1. Synaptic plasticity models in brain development and neurodevelopmental disorders
- Author
-
Teramoto Kimura, SY, Vogels, T, and Butt, S
- Abstract
Mathematical and computational models are powerful tools that can help us gain a deeper understanding about the world around us. In Neuroscience, however, there is still debate as to what the role of models is in research. In this thesis we build models of synaptic plasticity of varying degrees of complexity, and at different levels of abstraction in order to generate experimentally testable hypotheses and propose new ways of thinking about synaptic plasticity in development and in neurodevelopmental disorders. We applied our models to two datasets. First, we studied a transient translaminar cortical circuit that connects layer 5b (L5b) interneurons with L4 excitatory spiny stellate neurons (SSN), or the L5b-L4 loop, that is present in the mouse barrel cortex during the first two postnatal weeks (Marques-Smith et al., 2016).. We used spiking neural networks and rate models with plastic connections to explore the mechanisms that drive these large connectivity changes, as well as the possible role these may play in the developing mouse barrel cortex. Second, we analysed extracellular plasticity data recorded from Ts65Dn animals, a mouse model for Down Syndrome. We used three abstract models to formalise and test the hypothesis that synapses in Ts65Dn animals are more saturated, i.e. the baseline synaptic weights are closer to their maximum. First, we used a theory of synaptic plasticity that explicitly fit the baseline and the maximum synaptic weight, then we fit a descriptive phenomenological model, and finally, we adapted the Moran process, a model from evolutionary biology, to describe extracellular plasticity experiments. The methodological differences between how experimental and theoretical work is carried out makes it challenging to integrate modelling into an experimental framework, and vice versa. However, through these five modelling examples, we argue that a successful model is not one that can capture all the features of the data, but rather one that can address the questions and hypotheses that cannot be studied with experiments alone.
- Published
- 2019