Abstracts

Rationale: In focal epilepsy, epileptogenesis refers to the complex phenomenon of network reorganization after injury to produce a state susceptible to recurrent spontaneous seizures. Various mechanisms have been proposed at the molecular and cellular anatomic level. Homeostatic synaptic plasticity (HSP) is a collection of intrinsic molecular mechanisms designed to “tune” the network to prevent runaway synaptic potentiation or depression, and likely plays a critical role in both recovery after injury as well as epileptogenesis. In this study, we explore network-level mechanisms of epileptogenesis via computer modeling in an artificial neural network, both acutely after lesioning, and chronically after HSP has taken place. Methods: We created a fully-connected 2D network of 104 “neural units” arranged in a hexagonal configuration. Each unit was connected to others with a difference of Gaussians weight profile, mimicking an “on-off” receptive field, such that excitation and inhibition were balanced. Neural units fired with simple input summation exceeding an arbitrary threshold and had a post-firing refractory period. We lesioned the network to produce a focal area of approximately 70% neuronal dropout with different lesion shapes (uniform, Gaussian, and combination uniform-Gaussian lesions). Discrete-time network simulations were performed for both acute and chronic lesion models with both impulse and stochastic stimulation. To produce chronic network behavior, we mimicked an HSP-like mechanism by renormalizing excitatory input weights after lesioning, and simulated the resultant network behavior. Results: The unlesioned network model exhibited propagatory behavior (bursting) only below a critical neural firing threshold T­­c. After lesioning, propagatory behavior was suppressed in the lesion even for thresholds < Tc. After HSP, propagatory behavior was enhanced in the lesion due to increased temporal summation, and was seen at thresholds > Tc. Additional self-sustaining propagatory behavior (seizures) was present at thresholds > Tc.  Different lesion geometries influenced whether or not propagatory activity escaped the lesion; for the same neural firing threshold ranges, with more “focal seizures” seen at a threshold ranges with circular uniform lesions, and more “generalized seizures” seen with Gaussian lesions. In combination circular and Gaussian lesions, escape of seizure activity occurred preferentially through the Gaussian region. Conclusions: Epileptic activity in an artificial neural network can be explained by synaptic homeostasis after lesioning and neural unit dropout. Chronically after HSP-like mechanisms, propagatory, seizure-like activity is enhanced via increased temporal summation. Focal seizures in this model are seen with a well-defined “border” between more deafferented neural units and those that remain well-connected. Escape of focal seizures from an area of lesioning occurs preferentially along paths without such borders. Funding: None