LM1-RM1 rs-connectivity

Preprocessing and Nuisance Regressors. Physiologic information (i.e., heartbeat and respiration) collected during the rs-fMRI scan was regressed from the rs-fMRI images using the RETROICOR algorithm⁴⁹ in AFNI. Subsequent preprocessing and data analyses were performed using FSL⁴⁸ version 5.0. Preprocessing of the rs-fMRI images was performed using the FMRI Expert Analysis Tool (FEAT). Preprocessing steps included head motion correction, slice time correction to the middle slice, spatial smoothing with a Gaussian kernel of 6 mm full width at half maximum, and high pass temporal filtering at 0.01 Hz. The Brain Extraction Tool (BET) in FSL was also used to remove non-brain tissue (e.g., skull).

Time series for nuisance regressors were extracted from the rs-fMRI data. Nuisance regressors included the time series for cerebrospinal fluid (CSF), white matter (WM), and six head motion parameters. To obtain the time series of CSF and WM, we used FMRIB’s Automated Segmentation Tool (FAST) to segment the T1-weighted image into grey matter, WM, and CSF. This enabled us to create masks of the CSF and WM, which were eroded twice to ensure that the only voxels included in these masks were voxels representing CSF and WM. After, we used FMRIB’s Linear Image Registration Tool (FLIRT) to transform the CSF and WM masks from structural to functional space. From the preprocessed rs-fMRI data we extracted the mean time series of CSF and WM from voxels within the CSF and WM masks, respectively. To obtain the time series of the six head motion parameters, we used FSL’s MCFLIRT. In another FEAT analysis, the mean time series for CSF, WM, and six head motion parameters were inputted in the general linear model (GLM) as regressors of no interest. The residual of this analysis was used to determine LM1-RM1 rs-connectivity described below in the subsection LM1-RM1 Resting State Connectivity.

Seed Masks. To calculate LM1-RM1 rs-connectivity, we created seed masks for LM1 and RM1. The first author of this study (TKL) created the seed masks independently from the neurologist (KH) who traced the stroke lesions. Given our interest in the motor outcome of both the arm and hand, the seed masks were derived using two peak coordinates from arm/elbow and hand/finger fMRI paradigms reported in a meta-analysis⁵⁰ [Fig. 2B]. For the LM1 seed, the average peak coordinate for arm/elbow representation was (−28,−24, 62) [Supplementary Table ^S2A, coordinates in MNI space] and the average peak coordinate for hand/finger representation was (−36,−20, 56) [Supplementary Table ^S2B, coordinates in MNI space]. We then applied a 6 mm radius around each coordinate and added the two seed masks together to create a single LM1 seed. To create the RM1 seed mask, we flipped the peak coordinates used for the LM1 seed along the mid-sagittal plane such that the peak coordinates were on the right side of the MNI brain template [arm/elbow = (28, −24, 62); hand/finger = (36,−20, 56)]. The LM1 and RM1 seed masks were then transformed from standard MNI space to functional space using FNIRT.

LM1-RM1 Resting State Connectivity. Mean time series of LM1 and RM1 seeds were extracted from the residual rs-fMRI data, and used to compute the Pearson’s correlation (r-value) [Fig. 2B]. Two participants (s01 and s08) had partial overlap between their lesion and RM1 seed. For s01, 34% of the RM1 seed mask overlapped their lesion, whereas for s08, 10% of the RM1 seed mask overlapped their lesion. We included these two participants in the statistical analyses since the perilesional tissue in the right M1 region appears mostly intact for these participants. To yield data approximating a standard normal distribution, we transformed each participant’s LM1-RM1 rs-connectivity r-value into a standard z-score. These z-scores were inputted in the hierarchical regression analyses described in Statistical Analyses.