NEURONAL REPRESENTATION OF SONG VARIABILITY IN SONGBIRD BASAL GANGLIA-CORTICAL LOOP

The basal ganglia–thalamo–cortical (BGT) circuit is critical for motor skill acquisition across vertebrates, including speech learning in humans, a process thought to rely on reinforcement learning (RL). Songbirds provide a powerful model for studying skill learning, as they acquire their songs by imitating adult tutors through trial and error. Their specialized song system, comprising both a BGT loop and a cortical motor pathway, underlies song learning and production.
Within this circuit, the lateral magnocellular nucleus of the anterior neostriatum (LMAN) serves as the cortical output of the BGT loop, injecting variability into song production via projections to the motor pathway. RL models propose that LMAN sends an efference copy of vocal variability to Area X, the songbird basal ganglia nucleus homologous to mammalian motor circuits. However, the functional organization of LMAN and Area X, as well as how vocal variability is encoded within these regions, remains poorly understood.
To investigate these questions, we examined functional connectivity and neural dynamics in LMAN and Area X during singing using chronic and acute in vivo electrophysiology in zebra finches. We found no straightforward linear relationships between single-neuron activity and acoustic song features. Instead, population-level analyses and nonlinear models may better capture the underlying neural representations. In parallel, we are mapping functional connectivity from LMAN to Area X to estimate the neural population size required to encode vocal variability. Together, these findings provide insights on how BGT circuits represent vocal features and inform broader mechanisms of sensorimotor learning and speech acquisition.