Motor-skill learning and dendritic spine plasticity

Whether we are tying our shoelaces or scoring in a basketball game, we rely on the brain's ability to learn a movement and then produce it reliably, the same way, every time. Skills like these are not stored in one place. They emerge from an interplay between the cortex and structures buried beneath it, and among these the basal ganglia are essential: they are what allow a movement to be acquired through practice and then consolidated into something automatic. Within them, a population called medium spiny neurons receives the bulk of incoming information, and the tiny protrusions studding their branches — the dendritic spines — are where connections between neurons are strengthened or weakened. They are, in a real sense, the physical substrate on which a learned skill is written.

We know that these neurons change how they behave as an animal learns. What remains unclear is whether that functional change is accompanied by a physical one: whether the spines themselves are remodeled, growing in number, changing in size, or redistributing along the branch. Answering it is difficult less for conceptual reasons than for practical ones. You need an animal that has demonstrably learned something specific, a way to find the neurons that were actually involved in learning it, and enough resolution to see structures near the limit of what a light microscope can resolve. Much of my thesis was building that chain end to end.

I began with the behavior, training mice on a self-paced operant task in which a reward followed a fixed sequence of lever presses, and changing the lever access so that the task progressively required the animal to use one forelimb rather than the other. That constraint is what makes the rest of the analysis possible: it gives you a movement that is both learned and lateralized, so that the trained side of the brain can later be compared against the untrained one. With practice, the animals did what skill learning predicts, becoming faster and organizing their presses into consistent, stereotyped sequences performed predominantly with a single paw.

The next problem was finding the neurons that mattered. For this I used c-Fos, a protein expressed by neurons shortly after they have been strongly active, which acts as a molecular fingerprint of recent engagement. Rather than counting cells by hand in a few chosen regions, I combined immunostaining with an unbiased whole-brain detection and atlas-mapping pipeline, letting me survey activity across the entire brain and ask, without deciding the answer in advance, which regions were engaged by learning. The picture that emerged was subtle but consistent, and it fell where a lateralized motor skill would predict.

Finally, I established the methods to see the spines themselves. This meant optimizing protocols for high-resolution confocal imaging with deconvolution and dedicated software for classifying spine morphology.