Abstracts

Rationale: Neurostimulators may act to prevent focal seizures, although the mechanism is unknown. We hypothesize that they cause stimulation-induced network remodeling that increases seizure threshold. Here, we employ computational modeling to understand the relationship between network structure and seizure threshold. As our first step, we aim to identify network motifs that lower seizure threshold.

Methods: We adapt a previously published model of seizures,¹ which tracks four dynamical variables for each of its neurons: membrane voltage, firing threshold, intracellular chloride concentration, and conductance of the slow afterhyperpolarization current. Interneuronal connection strength decreases with distance, following a "Mexican Hat" structure. Inhibitory neurons are implicitly modeled. Model seizures are generated by adding background noise. We implement spike-timing dependent plasticity (STDP), which mathematically describes how the network’s excitatory connections are updated after a model seizure. Each neuron’s total excitatory output is held constant, so STDP only affects where a neuron’s output is directed. Seizure threshold is then arbitrarily defined as the level of noise at which probability of seizure is 50%, and we examine seizure thresholds before and after network structure is updated by STDP.

Results: Model seizures reproduce core phenomenological aspects of human seizures, namely: (1) an outwardly expanding tonic wavefront and (2) a back-propagating clonic core. These two features have been identified previously in microelectrode array recordings of human seizures. Network updating by STDP after a single seizure decreases seizure threshold by an average of 55% (range 23-81%). Seizure-induced structural changes are also highly heterogenous. However, we identify two structural motifs that are sufficient to explain >90% of the variance in updated network structures. One motif captures how neural activity is passed across the entire sheet. The other describes how neural energy is focused inward, towards the ictal core. Every updated network can be expressed as a combination of these two motifs, and seizure threshold varies parametrically across their combinations.

Conclusions: We show that network structure affects seizure threshold, even in the absence of changes in total excitatory output. We then demonstrate core network motifs that control our model’s seizure threshold. We conclude that network structure is itself a viable target for manipulation by neurostimulators. In the future, we hope to implement an in silico neurostimulator to identify stimulation strategies that ameliorate epileptogenic structural motifs, thereby increasing seizure threshold, in line with our hypothesis.

Reference:
1. Liou JY et al. A model for focal seizure onset, propagation, evolution, and progression. Elife. 2020 Mar 23;9: e50927.

Funding: NINDS Research Education Grant (R25), awarded to Columbia University