Neural Correlates of Multisensory Integration for Feedback Stabilization of the Wrist

Robust control of action relies on the ability to perceive, integrate, and act on information from multiple sensory modalities including vision and proprioception. How does the brain combine sensory information to regulate ongoing mechanical interactions between the body and its physical environment? Some behavioral studies suggest that the rules governing multisensory integration for action may differ from the maximum likelihood estimation rules that appear to govern multisensory integration for many perceptual tasks. We used functional magnetic resonance (MR) imaging techniques, a MR-compatible robot, and a multisensory feedback control task to test that hypothesis by investigating how neural mechanisms involved in regulating hand position against mechanical perturbation respond to the presence and fidelity of visual and proprioceptive information. Healthy human subjects rested supine in a MR scanner and stabilized their wrist against constant or pseudo-random torque perturbations imposed by the robot. These two stabilization tasks were performed under three visual feedback conditions: “No-vision”: Subjects had to rely solely on proprioceptive feedback; “true-vision”: visual cursor and hand motions were congruent; and “random-vision”: cursor and hand motions were uncorrelated in time. Behaviorally, performance errors accumulated more quickly during trials wherein visual feedback was absent or incongruous. We analyzed blood-oxygenation level-dependent (BOLD) signal fluctuations to compare task-related activations in a cerebello-thalamo-cortical neural circuit previously linked with feedback stabilization of the hand. Activation in this network varied systematically depending on the presence and fidelity of visual feedback of task performance. Addition of task related visual information caused activations in the cerebello-thalamo-cortical network to expand into neighboring brain regions. Specific loci and intensity of expanded activity depended on the fidelity of visual feedback. Remarkably, BOLD signal fluctuations within these regions correlated strongly with the time series of proprioceptive errors—but not visual errors—when the fidelity of visual feedback was poor, even though visual and hand motions had similar variability characteristics. These results provide insight into the neural control of the body’s physical interactions with its environment, rejecting the standard Gaussian cue combination model of multisensory integration in favor of models that account for causal structure in the sensory feedback