Multiscale Dynamics: from single neurons to large-scale networks Modeling and applications with a focus on Alzheimer’s Disease

Understanding the brain's function and pathology necessitates a multiscale approach, asneurological disorders stem from disruptions at the cellular level that cascade to alterlarge-scale network dynamics. This thesis integrates computational modeling across microand macroscales to investigate neural dynamics in both health and disease, with a core focuson Alzheimer's disease (AD). The work is organized into three complementary studies. First, we investigated the cellular mechanisms underlying the formation of hippocampal Place Cells (PCs). We developed a morphologically and biophysically realistic computational model of a CA1 pyramidal neuron to examine how it acquires spatially selective firing. The model demonstrates that place-field emergence results from the coordinated activation of two different inputs: a distal input from the entorhinal cortex (EC) and a proximal input from the CA3. This co-activation selectively potentiates specific synapses, leading the neuron to codefor a cue location. The model reproduces key experimental observations, including place-field stability across trajectories, stability after environment rotation, and place-field remapping. To date, it represents the only cellular model that mechanistically explains PC formation with biophysical and morphological accuracy, suggesting a potential mechanismthrough which early hippocampal impairment in AD could disrupt spatial navigation. Next, we studied Alzheimer’s disease (AD) at the macroscale using source-reconstructed EEG from 28 patients and 29 healthy controls. We quantified brain fluidity, a measure of the brain’s ability to flexibly reconfigure functional connectivity over time. AD patients showed frequency-specific disruptions: fluidity increased in the theta band but decreased in the beta band. Reduced beta-fluidity significantly correlated with higher cerebrospinal tau levels andpredicted cognitive performance better than standard biochemical markers, indicating that loss of temporal flexibility is a signature of AD-related network disconnection. Finally, we examined multiple sclerosis (MS). In 38 participants (18 MS and 20 controls), we combined MEG resting-state activity with MRI tractography and used large-scale brain models to estimate how tract lesions translate into conduction delays. Each brain region was modeled as a Stuart–Landau oscillator coupled by subject-specific connectomes. Through Bayesian model inversion, we showed that the inferred lesion–delay parameter stronglycorrelates with individual alpha-peak slowing in MEG and predicts clinical disability, whereas total lesion volume alone does not. This provides the first tract-specific quantitativemapping between myelin damage and functional slowing. Across these three studies, the thesis demonstrates how cellular alterations, loss of functional flexibility, and structural disconnection impair the timing and integration of neural information. The results provide a framework for the development of mechanistic biomarkersand suggest that multiscale models could contribute to individualized predictions of disease progression and therapeutic responses.