Abstract
The structure of population activity in the dorsolateral striatum during the performance of motor sequences has not been characterized and it is unclear if striatal ensembles encode (predict) kinematic parameters defining how sequences are executed. Here we analyzed hundreds of striatal spike trains from naive and trained rats performing a running sequence. We found that the population response was composed of a diversity of phasic modulations covering the entire sequence. The accuracy of kinematics encoding by single neurons was around chance level but improved when neuronal ensembles were considered. The distribution of single-neuron contributions to ensemble encoding was highly skewed with a minority of neurons responsible for most of the encoding accuracy. Importantly, running speed ensemble encoding improved after learning. We propose that during motor learning, striatal ensembles adjust their task representation by tuning the activity of a minority of neurons to the kinematic parameters most relevant to motor performance.