The nervous system maps high-dimension sensory inflow to low-dimension motor outputs during postural responses J. Lucas McKaya and Lena H. Tinga,b a Georgia Institute of Technology and bEmory University, Atlanta, GA, USA We have proposed that muscle activity of the automatic postural response (APR) is determined by the activation of a small number of muscle synergies [1]. When a postural perturbation is caused by horizontally translating the support surface, joints across the body are perturbed, stabilized initially by viscoelastic properties of active muscles [2]. During this 30 ms disturbance period, sensory information about the disturbance is used by the central nervous system (CNS) to pattern the APR, which becomes observable in electromyographic (EMG) activity at 50 ms and in mechanical variables at about 120 ms in cats. Here, we investigate at what stage in this causal chain of events the low-dimension of APRs originates. One possibility is that the dimension is defined peripherally: the 2 dimensions of the planar perturbation are simply transferred through the biomechanical, sensory, and motor systems, causing an equally low-dimension response. However, during standing balance, gravity acts perpendicular to the horizontal perturbations, causing biomechanical disturbances throughout the body. Disturbance effects may also depend on the state of the animal [2]. If these factors introduce complexity into the sensory inflow generated during the disturbance period, then the low-dimension of the evoked response must be defined in the central sensorimotor transformation. We tested these two predictions by comparing the dimension of the biomechanical disturbances, EMG responses, and biomechanical responses elicited by multidirectional postural perturbations. We analyzed biomechanical disturbances (32 joint angles, 32 joint angular velocities, 12 ground reaction forces), postural responses (16 hindlimb EMGs), and biomechanical responses to support-surface translations in 3 cats (12 or 16 perturbation directions, ~5 trials per direction). For each trial, mean values of each variable during the disturbance and response periods [3] were found. Three different principal components analyses (PCA) were performed. First, the dimension of each data type (e.g., forces, joint angles, EMGs) was estimated separately by determining the minimum number of PCs required for R2≥0.90 [4]. Second, to control for different matrix sizes, we compared the different data types by sampling 12 variables from each and estimating the dimension in 1000 such subsets (Fig. 1). Third, we compared the dimension of the biomechanical disturbance, the evoked EMG, and the biomechanical response by analyzing 100 subsets that randomly sampled any 12 disturbance biomechanical variables, any 12 EMGs, and any 12 response biomechanical variables (Fig. 2). All datasets associated with the disturbance period had greater dimension than the planar perturbation. First, joint angle disturbances were higher dimension than EMG responses (5.6±3.9 vs. 3.7±0.6, µ±σ across cats). Disturbance forces also had higher dimension than response forces (3.4±0.5 vs. 1.0±0.0, p<0.01). Second, these differences were maintained when we controlled for the number of variables in each data type in the second PCA (e.g., Fig. 1, F(6,22) = 4.76, p<0.01). Third, when all biomechanical variables were considered together, the biomechanical disturbance (5.0±0.6) had greater dimension than the EMG response (3.0±0.5) and the subsequent biomechanical response (3.7±0.8; ANOVA, F(2,895)=892.5, p<<0.01; Fig. 2). Our data suggest that the nervous system maps higher-dimensional sensory inflow to a lower-dimensional motor output during postural responses. Planar perturbations are made more complex in the redundant kinematic chain, and are not equivalent to planar reaching tasks [5], where the mapping from joint torques to kinematics is deterministic. Postural responses may therefore provide a unique paradigm for differentiating neural and biomechanical dimensionality constraints within the context of a single motor task [6].

Figure 1. Scree plots of PCA reconstruction R2 for postural disturbance and response variables in cat Sq. 95% R2 confidence intervals shown were created by resampling 12 random variables from each dataset 1000 times and performing PCA. Biomechanical variables were examined during disturbance (red) and response (black) periods. Top row: The dimension the kinematic variables of joint angle and joint velocity was approximately 4, and did not differ in the disturbance and response periods. Bottom left: The dimension of force variables decreased drastically between disturbance and response periods, from 3 to 1. Bottom right: EMG was examined only during the response period and had dimension 3.

Figure 2. Dimension estimates of the total biomechanical disturbance, EMG response, and total biomechanical response. Mean dimensionality estimates and confidence intervals were created by repeatedly resampling (100 times for each cat) any 12 disturbance biomechanical variables, any 12 EMGs, and any 12 response biomechanical variables, and performing PCA. The biomechanical disturbance had greater dimension (5.0±0.6) than the EMG response (3.0±0.5) and the subsequent biomechanical response (3.7±0.8; ANOVA, F(2,895)=892.5, **p<<0.01). [1] G. Torres-Oviedo, J. M. Macpherson, and L. H. Ting, "Muscle synergy organization is robust across a variety of postural perturbations," Journal of Neurophysiology, vol. 96, pp. 1530-1546, SEP 2006. [2] N. E. Bunderson, T. J. Burkholder, and L. H. Ting, "Reduction of neuromuscular redundancy for postural force generation using an intrinsic stability criterion," J Biomech, vol. 41, pp. 1537-44, 2008. [3] L. H. Ting and J. M. Macpherson, "Ratio of shear to load ground-reaction force may underlie the directional tuning of the automatic postural response to rotation and translation," Journal of Neurophysiology, vol. 92, pp. 808-23, Aug 2004. [4] V. C. Cheung, A. d'Avella, and E. Bizzi, "Adjustments of motor pattern for load compensation via modulated activations of muscle synergies during natural behaviors," J Neurophysiol, vol. 101, pp. 1235-57, Mar 2009. [5] I. Kurtzer, J. A. Pruszynski, T. M. Herter, and S. H. Scott, "Primate Upper Limb Muscles Exhibit Activity Patterns That Differ From Their Anatomical Action During a Postural Task," Journal of Neurophysiology, vol. 95, pp. 493-504, January 1, 2006 2006. [6] R. Gentner and J. Classen, "Modular organization of finger movements by the human central nervous system," Neuron, vol. 52, pp. 731-42, Nov 22 2006.