| Abstract No.: | A-D1143 |
| Country: | Canada |
| | |
| Title: | CONVERTING ERROR SIGNALS INTO PREDICTIVE CHANGES IN MOTOR CORTEX |
| | |
| Authors/Affiliations: | 2 Andrea Green*; 1 Reza Shadmehr; 2 John Kalaska;
1 Johns Hopkins University, Baltimore, MD, USA; 2 Université de Montréal, QC, Canada
|
| | |
| Content: | Objectives: As a new motor skill is learned, error signals are used to modify neural circuits so that the motor system functions in an increasingly predictive mode. Previous studies have shown that when novel forces act on the hand during reaching, neural populations in cortical motor areas exhibit learning-related changes in activity (Li et al., 2001; Padoa-Schioppa et al., 2004). However, the process by which sensed errors are transformed into predictive changes remains to be investigated. The goal of this study was to investigate whether there is a relationship between a cell’s directional tuning properties, the error signals it receives and the learning-related changes observed following acquisition of a new dynamic motor skill.
Materials and Methods: We recorded neural responses in the primary motor (M1) and caudal dorsal premotor (PMd) cortex of a rhesus monkey (Macaca mulata) during reaching movements made with a robotic manipulandum (IMT, Cambridge, MA). The baseline tuning properties of each cell were first characterized as the monkey made unperturbed reaching movements in 8 directions. The error signals encoded by each neuron were then documented during blocks of reaching movements in which the trajectory was perturbed by an unlearnable random series of clockwise (CW) and counterclockwise (CCW) velocity-dependent forces applied by the robot perpendicular to the direction of motion (i.e., curl force field). Finally, to examine how error signals are converted into predictive compensatory changes, the monkey was trained continuously in one curl field (CW or CCW) while reaching either in the movement direction in which the cell experienced the largest error signal, or in its preferred direction.
Results: During unlearnable perturbations, caudal M1 neurons exhibited robust activity changes relative to baseline (error signals) at latencies as short as 100 ms. The largest error signals were observed during movements roughly orthogonal to the cell’s preferred direction. Learning in the maximum-error direction gave rise to compensatory response changes in about 70% of the cells. However, changes were mainly limited to the period after movement onset. In 47% of cells these changes could nonetheless be considered predictive in the sense that they occurred in a period before proprioceptive feedback (i.e., within 100 ms after movement onset). In more rostral penetrations, M1/PMd cells showed little response to error. However, these cells developed learning-related changes in the delay period before movement onset.
Conclusions: We conclude that: 1) The caudal M1 neurons that experience the strongest error signal are not those most active during the movement in which the error is sensed but rather those most influential in correcting the movement; 2) Predictive changes in these cells occur after movement onset, suggesting that they may be conveyed mainly via internal feedback pathways; 3) In more rostral M1/PMd cells, learning-related changes occur before movement onset but are less well correlated with an error signal, suggesting they are feed-forward changes conveyed from other brain areas.
|
| | |
| Back |
|
