Effect of CB-tDCS on Motor Behavior During the Training Phase

The analysis of the training sessions (D1S1, D1S2, D2S1, D2S2) indicated a significant main effect of STIMULATION F(1, 616.1) = 15.53, p < .001, ηp2 = .02 and of SESSION F(3, 616.2) = 3.83, p = .01, ηp2 = .02, but no STIMULATION × SESSION interaction F(3, 616.1) = 1.82, p = .142, ηp2 = .01. Specifically, during MF-stimulation, the subjects demonstrated a globally better motor performance (smaller AUC of the movement trajectory for correct trials) in comparison to control (Fig. 2a). To further quantify the effects of the study intervention, in the next step, we analyzed the sessions with cerebellar stimulation (active vs. sham CB-tDCS) separately, namely D1S2 and D2S2. The results indicated a significant effect of STIMULATION F(1, 300.2) = 8.22, p = .004, ηp2 =.03, of SESSION F(1, 300.3) = 7.79, p = .006, ηp2 = .03 and a significant SESSION × STIMULATION interaction F(1, 300.3) = 5.27, p = .022, ηp2 = .02. Post hoc pairwise comparisons indicated that the effect of STIMULATION was significant for SESSION D1S2 t(287.9) = 3.66, p = .002, but not for D2S2 t(286.7) = 0.4, p = .979, which suggests a learning-phase specific effect of CB-tDCS. Furthermore, the effect of SESSION was significant for the sham stimulation group t(49.2) = 3.69, p = .003, but not for the real stimulation group t(58.8) = 0.34, p = .986, indicating a CB stimulation effect during the early stages of motor learning that remains stable over time and an eventual “catching-up” in performance in the control group.

Retention was measured during two follow-up visits conducted 1 day and about 10 days after the training phase. The data were corrected by subtraction of the last block of the last training session to ensure a comparison of actual retention of the learned sequence with respect to the end of the training phase. Results showed no significant effect of STIMULATION F(1, 90.7) = 1.15, p = .286, ηp2 = .04, or of FU F(1, 19.5) = 0.37, p = .55, ηp2 = .002, or a STIMULATION × FU interaction F(1, 90.7) = 0.05, p = .827, ηp2 = .001, see Fig. 2b.

To mitigate carry-over effects, after crossing over to the remaining stimulation condition, a wash-out phase was respected (mean: 35 days, range: 13 to 51 days). We were not able to detect differences for the baseline evaluation (MF-stimulation: U = 22.00, p = 0.25; Control: U = 18.00, p = 0.66) and the linear slopes (U = 11.00, p = 0.54 for both groups) fitted through the training data contrasting the before to the after cross-over phase. The complementary Bayesian analysis indicated that it was more likely that the baselines (MF-stimulation: BF01 = 1.44; Control: BF01 = 1.91) and slopes (MF-stimulation: BF01 = 1.70; Control: BF01 = 1.64) are equal than different. This makes a considerable carry-over effect unlikely. In addition to aggregated whole-group data, exemplary movement trajectories of a single participant sampled in the early and late training phase are depicted in Fig. 2c, d.

Analysis of Temporal Subcomponents of Learning

To analyze offline learning, the within-day offline analysis was separated from the overnight offline analysis. This was done because of different stimulation paradigms (M1 vs. CB) and the additional factor of sleep during the overnight offline learning [34]. The analysis of the within-day offline learning between D1S1 and D1S2, and between D2S1 and D2S2 showed no significant effect of STIMULATION F(1, 33) = 0.002, p = .962, ηp2 = .003, or of TIMING F(1, 33) = 0.095, p = .760, ηp2 < .001, nor an interaction between STIMULATION × TIMING F(1, 29) = 0.27, p = .607, ηp2 = .01. The overnight offline learning between session D1S2 and D2S1 showed no effect for STIMULATION F(1, 8.5) = 2.01, p = .192, ηp2 = .19.

Impact of Baseline Motor Ability on Stimulation Response

The behavioral data of the training phase were separated by a median split into a low and high-performer group based on their baseline performance to investigate if motor ability at baseline impacts subjects’ response to MF-stimulation. The linear mixed-effects model included behavior (AUC) as the dependent variable. The independent variables were TIMING (D1S1, D1S2, D2S1, D2S2) and PERFORMANCE (high vs. low). The results showed a significant main effect for TIMING F(3,32.7) = 4.18, p = .013, ηp2 = .28, for PERFORMANCE F(1, 10.9) = 10.14, p = .009, ηp2 = .48 and a trend for an interaction between TIMING × PERFORMANCE F(3, 270.7) = 2.52, p = .059, ηp2 = .03. Post hoc pairwise comparisons showed that MF-stimulation resulted in a stronger enhancement of motor behavior in the low performer group compared with the high performers in training session D1S1 t(15.4) = 2.59, p = .02, in D1S2 t(15.7) = 2.26, p = .038, in D2S1 t(15.9) = 3.13, p = .006 and D2S2 t(15.8) = 3.2, p = .006. This points towards an ability dependence of the induced stimulation effect, please see Fig. 3a.

Fig. 3

Motor ability-dependent effects of CB-stimulation. a The performance in the behavioral task during the active MF stimulation sessions only. The groups have been separated into high vs. low performer (“Perform”) groups based on the baseline performance. b The performance during the CB-stimulation sessions only. Groups are divided into MF-stimulation (“MF-Stim”) vs. control and high vs. low performance (“Perform”) during the preceding baseline session. *Significant difference between the respective contrast (p < .05). For an additional figure depicting the individual data points per subject, please see the supplementary material Fig. S2

To further explore the stimulation sensitivity, the active vs. sham stimulation conditions during the CB-stimulation sessions (D1S2, D2S2) were compared. The high vs. low performers were related to their respective stimulation conditions, creating four separate groups for comparison: “MF-Stim–high Perform,” “MF-Stim–low Perform,” “Control–high Perform,” and “Control–low Perform”. The results showed a significant main effect for TIMING F(1, 303.3) = 9.64, p = .002, ηp2 = .03 and for GROUPS F(3, 27.6) = 14.17, p < .001, ηp2 = .61, and an interaction effect for TIMING × GROUPS F(3, 393.4) = 5.49, p = .001, ηp2 = .05. There was a significant difference in behavioral performance comparing MF-stimulation to control stimulation in the low performer group during D1S2 t(297.9) = − 6.25, p < .001; this effect did not remain in D2S2 t(297.4) = − 2.37, p = .085. However, there was no effect of MF-stimulation vs. control stimulation in the high performers during D1S2 t(295) = 0.45, p = .970 and during D2S2 t(295.9) = 1.48, p = .452. This indicates that the observed CB-tDCS effect during the early training phase on the group level was driven by a high stimulation protocol susceptibility for subjects with a lower baseline motor ability, please see Fig. 3b.

Impact of Intracortical Inhibition and Facilitation of the Motor Cortex and Stimulation Response

The TMS-based metrics measured at the beginning of D1S1 showed no significant differences for the TP peak-to-peak amplitudes, the percentage of maximal stimulator output required to obtain adjusted TPs, SICI, or ICF between the before and after cross-over visits; for details, please see Table 2. This points towards a reliable adjustment of the TMS parameters, which assured that the metrics were obtained at a comparable range of the respective recruitment curves.

Table 2 Overview of the achieved adjustment for the TMS parameters. Table shows the mean and standard error of mean (SEM) in brackets of the different TMS parameters before and after cross-over. Paired samples t-test comparisons between before and after cross-over sessions are shown in the statistics column

To evaluate if the assessed metrics contain information that determines the subsequent response to stimulation, the data were divided into two groups based on a median split. Only the CB-stimulation sessions (D1S2, D2S2) were considered for the analysis. Following this procedure, we obtained two subgroups per assessed TMS metric, weak vs. strong inhibition for SICI and weak vs. strong facilitation for ICF. These factors were grouped based on the stimulation condition, resulting in four groups: “MF-Stim–strong SICI or ICF”; “MF-Stim–weak SICI or ICF”; “Control–strong SICI or ICF”; “Control–weak SICI or ICF”.

For SICI the results indicated a significant main effect for TIMING F(1, 222) = 5.35, p = .022, ηp2 = .02, and for GROUPS F(3, 224.9) = 25.29, p < .001, ηp2 = .25. There was no interaction between TIMING × GROUPS F(3, 222) = 1.19, p = .314, ηp2 = .02. Post hoc pairwise comparisons showed a better performance for strong vs. weak inhibition with MF-stimulation during session D1S2 t(219.8) = 5.69, p < .001 and D2S2 t(220.4) = 4.63, p < .001. There was better performance for strong vs. weak inhibition with control stimulation during D1S2 t(222.9) = 7.4, p < .001, and D2S2 t(222.6) = 6.84, p < .001. The participants with weak inhibition performed significantly better with MF-stimulation vs. control stimulation during D1S2 t(220.3) = − 3.56, p = .011, but not during D2S2 t(218.4) = − 2.61, p = .159. There was no difference between MF-stimulation vs. control stimulation in the strong inhibition group during D1S2 t(215.2) = − 1.29, p = .903 or during D2S2 t(215.2) = 0.7, p = .997, please see Fig. 4a.

Fig. 4

Relationship of ppTMS-derived metrics and stimulation response. Groups were separated based on the level of inhibition for SICI, respectively facilitation for ICF and applied stimulation condition: MF-Stimulation (“MF-Stim”) vs. control stimulation (“Control”). Only the sessions, in which active CB-stimulation or sham was applied (D1S2 or D2S2), were considered. a Baseline SICI strong vs. weak inhibition in relation to task performance. b Baseline ICF strong vs. weak facilitation in relation to task performance. *Significant difference between the respective contrast (p < .05). For an additional figure depicting the individual data points per subject, please see the supplementary material Fig. S3

The ICF results showed a significant main effect for TIMING F(1, 221.9) = 5.55, p = .019, ηp2 = .02 and for GROUPS F(3, 208.6) = 3.23, p = .024, ηp2 = .04, but not an interaction between TIMING × GROUPS F(3, 221.9) = 1.37, p = .252, ηp2 = .02. On visual inspection, there seemed to be an indication that the participants with strong facilitation perform better than the participants with weak facilitation in the MF-stimulation but not the control condition. However, post hoc comparisons demonstrated no significant differences between any of the groups or for any of the timings, please see Fig. 4b.

For an exploratory sub-analysis contrasting stimulation response based on the presence versus absence of stable upper limb motor evoked potentials (MEPs) of the affected limb, please see the supplementary material section “impact of corticospinal tract integrity on stimulation response” and Fig. S4. In brief, this exploratory analysis may indicate a higher stimulation response in stroke survivors with a no-MEP status.