Differential Transcript Expression and Alternative RNA Splicing Patterns to Differentiate Focal vs. Generalized-Onset...

Retrospective diagnosis of a seizure type is pivotal for effective management and treatment of epilepsy. Previously, we demonstrated that RNA signatures could discriminate between non-epileptic spells and epileptic seizures. Here, we investigate the utility of alternative RNA splicing to distinguish generalized versus focal epileptic seizures. Blood samples were collected at baseline, 4-6 h post-seizure, and at discharge from 27 patients undergoing video-electroencephalogram (vEEG) monitoring at the Emory University Hospital. Epileptologists determined seizure classification through vEEG data review. RNA was extracted, sequenced, and analyzed for RNA expression and transcript usage. Classification models were generated to distinguish between patients who had a focal or generalized seizure. The study shows transcriptomic profile changes following EEG-verified focal and generalized seizures. Compared to baseline, focal seizure exhibits limited changes in transcriptomic expression 4-6 h post-seizure and discharge samples. In contrast, generalized seizures demonstrated a broader transcript response, with 74 differentially expressed transcripts at 4-6 h and 70 at discharge. The changes were also evident across different time points between focal and generalized seizure. The study for the first time described the landscape of isoform switching in seizure type. Notably, significant isoform switching without differences in gene expression was observed. We identified 2689 isoform switches linked to 1249 genes among which 742 genes were sensitive to nonsense-mediated mRNA decay (NMD). Significant switches were observed in genes such as CORO1C, ZBTB44, SNHG1, and RPS17. Notably, we also observed novel isoforms, including CD300 (MSTRG.26116.1), RNF216 (MSTRG.52862.7), and RN7SL1 (MSTRG.17010.3) which exhibited significant switching, revealing potential new regulators of gene expression. Differentially expressed transcripts were utilized as classifiers for machine learning (ML) modeling using random forest (rf) and radial support vector machine (rSVM) algorithms, achieving ~ 83% accuracy in classifying generalized seizures, and multivariate adaptive regression splines (mars) algorithm achieving 100% accuracy in identifying focal seizure events. Our findings of blood transcript expression changes, including isoform switch analysis, underscore the potential of blood-based transcriptome analysis for retrospectively distinguishing seizure types and identifying biomarkers for epilepsy management.