Machine-learning classification of motor unit types in the adult mouse

Motor unit diversity arises from differences in the contractile properties of muscle fibres and the intrinsic electrical properties of their motoneurons. In mice, however, this relationship has not been quantitatively defined, and conventional classification often relies on subjective thresholds. Here, we combine in vivo intracellular recordings with supervised and unsupervised machine-learning methods to test whether motoneuron electrophysiology can predict the physiological identity of mouse motor units. Unbiased clustering identified four groups corresponding to slow (S), fast fatigue-resistant (FR), intermediate (FI) and fast fatigable (FF) types. A multinomial logistic regression model performed well, with most errors occurring between FI and FF types, which showed substantial overlap. Reducing the task to three classes improved accuracy. Feature selection revealed that four electrophysiological properties (input conductance, rheobase, afterhyperpolarization duration and maximal frequency) were sufficient for high predictive performance. Overall, this study provides a quantitative description of mouse motor unit properties and a framework for incorporating motor unit diversity into future investigations of neuromuscular physiology and disease. KEY POINTS: Motor units are traditionally classified as slow (S), fast fatigue-resistant (FR), fast intermediate (FI) or fast fatigable (FF) based on a handful of contractile properties, but in mice this classification has relied largely on subjective thresholds. We used unsupervised clustering of 40 contractile variables recorded in vivo to define motor unit types objectively in the adult mouse triceps surae. Motoneuron electrophysiological properties, including input conductance, rheobase, afterhyperpolarization duration and firing frequency, systematically varied across identified motor unit types. A multinomial logistic regression model predicted motor unit type from motoneuron electrical properties with good accuracy, particularly for slow and fast fatigue-resistant units. These results establish quantitative criteria linking motoneuron excitability to muscle contractile phenotype in mice, providing a framework for studying motor unit diversity in health and neuromuscular disease.