Encoding of natural variability in a dexterous motor skill over multiple days in the cortex of freely behaving mice

Skilled, goal-directed movements exhibit trial-to-trial variability even in experts, particularly in response to dynamic environmental conditions or when perfect repetition is not required for success. Identifying where, to what extent, and how stably this variability is encoded in the nervous system is essential for understanding how learned movements are robustly maintained over time yet flexibly executed on each trial. We record calcium fluorescence activity in forelimb motor cortex (M1), a key node in the multi-areal network responsible for movement control, in freely-moving mice of both sexes as they performed a self-paced, precision reach-to-grasp task. High trial counts and rich single-trial variability enable rigorous statistical analysis of moment-to-moment movement encoding across matched behavioral sets over five days. Approximately 80% of recorded neurons significantly encoded paw, digit, and head movements during reaching, as quantified using linear models. Across days, encoding similarity shows a small but measurable decline that increases with the interval between recording sessions. This drift is heterogeneously distributed across the population, with many neurons retaining high encoding similarity even in sessions four days apart, as assessed using shuffle controls and comparison to encoding for trial-averaged movements. Thus, over the timescale examined, M1 is capable of maintaining stable encoding of movement details at the level of single cells, even for complex, sensory-guided tasks like reach-to-grasp. Together, these results raise the question of whether downstream circuits support consistent behavior by preferentially relying on neurons with greater stability or instead through population-level readout that is robust to a modest level of representational change.

Skilled actions must be both precise and flexible, adapting to changing situations and environments. How the brain achieves consistent performance when the circumstances of each attempt rarely repeat remains poorly understood. By tracking hundreds of neurons in mouse motor cortex during a precision reach-to-grasp task, we show that individual cells stably encode fine details of paw, digit, and head movements across multiple days. This stability provides a reliable foundation for consistent execution while supporting adaptability in real-world contexts. Our findings indicate that motor cortex integrates postural information with skill-specific forelimb and digit signals, revealing how single neurons support robust yet flexible motor control, a capacity that remains a major challenge for artificial systems.