Seizure dynamics in humans, animal models, and in vitro systems generally unfold over a relatively slow timescale, typically lasting 100 seconds or more, both in terms of seizure duration and the intervals between seizures. Researchers propose that both the initiation and termination of seizures are governed by a slowly changing variable. When this variable falls within an optimal range, the likelihood of a seizure occurring becomes non-zero. In such a state, seizures can be triggered by external stimuli, activity from other brain areas, or intrinsic neural noise. Conversely, if the slow variable lies outside this range, the system is unable to support seizure activity. The precise biophysical nature of this slow variable remains unclear, but it is likely influenced by multiple interacting mechanisms. These may include activity-driven changes in extracellular potassium levels, metabolic factors like oxygen and NADH, short-term synaptic depression and recovery, and afterhyperpolarization following bursts of neural activity.

In this study, a population-level neural model was developed incorporating both synaptic depression and afterhyperpolarization mechanisms. The model is adapted from the seizure modeling framework originally proposed by Wendling et al., which itself builds on foundational work by Wilson and Cowan. It focuses solely on excitatory neurons, where the firing rate depends nonlinearly on membrane voltage. The voltage is influenced by excitatory synaptic inputs, modeled through a time-dependent afferent firing rate and synaptic response dynamics.

To enhance realism, short-term synaptic depression was added based on experimentally derived parameters, capturing the transient reduction in synaptic strength following activity. Additionally, the model includes afterhyperpolarization—an inhibitory process where neurons become less excitable after bursts of activity—also informed by experimental data. Random noise was used to drive the system, simulating spontaneous neural fluctuations. Simulations were conducted in MATLAB Simulink, with a cortical slice’s pyramidal layer represented as a network of 81 interconnected units. Each block in the model represents a micro-culture within the pyramidal layer of a cortical slice, with the capacity to generate its own seizure-like activity. These blocks are interconnected, with each unit feeding back into itself and also influencing its neighboring blocks through synaptic connections. The strength of these connections, or synaptic weights, depends on the physical distance between the blocks—closer blocks have stronger interactions.

To ensure the entire network exhibits seizure dynamics, the synaptic weights were calibrated so that the combined activity of all blocks results in a seizing network state. The spatial decay of synaptic influence was modeled using distance-dependent weights, derived from experimental measurements.

In experiments, electrical stimulation was applied to specific subregions of the hippocampus—either CA3 or CA1—and the resulting evoked activity was measured across the pyramidal layer. By analyzing how the response diminished in distant regions, an exponential decay curve was fitted to determine the spatial spread of activity. This provided a length constant, representing how far the effects of stimulation propagate. Based on these fits, a length constant of 150 μm was used for CA3 and 100 μm for CA1 in the model.

These experimentally derived parameters ensured that the spatial dynamics of the simulated network closely reflected observed biological behavior under control conditions, where no spontaneous or evoked seizures were present.