Model-Based Design of Subthalamic Nucleus Neurons

Abstract

The subthalamic nucleus (STN) plays an important role in the basal ganglia network and is a major surgical target in treatment of neurological disorders such as Parkinson’s disease (PD). Despite decades of clinical and experimental research, questions remain on the biophysical basis underlying unique features of STN neural activity and the generation of electrophysiological biomarker signals for disease states. Computational modeling is a promising tool to bridge these gaps. However, existing STN neuron models often lack morphological realism or fail to faithfully reproduce key electrophysiological characteristics observed in experimental recordings.This dissertation introduces a suite of STN neuron models ranging from a simplified four-compartment model to detailed models with realistic morphologies. We used a genetic algorithm (GA) as the foundation for our optimization efforts, and developed a new hybrid strategy that expands both the efficiency and capability of the optimization. This new hybrid optimization strategy integrates GA and simulation-based inference (SBI), which provides useful probability distributions for each parameter. This feature facilitates investigation of the mapping between electrophysiological features to biophysical parameters in the models. The initial step of the work developed a rodent STN neuron model reconstructed with the addition of axon compartments. The ion channel conductances were then optimized using a GA to fit the parameters to STN neuron firing characteristics, including spontaneous firing, rebound bursts, and frequency–current (F–I) relationships. The resulting model provided a significant improvement over the original Gillies and Willshaw (GW) model and demonstrated robustness and adaptability across additional diverse rodent STN neuron morphologies. The next step of the work developed a hybrid GA–SBI optimization strategy to improve parameter exploration and utilize SBI to capture the probability distribution of the parameter space. This hybrid method more efficiently explored the high-dimensional parameter space and generated better results when compared to GA and/or SBI only methods. Using the new algorithm, we are able to build a simple four compartment STN neuron model that provided an excellent fit to experimental recordings. We then used the model system to investigate the roles of selected ion channels on important electrophysiology characteristics such as AHP shapes and F-I curve modulations by leveraging the probability distributions generated by the algorithm. This process provided critical information to validate the biophysical realism of the new model, and enabled a flexible way to generate diverse models with different firing characteristics without re-running the entire optimization process. Finally, a non-human primate STN neuron model was developed using a 3D reconstruction from macaque STN neuron published by Sato et al. (2000). We applied the optimized hybrid framework to this morphology and improved the original Miocinovic primate STN neuron model in almost every aspect (Miocinovic et al., 2006). This updated primate STN neuron model represents a more accurate building block for future STN neuron simulation and the associated network activities in clinical neuromodulation analyses. Together, this dissertation presents a model-based design of STN neurons that bring together detailed neuronal biophysics and advanced optimization methods. The models and methods developed here aim to improve both the computational simulations of STN neurons and the overall capability and interpretability of algorithms used to construct biophysical models in computational neuroscience.