Attractor dynamics of a whole-cortex network model predicts emergence and structure of fMRI co-activation patterns in the mouse brain

Diego Fasoli, Ludovico Coletta, Daniel Gutierrez‐Barragan, Alessandro Gozzi, Stefano PanzeribioRxiv 2022 · DOI

TL;DR

This paper introduces a whole-cortex computational model of the mouse brain that successfully predicts the emergence and structure of fMRI co-activation patterns (CAPs) from underlying anatomical connectivity. The model reveals that these dynamic brain states, known as attractors, are crucial for understanding resting-state fMRI and are significantly shaped by the directed nature of neural connections.

What Problem Does This Paper Address?

Resting-state fMRI (rs-fMRI) reveals complex, dynamic patterns of brain activity that change on very short timescales (seconds). A prominent feature of this dynamic is the emergence of recurring "co-activation patterns" (CAPs), which are consistent spatial profiles of brain region activity. However, the underlying mechanisms generating these CAPs and how they arise from the brain's anatomical wiring have remained unclear. Specifically, it's debated whether CAPs are driven by subcortical structures or emerge from the intrinsic interactions within the cortex itself. This paper aims to bridge this gap by developing a computational model that links anatomical connectivity to the dynamic emergence of CAPs.

Methods

The authors developed a whole-cortex neural network model of the mouse brain. Key methodological aspects include:

  • Anatomical Connectivity: The model incorporates a directed, voxel-scale anatomical connectivity matrix derived from the Allen Mouse Brain Atlas. This directed nature of connections is a crucial feature, as it reflects the actual flow of information between brain regions.
  • Neural Dynamics: Each cortical region is modeled with both excitatory (E) and inhibitory (I) neural populations. These populations interact with non-linear activation functions, mimicking the spiking activity of neurons.
  • Neurovascular Coupling: The model translates neural activity into fMRI signals using a simplified neurovascular coupling function, specifically modeling fMRI as the total postsynaptic current to excitatory populations.
  • Attractor Dynamics: The core of the model's analysis lies in identifying "attractors" – stable or cyclical patterns of network activity that the system tends to converge to. The researchers developed algorithms to detect and map these attractors.
  • Parameter Fitting: Model parameters were fitted to match static properties of empirical mouse rs-fMRI data, specifically time-averaged activity and functional connectivity matrices.
  • Attractor Mapping: A novel algorithm was developed to map empirical fMRI time series onto the model's attractor landscape, allowing for a direct comparison between model predictions and real brain activity.

Datasets

  • Empirical Resting State fMRI Data: The study utilized rs-fMRI data from 15 adult male C57Bl6/J mice, which were previously published and made available. This dataset was used for both fitting model parameters and validating model predictions.
  • Mouse Anatomical Connectivity Data: This data was derived from the Allen Mouse Brain Institute's connectome project, providing a detailed map of directed connections between cortical regions.

Key Insights and Ideas

  • Attractors as the Basis of CAPs: The study strongly suggests that CAPs are not arbitrary fluctuations but rather reflect the underlying "attractor dynamics" of the cortical network. These attractors represent stable or recurring states the brain tends to occupy.
  • Directed Connectivity is Crucial: The directionality of anatomical connections is not just a detail; it's fundamental. Removing this directionality significantly impoverished the model's attractor landscape and its ability to predict empirical brain dynamics.
  • Inter-hemispheric Coupling Matters: The strength of connections between the left and right hemispheres plays a critical role in shaping attractor dynamics, influencing their number and the emergence of non-homotopic (asymmetric) patterns.
  • Non-Homotopic Attractors and Brain Dynamics: The model predicts that even with a generally symmetric anatomical structure, the brain can exhibit transient, asymmetric activity patterns (non-homotopic attractors) on short timescales, which are masked by averaging over longer periods. This suggests a richer dynamic than static connectivity might imply.
  • CAPs as Emergent Properties: The findings support the idea that CAPs are an emergent property of cortico-cortical interactions, rather than being solely driven by external modulatory inputs.

Key Results

  • The whole-cortex model successfully generated a rich set of stationary and oscillatory attractors, mirroring the dynamic nature of empirical fMRI data.
  • Model-predicted attractors recapitulated the topographical organization of empirical fMRI CAPs.
  • Neglecting fiber directionality severely reduced the number of attractors and their explanatory power for CAPs.
  • Modifying inter-hemispheric connectivity strength (both increasing and decreasing it from real values) led to fewer attractors and reduced non-homotopic features, highlighting the importance of the specific connectivity found in the mouse brain.
  • The model's attractor dynamics were compatible with empirical fMRI time series, as demonstrated by the attractor mapping algorithm.
  • The model accurately predicted functional connectivity patterns after global signal regression, including anti-correlations that are not apparent in static FC.

Limitations and Future Work

  • Model Simplifications: While the model captures key features, it uses simplified representations of neural populations (binary activation) and neurovascular coupling.
  • Parameter Fitting: Parameters related to local excitation-inhibition balance and inter-hemispheric coupling were determined by best fit rather than direct anatomical measurement, though the paper explores their impact.
  • Future Directions: The authors suggest that their framework can be used to make empirically testable predictions about how alterations in brain connectivity (e.g., due to injury) might affect frame-by-frame brain dynamics. They also propose investigating the interaction between attractor dynamics and neuromodulatory systems. Further work could explore how behavioral states might influence attractor dynamics and validate these findings in human brains.

Summary updated

Abstract

Abstract Resting state fMRI activity in mammals exhibits rich dynamics on a fast, frame-by-frame timescale of seconds, including the robust emergence of recurring fMRI co-activation patterns (CAPs). To understand how such dynamics emerges from the underlying anatomical cortico-cortical connectivity, we developed a whole-cortex model of resting-state fMRI activity in the mouse. Our model implemented neural input-output nonlinearities and excitatory-inhibitory interactions within cortical regions, as well as directed anatomical connectivity between regions inferred from the Allen mouse brain atlas. We found that, even if the model parameters were fitted to explain static properties of fMRI activity on the timescale of minutes, the model generated rich frame-by-frame attractor dynamics, with multiple stationary and oscillatory attractors. Guided by these theoretical predictions, we found that empirical mouse fMRI time series exhibited analogous signatures of attractor dynamics, and that model attractors recapitulated the topographical organization of empirical fMRI CAPs. The model established key relationships between attractor dynamics, CAPs and features of the directed cortico-cortical intra- and inter-hemispheric anatomical connectivity. Specifically, we found that neglecting fiber directionality severely affected the number of model’s attractors and their ability to explain CAPs. Furthermore, modifying inter-hemispheric anatomical connectivity strength by decreasing or increasing it from the value of real mouse anatomical data, resulted in fewer attractors generated by cortico-cortical interactions and reduced non-homotopic features of the attractors generated by the model, which were important for better predicting empirical CAPs. These results offer novel theoretical insight into the dynamic organization of resting state fMRI in the mouse brain and suggest that the frame-wise BOLD activity captured by CAPs reflects an emerging property of cortical dynamics resulting from directed cortico-cortical interactions. Author summary Whole-brain activity at rest transitions on a timescale of seconds between stereotyped co-activation patterns (CAPs), each with a distinct spatial profile of co-activation across different brain regions. CAPs have been robustly reported across datasets and mammalian species, including humans and mice. However, the significance and origin of these patterns remain unknown. Here we studied the origin of CAPs using a computational model of the whole cortex based on real-world directed measurements of mouse cortical anatomical connectivity. We found that we could explain the formation and topography of CAPs in terms of attractors (that is, states the network tends to converge to) that reflect the information in the anatomical connections between cortical regions. Attractors and CAPs are significantly influenced by the directionality of connectivity (available in mice but not in humans) and the strength of inter-hemispheric coupling. Together, our findings provide a novel possible explanation of the mechanisms of CAP generation which relates to cortico-cortical connectivity, and suggest that CAPs are at least in part an emergent property of cortico-cortical interactions which can be generated also in absence of the action of arousing modulatory nuclei.
BibTeX
@misc{fasoli2022attractor,
  author = {Diego Fasoli and Ludovico Coletta and Daniel Gutierrez‐Barragan and Alessandro Gozzi and Stefano Panzeri},
  title = {{Attractor dynamics of a whole-cortex network model predicts emergence and structure of fMRI co-activation patterns in the mouse brain}},
  year = {2022},
  doi = {10.1101/2022.04.28.489908},
  howpublished = {bioRxiv},
}
Figure