Abstracts

Rationale: Sleep and memory disorders are common comorbidities in epilepsy, yet little is known about the micro-architectural changes in sleep in these patients. Spindles are part of a coordinated neural network essential for memory formation and occur during NREM stages when interictal epileptiform discharges (IEDs) are frequent. Therefore, spindles may be susceptible to disruption by epileptiform activity. Understanding these relationships during sleep could reveal new insights into mechanisms of cognitive impairment in epilepsy and lead to novel therapeutic approaches.

Methods: We studied 48 patients with temporal lobe epilepsy (TLE) and 20 patients without epilepsy (controls, HC) who were previously admitted to the Epilepsy Monitoring Unit at Massachusetts General Hospital. Continuous scalp EEG was obtained in all subjects, and TLE subjects also had foramen ovale (FO) electrodes placed to record medial temporal lobe activity. Previously validated deep learning algorithms for automated sleep staging and spike detection on FO electrodes were used (Jaoude et al. 2020 Sleep; Jaoude et al. 2020 Clin Neurophysiol). Data were pre-processed and analyzed in Matlab to quantify sleep parameters, including sleep stages and sleep fragmentation (# sleep transitions to awake or drowsy states). A wavelet-based automatic spindle detector was used to quantify spindle density and parameters. Statistical analyses included direct group comparisons (two-sample t-test, alpha 0.05), and analyses of the associations of sleep parameters with IEDs (linear regression).

Results: Subjects in this study were of similar age (means ± std: TLE 41.2 ± 15.7 vs. HC 42.6 ±17.5 years). The percentages of total sleep times spent in each sleep stage were not significantly different between TLE and HC. There was also no significant difference in sleep fragmentation. Among TLE patients, sleep transitions and sleep fragmentation measures were not significantly associated with medial temporal lobe IED frequency. In contrast, spindle density (# / minute) was significantly lower in TLE compared to HC (means ± std: TLE 2.0 ± 0.78 vs. HC 2.8 ± 0.8, p< .001). This was observed throughout fronto-central channels on scalp EEG and was characterized by a ~30% reduction in many channels.