Cortical closed-loop brain-machine interface requires biomimetic sensory feedback

New and improved neuroprosthetics offer great hope for motor-impaired human patients to regain autonomy. One obstacle facing current technologies is that fine motor control requires near-instantaneous somatosensory feedback. The way forward is to artificially recreate the rich, distributed feedback generated by natural movements. Here, we hypothesize that incoming sensory feedback needs to follow biomimetic rules in order to be efficiently integrated by motor circuits. We have developed a rodent closed-loop brain-machine interface where head-fixed mice were trained to control a virtual cursor by modulating the activity of motor cortex neurons. Artificial feedback consisting of precise optogenetic stimulation patterns in the primary somatosensory cortex coupled to the motor cortical activity was provided online to the animal. We found that learning occurred only when the feedback had a topographically biomimetic structure. Shuffling the spatiotemporal organization of the feedback prevented learning the task. These results suggest that the patterns of inputs that are structured by the body map present in the primary somatosensory cortex of all mammals are essential for sensorimotor processing and constitute a backbone that needs to be considered when optimizing artificial sensory feedback for fine neuroprosthetic control.