Plasticity in lateral amygdala after Pavlovian fear conditioning -- A Computational Study

Computational neuroscience provides tools to abstract and generalize principles of neuronal function using mathematics and computers. This dissertation reports a computational model of a specific neuronal sub-circuit that provide insights into fear memory formation in vertebrates, including methods to develop biologically realistic reduced order models of single neurons.;One of the main contributions of the dissertation is an explanation of how and why certain neurons are recruited into a memory trace. For this, we developed a biophysical model of the rodent lateral amygdala (LA) and then examined how particular LA neurons are assigned to the fear memory trace, i.e., how fear memory is formed in a rodent brain, after Pavlovian fear conditioning. The model revealed that neurons with high intrinsic excitability are more likely to be integrated into the memory trace but that competitive synaptic interactions also play a critical role. We also examined the relative contributions of plasticity in auditory afferent (thalamic, cortical) neurons vs. within LA. This revealed that plasticity in afferent pathways to LA is required for fear memory formation, but that once formed, the plasticity in afferent pathways was not needed. The model then provided insights into how 'competition' was implemented at the single cell level, including the role of excitatory connections among neurons, of disynaptic inhibition, and of neuromodulation. These principles should also apply to other forms of memory in brains. We then investigated another related concept of specificity of memory, i.e., how does memory of fear to a particular tone prevented from being elicited by another tone. Analysis showed that formation of tone fear memory in LA involves plasticity in intrinsic excitatory and inhibitory connections within LAd and this intrinsic plasticity also ensures specificity for that memory.;Neuronal network models presently use simplified single cells models with either one or two compartments. This is largely due to the fact that computational overheads become prohibitive with more detailed models. We also report a procedure to develop a reduced order model matching passive properties, current injection traces, and preserving some synaptic integration features. Comparisons are made at both single cell and at 100- cell network model levels. Analysis showed that a model with three compartments provides a good compromise between biological realism and ease of computation.