Unveiling neural code with a low dimensional model of birdsong production

A. AMADOR, Y. SANZ PERL, G.B. MINDLIN AND D. MARGOLIASH

Salt Lake City, UT

Conferencia; Computational and Systems Neuroscience (COSYNE); 2014

Songbirds are a well-studied example of vocal learning that allows integrating neural and peripheral recordings with a precisely quantifiable behavior. Although neural activity in the premotor forebrain nucleus HVC has been related to song acoustics in auditory playback experiments, it remains unresolved whether neural activity is related to song spectral structure during singing. To address this issue, we worked with a minimal physical model for birdsong production, having as an output a synthetic song. Each syllable was coded in terms of parameters related to air sac pressure and tension of the syringeal labia, defining motor ?gestures?. To validate this model, we assessed responses of HVC neurons to song playback in sleeping birds, as HVC neurons exhibit highly selective responses to the bird´s own song (BOS). Remarkably, the mathematical model was able to elicit responses strikingly similar to those for BOS, with the same phasic-tonic features. These results demonstrate that a low dimensional model representing an approximation of peripheral mechanics is sufficient to capture behaviorally relevant features of song.Analyzing the HVC neurons responses to playback of each bird?s own song, we observed that projection neurons were excited and interneurons were suppressed, with near-zero time lag, at the times of gesture extrema (defined as beginning, end or maxima of gestures). In this way, HVC neurons precisely encode the timing of extreme points of movement trajectories. We confirm these results with HVC recordings in singing birds. Given that HVC activity occurs with near synchrony to behavioral output, we propose that the activity of HVC neurons cannot be premotor as the accepted paradigm proposes, and therefore another model needs to be defined.