The conditional knockout (KO) of Ndufs4 in only GABAergic interneurons leads to a severe epilepsy phenotype, suggesting GABAergic interneurons drive the severe and often fatal epilepsy phenotype commonly reported in Leigh Syndrome (LS) patients. Dysfunctions or loss of parvalbumin (PV) interneurons, a subtype of GABAergic interneurons, have been shown to play a key role in the mechanisms of various forms of epilepsy both in human and animal models. The present study aims to target PV interneurons. We hypothesized that KO of Ndufs4 in PV interneurons will cause dysfunctions or loss of PV neurons leading to epilepsy in our cell-specific model of LS. Experimental mice models with Ndufs4^flx/flx/PVCre^flx/+ genotype for the mutants, and Ndufs4^flx/flx/PVCre^+/+ genotype for the controls were used. For imaging experiments, Ndufs4^flx/flx/Ai14^flx/+/PVCre^flx/+ were used for mutants and Ndufs4^+/+/Ai14^flx/+/PVCre^flx/+ were used for controls. Seizure susceptibility was assessed by recording occurrence, frequency and duration of seizures and epileptiform events. Mice susceptibility to provoked seizures was examined by the pentylenetetrazol (PTZ) challenge. Assessment of cell loss was tested in imaging studies. Ai14-labeled PV interneurons in key areas associated with epilepsy were counted between the two groups. Finally, to assess motor dysfunctions comorbid to epilepsy, I tracked the movement of mice of both genotypes. Our results showed PV mutants had an increase in the frequency of spontaneous myoclonic seizures and interictal spikes on electroencephalograms (EEGs). There was no difference in seizure susceptibility to PTZ seizures between mutants and controls, nor any major impairments in locomotor activity or anxiety like behavior in PV mutants. Finally, no cell loss changes in PV mutants were detected. In conclusion, PV mutants display a mild seizure phenotype with no cognitive or motor abnormalities, suggesting targeted Ndufs4 KO in PV interneurons drives a small portion of the severe epilepsy phenotype observed in LS.

Epilepsy is a neurological disorder characterized by the presence of seizures (periods of abnormally synchronized neural hyperactivity) and interictal spikes (transient abnormal neural synchronization that occurs between seizures). Different genetic mutations and backgrounds lead to different forms of epilepsy, which in turn may lead to different manifestations of epileptiform neural activity. I used machine learning (ML) to detect interictal spikes in mouse models of different epilepsies. I used neural activity previously recorded in mice using two electrocorticographic (ECoG) electrodes and one electromyographic (EMG) electrode. I used data from mouse models of Dravet syndrome (DS; Heterozygous Scn1a gene deletion), focal cortical dysplasia (FCD; Pik3ca gene mosaic), Leigh Syndrome (LS; GABAergic Ndufs4 knockout) and, Alzheimer's Disease (AD; Increased beta-amyloid production), as well as wild type (WT) control. I used recordings binned into 10 second interictal spikes. I then used a computer algorithm that extracted 96 features - events that characterize ECoG and EMG electrical signals. These features and the manually identified interictal spikes were used to train several ML models to score unidentified interictal spikes in the remaining recorded data. The best performing ML algorithm had a mean test accuracy between 60% and 80% for each of the different models of epilepsy, but the features it used were different in each epilepsy mouse model. These results suggest that, while our ML-based method may capture epileptic activity with high accuracy, its success relies on features that are characteristic of each type of epilepsy. These results suggest the potential need to utilize different ML models for different forms of epilepsy in order to attain the highest possible accuracy if used for real-time interictal spike detection and potential seizure forecasting.