Neural Control of Movement - 2018

Intro

We make hundreds of reaching movements every day that exhibit an amazing consistency. This consistency has led to the proposal that there is a certain cost to making an arm reaching movement that humans attempt to minimize and that this cost is conserved across individuals. Energy costs likely plays a role in partly determining this cost and ultimately determining movement kinematics with arm reaching. However, objective energy costs, that is metabolic cost, can be difficult to measure. Thus a proxy for metabolic cost is often used. Here we examine how five model-based neuromechanical variables, often used to represent energy costs, relate to the measured metabolic cost of reaching.

Methods

We developed a biomechanically accurate model of the arm that estimates multiple biomechanical variables and related them to experimentally measured metabolic cost of. Experiment: Metabolic rate was collected from eight subjects that made 10 cm reaches. Subjects made reaches to four different targets at 45, 135, 225, and 315 degrees from the right horizontal. Four masses, 0, 5, 10, and 20 lbs, were added to the hand. Reaches were made at six different velocities. Model: The biomechanical model of the arm consists of two joints (shoulder and elbow), two arm segments (upperarm and forearm), and eight muscles. The model simulated planar reaching movements from a minimum jerk trajectory using the experimentally constrained movement durations. Using inverse dynamics, we calculated joint torques from the trajectory to four simulated targets with four different added masses at the hand. To calculate the muscle force, we distribute the force across muscles according to different minimization functions. Once muscle force has been determined, we calculated muscle active state and determined neural drive to the muscle. Lastly, a model for energetic expenditure was used for an estimate of metabolic rate at each time point. We then integrated the time-series data for joint torque, muscle force, muscle active state, neural drive, and estimated metabolic rate for each mass and speed condition and fit these sums to collected metabolic data using either a linear or quadratic function.

Results

The best fit linear to measured metabolic cost was the energy expenditure model with an R2 value of 0.71. A common function for estimating cost of arm reaching, the rate of torque squared, performed significantly worse (R2=0.49). The three remaining variables tested, force, active state, and neural drive, provided similarly low explanatory power (R2 values 0.47, 0.48, and 0.59 respectively). Nonetheless, these variables are able to predict the general shape of the collected metabolic cost. However, they are poor at describing differences between conditions and fail at faster speeds. Our analysis highlights the importance of exercising caution when representing metabolic cost via model-derived biomechanical variables.