Learning Task-Related Activities From Independent Local-Field-Potential Components Across Motor Cortex Layers.

Gonzalo Martín-Vázquez,Yoshikazu Isomura,Toshitake Asabuki,Tomoki Fukai

doi:10.3389/fnins.2018.00429

Abstract

Motor cortical microcircuits receive inputs from dispersed cortical and subcortical regions in behaving animals. However, how these inputs contribute to learning and execution of voluntary sequential motor behaviors remains elusive. Here, we analyzed the independent components extracted from the local field potential (LFP) activity recorded at multiple depths of rat motor cortex during reward-motivated movement to study their roles in motor learning. Because slow gamma (30–50 Hz), fast gamma (60–120 Hz), and theta (4–10 Hz) oscillations temporally coordinate task-relevant motor cortical activities, we first explored the behavioral state- and layer-dependent coordination of motor behavior in these frequency ranges. Consistent with previous findings, oscillations in the slow and fast gamma bands dominated during distinct movement states, i.e., preparation and execution states, respectively. However, we identified a novel independent component that dominantly appeared in deep cortical layers and exhibited enhanced slow gamma activity during the execution state. Then, we used the four major independent components to train a recurrent network model for the same lever movements as the rats performed. We show that the independent components differently contribute to the formation of various task-related activities, but they also play overlapping roles in motor learning.

Highlights

While task-related neural activities have been studied in different layers of cortical microcircuits (de Kock and Sakmann, 2009; Isomura et al, 2009; Harris and Mrsic-Flogel, 2013; Masamizu et al, 2014; Manita et al, 2015; Takeda et al, 2015), direct recordings of presynaptic inputs to local cortical circuits are still technically challenging in behaving animals
To identify independent sources of activity contributing to the local field potential (LFP), we applied independent component analysis (ICA) to the raw LFP from the whole recording and obtained spatial voltage distribution and time course of each independent component (IC)
The results obtained were qualitatively the same when applying ICA only to the movement periods, so we show the results for the whole recording analysis because it can be extracted more information of the independent activity

Summary

Introduction

While task-related neural activities have been studied in different layers of cortical microcircuits (de Kock and Sakmann, 2009; Isomura et al, 2009; Harris and Mrsic-Flogel, 2013; Masamizu et al, 2014; Manita et al, 2015; Takeda et al, 2015), direct recordings of presynaptic inputs to local cortical circuits are still technically challenging in behaving animals. The lack of input information makes it difficult to address how inputs from different brain regions contribute to the learning of task-related cortical activities We ask this question by using the local field potentials (LFPs) recorded from the motor cortex of the rats performing a voluntary sequential arm movement (Isomura et al, 2009). If neural activities in different regions targeting the motor cortex are partly correlated with one another, the components extracted by ICA would not represent exact inputs from different brain areas. They can be, at least approximately, regarded as independent inputs converging to the primary motor cortex through multiple synaptic pathways

Methods

Results

Conclusion