Understanding cerebellar disorders: a computational investigation of ataxia through spiking neural networks

This thesis presents a comprehensive computational modelling approach to understand spinocerebellar ataxias (SCAs), a group of neurodegenerative disorders characterized by progressive motor dysfunctions and the current lack of effective therapies. We developed a point-neuron circuit model (a spiking neural network - SNN) of the mouse cerebellum, with a detailed biophysical reconstruction and refined versions of the Extended Generalized Leak Integrate and Fire tuned on each neural population. We have employed an advanced Python-based modelling framework - the Brain Scaffold Builder (BSB) – able to interface with the SNN simulator NEST, to reconstruct the cerebellar network, that mirrors key aspects of cerebellar architecture and connectivity, and to simulate the network dynamics. We have introduced ataxic-specific alterations of some neural mechanisms at structural (Purkinje cell -PC- density and PC “dendritic complexity”) and the corresponding functional level. The network attributes and parameters are easy to be reconfigured, since the user-friendly scaffold modelling approach in BSB. Some of the alterations were introduced indirectly, identifying a model equivalent available in the SNN. The typical flow of the disease was followed, starting for instance, with the reduction of the PC “dendritic complexity” by developing a pruning algorithm on the PC dendritic tree arborization, which impacts on synaptic density, and, therefore, on excitatory/inhibitory balance on PCs. Then the simulations proceeded with a reduction in PC density and, subsequently, dendritic tree complexity (DCI) and size variables were merged to reproduce the more advanced stages of the disease. Firstly, number of convergences/divergences and synapses where analysed: they highlighted the loss of important connections between PCs and surrounding cells. Consequently, the corresponding functional alterations were examined. These results confirmed our initial hypotheses from the structural modifications: as the density, DCI, and size decrease, the DCN become progressively more disinhibited due to a reduction in inhibitory input from PCs. Moreover, the impact of shrinkage appears more pronounced than that of DCI; notably, a modest shrinkage (around 25%) enhances the effect of DCI, whereas a 50% reduction results in a minimal impact, being the size too modest for connections to be maintained. Eyeblink conditioning experiments were also performed, during which we simulated inputs from mossy fibers (msf) and climbing fibers (CF), and recorded the responses of PCs and deep cerebellar nuclei (DCN). In this case as well, we observed that the severity of network alterations was predominantly determined by PC density, followed by cell size, and ultimately by DCI. These modifications yielded promising results consistent with observed pathological phenomena in SCAs, thereby reflecting the validity and robustness of our modelling approach. The model holds significant potential for further refinement. Indeed, with the integration of increasingly detailed experimental data from ataxia, this framework could evolve into a more realistic and predictive tool for investigating the complex mechanisms underlying SCAs. This could also lead to design and testing of therapeutic strategies: in future the validated cerebellar ataxic microcircuit can be embedded in whole-brain simulators in order to investigate the impact of altered cerebellar activity on the other brain regions.