Introduction

The clinical profile of motor deficit after ischemic stroke (IS) may vary greatly, which can not be fully interpreted by the lesion topology. Cortical reorganization and the functional remodeling involving the unaffected brain also shape the overall disease dynamic [1,2]. This study aims to explore the clinical phenotype polymorphism of IS based on the inter-network functional connectivity.

Methods

This study was approved by the local Institutional Review Board. Fifty three right-handed subjects with unilateral somatosensory impairment as a result of the first-ever single IS lesion (M/F = 38/15, aged 58.53±14.68 years) and 34 age-matched healthy controls were consecutively recruited (M/F= 16/18, aged 58.12±9.93 years). Resting-state fMRI data were acquired (1.5T, Siemens, Erlangen, Germany; 12-channel phased array head coil) with an echo-planar imaging (EPI) sequence and typical parameters of TR/TE 3000/30 ms, flip angle 90°, matrix 128×128, slice thickness 3mm, bandwidth 1395 Hz/pixel, 80 volumes. In addition, MPRAGE and DWI images were acquired to facilitate lesion identification and data preprocessing. IS lesion was manually determined on DWI images for each participant, the resulted binary lesion mask was then coregistered to MPRAGE image. fMRI data were preprocessed using DPARSF with standardized procedures [3]. Group independent component analysis (ICA) was performed using previously developed toolbox GIFTv3.0b (https://trendscenter.org/software/). ICA was conducted for healthy control (HC) and groups with lesions in the left and right hemisphere, respectively. One-sample t test was conducted on each identified component of all subjects to create a sample-specific component map and network mask (FDR corrected p < 0.05, cluster size > 10 voxels). The inter-network functional connectivity (FC) matrix was created by calculating the pair-wise Pearson’s correlation coefficient between the identified ICs and transformed to Fisher’s Z score. K-means clustering method was applied to estimate the inter-network connectivity patterns with a 100 times iteration to reduce the bias of initial random selection of cluster centroids.

Results

Six components were identified by spatial correlation analysis between the selected ICs and the reported template (4): anterior and posterior default mode network (aDMN/pDMN), left and right frontoparietal network (LFPN/RFPN), auditory network (AN), and visual network (VN). The inter-network FC matrices of IS subjects were optimally clustered into two patterns (cluster I/II = 23/30) with nine pairs of FC showing significant inter-cluster difference. The FC between DMN and LFPN or RFPN was positive for cluster I but negative for cluster II; FC between aDMN and pDMN was weak and negative for cluster I, but strong and positive for cluster II; FC between AN and LFPN or RFPN and the FC between of aDMN and VN were negative for cluster I but weakly positive for cluster II. Both patient clusters were characterized with weaker inter-network connectivity as compared with HCs (Figure A-F). In particular, FC between LFPN and RFPN of two patient clusters was significantly weaker (Two sample t-test, p<0.05, FDR correction). Connectivity between VN and LFPN or RFPN was strong and negative for HCs, but weak and positive for patient clusters. In addition, FCs within DMN, FC between aDMN--LFPN as well as FC between pDMN and VN in Cluster I were weaker and opposite in the correlation signs compared to HCs. Connectivities between FPN and pDMN or AN as well as the FC between aDMN-VN of Cluster II were weaker and opposite in the correlation signs compared to HCs. Cluster I was associated with a significantly higher frequency of right fronto-parietal involvement as compared with cluster II (37.0% vs. 3.8%, p<0.05). No other difference was identified between the patient clusters (stroke age, blood pressure, NIHSS score, status of atherosclerosis, Mann-Whitney U test, p>0.05).

Discussion and Conclusions

Divergent patterns of inter-network functional connectivity were identified in this cohort with unilateral motor dysfunction after ischemic stroke, indicating a polymorphism of the clinical phenotype of the disease that possibly underpinned by the altered interaction between DMN and LFPN as well as the audiovisual networks. In addition, involvement of the right frontoparietal cortices showed particular significance in the diversification of the clinical phenotype of motor impairment. Inter-network analysis provides a new insight of into better disease dynamic characterization and itemized management of ischemic stroke.

Acknowledgements

This study was partially supported by GJHZ20180928120207356 and NSFC81627901.