Practice makes perfect: how does the brain encode exploratory learning versus learned performance?

In Chapter 11 “Memory, Learning, and Synaptic Plasticity”, we learned that memory is stored primarily as synaptic connection strengths in neural circuits. Further, through a diverse repertoire of studies using models ranging from Aplysia (sensitization: gill-withdrawal reflex), Drosophila (olfactory conditioning) to rodents, monkeys and humans (reward-based reinforcement learning, cognitive learning), we learned that there exist various plasticity mechanisms to modify synaptic weight matrices, allowing diverse forms of learning to occur. 

Notably, learning complex skills such as music and athletics requires intensive practice. While the final performance in front of an audience may be precise and reliable, the learning that led to this learned result involves exploration, experimentation, and a lot of mistake-making. What neural mechanisms underlie such exploratory learning? Specifically, how do neurons encode and regulate motor variability during practice versus performance? A study by Singh Alvarado et al. (2021) probed this question by studying some of the most impressive pupils in nature–the zebra finches.

Male zebra finches learn to produce courtship songs. Although they perform stereotyped songs in front of females during courtship, male zebra finches sing much more varied songs when practicing alone. In their study, Singh Alvarado et al. imaged spiny neurons (SNs) in the basal ganglia (sBG) using miniature microscopes in freely-singing adult male zebra finches. The authors found that during song practice, SNs are highly variable: the specific timing and participation of active neurons varied across consecutive song renditions (Fig. 1a-c). This dynamic ensemble activity of SNs was specific to singing, and such variability in neural activity was not observed in the projection neurons in HVC (HVC PNs), a major source of song-related input to sBG. In contrast, during performance in front of females, SN calcium signals are strongly suppressed (Fig. 1a-c), while singing-related activity of HVC cells and their axon terminals in sBG displayed little changes between practice and performance. When SNs were optogenetically suppressed during song practice, vocal variability was greatly reduced, resulting in a song that is more characteristic of performance. Using the variational autoencoder (VAE), an unsupervised machine learning model that compresses the high-dimensional sound spectrogram data of zebra finch song, Singh Alvarado et al. showed that distinct variants of SN neural ensemble activity map onto distinct song practice variants. Together, these results suggest that the dynamics in SN activity encode and drive vocal exploration during practice.

To further explore what signal regulates SN activity and subsequently influences vocal variability, Singh Alvarado et al. infused either noradrenaline or dopamine antagonists into the sBG of adult male finches during practice and female-directed performance. Blocking dopamine D1 receptors in the sBG did not change song variability during either practice or performance. However, when α-adrenergic receptors in the sBG were blocked, female-directed songs became more variable, similar to songs normally produced during practice. Conversely, when α-adrenergic receptors in the sBG were activated, practice songs became less variable while other song features remained unaffected. Moreover, using a combination of fiber photometry and reverse drug infusion, Singh Alvarado et al. showed that while blocking D1 receptors in the sBG did not alter SN activity levels during either practice or performance, infusing noradrenaline or an α-adrenergic receptor agonist into the sBG significantly reduced SN calcium signal during practice but not during female-directed performance (Fig 1. d-e). Finally, ex vivo whole-cell current clamp recordings from identified SNs in brain slices showed that noradrenaline suppressed direct-current-evoked action potential activity in SNs, which could be reversed by blocking α-adrenergic receptors. Meanwhile, blocking α-adrenergic receptors in naive slices strongly increased direct-current-evoked action potential responses and input resistances in SNs. Altogether, these results suggest that noradrenergic signaling directly suppresses SN activity, thereby reducing male zebra finch song variability and promoting precise song performance in front of females. 

Indeed, learning is not a one and done process. Variability in practice is both normal and critical for achieving a precise performance at the end. Besides providing a beautiful brain mechanism that underlies practice variability vs. performance precision, the study by Singh Alvarado et al. also prompted many more interesting questions: how does the brain balance practice variability and performance precision? Specifically, how is the dynamics of noradrenergic signaling regulated? We still have much more to learn about learning!

Figure 1. a, Example ROI SN activity aligned to undirected (blue) and directed (red) singing (blue and red bars, song motifs). b, Pitch variability across undirected (U) and directed (D) singing. c, Average SN population activity for undirected and directed singing. d, Experimental approach for simultaneous SN photometry and drug infusion experiments. e, Example SN photometry recordings from one bird during infusion of clonidine (a2 adrenergic receptor agonist) into the sBG; dashed line indicates first syllable of motif during infusion of clonidine into the sBG. (Adapted from Singh Alvarado et al. (2021))

Reference:

Singh Alvarado J, Goffinet J, Michael V, Liberti W 3rd, Hatfield J, Gardner T, Pearson J, Mooney R. Neural dynamics underlying birdsong practice and performance. Nature. 2021 Nov;599(7886):635-639. doi: 10.1038/s41586-021-04004-1. Link