Millisecond-scale motor coding precedes sensorimotor learning in songbirds

The changes in neural activity that underlie motor skill acquisition during development are unknown. Juvenile songbirds learn their songs by an imitative process, and after learning adults use a millisecond-precise spike timing code to control vocal acoustics. Current theories suggest that developmental changes in neural firing rates, rather than precisely timed spike patterns, underlie the emergence of learning. Here we tested the hypothesis that songbirds transition from a rate-based to a spike-timing-based motor code during the process of vocal development. To do so, we recorded vocal motor neurons across development as individual Bengalese finches learned their songs. Contrary to our hypothesis, we found that despite dramatic changes in firing statistics during development, millisecond-scale spike pattern codes for vocal acoustics are present throughout all stages of vocal development. Furthermore, firing rate fluctuations are no more predictive of song output in young learners than in expert adults. The dramatic changes in spiking statistics observed during song learning therefore do not reflect a developmental change in the timescale of motor coding, but instead signals the selection of a particular subset of precisely timed spike patterns. We speculate that such patterns are favored because they most reliably modulate behavior.