Multiple forms of neural processing when repeating voluntary thumb flexions

There is considerable trial-to-trial variability in single cortical neurons when performing the same action repeatedly. One possibility is that neural populations are more stable in representing actions; alternatively, they too may be distinctly engaged from trial-to-trial. To address the nature of variability in large neural populations, we captured the EEG signals time-locked to repeated voluntary thumb flexion movements (∼500 repetitions, 23 subjects). By using non-negative matrix factorization on the low-frequency sensorimotor cortical signals, we quantified the trial-to-trial motor-related potentials (MRPs) in terms of prototypical signals (meta-MRPs) and their corresponding prominence at each trial (meta-trials). Clustering the meta-MRPs across the sampled population revealed 5 distinct signal patterns. There were brain-wide correlates of these meta-MRP clusters. Cortical hemispheres were distinctly recruited from trial-to-trial as certain clusters were accompanied by bilateral motor negativity while others were characterized by ipsilateral motor negativity. The sensory feedback too was distinctly processed from trial-to-trial as the central post-movement positivity was present only for certain clusters. A poorly understood pre-motor positivity accompanied all clusters albeit varying in their timing from trial-to-trial. These patterns – including the time-varying positivity preceding the movement – were rendered invisible by the traditional averaging of the signals. We suggest that the variability in EEG signals is not just noise but a consequence of distinct activation patterns deployed by the cortex. We support the idea that the cortex flexibly switches between distinct forms of neural processing to achieve the same behavioral goals.