Title: The Impact of Learning on Goal Encoding in Premotor Cortex

Date: Thursday, November 19th, 2015

Time: 11:00am

Location: 2nd floor Auditorium, Learning Research and Development Center (LRDC)

Department: Pitt Bioengineering

Advisor: Aaron Batista, PhD

Abstract: 

The dorsal aspect of the premotor cortex (PMd) is a key node in the cortical pathway for visually-guided reaching. As such, one of the functions it may contribute to is the conversion from visual (input) to muscle (output) coordinates. There is debate about where in the eye-to-arm reference frame transformation PMd sits. My central question in this debate is whether or not the task that the animal is trained to perform may affect the tuning properties of PMd neurons. To address this we recorded from PMd of two Rhesus monkeys while they performed a delayed reaching task. The animals were not trained to fixate. We found that tuning in PMd for reach target location relative to the direction of gaze was quite weak. We recorded neural activity using a 96-channel Blackrock microelectrode array. Nine sessions with 447 well-isolated neurons were analyzed. We first used a planar regression to determine neural tuning. We found that tuning to the location of the target relative to the hand (target-hand reference frame; TH) exhibited a statistically significant regression fit (p <0.05). We also found significant regression fits to the location of the target relative to the eye (target-eye reference frame; TE). However, we recognized that this measurement of TE could be an overestimate. We reasoned that, if the animals consistently looked at their hand during the task, then there would be spurious TE tuning. This is true for any consistent gaze behavior. We found that both animals exhibit consistent gaze behavior patterns during the task, and this meant that at least some of the TE tuning we observed might be an artifact of the nonuniform gaze behavior. To check for this, we performed two additional analyses. We used a partial regression analysis to first remove the tuning due to one reference frame so we could investigate whether the residual variance was tuned in the other reference frame. When the effect of TH was removed, only 36% of our cells exhibited significant but weak tuning to TE. In our second analysis, we simulated cells whose tuning were entirely determined by TH tuning. We built a simulated neural population with TH tuning measured from the real data, but no TE tuning. When those simulated neurons were analyzed like the real data, using the monkeys’ actual gaze behavior, we found they exhibited TE tuning. The amount of TE tuning recovered from the simulation was comparable to the amount of TE tuning measured in the real data. This means that even when no TE tuning is present, TH tuning and nonuniform gaze behavior can produce spurious TE tuning. Our results suggest that neural tuning to the target location relative to the eye is inherently quite weak, weaker still than has been appreciated so far. The TE tuning which others have reported in PMd may arise once animals have been trained to fixate.