Nearly all living creatures maintain learned motor skills over long time-scales–for days, years or even decades. However, little is known about the mechanistic basis of this stability. Some propose that while motor skills can remain stable for years, the individual neurons controlling them may significantly change their firing properties over the course of hours. Others contend the tuning of individual neurons is as stable as the motor skill itself. Merging these two viewpoints, the central hypothesis of this proposal is that the brain encodes learned behaviors on two distinct levels - a mesoscopic level that is highly stable, and a microscopic level in which single neurons change and are influenced by the recent history of reward. In other words, the stability of a memory is rooted not in single neuron stability, but in the stability of network patterns that persist in spite of drifting individual neuron dynamics.
We address this question in one of the most stable of all animal behaviors–birdsong. The neural circuits that underly song behavior are well defined, extensively studied, and in key respects homologous to the cortico-basal ganglia circuits underlying sensory-motor learning in mammals. For this project, the key value of the songbird is the stability of its natural motor behavior. A songbird can sing the same learned song with great precision for years. Our preliminary data reveals that single neurons in pre-motor cortex drift in their tuning over a time-scale of hours to days. However, our data also shows that the pattern of activity measured by local field potentials (LFPs, which monitor the activity of hundreds of neurons) in premotor cortex persists for months. This project will investigate the stability of neural coding in premotor cortex at the level of LFPs and through long term ensemble measurements with single neuron resolution. We will then analyze how changes in single neurons and ensembles are driven by performance errors. These experiments hold the potential to reveal single neuron rules underlying sensory-motor learning, and address long standing questions about the nature of memory stability for skilled movements.
Aim 1) To test the hypothesis that network dynamics in song pre-motor cortex is stable. Songbird motor cortex is one of the most intensively studied neural sequence generators, but a lack of multi-channel recordings in this area has limited an understanding how network dynamics controls single neuron firing. Preliminary data reveals that two length scales are fundamental to the description of the circuit dynamics. On the larger scale, (measuring the ensemble activity of hundreds of neurons by LFPs), coherent zones of inhibition cycle at 30 Hz. On the smaller scale (measuring the activity of single neurons), this 30Hz inhibitory rhythm gates excitatory cells that fire with millisecond precision. We hypothesize that the spatio-temporal pattern of inhibition is stable over months in adult birds, and also stable to perturbation of sensory feedback induced by nerve injury.
Aim 2) To test the hypothesis that the tuning of single neurons in song pre-motor cortex is unstable. While ensemble dynamics measured by LFPs and song behavior are stable in adult birds, preliminary data indicates that individual neurons in pre-motor cortex shift their tuning properties over time-scales of days. This aim will explore the dynamics of single neuron drift for stable adult song using genetically encoded calcium indicators in singing birds. Experiments will then examine whether the rate of change of single neurons depends on the quality of sensory feedback, and whether or not this drift in pre-motor cortex requires intact basal ganglia inputs.
Aim 3) To test the hypothesis that the firing of single neurons in pre-motor cortex is shaped by the recent history of time-correlated reward signals. This experiment will apply conditional sensory feedback triggered by neural activity in singing birds. This is accomplished through a brain-computer interface that delivers brief bursts of white noise that are precisely timed relative to neural firing. Neurons active during motor performance are thought to be influenced by sensory feedback in a narrowly defined time-window that corresponds to the sensorimotor loop delay between neural firing and sensory feedback. By parametrically changing the timing of sensory errors relative to spike times, this experiment will reveal the details of how single neurons adjust their activity to minimize vocal errors.