From May 2014 to December 2018, a total of 41 patients with clinically diagnosed RRMS (17 in acute phases and 24 in remitting phases) were recruited at the First Affiliated Hospital of Nanchang University, according to the 2010 revised McDonald’s criteria . Twenty-three appropriately matched (age, sex, and education) subjects served as healthy controls (HCs). The exclusion criteria for the subjects were the presence or history of traumatic brain injuries, tumour or stroke based on conventional MRI data. All subject have self-reported being right-handed.
All the patients underwent neuropsychological evaluations, including the Expanded Disability Status Scale (EDSS), Modified Fatigue Impact Scale (MFIS), and Paced Auditory Serial Addition Test (PASAT).
Image acquisition and preprocessing
All the subjects were imaged with a 3.0 T MRI scanner (Trio Tim; Siemens, Munich, Germany) using an eight-channel phased array head coil. The following sequences of the brain were acquired: 1) T2*-weighted gradient echo sequence (TR/TE = 2,000/30 ms, flip angle = 90°, FOV = 200 × 200 mm, matrix = 64 × 64, 30 interleaved axial slices with a 4 mm thickness and an interslice gap of 1.2 mm, number of time points = 240); 2) T2-weighted turbo spin-echo imaging; and 3) three-dimensional T1-weighted imaging. During the RS-fMRI scanning, the subjects were instructed simply to rest with their eyes closed, not to think systematically, and not to fall asleep.
All the preprocessing steps were carried out using the MATLAB 2012a platform (MathWorks, Inc., Natick, MA, USA). The standard preprocessing procedures have been described in our previous studies . The main steps of preprocessing included the first ten volumes of each session were discarded for the equilibrium state of the echo signal, slice correction, head realignment, spatial normalization to MNI space with high-resolution T1WI registration, resampling to 3 mm isotropic voxels, and 6 mm smoothing. Besides, subjects with head movement in the cardinal directions (x, y, z) >2 mm and a maximum rotation (x, y, z) >2° were excluded.
We performed a group spatial ICA on the preprocessed data of the patients with RRMS and normal controls using the Group ICA of fMRI Toolbox (GIFT, http://icatb.sourceforge.net/groupica.htm). We chose a relatively high model order ICA (number of components, C = 75), as previous studies have demonstrated that such models yield refined components  and a highly stable ICA decomposition . In the group ICA, the mean independent components of all the subjects, the corresponding mean time courses and the independent components for each subject were obtained from the group ICA separation and back reconstruction to ensure that all the subjects had the same components . After standard preprocessing of the group ICA results, from 75 components, we identified fourteen RSNs via a template-matching algorithm based on the maximum spatial correlation value. These functional templates were provided by Shirer et al.. Then, one-sample t-test of the group-wise spatial maps was performed by Data Processing Assistant for Resting-State fMRI Advanced Edition (version 2.2).
The GCA in a spectral method that is used to elucidate the causal relationships between two or more stationary variables for brain networks . Based on the principle of Granger causality, if incorporating the past values of time series X improves the future prediction of time series Y, then X is said to have a causal influence on Y . Granger causality analysis was accomplished using the functional network connectivity toolbox (http://icatb.sourceforge.net/). According to a given interval and an order selection criterion, the optimal order of the autoregressive model was selected, which was not constant and varied within the interval for every mutual relationship. We preferred the Schwartz Bayesian criterion to determine the optimal order of the autoregressive model and to obtain the smallest mean-squared prediction error of the fitted autoregressive model. In this study, the GCA was compared between the acute phase of RRMS and healthy controls and between the remitting phase of RRMS and healthy controls. The statistical significance level was set at a P-value less than 0.05 with false discovery rate (FDR) correction.
We performed statistical analyses of the demographic, clinical, and relationship data by using SPSS software (version 22, 2013; IBM, Chicago, Ill). We investigated Pearson’s correlation between abnormal Granger causality coefficients and neuropsychological characteristics in RRMS while controlling for age and sex. The relationship was significant if the P-value was below 0.05.