Neural Circuit Mechanisms Underlying Skill Learning, Adaptation, and Maintenance

Abstract

Part I

Mastering a motor skill, such as a playing the guitar, requires precisely controlling both spatial and temporal aspects of motor output – that is, what movements to perform when. While it is generally assumed that these aspects are acquired through the same learning processes and in the same circuits, there is also evidence that the brain can control them independently. But if that’s true, how is such modularity in motor control and learning implemented in neural circuitry? To probe this question, we developed a paradigm that ‘trains’ songbirds to change either spatial or temporal aspects of their vocal output and showed that learning in the two domains is implemented in distinct neural circuits. This dissociation extended to premotor nucleus HVC, which we showed encodes changes to temporal but not spectral song structure. Such functional modularity, i.e. different circuits learning and implementing different aspects of motor control, could serve to overcome the limitations of reinforcement learning algorithms in dealing with large task domains.

Having identified key mechanisms by which an acquired motor skill can be modified, we then turned to investigate the mechanisms underlying the formation of circuits during the initial acquisition of a motor skill. The neural circuits controlling learned behaviors develop under genetic constraints and in response to environmental influences. Recent studies have provided an unprecedentedly detailed view of the circuit- and synaptic-level changes that accompany complex motor learning, but have left unexplored how environmental factors influence the formation of the neural circuits underlying motor skills. To address this, we investigated how the lack of a behavioral model affects normal motor circuit development in songbirds, a question with relevance for developmental disorders associated with deficits in imitation. We found that the primary difference in circuit formation was delayed and decreased pruning at a synapse that is a principal locus of learning. We show that this difference in synapse refinement is consistent with it being the principal mechanism driving reduced temporal precision of song and the underlying motor program. Intriguingly, our finding of impaired synapse formation mirrors what has been suggested in previous studies of autism.

Part II

Assigning function to brain areas is a principal aim of neuroscience that is often pursued by rapidly and reversibly manipulating neural activity in behaving animals. An important assumption underlying this experimental regime is that consequent behavioral changes reflect the function of the targeted circuits. In Part II of this dissertation, we demonstrate that this assumption is problematic in that it fails to account for indirect effects on the independent functions of circuits downstream of the targeted area. Transient inactivation of sensorimotor area Nif in songbirds and motor cortex in rats severely disrupts courtship songs and task-specific movement patterns – learned skills that recover spontaneously after permanent lesions of the same areas. How can a brain area be both essential for behavior execution (as assayed by the now preferred method, transient perturbation) and not (as assayed by the traditional method, lesions)? We resolve this seeming paradox in songbirds, showing that sudden silencing of Nif disrupts song and neural dynamics within HVC, a downstream song control nucleus. In parallel with song recovery, the off-target effects resolved within days of lesion, a recovery consistent with homeostatic regulation of neural activity within HVC. These finding have broad implications for how neural circuit manipulations are interpreted and for understanding the mechanisms supporting functional recovery following brain injury.