Participants
Two groups of subjects were recruited for this study: 28 participants (mean±SD: 24.38±3.96 years, 16 women) completed the ROT task session and 14 subjects (mean±SD: 24.99±5.43 years, 10 women) participated to the MOT task session (see below). Data from part of these subjects were presented in previous publication 15,38.
All subjects were right-handed, had normal or corrected vision and no history of disorders affecting the nervous system. The study was approved by the CUNY University Integrated Institutional Review Board (UI-IRB) and registered with the US Department of Health and Human Services Office for Human Research Protections. The experiment was performed in accordance with the ethical principles of the Declaration of Helsinki and its subsequent amendments. Each participant signed an IRB-approved informed consent form before completing the experiment.
Experimental design
Participants were comfortably seated in a sound-shielded room in front of a computer display and fitted with a 256-channel EEG Geodesic Sensor Net (Electrical Geodesics Inc., Eugene, OR).
Both the ROT and MOT groups underwent a baseline assessment (mov0) before and after (mov3) completing either three one-hour blocks of ROT, an implicit motor learning task, or MOT, a control task with the same features of ROT but with negligible learning components.
Tasks and test
The testing apparatus and the instructions to the subjects were the same for the two tasks (ROT and MOT) and the test (mov) and are detailed in previous publications 14,15,39. Briefly, in all tasks and test, subjects performed out-and-back movements on a digitizing tablet starting from a central starting point to a target presented from 400 ms as a blackening circle on a screen placed in front of the subjects. Instructions were to move after the target presentation, as soon and fast as possible, without corrections and to reverse direction within the target circle without stopping. The cursor position on the screen and the central starting point were always visible. Target presentation was random in all tasks and test.
The tasks and the test differed in the following characteristics: i) the time interval between two target presentations was 3 s in the test and 1.5 s in the two tasks; ii) in the mov test, the targets were presented at three distances (4, 7 and 10 cm) in eight directions (45° separation) with their radius varying with their distance from the center (0.5 cm, 0.88 cm, 1.25 cm, respectively). A total of 96 targets were presented in each mov test.
In the two tasks, ROT and MOT, the target array consisted of eight, radially arranged, empty circles, all at 4 cm from a center point. In order to probe implicit learning processes, in ROT the direction of the cursor on the screen was rotated relative to the direction of the hand on the tablet in steps of 10°, 20° or 30° each, either clockwise or counterclockwise, starting from 0° (no rotation) up to a maximum of 60°. For each block, we run 21 sets for ROT and 20 sets for MOT of 56 reaching movements with 30 s inter-set interval. Crucially, all three ROT blocks ended with 112 movements without rotations to avoid carry-over effects on the subsequent mov test. In the MOT task all the sets had no imposed rotation.
Kinematic data recording and analyses
The (x,y) coordinates of each movement’s trajectory were recorded with a custom-designed software by E.T.T. s.r.l., MotorTaskManager, Genoa, Italy (http://www.ettsolutions.com) and analyzed using an ad-hoc Matlab-based pipeline. First, we filtered the coordinates with a Butterworth filter and then computed the first, second and third derivative of the trajectory to obtain velocity, acceleration and jerk for all the movements.
Following previous publications 16,31, several measures were computed for each movement; in this study we focused on: reaction time (time from target appearance to movement onset), movement time (duration of the outgoing movement), and amplitude of peak velocity.
Movements with kinematic measures outside two SD and those rejected from EEG preprocessing were excluded from further analyses. Importantly, to allow for a proper task comparison unbiased by the rotation learning occurring in the ROT task, only the movements with zero-degree imposed rotation were included in the behavioral performance analyses. Therefore, for both the ROT and MOT tasks, we extracted the kinematic characteristics of the first and last 30 movements of each block (ROT1, ROT3, MOT1, MOT3).
EEG Recording and preprocessing
High density (HD) EEG data were acquired using a 256-channel HydroCel Geodesic Sensor Net (Electrical Geodesic Inc.) with a Net Amp 300 amplifier (250 Hz sampling rate, online reference electrode: Cz) and Net Station version 5.0 software. Impedances were kept below 50 kΩ throughout the recording to preserve a good signal-to-noise ratio. The entire preprocessing was carried out using the public Matlab toolbox EEGLAB version 13.6.5b 40,41. The EEG continuous signal was FIR filtered between 1 and 80 Hz and Notch filtered at 60 Hz (59-61 Hz).
Recordings were then segmented in 4-s epochs centered on target onset, resulting in a total of 96 epochs for mov, and 1176 epochs for ROT and 1120 MOT tasks. A manual visual inspection of the data was carried out to remove epochs and channels containing sporadic artifacts. After trial rejection, the average number of trials per subject was 70.09 ±18.93 and 77.75±13.50 for ROTmov and MOTmov, respectively. For the ROT and MOT tasks, we kept an average of 1001.24±93.28 epochs for ROT and 941.38±87.52 epochs for MOT.
Independent Component Analysis (ICA) with Principal Component Analysis (PCA)-based dimension reduction (max 108 components) was applied to remove stereotypical artifacts (e.g., eye blinks, heartbeat, and muscular activity).
We retained an average of 16.26±6.91 and 13.50±3.17 components for the ROTmov and MOTmov recordings, and 16.22±6.25 and 19.46±7.24 components for the ROT and MOT tasks, respectively.
Electrodes rejected due to artifacts or poor signal quality were reconstructed using spherical spline interpolation, whereas those located on the cheeks and neck were removed from later analysis, resulting in 180 electrodes. Finally, the signal was re-referenced to common average.
For the purposes of our investigation, we focused our analyses on the first and last blocks for both the test (mov0 and mov3) and the tasks (ROT/MOT1 and ROT/MOT3). All the subsequent analyses were carried out using the MATLAB Toolbox Fieldtrip 42.
EEG data analyses
In order to avoid confounding effects from mis-executed movements, after the preprocessing we discarded epochs representing movements whose kinematic parameters exceeded two SD and time-locked the remaining trials to movement onset (-1 to 2.5 s).
For both tasks (ROT and MOT) and test (ROTmov and MOTmov) the signal was decomposed into their time-frequency representations by convolving the signal with complex Morlet Wavelets at linearly spaced frequencies (1-55 Hz, 0.5 Hz bins) and increasing number of cycles (3 to 10 cycles).
In the analyses of ROT and MOT tasks, we first explored whether the practice-related within-block changes in beta oscillatory activity (13.5 25 Hz) would differ between the two blocks and between the two tasks. To avoid the confounding effects of the imposed-rotation that was implemented in ROT, only the trials corresponding to zero-degree rotation were included. Thus, for both the ROT and MOT tasks, the within-block increase is represented by the difference between the last and first two zero-degree sets of each block (ROT1, MOT1, ROT3, MOT3). Importantly, for each block, the first and last trials were normalized by subtracting and dividing the average signal of the entire time-window of all trials.
As the mov test was implemented to assess the spectral changes occurring after extensive ROT or MOT practice, the signal was normalized by subtracting and dividing each trial by the average signal of all the trials in the entire time-window of the baseline test (mov0).
Because their average beta power was exceeding 2 SD of the group average we removed: three subjects from ROT, one subject from MOT and one subject from ROTmov. Thus, the resulting sample size was 25 and 13 subjects for the ROT and MOT tasks, and 27 and 14 subjects for ROTmov and MOTmov, respectively.
Statistical analysis
Beta power analysis
For the ROT and MOT tasks and their respective mov tests (ROTmov and MOTmov), analyses on practice-related changes in average beta power were conducted with Bonferroni-corrected Monte Carlo non-parametric permutation statistics (10000 permutations).
Non-parametric paired t-test permutation analysis on the between-blocks changes in practice-related beta power increase (Block1last-first vs Block3 last-first) were first run on the two tasks separately. The alpha threshold was set at 0.01 for the first block (ROT1 and MOT1) and, due to lack of statistically significant results with alpha=0.01, at 0.05 for the third block (ROT3 and MOT3). Further, in order to unveil possible differences in the practice-related beta power increase in the two tasks, independent-samples t-test permutation analyses were run to compare Block1 (ROT1 and MOT1) and Block3 (ROT3 and MOT3) practice-related changes in the two groups (alpha=0.05).
For the ensuing mov tests (ROTmov and MOTmov), the same approach was followed. Paired t-test permutation analyses (alpha = 0.05) were run to characterize between-block changes (mov0 vs mov3) in beta oscillatory activity in the two groups separately. These analyses were followed by an unpaired t-test statistic to directly assess differences in practice-related beta power changes between the two groups (ROTmov0-ROTmov3 vs MOTmov0-MOTmov3, alpha=0.05).
Beta modulation analysis
Following our previous publications 15,39, movement-related beta modulation analyses were conducted using a personalized approach.
For each participant, we run time-frequency representations within the beta frequency range (13.5-25 Hz) using Complex Morlet Wavelets at linearly spaced frequencies (0.5 Hz bins, 10 cycles) on mov0, normalizing the signal by the average of beta power over the entire epoch. Next, the beta ERD and ERS peak amplitude and timing were computed over three broad regions corresponding to electrodes located on the frontal, left, and right sections of the EEG net. Peak ERD was defined as the minimum value of beta power between 100 ms before movement onset to 950 ms after, whereas the peak ERS was the maximum value in the 700 to 2500 ms time range. The beta ERS-ERD peak-to-peak difference (beta modulation depth) was consequently computed for each broad region to identify the electrode with the maximum beta modulation depth and the six neighbor ones. Throughout the paper, these electrode selections are denoted as Frontal, Left, and Right Regions of Interest (ROIs). The same procedure was carried out for both ROT1 and MOT1 with the following time intervals: -200 to 700 ms for the peak ERD and 500 to 1200 ms for the peak ERS.
For both the mov test and ROT/MOT tasks, time-frequency analyses were carried out on the selected ROIs (1:55 Hz, 0.5 Hz bins, 3:10 wavelet cycles) and normalized by subtracting and dividing by the average power over the entire time-window for all the trials. Peak beta ERS, ERD, modulation depth magnitude, as well as the ERS and ERD peak timing values were finally extracted for subsequent statistical analysis.
Kinematics and mov EEG indices
In order to ascertain whether a parametric test was the appropriate statistical tool to test our data, both Shapiro–Wilks tests and Kolmogorov–Smirnov tests were run on the standardized residuals of the behavioral and EEG analyses to check for normality.
As no violation was observed for all the behavioral indices, mixed-model ANOVAs (with Blocks as within-subjects factor and Group as between-subjects factor) were run to test for any practice effect on mov kinematics.
For what concerns the EEG indices (peak beta ERS and ERD amplitude, beta modulation depth), as normality assumption was violated (p < 0.05), Wilcoxon Signed Ranks Tests were first run on ROTmov and MOTmov separately to check for blocks differences in each ROI (mov3 vs mov0). Between-groups differences were assessed for each ROI on the difference between mov0 and mov3 (mov3- mov0) with Kruskal-Wallis Tests.
Source analysis
To identify the source responsible for the observed practice-related power changes, we also estimated the sources of beta oscillatory activity during a broad post-movement time window (0.7-2 s), where the beta ERS typically occurs.
For this purpose, we applied a beamforming approach, the Dynamical Imaging of Coherent Sources (DICS) method, and the estimates were calculated in the frequency domain 43.
We first computed the cross-spectral density (CSDs) matrices of the two blocks of interest (mov0 and mov3) using multitaper spectral estimates in the beta band (13.5-25 Hz) averaged over a broad beta ERS time window (0.7-2 s).
Since individual anatomical MRIs were not collected for this study, we applied a template volume conduction model of the head based on the boundary element method (BEM), a 3-compartment (scalp, skull and brain) model provided by Fieldftrip 42. The BEM model and standard EEG electrode positions were co-registered by projecting all electrodes to the nearest point on the head surface mesh and computing a bilinear interpolation matrix from vertices to electrodes. The bioelectric forward problem was formulated as a leadfield matrix, where each column corresponds with the potential distribution on all channels for one of the x,y,z orientation of the dipole.
Source reconstruction was performed on each subject using a spatial filter computed on the combined mov0 and mov3 CSD matrices; the resulting source was then contrasted as follows: (mov3 – mov0)/mov0. Once each subject’s source was reconstructed, the grand-averages of ROTmov and MOTmov sources were statistically compared to highlight whether the differences observed on the channel-level could also be observed at the source level.
Non-parametric Monte Carlo permutation test with Bonferroni correction (10000 permutations, alpha=0.05) was applied. To identify the corresponding MNI coordinates of the significant voxels, the statistic output was interpolated with the Brainnetome Atlas 44, a cross-validated atlas based on structural and functional connectivity measures.