Final Report Summary - NMVLRBG (Neural mechanisms of vocal learning: role of the basal ganglia) Human speech is a complex sensorimotor skill and vocal learning is one of the most striking cognitive abilities of the brain. As other motor skills, vocal learning naturally emerges from our experience and vocal production is performed without awareness. Such complex motor skill learning requires the acquisition of a given movement sequence, which is learned slowly over several training sessions. Cortical and basal ganglia (BG) networks are involved in the acquisition of motor skills, and their operation is likely optimized by learning mechanisms, each unique to the cortex and BG. How do BG and cortical network interact to successfully imprint long-lasting memories of optimal sensorimotor transformation underlying motor skills? If the BG drive learning in the premotor network, how can motor skill rely only on the premotor network after learning? Songbirds use learned vocalizations to communicate and offer a unique model to study the neural mechanisms of vocal learning. The goal of the present project is to reveal the microscopic (cellular, neuronal) factors that underlie sensorimotor learning in general and vocal learning in particular. To this end, we investigate the neural mechanisms of BG-driven sensorimotor learning in songbirds. Our project combines an experimental approach (chronic neural recording in awake bird, manipulation of single neuron activity through light-activated ion channels) and a theoretical one (development of a mathematical model of the song system).Long-term applications of the proposed research project are numerous, from medicine to robotic. They include, but are not limited to, the discovery of original therapies for language disorders or the design of new algorithms for learning in artificial neural network.On the theoretical side, our studies have revealed that learning the coordination of multiple gestures, the adaptation process for the different gestures can interfere destructively or constructively depending on the similarities between the sensory representation of gestures and the overlap in their neuronal representations. Destructive interferences can result in a drastic slowdown of the adaptation. As a result of interference, the time to adapt varies non-linearly with the number of gestures learned. We also demonstrate how shaping the reinforcement signals or shaping the task can accelerate the learning dramatically by reducing destructive interferences. We argue that experimentally investigating the dynamics of reward-driven sensorimotor adaptation for more than one gesture can shed light on the underlying learning rules. Additionally, we have revealed that there are universal temporal features during babbling-like vocalizations in human and non-human vocal learners and demonstrate how emerging spatiotemporal neuronal variability can produce such behavioral variability. The results points toward a general mechanism for producing universal babbling behavior based on a simple network architecture recently revealed in songbirds. On the experimental side, we have shown the implication of the BG-thalamo-cortical loop in seasonal song plasticity in songbirds. Indeed, the BG loop participates to the additional acoustic variability observed in autumn when the birds resume singing after a silent summer. This is in-line with the role of the BG loop in adding frequency variability in song syllable during juvenile learning in zebra finches, and clearly points to a similar role of the loop during adult song plasticity. Moreover, we find that the global song structure of song (song duration, phrase duration and inter-syllable gap duration), which is clearly affected by seasonal plasticity, is also influenced by the BG loop. While this result is in contrast with the classical role assigned to the BG loop in closed-ended learner songbirds, it opens a new animal model for the study of BG function in adult plasticity. We are currently recording neuronal activity in the GB activity to reveal the underlying mechanisms. Finally, we found an unexpected projection from the deep cerebellar nuclei to the BG, opening a new line of research to investigate the function of the cerebellum and the cerebello-BG interaction in sensorimotor learning in the songbird model.
