Abstract
Cervical dystonia is a neurological disorder characterized by sustained, involuntary movements of the head and neck. Most cases of cervical dystonia are idiopathic, with no obvious cause, yet some cases are acquired, secondary to focal brain lesions. These latter cases are valuable as they establish a causal link between neuroanatomy and resultant symptoms, lending insight into the brain regions causing cervical dystonia and possible treatment targets. However, lesions causing cervical dystonia can occur in multiple different brain locations, leaving localization unclear. Here, we use a technique termed ‘lesion network mapping’, which uses connectome data from a large cohort of healthy subjects (resting state functional MRI, n = 1000) to test whether lesion locations causing cervical dystonia map to a common brain network. We then test whether this network, derived from brain lesions, is abnormal in patients with idiopathic cervical dystonia (n = 39) versus matched controls (n = 37). A systematic literature search identified 25 cases of lesion-induced cervical dystonia. Lesion locations were heterogeneous, with lesions scattered throughout the cerebellum, brainstem, and basal ganglia. However, these heterogeneous lesion locations were all part of a single functionally connected brain network. Positive connectivity to the cerebellum and negative connectivity to the somatosensory cortex were specific markers for cervical dystonia compared to lesions causing other neurological symptoms. Connectivity with these two regions defined a single brain network that encompassed the heterogeneous lesion locations causing cervical dystonia. These cerebellar and somatosensory regions also showed abnormal connectivity in patients with idiopathic cervical dystonia. Finally, the most effective deep brain stimulation sites for treating dystonia were connected to these same cerebellar and somatosensory regions identified using lesion network mapping. These results lend insight into the causal neuroanatomical substrate of cervical dystonia, demonstrate convergence across idiopathic and acquired dystonia, and identify a network target for dystonia treatment.
Introduction
Cervical dystonia is a chronic neurological disorder characterized by sustained and involuntary contractions of the neck muscles, and is the most common form of focal dystonia (Xiao et al., 2012). Cervical dystonia has traditionally been ascribed to dysfunction of the basal ganglia (Galardi et al., 1996; Naumann et al., 1998), but abnormalities have been observed in many other brain regions including the cerebellum (Batla et al., 2015), prefrontal cortex (Li et al., 2017), midbrain (Holmes et al., 2012), motor cortex (Richardson, 2015), and somatosensory cortex (Prudente et al., 2016). This has led to the suggestion that cervical dystonia is a ‘network disorder’ resulting from dysfunction in multiple different brain regions (Jinnah et al., 2006). However, the key nodes of this network have yet to be identified. Further, it remains unclear which brain regions are causative and which are compensatory or incidental correlates.
Occasionally, a focal brain lesion can cause symptoms that are nearly identical to those observed in idiopathic cervical dystonia (LeDoux et al., 2003; Albanese et al., 2013). Although these cases of acquired cervical dystonia are rare compared to cases of idiopathic cervical dystonia (LeDoux et al., 2003), they are uniquely valuable because lesions allow for causal links between the damaged brain region and resultant symptoms (Adolphs, 2016; Fox, 2018). However, lesions causing cervical dystonia can occur in numerous brain locations, spanning the cerebellum, medulla, pons, midbrain, and basal ganglia (LeDoux et al., 2003). Further, symptoms can emerge not only from the lesion itself, but also from the effect of the lesion on remote but connected brain regions, a phenomenon referred to as diaschisis (von Monakow, 1914; Carrera et al., 2014). These factors complicate the localization of cervical dystonia symptoms based on focal brain lesions alone.
Recently, we validated a technique termed ‘lesion network mapping’, which can link lesions in different locations to a common brain network (Boes et al., 2015). Rather than focusing solely on lesion location, this technique uses a database of normative resting state functional connectivity MRI (rs-fcMRI) scans to identify the network of brain regions connected to each lesion location. This technique has lent insight into the localization of multiple neuropsychiatric symptoms (Fox, 2018), including other movement disorders (Fasano et al., 2016; Laganiere et al., 2016; Joutsa et al., 2018a), and may help identify therapeutic targets for brain stimulation therapies (Joutsa et al., 2018a, b). Here, we apply this approach to lesions causing cervical dystonia. We then go beyond prior lesion network mapping studies by investigating whether the neuroanatomical substrate of cervical dystonia derived from focal brain lesions is also abnormal in patients with similar symptoms, but without brain lesions.
Materials and methods
Case selection
Cases of lesions causing cervical dystonia were identified from a systematic search of Pubmed in January 2017 using the combination of synonyms of the following terms: cervical dystonia; torticollis; lesion; infarct; tumor; magnetic resonance imaging; and computerized tomography. The exact search syntax is provided in the Supplementary material. Reference lists of selected articles were searched for possible cases missed in the initial search. Inclusion criteria were: (i) neurological examination documenting cervical dystonia that was thought to be caused by an intraparenchymal brain lesion(s); and (ii) a figure or image showing the lesion location in sufficient clarity for it to be traced onto a standard brain atlas. Exclusion criteria were: (i) lesions in children aged <10 years, given that in these cases the brain is not sufficiently developed to resemble the standard adult brain; and (ii) lesions of the CNS but outside the brain (e.g. meningioma). As the emergence of dystonia may be delayed by months or even years following a brain insult (Scott et al., 1996; LeDoux et al., 2003), we did not apply a strict time limit for the onset of symptoms post-lesion. Based on these criteria, it is important to note that not all lesions causing cervical dystonia found in our search, which would be eligible based on clinical description (LeDoux and Brady, 2003), were eligible for inclusion in the current analysis.
Lesion network mapping
The network of regions functionally connected to each lesion location was identified using previously described methods (Boes et al., 2015; Darby et al., 2018a). First, lesions from published images were traced by hand onto a standardized brain atlas (2 × 2 × 2 mm MNI152 brain) using FSLview software (version 5.0.9) (Jenkinson et al., 2012). This approach generates only 2D slices of 3D lesions, but prior work has shown that the resulting connectivity maps are nearly identical (Boes et al., 2015; Darby et al., 2018a). Second, rs-fcMRI maps were created for each lesion using a standard seed-based approach, leveraging rs-fcMRI data from a normative dataset of 1000 healthy young adults (Yeo et al., 2011; Holmes et al., 2015). The time course of the average blood oxygen level-dependent signal within the lesion volume was extracted for each participant in the normative cohort and correlated with all brain voxels. Resulting individual r-maps were Fisher z-transformed, which were then used to generate a single connectivity t-map for each lesion. For step three, connectivity maps for each lesion were thresholded at a t-value of ±7 [corresponding to whole brain family-wise error (FWE)-corrected P < 10−6], binarized (functionally connected or not, positive and negative connectivity separately as they may have different biological interpretation), and then overlapped to identify voxels connected to all 25 of our lesion locations causing cervical dystonia (Fig. 1). This three-step technique is summarized in Fig. 2.
Lesion locations causing cervical dystonia. A systematic literature search identified 25 cases of cervical dystonia with an identifiable lesion location that could be traced onto a standard brain atlas. Case numbers correspond to those in Table 1 and Supplementary Table 1, which provides additional clinical details of cases listed in Table 1.
Lesion network mapping technique. In step one, lesions causing cervical dystonia were traced onto a standard atlas. In step two, connectivity between each lesion location and the rest of the brain was computed using a normative dataset of resting state functional connectivity scans from 1000 healthy individuals, and a standard seed-based approach. In step three, functional connectivity maps were thresholded, binarized (functionally connected or not), and overlapped to identify voxels connected to the greatest number of lesion locations.
We also ran a number of lesion network mapping subanalyses, excluding cases with ataxia or dysmetria (n = 11), head tremor (n = 6), hemiparesis (n = 9), dystonia symptoms outside of cervical regions (n = 6), and excluding cases not caused by ischaemic stroke (n = 15), to check that our findings were not being driven by these cases.
Specificity
To test for specificity to cervical dystonia, we compared our results to two ‘control’ datasets of lesions not causing cervical dystonia, as described previously (Joutsa et al., 2018a). First, we used a ‘non-specific’ dataset of lesions that were distributed throughout the brain without a common neuropsychiatric phenotype (n = 135) (Corbetta et al., 2015). Second, we used a ‘movement disorders’ dataset, consisting of 73 lesions causing movement disorders other than dystonia: asterixis (n = 30) (Kim, 2001; Laganiere et al., 2016); hemichorea-hemiballismus (n = 29) (Laganiere et al., 2016), and freezing of gait (n = 14) (Fasano et al., 2016).
We compared our network maps from lesions causing cervical dystonia to these two control lesion datasets using two statistical methods: (i) a Liebermeister test, using voxel-based lesion-symptom mapping (VLSM) (Rorden et al., 2007); and (ii) a two-sample t-test, using Statistical Parametric Mapping (SPM12; http://www.fil.ion.ucl.ac.uk/spm/software/spm12/) (Ashburner, 2012). Both statistical approaches identify voxels that are significantly more or less connected to cervical dystonia lesion locations than control lesion locations. The difference between these approaches is that the Liebermeister test analyses voxels in a binary fashion (functionally connected or not), and is more commonly used in lesion analyses, while the t-test takes into account the strength of the connection, and is more commonly used in functional neuroimaging (Fasano et al., 2016). Because the Liebermeister test is used for binary image analyses, the group comparisons were conducted separately for positive and negative connectivity maps. Correction for multiple comparisons was conducted using whole brain voxel-level FWE for t-tests and false discovery rate (FDR) for Liebermeister tests across the whole brain voxels showing at least 10% overlap in the whole sample. Corrected P-values <0.05 were considered significant. Specificity analyses were restricted to voxels within the cervical dystonia lesion network map [i.e. regions that were functionally connected to >90% (at least 23/25) of the lesions as shown in Fig. 3].
Lesion network mapping of cervical dystonia. (A) Regions positively correlated (orange/yellow) or negatively correlated (blue/green) to lesion locations causing cervical dystonia (thresholded at 90% or 23/25 cases). From left to right: thalamus (z = 10); globus pallidus (z = −2); midbrain (z = −13); cerebellum (z = −32), and somatosensory cortex (projected onto the brain surface). (B) Regions that are both sensitive and specific to lesions causing cervical dystonia, with significantly greater (positive or negative) functional connectivity to lesion locations causing cervical dystonia compared to control lesion locations (conjunction across four separate specificity analyses).
Regions of interest
To identify regions whose connectivity was both sensitive and specific to lesion locations causing cervical dystonia, we performed a conjunction analysis of the above maps. These regions of interest comprised of voxels that were connected to >90% of lesion locations causing cervical dystonia, and also specific to cervical dystonia across all four specificity analyses above (two control groups × two statistical tests). Because the somatosensory cortex cluster that survived all four specificity tests was very small (14 voxels with the centre of gravity at −8 −43 75 mm; Supplementary Fig. 1D), voxels surviving three of the four specificity analyses were used to define the somatosensory region of interest.
The resultant cerebellar and somatosensory regions of interest were then used in three analyses. First, we used a linear model to test whether connectivity between lesion locations and these regions of interest were independent or redundant predictors of lesion-induced cervical dystonia. Note that our method of selecting these regions of interest requires that connectivity to each region of interest alone be a predictor of cervical dystonia, but does not tell us whether these are independent predictors when combined in a linear model. Second, we used these regions of interest to generate a network map that, by definition, encompasses lesion locations causing cervical dystonia. To generate this map, we identified all voxels positively correlated with our cerebellar region of interest, all voxels negatively correlated with our somatosensory region of interest, thresholded each map (t ≥ ±7, voxelwise FWE corrected P < 10−6), and identified voxels meeting both criteria. Lesion locations were overlaid on this map for illustrative purposes. Finally, we used these regions of interest to test whether these same regions, identified based on brain lesions, were abnormal in idiopathic cervical dystonia.
These regions of interest were localized in greater anatomical detail using the Anatomy toolbox within SPM 12, using cerebellar (Schmahmann et al., 1999; Diedrichsen, 2006; Diedrichsen et al., 2009), motor cortex (Geyer et al., 1996), and somatosensory cortex (Geyer et al., 2000; Grefkes et al., 2001) atlases. The cerebellar atlas uses nomenclature of Schmahmann et al. (1999), and also includes updates provided by Diedrichsen (2006) and Diedrichsen et al. (2009) to determine fissure and lobule locations.
Relevance to idiopathic cervical dystonia
Our cerebellar and somatosensory regions of interest were used as seed regions to compare functional connectivity patterns between 39 idiopathic cervical dystonia patients and 37 control subjects, in a dataset collated from two previously published rs-fcMRI studies of idiopathic cervical dystonia (Delnooz et al., 2013; Prudente et al., 2016). The preprocessing of the rs-fcMRI data followed conventional methods and guidelines, including global signal regression (Fox et al., 2010; Murphy et al., 2016), but added an extra artefact-reduction step modified from prior principal component analysis-based approaches (Behzadi et al., 2007) (Supplementary material).
Functional connectivity of patients and controls was compared using non-parametric permutation interference with threshold-free cluster enhancement implemented in FSL software (Jenkinson et al., 2012). Permutation/randomization-based correction for multiple comparisons was selected to avoid inflated type I error rate often associated with parametric cluster-level correction (Winkler et al., 2014; Eklund et al., 2016). Because patients often move more than controls, two metrics of in-scanner movement were included as subject-level covariates (relative frame-to-frame motion and cumulative frame-wise transposition) in addition to dataset (Fox et al., 2010). FWE corrected P-values <0.05 were considered significant. The z-transformed values were extracted from all of the significant clusters to illustrate the direction of connectivity (positive or negative). Cumulative and relative in-scanner movement was compared between the groups using two-sample t-tests. P-values <0.05 were considered significant.
To assess specificity, we repeated this analysis using control regions of interest derived from prior lesion network mapping studies of other neurological symptoms (Boes et al., 2015; Fasano et al., 2016; Fischer et al., 2016; Laganiere et al., 2016; Darby et al., 2017; 2018a, b; Joutsa et al., 2018a). Control regions of interest were derived in the same way as our dystonia regions of interest, based on stronger connectivity to lesions causing a neurological symptom versus control lesions not causing the symptom. We identified 19 control regions of interest from eight previous papers, covering 11 different neurological symptoms (MNI coordinates of each control region are provided in the Supplementary material). Coordinates of one region of interest were not reported in the original study (Laganiere et al., 2016), and were identified through visual comparison with an atlas brain. For this analysis, a 5-mm radius sphere was generated at each coordinate, including centre of gravity coordinates for our cerebellar and somatosensory regions of interest. For each region of interest, we repeated the above permutation-based analysis to identify differences in connectivity between patients with idiopathic cervical dystonia and healthy controls. This resulted in a statistical map of T-values for each region of interest. To quantify the overall magnitude of these connectivity abnormalities, we computed the average absolute T-value of all brain voxels. We compared the average absolute T-values of the 19 control regions of interest to those from our two dystonia regions of interest (cerebellar and somatosensory) using two-sided one-sample t-tests with the null hypotheses that the control regions of interest do not differ from either of the cervical dystonia regions of interest.
Relevance to deep brain stimulation treatment
Clusters of voxels near the globus pallidus significantly associated with clinical response to deep brain stimulation (DBS) for dystonia were extracted from a recent study (Reich et al., 2019). Briefly, this study examined DBS electrode locations and stimulation sites from 105 patients with dystonia (53 cervical dystonia, and 52 generalized or segmental dystonia patients). Patients were categorized as having a ‘good’ or ‘poor’ DBS response based on improvement in Toronto Western Spasmodic Torticollis Rating Scale (TWSTRS) score (cervical dystonia), or Burke-Fahn-Marsden Dystonia Rating Scale (generalized or segmental dystonia). Voxels significantly associated with good clinical response were identified for the full cohort of dystonia patients (P < 0.01), and also separately for subjects with cervical dystonia (P < 0.05). Although not emphasized in the paper by Reich et al., there was also a cluster of voxels significantly associated with poor DBS response in the full dystonia cohort (P < 0.01). We tested whether ‘good’ clusters were functionally connected to our dystonia regions of interest, and whether this connectivity in the full dystonia cohort was significantly greater than for the ‘poor’ cluster, using our resting state functional connectivity dataset from 1000 healthy young adults (Yeo et al., 2011; Holmes et al., 2015). Finally, we performed a voxelwise analysis to identify voxels significantly connected to the ‘good’ cluster, controlling for connectivity to the ‘poor’ cluster using partial correlation. After z-transformation, the significance of the correlations was calculated using two-sided one-sample t-tests, and differences in connectivity from ‘good’ versus ‘poor’ clusters to our regions of interest were analysed using two-sided paired t-tests. Correlation to our cerebellar and somatosensory regions of interest, and to all brain voxels, was calculated as with lesion analyses, described in the previous paragraphs.
Data availability
Data are available from the corresponding authors upon request.
Results
Lesions causing cervical dystonia
We identified 25 cases of lesion-induced cervical dystonia that met our inclusion/exclusion criteria (Supplementary Table 1, Table 1 and Fig. 1). Lesions occurred in a number of different brain locations including the cerebellum (11 lesions), brainstem (n = 9), basal ganglia (n = 8), thalamus (n = 1), and occipital lobe (n = 1). Some patients had lesions in multiple locations.
Case characteristics of lesions causing cervical dystonia
| Case | Authors | Age/gender | Lesion type | Lesion location | Head/neck position | CD symptom latency |
|---|---|---|---|---|---|---|
| 1 | LeDoux and Brady (2003) | 55/M | Haemorrhage | R pons | L rotation | 12 h |
| 2 | LeDoux and Brady (2003) | 42/F | Cyst | L CB/pons | L rotation, R laterocollis | 3 to 4 months |
| 3 | LeDoux and Brady (2003) | 67/F | Infarct | Pons/midbrain | R latero- and anterocollis | Several days |
| 4 | LeDoux and Brady (2003) | 72/M | Infarct | Pons, L thalamus, L occipital | L rotation, retrocollis | 1 day |
| 5 | Isaac and Cohen (1989) | 28/M | Haemorrhage | R putamen | L rotation, R shoulder elevation | 4–5 years |
| 6 | Plant et al. (1989) | 30/F | MS plaques | R midbrain, R CB | L rotation | 1 year |
| 7 | Tranchant et al. (1991) | 53/F | Angioma | R CB | L rotation, antero- and laterocollis | 3 years |
| 8 | Molho and Factor (1993) | 68/F | Infarct | L putamen | R rotation, L laterocollis | Acute onset, 1 year before scan |
| 9 | Molho and Factor (1993) | 41/F | Infarct | L putamen | R rotation, R latero- and anterocollis | Acute onset, 3 years before scan |
| 10 | Schulze-Bonhage and Ferbert (1995) | 40/M | Glioma | R BG and frontoparietal WM | R laterocollis | 2 years before lesion was detected |
| 11 | Schwartz et al. (1995) | 63/M | Infarct | R BG and IC | R rotation, L laterocollis | Started gradually weeks before scan |
| 12 | Kajimoto et al. (2004) | 84/F | Infarct | L medulla | R laterocollis | 10 days |
| 13 | Loher and Krauss (2009) | 31/M | Haemorrhage | R midbrain, pons, CB. | R laterocollis, L rotation | 3 months |
| 14 | Loher and Krauss (2009) | 42/M | Haemorrhage | Midbrain and pons | R laterocollis, L rotation | 4 months |
| 15 | Loher and Krauss (2009) | 56/M | Haemorrhage | L pons, L CB | R laterocollis, L rotation | 1 month |
| 16 | Chang et al. (2002) | 23/M | Haemorrhage | L GPi | L rotation | 3 years |
| 17 | Zadro et al. (2008) | 48/F | Infarct | L CB | R rotation, anterocollis | 1 to 2 days |
| 18 | Usmani et al. (2011) | 37/M | Haemorrhage | Vermis, R CB | L rotation | 15 months |
| 19 | O’Rourke et al. (2006) | 35/F | Infarct | L and R CB | R rotation | 3 days |
| 20 | Batla et al. (2015) | 56/F | Tumour | L CB | R rotation | Information unavailable |
| 21 | Batla et al. (2015) | 33/M | Cyst | R CB | R rotation | Information unavailable |
| 22 | Batla et al. (2015) | 58/M | Infarct | L CB | R rotation | Information unavailable |
| 23 | Batla et al. (2015) | 29/M | Cyst | L CB | L rotation | Information unavailable |
| 24 | Kirton and Riopelle (2001) | 60/F | Infarct | L and R GP | Rotation, antero-, retro- and laterocollis | Several years |
| 25 | Lambrecq et al. (2010) | 23/M | Tumour | R BG, WM and ventricle | Anterocollis | Acute onset, scan taken within days |
| Case | Authors | Age/gender | Lesion type | Lesion location | Head/neck position | CD symptom latency |
|---|---|---|---|---|---|---|
| 1 | LeDoux and Brady (2003) | 55/M | Haemorrhage | R pons | L rotation | 12 h |
| 2 | LeDoux and Brady (2003) | 42/F | Cyst | L CB/pons | L rotation, R laterocollis | 3 to 4 months |
| 3 | LeDoux and Brady (2003) | 67/F | Infarct | Pons/midbrain | R latero- and anterocollis | Several days |
| 4 | LeDoux and Brady (2003) | 72/M | Infarct | Pons, L thalamus, L occipital | L rotation, retrocollis | 1 day |
| 5 | Isaac and Cohen (1989) | 28/M | Haemorrhage | R putamen | L rotation, R shoulder elevation | 4–5 years |
| 6 | Plant et al. (1989) | 30/F | MS plaques | R midbrain, R CB | L rotation | 1 year |
| 7 | Tranchant et al. (1991) | 53/F | Angioma | R CB | L rotation, antero- and laterocollis | 3 years |
| 8 | Molho and Factor (1993) | 68/F | Infarct | L putamen | R rotation, L laterocollis | Acute onset, 1 year before scan |
| 9 | Molho and Factor (1993) | 41/F | Infarct | L putamen | R rotation, R latero- and anterocollis | Acute onset, 3 years before scan |
| 10 | Schulze-Bonhage and Ferbert (1995) | 40/M | Glioma | R BG and frontoparietal WM | R laterocollis | 2 years before lesion was detected |
| 11 | Schwartz et al. (1995) | 63/M | Infarct | R BG and IC | R rotation, L laterocollis | Started gradually weeks before scan |
| 12 | Kajimoto et al. (2004) | 84/F | Infarct | L medulla | R laterocollis | 10 days |
| 13 | Loher and Krauss (2009) | 31/M | Haemorrhage | R midbrain, pons, CB. | R laterocollis, L rotation | 3 months |
| 14 | Loher and Krauss (2009) | 42/M | Haemorrhage | Midbrain and pons | R laterocollis, L rotation | 4 months |
| 15 | Loher and Krauss (2009) | 56/M | Haemorrhage | L pons, L CB | R laterocollis, L rotation | 1 month |
| 16 | Chang et al. (2002) | 23/M | Haemorrhage | L GPi | L rotation | 3 years |
| 17 | Zadro et al. (2008) | 48/F | Infarct | L CB | R rotation, anterocollis | 1 to 2 days |
| 18 | Usmani et al. (2011) | 37/M | Haemorrhage | Vermis, R CB | L rotation | 15 months |
| 19 | O’Rourke et al. (2006) | 35/F | Infarct | L and R CB | R rotation | 3 days |
| 20 | Batla et al. (2015) | 56/F | Tumour | L CB | R rotation | Information unavailable |
| 21 | Batla et al. (2015) | 33/M | Cyst | R CB | R rotation | Information unavailable |
| 22 | Batla et al. (2015) | 58/M | Infarct | L CB | R rotation | Information unavailable |
| 23 | Batla et al. (2015) | 29/M | Cyst | L CB | L rotation | Information unavailable |
| 24 | Kirton and Riopelle (2001) | 60/F | Infarct | L and R GP | Rotation, antero-, retro- and laterocollis | Several years |
| 25 | Lambrecq et al. (2010) | 23/M | Tumour | R BG, WM and ventricle | Anterocollis | Acute onset, scan taken within days |
BG = basal ganglia; CB = cerebellum; CD = cervical dystonia; GPi = globus pallidus interna; IC = internal capsule; L = left; MS = multiple sclerosis; R = right; WM = white matter.
Case characteristics of lesions causing cervical dystonia
| Case | Authors | Age/gender | Lesion type | Lesion location | Head/neck position | CD symptom latency |
|---|---|---|---|---|---|---|
| 1 | LeDoux and Brady (2003) | 55/M | Haemorrhage | R pons | L rotation | 12 h |
| 2 | LeDoux and Brady (2003) | 42/F | Cyst | L CB/pons | L rotation, R laterocollis | 3 to 4 months |
| 3 | LeDoux and Brady (2003) | 67/F | Infarct | Pons/midbrain | R latero- and anterocollis | Several days |
| 4 | LeDoux and Brady (2003) | 72/M | Infarct | Pons, L thalamus, L occipital | L rotation, retrocollis | 1 day |
| 5 | Isaac and Cohen (1989) | 28/M | Haemorrhage | R putamen | L rotation, R shoulder elevation | 4–5 years |
| 6 | Plant et al. (1989) | 30/F | MS plaques | R midbrain, R CB | L rotation | 1 year |
| 7 | Tranchant et al. (1991) | 53/F | Angioma | R CB | L rotation, antero- and laterocollis | 3 years |
| 8 | Molho and Factor (1993) | 68/F | Infarct | L putamen | R rotation, L laterocollis | Acute onset, 1 year before scan |
| 9 | Molho and Factor (1993) | 41/F | Infarct | L putamen | R rotation, R latero- and anterocollis | Acute onset, 3 years before scan |
| 10 | Schulze-Bonhage and Ferbert (1995) | 40/M | Glioma | R BG and frontoparietal WM | R laterocollis | 2 years before lesion was detected |
| 11 | Schwartz et al. (1995) | 63/M | Infarct | R BG and IC | R rotation, L laterocollis | Started gradually weeks before scan |
| 12 | Kajimoto et al. (2004) | 84/F | Infarct | L medulla | R laterocollis | 10 days |
| 13 | Loher and Krauss (2009) | 31/M | Haemorrhage | R midbrain, pons, CB. | R laterocollis, L rotation | 3 months |
| 14 | Loher and Krauss (2009) | 42/M | Haemorrhage | Midbrain and pons | R laterocollis, L rotation | 4 months |
| 15 | Loher and Krauss (2009) | 56/M | Haemorrhage | L pons, L CB | R laterocollis, L rotation | 1 month |
| 16 | Chang et al. (2002) | 23/M | Haemorrhage | L GPi | L rotation | 3 years |
| 17 | Zadro et al. (2008) | 48/F | Infarct | L CB | R rotation, anterocollis | 1 to 2 days |
| 18 | Usmani et al. (2011) | 37/M | Haemorrhage | Vermis, R CB | L rotation | 15 months |
| 19 | O’Rourke et al. (2006) | 35/F | Infarct | L and R CB | R rotation | 3 days |
| 20 | Batla et al. (2015) | 56/F | Tumour | L CB | R rotation | Information unavailable |
| 21 | Batla et al. (2015) | 33/M | Cyst | R CB | R rotation | Information unavailable |
| 22 | Batla et al. (2015) | 58/M | Infarct | L CB | R rotation | Information unavailable |
| 23 | Batla et al. (2015) | 29/M | Cyst | L CB | L rotation | Information unavailable |
| 24 | Kirton and Riopelle (2001) | 60/F | Infarct | L and R GP | Rotation, antero-, retro- and laterocollis | Several years |
| 25 | Lambrecq et al. (2010) | 23/M | Tumour | R BG, WM and ventricle | Anterocollis | Acute onset, scan taken within days |
| Case | Authors | Age/gender | Lesion type | Lesion location | Head/neck position | CD symptom latency |
|---|---|---|---|---|---|---|
| 1 | LeDoux and Brady (2003) | 55/M | Haemorrhage | R pons | L rotation | 12 h |
| 2 | LeDoux and Brady (2003) | 42/F | Cyst | L CB/pons | L rotation, R laterocollis | 3 to 4 months |
| 3 | LeDoux and Brady (2003) | 67/F | Infarct | Pons/midbrain | R latero- and anterocollis | Several days |
| 4 | LeDoux and Brady (2003) | 72/M | Infarct | Pons, L thalamus, L occipital | L rotation, retrocollis | 1 day |
| 5 | Isaac and Cohen (1989) | 28/M | Haemorrhage | R putamen | L rotation, R shoulder elevation | 4–5 years |
| 6 | Plant et al. (1989) | 30/F | MS plaques | R midbrain, R CB | L rotation | 1 year |
| 7 | Tranchant et al. (1991) | 53/F | Angioma | R CB | L rotation, antero- and laterocollis | 3 years |
| 8 | Molho and Factor (1993) | 68/F | Infarct | L putamen | R rotation, L laterocollis | Acute onset, 1 year before scan |
| 9 | Molho and Factor (1993) | 41/F | Infarct | L putamen | R rotation, R latero- and anterocollis | Acute onset, 3 years before scan |
| 10 | Schulze-Bonhage and Ferbert (1995) | 40/M | Glioma | R BG and frontoparietal WM | R laterocollis | 2 years before lesion was detected |
| 11 | Schwartz et al. (1995) | 63/M | Infarct | R BG and IC | R rotation, L laterocollis | Started gradually weeks before scan |
| 12 | Kajimoto et al. (2004) | 84/F | Infarct | L medulla | R laterocollis | 10 days |
| 13 | Loher and Krauss (2009) | 31/M | Haemorrhage | R midbrain, pons, CB. | R laterocollis, L rotation | 3 months |
| 14 | Loher and Krauss (2009) | 42/M | Haemorrhage | Midbrain and pons | R laterocollis, L rotation | 4 months |
| 15 | Loher and Krauss (2009) | 56/M | Haemorrhage | L pons, L CB | R laterocollis, L rotation | 1 month |
| 16 | Chang et al. (2002) | 23/M | Haemorrhage | L GPi | L rotation | 3 years |
| 17 | Zadro et al. (2008) | 48/F | Infarct | L CB | R rotation, anterocollis | 1 to 2 days |
| 18 | Usmani et al. (2011) | 37/M | Haemorrhage | Vermis, R CB | L rotation | 15 months |
| 19 | O’Rourke et al. (2006) | 35/F | Infarct | L and R CB | R rotation | 3 days |
| 20 | Batla et al. (2015) | 56/F | Tumour | L CB | R rotation | Information unavailable |
| 21 | Batla et al. (2015) | 33/M | Cyst | R CB | R rotation | Information unavailable |
| 22 | Batla et al. (2015) | 58/M | Infarct | L CB | R rotation | Information unavailable |
| 23 | Batla et al. (2015) | 29/M | Cyst | L CB | L rotation | Information unavailable |
| 24 | Kirton and Riopelle (2001) | 60/F | Infarct | L and R GP | Rotation, antero-, retro- and laterocollis | Several years |
| 25 | Lambrecq et al. (2010) | 23/M | Tumour | R BG, WM and ventricle | Anterocollis | Acute onset, scan taken within days |
BG = basal ganglia; CB = cerebellum; CD = cervical dystonia; GPi = globus pallidus interna; IC = internal capsule; L = left; MS = multiple sclerosis; R = right; WM = white matter.
Lesion network mapping
Each lesion location was converted into a lesion network map, and regions functionally connected to all or most lesion locations causing cervical dystonia were identified (Fig. 2). Despite heterogeneity in lesion locations, all lesions causing cervical dystonia were part of a single functionally connected brain network. All 25 lesion locations were functionally connected (positively correlated) to the cerebellar vermis, dentate nucleus, cerebellar cortex, and midbrain (Table 2), and over 90% of lesion locations were functionally connected to the thalamus and globus pallidus (Fig. 3A). All 25 lesion locations were also functionally connected, but negatively correlated, to the right somatosensory cortex (Table 2), and over 90% of lesion locations were connected to the somatosensory cortex bilaterally, extending slightly into the motor cortex (Fig. 3A). Medial and lateral clusters were found within the somatosensory cortex, consistent with previous reports of both a medial and lateral representation for the neck within the homunculus (Prudente et al., 2015, 2016). Some smaller clusters of (positive and negative) functionally connected voxels were also present (Supplementary Fig. 2). Results were unchanged when excluding cases with ataxia or dysmetria, head tremor, hemiparesis, dystonia symptoms outside of cervical regions, or cases not caused by ischaemic stroke (Supplementary Fig. 3).
Brain regions functionally connected to 25/25 lesions causing cervical dystonia
| Voxels | x | y | z | Brain region |
|---|---|---|---|---|
| Regions positively connected with lesions | ||||
| 111 | −6 | −52 | −36 | Medial cerebellum |
| 38 | −33 | −55 | −29 | Left cerebellar cortex |
| 22 | 14 | −54 | −36 | Medial cerebellum |
| 9 | 8 | −17 | −14 | Midbrain |
| 9 | 1 | −25 | −12 | Midbrain |
| 7 | 35 | −52 | −30 | Right cerebellar cortex |
| 3 | 0 | −20 | −13 | Midbrain |
| 2 | 31 | −58 | −28 | Right cerebellar cortex |
| 1 | 26 | −24 | −4 | Right lateral geniculate nucleus |
| Regions negatively connected with lesions | ||||
| 32 | 45 | −30 | 58 | Somatosensory cortex |
| Voxels | x | y | z | Brain region |
|---|---|---|---|---|
| Regions positively connected with lesions | ||||
| 111 | −6 | −52 | −36 | Medial cerebellum |
| 38 | −33 | −55 | −29 | Left cerebellar cortex |
| 22 | 14 | −54 | −36 | Medial cerebellum |
| 9 | 8 | −17 | −14 | Midbrain |
| 9 | 1 | −25 | −12 | Midbrain |
| 7 | 35 | −52 | −30 | Right cerebellar cortex |
| 3 | 0 | −20 | −13 | Midbrain |
| 2 | 31 | −58 | −28 | Right cerebellar cortex |
| 1 | 26 | −24 | −4 | Right lateral geniculate nucleus |
| Regions negatively connected with lesions | ||||
| 32 | 45 | −30 | 58 | Somatosensory cortex |
Brain regions functionally connected to 25/25 lesions causing cervical dystonia
| Voxels | x | y | z | Brain region |
|---|---|---|---|---|
| Regions positively connected with lesions | ||||
| 111 | −6 | −52 | −36 | Medial cerebellum |
| 38 | −33 | −55 | −29 | Left cerebellar cortex |
| 22 | 14 | −54 | −36 | Medial cerebellum |
| 9 | 8 | −17 | −14 | Midbrain |
| 9 | 1 | −25 | −12 | Midbrain |
| 7 | 35 | −52 | −30 | Right cerebellar cortex |
| 3 | 0 | −20 | −13 | Midbrain |
| 2 | 31 | −58 | −28 | Right cerebellar cortex |
| 1 | 26 | −24 | −4 | Right lateral geniculate nucleus |
| Regions negatively connected with lesions | ||||
| 32 | 45 | −30 | 58 | Somatosensory cortex |
| Voxels | x | y | z | Brain region |
|---|---|---|---|---|
| Regions positively connected with lesions | ||||
| 111 | −6 | −52 | −36 | Medial cerebellum |
| 38 | −33 | −55 | −29 | Left cerebellar cortex |
| 22 | 14 | −54 | −36 | Medial cerebellum |
| 9 | 8 | −17 | −14 | Midbrain |
| 9 | 1 | −25 | −12 | Midbrain |
| 7 | 35 | −52 | −30 | Right cerebellar cortex |
| 3 | 0 | −20 | −13 | Midbrain |
| 2 | 31 | −58 | −28 | Right cerebellar cortex |
| 1 | 26 | −24 | −4 | Right lateral geniculate nucleus |
| Regions negatively connected with lesions | ||||
| 32 | 45 | −30 | 58 | Somatosensory cortex |
Connectivity to the cerebellum and somatosensory cortex was specific to lesion locations causing cervical dystonia, compared to control lesion locations, independent of the statistical approach and control dataset (Fig. 3B). We performed a conjunction analysis to identify regions whose connectivity was both sensitive and specific to lesion locations causing cervical dystonia. This identified a region of interest in the cerebellum, centred on the vermis of lobule IX (MNI coordinates 1 −54 −34 mm) and a region of interest in the somatosensory cortex/Brodmann’s area 1 (MNI coordinates right hemisphere centre of gravity: 45 −24 60 mm; and left hemisphere centre of gravity: and −45 −28 59 mm) (Fig. 3B). See Supplementary Table 2 for greater anatomical detail of these region of interest locations, and Supplementary Figs 4 and 5 for overlay of our cervical dystonia regions of interest on cerebellar (Schmahmann et al., 1999; Diedrichsen, 2006; Diedrichsen et al., 2009), and sensorimotor cortex atlases (Geyer et al., 1996, 2000; Grefkes et al., 2001). Our cerebellar region of interest, connected to lesions causing cervical dystonia, was in a different region of the cerebellum than a previously published region of interest connected to lesions causing freezing of gait (Fasano et al., 2016) (Supplementary Fig. 6).
To test whether connectivity from the lesions to cerebellum and somatosensory cortex regions of interest were independent or redundant predictors of lesion-induced cervical dystonia, we included both as factors in a logistic regression model with a binary outcome variable (cervical dystonia lesion versus control lesion). Cerebellum region of interest connectivity was a strong independent predictor of cervical dystonia (P = 0.002), while somatosensory cortex region of interest connectivity fell just short of our statistical threshold for independence (P = 0.051).
By definition, connectivity with our cerebellar and somatosensory regions of interest defines a network that encompasses lesion locations causing cervical dystonia while avoiding control lesions. To illustrate this, we constructed a map of voxels both positively correlated with our cerebellar region of interest and negatively correlated with our somatosensory region of interest. As expected, our lesion locations causing cervical dystonia fell within this topographic distribution (Fig. 4), although one lesion fell just at the boundary of this network (Case 8).
Lesions causing cervical dystonia are part of a commonly connected brain network. The combination of positive connectivity to our cerebellum region of interest and negative connectivity to our somatosensory region of interest defines a network of regions (blue) that encompasses 24 of 25 lesion locations causing cervical dystonia (red). Case 8 lesion location falls immediately adjacent to this network.
Relevance to idiopathic cervical dystonia
The aforementioned cerebellar and somatosensory regions of interest, identified based on brain lesions, were used as seed regions to test whether these same regions were abnormal in patients with idiopathic cervical dystonia. Our seed region of interest in the cerebellum showed abnormal connectivity to regions in the lateral sensorimotor cortex and operculum (Fig. 5A and Supplementary Table 3). Our seed region of interest in the somatosensory cortex showed abnormal connectivity to regions in the basal ganglia, thalamus, anterior cingulate, occipital cortex, and sensorimotor cortex (Fig. 5B and Supplementary Table 3). Each of these connectivity abnormalities involved a loss of normal negative or positive connectivity (Fig. 5).
Relevance to idiopathic cervical dystonia. Connectivity with our lesion-derived cerebellar region of interest (A, red) and somatosensory region of interest (B, blue) is abnormal in patients with idiopathic cervical dystonia. Patients with idiopathic cervical dystonia had a loss of negative functional connectivity from our cerebellar region of interest to the right sensorimotor cortex (z = 38) (orange/yellow) (A). Patients with idiopathic cervical dystonia also had loss of negative connectivity from our somatosensory region of interest to regions in the thalamus/basal ganglia and anterior cingulate (z = 12) (orange/yellow), and loss of positive connectivity to the sensorimotor and occipital cortex (z = 27) (blue/green) (B). All images were corrected with threshold-free cluster enhancement PFWE < 0.05. Corresponding average (SEM) Fischer z transformed correlation coefficients (Fz) are shown in box and whisker plots. Middle line within plot represents median, and cross shows mean. CD = idiopathic cervical dystonia patients; HV = healthy volunteers.
Our two regions of interest derived from brain lesions causing cervical dystonia showed greater abnormalities in idiopathic cervical dystonia patients than 19 control regions of interest derived from lesions causing other neurological symptoms (cerebellar region of interest versus control regions P < 0.001; somatosensory region of interest versus control regions P < 0.001). Average absolute t-values for all regions of interest are presented in Supplementary Table 4.
There was no difference in cumulative (P = 0.69) or relative (P = 0.22) in-scanner movement between cervical dystonia patients and healthy volunteers.
Relevance to deep brain stimulation treatment
Finally, we examined whether our cerebellar and somatosensory regions of interest, derived from focal brain lesions and abnormal in patients with idiopathic cervical dystonia, were relevant to DBS treatment (Fig. 6). Our cerebellar region of interest was positively connected to DBS sites associated with good clinical response in cervical dystonia patients (P < 0.001) and in dystonia patients in general (P < 0.001), and significantly more connected to DBS sites associated with good, compared to poor, clinical response (P = 0.002). Our somatosensory region of interest was negatively connected with the optimal DBS site for treating cervical dystonia (P < 0.001) and dystonia in general (P < 0.001) and significantly more negatively connected to DBS sites associated with good, compared to poor, responses (P < 0.001). Functional connectivity with these DBS sites matched the spatial topography of our cervical dystonia network derived from focal brain lesions (Fig. 6).
Relevance to deep brain stimulation. Lesions causing cervical dystonia are negatively connected to the somatosensory cortex and positively to the cerebellum (A). Globus pallidus interna DBS locations associated with good clinical response in cervical dystonia (B) and all dystonia patients (C) also show negative functional connectivity to the somatosensory cortex and positive functional connectivity to the cerebellum. Voxels associated with good response also show similar connectivity profile when controlling for voxels associated with poor response (D).
Discussion
There are several noteworthy findings. First, lesions causing cervical dystonia are found in heterogeneous brain locations, but are part of a single functionally connected brain network. Second, this network is defined by positive connectivity to the cerebellum and negative connectivity to the somatosensory cortex, a pattern that is specific to lesions causing cervical dystonia compared to control lesions. Finally, this network is abnormal in patients with idiopathic cervical dystonia, and also matches the connectivity pattern of DBS sites associated with dystonia symptom improvement. These findings suggest a shared neuroanatomical network for cervical dystonia independent of symptom aetiology, and illustrate how lesion network mapping can guide the search for brain abnormalities and treatment targets in non-lesion patients with similar neurological symptoms.
Lesion network mapping in cervical dystonia
It is well known that lesions causing cervical dystonia can occur in different brain locations (LeDoux et al., 2003), and connectivity with the lesion locations has been hypothesized to play a role in explaining this phenomenon (LeDoux et al., 2003; Prudente et al., 2014). Lesion network mapping allows for direct testing of this hypothesis by integrating brain connectivity into lesion analysis (Boes et al., 2015; Fox, 2018). This technique allows one to localize lesion-induced symptoms to networks, rather than individual brain regions, and has proven useful in localization of other movement disorders (Fasano et al., 2016; Laganiere et al., 2016; Joutsa et al., 2018a). In the present study, we localize cervical dystonia to a single brain network defined by connectivity to the cerebellum and somatosensory cortex.
The cerebellum in cervical dystonia
It has been suggested that cervical dystonia may arise from dysfunction of the cerebellum given its role in integrating motor and proprioceptive inputs to coordinate movement (LeDoux et al., 2003; Jinnah et al., 2006). This hypothesis is supported by functional MRI abnormalities in the cerebellum in cervical dystonia patients (Prudente et al., 2016; Li et al., 2017), Purkinje cell loss on human autopsy (Prudente et al., 2013), and rodent studies causing dystonia via the manipulation of the cerebellum (Pizoli et al., 2002; Calderon et al., 2011). The present study adds to this previous work by showing that all lesion locations causing cervical dystonia are connected to the cerebellum, including the cerebellar cortex, vermis, and dentate nucleus (Fig. 3, Supplementary Table 2 and Supplementary Fig. 4).
Several studies have found normal cerebellar function or connectivity in idiopathic cervical dystonia patients, or abnormal cerebellar function only in cervical dystonia patients with tremor (Delnooz et al., 2013; Sadnicka et al., 2014; Antelmi et al., 2016; Bologna et al., 2016; Avanzino et al., 2018). Here, we found that all lesions causing cervical dystonia were connected to the cerebellum, including cases with no head tremor (Supplementary Fig. 3D). We also found abnormal cerebellar connectivity in our idiopathic cervical dystonia resting-state functional MRI dataset, composed of patients with minimal or no head tremor (Fig. 5) (Delnooz et al., 2013; Prudente et al., 2016). One possible explanation for these discordant findings is that cervical dystonia involves a specific region within the cerebellum, and other cerebellar regions can appear normal, or abnormal only in patients with tremor. Another possibility is that our lesion-based approach and larger cohort size allowed for increased sensitivity for cerebellar abnormalities in cervical dystonia.
The somatosensory cortex in cervical dystonia
Prior work also implicates the somatosensory cortex in the pathophysiology of cervical dystonia by demonstrating hyperactivity during head rotation (Prudente et al., 2016), increased plasticity to sensorimotor stimuli (Kojovic et al., 2013; Koch et al., 2014), and disinhibition (Inoue et al., 2004). It has been hypothesized that dystonia may result from increased proprioceptive input to the somatosensory cortex, leading to ‘motor overflow’ and co-contraction of muscles (Hallett, 2011; Kaňovský et al., 2011). Our finding that lesion locations are negatively correlated to the somatosensory cortex may be consistent with this hypothesis. The interpretation of negative correlations seen with fcMRI remains a matter of debate (Murphy et al., 2016); however, negative correlations may represent brain regions that are suppressed during activation of competing regions (Fox et al., 2005). Based on this model, a lesion causing cervical dystonia could result in a loss of the normal suppressive input from the lesion location to the somatosensory cortex, and therefore hyperactivity in this region.
Similar results are seen in lesion-induced hallucinations. Specifically, lesions causing visual or auditory hallucinations are negatively correlated with visual and auditory cortices, respectively (Boes et al., 2015). Other similarities exist between cervical dystonia and hallucinations, including hyperactivity in the relevant sensory cortical area (Prudente et al., 2016; Zmigrod et al., 2016), and symptom improvement with sensory input. For example, visual and auditory hallucinations can improve with visual and auditory input (Teunisse et al., 1996; Corlett et al., 2009), while cervical dystonia can improve with sensory or proprioceptive input, the so-called ‘geste antagoniste’ or sensory trick (Naumann et al., 2000; Schramm et al., 2004). The notion that cervical dystonia may be a form of sensory or proprioceptive hallucination is highly speculative, but a testable hypothesis motivated by the present findings.
The basal ganglia in cervical dystonia
Our findings emphasize the importance of the cerebellum and somatosensory cortex in defining the cervical dystonia network, but do not discount a role for the basal ganglia or other brain regions (Neychev et al., 2011). For example, 24 of 25 cervical dystonia lesion locations were connected to the globus pallidus (Fig. 3A), an effective DBS target for cervical dystonia (Volkmann et al., 2014). However, unlike the cerebellum and somatosensory cortex, connectivity to the basal ganglia was not specific to lesions causing cervical dystonia. This is not surprising given the role of the basal ganglia in other lesion-induced symptoms, including other movement disorders included in our ‘control lesions’ sample (Fasano et al., 2015; Laganiere et al., 2016). Similarly, treatments targeting the basal ganglia such as DBS are not specific to cervical dystonia, but are also effective for other movement disorders (Follett et al., 2010). As such, connectivity to the cerebellum and somatosensory cortex are the most sensitive and specific markers of lesion-induced cervical dystonia, but this does not discount the involvement of the basal ganglia in cervical dystonia.
A two-hit model of cervical dystonia
The involvement of two distinct brain regions differs from previous ‘lesion network mapping’ studies of movement disorders where lesion locations were characterized by connectivity to just a single location (Fasano et al., 2016; Laganiere et al., 2016; Joutsa et al., 2018a). Results in cervical dystonia are similar to lesion network mapping of more complex symptoms such as hallucinations (Boes et al., 2015), delusions (Darby et al., 2017), and criminality (Darby et al., 2018a), in which lesion locations were positively connected to one brain region and negatively connected to another. Connectivity of lesion locations to two different regions is consistent with two-hit models of symptom generation. For example, delusions are thought to require both a disruption in sensory processing and belief evaluation (Coltheart, 2010). A two-hit model has previously been proposed for dystonia (Schicatano et al., 1997; Jinnah et al., 2006; Neychev et al., 2008), but these models usually implicate the cerebellum and basal ganglia. Our results suggest that cervical dystonia symptoms may be caused by combined dysfunction of the cerebellum and somatosensory cortex.
Relevance to idiopathic cervical dystonia
One of our most important findings is the demonstration that lesion network mapping can guide analyses of patients with similar symptoms but without brain lesions, to identify a common neuroanatomical substrate. Idiopathic and acquired cervical dystonia can be indistinguishable clinically (LeDoux et al., 2003); however, unlike acquired cervical dystonia where symptoms are causally linked to a lesion location, the brain regions causing idiopathic cervical dystonia are difficult to isolate. Neuroimaging studies have implicated many different regions and connections between regions (Delnooz et al., 2013; Prudente et al., 2016; Li et al., 2017), and determining which abnormalities are causing symptoms, compensating for symptoms, or incidentally correlated with symptoms can prove difficult or impossible. By starting with brain lesions, we identified a network causally linked to cervical dystonia, defined by connectivity to the cerebellum and somatosensory cortex. Connectivity with these two regions thus defines a distributed brain network that encompasses lesion locations causing cervical dystonia. The fact that connectivity with these same two regions is abnormal in idiopathic cervical dystonia suggests a shared neuroanatomical substrate for idiopathic and acquired cervical dystonia. Note that connectivity need not be abnormal between these two regions to establish this convergence, as it is connectivity between each region and all other brain voxels that defines the cervical dystonia network. Finally, the fact that these two regions were significantly more abnormal than 19 other control regions suggests that lesion network mapping can help identify the location of key abnormalities in patients with similar symptoms but who do not have brain lesions.
This shared neuroanatomical substrate generates testable hypotheses for identifying and refining therapeutic targets in cervical dystonia. For example, DBS to the globus pallidus is effective for many but not all patients with cervical dystonia (Kiss et al., 2007; Volkmann et al., 2014). Here we show that globus pallidus DBS electrode locations associated with good clinical response have positive connectivity to the cerebellum and negative connectivity to the somatosensory cortex. This importance of brain connectivity in mediating DBS response is reminiscent of recent work in Parkinson’s disease (Horn et al., 2017; Joutsa et al., 2018a). Similarly, transcranial magnetic stimulation to the lateral cerebellum has shown some promise in patients with cervical dystonia (Koch et al., 2014), and this target could possibly be refined based on the current results. Finally, the present results highlight the somatosensory cortex as a potential therapeutic target easily amenable to non-invasive brain stimulation. Though this target has yet to be tried in cervical dystonia to our knowledge, there is some evidence that this target may provide benefit to patients with hand dystonia (Havrankova et al., 2010).
Limitations
A number of limitations should be acknowledged. First, although we conducted a systematic search to collect a representative sample of brain lesions causing cervical dystonia, we cannot exclude a publication bias, as lesions in locations previously linked to cervical dystonia may be more likely to be reported. Second, there are potential limitations regarding lesion network mapping, such as drawing lesions by hand, using 2D instead of real 3D lesions, and the use of a normative connectome dataset. However, these limitations have been addressed in detail previously and found to have little impact on lesion network mapping results (Boes et al., 2015; Darby et al., 2018a). Next, interpretations based on functional connectivity data are based on indirect evidence, which constrains the causal interpretation of lesion network mapping findings (Fox, 2018), and of functional connectivity abnormalities observed in patient populations (Fox et al., 2010; Delnooz et al., 2013; Prudente et al., 2016). Finally, there have been numerous brain regions implicated in idiopathic cervical dystonia based on neuroimaging (Prudente et al., 2016; Li et al., 2017), and it is yet to be determined whether the subset of regions connected to causal brain lesions identified in this study will prove more central to symptom pathophysiology, or more useful as treatment targets.
Conclusions
Lesion locations causing cervical dystonia are part of a common brain network defined by connectivity to the cerebellum and the somatosensory cortex. This network, identified based on brain lesions, is abnormal in patients with idiopathic cervical dystonia, and aligns with effective DBS sites. We suggest a shared substrate for idiopathic and acquired cervical dystonia, propose a two-hit model of cervical dystonia symptoms, and provide testable hypotheses for improving treatment.
Abbreviations
Acknowledgement
The authors wish to thank Prof. Jens Volkmann for sharing DBS localization data, and for valuable feedback on manuscript drafts.
Funding
The present work was supported by the Dorothy Feiss Dystonia Research Fund and the Dystonia Medical Research Foundation. D.C. was supported by a Victoria Fellowship awarded by the Veski Foundation. J.J. was supported by the Academy of Finland (Grant #295580), the Finnish Medical Foundation, and the Orion Research Foundation. M.D.F. was supported by K23 NS083741, R01 MH113929, the Nancy Lurie Marks Foundation, the National Parkinson’s Foundation, and the Mathers Foundation. C.D. receives support from Prinses Beatrix fund and Hersenstichting. B.P.C.vdW. receives research support from Hersenstichting, ZonMW, Bioblast Pharma, and Radboud University Medical Centre. H.L. is supported by the NIH grants 1R01NS091604, P50MH106435, Beijing Municipal Science & Technology Commission grant No. Z161100002616009; and National Natural Science Foundation of China grant No. 81790652. J.J. reports travel grants from Abbvie and Orion. H.J. has active or recent grant support from the US government (National Institutes of Health), private philanthropic organizations (the Benign Essential Blepharospasm Research Foundation, Cure Dystonia Now), academically-oriented institutions (the Dystonia Study Group), and industry (Cavion Therapeutics, Ipsen Pharmaceuticals, Retrophin Inc.). H.J. has served on advisory boards or as a consultant for Abide Therapeutics, Allergan Inc., Psyadon Pharmaceuticals, Retrophin Inc., Saol Therapeutics, and Medtronic Inc. H.J. has received honoraria or stipends for lectures or administrative work from the American Academy of Neurology, the Dystonia Medical Research Foundation, the International Neurotoxin Society, the International Parkinson’s Disease and Movement Disorders Society, The Parkinson’s Disease Foundation, and Tyler’s Hope for a Cure. H.J. serves on the Scientific Advisory Boards for Cure Dystonia Now, the Dystonia Medical Research Foundation, Lesch-Nyhan Action France, and Tyler’s Hope for a Cure. He also is principle investigator for the Dystonia Coalition, which receives the majority of its support through NIH grant TR001456 from the Office of Rare Diseases Research at the National Center for Advancing Translational Sciences, and previously NS065701 from the National Institutes of Neurological Disorders and Stroke. The Dystonia Coalition has received additional material or administrative support from industry sponsors (Allergan Inc. and Merz Pharmaceuticals) as well as private foundations (The American Dystonia Society, Beat Dystonia, The Benign Essential Blepharospasm Foundation, Cure Dystonia Now, Dystonia Inc., Dystonia Ireland, The Dystonia Medical Research Foundation, The European Dystonia Federation, The Foundation for Dystonia Research, The National Spasmodic Dysphonia Association, and The National Spasmodic Torticollis Association). M.M.R. was supported by the Interdisciplinary Center for Clinical Research (Z-3/64) of the University Hospital Wuürzburg and the German section of the International Federation of Clinical Neurophysiology.
Competing interests
The authors report no competing interests.
References
Author notes
Daniel T. Corp and Juho Joutsa authors contributed equally to this work.
