Multiscale Modeling of the Brain as a Function of Metabolic Constraints

Multiscale Modeling. Our computational neuroscience modeling of network connectivity under energy constraints integrates scales that range from neurons and astrocytes all the way up to emergent dynamics seen with fMRI, in predicting individual trajectories.

Our project is structured such that empirically-established macroscopic (human whole-brain neuroimaging) features both motivate and are motivated by mechanistic neuronal experiments in animal models. Connecting the two scales is a multiscale computational model, which aims to demonstrate how even minute changes in neurons’ access to energy create emergent effects at the whole-brain scale, laying the foundation for an eventual impact in cognitive functioning. One important motivation for this type of computational modeling, beyond the mechanistic integration of our results, is to better understand, and thus predict, individual clinical trajectories with respect to brain aging. Previous efforts at multiscale biomimetic computational modeling have succeeded in recording fMRI-scale dynamics from neuronal parameters. However, in spite of the fact that the brain is exquisitely sensitive to fluctuations in energy availability, these models do not explicitly model metabolic parameters (e.g., glycogen-buffering via astrocytes). To expand multiscale models to include metabolic components, we are building upon the platform of The Virtual Brain (TVB), a neuroinformatics tool developed as part of the EU’s BlueBrain/Human Brain Project, which enables simulation of the human cortex from neural mass to fMRI scales. This expansion requires significant modification of two types. At the neuronal scale, we are incorporating metabolic parameters established by our animal experiments into the standard sets of equations used to describe neuron firing dynamics (e.g., Hodgkin-Huxley, Wilson-Cowan, Jansen-Rit). At the fMRI scale, we are modifying networks to include “payoff” weights with respect to functional processing as well as “cost” weights with respect to glucose utilization. This is necessary in order to understand the processes by which the brain functionally reorganizes (as per our network instability results) when faced with changes in available energy. As a first step, we developed a fitting algorithm, using our fMRI data to constrain neuron-scale parameters. While this work is just starting, preliminary results are encouraging, as the algorithm recovers a parameter for axonal conduction velocity to within approximately 5% of values established by our animal experiments.