Basic Modes of Spontaneous Activity Capture Spatiotemporal Patterns of the Global Signal at Rest

1791 

Abstract Submission 

Xi Chen¹, Xuhong Liao^1,2

¹School of Systems Science, Beijing Normal University, Beijing, China, ²Beijing Key Lab. of Brain Imaging and Connectomics, Beijing Normal University, Beijing, China

Xi Chen  
School of Systems Science, Beijing Normal University
Beijing, China

Xuhong Liao  
School of Systems Science, Beijing Normal University|Beijing Key Lab. of Brain Imaging and Connectomics, Beijing Normal University
Beijing, China|Beijing, China

Recent neuroimaging connectomics has revealed intrinsic human brain functional networks with non-trivial topological properties based on the resting-state functional MRI (R-fMRI) [1]. Of note, connectivity patterns of the functional networks are affected by the data preprocessing and network analysis strategies, among which global signal regression is a longstanding controversial issue [2]. One main concern is that global signal regression better reduces the influence of head motion [3] but induces anti-correlations in networks [4]. However, the spatiotemporal pattern of the global signal itself and how it affects functional connectivity (FC) patterns require further elucidation. Here, we applied a novel eigen-microstate analysis [5,6] on the R-fMRI data from healthy young adults to identify typical activity patterns of the global signal and reveal how the global signal contributes to the functional network organization.

We selected the minimally preprocessed R-fMRI data from 700 healthy young adults (age range: 21-35 years, M/F: 304/396) in the Human Connectome Project (HCP) [7]. Additional preprocessing steps included removing the first 10-second volumes, linear detrending, nuisance regression, and temporal filtering (0.01-0.08Hz). To assess the potential influence of the global signal, we used two nuisance regression strategies, one with global signal regression (GSR) and the other not (NGSR). After preprocessing, we extracted time courses of 1000 cortical nodes [8] for each participant for the subsequent analysis.
First, we constructed group-level FC matrices in GSR and NGSR cases and compared their FC strength distributions and spatial patterns. Second, we performed the eigen-microstate analysis [5,6] on regional time courses to extract basic activity modes in both cases. The basic modes with high ranking are regarded to dominantly contribute to the spatiotemporal variance of spontaneous activity over time. Next, we examined the spatial correspondence of the basic modes between two cases and assumed that the modes specific to the NGSR case capture the global signal influence. We further reconstructed the FC matrix in the GSR case by including basic modes in the NGSR case but excluding the basic modes specific to the NGSR case. Finally, we examined the spatial patterns of the global signal-specific modes by comparing them with that of the principal functional gradient [9]. We further examined the spatial association between the coactivation patterns of this mode and the differences in FC patterns between the two cases.

The FC matrices in the GSR and NGSR cases showed high spatial similarity (Fig. 1A, r=0.87, p<0.001) but differed in the FC strength (Fig. 1B). Specifically, the GSR strategy induced a leftward shift of the strength distribution (Fig. 1C).
We extracted the first ten basic activity modes in two cases and found they showed one-to-one spatial correspondence, except for an additional mode (i.e., first basic mode) identified in the NGSR case (Fig. 2A). The FC matrix reconstructed from the first ten basic modes in NGSR case, but excluding the first mode, was similar with the original FC matrix with GSR in terms of FC patterns (r=0.81, p<0.001) and strength distributions (Fig. 2B). The global signal-specific first mode showed the same signs of fluctuation amplitudes across regions and exhibited high spatial correlations with the principal functional gradient map (Fig. 2C, r=-0.83, pperm<0.001, spatial autocorrelation corrected). Moreover, this basic mode corresponded to positive coactivations between regions, which was spatially similar to the differences in FC patterns between the two cases (Fig. 2D, r=0.70, p<0.001).

We identified a global signal-specific basic activity mode, which follows a hierarchical organization and reflects inter-regional positive correlations. We speculate that the GSR strategy will remove the contribution of this mode and thus aggravates anti-correlations in the functional networks.

fMRI Connectivity and Network Modeling ¹

Motion Correction and Preprocessing

BOLD fMRI ²

FUNCTIONAL MRI

Other - Functional connectivity; Global signal; Spontaneous activity