Challenges and new approaches to proving the existence of muscle synergies of neural origin

Abstract

Muscle coordination studies repeatedly show low-dimensionality of muscle activations for a wide variety of motor tasks. The basis vectors of this low-dimensional subspace, termed muscle synergies, are hypothesized to reflect neurally-established functional muscle groupings that simplify body control. However, the muscle synergy hypothesis has been notoriously difficult to prove or falsify. We use cadaveric experiments and computational models to perform a crucial thought experiment and develop an alternative explanation of how muscle synergies could be observed without the nervous system having controlled muscles in groups. We first show that the biomechanics of the limb constrains musculotendon length changes to a low-dimensional subspace across all possible movement directions. We then show that a modest assumption--that each muscle is independently instructed to resist length change--leads to the result that electromyographic (EMG) synergies will arise without the need to conclude that they are a product of neural coupling among muscles. Finally, we show that there are dimensionality-reducing constraints in the isometric production of force in a variety of directions, but that these constraints are more easily controlled for, suggesting new experimental directions. These counter-examples to current thinking clearly show how experimenters could adequately control for the constraints described here when designing experiments to test for muscle synergies--but, to the best of our knowledge, this has not yet been done.

Figure 1. The nervous system does not need to control muscles in groups (muscle synergy hypothesis) to observe low-dimensional EMG.

A. The behavioral approach to muscle synergies. Setup: A limb with more muscles than mechanical degrees-of-freedom (DOF). Experiment: the limb moves voluntarily (or is moved externally) in a large number of directions to span its workspace. Observation: The set of points in EMG space corresponding to each movement is in a low-dimensional subspace. Explanation: The nervous system has modules that activate muscles in groups to simply the control of movement. B. Alternative explanation 1: muscle synergies are movement related. The movement set induces a set of points in the space of musculotendon length change that is low dimensional (spanned by vectors composed of the muscle moment arms grouped by DOF). We perform a thought experiment by assuming that muscles are not controlled in groups by descending drive, but each muscle independently resists lengthening during small external perturbations of the endpoint. Only muscles lengthened by the perturbation will generate EMG; muscles shortened by the perturbation will not produce EMG. Using this thought experiment, we can generate simulated EMG, and we find that it is also low-dimensional. C. Alternative explanation 2: muscle synergies are feedforward-related. In this case, we imagine a limb producing forces in all directions at its endpoint. For each direction, the set of feasible muscle activations (assuming that each muscle can be activated between 0 and 1) can be calculated. These represent all the redundant activation vectors that will generate the same endpoint force. The set of all such feasible muscle activations across all directions is low-dimensional, as detected by PCA.

Figure 2. Experimental setup to demonstrate movement- and feedforward-related muscle synergies.

A. To demonstrate feedback-related muscle synergies, we connected all seven index finger tendons in a cadaver specimen via Nylon cords to computer-controlled rotational motors, and left the index finger free. B. The experimenter forced the index finger to move while 5 N of tension was maintained on each tendon by a feedback controller. We simultaneously recorded the tendon excursions induced by each movement. C. We generated two-dimensional (entire workspace) movements by moving the finger randomly in its workspace around a starting posture. D. To demonstrate feedforward-related muscle synergies, we used the same setup as above, but rigidly coupled the fingertip to a 6-DOF load cell. We applied a sequence of muscle coordination patterns (see text) to determine the feasible forces that the finger could generate using its musculature. E. We performed linear regression on the fingertip load cell readings using the tendon tensions as factors, thus identifying the force vector in endpoint space caused by 1 N of tension in each tendon.

Figure 3. Feedback-related muscle synergies as raw data.

A. The random movements that we produced with the fingertip are shown for both specimens and postures that we examined. The reference postures, from which excursions and simulated EMG are calculated, are shown by the circles. B. EMG is simulated using the thought experiment shown in Figure 1 across the time-series of data. The time series shown here corresponds to specimen 1 in the more extended posture. C. The simulated EMG was low-dimensional in all specimens and postures examined, with 2 principal components representing more than 80% of the simulated EMG variance in all cases.

Figure 4. Feedback-related muscle synergies generalize across posture: human leg model.

A. We simulated EMG for foot movements in 16 directions in a reference posture (only 8 of 44 muscles shown for clarity). B. We found that 80% of simulated EMG in the reference posture could be represented using only 5 PCA basis vectors (synergies). When we attempted to reconstruct simulated EMG from test postures scattered over the workspace using these 5 synergies, we found that generalization was expected over large portions of the workspace (more than 80% variance accounted for in each muscle). C. One example test posture is shown in which generalization would be expected to occur, without the muscle synergy hypothesis being true.

Figure 5. Feedforward-related muscle synergies: cadaveric data.

A. We estimated the feasible force set in the palmar-proximal plane of the index fingertip for each specimen and posture examined (only one specimen/posture shown for clarity). We then determined how many force levels (concentric circles) could fit within the feasible force set. B. For each possible force level in each specimen at each posture, we estimated the dimensionality of the set of possible coordination patterns for force vectors in 16 directions. We found that the set of coordination patterns for omnidirectional tasks were low-dimensional regardless of the force level, specimen, or posture chosen (error bars indicate non-parametric 95% confidence intervals). This demonstrates that, even if the limb produces omnidirectional force, the mechanical nature of such experiments will always produce low-dimensional EMG data of the type that has been interpreted to reveal synergies of neural origin.

Figure 6. Feedforward-related muscle synergies: leg model.

A. We used a simplified model with 14 muscles/muscle groups to make feasible the computation of all possible muscle coordination patterns for foot forces in different directions (8 muscles are illustrated for clarity). B. We found that the leg in this posture had a highly elongated feasible force set compatible with prior work , . We found all force levels (concentric circles) that could fit within the feasible force set. C. We found that the set of possible coordination patterns for forces in 16 directions was low dimensional for all force magnitude levels.

Figure 7. The question that we want to answer is “Can we find synergies of neural origin?”

We believe that the first question to ask is whether the experimental paradigm is related to movement or force. The key is to disambiguate synergies of neural origin from potential confounds. If the experimental task primarily involved movement, then a biomechanical model must be proposed that relates musculotendon length changes to muscle activation. This model can then be used to predict whether the observed muscle synergies are feedback-related, and thus not neural in origin. If the experimental task primarily involved force, it is necessary to ask if the force set was only in all directions, or covered all possible directions and magnitudes. If only all force directions were covered at a fixed magnitude, then a biomechanical model must be proposed to predict endpoint force from muscle activation. It can then be determined if the observed muscle synergies are feedforward-related because the possible muscle coordination patterns occupy a low-dimensional space. Finally, if the experimental force set covered all directions and magnitudes (the entire feasible force set), and muscle synergies are observed, these synergies can be attributed to the nervous system without a biomechanical model. This is because the possible muscle coordination patterns become the entire full-dimensional muscle space once every possible endpoint force output has been visited.