Spatial transcriptomics in epilepsy research: Early successes, opportunities, and challenges

Epilepsy affects more than 70 million people worldwide1 and remains an important cause of global disability and mortality.2 Despite treatment advances, >30% of all epilepsies remain refractory to medication.3-5 Improving the mechanistic understanding of epilepsy pathogenesis remains a critical unmet need.6 Epilepsy is often conceptualized as a disease of brain networks, not only as derangements within a connectome of neurons and network hubs,7 but also as pathological changes within an interactome of molecular processes among neurons and surrounding glia.8 Considering this conceptual framework,9 understanding the spatial relationship and interaction among cells within epilepsy networks is crucial. Transcriptomics, as a method of characterizing cellular gene expression through sequencing mRNA, has rapidly advanced understanding of human brain cellular diversity. Transcriptomic sequencing technology has quickly evolved through three stages: bulk sequencing, single-cell RNA sequencing (scRNA-seq), and spatial transcriptomics (ST; Figure 1).11 The evolution from bulk to single-cell sequencing12 has revealed thousands of regionally specific brain cell types,13 with individual cortical areas demonstrating >100 unique neuronal and nonneuronal subtypes.14 However, epilepsy is foundationally a network disease. Although single-cell transcriptomics can unlock new insights into epilepsy pathogenesis, spatially resolved gene expression characterization is needed for comprehensive molecular understanding of altered circuit function. ST addresses this by profiling single-cell gene expression in intact tissue, preserving spatial context critical for understanding neuronal–glial interactions, defining epileptogenic zone boundaries, and determining network-level expression changes. As ST becomes more accessible, building expertise and research communities to catalyze its use in epilepsy discovery is essential. At the 2024 American Epilepsy Society annual meeting, our team organized an Investigators Workshop highlighting ST in epilepsy research. We now formally summarize and expand that presentation in this article. ST technologies are categorized as either imaging-based or sequencing-based, with advantages and limitations of each summarized in Table 1. Imaging-based approaches use fluorescent probes to visualize specific genes directly in tissue sections. These methods achieve subcellular resolution (<300 nm) and include in situ sequencing (e.g., fluorescence in situ sequencing [FISSEQ]) and in situ hybridization (e.g., multiplexed error-robust fluorescence in situ hybridization [MERFISH]). MERFISH labels RNA with error-robust barcodes detected through successive rounds of fluorescent hybridization, whereas FISSEQ directly sequences RNA within fixed tissue with higher signal-to-noise ratio but lower throughput. Both strategies excel at detecting individual RNA molecules within single cells but are limited to preselected gene panels. Sequencing-based approaches use spatial barcodes on specialized arrays to capture mRNA from tissue sections, followed by next generation sequencing to profile the entire transcriptome. Spatial information is restored by aligning sequencing data with tissue images.12, 15, 16 Although these methods typically have lower resolution than imaging approaches, they offer unbiased, genome-wide gene expression profiling without requiring preselection of target genes. Recent advances have improved resolution, sensitivity, and scalability to accommodate larger samples. Individual platforms within both categories continue to evolve rapidly (Figure 2). Furthermore, hybrid approaches that combine the advantages of both methods are being developed, with some platforms adding multiomic (e.g., transcriptomic + proteomic) capabilities. A schematic of ST workflows is provided in Figure 3. ST has been successfully applied to mouse and human brain tissue in both physiological20-22 and pathological23 conditions, including epilepsy.24 However, ST requires careful study design due to high costs, lengthy processing times, and lack of technical standardization. Four key considerations guide ST study design (Table 2): experimental question, experimental groups, available resources, and anticipated results. The experimental question determines the optimal ST approach (Figure 4). Data-driven studies aim to characterize total spatial gene expression without specific hypotheses, making them discovery-oriented and best suited for nontargeted, sequencing-based methods that profile the whole transcriptome. These studies have generated valuable transcriptomic atlases.20, 21, 26, 27 Hypothesis-driven studies test specific predictions and typically utilize imaging-based ST methods for higher spatial resolution and sensitivity.28, 29 These require appropriate controls, which may leverage existing mouse30, 31 or adult human32 atlas datasets, or investigators may incorporate their own ST control group. Choosing appropriate control tissue is nontrivial, given variability among different brain regions and intersubject variability. Additional considerations, including species, sample type (fixed versus frozen), platform availability, and budget, ultimately guide platform selection. Investigators must consider a priori the types of data they will obtain and how they may be used to address their research questions. Critical preanalytical considerations include tissue quality, RNA integrity, and sample processing protocols, as RNA degradation significantly impacts data quality and interpretation. Surgical samples present unique challenges due to inherent variability in tissue handling, ischemic time, seizure activity prior to resection, and patient-specific factors (e.g., antiseizure medications, seizure frequency). Standardized protocols for tissue collection, processing, and storage are essential to minimize technical variability. Additionally, batch effects arising from processing samples across different times or conditions must be carefully controlled and computationally corrected during analysis. ST experiments frequently generate petabytes of data, which undergo multiple steps of data processing, so collaboration with a bioinformatics specialist experienced in large datasets is essential. Computational expertise is especially critical in higher order, inferred datasets like cell–cell interaction, which may be critical to understanding epileptic circuits. Quality control metrics should assess RNA integrity numbers, tissue morphology, sequencing depth, and spatial mapping accuracy before downstream analysis. Once differential gene expression lists, cell subtypes, and their spatial locations are identified, one can perform gene ontology analyses and gene set enrichment analyses to identify features of gene expression that provide insights into changes in molecular and cellular function.25, 33 Lastly, data validation is critical in ST and should be considered during study design (Table 3).19, 34 ST in preclinical and clinical epilepsy research remains nascent. Although bulk and single-cell techniques have been used in epilepsy models to define cellular heterogeneity,35, 36 glial dysfunction,37, 38 immune infiltration,39 and interneuron dysfunction,40 our literature search revealed few publications applying ST to the neurobiology of epilepsy (Table 4). Fewer of these studies utilized spatial data in their analyses, and we failed to find any ST studies assessing cell–cell interaction networks in epilepsy. Two preclinical studies utilized a mouse kainic acid model of temporal lobe epilepsy (TLE). The first integrated scRNA-seq, single-nucleus RNA sequencing, and ST to create a high-resolution atlas of hippocampal cellular subtypes, uncovering important TLE-induced molecular changes underlying epilepsy pathogenesis, including glial inflammatory responses, suppressed γ-aminobutyric acidergic synapses, and reactive neurogenesis.43 The second identified specific overexpression of C-C chemokine ligand 5 in seizure hippocampi and functionally validated this finding through maraviroc-induced pathway blockage, preventing microglia activation and neuron degeneration in the seizure mouse.42 The first study was limited by gene panel size, and both were limited by model generalizability. Two studies applied sequencing-based ST (10x Genomics Visium) to surgically resected human focal cortical dysplasia (FCD). The first study implicated dysplastic neuronal mTOR and autophagy pathways and balloon cell inflammatory pathways in FCD pathogenesis.24 The second synthesized a transcriptional catalog of type IIb FCD and identified noncanonical markers of signaling and neurotransmitter pathways in cytomegalic dysplastic neurons, noting SLC6A5 as a specific marker.44 Both studies were limited by small sample sizes, tissue heterogeneity, and lack of well-established control data. Finally, Busch et al., employed sequencing-based ST (GeoMx) to study comorbid memory loss in human TLE.41 Utilizing spatial data, investigators localized the strongest molecular signal to hippocampus CA3, an anatomical substructure critical to memory formation. BDNF was identified as a central hub in regulating CA3-related networks. As such early efforts expand, ST has the potential to reach beyond pathological characterization, translating to aid biomarker discovery, surgical decision-making, and therapeutic development. Histopathological diagnosis remains the gold standard in determining etiology and likely outcomes of various structural epilepsies. This is perhaps best demonstrated in TLE46, 47 and FCD.48-50 Current histopathological classification criteria rely upon features such as cellular morphology and density to diagnose and stratify. However, molecular techniques are increasingly incorporated into diagnostic practice.51 Molecular biomarkers have transformed clinical oncology, uncovering new treatment pathways and driving tumor reclassification,52, 53 as well as defining the borders between healthy and diseased tissue. Molecular techniques like single-cell transcriptomics have revealed mTOR/MAP kinase54 and other pathways involving synaptic function and calcium dynamics alterations in the formation of FCD.55 ST may likewise alter how structural epilepsy is diagnosed, potentially forcing a deeper understanding of epilepsy pathogenesis on an individual patient level. ST holds transformative potential for integration with conventional diagnostic modalities (magnetic resonance imagining [MRI], positron emission tomography [PET], electroencephalography [EEG]) to generate noninvasive molecular biomarkers. Building on MRI–transcriptomics integration in Alzheimer disease,56 similar frameworks could link preoperative neuroimaging to spatial molecular profiles, enabling molecular phenotyping of epileptogenic zones prior to surgery. Combining ST with multielectrode array recordings, which have revealed altered neuro- and gliogenesis in TLE,57 could elucidate molecular mechanisms of hyperexcitability with subcellular resolution. Nonspatial transcriptomic profiling from patch-clamp recordings58 and stereotactic EEG studies59 has demonstrated the feasibility of molecular–electrophysiological integration; incorporating spatial context could define molecular changes across epileptic networks. Conversely, well-established electrophysiological and neuroimaging techniques could help validate emerging ST findings (Table 3). However, substantial challenges remain. Data harmonization across platforms with different resolutions, coordinate systems, and formats requires sophisticated coregistration. Computational interpretability of integrated high-dimensional datasets, namely, millions of gene measurements combined with electrophysiological and imaging data, demands advanced analytical frameworks. Although deep learning shows promise,60 limited training datasets in epilepsy constrain current applications. Standardized data-sharing protocols, common data elements, and collaborative consortia are essential to build infrastructure for robust multimodal integration and realize ST's translational potential. Spatial transcriptomics, as part of a multiomic framework, represents a paradigm shift from descriptive pathology to mechanistically informed, spatially resolved molecular medicine in epilepsy. By revealing molecular disruptions, dysregulated cellular subtypes, and aberrant signaling pathways within epileptic circuits, ST can transform histopathological pattern recognition into mechanistic understanding. Clinically, colocalizing ST-derived molecular signatures with electrophysiological markers of hyperexcitability (EEG, intracranial recordings) and structural/metabolic abnormalities (MRI, PET) may improve prediction of treatment response, stratification of surgical candidates, and delineation of resection boundaries. Integration with artificial intelligence (AI) can identify previously unrecognized molecular subtypes and disease patterns from large-scale spatial transcriptomic datasets, enabling precision therapies. This mechanism-based, AI-augmented approach holds promise for improving seizure freedom rates, reducing surgical morbidity, and identifying disease-modifying therapies for patients with drug-resistant epilepsy. Donald J. Phillips: Conceptualization (co-lead); writing—original draft (co-lead); writing—review and editing (equal); visualization (main). Autumn S. Ivy: Conceptualization (co-lead); writing—original draft (co-lead); writing—review and editing (equal). Samuel Guzman: Writing—original draft (supporting); writing—review and editing (equal). Saman Hazany: Writing—original draft (supporting); writing—review and editing (equal). Xiangmin Xu: Conceptualization (supporting); writing—original draft (supporting); writing—review and editing (equal). None. The authors declare no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines. This commentary synthesizes published literature; no new data were generated. Readers can contact the corresponding author for details on accessing source data from original studies.