RNA capping (Dataset S12). In IN-SV2C, DEPs mapped to subtype-specific pathways: in FCD I, DEPs involved in double-stranded methylated DNA binding were significantly altered; in FCD II, the ATF6-mediated unfolded protein response pathway was the most prominent; and in FCD III, mitochondrial oxidative phosphorylation exhibited the most significant dysregulation (Dataset S12). Additionally, alterations in ammonia metabolism were evident in IN-SV2C neurons across all FCD subtypes. Specifically, in FCD III, the polyamine transport pathway was significantly dysregulated, whereas in FCD II, both the urea cycle and pyrimidine de novo biosynthesis pathways exhibited dysregulation. In FCD I, the kynurenine/tryptamine/serotonin pathway displayed marked perturbations (Dataset S12).

WGCNA further identified subtype-dependent disease-associated modules exhibiting prominent enrichment in interneuron populations (such as IN-PV and IN-SV2C) and specific excitatory neuron subtypes (SI Appendix, Fig. S9A). DO enrichment analysis of module genes, employing DisGeNet gene sets, revealed that beyond neuronal populations, vascular and glial cells were additionally implicated in epilepsy-related conditions—most notably in FCD II/III (Fig. 3F and SI Appendix, Figs. S9B and S10 A and B and Dataset S13). Concurrently, hypoxia pathways were dysregulated in these cell populations (SI Appendix, Fig. S9C). Hematoxylin-eosin staining further demonstrated vascular aberrations—including altered thickness, density, and morphology—in the lesional neocortex of FCDs (Fig. 3G). Aligning with prior studies, these findings establish a robust association between cerebrovascular disorders (e.g., ischemic stroke) and epilepsy (31, 32). Indeed, clinical neuroimaging analyses using arterial spin labeling (ASL) and positron emission tomography (PET) revealed marked reductions in cerebral blood flow and metabolic activity at FCD lesion sites (Fig. 3H).

Together, these results demonstrate that FCD subtypes are characterized by layer- and cell-type-specific transcriptional programs, providing a molecular basis for their clinical and pathological heterogeneity.

1.4. Cell-Type-Specific Heterogeneity and Molecular Stratification of FCDs.

To quantify transcriptional variability across FCD subtypes, we performed ANOVA-like differential expression analyses using likelihood ratio testing on metacell-based expression matrices (Fig. 4A). Inhibitory interneurons and glial populations exhibited greater transcriptional variability across FCD subtypes than projection neurons, with the most pronounced differences observed between FCD I and the other subtypes (Fig. 4B and Dataset S14). Consistently, CPA indicated that cell populations in FCD I were more extensively altered than those in FCD II/III, particularly among glial and endothelial cells (Fig. 4B).

Fig. 4.

Cell-type-specific heterogeneity and molecular stratification of FCDs. (A) Workflow for cross-subtype analysis. Pseudobulk matrices were used to identify DEGs via Likelihood Ratio tests (LRT) and evaluate subtype-specific markers by Receiver operating characteristic (ROC) analysis. Group-wise CPA was carried out for lesion sites from different FCD subtypes, using scProportionTest. (B) Summary of DEGs (Top) and group-wise CPA (Bottom) across FCD I, II, and III of the indicated comparisons for each cell type. DEGs were defined by P_adj< 0.0001. Significance in CPA was defined by │log₂FC│ > 0.58 and FDR < 0.05. Log₂FC, log2 fold change. (C) Discriminative performance of subtype-specific gene signatures evaluated using logistic regression. Labels in red indicated the concerned genes. (D) Real-time quantitative PCR validation of selected genes in isolated astrocytes and endothelial cells from FCD lesions. *P < 0.05, **P < 0.01, and ****P < 0.0001 by Student’s t test. (E) ROC analysis of cell-type-specific mRNA markers for discriminating FCD subtypes; AUC values were shown. AUC, area under the ROC curve. (F and G). (G) Immunofluorescence staining validation of candidate DEGs in astrocytes (F) and vascular cells (G). GFAP and ACTA2 mark astrocytes and SMCs, respectively; ZNF254 and ATF4 indicate candidate markers. (Scale bars, 50 μm.) (H and I) Quantification of GFAP⁺/ZNF254⁺ astrocytes (H) and relative ATF4 expression in endothelial cells (I). **P < 0.01 and ***P < 0.001 by paired Student’s t test.

Subtype-variable genes were enriched for disease-associated terms, most prominently in glial cells, L2/3 projection neurons, and vascular populations (SI Appendix, Fig. S11A). Astrocyte-associated subtype-variable genes were enriched for immune-related processes, including regulation of CD4⁺ T cell activation and T-helper cell differentiation, suggesting functional remodeling beyond canonical homeostatic roles (SI Appendix, Fig. S11B). Similar immune–glial interactions have been reported in cortisol-mediated neurotoxicity in poststroke dementia (33), suggesting engagement of conserved stress-responsive pathways across neurological disorders. Meanwhile, SMC-associated genes were enriched for VEGFR signaling and cerebral cortex cell migration pathways (SI Appendix, Fig. S11B), consistent with vascular alterations identified in prior analyses.

Pathway-level analyses revealed widespread disease-associated transcriptional perturbations across cell types, with notable enrichment in neuronal populations (SI Appendix, Fig. S12A). These perturbations were linked, in a cell-type-specific manner, to neurological and neuropsychiatric disorders, including epileptic encephalopathy, childhood absence epilepsy, autism spectrum disorder, and inflammatory brain conditions (SI Appendix, Fig. S12B). However, at the cellular level, we observed that vascular populations exhibited the most heterogeneity among FCD subtypes (Fig. 4B).

To evaluate whether subtype-associated transcriptional signatures could facilitate molecular stratification, we performed exploratory logistic regression analyses using single-gene features. Several genes exhibited consistent subtype-specific expression patterns, including ZNF254, DRG1, and ABHD17A in astrocytes, and ATF4 in endothelial cells (Fig. 4C and SI Appendix, Fig. S13 A and B). Considering the panlineage expression of these genes across brain cell types, astrocytes and endothelial cells were purified via flow cytometric sorting prior to experimental validation. Real-time quantitative PCR validated enriched expression of ZNF254 in astrocytes derived from FCD I lesions and ATF4 in endothelial cells from FCD II lesions (Fig. 4D). Receiver operating characteristic (ROC) curve analyses revealed these markers exhibited robust subtype-discriminatory capacity for FCD subtypes (Fig. 4E). Immunofluorescence staining further corroborated these patterns, revealing higher frequencies of GFAP⁺/ZNF254⁺ astrocytes in FCD I lesions and upregulated ATF4 expression in endothelial cells from FCD II lesions (Fig. 4 F–I and SI Appendix, Fig. S14A).

Furthermore, we identified disease-severity-associated (DSA) genes and pathways in FCDs. Most DSA genes were enriched in neuronal populations, notably excitatory neurons, including PAMR1 and DRC1 in L3-PRSS12, and KMT2A and SMIM20 in L4-MME neurons (SI Appendix, Fig. S14B). At the pathway level, excitatory neurons displayed a greater enrichment of DSA pathways, including mTOR and Ephrin signaling in L2/3-FREM3 neurons, Wnt-mediated presynaptic organization in L4-MET neurons, and FGF signaling in L4-MME neurons (SI Appendix, Fig. S1 C and D and Dataset S15).

Together, these findings reveal extensive cell-type-specific transcriptional heterogeneity across FCD subtypes and identify astrocytic and vascular molecular signatures associated with refined pathological stratification.

2. Discussion

FCD is a leading structural cause of drug-resistant epilepsy, yet the extent to which distinct cellular programs converge or diverge across FCD subtypes remains incompletely defined (34, 35). By profiling approximately 500,000 nuclei from paired lesional and perilesional cortical tissues spanning FCD types I–III, we generated a large-scale, cell-type-resolved transcriptomic atlas of human FCD. This dataset reveals that epileptogenic lesions are shaped by coordinated transcriptional remodeling across neuronal, glial, and vascular compartments, extending current neuron-centric models of FCD pathogenesis.

At the neuronal level, we uncover layer- and subtype-specific vulnerabilities that provide molecular insights into classical histopathological features of FCDs. Subtype-dependent differences in L2/3-CUX2 excitatory neuron abundance may reflect differential disruption of cortical lamination (36), whereas deep-layer projection neurons show selective vulnerability in FCD I. Across all subtypes, inhibitory IN-SV2C interneurons exhibit the most extensive transcriptional dysregulation, implicating impaired synaptic vesicle cycling and GABAergic signaling as conserved contributors to circuit instability. In parallel, subtype-specific pathways were identified within IN-SV2C, including mitochondrial oxidative phosphorylation in FCD III, ATF6-mediated unfolded protein response in FCD II, and double-stranded methylated DNA binding in FCD I.

Notably, FCD I and FCD III shared a subset of DEGs across multiple neuronal populations, particularly in IN-PV (e.g., GRID2, RELN, NELL1), IN-SV2C (e.g., IQGAP2, RYR3), and L4-ARHGAP neurons (e.g., SMOC2, CUX2). These shared DEGs were strongly enriched in epilepsy-related processes, synaptic signaling, and relevant biological pathways, suggesting partially convergent pathogenic mechanisms across FCD subtypes. Furthermore, ammonia metabolism emerged as a consistently altered pathway in IN-SV2C across all FCD subtypes. Given that dysregulated ammonia metabolism induces neurotoxicity by impairing energy homeostasis, neurotransmission, and mitochondrial integrity (37, 38), whether IN-SV2C-specific ammonia metabolic alterations actively contribute to FCD pathogenesis warrants further investigation.

Beyond neuronal mechanisms, we observed pronounced cell-type-specific heterogeneity across FCD subtypes. Astrocytes, endothelial cells, and SMCs were preferentially altered in FCD II, consistent with our prior findings (19). Augur-based prioritization further demonstrated greater cell-type-specific separability within individual FCD lesions compared to the global dataset, emphasizing the necessity of subtype-stratified analytical approaches. Collectively, these data support vascular malformation as a unifying feature across FCD subtypes. Functional enrichment analyses of astrocytic and endothelial DEGs further revealed associations with neuronal developmental processes, including central nervous system myelination. Disruption of cerebrovascular development and oxygen homeostasis has long been implicated in epileptogenesis (39, 40) and intersects with pathways such as mTOR signaling, which is tightly linked to cortical malformations and neuronal hyperexcitability (3, 36). This convergence suggests that nonneuronal mechanisms may amplify or sustain epileptogenic circuits in FCDs. From a translational perspective, the predominance of glial and vascular pathways among the most consistently dysregulated programs has important therapeutic implications. Current antiseizure medications primarily target neuronal excitability (41); however, our findings suggest that modulation of astrocytic metabolism, neurovascular integrity, or hypoxia-related signaling may represent complementary therapeutic avenues. By defining recurrent, cell-type-resolved molecular programs in human FCD tissues, this atlas provides a framework for mechanism-informed therapeutic exploration.

The extensive transcriptional heterogeneity observed across FCD subtypes also underscores the limitations of morphology-based classification alone. Lesions with similar histopathological diagnoses can diverge substantially at the molecular level. We identified subtype-associated astrocytic and vascular signatures, validated at the protein level, indicating that nonneuronal molecular features may complement existing criteria for refining FCD stratification. Specifically, ZNF254, DRG1, and ABHD17A in astrocytes, and ATF4 in endothelial cells, exhibited consistent subtype-associated expression patterns. RT–qPCR and immunofluorescence confirmed preferential expression of ZNF254 in astrocytes from FCD I lesions and ATF4 in endothelial cells from FCD II lesions, supporting their potential utility as molecular biomarkers.

Several neuronal DEGs, including PAMR1 and DRC1 in L3-PRSS12 neurons and KMT2A and SMIM20 in L4-MME neurons, correlated with disease severity. KMT2A, a histone methyltransferase implicated in neurodevelopmental disorders with epilepsy (42, 43), and SMIM20, which encodes the precursor of the neuropeptides phoenixin-14/20 involved in neuronal and behavioral regulation, represent plausible modulators of disease progression (44), although their precise roles in FCD require further functional validation. At the pathway level, excitatory neurons also displayed multiple DEPs correlated with disease severity—excluding the mTOR signaling pathway (in L2/3-FREM3)—such as Wnt (in L4-MET) and FGF signaling (in L4-MME). Previously, ephrin signaling intersects with the mTOR pathway at multiple nodes in different contexts (45), which functions in shaping excitatory circuit loops and recurrent seizure activity (46). In L2/3-FREM3, Ephrin signaling demonstrated a weak negative correlation with FCD severity, whereas mTOR signaling showed a positive correlation with disease severity, implying a potential regulatory function of Ephrin signaling. Considering the weak negative correlation between Ephrin signaling and disease severity, we infer that Ephrin-mediated modulation of mTOR signaling does not constitute a dominant mechanism. In 2015, Atsushi et al. demonstrated that the divergent canonical Wnt signaling pathway regulates presynaptic active zone organization in Drosophila (47). Beatrice et al. showed that ectopic expression of FGF-2/BDNF in the lesioned hippocampus of epileptic rat models enhanced neurogenesis, ameliorated neuronal damage, and attenuated epileptogenesis (48). Taken together with our findings, cell-type-specific Wnt and FGF signaling may pathologically contribute to FCD pathogenesis.

Several limitations warrant acknowledgment, including the utilization of the perilesional cortex as a reference tissue and the relative nature of snRNA-seq-based quantification. Meanwhile, during single-cell sequencing, technical and biological factors impact nuclear capture efficiency, potentially leading to preferential loss or underrepresentation of other neuronal subtypes within the dysplastic cortex (49). Accordingly, single-nucleus transcriptomic proportion estimates mirror the relative cellular representation within the captured nuclei. In addition, causal relationships have yet to be elucidated via targeted functional studies.

In summary, this study presents a comprehensive single-nucleus transcriptomic atlas of human FCD subtypes I–III, revealing coordinated neuronal, glial, and vascular remodeling underlying disease heterogeneity. By integrating cell-type-resolved molecular programs with established pathological frameworks, our work advances the biological understanding of FCDs and lays a foundation for refined molecular stratification and mechanism-guided therapeutic strategies.

3. Materials and Methods

3.1. Study Design.

This study aimed to systematically identify disease-associated cell types and delineate cellular and molecular heterogeneity across FCD subtypes using 10x snRNA-seq. Paired lesional core and adjacent perilesional cortical samples were obtained from 34 patients undergoing epilepsy surgery, including FCD type I (n = 10), type II (n = 15), and type III (n = 9), enabling within-patient comparisons (Dataset S1).

Raw sequencing data were processed using the Cell Ranger pipeline, followed by removal of ambient RNA contamination with CellBender. Downstream quality control, dimensionality reduction, clustering, and marker identification were performed using Seurat (v4.0). Cell-type annotation was conducted using Clustifyr with a curated human cortical single-nucleus reference and further refined based on canonical marker genes.

To identify disease-associated cell types and transcriptional programs, we applied a multilevel analytical framework integrating relative cell-type abundance and prioritization analyses, pseudobulk differential expression analysis, pathway-level enrichment analysis, and WGCNA. To interrogate intersubtype heterogeneity, lesional samples from different FCD subtypes were directly compared, and subtype-discriminative genes were further evaluated using logistic regression and ROC analysis.

3.2. Sample Information.

Detailed clinical, imaging, and pathological information is provided in Dataset S1. During the interictal or postictal period, epileptogenic zones were localized preoperatively using comprehensive electroclinical evaluation and multimodal neuroimaging, including ASL-MRI, and positron emission tomography MRI (PET-MRI). Diagnoses were made according to ILAE criteria and confirmed by postoperative histopathology. Perilesional tissue was defined as cortex adjacent to the primary lesion without overt FCD-defining histopathological features and served as a patient-matched internal control. MRI demonstrated cortical malformations, thickening, and blurring at the gray-white matter junction. PET-MRI revealed focal hypometabolism, and ASL perfusion MRI demonstrated regional cerebral blood flow reductions in lesion areas. For quantification, the symmetrical side of the lesion in each scan of a patient served as control.

All procedures were approved by the Ethics Committee of the Children’s Hospital affiliated with Fudan University (IRB No. 2021-77). Written informed consent was obtained from the legal guardians of all participants.

Brain tissues were homogenized in ice-cold nuclei lysis buffer containing 0.25 M sucrose, 25 mM KCl, 5 mM MgCl₂, 20 mM Tricine-KOH, 1 mM DTT, 0.15 mM spermine, 0.5 mM spermidine, protease inhibitor cocktail, and 0.3% Igepal CA-630. Nuclei were purified using iodixanol density-gradient centrifugation (25, 30, and 40%) at 4 °C for 30 min at 4,500 rpm. Nuclei at the 30 to 40% interface were collected, washed three times, and resuspended in phosphate-buffered saline (PBS) supplemented with an RNase inhibitor.

Nuclear integrity and concentration were assessed using trypan blue staining and acridine orange/propidium iodide (AO/PI) staining. Purified nuclei were loaded onto the Chromium Controller (10X Genomics) for library preparation.

3.3. snRNA-Seq and Data Processing.

snRNA-seq libraries were prepared using the Chromium Next GEM Single Cell 3′ Kit v3.1 (10X Genomics) and sequenced on an Illumina NovaSeq 6000 platform. Reads were aligned to the human reference genome (GRCh38.p14) using Cell Ranger (version 8.0), with intronic reads retained to capture nuclear transcripts.

Quality control was performed in Seurat (version 4.0). Nuclei with 500 to 8,000 detected genes and <10% mitochondrial gene content were retained. Doublets were identified via both DoubletFinder (version 2.0.6, https://rdrr.io/github/chris-mcginnis-ucsf/DoubletFinder/) and scDblFinder (version 1.24.0, https://bioconductor.org/packages//release/bioc/html/scDblFinder.html) with the default settings, and nuclei classified as doublets by both methods were excluded. After filtering, 487,286 high-quality nuclei were retained for downstream analyses (50).

Batch effects across samples were corrected using Harmony (version 1.2.4) (51), followed by dimensionality reduction and visualization with UMAP. Unsupervised clustering was performed using the FindNeighbors and FindClusters functions (resolution = 1.0).

Cell-type annotation was conducted using Clustifyr (version 1.22.0, https://bioconductor.org/packages//release/bioc/html/clustifyr.html) with three independent reference datasets and manually refined using canonical markers: inhibitory neurons (GAD1/2, PVALB, SST, VIP, SV2C), excitatory neurons (SYT1, RBFOX3, CUX2, SATB2, RORB, TLE4), glial cells (SLC1A2, GFAP, PLP1, PDGFRA, PTPRC), and vascular cells (CLDN5) (25, 34, 52). Cluster marker genes were identified using Seurat’s FindAllMarkers function with the Wilcoxon rank-sum test (average log FC > 0.5, detection rate > 50%). The enrichment score of canonical markers for different cell types was calculated by the AddModuleScore function of Seurat.

3.4. Gene-Level Differential Expression and Enrichment Analysis.

In the DEG analysis, the pseudobulk-based method was employed using the Muscat R package (version 1.25.1, http://bioconductor.jp/packages/devel/bioc/html/muscat.html). After obtaining cluster-specific metacell matrices, we carried out DEG analysis using DESeq2 with a default setting (in a lesion–perilesion paired manner) or test = LRT, an ANOVA-like likelihood ratio test (LRT, for multiple-group comparison). Unless otherwise specified, the DEGs were defined by a cutoff of FC > 1.5 and P < 0.05 in two-group comparison, and P < 0.05 in multiple comparison.

Functional enrichment analysis of DEGs was conducted using EnrichR (version 3.4, https://cran.rstudio.com/web//packages/enrichR/). Gene sets from GeDiPNet (2023, http://gedipnet.bicnirrh.res.in/), DisGeNET (https://www.disgenet.org), and Jensen-DISEASES (http://diseases.jensenlab.org/), Kyoto Encyclopedia of Genes and Genomes (KEGG, Human, 2021), WikiPathways (Human, 2024), Pather (2016), etc., were employed for analysis. Unless otherwise specified, the significance was defined by a cutoff of odds ratio > 1.5 and P < 0.01.

3.5. Pathway-Level Analysis.

For pathway-level analysis, transcript-per-million (TPM)-normalized expression matrices were processed with the IOBR package to generate GSVA pathway activity scores. Differentially enriched pathways were identified using limma (version 3.66.0, https://www.bioconductor.org/packages/release/bioc/html/limma.html). Several gene set libraries were incorporated in these analyses, including: KEGG (Human, 2021), Gene Ontology (GO, 2025), WikiPathways (Human, 2024), Reactome Pathways (2024), GeDiPNet (2023), DisGeNET, Jensen-DISEASES, etc. Unless otherwise specified, significance was defined by a cutoff of |log2 FC| > 0.3 and P < 0.05 in paired two-group comparison; P < 0.01 in multiple comparison.

In the WGCNA, TPM-normalized pseudobulk expression matrices were used for module construction, using the default settings. Unless otherwise specified, the trait-associated modules were defined by │ρ│ > 0.45. For the functional enrichment of the gene modules, the similar gene set libraries in GSVA were used, and significance was defined by a cutoff of P_adj < 0.01. For the simplification of synapse-related terms, the simpleEnrichment R package was employed with the setting of measure = Sim_XGraSM_2013.

3.6. Cell-Level Analyses.

When analyzing shifts in cell counts across different populations in scRNA-seq datasets, we accounted for the compositional nature of the data, a property defined by the relative abundances of all features (e.g., cell types in our study) within a sample that always sum to unity (53). In other words, the proportion of cells in a given cluster is not arbitrary but instead exhibits a negative correlation with those of other clusters within the same sample, implying that genuine differential abundance in one cluster may induce spurious shifts in the apparent abundances of others. In our analysis, we performed CPA for each FCD subtype via the R package scProportionTest (https://github.com/rpolicastro/scProportionTest/) in both sample- and group-wise manners. In the group-wise analysis, comparisons were performed between both lesion vs. perilesion and subtypes. The significance was defined by a cutoff of FDR < 0.05 and │log₂FC│ > 0.58 (altered by 50%) (54).

To prioritize cell types responsive to biological perturbations in scRNA-seq datasets, we utilized the Augur R package (https://github.com/neurorestore/Augur) with default parameters. Augur relies on a machine learning framework that assigns a macroaveraged AUC across all pairwise comparisons to capture the relative magnitude of cell type responses to biological perturbations (55). AUC values range from 0.5 to 1.0, with higher values positively associated with the significance of cell-type responses to biological perturbations.

3.7. Immunofluorescence Staining (IF).

Fresh surgical brain specimens were fixed in 4% paraformaldehyde, cryoprotected in 30% sucrose, and sectioned at 30 μm thickness. Sections were permeabilized with 0.05% Triton X-100, blocked with donkey serum, and incubated with primary antibodies overnight at 4 °C. Primary antibodies included ZNF254 (Cat No. H00009534-B01P, Thermo Fisher Scientific, MA, USA), ATF4 (Cat No. 11815 T, Cell Signaling Technology, MA, United States), and GFAP (Cat No. ab68428, Abcam, United Kingdom). After washing, sections were incubated with fluorescent secondary antibodies, counterstained with DAPI, and mounted using fluorescence-preserving medium.

3.8. Flow Cytometric Sorting (Astrocyte and Endothelial cells).

Brain tissues were harvested under sterile conditions and rinsed with ice-cold PBS to remove residual blood and connective tissue. Samples were minced into approximately 1 to 2 mm³ fragments and enzymatically dissociated using collagenase (0.1 to 0.5 mg/mL) and papain (0.05 to 0.2 mg/mL) at 37 °C for 30 to 60 min with gentle agitation. Enzymatic digestion was terminated by adding an equal volume of medium supplemented with 10% fetal bovine serum (FBS).

The resulting cell suspension was passed through a 70 μm cell strainer to remove undigested tissue debris and centrifuged at 300×g for 5 min. For purification of astrocytes and endothelial cells, single-cell suspensions were incubated with fluorophore-conjugated antibodies against GFAP and CD31, respectively. GFAP⁺ astrocytes and CD31⁺ endothelial cells were isolated by fluorescence-activated cell sorting (FACS) or magnetic-activated cell sorting (MACS), yielding a final purity exceeding 95%.

3.9. RNA Extraction and Real-time qPCR.

The obtained cells were subjected to mRNA extraction using the TaKaRa MiniBEST Universal RNA Extraction Kit (Cat: 9767, Beijing, China), and first-strand cDNAs were synthesized with the TaKaRa First-Strand cDNA Synthesis Kit (Cat: 6110A, Beijing, China). mRNA levels were quantified on the ABI7500 real-time PCR system using TaKaRa TB Green Fast qPCR Mix (Cat: RR430S, Beijing, China). The expression levels of selected DEGs (including ZNF254, ATF4, etc.) were determined using the 2^−ΔΔCT method and normalized to ACTB (β-actin). Primers used are provided in Dataset S16.

3.10. Statistical Analyses.

Statistical analyses were performed using GraphPad Prism 9. Data are presented as mean ± SD. Unless otherwise specified, comparisons between two groups were conducted using a two-tailed Student’s t test. Statistical significance was defined as P < 0.05.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Dataset S05 (XLSX)

Dataset S06 (XLSX)

Dataset S07 (XLSX)

Dataset S08 (XLSX)

Dataset S09 (XLSX)

Dataset S10 (XLSX)

Dataset S11 (XLSX)

Dataset S12 (XLSX)

Dataset S13 (XLSX)

Dataset S14 (XLSX)

Dataset S15 (XLSX)

Dataset S16 (XLSX)

Data, Materials, and Software Availability

Raw single-nucleus RNA-seq data have been deposited in Genome Sequence Archive (GSA), China National Center for Bioinformation (HRA010445) (56). All other data are included in the manuscript and/or supporting information.