Motor training‐related brain reorganization in patients with cerebellar degeneration

Abstract Cerebellar degeneration progressively impairs motor function. Recent research showed that cerebellar patients can improve motor performance with practice, but the optimal feedback type (visual, proprioceptive, verbal) for such learning and the underlying neuroplastic changes are unknown. Here, patients with cerebellar degeneration (N = 40) and age‐ and sex‐matched healthy controls (N = 40) practiced single‐joint, goal‐directed forearm movements for 5 days. Cerebellar patients improved performance during visuomotor practice, but a training focusing on either proprioceptive feedback, or explicit verbal feedback and instruction did not show additional benefits. Voxel‐based morphometry revealed that after training gray matter volume (GMV) was increased prominently in the visual association cortices of controls, whereas cerebellar patients exhibited GMV increase predominantly in premotor cortex. The premotor cortex as a recipient of cerebellar efferents appears to be an important hub in compensatory remodeling following damage of the cerebro‐cerebellar motor system.


| INTRODUCTION
Cerebellar ataxia affects the coordination and control of gait, posture, upper limb movements, oculomotor function and speech. Ataxia results from focal lesions, such as a stroke, or from a progressive neurodegenerative process. While patients with cerebellar stroke frequently show a good recovery, degenerative cerebellar disease leads to a progressive loss of motor function. Despite several attempts, no drug treatment is currently available that ameliorates the symptoms of cerebellar ataxia (Ilg et al., 2014). Noninvasive and invasive brain stimulation methods have gained interest, but robust and reproducible effects of improving motor function have not been shown (Benussi et al., 2018;Hulst et al., 2017;Miterko et al., 2019). Antisense oligonucleotide therapy may be available in the near future, but will apply only for a subset of trinucleotide repeat disorders (Scoles & Pulst, 2019).
Currently, available treatment consists mainly of physical therapy (Ilg et al., 2014;Ilg & Timmann, 2012), but it has been questioned whether such therapy is a useful treatment given that the cerebellum itself is essential for implicit motor learning and such learning becomes impaired in cerebellar disease (Bastian, 2006;Saywell & Taylor, 2008;Thach & Bastian, 2004). However, recent evidence documented that motor training can improve motor function in patients with cerebellar degeneration (Burciu et al., 2013;Ilg et al., 2009;Keller & Bastian, 2014;Miyai et al., 2012). Yet, there is still a lack of physical rehabilitation training programs that take knowledge about cerebellar pathophysiology and the opportunities afforded by residual sensorimotor function into account.
With respect to training, it is well documented that sensorimotor learning involves the synergistic engagement of explicit and implicit learning processes Taylor, Krakauer, & Ivry, 2014). Explicit learning is often equated with strategic learning.
There is some evidence that cerebellar patients can make use of learning strategies during visuomotor reach adaptation (Taylor, Klemfuss, & Ivry, 2010). However, it is unknown, if additional explicit verbal feedback about movement errors and instruction on how to control for them may aid learning of a sensorimotor skill in patients with cerebellar degeneration. If cerebellar patients could indeed benefit from explicit verbal error feedback or knowledge-of-results during training, conventional physical therapy may incorporate this approach to yield better results.
Another aspect of motor learning, which has received little attention in the rehabilitation of degenerative ataxias, relates to the role of proprioception. Given the fact that the cerebellum receives massive proprioceptive afferents through the spinocerebellar tracts (Bloedel, 1973), and given the vital role of proprioceptive information for motor control, it becomes plausible that a training scheme with a focus on the proprioceptive cues could be of help for cerebellar patients (Aman, Elangovan, Yeh, & Konczak, 2015;Saywell & Taylor, 2008). There are reports that proprioception remains intact as cerebellar patients do not show abnormalities in passive position sense tasks, where the limb is passively moved (Bhanpuri, Okamura, & Bastian, 2012;Maschke, Gomez, Tuite, & Konczak, 2003). However, active position sense during voluntary movement becomes impaired in cerebellar degeneration (Bhanpuri et al., 2012), casting doubts, whether a proprioceptive-focused learning is still possible in patients presenting with degenerative ataxia. Thus, it remains an open question, if these patients can still effectively make use of proprioceptive information to guide motor learning. Finally, and equally important, the underlying neuroplastic changes during training associated with residual learning or compensatory forms of motor learning are only incompletely understood in patients with cerebellar degeneration.
To address these knowledge gaps, we designed a training regimen for a group of people with degenerative ataxia. The main goals of this study were threefold: First, to investigate if the effects of visuomotor training can be enhanced by providing additional explicit motor performance feedback. Second, to gain an understanding if the ability to use proprioceptive error feedback during motor learning is still intact in people with cerebellar degeneration. To that effect, we exposed patients to a training regimen without vision that purely relied on proprioceptive feedback. Third, to delineate the possible neuroplastic changes associated with such learning. We used neuroimaging before and after training and performed a voxel-based morphometry (VBM) analysis to obtain information on the cerebellar and extracerebellar neural correlates of such training.

| Participants
A total of 41 patients with cerebellar degeneration and 44 neurologically healthy controls participated in the study. One patient and two controls dropped out prematurely because of acute illness unrelated to the study. Two controls had to be excluded from analysis because of incidental findings on brain magnetic resonance imaging (MRI).
Hence, data from 40 patients (mean age 55 ± 11.4 years, 19 males) and 40 sex-and age-matched controls (mean age 55.9 ± 10.9 years, 20 males) were included for analysis. Pretraining behavioral and MRI data of 30 patients and 30 controls of the present study population has been reported in a previous study by our group (Draganova et al., 2021).
All patients were diagnosed with a pure form of cerebellar cortical degeneration, primarily as spinocerebellar ataxia type 6 (SCA6), autosomal dominant cerebellar ataxia type 3 (ADCA III), and sporadic adult-onset ataxia (SAOA) of unknown etiology. The severity of ataxia was assessed by the clinical Scale for the Assessment and Rating of Ataxia (SARA; Schmitz-Hübsch et al., 2006). Patients and matched controls were pseudorandomly assigned to one of four training conditions (see below). The four subgroups of patients were matched for sex, age, and clinical ataxia rating (SARA) scores. Characteristics of individual patients and matched controls are detailed in Table 1. All participants were right-handed as assessed by the Edinburghhandedness scale (Oldfield, 1971). The study was approved by the Ethics Committee of the Essen University Medical Center. Oral and written informed consent was obtained from all participants prior to testing.
T A B L E 1 Clinical characteristics of cerebellar patients and matched controls. Patients and controls are grouped based on their assignments to the four training subgroups. Severity of ataxia was rated using the SARA (range SARA score 0-40; maximum SARA score = 40; Schmitz-Hübsch et al., 2006). SCA 6, 8, 14 = spinocerebellar ataxia Types 6, 8, and 14; SAOA = sporadic adult-onset ataxia; ADCA III = autosomal dominant cerebellar ataxia type III (pure cerebellar type); EOCA = early onset cerebellar ataxia. All patients suffered from cerebellar degeneration, and all patients presented with a pure cerebellar phenotype. Subject IDs refer to the order the patients and controls were recruited and assigned to the respective subgroups. For details on training conditions, see

| Apparatus
Participants performed elbow flexion movements in the horizontal plane employing a one degree of freedom single-joint manipulandum as described in Draganova et al. (2021).
The manipulandum allowed the execution of precise goal directed movements without the need of the user to compensate for gravity ( Figure 1).  All participants received information about the movement goal (i.e., the target) either visually, and/or as verbal feedback (i.e., "on target," "target has not been reached") when vision was blocked.
Online proprioceptive feedback was always available.

| Training sequence
Training was performed on five consecutive days. Targets were at

| Assessment of training
On the days before and after training, motor performance was measured similar to the Vision Only condition. That is, vision was available, but no explicit forms of movement feedback were given, to allow direct comparison of motor performance outcomes between the four training conditions. Targets had a width of 0.3 , and were at 10, 25, and 50 of elbow flexion. Participants performed 10 trials per target amplitude (30 trials total). Targets were presented in pseudorandom order with maximal two consecutive movements of the same target amplitude. The order of targets was the same across all participants.
T A B L E 2 Summary of the four training groups. Feedback types were different for each training condition. Online proprioceptive feedback was available during motion, either with or without explicit feedback. During the conditions with vision, learning was driven by visual and proprioceptive inputs. No vision conditions constituted forms of proprioceptive training. Information about the movement goal (i.e., whether target has been reached or not) was available either through vision (condition 1, 2), or through explicit verbal feedback (condition 3, 4). Verbal feedback about the magnitude and direction of the movement error was provided after movement execution only in the "+ Exp Feedb" conditions (e.g., "Target was undershoot by xx degrees. Increase movement by x degrees.")

| Measurements
The calculation of the behavioral parameters is described in detail in Draganova et al. (2021) and is reproduced below. Data from the optical encoder were processed offline by custom written software in MATLAB (MathWorks, Natick, MA). For each trial the absolute JP error (JPE) and the relative JPE (RJPE) with respect to each target amplitude was computed. RJPE was calculated based on the instantaneous JP and JPE over a period of 408 samples, which corresponded to the 4 s holding period of the arm (i.e., after the transport phase of the movement was completed): where N = 408 is the number of sampling points covering 4 s holding period when the target was reached, TA is the corresponding amplitude 10, 25, or 50 . RJPE was expressed in percentage of the amplitude (i.e., RJPE was multiplied by 100). In addition, the peak velocity (V max ) during the transport phase of the pointing movement was determined.
All movement trials were visually inspected for data integrity prior to inclusion in the analysis. In the pretraining and post-training assessments, one trial each had to be excluded in four controls and three patients because of technical errors. In the training sessions, no trials had to be excluded. For each participant and target amplitude, means (M) and SDs of RJPE were calculated on each day (pre, post, and the five training days).

| Statistical analysis
The primary outcome parameter was the difference in mean RJPE (RJPE M for each of the movement amplitudes before and after training. Secondary outcome parameter was the SD of RJPE (RJPE SD) calculated for each of the movement amplitudes. In addition, trainingrelated changes of RJPE M and RJPE SD were assessed across the 5 days of training.
The data for RJPE M and RJPE SD were not normally distributed.
Therefore, they were modeled and analyzed using the nonparametric rank-based analysis of variance (ANOVA)-type test for factorial longitudinal data using the statistical software packages nparLD (http:// www.R-project.org/) and SAS (Domhof & Langer, 2002;Noguchi, Gel, Brunner, & Konietschke, 2012). The underlying treatment effects are so-called relative effects, also known as Wilcoxon-Mann-Whitney effects p X = P(X < Y), where X denotes the factor level of interest and Y denotes the fixed reference (mean) distribution. The effects display the order of the data across all groups: If p X < p Z , then the data under condition X tends to be smaller than those measured under condition Z. If p X = p Z , then none of the data under conditions X and Z tend to be smaller or larger. The nonparametric rank-based method allows reliable conclusions when sample sizes are small. Since the procedure is solely based on ranks of the data, presence of outliers do not affect the outcome. 2.5.1 | Voxel-based morphometry VBM analysis was performed considering the whole brain using the MNI normalization procedure (Ashburner & Friston, 2000 uk/spm). The post-training scans were registered to the baseline (pretraining) scans for each participant separately. A mean of the realigned images was generated for each participant and used for bias correction for field inhomogeneity between the different time points. Based on the segmentation (tissue classification) of the mean image, using tissue probability maps, tissue was classified into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF). Using two transformations, linear (12 parameters affine) and nonlinear transformations (warping), the mean image was registered to match a standard template (DARTEL) within a unified model (Ashburner & Friston, 2005).

Considering RJPE M and RJPE
The spatial normalization parameters estimated during this step resulted in spatial deformation fields. The latter were applied to the GM segmentations of the images of both time points (pretraining and post-training). To correct the volume changes after spatial normalization, GM density segments were modulated by the Jacobian determinants as derived from the spatial normalization's deformation parameters (Kurth, Thompson, & Luders, 2018 Burciu et al., 2013;Taubert et al., 2010). Whole brain analysis was performed and results are reported using an exploratory, uncorrected threshold of p = .001. To partially correct for multiple comparisons the expected voxels per cluster (<k>) calculated based on random field theory in SPM12 were used as a cluster size threshold (cf. Burciu et al., 2013).
Assignment of peak MNI coordinates and brain clusters to brain areas was done using the Julich-Brain Cytoarchitectonic Atlas (JuBrain), based on the maximum probability map (MPM) (Eickhoff et al., 2007). In case, the respective brain area was not part of the current JuBrain atlas the automated anatomical labeling (AAL3) atlas was used instead (Rolls, Huang, Lin, Feng, & Joliot, 2020;Tzourio-Mazoyer et al., 2002).
Next, significant brain cluster regions considering the main effects     Table 4. Group mean data are given in  Table 4.  In controls, on the other hand, GM increase was observed primarily within the right cuneus (including areas hPO1 and hIP8 in the posterior intraparietal sulcus with some extension to area 7P in the T A B L E 3 Summary of statistical results considering pretraining and post-training assessments. Nonparametric rank-based ANOVA-type tests for factorial longitudinal data were applied. Degrees of freedom were adjusted in case variances differed  To analyze whether GMV increases were correlated with improved motor performance, we asked the question how many cerebellar patients and how many controls showed both an increase of GMV and improved RJPE M in the three VOIs. Increase in GMV was defined as a difference of GMV in a given VOI that was larger at posttraining when compared to pretraining. No threshold was used. As shown in Figure 6a,d, we found that 55% (  Abbreviations: ANOVA, analysis of variance; RJPE, relative joint position error.

| Differences in GMV between cerebellar patients and controls at baseline
Finally, we were interested if GMV differences existed in cerebellar patients and controls already at baseline. As expected, patients with cerebellar degeneration exhibited smaller GMV in most parts the cerebellar cortex (Figure 7a,b). Differences were most marked in anterior cerebellar lobe and adjacent lobule VI, as well as in the vermis (threshold: p FWE < .05; see also Table 5 Table 6).

| DISCUSSION
This study investigated to what extent upper limb motor learning is preserved in cerebellar degeneration. We systematically investigated, if the provision of explicit verbal feedback could "boost" the learning outcome for cerebellar patients. In addition, we examined if such motor learning was driven primarily by visual feedback, or if these patients were still able to use proprioceptive information as an error feedback signal. The three main findings of the study are the following: First, explicit verbal feedback did not enhance visuomotor learning in the cerebellar patient group. Second, in our sample of patients, who presented with mild to severe degenerative ataxia, proprioceptive-based motor learning was preserved. Third, as a neural correlate of motor learning the control group exhibited an increase in GMV (GMV) most prominently in visual association cortices, while motor learning in the cerebellar patients was associated with a GMV increase in premotor cortex. Results corroborate previous findings of our group in a balance training task in patients with cerebellar degeneration (Burciu et al., 2013), and suggest that compensatory remodeling primarily takes place in those cerebral motor areas that receive strong efferent projections from the cerebellum.

| Explicit verbal feedback does not aid motor learning in cerebellar disease
Different to our expectation, additional explicit verbal feedback about movement errors and instruction on how to control for them, did not lead to superior learning neither in patients with cerebellar degeneration nor in controls. This finding is at odds with earlier reports showing that cerebellar patients can use explicit information to apply cognitive strategies during visuomotor adaptation to minimize movement error (Taylor et al., 2010;Wong, Marvel, Taylor, & Krakauer, 2019). In one experiment (Taylor et al., 2010), cerebellar patients received incongruent visual feedback (cursor and physical hand position were shifted by 45 ) and successfully applied a À45 F I G U R E 5 Training-related increases of gray matter volume (GMV) in patients with cerebellar degeneration (Cer) and healthy controls (Con). (a) t-Contrast pretraining < post-training in the group of all cerebellar patients, (b) t contrast pretraining < post-training in the group of all healthy controls. (c) F contrast of the interaction time (pretraining < post-training) and group (cerebellar vs. controls). Voxelbased morphometry (VBM) data are shown at an exploratory threshold of p < .001 overlaid on the mean smoothed gray matter (GM) segmentation image in all cerebellar patients (a), all controls (b), and all patients and controls (c). In each panel, VBM clusters are shown superimposed on coronal, sagittal and axial sections (upper part), as well as axial sections (lower part), the latter being the same in (a-c) for direct comparison. Small inserts show mean GMV and SEs in the largest cluster of the given contrast assessed in individual cerebellar patients and controls pretraining and post-training strategy to compensate for the visual error. That is, they were able to use a cognitive strategy. In our case, participants did not have to adapt to an external perturbation. They received explicit verbal feedback about the magnitude and the direction of the movement error to optimize motor outcome in a skill (i.e., pointing as precisely as possible). This explicit feedback augmented existing visual and proprioceptive feedback about the arm position. The fact that neither controls nor cerebellar patients benefitted from this feedback suggests that this feedback was redundant, as it did not enhance performance. It also implies that cerebellar patients cannot easily substitute possible deficits in visual (Maschke, Gomez, Tuite, Pickett, & Konczak, 2006) or proprioceptive perception (Bhanpuri et al., 2012) through the use of explicit verbal error feedback to improve motor performance. It is plausible that the implicit, cerebellar-dependent components of learning prevailed (Kim, Ogawa, Lv, Schweighofer, & Imamizu, 2015;McDougle, Bond, & Taylor, 2015;Smith, Ghazizadeh, & Shadmehr, 2006 Note: Results of whole brain analysis reported at an exploratory, uncorrected threshold of p = .001, partially corrected for multiple comparisons using predetermined cluster sizes (<k>, expected voxels per cluster). k E = voxels per cluster; MNI coordinate = Montreal Neurological Institute coordinates. Abbreviations: GMV, gray matter volume; MPM, maximum probability map; SPL, superior parietal lobe. a AAL3 atlas labels (Rolls et al., 2020): Frontal_Sup = superior frontal gyrus, dorsolateral; Supp_Motor_Area = supplementary motor area; Post_central = postcentral gyrus; Cuneus = cuneus; Frontal_Mid = middle frontal gyrus; Cerebellum_Crus1 = Crus I of cerebellar hemisphere. b JuBrain atlas labels (Eickhoff et al., 2007): area 6d2, area 6d3 = dorsolateral premotor areas; 6mr/preSMA = supplementary motor area; area 2 = primary sensory cortex; areas 7PC, 7P = areas the SPL; areas hPO1, hIP8 = areas in the posterior intraparietal sulcus.
c Probabilities for all histological data found at the position of this voxel (Eickhoff et al., 2007; see also https://www.fz-juelich.de/SharedDocs/Downloads/ INM/INM7/EN/SPM_Toolbox/Manual.pdf?__blob=publicationFile). d Relative extent (i.e., percentage) of cluster assigned to a cytoarchitectonic area based on the cytoarchitectonic MPM (Eickhoff et al., 2007). Comparing all patients with cerebellar degeneration and all healthy control participants. The t contrast cerebellar (Cer) patients < controls (Con) is shown in blue/ green colors and the t contrast Con > Cer in red/yellow colors. Significant differences are shown (a) superimposed on coronal, sagittal, and axial sections of the whole brain map (calculated as mean of cerebellar and control group gray matter [GM] images), and (b) superimposed on a flat map of the cerebellum (Diedrichsen & Zotow, 2015) at a threshold of p < .

| Training-related GM increases in premotor cortex
The most important finding of the present study was that training effects in cerebellar patients were related to GM increases primarily within the premotor cortex. This result is in good accordance with a previous study of our group, which also found training-related GM increases in the premotor cortex in a postural training task (Burciu et al., 2013). Premotor cortex is involved in the generation of motor plans based on visuospatial information from the parietal cortex, and in motor learning (Hardwick et al., 2015;Mazurek & Schieber, 2017). Both the present arm movement task and the previous balance task involved movements to visual targets, in the latter by moving the center of gravity on a force platform. Knowing that premotor cortex receives large efferent projections from the cerebellum (Bostan, Dum, & Strick, 2013), a GM increase in premotor cortex may be an attempt of the cortical targets of cerebellar output to compensate for the altered cerebellar signals. This implies that GMV increase in premotor cortex constitutes a neurostructural response to altered efferent input. From a computational motor control perspective, it may be viewed as an attempt by the system to maintain network integrity. Given that cerebellar output signals modulate motor planning, a GMV increase associated with cerebellar degeneration could be understood as an effort to add neural resources to interpret increasingly noisier cerebellar efferent signals and to use such signals for motor planning and learning (e.g., as a predictive or error feedback signal). Predictive ability is not limited to the cerebellum (Sokolov, Miall, & Ivry, 2017) and may also be a function of premotor cortex (Stadler et al., 2012). Premotor cortex has also been reported to compensate for M1 lesions due to stroke for upper limb (for review, see Kantak, Stinear, Buch, & Cohen, 2012) and gait function (Miyai et al., 2002). Thus, premotor cortex appears to be an important hub in compensatory remodeling following damage of the motor system.
The present findings may be relevant for clinical practice. For example, premotor cortex may be a possible target for noninvasive brain stimulation (NIBS) to enhance the effects of motor training in cerebellar patients. So far, studies using NIBS focused on stimulating the cerebellum, M1, or the spinal cord (Benussi et al., 2017;Benussi et al., 2018;Hulst et al., 2017).
In addition to premotor cortex, we saw learning-related GM increase in the SMA. The SMA is part of the basal ganglia circuit involved in motor learning, and there is recent evidence that the cerebellum has direct anatomical connections not only with PM, but also with SMA (Bostan et al., 2013). In good agreement with the present study, dynamic causal modeling analysis of fMRI data showed that the SMA was involved in motor learning in patients with degenerative cerebellar disease (Tzvi et al., 2017). One may assume that those cere-  (Burciu et al., 2013), parts of the cerebello-corticalmotor-loop unaffected by the disease showed the most significant learning-related GM increase. Note: Cerebellar patients < controls: results of whole brain analysis corrected at a corrected threshold of p < .05, FWE corrected; cerebellar patients > controls results of whole brain analysis reported at an exploratory, uncorrected threshold of p = .001, partially corrected for multiple comparisons using predetermined cluster sizes (<k>, expected voxels per cluster). k E = voxels per cluster; MNI coordinate = Montreal Neurological Institute coordinates. Abbreviation: MPM, maximum probability map.
The pattern of learning-related GM increases was very different in controls, where most learning-related changes were found in visual associative areas. The same area has been found to show changes in regional brain morphology related to learning of a more complex visuomotor task, that is juggling, in healthy participants by Scholz, Klein, Behrens, and Johansen-Berg (2009) and Draganski et al. (2004).
Finally, similar to our previous study, cerebellar GMV increase related to motor learning was absent in cerebellar patients and scant in controls. This does not exclude that learning-related plastic changes take place within the cerebellum in cerebellar patients, which has been shown in histological data of training studies in mouse models of cerebellar degeneration (e.g., Fucà et al., 2017).

| Limitations
The main limitation of our study is weak statistical power. Although the total number of patients (N = 40) was comparatively large, there were only 10 participants per training group given the betweensubject design. We cannot exclude that differences between training conditions become obvious in larger patient populations. Furthermore, imaging data analysis was only partially corrected for multiple comparisons. However, because our main finding of training-related GMV increase in cerebellar patients' premotor cortex are supported by previous data of our group (Burciu et al., 2013), we believe that our current findings are valid.
Distribution of diagnoses differed between training subgroups (Table 1). However, great care was taken to enroll only patients with a pure form of cerebellar cortical degeneration. Thus, although the etiology differed between patients, the underlying pathology was the same. Furthermore, we assured that there was significant overlap between groups with all groups including patients with SCA6, ADCAIII, and SAOA. Therefore, it is unlikely that the comparatively small differences in distribution of diagnoses between subgroups had a significant impact on our results.
Another limitation is training duration. Five days is a comparatively short training duration. Effects of the training conditions may become more pronounced with longer training duration. Finally, elbow flexion is a relatively simple, single-joint movement. Because arm ataxia becomes more pronounced in multijoint movements (Bastian, Martin, Keating, & Thach, 1996), findings need to be validated in future studies using more complex movements.

| CONCLUSIONS
Our data confirm that patients with cerebellar degeneration still benefit from motor training. We found no evidence that providing additional explicit verbal feedback effectively "boosts" sensorimotor learning. Consequently, there is still a need to further understand under which conditions cerebellar patients may benefit from explicit movement instructions. In contrast, the same patients effectively used proprioceptive information for motor learning when vision was blocked. Most importantly, our data provide additional evidence that premotor areas are involved in compensatory processes in cerebellar disease. Future studies are needed to understand to which extent premotor cortex can functionally compensate for cerebellar dysfunction.