Neural population mechanisms of reward-mediated performance

 Across the animal kingdom, heightened rewards invigorate movement behavior. Mice,  monkeys, and humans alike tend to make movements more quickly and accurately when  greater stakes are in play (think game time versus practice). Yet, when these stakes are  too great, humans in particular have demonstrated a propensity to “choke under  pressure”: to fail when we want to succeed the most. Clearly, motivation has a  complicated relationship with motor performance, yet the neural mechanisms linking  changes in rewards at stake to altered behavior remain poorly understood. 

What has been previously demonstrated is that signals of reward proliferate near-globally  throughout the brain. This includes the motor cortex, a region of the brain responsible for  many of the brain’s descending projections to the spinal cord that control voluntary limb  movements. To date, the studies of reward signals in motor cortex have effectively been  limited to documenting that reward can increase or decrease motor cortical activity, not  yet probing its role in reward-mediated performance. In parallel with these studies,  advances in neural recording technologies and accompanying theory have yielded  mechanistic insights of how motor cortex controls movements. By being able to analyze  the activity of hundreds of simultaneously recorded neurons and combining recordings  across days, we now are well-poised to probe the motor cortical mechanisms of reward-mediated performance. 

The focus of this thesis is to study how the motor cortex translates changes in rewards  into changes in performance. To do this, we will examine the spiking activity of  populations of neurons in the primary and dorsal pre-motor cortex recorded from monkeys  as they performed challenging arm movement tasks. In the first part of the thesis, we will  demonstrate that monkeys, like humans, choke under pressure. We trained five monkeys  to perform a prepare-then-reach to target task, where the animals were cued what reward  they would receive for a fast and accurate reach to a displayed target location before they  were permitted to move. While higher reward cues to an extent led to faster and more  accurate reach behavior, the monkeys all choked under pressure for rare and high?magnitude “Jackpot” rewards by reaching too cautiously. This produces an “inverted-U”  characteristic relating performance with reward. 

In the second part, we will explore the animals’ neural activity and identify a potential  neural mechanism for choking under pressure: failure in movement preparation. We will  demonstrate that while the primary effect of reward on motor cortical preparatory activity  appears to be monotonic with the magnitude of cued reward, an interaction between  reward size and reach direction information in motor cortex occurs. This interaction is an  expansion-then-collapse of neural preparatory states for different reach targets as a function of reward, where the decodability of the upcoming movement exhibits the same  inverted-U shaped function with reward as task success rates do. We will demonstrate  that this, along with other motor cortical signals, support the idea that Jackpot rewards  push average motor cortical preparatory states beyond an optimal zone for the upcoming  reaching movement.

 The third and final part of the thesis will more broadly explore the nature of reward signals  in motor cortex with a goal of understanding what they are and how they do (and don’t)  impact behavior. We find that these motor cortical reward signals are not well-explained  by the changes we see in behavior, nor by arousal-like internal state signals. We then  document how reward encoding evolves as the animal performs a task, the reward  signals’ relationships with task-related variability, and how reward alters neural activity  patterns intrinsic to the task in ways that relate to behavior.  

Overall, this thesis characterizes the encoding of reward-like signals in motor cortex and  explores their influence on upcoming movement behavior.