UC San Diego

Seizures reflect abnormal synchronized activity of a neuronal network, however, the activity dynamics preceding seizure onset are still poorly understood and algorithms for seizure prediction typically rely on local field potential recordings. Recent research has asked whether single-unit recordings might improve the predictability of seizures. Neuronal activity was found to change inconsistently before behavioral seizure onset such that an increase in variability was observed in the minutes before the onset. GABAergic interneurons integrate excitatory inputs from local and afferent networks. Such convergence of input may allow interneurons to be more sensitive to widespread neural synchrony that leads to seizures compared to principal cells. We asked whether interneurons might be more reliable predictors of seizures than principal cells. To record activity patterns of interneurons and principal cells before behavioral seizures, we used the repeated low-dose kainate model of chronic temporal lobe epilepsy in male Wistar rats. We implanted animals that developed epilepsy (>_ 2 spontaneous seizures) with tetrode arrays and video monitored the rats (n=3) while local field potentials and single unit activity were recorded from CA1 and CA3. We identified a total of 20 behavioral seizures (rat 1: 4 seizures across 4 days, rat 2: 5 seizures across 3 days, rat 3: 11 seizures across 5 days). We assessed the activity patterns of hippocampal pyramidal cells (n=193) and interneurons (n=69) during the 5 minutes preceding the behavioral seizure onset. First, we characterized baseline firing rates during 100 second epochs. We found that only ~4% of principal cells exhibited firing rates that deviated by more than 3 standard deviations from the baseline in the minutes before the seizure onset, whereas ~25% of interneurons deviated from baseline up to 2 minutes before seizure onset. The average increase in firing rate of all interneurons was +60% during the 2 minutes before the seizure onset compared to baseline. Interneurons are thus a much better predictor of seizures during the minutes before the seizure onset. Because of the large effect size, even small numbers of recorded interneurons can reliably predict a seizure. Together, these data suggest that knowing the change in firing from a previous seizure improves prediction, but that the gains from using interneurons for detection algorithms would be much more robust