A landscape view of motor planning

STEFANO FERRAINA; DEL GIUDICE, PAOLO; BAGLIETTO, GABRIEL

Washington

Conferencia; Society For Neuruoscience 2017; 2017

We recently described (Baglietto et al, Plos One 2017) an approach to the analysis of multi-dimensional time series, such as those obtained from multiple simultaneous electrophysiological recordings, based on a density-based clustering (DBC), and illustrated in a modeling study its potential for efficient representation, inference and complexity estimates. In this contribution we used DBC to analyze the spatio-temporal organization of cortical activity patterns during the execution of a ?countermanding? task, in which a monkey must perform a reaching movement to a target on a screen (?no-stop trials?), unless a stop signal prescribes to withhold the movement (?stop-trials?); no-stop and stop trials are randomly intermixed, and the stop signal occurs at random times within the monkey?s reaction time.  In a previous study using the same task (Mattia et al, J. Neuroscience 2013), a spectral analysis of multi-unit activity (MUA) from single electrode recordings of the dorsal premotor cortex (PMD) suggested that the development of the motor plan proceeds as a cascade of sudden UP-DOWN/DOWN-UP transitions in the local cortical neural population (module). The analysis suggested that transitions would be autonomously generated in most excitable modules, and would propagate to less excitable ones. In their relation with the completion of the motor plan, the time patterns of transitions were predictive of the movement time in no-stop trials, and of the correct or incorrect behavior in stop trials. Here we present preliminary result of the analysis of simultaneous MUA recordings from a 96-electrodes grid in PMD during the same countermanding task. We perform DBC on the 96-dimensional MUA time series, allowing to code it as a discrete sequence of clusters? centroid labels (so each label represents a 96-elements vector of MUA values in a given time bin, which is the centroid of the cluster currently visited by the neural dynamics). This coding offers a clean and compact view of the spatio-temporal dynamics, which confirms the picture suggested by the previous analysis and offers an explicit representation of the task-related activation sequences of cortical modules. Besides, to inquire into long memory effects in the dynamics, from the sequence of centroid labels we first estimated the matrix of transition probabilities between clusters, from which we constructed surrogate Markov centroid sequences. For both the real and surrogate sequences we then computed the Lempel-Ziv complexity, and defined an index measuring their relative complexities, thereby obtaining a quantitative measure of long-memory, non-Markov dynamic components.