Stimulating about 20 mouse motor cortex neurons evokes a movement like the learned lever press

The trial-to-trial scatter is the most informative part of the design. Stimulation did not give the same result every time. The population activity it triggered varied from trial to trial, and the induced movements varied along with it ^[6]. Both variabilities were predicted by the state of M1 activity just before stimulation ^[7]. A fixed trigger with a variable outcome points to the cortex's starting condition as part of the cause. The authors conclude that M1 can produce the learned activity if it is in the best starting state and is then triggered by precise inputs ^[11].

The relationship is correlational, measured only within stimulated trials.

The two negative results close off cheaper explanations. A generic twitch evoked by lighting up motor cortex cells would have shown up in the task-disengaged mice, where nothing happened ^[9]. A generic 20-cell kick would have worked on neurons that fire after movement onset, and it did not ^[10]. The effect needed the right cells in an engaged animal. The movement-free trials add one more piece: if the learned activity appears without a movement, movement feedback cannot be what produces it ^[8].

The study does not test whether M1 does this in unperturbed behavior. Stimulation shows that a small, well-chosen ensemble is sufficient to start the pattern under permissive conditions. It does not show that those cells start the press in a mouse that is simply working for a reward, or that silencing them would stop it. The dispute the authors set out to address is described as open ^[2], and the paper's own reference list shows both sides: it includes Kawai et al. (2015), titled "Motor cortex is required for learning but not for executing a motor skill", and Guo et al. (2015), titled "Cortex commands the performance of skilled movement" ^[12].

The scope is narrow. The work was done in mice performing a lever-pressing task, with stimulation confined to functionally defined ensembles in layer 2/3 ^[3]. Nothing here extends to people, or to other layers or movements. The abstract does not include how much of the variance the initial state explains, how many mice were used, how often stimulation produced a movement, or how closely the induced movements matched the learned one. Those figures decide how large the effect is, and they are in the full paper rather than the abstract.