Work

A multi-frequency whole-brain neural mass model with homeostatic feedback inhibition

Year: 2026

Type: article

Abstract: Whole-brain models are valuable tools for understanding brain dynamics in health and disease by enabling the testing of causal mechanisms and identification of therapeutic targets through dynamic simulations. Among these models, biophysically inspired neural mass models have been widely used to simulate electrophysiological recordings, such as MEG and EEG. However, traditional models face limitations, including susceptibility to hyperexcitation, which constrains their ability to capture the full richness of neural dynamics. Here, we developed and characterized a new version of the Jansen-Rit neural mass model aimed at overcoming these limitations. Our model incorporates inhibitory synaptic plasticity (ISP), which adjusts inhibitory feedback onto pyramidal neurons to clamp their firing rates around a target value. Further, the model combined two subpopulations of neural cortical columns oscillating in α and γ, respectively, to generate a richer EEG power spectrum. We analyzed how different model parameters modulate oscillatory frequency and connectivity. We considered a model’s showcase, simultaneously fitting EEG and fMRI recordings during NREM sleep. Bifurcation analysis showed that ISP increases the parameters’ range in which the model exhibited sustained oscillations; the target firing rate acts as a bifurcation parameter, moving the system across the bifurcation point, producing different oscillatory regimes, from slower to faster. High frequency activity emerged from low global coupling, high firing rates, and a high proportion of γ versus α subpopulations. Importantly, ISP was necessary in the multi-frequency model to successfully fit EEG functional connectivity across frequency bands. Finally, ISP-controlled reductions in excitability reproduced both the slow-wave activity and the reduced connectivity in NREM sleep. Altogether, our model is compatible with biological evidence of the effects of E/I balance on modulating brain rhythms and connectivity, as observed in sleep, neurodegeneration, and chemical neuromodulation. This biophysical model with ISP provides a springboard for realistic brain simulations in health and disease. Author Summary Macroscale brain activity can be captured using techniques like EEG and fMRI. However, the granular or more detailed activity of neurons and neural masses is inaccessible. A solution is the use of whole-brain models, although they are not free from limitations, they can simulate EEG and fMRI recordings from mathematical equations and empirical data. One first limitation in these models is hyperexcitation. When the coupling between brain areas increases, brain areas might become aberrantly hyperexcitable if no compensatory mechanisms are considered. To address this, we introduce a mechanism in the model that dynamically modifies feedback inhibition to compensate for this excitability increase when running simulations. A second limitation is that many models fail to reproduce the spectral richness of EEG signals. EEG recordings reflect interweaving slower and faster rhythms, and some traditional models of EEG fail in capturing the spectral range of electrophysiological recordings. Here, we addressed this by combining two subpopulations of cortical columns within single brain areas, each one oscillating within the α and γ bands of EEG. Their combined activity generates EEG oscillations resembling the slower rhythms observed during sleep, and the faster ones triggered by increased attentional load. We ran different types of simulations and analyses to fully characterize our model. We observed that controlling system excitability is necessary to fully capture EEG connectivity and to simultaneously reproduce the EEG power spectrum and fMRI dynamics. Moreover, we showed that reduced/increased brain excitability is the cause of the emergence of the slowest/fastest EEG rhythms. The model can be used to characterize how connectivity and brain dynamics are altered in different types of conditions, such as chemical neuromodulation, drug delivery, altered states of consciousness, and neurodegenerative disorders. Our model is open access, well-documented, and introduced with tutorials, in the way to make it accessible to the whole neuroscience community.

Source: PLoS Computational Biology

DOI: 10.1371/journal.pcbi.1013463