Neural Agonist-Antagonist Coupling in the Absence of Mechanical Coupling after Targeted Muscle Reinnervation

TL;DR

The work addresses the disruption of neuromechanical coupling after limb amputation and targeted muscle reinnervation by recording motor-unit activity from reinnervated muscles during agonist–antagonist tasks. Using high-density intramuscular arrays and blind source separation, the authors demonstrate persistent, functional coupling between agonist and antagonist pathways, evidenced by shared motor units and task-dependent neural structure. The study introduces NNMF-based neural manifolds to quantify common input and shows a minimal latent dimensionality of at least two for agonist–antagonist pairs, implying residual central coordination that can inform prosthetic control and proprioceptive feedback strategies. These findings suggest that despite biomechanical decoupling, the nervous system maintains coordinated commands, offering avenues for advanced, impedance-modulated prostheses and richer sensory substitution.

Abstract

Following limb amputation and targeted muscle reinnervation (TMR), nerves supplying agonist and antagonist muscles are rerouted into separate targeted muscles, disrupting natural neuromechanical coupling between muscle groups. Using high-density intramuscular microelectrode arrays in reinnervated muscles, we show that neural signals for agonist and antagonist tasks remain functionally coupled: motor units active during agonist tasks were also recruited during corresponding antagonist tasks, despite no visual feedback on coactivation being provided.

Figures (3)

• Figure 1: Micro-electrode arrays record residual neural coupling in muscles reinnervated by transferred polyfascicular nerves during agonist and antagonist tasks.a-b, In persons with intact upper limbs, the neural control of movement, e.g., wrist flexion and extension, involves the activation of groups of agonist and antagonist muscles spanning relevant articular joints. c-d, Loss of peripheral tissue after amputation leads to a disruption of the neuro-mechanical coupling between agonist and antagonist muscles of the upper limb. A transhumeral patient had the residual ulnar (blue), median (orange) and radial (green) nerves transferred to targeted muscles via Targeted Muscle Reinnervation (TMR) surgery. The participant was asked to perform agonist-antagonist tasks with his missing limb (e.g., wrist flexion and extension) while the intramuscular activity of all reinnervated muscles was recorded using micro-electrode arrays (panel f) inserted percutaneously into the muscles. The median across-channel EMG activity of the reinnervated muscle defined as the agonist for a given task (e.g., the upper portion of the pectoralis major reinnervated by the ulnar nerve for wrist flexion), was displayed on a screen as real-time visual feedback. In panels c and d, the participant performed different tasks and thus received visual feedback from different micro-electrode arrays. The participant had to match a target trapezoidal activation profile with his muscle activity. e, In an offline analysis, for each task, we decomposed the neural signals recorded by the "agonist" and "antagonist" micro-electrode arrays (blue and green) into constituent motor unit spike trains by blind-source separation. Corresponding motor unit spike trains are shown in the top (shaded blue) and bottom (shaded green) figure, during wrist flexion and extension. Tracking of motor units across the two tasks revealed that during wrist flexion, the micro-electrode array recording the activity of the "antagonist" muscle detected motor units that were functionally relevant and matched some of those recruited when the subject performed the corresponding antagonist task, i.e., wrist extension. For example, all 3 motor units detected by the "antagonist" (green) micro-electrode array during wrist flexion were recruited during wrist extension, when the TMR muscle had an antagonist role. f, Schematic of high-density micro-electrode array muceli2022blind consisting of 40 channels (diameter of 140µm) linearly distributed over a length of 2 with inter-electrode distance of 500µm. Note that in panels A and B, a representative agonist and antagonist muscle pair is shown for simplicity. The neural structures and micro-electrode array are not represented with accurate scaling to improve clarity.
• Figure 2: a,b,c, For each patient (P1, P2, P3) and reinnervated muscle (TMRX), the diagrams detail the relation between recorded tasks in terms of the number of shared motor units. Given a task ("recorded task"), a TMR muscle may have an agonist/antagonist role depending on its rerouted nerve. For each task, an agonist and an antagonist TMR muscles were identified, and the intramuscular signals were recorded. E.g., when P2 performed wrist flexion, TMR1 provided agonist signals and thus this task is labelled as "agonist". TMR3 of P2 provided agonist neural signals for wrist extension. Wrist extension task was thus labelled as agonist in TMR3, but as antagonist in TMR1. The diagrams show the relationship between "agonist" motor units and motor units recruited when the antagonist task was performed (squares highlighted in yellow). It can be observed that some of the motor units recruited during agonist tasks were also active during the corresponding antagonist task (e.g, the same TMR muscle had an agonist and antagonist role). d, Total number of motor units identified per movement when this had an agonist or antagonist role. E.g., two and one MUs were recruited by P1 during Pronation (agonist function) and Pronation (antagonist function, i.e., recorded during wrist supination).
• Figure 3: Discharge properties of motor units in reinnervated muscles with agonist and antagonist roles for a given task and neural manifold analysis. In panels a-d, we show the spike trains of motor units reliably decomposed using a micro-electrode array into the targeted muscle reinnervated by the median nerve (a-b) and radial nerve (c-d). These nerves carry neural signals for elbow flexion and extension tasks, respectively. Hence, during elbow flexion, "agonist" motor units are displayed in panel (a), whereas antagonist ones are shown in panel (c). Viceversa, during elbow extension, the radial nerve carries "agonist" signals (d) and the median nerve antagonist ones (b). For each reinnervated muscle, we tracked and colour-coded motor units active during both elbow flexion and extension tasks. In panel a, motor unit 4 (blue) is also active during elbow extension (light blue in panel b). Similarly, motor units 1-3, which have an agonist role for elbow extension (d, blue), are recruited during elbow flexion (panel c, light blue). e, we computed the mean discharge rate of each motor unit active in both a and b or c and d, as the mean of the inverse of inter-spike-intervals during the plateau phase of the contraction in the "agonist" reinnervated muscle. The discharge rate of motor units in the "antagonist" reinnervated muscle is computed within the same time window. We assessed whether motor units had a different mean discharge rate during corresponding agonist and antagonist tasks. The distribution of mean discharge rates of motor units with an agonist role (color coded in blue as in panels a and d) is displayed in the "Agonist Role" group, while antagonist motor units (light blue in panel b and c) are shown in the "antagonist role" group. Each group contains 18 points (motor units across all tasks). The two-sided Wilcoxon signed-rank test is used to assess statistically significant differences between the two groups (p-value $<$ 0.005). Statistical significance is indicated with a bar and asterisk. f, For each task, we consider the average mean discharge rate of motor units recorded in the agonist and correspondent antagonist reinnervated muscle. All detected motor units are used, e.g., the average mean firing rate of motor units in panel a and motor units in panel c constitute a pair of values in "Agonist TMR" and "Antagonist TMR". g, Neural manifold analysis for exemplary task for P2. On the left, a 15-dimensional MU spare (V) is obtained by concatenating smoothed discharge rates of motor units recruited during pronation and supination in TMR1 (black) and TMR2 (blue). The dimensionality of the latent space (H) embedded in V is estimated using Non-Negative matrix Factorization (NNMF). The time-varying latent signals are shown on the left for the estimated dimensionality equal to 3. h,$R^{2}$-curve obtained by applying NNMF on $V$ with increasing latent space dimensionality (from 1 to 5). A dimensionality of 3 (dashed line) was estimated when considering MUs recruited during pronation, and supination only, as well as when considering the concatenated data (See Methodology for details).