Botulinum toxin injection changes resting state cerebellar connectivity in cervical dystonia

In cervical dystonia, functional MRI (fMRI) evidence indicates changes in several resting state networks, which revert in part following the botulinum neurotoxin A (BoNT) therapy. Recently, the involvement of the cerebellum in dystonia has gained attention. The aim of our study was to compare connectivity between cerebellar subdivisions and the rest of the brain before and after BoNT treatment. Seventeen patients with cervical dystonia indicated for treatment with BoNT were enrolled (14 female, aged 50.2 ± 8.5 years, range 38–63 years). Clinical and fMRI examinations were carried out before and 4 weeks after BoNT injection. Clinical severity was evaluated using TWSTRS. Functional MRI data were acquired on a 1.5 T scanner during 8 min rest. Seed-based functional connectivity analysis was performed using data extracted from atlas-defined cerebellar areas in both datasets. Clinical scores demonstrated satisfactory BoNT effect. After treatment, connectivity decreased between the vermis lobule VIIIa and the left dorsal mesial frontal cortex. Positive correlations between the connectivity differences and the clinical improvement were detected for the right lobule VI, right crus II, vermis VIIIb and the right lobule IX. Our data provide evidence for modulation of cerebello-cortical connectivity resulting from successful treatment by botulinum neurotoxin.


Methods
The diagnosis of CD was determined following a comprehensive neurological examination by a movement disorders specialist, based on history of typical clinical symptoms for at least 12 months and polyelectromyographic examination of neck muscles. All subjects had a recent magnetic resonance imaging (MRI) of the brain with no structural abnormality. Each patient was informed in detail about the goal and the course of investigation and signed an informed consent form. The study protocol was approved by the local ethics committee (Ethics Committee of the University Hospital and the Faculty of Medicine and Dentistry of Palacký University Olomouc, Czech Republic), in accordance with the principles and recommendations of the Declaration of Helsinki, 1975 and later revisions.
Patients. Seventeen patients with CD indicated for treatment with BoNT were enrolled (14 female, aged 50.2 ± 8.5 years, range 38-63 years). Clinical and functional MRI examinations were carried out immediately before and 4 weeks after the first BoNT injection.
The severity of CD was evaluated using the Toronto Western Spasmodic Torticollis Rating Scale (TWSTRS) 8 at two sessions: at Week 0 (W0-on the day of the first fMRI examination immediately before the BoNT injection) and at Week 4 (W4-on the day of the second fMRI examination four weeks after the BoNT injection). The BoNT treatment was indicated and carried out according to national and international standards and there was no investigational treatment involved. In all patients, the injected muscles were determined on the basis of a polyelectromyographic examination, provided by 4-channel Keypoint workstation, Medtronic, Minneapolis, MN, USA. The details of the electromyographic examination and BoNT injection were described in our previous work 33 . All patients were treated with onabotulinum toxin type A (Botox,Allergan, Inc, Irvine, CA, USA in concentrations of 25 IU/ml. The demographic and clinical data of the patients are presented in Table 1. Data acquisition. The acquisition of MRI data was performed at W0 and W4 visits using 1.5-T scanners (Siemens Aera, Avanto, and Symphony, Erlangen, Germany) with standard head coils. To avoid any possible effects due to scanner used, the schedule was either matched or counter-balanced. To provide maximum comfort and minimize head motion, the patient's head was immobilized with cushions. During the acquisition, patients were asked to lie still with their eyes closed and not to think about anything in particular. The MRI protocol consisted of a functional T 2 * -weighted BOLD image acquisition using a gradient-echo echo-planar imaging (EPI) sequence with 30 axial slices parallel to the anterior commissure-posterior commissure line, 5 mm slice thickness, repetition time/echo time (TR/TE) = 2500/40 ms, flip angle 80°, field of view (FoV) = 220 mm, matrix 64 × 64, resolution 3.4 × 3.4 × 5.0 mm, 192 volumes. Furthermore, gradient-echo phase and magnitude field map images with identical geometry were acquired to allow correction of the B 0 imaging distortions. For anatomical reference, a high-resolution three-dimensional MPRAGE scan was also acquired.
Data Pre-processing. Initially, the fMRI data were checked for susceptibility or severe motion artifacts, but no subject had to be excluded. The statistical analysis of BOLD time-series was performed in FEAT Version 6.00, part of FSL (FMRIB's Software Library, www. fmrib. ox. ac. uk/ fsl), version 5.0.9 29 . The built-in pre-processing pipeline included: correction of B 0 distortions, motion correction using MCFLIRT, non-brain removal, and spatial smoothing using a Gaussian kernel with 8.0 mm FWHM. In each patient, affine registration matrices between the functional images from either session and a single anatomical reference image were obtained. The anatomical reference was chosen based on a quality assessment of both available T 1 -weighted scans. Finally, a non-linear registration of the anatomical image to the MNI 152 standard space using FNIRT was calculated 24  www.nature.com/scientificreports/ Next, residual motion-related artifacts were regressed out from functional time-series using ICA-AROMA noise component classification tool 48 and high-pass temporal filtering with sigma = 60.0 s was applied. In a parallel pre-processing pipe-line, the same ICA-AROMA noise components were regressed from data with no spatial smoothing applied. This unsmoothed dataset was used to extract both the signal of interest and non-neuronal nuisance signal.
Extraction of time-series. The time-series for seed-based connectivity were extracted using anatomically defined ROIs based on the FSL built-in Probabilistic Cerebellar Atlas parcellation with 50% probability threshold 15 , see Fig. 1, Panel F. At this threshold, mask of label 9 (Vermis Crus I) contained 0 voxels and was therefore discarded. Next, binarized masks were transformed into each individual's functional space using an inverted non-linear transformation field estimated in the pre-processing pipeline, resampled using trilinear interpolation and thresholded at 0.5. After the resampling, mask of label 15 (Vermis VIIb) contained 0 voxels in number of subjects and was discarded as well. The final set of ROIs consisted therefore of 26 regions. In each of the ROIs, the first eigenvariate was extracted and used as a representative de-meaned time-series. To provide a broader context for the changes in the cerebellar networks, we performed a supplementary exploratory connectivity analysis using several regions of interest in the cerebral cortex and basal ganglia selected a priori as a reference. The specific procedures of the supplementary analysis are described in Supplementary Methods, the remaining procedures were identical to the main analysis.
Additionally, nuisance signal from six sources in the white matter and one source in the lateral ventricles was obtained as described elsewhere 27 .
Statistical analysis of imaging data. The seed-based functional connectivity analysis was carried out using FILM 55 . For each ROI, a separate single-subject analysis was performed with a single regressor of interest and its temporal derivative to account for non-uniform slice timing and hemodynamic delay, 6 estimated motion parameters and 6 nuisance signal regressors from the white matter and 1 from the ventricles (see Extraction of Time-series section).
In the main group analysis, (1) we analyzed changes in functional connectivity, controlling for each individual's change in TWSTRS score from W0 to W4 (before and after BoNT-A) by including change in TWSTRS as a covariate; and (2) we also investigated the relationship between the change in functional connectivity with the change in TWSTRS score. To this end, pair-wise within-subject differences were modeled with an additional (explanatory) variable consisting of the corresponding within-subject changes in TWSTRS. The purpose of the (explanatory) variable was twofold: (1) to serve as a regressor (covariate) controlling for the individual variability in clinical effect when assessing the average changes in functional connectivity, and (2) to serve as a predictor of changes specifically linked to the clinical effect of BoNT-A. The second approach exploits the individual variability in treatment response, which is less likely to be affected by other factors that could bias the average differences. To incorporate both pair-wise differences and the TWSTRS change regressor/predictor, a two-step analysis was employed. Within-subject differences were first computed using a fixed effects analysis. The estimated beta values and variances were then carried over to the final mixed effects (fixed effects + random effects) analysis with group mean as the first regressor and the TWSTRS change as the second one. The final step thus  54 . The whole-brain analysis was limited to the MNI standard brain mask 24 minus a white-matter mask as described elsewhere 27 . The Z (Gaussianized T statistic images were thresholded using clusters determined by Z > 3 and family-wise error (FWE and Bonferroni corrected (accounting for the number of ROI and two contrasts per contrast pair cluster significance threshold was p < 0.00096 (calculated as 0.05 / [26 * 2]. In the exploratory evaluation of the non-hypothesized cortical and subcortical ROIs (supplementary analysis, we applied the same Bonferroni correction as in the main analysis (see Supplementary Methods, similar to the exploratory approach described by Delnooz et al. 12 .

Post-hoc plots.
To determine the type of change detected using the group contrasts, such as loss of positive or negative functional connectivity, gain of positive or negative connectivity, or both, a post-hoc analysis was performed using the masks of significant clusters in group analysis. First, mean Z scores were extracted from individual statistical maps using Featquery tool (part of FSL), which incorporates the back-transformation of the cluster masks from the standard into the individual functional space. Finally, the values were visualized as box plots (contrasts W0 > W4 and W4 > W0) or scatter plots with linear fit (contrasts Positive Correlation with TWSTRS Change and Negative Correlation with TWSTRS Change).

Post-hoc visualization of the average connectivity of cerebellar ROIs.
To determine the functional role of the ROIs with significant effects in the main analysis, a post-hoc connectivity analysis was performed. The average effects over the two sessions (W0 + W4) were pooled to visualize the functional networks associated with individual ROIs regardless of treatment. No Bonferroni correction was applied to the post-hoc statistical maps (mean connectivity) produced solely for visualization purposes. Additionally, to allow interpretation, the cluster-forming threshold was adjusted retrospectively to produce visually comparable co-activation levels across all functional connectivity maps.

Results
Clinical data. All patients were injected into the muscles identified by clinical evaluation and by polyelectromyography. The mean total dose of onabotulinumtoxin per patient was 161.8 ± 39.6 IU. The mean value of TWSTRS at Week 0 was 14.3 ± 3.1 and, at Week 4, it was 8.7 ± 3.2 (p = 0.00002, one-sided paired t-test), see Table 1. The significant decrease in TWSTRS four weeks after injections suggests a good clinical effect of BoNT treatment.
Imaging data. The first pair of contrasts (W0 > W4 and W4 > W0) revealed a significant decrease in connectivity (W0 > W4) between the vermis lobule VIIIa and the left dorsal mesial frontal cortex (Fig. 2, panel A). The functional connectivity changed from positive to negative after treatment (Fig. 2, panel B). No other ROIs showed average changes in functional connectivity after BoNT-A treatment. This result represents an average effect of the BoNT-A treatment after correcting for inter-individual differences in the clinical effect (determined as a change in TWSTRS). In other words, the functional connectivity decreased after treatment regardless of actual clinical improvement.
Changes in functional connectivity that were correlated with the individual differences in clinical effect were assessed using the second set of contrasts (Positive Correlation with TWSTRS Change and Negative Correlation with TWSTRS Change). Positive correlations between the connectivity differences (W0 > W4) and the clinical improvement (TWSTRS W0 > W4) were detected for the right lobule VI, right crus II, vermis VIIIb and the right lobule IX (Fig. 3). No significant negative correlation was observed after Bonferroni correction. All clusters of treatment-related differences and correlations (including those not reaching significance after correction) are summarized in Table 2.  Supplementary Table S1.
The post-hoc visualization of average connectivity (W0 + W4) for those ROIs with significant clusters in the main analysis is displayed in Fig. 1. Lobule VI was functionally associated predominantly with the bilateral premotor cortices, supplementary motor area (SMA), anterior cingulate cortex, and bilateral intraparietal sulcus; the right crus II, vermis VIIIb and right lobule IX showed mainly connectivity with the bilateral posterior cingulate cortex, precuneus, parietooccipital and mesial prefrontal cortex; vermis VIIIa showed mainly connectivity with the anterior cingulate cortex and the right temporoparietal junction (Fig. 1).

Discussion
In presented study, we investigated alterations in cerebellar functional connectivity and the possible modifying effect of BoNT injections in CD by means of resting-state fMRI and atlas-based parcellation.  www.nature.com/scientificreports/ It has been repeatedly suggested that CD is a network disorder 4,35,39,46 . Previous studies reported rather diffuse or non-overlapping structural 25,44,49 and functional 10,35,47 changes in multiple brain regions, including the sensorimotor cortices, basal ganglia, thalamus, and, more recently, cerebellum. In fact, it has been suggested that dystonia is caused by a combined dysfunction of several network nodes or their abnormal connectivity 39 . However, it remains unknown, which option describes the best the pathophysiology of dystonia 35 .
The role of the cerebellum in the pathophysiology of CD is still controversial as it is not clear whether functional changes in the cerebellum are the source or a consequence of dystonia 38 . Still, cerebellum showed consistent post-mortem histopathological changes across 6 patients with CD 45 and was one of the two brain regions specifically associated with secondary CD 9 . Although damage to the cerebellum is usually associated with negative symptoms (a loss of function), it can be argued that different kinds or localizations of lesions may produce different clinical presentations based on the affected pathways 38 . Abnormalities in the anterior lobe of the cerebellum may be associated with dysfunction of the cortical sensorimotor areas in dystonia 43 . Filip et al. 20 have found cerebellar dysfunction mainly localized in the left posterior hemisphere. They have also demonstrated that CD was associated with decreased functional connectivity between the left cerebellar lobule VI and the contralateral prefrontal cortex, and between the left cerebellar crus I and the ipsilateral middle temporal gyrus. However, the study by Filip et al. 20 analyzed cerebellar connectivity during a visuospatial task and only from a limited number of seeds (lobule VI, crus I and vermis VIIb), therefore the results are not directly comparable to ours.
In this study, we observed that resting-state connectivity from a number of cerebellar seeds in the posterior vermis and right posterior cerebellar hemisphere decreased with treatment. Moreover, the higher was the clinical benefit of the treatment, the larger was the decrease in functional connectivity (Fig. 3). The cerebellar seeds associated with these changes were found in the following areas: right lobule VI, right crus II, vermis VIIIa and VIIIb, and right lobule IX.
In agreement with previous data in healthy individuals 7 , the mean functional connectivity maps in Fig Among the observed treatment-related effects, one stood out: the connectivity from the vermis VIIIa to the left superior frontal gyrus decreased irrespective of the individual differences in the clinical outcome (Fig. 2). Superior frontal gyrus (BA6/9/10) has been reported as one of the regions showing higher resting-state functional connectivity in CD than in the control group, for a compound cortico-subcortical motor seed 12,35 or sensorimotor cortex 19 , and has also manifested different structural connectivity pattern 26 and increased gray matter volume 37 in other focal dystonias. Our treatment-related effect might be relevant, though possibly related to secondary phenomena following the BoNT treatment, e.g., behavioral or affective changes associated with motor improvement.
However, the remaining significant changes in connectivity scaled with the degree of clinical motor improvement. The change in TWSTRS was either correlated with the strength of intra-cerebellar functional connections, as in the case of seeds in the right lobule VI and right lobule IX, or with cerebro-cerebellar connections, as in the case of seeds in the right crus II and vermis VIIIb (Fig. 3).
The right lobule VI, which is predominantly associated with the sensorimotor control areas, showed in good responders decreasing connectivity with the right crus II (Fig. 3, panel B), a cerebellar region associated with the default mode network 7 . In contrast, the right lobule IX showed decreasing connectivity with the left crus II (Fig. 3, panel D). In this case, both regions are functionally associated with the default mode network 7 .
With respect to the cerebro-cerebellar connections, the clinical improvement (change in TWSTRS) was correlated with the functional connectivity strength between the right crus II and the right lateral prefrontal cortex, i.e., a component of the frontoparietal control network 53 . Furthermore, correlation with change in TWSTRS was observed between the vermis VIIIb and clusters in the right temporoparietal junction, left middle frontal gyrus, and right premotor cortex, which represent nodes of the ventral attention, frontoparietal control, and dorsal attention networks, respectively 53  In summary, BoNT injection followed by successful control of CD signs led to (1) decrease in intrinsic connectivity in the posterior cerebellum, including connectivity between motor and more cognitive regions of the cerebellum. Furthermore, (2) the vermis VIIIa and VIIIb, implicated in motor control, showed decreased connectivity with bilateral dorsolateral frontal and premotor and right temporoparietal cortices, and (3) the crus II, predominantly involved in cognitive functions, showed decreased connectivity with the right prefrontal cortex. www.nature.com/scientificreports/ The decreasing cerebro-cerebellar connectivity after treatment is in line with evidence for abnormally "increased" (i.e., less negative) connectivity in CD between the sensorimotor cortex and a cerebellar seed in vermis defined as an overlap of dystonia-related brain lesions 9 . Our findings also have to be considered in the context of previous published papers on central treatment effects in idiopathic dystonia e.g., 12,18,38 .
Our previous study showed decreased activation during complex finger tapping task throughout the sensorimotor system after BoNT treatment, including the anterior cerebellum 38 . Although this could be viewed as decreasing task-driven cerebro-cerebellar connectivity, thus compatible with our current results, a direct comparison of task-related and resting-state data is not straightforward. A previous resting-state fMRI study in CD patients that assessed BoNT effect revealed no difference in the cerebellar network 12 . However, cerebellar connectivity was assessed in that study on a global scale, de facto using a single cerebellar seed. Though reasonable in search for large effects, such approach could possibly miss changes on a fine scale of individual cerebellar lobuli.
Further evidence for the role of the cerebellum comes from structural imaging studies. In patients with various types of dystonia chronically treated with deep brain stimulation (DBS), voxel-based morphometry revealed increased gray matter density in the cerebellar vermis, supplementary motor area (SMA) and anterior cingulate cortex, which was more profound in good responders 18 . Together with our data, this study suggests a prominent role of the cerebellum and its frontal connections in the normalization of the motor function, though the exact mechanisms in DBS and BoNT may considerably differ.
It should be acknowledged that the mechanism of BoNT effect on central structures has not been fully elucidated. The established mechanisms are summarized by Marchand-Pauvert et al. 36 , including blockade of the gamma motor endings, plasticity evoked by blockade of the neuromuscular transmission, and retrograde transport and transcytosis of BoNT. Although it is not entirely clear, which of the mechanisms plays a key role in the clinical improvement after BoNT, there is multiple evidence from recent studies about modulation of activity of various structures in the central nervous system after BoNT both in motor and sensory areas. We suggest that in focal dystonias, BoNT-induced effects encompass complex mechanisms beyond chemodenervation of the injected muscles 5 .
In conclusion, the presented data provide evidence for modulation of cortico-cerebellar connectivity resulting from successful treatment by BoNT.