Selectively disrupted sensorimotor circuits in chronic stroke with hand dysfunction

Abstract Aim To investigate the directional and selective disconnection of the sensorimotor cortex (SMC) subregions in chronic stroke patients with hand dysfunction. Methods We mapped the resting‐state fMRI effective connectivity (EC) patterns for seven SMC subregions in each hemisphere of 65 chronic stroke patients and 40 healthy participants and correlated these patterns with paretic hand performance. Results Compared with controls, patients demonstrated disrupted EC in the ipsilesional primary motor cortex_4p, ipsilesional primary somatosensory cortex_2 (PSC_2), and contralesional PSC_3a. Moreover, we found that EC values of the contralesional PSC_1 to contralesional precuneus, the ipsilesional inferior temporal gyrus to ipsilesional PSC_1, and the ipsilesional PSC_1 to contralesional postcentral gyrus were correlated with paretic hand performance across all patients. We further divided patients into partially (PPH) and completely (CPH) paretic hand subgroups. Compared with CPH patients, PPH patients demonstrated decreased EC in the ipsilesional premotor_6 and ipsilesional PSC_1. Interestingly, we found that paretic hand performance was positively correlated with seven sensorimotor circuits in PPH patients, while it was negatively correlated with five sensorimotor circuits in CPH patients. Conclusion SMC neurocircuitry was selectively disrupted after chronic stroke and associated with diverse hand outcomes, which deepens the understanding of SMC reorganization.


| INTRODUC TI ON
Stroke remains the primary reason for adult disability, 1 and hand function recovery is vital for survivors to regain functional independence. 2 Motor outcomes following stroke have been found to be associated with infarction size, 3 lesion topography, 4,5 gray matter plasticity, 6 corticospinal tract integrity, 7 functional network connectivity, 8,9 and frequency-specific local oscillations. 10 Benefiting from these neuroimaging discoveries in stroke populations, recent studies have suggested that modulating key sensorimotor nodes by noninvasive brain stimulation [11][12][13][14][15] could promote motor recovery after stroke. Hence, neuroimaging opens the door for understanding the pathophysiology of motor deficits following stroke and may inspire progress in personalized, neurobiologically informed neuromodulation. 16,17 Resting-state functional magnetic resonance imaging (fMRI) can non-invasively explore intrinsic human brain activity, 18 and connectivity analysis provides an effective framework for understanding information interactions among brain regions after stroke. 19 In well-recovered stroke patients, although activation patterns are close to those in healthy controls, the network connectivity is aberrant. 20 Plastic changes in functional connectivity throughout the sensorimotor regions have been demonstrated to be associated with upper extremity dysfunction and recovery following stroke. 16,21 In cross-sectional studies, disrupted interhemispheric sensorimotor connectivity was positively correlated with motor dysfunction, 22 while it was inversely affected by the lesion load of the corticospinal tract. 23,24 In longitudinal studies, connectivity between the ipsilesional primary motor cortex (PMC) and contralesional supplementary motor area (SMA) in the early days could predict long-term motor outcomes after stroke, 25 and dynamic connectivity changes in the cerebrocerebellar circuits were accompanied by spontaneous recovery in stroke patients with pontine 26 or subcortical 27 infarcts.
In neurorehabilitation studies, regulating the bilateral PMC through bihemispheric transcranial direct current stimulation, 15 priming the ipsilesional PMC through intermittent theta burst stimulation, 11 and inhibiting the contralesional PMC through repetitive transcranial magnetic stimulation 13 could facilitate motor recovery following stroke. However, the direction of functional interactions between the sensorimotor cortex (SMC) and whole brain following stroke is less clear.
Effective connectivity (EC) is mostly employed for task-evoked fMRI data, including dynamic causal modeling, psychophysiological interactions, structural equation modeling, and Granger causality analysis. 28 In contrast to the non-directional characteristic of functional connectivity, EC analysis can delineate the causal influences among brain regions. Using dynamic causal modeling, 13,[29][30][31] several milestone studies have investigated EC among key sensorimotor areas following stroke. Rehme et al. reported that the interhemispheric coupling between the bilateral PMC was associated with illness duration 29 and the severity of deficits. 30 Grefkes  Another study investigating resting-state EC among the frontoparietal area and sensorimotor system found that stroke patients showed decreased influences from the superior parietal lobule to both the PMC and SMA in the lesioned hemisphere. 32 However, as a data-driven exploratory method, Granger causality analysis has rarely been used in stroke studies. The SMC involves a wide spectrum of integrated motor functions and can be divided into seven subregions in each hemisphere. 33 Given the different functions of SMC subregions and their associations with motor deficits after stroke, we investigated whether the resting-state EC patterns of the SMC subregions suffer selective disruption in chronic subcortical stroke patients with hand dysfunction.
Here, we first defined the seven SMC subregions on the basis of the probabilistic cytoarchitectonic atlases for each hemisphere and then calculated the whole-brain resting-state fMRI EC patterns for each SMC subregion in each participant. Next, we examined EC differences between all stroke patients and healthy participants and between stroke subgroups with different hand outcomes. Finally, brain-behavior correlations between EC patterns and hand performance were also explored. and (e) dextromanuality as evaluated by the Edinburgh Handedness K E Y W O R D S effective connectivity, functional reorganization, Granger causality analysis, resting-state functional magnetic resonance imaging, stroke Inventory. We excluded those patients from this study who had MRI contraindications, severe cognitive impairment/aphasia/neglect, and unstable illness states (eg, serious atrial fibrillation and multiple organ failure). Healthy participants who had no neuropsychiatric history or cognitive impairments were recruited from the local community.

| Recruitment of participants
As per previous studies, 8,10 we used the Paretic Hand Scale (see Supporting Materials) to divide stroke patients into the partially (PPH) and completely (CPH) paretic hand subgroups. This scale was specifically designed to evaluate the practical function of the hand in everyday life. Stroke patients who could finish one or more tasks were categorized as having PPH, while those could not finish any task were classified as having CPH.

| Behavioral assessment
The Hand and Wrist subscale of the Fugl-Meyer Assessment (FMA-HW) was used to evaluate paretic hand performance in all stroke patients before fMRI scanning. 8 The FMA-HW subscale, which was regarded as the primary measurement, consists of a wrist section (five items) and a hand section (seven items), with a possible score ranging from 0 to 24.

| Collection of imaging data
Imaging data were acquired using a 3-Tesla scanner (SIEMENS Trio, Germany). T1-weighted images were collected using an MPRAGE

| Mapping lesion overlap
We first used MRIcron software (https://people.cas.sc.edu/rorde n/mricr on/insta ll.html) to delineate the lesion profiles of each stroke patient on T2-weighted images (see Supporting Materials).
Then, the T2-weighted lesion masks of all stroke patients were standardized to the MNI space. Finally, we summed each resampled lesion mask with a resolution of 1 × 1 × 1 mm 3 to establish the lesion map ( Figure 1).

| Imaging data preprocessing
We employed DPABI software (http://rfmri.org/DPABI) to preprocess the resting-state functional imaging data. 34 The processing steps involved (a) deletion of the first ten volumes, (b) correction of slice timing, (c) realignment of head motion, (d) standardization to the MNI space using the unified segmentation of structural images, (e) spatial smoothing (FWHM = 6 mm), (f) linear detrending, and (g) bandpass filtering (0.01-0.1 Hz). Finally, we regressed out the six head motion parameters, global mean signal, cerebrospinal fluid signal, and white matter signal. During image preprocessing, no participants were discarded based on the predefined criteria of head motion (exceeding 2 mm/degree). To eliminate the influences of head motion on the EC results, 35 we regressed out the framewise displacement in all subsequent between-group statistical analyses.

| Definition of the SMC subregions
We defined seven SMC subregions in the ipsilesional and contralesional hemispheres on the basis of the probabilistic cytoarchitectonic F I G U R E 1 Lesion overlap map for all stroke, PPH, and CPH patients. The color bar indicates the frequency of patients having lesions in each voxel in the left (ipsilesional) hemisphere. CPH, completely paretic hand; PPH, partially paretic hand atlas, as integrated in SPM12 software (https://www.fil.ion.ucl. ac.uk/spm/softw are/spm12/). 33 For each hemisphere, the seven SMC subregions included premotor_6, PMC_4a, PMC_4p, primary somatosensory cortex_1 (PSC_1), PSC_2, PSC_3a, and PSC_3b ( Figure 2).

| EC analysis of the SMC subregions
The Granger causality analysis module within REST software (http:// restf mri.net/forum/ REST_V1.8) was used to generate voxelwise seed-based EC maps. 36 We employed the vector autoregression model to perform the Granger causality analysis. First, we performed a bivariate coefficient Granger causality analysis to obtain the EC maps for all participants and then converted these maps to z-values using Fisher's conversion. Next, we employed one-sample t-tests to obtain the group EC patterns of each of the seven SMC subregions in each hemisphere.

| Statistical analysis
To analyze the baseline data of all participants, SPSS software (version 25.0, IBM Inc.) was used. We first performed the Shapiro-Wilk test to assess the normality of all continuous variables (age, duration of illness, lesion volumes, FMA-HW score, and framewise displacement). Then, we found that only age was distributed normally and thus was analyzed by the two independent samples t-test, while the other variables were distributed abnormally and were analyzed by the Mann-Whitney test. For the sex ratio and stroke type, we employed the chi-square test to analyze the differences between groups. To infer the EC differences between groups, we used twosample t-tests to explore disrupted EC between all stroke patients and healthy participants as well as between CPH patients and PPH patients with sex, age and framewise displacement as covariates. The AlphaSim method with a p < 0.0001 was adopted to perform multiple comparisons (voxel p = 0.001, cluster ≥ 43, FWHM = 7.9 mm, with a gray matter mask). Finally, we used multiple linear regression analysis within SPM12 software (https://www.fil.ion.ucl.ac.uk/spm/ softw are/spm12/) to explore associations between the EC patterns of each subregion and the FMA-HW scores in all stroke patients with sex, age, and framewise displacement as covariates. We also performed the same analysis in both the PPH and CPH subgroups. The AlphaSim method with a p < 0.01 was adopted to perform multiple

| Basic demographic and clinical data
Sixty-five patients with left subcortical chronic stroke (32 PPH vs. 33 CPH) and forty healthy controls were recruited. We found that there were significant differences in the sex ratio (p = 0.009) and no significant differences in age (p = 0.671) or framewise displacement (p = 0.089) between stroke patients and healthy controls.
Furthermore, we found that there were no significant differences in age (p = 0.811), sex ratio (p = 0.110), stroke type (p = 0.540), duration of illness (p = 0.305) or framewise displacement (p = 0.823) F I G U R E 2 In total, fourteen seeds were defined by sensorimotor subregions based on the cytoarchitectonic atlas. The left panel shows seven ipsilesional (left) hemisphere seeds, and the right panel shows seven contralesional (right) hemisphere seeds between the PPH and CPH subgroups. However, the lesion volumes of the CPH patients were significantly larger than those of the PPH patients (p = 0.006), and the FMA-HW scores of the PPH patients were significantly higher than those of the CPH patients (p < 0.001) (  Figure 3).

| Distinct EC patterns between stroke patients with CPH and PPH
Compared with the CPH patients, the stroke patients with PPH demonstrated decreased EC from the ipsilesional inferior parietal lobe to the ipsilesional premotor_6 and ipsilesional PSC_1 and from the ipsilesional inferior temporal gyrus and contralesional middle frontal cortex to the ipsilesional PSC_1 (Table 2, Figure 5).

TA B L E 1 Demographic and clinical data of participants recruited in this study
Note: Values expressed as the mean ± SD; the superscript a indicates the chi-square test, b indicates two independent sample t-test, and all others are Mann-Whitney tests.
The bold value used to highlight the significant p values.

| DISCUSS ION
The challenge in post-stroke neuroimaging studies is to identify the intervention targets 38 and predict the long-term outcomes. 39 Using Granger causality analysis of resting-state fMRI data in chronic stroke patients, we found that the EC patterns of SMC subregions were selectively disrupted and correlated with hand dysfunction. Most importantly, the correlations between EC patterns and hand performance in stroke patients with PPH were positive, while those in stroke patients with CPH were negative.
These findings indicate that injury of the subcortical motor pathway results in specific disruption of sensorimotor circuits, which may inspire the development of neuroimaging biomarkers 40 and stimulation targets 41 in neurorehabilitation practice after chronic stroke.

| Disrupted sensorimotor circuits following chronic stroke
Usually, stroke patients with severe motor impairments exhibit hyperactivation of the sensorimotor system. [42][43][44] In fact, activation in the ipsilesional premotor and primary motor areas 45 without recruitment of contralesional activity 46 is related to good motor outcomes.
However, chronic stroke patients who receive bilateral arm training demonstrate a recovery-associated increase in activation in the highlighting the importance of interhemispheric disturbances among sensorimotor regions for hand dysfunction observed in chronic stroke patients. 9,22 Furthermore, these data suggest a functional relevance of the disrupted influences from the ipsilesional to contralesional somatosensory cortices due to the strong brain-behavior correlations.
Stroke patients with motor deficits typically show recruitment of non-motor regions (eg, the prefrontal, parietal, and temporal lobes), with the consensus that greater activation of non-motor areas leads to poorer functional recovery. 50 The anterior precuneus is closely related to sensorimotor processing. 51 One longitudinal study indicated that hyperactivation in the precuneus is correlated with slower motor recovery following stroke. 52 Here, we demonstrated that not only task-related activation of the precuneus but also EC strength from the bilateral precuneus to the ipsilesional PSC_2 is increased in chronic stroke patients. Furthermore, EC strength from the contralesional PSC_1 to the contralesional precuneus is negatively correlated with paretic hand performance.
The precuneus has extensive connections with the SMC system, which plays important roles in visual goal-directed hand movements. 53 Thus, our data suggest that the exchange of information between the somatosensory cortex and precuneus is needed for chronic stroke patients to support visual processing during affected hand movements. Complex motor tasks, for example, novel and skilled sequential hand movements, often require audiomotor processing support from the bilateral temporal gyrus. 54

| Differently disrupted sensorimotor circuits between PPH and CPH patients
It is well known that the premotor cortex gives rise to the corticoreticulospinal tract and is specifically related to proximal movement, 58 which is a good substitution for hand function recovery after mild to moderate stroke. 50 The premotor cortex involves transferring sensory stimuli into motor programs, 59

| Dissociated sensorimotor circuits correlated with different hand outcomes
Two different functional reorganizations have been reported after rehabilitation in chronic stroke patients. 67 Specifically, patients with intact or damaged PMCs and their descending motor pathway show decreased or increased activation in the ipsilesional SMC. Interestingly, we also found two brain-behavior correlations in which paretic hand performance is positively correlated with EC patterns in stroke patients with PPH but negatively correlated with EC patterns in CPH patients. In stroke patients with PPH, convergence of positive correlations to the contralesional PSC_1 indicates that this region holds a highly important function within the sensorimotor network configurations (eg, brain hub) to drive motor recovery. 68 However, in stroke patients with CPH, distributed negative correlations from non-motor regions (eg, the visual cortex) may represent a compensatory cognitive strategy, for example, visuospatial processing, to sustain poor hand outcomes. Except for physiological processes (eg, decreased GABAergic inhibition and increased NMDA facilitation), 66 excessive activation within the SMC system during paretic hand movement is primarily determined by the structural integrity of the corticospinal tract in stroke patients. 42,43 Therefore, we speculate that the different injury loads of the corticospinal tract between CPH and PPH patients may be the important reason for these dissociated positive and negative correlations. 69

| Limitations and future considerations
First, stroke patients show a heavy male predominance because endogenous estrogen can exert neuroprotective effects for premenopausal women away from a higher risk for stroke. 70 To address this bias, we regressed out sex in the statistical analyses. Second, this was not a longitudinal/interventional study, making it difficult to infer the dynamic evolution of EC patterns in SMC subregions during the process of motor recovery. Third, considering the enormous values of injured corticospinal tracts for predicting motor recovery in stroke patients, 7,71 it will be promising to combine structural and functional biomarkers for the prediction of treatment responses.
Finally, although modulating bilateral PMC targets has been shown to be beneficial for stroke patients with hand dysfunction, 11,[13][14][15]72 future studies might consider additional targets found in this study (eg, the postcentral gyrus) to design neuromodulation experiments. 41,73

| CON CLUS ION
In this study, we systematically investigated EC patterns between sensorimotor subregions and the whole brain after chronic stroke.
We found that large-scale sensorimotor circuits are selectively disrupted and that dissociated motor-related neurocircuitry is associated with different hand outcomes in chronic stroke patients, which has rarely been reported in previous studies. Our findings indicate that these disrupted sensorimotor circuits might be considered potential neuroimaging biomarkers and stimulation targets to repair lesion-induced abnormal motor networks, 74 which in turn, facilitate hand rehabilitation after chronic stroke.