The large‐scale structural connectome of task‐specific focal dystonia

Abstract The emerging view of dystonia is that of a large‐scale functional network disorder, in which the communication is disrupted between sensorimotor cortical areas, basal ganglia, thalamus, and cerebellum. The structural underpinnings of functional alterations in dystonia are, however, poorly understood. Notably, it is unclear whether structural changes form a larger‐scale dystonic network or rather remain focal to isolated brain regions, merely underlying their functional abnormalities. Using diffusion‐weighted imaging and graph theoretical analysis, we examined inter‐regional white matter connectivity of the whole‐brain structural network in two different forms of task‐specific focal dystonia, writer's cramp and laryngeal dystonia, compared to healthy individuals. We show that, in addition to profoundly altered functional network in focal dystonia, its structural connectome is characterized by large‐scale aberrations due to abnormal transfer of prefrontal and parietal nodes between neural communities and the reorganization of normal hub architecture, commonly involving the insula and superior frontal gyrus in patients compared to controls. Other prominent common changes involved the basal ganglia, parietal and cingulate cortical regions, whereas premotor and occipital abnormalities distinctly characterized the two forms of dystonia. We propose a revised pathophysiological model of focal dystonia as a disorder of both functional and structural connectomes, where dystonia form‐specific abnormalities underlie the divergent mechanisms in the development of distinct clinical symptomatology. These findings may guide the development of novel therapeutic strategies directed at targeted neuromodulation of pathophysiological brain regions for the restoration of their structural and functional connectivity.

Task specificity in dystonia is a clinically well-documented phenomenon, with symptoms often causing long-term psychological stress, psychiatric comorbidities, social isolation, and professional disability.
Although the exact pathophysiology of focal dystonia, including TSFDs, is unclear, it is currently being viewed as a disorder of the large-scale functional connectome (e.g., Battistella, Termsarasab, Ramdhani, Fuertinger, & Simonyan, 2017;Conte et al., 2019;Fuertinger & Simonyan, 2018;Neychev, Gross, Lehericy, Hess, & Jinnah, 2011;Schirinzi, Sciamanna, Mercuri, & Pisani, 2018;Simonyan, 2018;Zoons, Booij, Nederveen, Dijk, & Tijssen, 2011). The common, unifying features of the functional connectomes in different forms of focal dystonia include disorganization of the basal ganglia-thalamo-cortical community, abnormal distribution of influential regions of information transfer (hubs) in sensorimotor regions and thalamus, and reduced connectivity within the sensorimotor and frontoparietal regions. Moreover, a greater extent of functional alterations involving sensorimotor and executive cortical regions vs. subcortical structures are distinct features of TSFDs, such as LD and WC, compared to non-task-specific dystonias, such as cervical dystonia and blepharospasm . Different forms of TSFDs are further characterized by abnormalities in the functional specialization of network hubs that are responsible for the various levels of sensorimotor and executive control during production of affected motor behaviors (Fuertinger & Simonyan, 2018).
Despite these advances in identifying functional network properties in TSFDs, their structural underpinnings are less well understood.
Microstructural alterations have been reported as gray matter volumetric and cortical thickness changes in the basal ganglia, thalamus, cerebellum, sensorimotor and parietal cortex, as well as white matter aberrations along the cortico-striato-pallido-thalamic and cerebellothalamo-cortical pathways (e.g., Bianchi et al., 2017;Bianchi, Fuertinger, Huddleston, Frucht, & Simonyan, 2019;Delmaire et al., 2007;Delmaire et al., 2009;Garraux et al., 2004;Granert et al., 2011;Ramdhani et al., 2014;Simonyan et al., 2008;Simonyan & Ludlow, 2012). Moreover, significant relationships have been established between abnormalities in brain activation and gray matter structural organization of the primary somatosensory, superior temporal, inferior frontal cortical regions, and cerebellum (Simonyan & Ludlow, 2012), pointing to multi-domain structure-functional interactions underlying the TSFD pathophysiology. However, it remains unknown whether these structural changes represent abnormal nodes of the large-scale structural dystonic network or rather have only local impact by underlying regional functional abnormalities. If former, it is critical to establish how structural network alterations are further specialized to contribute to distinct clinical symptomatology of different forms of TSFD.
In this study, we used diffusion-weighted imaging (DWI) and graph theoretical analysis to identify the overall architecture of the large-scale structural network and determine its common and distinct alterations in two clinically different forms of TSFD -WC and LD, compared to healthy individuals. We hypothesized that the large-scale structural connectome is altered in focal dystonia, and its abnormalities closely follow those of the functional connectome. In particular, we postulated that structural network abnormalities commonly involve subcortical structures, such as the basal ganglia and cerebellum, in both forms of TSFD, whereas distinct nodal changes within sensorimotor and executive cortical regions represent specialized abnormalities of the fine motor control that is impaired in each TSFD form.

| Study subjects
A total of 48 subjects participated in the study, including 17 patients with LD (10 females/7 males, mean age 56.6 ± 13.2 years), 15 patients with WC (9 females/6 males, mean age 53.7 ± 11.7 years), and 16 healthy controls (10 females/6 males, mean age 53.4 ± 12.2 years) ( There were no significant differences (p ≥ .21) in age, sex, dystonia duration, onset, or severity between the groups based on two-sample t tests or Fisher's exact tests, as appropriate (Table 1).
All participants gave their informed written consent before study participation, which was approved by the Internal Review Board of Partners HealthCare Research Program. Data from some participants were used in our previous studies, which examined regional alterations in gray matter organization, white matter integrity, and resting-state functional connectivity (Bianchi et al., 2019;Fuertinger & Simonyan, 2018).

| Image acquisition
A whole-brain high-resolution MRI was performed on a 3.0 T Siemens Skyra scanner with a 32-channel head coil. Uniform T1-weighted images were acquired using a 3D magnetization prepared rapid acqui- (1) The uniform T1-weighted image was skull-stripped by segmenting gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF), multiplying the resulting whole-brain mask to the image; (2) the anatomical scan was transformed to match the AFNI standard reference template in Talairach-Tournoux space; (3) motion and eddy current distortions were corrected for both the anterior-posterior (AP) and posterior-anterior (PA) diffusion-weighted imaging (DWI) data sets using a T2-weighted imitation image. AP and PA were combined to generate the final DWI reconstruction; (4) the diffusion ellipsoids and parameters were calculated using a nonlinear fitting method; (5) whole-brain coverage with 116 anatomical regions of interest (ROIs) based on the macrolabel atlas was transferred into individual DWI space; (6) deterministic fiber tracking was performed in the native space between all pairs of ROIs; (7) a 116 × 116 adjacency matrix for each subject was created based on the normalized number of streamlines, thresholded, and used as the measure of structural pairwise connectivity between all ROI pairs (i.e., the nodes of the network) number of averages = 2). A whole-brain DWI was acquired with

| Image preprocessing
Data preprocessing was performed using a combination of SPM12, FSL, TORTOISE, and AFNI software packages ( Figure 1). DWI data were corrected for motion, eddy current distortions and susceptibility-induced artifacts in both AP and PA directions. The AP and PA datasets were then combined using geometric averaging to generate the final corrected DWI dataset (Irfanoglu et al., 2015), F I G U R E 2 Overall structural neural community architecture. (a) The 116 × 116 matrices show averaged number of streamlines between each pair of brain regions in healthy controls, patients with laryngeal dystonia and writer's cramp, respectively. The neural community partition is based on the normalized number of streamlines between each pair of regions. The modules of the network are ordered and visualized according to the community structure. (b) The connectograms shows the same modules as in (a), with nodes (circles) labeled according to their regions and the degree/strength hubs (larger circles, bold labels) distributed within each module. 3D brain view in the center of the connectograms shows the spatial distribution of neural communities, with spheres representing hubs in each module. following which the diffusion ellipsoids and parameters were calculated with a nonlinear fitting method (Taylor & Saad, 2013). The uncertainty estimates were calculated using jackknife resampling.
Next, the individual T1-weighted images were skull-stripped, normalized to the standard template, and segmented into 116 anatomical regions of interest (ROIs) based on the macrolabel atlas (Eickhoff et al., 2005). In each subject, the T1-weighted image and ROIs were registered to the individual DWI scan in its native space for proper alignment between the datasets  using the fat_proc_map_to_dti program of AFNI software. For this, a

| Network integration
The measures of characteristic path length and global efficiency were used to determine the ability of the structural network to efficiently integrate information from distributed brain regions (Sporns, 2013).
The characteristic path length is computed as the average shortest path between all pairs of nodes. Inversely related to this measure is global efficiency, that is the average inverse shortest path length in the network (Rubinov & Sporns, 2010). The values from both measures were normalized with the total weights (Cheng et al., 2012).

| Network segregation
The measures of clustering coefficient and network modularity were used to examine the network capacity for specialized processing within interconnected brain regions (Sporns, 2013). The clustering coefficient is calculated as the likelihood of neighboring nodes to form segregated groups of nodes, also normalized with the total weights.
Between-group statistical differences were determined using two-way analysis of variance (ANOVA) with two factors: subject groups (LD, WC, controls) and examined graph measures (characteristic path length, global efficiency, clustering coefficient) at overall significance of p ≤ .05. If the overall group effect or its interaction with the graph measure was statistically significant, the follow-up post hoc univariate F-tests were computed to determine the differences between the groups.
Another commonly used graph measure is network modularity, which is a data-driven approach computed without an a priori set number of network decomposition modules. Modules are defined as segregated neural communities with dense inter-modular and weak intramodular connections. We used a multi-iterative (n = 100) generalization of the Louvain community detection algorithm (Blondel, Guil-laume, Lambiotte, & Lefebvre, 2008), which subdivided the network F I G U R E 4 Significant regional alterations of the structural connectome of dystonia patients compared to healthy controls. 3D brain renderings and bar graphs show significant regional changes in nodal degree, nodal strength, and betweenness centrality in (

| Nodal influence and the formation of hubs
To assess the nodal influence within the network, we computed the measures of nodal degree, strength, and betweenness centrality (Rubinov, Sporns, van Leeuwen, & Breakspear, 2009). Nodal degree (ki) is calculated as the number of links connected to the node, and nodal strength (si) is assessed as the sum of weights of links connected to the node. Betweenness centrality (bi) is the fraction of all shortest paths in the network that pass through a given edge, which is computed by converting the weighted connection matrix to a connection-length matrix and normalizing it using the factor (n − 2) (n − 1). This measure reflects the probability of information transfer through a given node along the shortest path between two random nodes (Brandes, 2001). As such, edges with higher values suggest participation in a large number of shortest paths and are of higher importance for controlling the information flow. Between-group statistical differences in nodal degree, strength, and betweenness centrality were determined using nonparametric permutation tests with 10,000 iterations at p ≤ .016 to correct for multiple comparisons (Nichols & Holmes, 2002).
Network hubs were determined based on nodal degree and strength of at least one SD greater than the average total degree and strength of the group network. The hubs were classified into provincial (i.e., linking nodes within a module with PI = 0.3-0.75) and connector (i.e., linking nodes between the modules with PI ≤ 0.3) (van den Heuvel & Sporns, 2011).

| Clinical correlates of network alterations
The relationship between clinical features of LD and WC, as described above and Table 1, with significantly abnormal network measures was assessed using Spearman's rank correlation coefficients at p ≤ .05. T A B L E 3 Differences in local graph metrics between TSFD patients and healthy controls

| RESULTS
Regional alterations in white matter integrity using tract-based spatial statistics, gray matter organization using voxel-based morphometry and cortical thickness analysis, as well as resting-state functional connectivity using independent component and graph theoretical analyses in LD and WC patients were reported in our previous studies Bianchi et al., 2019;Fuertinger & Simonyan, 2018).
The overall large-scale structural architecture was comparable between the dystonic and healthy states, forming five different neural Changes in network modular structure of TSFD patients were further instigated by abnormal hub formation. Based on nodal degree, both TSFD patients and healthy controls shared hubs in the right caudate nucleus, bilateral hippocampus, precuneus, putamen, and thalamus (Table 2). Based on nodal strength, all patients and controls shared hubs in the right precuneus, bilateral putamen, thalamus, posterior cingulate cortex, and cerebellar vermis (Table 2). Notably, the right posterior cingulate hub was downgraded from its connector status in healthy controls to the provincial status in both LD and WC patients. In addition, hubs in the right putamen and left posterior cingulate cortex were downgraded from their connector influence in healthy controls to provincial influence in WC patients, thus affecting the network information flow passing through these regions. Compared to healthy controls, both LD and WC patients commonly lost the left insular hub but gained the right superior frontal gyrus as degree connector hub (Figure 3a, Table 2).
TSFD-form specific hub alterations were as follows. LD patients distinctly gained the left anterior cingulate cortex (ACC) as strength connector hub and lost the left caudate nucleus as degree connector hub compared to healthy controls and WC patients (Figure 3a,  (Figure 3a, Table 2). Thus, alterations of the TSFD structural connectome were characterized by abnormal nodal migration across the neural communities and both common and distinct patterns of abnormal hub formation within these communities in LD and WC patients. At the regional level, structural networks in both forms of TSFD showed significant alterations in nodal degree, strength, and betweenness centrality compared to healthy controls ( Figure 4, Table 3). LD connectome was characterized by increased nodal degree and betweenness centrality in the supplementary motor area (SMA) and decreased betweenness centrally in the right superior parietal lobule (all p ≤ .016) (Figure 4a, Table 3). Conversely, WC patients showed decreased measures of nodal degree, strength or betweenness centrality in the bilateral ACC, insula, and right paracentral lobule (all p ≤ .016) (Figure 4b, Table 3). Thus, network nodes were distinctly altered in LD and WC patients compared to healthy controls, further pointing to dystonia-form specific neural changes.

| Clinical correlates of network alterations
There were no significant relationships between the duration of dystonia and the severity of either WC (p ≥ .23) or LD (all p ≥ .24), as well as between the age of LD onset and its severity as assessed by BFM movement and disability scores (p ≥ .21). The age of WC onset showed a significant negative correlation with the BFM disability score (R s = −0.76, p = .002) but not the BFM movement score (p = .47).
In addition, significant correlations were found between abnormal nodal degree of the left caudate nucleus and LD duration (R s = −0.50, p = .041) and nodal degree of the left insula and LD age of onset (R s = 0.51, p = .035) (Figure 3b). In WC, the symptom severity significantly correlated with abnormal nodal degree of the right globus pallidus (disability score: R s = 0.5, p = .04; movement score: R s = 0.63, p = .015), whereas the age of dystonia onset showed a correlation with abnormal nodal strength of the left SOG (R s = 0.51, p = .035) ( Figure 3b).

| DISCUSSION
Dystonia has been long considered a basal ganglia disorder (Berardelli et al., 1998;Defazio, Berardelli, & Hallett, 2007), with recent evidence suggesting the presence of additional abnormalities in the function of higher-order sensorimotor and associative cortical areas, especially in patients with TSFDs, such as LD and WC Fuertinger & Simonyan, 2018;Gallea, Horovitz, Ali Najee-Ullah, & Hallett, 2016). Mechanistic alterations of the functional connectome in these dystonias have been demonstrated to involve a top-down disruption of the sensorimotor network due to hyperexcitable parietal-basal ganglia connectivity (Battistella & Simonyan, 2019) and abnormal increases of striatal dopamine release contributing to the altered balance between the direct and indirect basal ganglia pathways during production of dystonic behaviors (Berman, Herscovitch, Hallett, & Simonyan, 2010;Simonyan, Berman, Herscovitch, & Hallett, 2013). The present study demonstrates that, in addition to profoundly altered functional network in focal dystonia, its structural The loss of insular hub in the LD and WC structural connectomes is in line with a similar deficit found in the functional connectome of these patients . Other neuroimaging studies reported abnormal activity during speaking in LD and writing in WC (Ali et al., 2006;Ceballos-Baumann, Sheean, Passingham, Marsden, & Brooks, 1997;Lerner et al., 2004;Peller et al., 2006;Simonyan & Ludlow, 2010), decreased cortical thickness linked to distinct clinical phenotypes of LD (Bianchi et al., 2017), and changes in GABA A receptor density in WC (Gallea et al., 2018;Peller et al., 2006). The insula is an important cortical outflow hub, being involved in various cognitive and sensorimotor behaviors, including generation of internal representations of intended movements (Karnath, Baier, & Nagele, 2005;Menon & Uddin, 2010;Sridharan, Levitin, & Menon, 2008). Our finding of the loss of the insula as network connector hub in both LD and WC, as well a significant correlation between its abnormal connectivity and LD age of onset, may suggest the failure of this region to coordinate the information flow between neural communities that participate in the control of sensorimotor processing during movement planning. Furthermore, together with the hub emergence in the SFG and nodal changes in the SMA and ACC, abnormal insular participation within the network may reflect abnormal monitoring of internal movement representations, decision making and working memory during performance of dystonia-affected behaviors (Bush et al., 2002;Bush, Luu, & Posner, 2000;Daw, O'Doherty, Dayan, Seymour, & Dolan, 2006;Kovach et al., 2012;Pochon et al., 2002;Xu et al., 2013;Zeuner et al., 2016).
Commonly altered hub formation in the basal ganglia may play an important role in further facilitation of abnormal traffic within LD and WC structural networks. Specifically, significant relationships between decreased connectivity of the caudate nucleus and LD duration as well as increased connectivity of the globus pallidus and WC severity suggest that these regions may take part in abnormal control of goaldirected motor behaviors and altered suppression of error feedback monitoring (Redgrave et al., 2010). It is important to note that the globus pallidus is currently defined as an effective deep brain stimulation (DBS) site in patients with dystonia (Volkmann et al., 2012). While the mechanisms of its neuromodulatory effects remain unclear, it is possible that the therapeutic outcome of pallidal DBS might, in part, be due to corrective reversal of altered pallidal connectivity.
Other regions that were commonly altered in the LD and WC connectomes were the cingulate and parietal cortical areas. Changes in structural connectivity of the cingulate cortex points to the aberrant motor action selection and error correction (Arrighi et al., 2016;Holroyd & Coles, 2002)  The connectomes in each form of focal dystonia were further characterized by a set of distinct alterations of hubs and nodes.
Regional abnormalities in the LD structural network most prominently involved the SMA, which is known to control action preparation, initiation and selection (Bonini et al., 2014;Swann et al., 2012) during speech production (Fuertinger, Horwitz, & Simonyan, 2015 (Suri et al., 2018). While secondary dystonias differ from isolated task-specific dystonia, such as LD and WC, in their causative mechanisms, lesion studies have traditionally provided important insights into causative brain function (Adolphs, 2016), including the significance of the basal ganglia in dystonia pathophysiology (Marsden, Obeso, Zarranz, & Lang, 1985). As such, similarities in brain alterations between secondary and isolated dystonias may point toward common underlying pathways involved in the occurrence of dystonic symptoms in general. More recently, another study in patients with focal dystonia, including WC, demonstrated that occipital regions contribute to the formation of aberrant network kernel (Fuertinger & Simonyan, 2018) and, together with premotor and parietal regions, support processing of visual temporal discriminatory stimuli in TSFD patients (Maguire, Reilly, & Simonyan, 2020). Future studies are warranted to conduct a detailed investigation of the involvement of the occipital region in the pathophysiology of dystonia.
The absence of graph measure abnormalities in the primary motor cortex and cerebellum may seem at first a counterintuitive finding given the fact that TSFDs are movement disorders. Notably, graph analysis is a data-driven methodology applied to the wholebrain data versus data-driven methods applied to a given region or network as in case of region-of-interest or seed-based studies described in the previous reports, which defined the presence of functional and structural alterations in these regions (Neychev et al., 2011;Simonyan, 2018;Zoons et al., 2011). Our current findings suggest that, at the level of a whole-brain network, other brain regions that are involved in the control of movement planning, preparation, and integration of sensorimotor information may play a more prominent role in the formation of the TSFD structural connectome than the primary motor cortex and cerebellum. This finding is in line with our recent study, which showed that functional alterations in premotor-parietal-basal ganglia circuitry precede those in the primary motor cortex, and the network disruption likely occurs well before the dystonic behavior is produced by the primary motor cortex (Battistella & Simonyan, 2019).
In conclusion, our data provide new evidence of abnormal largescale structural architecture in focal dystonia and propose that TSFD is a network disorder at both structural and functional levels. As several studies have suggested that non-invasive neuromodulation approaches, such as repetitive transcranial magnetic stimulation and transcranial direct current stimulation, modulate brain networks rather than only local targets of stimulation (To, De Ridder, Hart Jr., & Vanneste, 2018), the detailed knowledge of large-scale network organization in dystonia may prove useful in defining novel targets for therapeutic neuromodulation in this disorder.

ACKNOWLEDGMENTS
We thank Paul Taylor, PhD, for his assistance with the analysis of whole-brain tractography. This study was funded by the National Institute of Neurological Disorders and Stroke (R01NS088160 grant to KS).

CONFLICT OF INTEREST
The authors declare no conflicts of interest.

DATA AVAILABILITY STATEMENT
Upon the acceptance of this manuscript, the research data used in this study will be archived in the figshare public repository. Analytic codes used in this study are publicly available at https://simonyanlab.hms. harvard.edu/resources.