Differential effects of anodal and dual tDCS on sensorimotor functions in chronic hemiparetic stroke patients

BACKGROUND AND PURPOSE
Previous tDCS studies in chronic stroke patients reported highly inconsistent effects on sensorimotor functions. Underlying reasons could be the selection of different kinematic parameters across studies and for different tDCS setups. We reasoned that tDCS may not simply induce global changes in a beneficial-adverse dichotomy, but rather that different sensorimotor kinematics are differentially affected. Furthermore, the often-postulated higher efficacy of bilateral-dual (bi-tDCS) over unilateral-anodal (ua-tDCS) could not yet be demonstrated consistently either. We investigated the effects of both setups on a wider range of kinematic parameters from standardized robotic tasks in patients with chronic stroke.


METHODS
Twenty-four patients with arm hemiparesis received tDCS (20min, 1 mA) concurrent to kinematic assessments in a sham-controlled, cross-over and double-blind clinical trial. Performance was measured on four sensorimotor tasks (reaching, proprioception, cooperative and independent bimanual coordination) from which 30 parameters were extracted. On the group-level, the patterns of changes relative to sham were assessed using paired-samples t-tests and classified as (1) performance increases, (2) decreases and (3) non-significant differences. Correlations between parametric change scores were calculated for each task to assess effects on the individual-level.


RESULTS
Both setups induced complex effect patterns with varying proportions of performance increases and decreases. On the group-level, more increases were induced in the reaching and coordination tasks while proprioception and bimanual cooperation were overall negatively affected. Bi-tDCS induced more performance increases and less decreases compared to ua-tDCS. Changes across parameters occurred more homogeneously under bi-tDCS than ua-tDCS, which induced a larger proportion of performance trade-offs.


CONCLUSIONS
Our data demonstrate profound tDCS effects on sensorimotor functions post-stroke, lending support for more pronounced and favorable effects of bi-tDCS compared to ua-tDCS. However, no uniformly beneficial pattern was identified. Instead, the modulations varied depending on the task and electrode setup, with increases in certain parameters occurring at the expense of decreases in others.

Stroke constitutes a leading cause of acquired disability in higher age [1] as the recovery of cognitive 57 and motor functions remains incomplete for most patients [2], impairing their life quality [3]. It is, 58 therefore, imperative to explore new avenues to improve rehabilitation after stroke. 59 Transcranial direct current stimulation (tDCS) is a technique that is safe to administer in patients 60 [4] and is increasingly applied as neuromodulatory adjuvant to neurorehabilitation. Unilateral-anodal 61 tDCS (ua-tDCS), aiming at the facilitation of the ipsilesional sensorimotor cortices, is the most 62 extensively studied setup and has been shown to improve a variety of motor outcomes measures in 63 stroke patients [5]. Similarly, unilateral-cathodal tDCS has been reported to effectively modulate 64 performance by inhibiting overactive contralesional cortices [6]. Dual or bilateral tDCS (bi-tDCS) 65 combines the facilitation of the ipsilesional cortices (anodal component) with the inhibition of the 66 contralesional cortex (cathodal component) and has thereby been shown to induce stronger effects 67 than either unilateral setup [7]. Other studies, however, did not demonstrate these canonical [8] 68 performance modulations for either setup [9-11] and recent meta-analyses and systematic reviews 69 They were briefed about study procedures and intent and provided written informed consent in 93 accordance with the Declaration of Helsinki. Twenty-four patients (16 males, mean age: 60.2 ± 12.4 94 years) with first stroke occurrence (13 right-affected) and mild to moderate upper extremity 95 hemiparesis at least 6 months post-stroke were recruited.  Table I) were acquired by experienced staff. Enrolled 105 patients were familiarized with the experimental setup, equipment and tested all kinematic tasks 106 prior to data collection. The effects of stimulation were assessed in three test sessions, each 107 separated by at least one week to avoid carry-over effects of tDCS [16]. The experimental procedures 108 and kinematic assessments were identical in each session apart from the applied tDCS setup ( Figure  109 1.B). The order of sessions (tDCS setups) and kinematic tasks was pseudo-randomized across 110 patients: respective randomization lists were generated a priori in MATLAB 9.3. (R2017b, 111 MathWorks, Inc., Natick, MA, USA) and patient IDs were filled successively when a new patient was 112 enrolled. Neuroimaging data were acquired for half of the sample before and the other half after the 113 behavioral assessments. All data collection was conducted by the same blinded experimenter. A 114 second experimenter allocated patients to the randomization lists, supervised the stimulator during 115 testing and assisted with preparations but did otherwise not interact with patients. 116 < Insert Figure 1 here > 117

Kinematic Assessments 119
All kinematic assessments were performed on a second-generation KINARM exoskeleton lab (BKIN 120 Technologies, Canada, Figure 1.C) using Dexterit-E 3.6.2 [17]. KINARM can initiate upper extremity 121 movements and reliably record in-plane movements in high temporal and spatial resolution [18,19]. 122 Sessions (Figure 1.B) commenced with a questionnaire on life-style variables and visual 123 analogue scales for attention, wakefulness and pain. After tDCS preparations, the exoskeleton was 124 adjusted to fit individual limb segment lengths and the system was calibrated. Adjustment and 125 calibration parameters were recorded to avoid a measurement bias between sessions (tolerance: ± 126 1cm). Task instructions were repeated in a standardized way to maximize compliance. were not performed as those were beyond the focus of this paper. 174 To control the quality of the repeated measures design and to avoid potential confounders 175 for tDCS effects, lifestyle variables like individual circadian rhythm, sleep, physical activity, substance 176 consumption as well as head measurements, electrode resistances, blinding and changes in visual 177 analogue scales for levels of attention, wakefulness and pain were acquired. Detailed descriptions 178 and results are provided in Supplemental Material Section 2. In short, there were no significant 179 differences regarding lifestyle variables, electrode placement parameters or levels of attention, 180 wakefulness and pain -including their change within sessions -between tDCS setups. No tDCS side-181 effects were reported. 182

Data Analysis 185
Preprocessing 186 To quantify the modulation of sensorimotor performance due to tDCS, session means across all trials 187 were calculated for all kinematic parameters of each task using Dexterit-E. To comprehensively 188 assess overall sensorimotor performance, all parameters that met the following criteria were to be 189 included into statistical analysis: parameters were required to (i) provide information which was not 190 otherwise covered by other parameters and (ii) changes in parameters were required to be 191 identifiable as performance increases or decreases. Accordingly, a subset of 30 parameters across 192 tasks was selected a priori.

t-Tests 214
To assess the setup-specific effect patterns, differences between the two setups and sham 215 were calculated using paired t-tests. For each parameter, t-test results were classified as (1) 216 performance increases, (2) decreases or (3) non-significant changes compared to sham. Frequencies 217 for each category were counted task-wise and across all tasks. Permutation tests (5000 iterations) 218 were performed to control for falsely positive statistics [39]. The permutations were synchronized 219 across tests to account for possible collinearities between parameters. Refer to Supplemental 220 Material Section 3 for parameter directionality interpretations, statistical assumption tests and 221 permutation procedures. The ratios between performance increases and decreases under 222 stimulation over the total amount of significant parameters was compared using Fisher's exact tests. 223 To assess a superiority between the two setups directly, paired t-tests were calculated for 224 the performance under ua-tDCS vs. bi-tDCS analogously to the procedure described above. These 225 differences do not account for baseline performance. As superior performance could either mean 226 actual better performance or less detrimental performance, caution is warranted when interpreting 227 these results. 228

Inter-Parameter Correlations 230
To further investigate the pattern of changes on the individual-level, relationships among changes 231 between parameters were assessed task-wise. Change scores from sham to stimulation were 232 calculated for ua-tDCS and bi-tDCS on each parameter using the formula [40] △% stimulation = 233 (performance sham -performance stimulation ) / performance sham * 100 in SPSS. Change scores were 234 correlated across all parameters for each task using Kendall's Tau. Resulting positive coefficients 235 were considered mutual performance increases if most (> 50% of cases) underlying change scores 236 were △% > 0, and mutual decreases if most were △% < 0. As negative coefficients indicate that an 237 increase in one parameter comes at the "cost" of decreases in another, those were considered inter-238 parametric trade-offs ( Figure  5.B). 239  Permutation tests revealed that the total number of significant comparisons was always 257 larger than the number of corresponding significant comparisons (both regardless of the direction of 258 the difference) obtained after permuting (corresponding to p < .001). 259

240
The superiority comparisons indicated no significant difference in ratios between the two 260 setups (p = .379). 261

275
The present study investigated online effects of ua-tDCS and bi-tDCS on paretic arm performance 276 across different sensorimotor tasks testing reaching movements, proprioceptive performance and 277 bilateral coordination in chronic stroke patients. The data demonstrate considerable modulations of 278 sensorimotor functions which, however, did not occur uniformly across either the four robotic tasks 279 or their kinematic parameters: the overall group-level effects were task-dependent with a higher 280 proportion of performance improvements in the unimanual reaching and independent bimanual 281 coordination and deteriorations in the proprioception and cooperative bimanual coordination tasks. 282 Moreover, effects were different between setups. While there were more parameters affected 283 during bi-tDCS, a direct comparison using Fisher´s exact test did not exhibit significant results but 284 conveyed stronger performance increases during bi-tDCS as compared to ua-tDCS. At the individual-285 level, correlations revealed that tDCS-induced changes in single parameters are complexly related to 286 changes in the other parameters of a respective task and that these change patterns are, task-and 287 setup-specific rather than uniform, contrary to the postulated canonical effects of tDCS. Indeed, our 288 results convey the notion that tDCS in stroke patients induces more complex sensorimotor changes 289 when investigated at a kinematic level: an increase in one parameter might come at the expense of 290 decrease in other (trade-off). Some authors suggested previously that behavioral enhancement 291 induced by non-invasive brain stimulation might the consequence of resource allocation and 292 therefore be accompanied by deterioration in other functions. Indeed, our data convey the notion 293 that this phenomenon -at least partly -might play a role in patients with focal lesions that exhibit 294 even more constrained neural resources. While these results show how intricately specific kinematic 295 aspects of a sensorimotor task are interrelated, they also evince how the selection of outcome 296 parameters can strongly influence the evaluation of tDCS effects. Thus, our findings both confirm and 297 unify the heterogeneous evidence from previous studies and provide a new perspective for the investigation of sensorimotor tDCS effects. On the one hand, our findings are promising for the 299 rehabilitation of sensorimotor deficits after stroke, if they can be translated into training protocols 300 supplemented by tDCS that induce similar and long-lasting effects. On the other hand, the data 301 demonstrates that the optimal tDCS setup will probably have to be chosen individually together with 302 a personalized prioritization of the kinematic aspects to be trained. To make such a personalization 303 reliable and accurate, additional multi-parametric studies will be necessary. 304 We deployed the multivariate t-test approach for a more detailed description of group-level 305 effect patterns as compared to, for instance, multiple analysis of variance or machine learning 306 classifiers (Supplemental Material Section 3.3). 307 Altogether, both setups induced substantial performance changes on the group-level, 308 confirming our first hypothesis. Concretely, bi-tDCS exhibited both a more pronounced, as more 309 parameters were significantly and more strongly changed irrespective of their directionality, and 310 more beneficial effect than ua-tDCS, as more performance increases and less decreases occurred. 311 These findings confirm our second hypothesis and the general trend in previously published studies 312 [7, 41-43], although performance decreases in the magnitude observed in our data have not been 313 previously reported for either setup. 314 The four robotic tasks investigated different sensorimotor functions and were task-and setup-315 specifically affected: while the highest proportional increases were identified for VGR and IBC, APM 316 and CBC were mainly negatively affected by both setups. Bi-tDCS exhibited a more beneficial effect 317 pattern (i.e., more increases, less decreases) in all individual tasks when compared to ua-tDCS 318 directly. 319 The simultaneous investigation of multiple parameters per task allowed us to address their 320 interdependence on the individual-level at which task-and setup-specific group-level results were showed no significant modulations on the group-level, underscoring the large inter-individual 323 variability of tDCS effects [44] that can be difficult to capture with conventional mean-based 324 approaches. A more homogeneous modulation across parameters was induced by bi-tDCS, as more 325 positive relationships were identified in all tasks. However, the negative correlations will be of 326 particular interest for future investigations as they can be interpreted to reflect trade-offs between 327 parameters: for instance, a more accurate aiming towards a target (initial direction angle) could 328 come at the "cost" of higher reaction times in the movement planning phase of VGR. If in an 329 individual patient accuracy is considered the primary rehabilitation outcome, such trade-offs in 330 reaction times would therefore unlikely be considered a negative outcome. As a higher proportion of 331 trade-offs occurred for ua-tDCS, the more pronounced group-level performance decreases should 332 therefore only be cautiously interpreted as true performance deteriorations. An alternative 333 interpretation could be that ua-tDCS induces less homogenous effects with more performance shifts 334 between kinematic parameters while bi-tDCS induced rather homogeneous modulations. However, if 335 only trade-offs had occurred across parameters, the overall effects of tDCS would not be considered 336 substantial but rather represent a mere net-zero-sum shift across parameters [45]. However, this can 337 be excluded based on the substantial proportion of positive correlations. Taken together, these 338 results confirm our last hypothesis of complex (i.e., non-uniform) and inter-related effects across 339 task parameters that were specific to the investigated robotic tasks. 340 As no other multi-parametric study combining tDCS and robotic sensorimotor assessments in 341 stroke patients has been performed, it is difficult to relate our findings to previous work. Only three 342 studies compared the two setups directly in single clinical trials yet: Mahmoudi et al. [43] showed 343 beneficial effects of ua-tDCS and bi-tDCS on the Jebsen- Taylor  difficulty. Therefore, they likely engage different neural processes involving distinct brain networks. 380 For example, VGR is a task that tested unilateral visually guided reaching movements of the affected 381 limb. The IBC task is a bilateral task, in which both affected and unaffected arm are required to 382 perform reaching movements, with both limbs operating individually in an alternating fashion. While 383 the CBC task is also a bilateral task, here by contrast to the IBC, both limbs cooperate simultaneously 384 in order to achieve the goal. Therefore, this task requires simultaneous coordinated bilateral 385 movements and might have a stronger dual-task performance component compared to IBC and 386 therefore a stronger cognitive load. The APM, lastly, is a task that assesses proprioceptive 387 performance of the affected limb. The task design involves a sequence of events with a passive 388 movement of the affected arm followed by a matching movement of the unaffected limb in the 389 absence of visual control. Therefore, also working memory and attention components are required 390 for the performance of this task. 391 While in VGR and IBC, tDCS rather induced performance increases, in APM and CBC, by 392 contrast, both setups induced mostly performance decreases. In particular, we consider the 393 following reasons for such decreases: Firstly, the tasks differ in their level of sensory information 394 necessary for task performance for cooperative inter-limb coordination [21,23]. In consequence, it is 395 likely that these two tasks, and APM in particular, have higher demands on working memory and

Relationships Induced by tDCS (Individual-Level).
A Count of parameters affected by tDCS provided as percentages [%] of performance change categories for ua-tDCS (left), bi-tDCS (middle) and in direct comparison between ua-tDCS and bi-tDCS (right) across all (upper charts) and separately for each task (lower charts).
Statistics for each parameter provided in Table 1.
B Interpretation key to inter-parametric change score relationships.

Disclosures and Conflicts of Interest
We wish to confirm that there are no known conflicts of interest associated with this publication and there has been no significant financial support for this work that could have influenced its outcome.
We confirm that the manuscript has been read and approved by all named authors and that there are no other persons who satisfied the criteria for authorship but are not listed. We further confirm that the order of authors listed in the manuscript has been approved by all of us.
We confirm that we have given due consideration to the protection of intellectual property associated with this work and that there are no impediments to publication, including the timing of publication, with respect to intellectual property. In so doing we confirm that we have followed the regulations of our institutions concerning intellectual property.
We further confirm that any aspect of the work covered in this manuscript that has involved human patients has been conducted with the ethical approval of all relevant bodies and that such approvals are acknowledged within the manuscript.
We understand that the Corresponding Author is the sole contact for the Editorial process (including Editorial Manager and direct communications with the office). He is responsible for communicating with the other authors about progress, submissions of revisions and final approval of proofs. We confirm that we have provided a current, correct email address which is accessible by the Corresponding Author and which has been configured to accept email from muffel@cbs.mpg.de.
Signed by all authors