Abstracts

Rationale: Patients with mesial temporal lobe epilepsy frequently do not become seizure free with medications and many are not surgical candidates. While open-loop electrical stimulation of the hippocampus has shown promise, intelligent control systems capable of learning how the brain responds to stimulation and adapting accordingly have the potential to deliver more effective and efficient therapy. However, the behavior of these systems is poorly understood, as studying them exclusively in vivo is time and cost prohibitive. A more tractable approach is leveraging a computational platform that couples a model of hippocampal dynamics with an intelligent control system to produce solutions that can be rigorously validated in vivo. Here we present the first piece of this platform – a computational model of the hippocampus with functionally distinct but connected CA3 and CA1 regions – and demonstrate the effects of stimulation in the modelMethods: A neural mass model of the hippocampus based on work by Wendling et al. was extended to include two populations representing the CA3 and CA1 regions of the rat hippocampus (Wendling et al., 2005). Each population contains three neuronal masses with different dynamics: excitatory, fast inhibitory and slow inhibitory. The two populations are connected through an excitatory projection from CA3 to CA1 and an inhibitory projection from CA1 to CA3 (Figure 1). The dynamics of the model are described by a set of 22 differential equations with parameters for the synaptic gain and delays of each neuronal mass and the connectivity between the masses. The inputs are background Gaussian noise and the outputs are the summed post-synaptic potentials in each population (CA3 or CA1) representing local field potentials. The synaptic gains of the model were fit using a global optimization algorithm to minimize the difference between the frequency spectrum of the model output and an in vivo recording.Results: Stimulation of the CA3 population at 7Hz, 17Hz and 35Hz induced an increase in power at the respective stimulation frequency, and smaller increases in power at harmonic multiples in CA3, and distributed changes in the spectrogram of CA1 concentrated at 1-5Hz (Figure 2A). Decreasing the synaptic gain of the inhibitory neurons of CA3 produced sporadic spikes that propagated to CA1. Stimulation of CA1 showed no substantial increase in power at 7Hz. However, stimulation at 17Hz and 35Hz caused increased power at these respective frequencies, along with the harmonics observed in CA3 stimulation. Additionally, the stimulation at 17Hz and 35Hz exhibited a substantial decrease in the amplitude of the CA3 signal and suppressed the sporadic spiking (Figure 2B).Conclusions: These results lay the foundation for a platform to investigate and develop intelligent neural control systems that can be generalized to in vivo settings and can be validated through ongoing in vivo experiments.