Spinal-Like Regulator

Jared Goodner, John Sunwoo

Most studies of arm and hand movements in primates have attempted to relate neural activity in motor cortex to simple features of the movement (e.g. position, force, individual muscle activity). But motor cortex does not control those features directly; it acts through the complex circuitry of the spinal cord and the mechanics of the musculoskeletal system. “The Circuitry of the Human Spinal Cord” has been well-described for simple antagonist pairs of muscles but it isn’t clear how it can be extended to multimuscle, multi-degree-of-freedom linkages. We have modeled a two-axis, four-muscle wrist joint with realistic musculoskeletal mechanics and proprioceptors (spindle primary and GTO afferents) and a network of spinal circuitry based on the classical types of interneurons (propriospinal, monosynaptic Ia- excitatory, reciprocal Ia-inhibitory, Renshaw inhibitory and Ib-inhibitory pathways) and their supraspinal control (via biasing activity, presynaptic inhibition and fusimotor gain).  All interneurons and motoneurons had sigmoidal transfer functions that are in general agreement with cellular physiology and whose nonlinear properties lead to the types of “reflex gating” described when biological systems switch between behavioral states. The modeled system has a very large number of control inputs (~200), not unlike the real spinal cord that the brain must learn to control to produce desired behaviors. None of the corticomotoneuronal control inputs projected directly to the spinal motoneurons in our model but the patterns of muscle coordination and movement emerged largely from the integration of descending control and afferent feedback in spinal interneurons such as the propriospinal system.

 


We found that relatively simple and generally intuitive step-changes in a small subset of descending controls could reproduce very different but well-described behaviors.  Furthermore, details of stability and temporal patterning of muscle recruitment were surprisingly realistic, even though we made little or no attempt to optimize the values of the many other control signals and we did not modulate any of the control signals.  In fact, these behaviors appear to be robust emergent properties of the complete set of spinal circuitry as modeled.  Such properties would substantially change and generally simplify the learning of motor tasks by the brain. We then optimized the control inputs using gradient descent and hill-climbing algorithms that minimized a simple kinematics cost function. Even though the inputs were selected randomly during the initial simulation, the algorithm converged rapidly to produce physiologically realistic outputs and the solution space was found to consist of rather broad and stable local minimas. More systematic analysis is underway to determine the range of tasks and behaviors that can be modeled by this system. We are also working on a method to couple actually recorded M1 activity to our model to replicate recorded behaviors in monkeys.