Metabolic brain covariant networks as revealed by FDG-PET with reference to resting-state fMRI networks

Abstract

The human brain is inherently organized as separate networks, as has been widely revealed by resting-state functional magnetic resonance imaging (fMRI). Although the large-scale functional connectivity can be partially explained by the underlying white-matter structural connectivity, the question of whether the underlying functional connectivity is related to brain metabolic factors is still largely unanswered. The present study investigated the presence of metabolic covariant networks across subjects using a set of fluorodeoxyglucose ((18)F, FDG) positron-emission tomography (PET) images. Spatial-independent component analysis was performed on the subject series of FDG-PET images. A number of networks that were mainly homotopic regions could be identified, including visual, auditory, motor, cerebellar, and subcortical networks. However, the anterior-posterior networks such as the default-mode and left frontoparietal networks could not be observed. Region-of-interest-based correlation analysis confirmed that the intersubject metabolic covariances within the default-mode and left frontoparietal networks were reduced as compared with corresponding time-series correlations using resting-state fMRI from an independent sample. In contrast, homotopic intersubject metabolic covariances observed using PET were comparable to the corresponding fMRI resting-state time-series correlations. The current study provides preliminary illustration, suggesting that the human brain metabolism pertains to organized covariance patterns that might partially reflect functional connectivity as revealed by resting-state blood oxygen level dependent (BOLD). The discrepancy between the PET covariance and BOLD functional connectivity might reflect the differences of energy consumption coupling and ongoing neural synchronization within these brain networks.

FIG. 1.

Metabolic network maps as revealed by spatial-independent component analysis (ICA) on the PET data. (A–J) display the 10 IC maps classified as meaningful network when 20 ICs were extracted. (K–O) illustrates additional 5 IC maps when 40 ICs were extracted. Each individual IC map was z-transformed and thresholded at z>1.96.

FIG. 2.

Resting-state network maps (A–I) as revealed by spatial ICA on the blood oxygen level-dependent (BOLD) data. Twenty components were extracted. Each individual IC map was z-transformed and thresholded at z>1.96.

FIG. 3.

Correlation matrices of 14 region of interests for the PET (A) and functional magnetic resonance imaging (fMRI) (B) datasets. (A) Intersubject correlation matrix of PET metabolism. (B) Mean resting-state BOLD time-series correlation matrix across subjects.

FIG. 4.

Correlations between two main nodes within each network as revealed by both resting-state BOLD data and PET data. For the BOLD data, the bars and error bars represent group mean and standard deviation of Fisher's z-scores corresponding to correlations of BOLD time series. For the PET data, the bars represent the Fisher's z-scores of intersubject PET correlations. *Statistically significant difference at p<0.05.