Neurophysiological modelling of network stimulation reveals distinct signatures across low- to higher-order networks

Poster Session C - Sunday, April 14, 2024, 5:00 – 7:00 pm EDT, Sheraton Hall ABC

Davide Momi¹, Zheng Wang¹, Sara Parmigiani³, Ezequiel Mikulan⁴, Sorenza Bastiaens^1,2, Parsa Oveisi^1,2, Kevin Kadak^1,2, Allison Waters⁵, Sean Hill¹, Andrea Pigorini⁴, Corey Keller³, John D Griffiths^1,2; ¹Centre for Addiction and Mental Health (CAMH), Toronto, ²University of Toronto, ³Stanford University, ⁴Università degli Studi di Milano, ⁵Icahn School of Medicine at Mount Sinai, New York

The human brain comprises several distinct ‘resting-state’ networks, which exhibit structured spontaneous activity patterns and are implicated in a range of cognitive functions. Prior research has revealed a hierarchical neurocognitive organization of these networks, following a continuous ‘principal cortical gradient’ (PCG; first eigenvector of rsfMRI functional connectivity) progressing from low-order unimodal sensory/motor regions, through to high-order multimodal regions. A key question for cognitive neuroscience is whether this topographical feature of brain organization influences how different brain regions respond to inputs and engage in information processing. This was investigated in the present study using intracranial neurostimulation in surgical patients, together with computational brain network modelling. We analyzed sEEG+hdEEG data from epilepsy patients undergoing intracortical single-pulse electrical stimulation. Using the Whole-Brain Modelling in PyTorch (WhoBPyT) Python library (github.com/griffithslab/whobpyt), we fit individualized neural models to these data. A ‘virtual lesion’ approach then evaluated which responses to stimulation in multi- vs. unimodal cortical areas are reliant on recurrent feedback connections. In both sEEG/hdEEG datasets, we found strongest global field power when high-order multimodal networks were stimulated than when low-order unimodal networks were, and that virtual lesions suppressed late responses significantly more for high-order than low-order regions. Interestingly, both of these effects followed a spatial trajectory and linear ordering, closely matching that of PCG. Our results suggest that cortical regions differ in their strategy for information processing and activity dynamics, with high-order multimodal cortex being more globally integrated and interdependent, and low-order unimodal cortex being relatively independent of external inputs from other brain regions.

Topic Area: METHODS: Neuroimaging