Brain's Energy Budget: Integrating Information Costs and Metabolic Demand Along the Brain Connectome

1925 

Abstract Submission 

Mahnaz Ashrafi Varnosfaderani¹, Gabriel Castrillón², Valentin Riedl³

¹Technical University of Munich, Munich, Germany, ²Friedrich-Alexander University, Erlangen, Germany, ³Technical University of Munich, Erlangen, Germany

Mahnaz Ashrafi Varnosfaderani  
Technical University of Munich
Munich, Germany

Gabriel Castrillón  
Friedrich-Alexander University
Erlangen, Germany

Valentin Riedl  
Technical University of Munich
Erlangen, Germany

The intricate human brain, a dynamic network of connections, dedicates a substantial portion of its energy budget, estimated at up to 90%, to neural signaling processes, especially synaptic transmission and action potentials. This energy-intensive process results in varying energy demands among regions, particularly those connected to highly active neighbors requiring increased signal integration. Traditional neuroimaging techniques, focused on structural and functional network topology, fall short in explaining energy variability across brain regions. Two regions with seemingly identical network topologies can have vastly different energy profiles, underscoring the need for a paradigm shift in brain network analysis. Our research proposes a novel perspective considering the topology and metabolic annotation of connected regions, defining a given region's 'activity importance' within the broader network. By focusing on the energy expended by each target region in integrating information from its connected 'source' regions, we aim to illuminate the complex interplay between connectivity and energy consumption in the brain, offering insights into how neural networks manage limited energy resources for information processing.

Main data: simultaneous FDG-PET/fMRI of twenty healthy right-handed subjects evenly split between genders, with an average age of 34.45 ± 5.06 years. Data was acquired by a 3T PET/MR scanner. Preprocessing was done by using a standard pipeline of C-PAC.
We propose a novel regional information cost (IC) model across the human brain cortex. We hypothesize that regions connected to more influential regions with higher energy consumption demand greater energy costs for information integration. Each region termed the target node, has an IC value based on regional energy consumption and the impact coefficient of its connected nodes called sources. The impact coefficient calculated by the mutual information (MI) network signifies the influence of each source on its respective target. IC for each target (t) is computed by a weighted summation of energy metabolic values of sources (Es) with respective impact coefficients (α(t,s))(Fig 1. a). Energy metabolic values (CMRglc) and MI network acquired from simultaneous fMRI/FDG-PET (Fig.1.b, c).

IC map across the cortex was identified on the group level, with peak values in the angular and frontal cortex (Fig. 1d). It is important to note that the CMRglc of each target is not included in its IC. We found a strong statistically significant correlation between the target's CMRglc and IC values at the group (group: r = 0.64, p-value < 1e-10) and individual levels (Fig. 1e, f). This finding indicates that 40% of the variance in the region's CMRglc values can be attributed to external parameters related to sources (Es and α(t,s)). To reinforce the robustness of our findings, we replicated the analysis using two other datasets (TUM.closed-eyes and Wien). We found a significant correlation between IC and CMRglc for targets using group data (scatter plot in Fig.1 g, h) and individual data (box plots in Fig.1 g, h) for two datasets. We did a network analysis and found the IC for each network. Higher cognitive networks such as DMN have higher IC and a stronger correlation with CMRglc.

We employ diverse datasets to demonstrate a highly significant correlation between the target's IC map and its actual CMRglc. This suggests that the IC is a reliable CMRglc map estimate. This model indicates that regions connected to active regions are exposed to higher information rates and demand higher energy. Our network analysis also approves this result, which represents that cognitive regions such as DMN integrating information have higher IC value.

fMRI Connectivity and Network Modeling

Methods Development ¹

Multivariate Approaches

PET Modeling and Analysis ²

Cerebral Metabolism and Hemodynamics

Cortex

Positron Emission Tomography (PET)

Statistical Methods