Morrow, M. M. and L. E. Miller. Prediction of muscle activity by populations of sequentially recorded primary motor cortex neurons. J Neurophysiol 89: 2279 -2288, 2003. First published December 18, 2002 10.1152/jn.00632.2002. We have adopted an analysis that produces a post hoc prediction of the time course of electromyogram (EMG) activity from the discharge of ensembles of neurons recorded sequentially from the primary motor cortex (M1) of a monkey. Over several recording sessions, we collected data from 50 M1 neurons and several distal forelimb muscles during a stereotyped precision grip task. Ensemble averages were constructed from 5 to 10 trials for each neuron and EMG signal. We used multiple linear regression on randomly chosen subsets of these neurons to find the best fit between the neuronal and EMG data. The fixed delay between neuronal and EMG signals that yielded the largest coefficient of determination (R 2 ) between predicted and actual EMG was 50 ms. R 2 averaged 0.83 for ensembles composed of 15 neurons. If, instead, each neuronal signal was delayed by the time of its peak cross-correlation with the EMG signal, R 2 increased to 0.88. Using all 50 neurons, R 2 under these conditions averaged nearly 0.97. A similar analysis was conducted with signals recorded during both a power grip and a precision grip task. Quality of the fit dropped dramatically when parameters from the precision grip for a given set of neurons were used to fit data recorded during the power grip. However, when a single set of regression parameters was used to fit a combination of the two tasks, the quality of the fits decreased by Ͻ10% from that of a single task.
I N T R O D U C T I O NSingle neuron recording methods applied to behaving animals have contributed dramatically to our understanding of the limb movement-related signals present in the CNS. Despite this success, however, recording neurons one at a time affords the smallest of windows through which to observe command signals that are encoded across tens or hundreds of thousands of neurons.Nearly 20 years ago, Georgopoulos and his colleagues published a now classic study in which they demonstrated that the discharge of ensembles of neurons recorded sequentially in the primary motor cortex (M1) during a highly stereotypic task could be used to make a post hoc "prediction" of the direction of hand motion (Georgopoulos et al. 1983). Since that time, many other studies have been published using similar methods to explore the nature of population coding of motor command signals (Caminiti et al. 1990;Fu et al. 1993;Kalaska et al. 1989;Moran and Schwartz 1999;Scott and Kalaska 1997;Shen and Alexander 1997; Taira et al. 1996). Although many of these experimental studies have concluded that M1 signals encode movement in a hand-centered coordinate system, several theoretical studies have shown that the directional modulation of M1 discharge may arise from a system that fundamentally controls muscle activation (MussaIvaldi 1988; Todorov 2000). While many studies have used spike-t...