Cellular mechanisms of learning in a closed-loop sensorimotor task – MotorSense

Considering the astronomic rewards granted to the best sports athletes, humans undoubtedly marvel at exceptional motor skills, to the point of justifying decades of intense training to reach top performance. Perhaps our fascination stems from our own experience that learning a new skill can be difficult, requiring utmost concentration and tenacity. It also brings forward the dramatic consequences of sensorimotor impairment.
Our brains learn to generate such motor commands while integrating seamlessly sensory cues, enabling us to reach an exquisite level of precision of our body movements. What are the neural mechanisms responsible for learning new sensorimotor skills, with such adaptation to ongoing events?
This question has been difficult to address because of the multiple pathways mediating the integration of sensory inputs with motor output on awake behaving animals. Among the brain areas involved, the primary motor cortex (M1) is known to be a key player in sensorimotor learning and execution. We hypothesize that learning new sensorimotor sequences requires neuronal plasticity at the motor cortex level, notably to integrate sensory inputs into motor commands in an adaptive way.
In the MotorSense project, we will explore this question in the mouse model, which allows to combine state-of-the-art recording techniques (2-photon calcium imaging and multi-channel electrophysiology) with peripheral and cortical stimulation methods while the mouse, held by the head, is resting or involved in an operant conditioning task. We will focus on the sensorimotor whisker system, known to be critical for many rodent behaviors.
Two complementary aspects will be addressed. Our first Aim will be to probe how single M1 neurons receive and integrate somatosensory inputs originating from distinct regions in the periphery. Based on previous findings, we will investigate whether M1 neurons extract global features from complex sequences of discrete tactile stimuli, particularly when these sequences are relevant for behavior. In a further step, we will test how this integration relies on direct cortico-cortical projections from the primary somatosensory cortex (S1), by replacing whisker deflections by direct cortical optogenetic stimulation of pyramidal neurons in S1. These experiments will allow us, on the one hand, to better understand the integrative principles of the somatosensory-motor cortical loop, and on the other hand to evaluate which neurons in the network are best candidates to become M1 driver neurons in the context of closed-loop sensorimotor learning.
Our second Aim will interrogate directly the ability of M1 neurons to drive such learning. In order to control both inputs and outputs of the cortical microcircuit, we will use a full closed-loop brain-machine interface developed recently. The mice will be learning a cortical neuroprosthetic task, where the actuator is an external device rather than a body part. On the motor side, the animal will learn to modulate the activity of individual M1 neurons to drive the actuator towards a target, in order to obtain a reward. On the sensory side, feedback signals will be delivered directly to S1 by spatiotemporal patterned optogenetics. In a first step, we will characterize which M1 driver neurons lead to optimal learning, with particular interest in neurons previously identified as potentially extracting relevant sensory information from distributed inputs. In a second step, we will interrogate directly the plasticity of sensory integration by M1 neurons, known to be dependent on the activity of a subtype of cortical interneurons.
Our research will reveal novel insights on sensorimotor computational strategies of cortical networks, benefiting our knowledge of fundamental human behavior. Beyond this, the expected insights will help design efficient neuroprostheses for motor-impaired patients, incorporating optimized tactile feedback.