Experimental Setup and Motor Unit Tracking.

A: Experimental setup. 16-channel high-density intramuscular electrode arrays were implanted into the FDI under ultrasound guidance. Image on the left shows ultrasound during insertion with FDI muscle (red) and HD-iEMG insertion cannula (yellow) outlined. The index finger was attached to a 3-axis load cell proximal to the PIP joint (more details see Supplementary figure S1). EMG signals from the arrays were decomposed into single motor unit activity. B: Motor unit tracking. Left: MUs were identified separately in abduction and flexion and then tracked by applying the MU filters obtained in one contraction direction onto the signals of the other direction and vice versa. Right: Two exemplary MUs 4 and 7 are shown. MUAPs obtained from spike-triggered-averaging of each channels EMG signals shows identical MUAP shapes in both contraction directions, as well as distinct centers of activation along the electrode array.

Recruitment variability within and across contraction directions.

A: FDI MU recruitment thresholds and orders were computed via averaging the level of EMG activity at each MUs first spike across multiple repeated contractions separately for both contraction directions. Smoothed spike trains show a reversal of recruitment order of the exemplary MUs 9 and 12 between abduction and flexion. The change in recruitment is accompanied by a corresponding change of DR. B: MU recruitment thresholds and or-ders between two trials of abduction (left) and flexion (right). Errorbars indicate standard deviation across contractions in each trial. Pearson R indicates highly consistent RTs and ROs within the same contraction direction. C: RT and RO between abduction and flexion. Data shows some MUs considerably change their RTs and position in the RO between the two contraction directions, resulting in lower Pearson correlation values. MU#9 and MU#12 are highlighted relating to panel A. D: Each participant’s median absolute RT within and across contraction directions. Lines connect same participant‘s values. Errorbars show mean ± SD. RT was significantly higher across than within directions (paired t-test, p = 0.0016). * p < 0.05, ** p < 0.01, *** p < 0.001

Modulation of DR and RT across contraction directions.

A: Exemplary participant‘s data, showing inverse correlation of DR and RT with linear regression lines. B: Correlation between RT and DR split into lower force (left, “F1”) and higher force (right, “F2”) trials showing all participants linear regression lines. No significant different correlation was found between abduction and flexion (Wilcoxon Signed-Rank test, F1 p = 0.35, F2 p = 0.75). C: Same MUs as in A. Difference in DR over difference in RT between abduction and flexion with corresponding linear regression. MUs that reduce their RT show increased DR and vice versa. D: Difference in DR over difference in RT split into lower and higher force trials showing all participants linear regression lines. A highly significant correlation of differences in RT and DR is found across participant in both force levels (repeated-measures correlation, p<0.001). * p < 0.05, ** p < 0.01, *** p < 0.001. test, F1: Z=-0.94, p=0.35, F2: Z=-0.31, p=0.75).

Influence of MU size on RT variability across contraction directions.

A: Across-direction normalized absolute ΔRT values, split into MUs below and above each participant‘s median MUAP am-plitude. Lines connect participant‘s values, errorbars show mean ± SD. No significant effect of MU size group on ΔRT values was observed (linear mixed models, p > 0.2). B: normalized not-absolute ΔRT across contraction direction of all participant‘s MUs split into MU size groups. Both groups showed a significant increase in RT from abduction to flexion (linear mixed models. < median: p = 0.034, > median: p = 0.002). However, many MUs still showed decreasing RTs from abduction to flexion, as shown by the negative data points. C: MU RTs in mV over the MUs amplitudes separated between abduction and flexion. Colors cor-respond to participants. No significant difference was found between the average Pearson correlations per contraction direction (paired t-test, p = 0.32). * p < 0.05, ** p < 0.01, *** p < 0.001.

Intramuscular coherence in abduction compared to flexion.

A: Coherence is calculated between two randomly selected subgroups of FDI MUs per contraction direction and averaged across 100 permutations, considering only firings during steady plateau phases. B: MU DRs in coherence protocol data across participants. C: Participant group average intramuscular coherence. Intramuscular z-coherence in abduction and flexion with delta (1-5 hz), alpha (5-13 Hz) and beta (13-30 Hz) bands highlighted. MUs show stronger beta band coherence during abduction. D: Across-participants coherence in delta (left), alpha (center) and beta (right) bands, expressed as area-under-curve above bias. No significant difference was observed in delta and alpha, whereas beta-band coherence was significantly reduced in flexion compared to abduction. * p < 0.05, ** p < 0.01, *** p < 0.001.

Detailed Experimental Setup.

A: Custom Dynamometer for 2D isometric index finger contrac-tions. Main features marked with red numbers and described. The setup features a 3-axis load cell capable of sensing the applied index finger force in all possible degrees of freedom. The hand and forearm posi-tions are fixed with adjustable restraints, thereby fixing the wrist position and restricting force application through wrist contractions. The thumb position is likewise fixed to ensure a constant posture throughout an experimental protocol. Top right: 2D force feedback as observed by the user, displaying the user’s force as a green dot and trajectory with force amplitude as distance from center and force direction as angular position. Target force is displayed in red. B: Analysis of all participant‘s directional accuracy throughout the con-tractions of the experimental protocols of the present study. Mean and standard deviation of the deviation of applied force to target force is shown, split into abduction and flexion contractions. For the recruitment protocol (left) the applied force directions of the full ramp contractions, including the concentric phase is considered. For the coherence protocol, only the force applied during the plateau phase is considered. In both protocols, force application was precise across participants, showing mean deviations close to 0 with low standard deviations. Mean deviations range from-0.85° to 0.51° in the recruitment protocol and-0.21 to 2.30° in the coherence protocol.

FDI activity amplitude in different contraction directions relative to abduction.

Across participants, the plots show FDI muscle activity levels relative to the abduction activity during different con-traction directions at matched relative force levels (left) and at identical absolute force levels at 5N and 15 N (center and right) per direction. In neither case did FDI activity levels match across directions, indicating the need for participant-and direction-specific target force amplitudes for the present study protocol. Dots represent each participant’s FDI EMG amplitudes with lines connecting the values corresponding to single individuals across force directions. Methods: 19 healthy participants (age 26.8 ± 4.5 years. 14 male, 5 female) performed isometric ramp contractions in index finger abduction (90°), flexion (180°) and two di-rections equally spaced in between (120° and 150°). Participants used the dynamometer setup described in the main text and Fig. S1. Maximum voluntary contractions (MVC) values were acquired for each direc-tion separately. Participants performed ramp contractions at 15% of the respective MVC in each direction (Protocol 1), as well as contractions at consistently increasing force until failure (Protocol 2). High-Density surface EMG signals (GR03MM1807, OT Bioelettronica, Turin, Italy) were captured from the FDI muscle. EMG amplitude was calculated as the mean root-mean-square (RMS) value across the full EMG grid during the contractiońs plateau phase for the 15% MVC ramp contractions (Protocol 1, left plot), and within a 10% force window around the 5N and 15N absolute forces (Protocol 2, center and right plot). All values were then normalized to the respective participant‘s value in index finger abduction.