Sigmoidal decoding from distinct M1 spiking populations and LFP band power for locomotion speed in mice

The extent to which the primary motor cortex (M1) encodes locomotion speed is relatively unexplored, with some studies suggesting linear or gain-modulated tuning. Here, we show that there are neurons in the mouse M1 where spiking relates to locomotion speed in a manner consistent with a sigmoidal state-transition model implemented by two functionally distinct neural populations. In addition, sigmoidal framework extends to local field potential (LFP) band power, enabling a direct within-animal comparison of spiking and LFP-based speed encoding. We recorded extracellular activity (5,889 single units, 384 channels) using chronic 32-channel laminar arrays in eight mice locomoting on a motorized treadmill over 8 wk, with actual speed tracked via DeepLabCut. Unsupervised clustering of temporal firing rate profiles identified two groups: speed-positively related (70.8%) and speed-inversely related (29.2%) units. Sigmoidal models of speed-positively related rate tuning significantly outperformed linear and quadratic alternatives for both populations, and the two clusters shared a common speed threshold (∼2.3 m/min) consistent with a shared subcortical locomotor gate. When exploring decoding, the minority speed-inversely related population demonstrated significantly higher decoding accuracy via inverse-sigmoid transformation compared with the larger speed-positively related population or all units combined, an advantage that generalized across animals via leave-one-animal-out cross-validation. LFP band power also exhibited sigmoidal tuning relative to locomotion speed, but decoded speed with lower fidelity. These findings suggest a push-pull sigmoidal architecture for spiking-based speed representation in M1 and demonstrate that LFP provides a complementary and stable signal for coarser speed estimation.NEW & NOTEWORTHY This study reveals that locomotion speed is encoded in mouse primary motor cortex through a sigmoidal state-transition mechanism carried by two functionally distinct spiking populations with a shared speed threshold. Using high-density laminar recordings and markerless motion tracking, we show that this framework extends to local field potentials and enables accurate, generalizable decoding. These findings establish a push-pull sigmoidal architecture and highlight implications for stable, calibration-light brain-machine interface design.