A neural geometry for forelimb proprioception in the cervical spinal cord

Precise, real-time somatosensory feedback is essential for coordinated movement. While the anatomy and physiology of these sensory pathways is well described, their neural code and its construction remains unclear. Here we show that neurons in the cervical spinal cord generate a precise neural representation of the forelimb's kinematic state using muscle and tendon sensory afferent inputs. We identify two classes of movement responsive neurons - the first encodes speed, position and direction of the limb, while the second exhibits precise firing at specific limb positions or grid-like firing patterns that tile space. Their composite population activity is constrained to a low dimensional manifold that is an ordered representation of the position and velocity of the limb. Ablating muscle and tendon sensory afferents, but not cutaneous sensory afferents, disrupts this neural manifold. Moreover, transient perturbations of muscle and tendon afferents in freely moving mice reaching to spatial targets cause end-point errors as predicted by the deficits in the neural code. Our findings demonstrate that spinal networks, one synapse from the periphery, perform the complex computations necessary to represent forelimb movement.