Session X6: The Neural Dynamics of Songbirds

Sleep and Learning

The neural basis of cognition represents a grand challenge problem involving multiple disciplines and approaches to the analysis of behavior. Song learning by juvenile songbirds such as zebra finches has proven to have considerable utility for exploring how behavior is represented at multiple levels of brain function. As classically described, young birds are exposed to a ``tutor'' (adult) song and commit that song to memory early in life, then engage in an extended period (weeks) of plastic singing as they slowly learn to match vocal output to the tutor song memory via auditory feedback. In recent years, the role of sleep in learning processes has been actively explored. Young birds isolated from adult songs, then suddenly given access to such songs at circa 40 days of age, show a sudden change in their singing behavior starting on the day following first exposure. Such birds sing songs that have less structure in the mornings than do the songs sung in the afternoons before or after that morning. This fluctuation is directly the result of sleep (not circadian rhythm), and the magnitude of fluctuation is positively correlated with the ultimate similarity to the tutor song. Examining spontaneous neuronal activity in certain brain structures during the night in sleeping adults shows ``replay'' of the patterns of activity the same neurons exhibit during daytime singing, and ``preplay'' of new patterns that will first be incorporated into daytime singing the following day. In experiments on juveniles, nighttime neuronal activity shows dramatic changes associated with song learning, even on the night after the first day of tutor song exposure (preceding changes in singing behavior). Offline processing, especially sleep, has been well documented to participate in memory consolidation in a very broad range of behaviors including in humans. Placing the bird song results in a theoretical framework thereby helps to inform a very broad range of phenomena.   

How the songbird brain listens to its own songs

Songbirds are capable of vocal learning and communication and are ideally suited to the study of neural mechanisms of auditory feedback processing. When a songbird is deafened in the early sensorimotor phase after tutoring, it fails to imitate the song of its tutor and develops a highly aberrant song. It is also known that birds are capable of storing a long-term memory of tutor song and that they need intact auditory feedback to match their own vocalizations to the tutor's song. Based on these behavioral observations, we investigate feedback processing in single auditory forebrain neurons of juvenile zebra finches that are in a late developmental stage of song learning. We implant birds with miniature motorized microdrives that allow us to record the electrical activity of single neurons while birds are freely moving and singing in their cages. Occasionally, we deliver a brief sound through a loudspeaker to perturb the auditory feedback the bird experiences during singing. These acoustic perturbations of auditory feedback reveal complex sensitivity that cannot be predicted from passive playback responses. Some neurons are highly feedback sensitive in that they respond vigorously to song perturbations, but not to unperturbed songs or perturbed playback. These findings suggest that a computational function of forebrain auditory areas may be to detect errors between actual feedback and mirrored feedback deriving from an internal model of the bird's own song or that of its tutor.   

The Origin of Time in the Songbird Motor Pathway

Many complex behaviors, like speech or music, have a hierarchical organization with structure on many timescales. How does the brain control the timing and ordering of behavioral sequences? Do different circuits control different timescales of the behavior? To begin answering these questions, we use temperature to manipulate the biophysical dynamics in different regions of the songbird forebrain involved in song production. We found that cooling premotor nucleus HVC (high vocal center) uniformly slows song speed by up to 40{\%} while only slightly altering the acoustic structure, whereas cooling downstream motor nucleus RA (robust nucleus of the arcopallium) has no observable effect on song timing, despite a marked affect of RA spiking activity. To better understand the circuit mechanisms of precise premotor timing, we perform intracellular recordings in RA-projecting HVC neurons during singing. Our observations suggest highly ordered dynamics within HVC which are consistent with a synfire-like neuronal architecture.   

Estimating Network Properties of the Adult Song Production Pathway

To analyze networks of neurons and predict their behavior with new stimuli, one must estimate the parameters of the network, including properties of the neurons at the nodes and the properties of the synaptic connections. One is always provided with sparse data, and from this one must estimate both the parameters and the full state of the system in order to make predictions. We will discuss methods for achieving this in both a deterministic and a probabilistic setting, and address how to succeed in developing the song production networks from electrophysiological data.   

The neural dynamics of song syntax in songbirds

Songbird is ``the hydrogen atom'' of the neuroscience of complex, learned vocalizations such as human speech. Songs of Bengalese finch consist of sequences of syllables. While syllables are temporally stereotypical, syllable sequences can vary and follow complex, probabilistic syntactic rules, which are rudimentarily similar to grammars in human language. Songbird brain is accessible to experimental probes, and is understood well enough to construct biologically constrained, predictive computational models. In this talk, I will discuss the structure and dynamics of neural networks underlying the stereotypy of the birdsong syllables and the flexibility of syllable sequences. Recent experiments and computational models suggest that a syllable is encoded in a chain network of projection neurons in premotor nucleus HVC (proper name). Precisely timed spikes propagate along the chain, driving vocalization of the syllable through downstream nuclei. Through a computational model, I show that that variable syllable sequences can be generated through spike propagations in a network in HVC in which the syllable-encoding chain networks are connected into a branching chain pattern. The neurons mutually inhibit each other through the inhibitory HVC interneurons, and are driven by external inputs from nuclei upstream of HVC. At a branching point that connects the final group of a chain to the first groups of several chains, the spike activity selects one branch to continue the propagation. The selection is probabilistic, and is due to the winner-take-all mechanism mediated by the inhibition and noise. The model predicts that the syllable sequences statistically follow partially observable Markov models. Experimental results supporting this and other predictions of the model will be presented. We suggest that the syntax of birdsong syllable sequences is embedded in the connection patterns of HVC projection neurons.