What’s the science?
We use feedback from rewards every day to learn new things. For example, if we are offered a mango and we have enjoyed several mangos previously, over time we might learn to favor mangos. Research on the neural underpinnings of this form of reward-based learning typically focuses on short-term learning (across several minutes). However, we don’t know what happens when learning occurs over several weeks time, as it might in many everyday situations. Gradual learning from reward feedback relies on a dopaminergic system in the brain, but short-term learning paradigms in typical experiments may also rely on short-term (‘working’) memory systems. This week in the Journal of Neuroscience, Wimmer and colleagues used behaviour and functional magnetic resonance imaging (fMRI) to understand the mechanisms underlying reward learning over a period of several weeks versus within a single session in humans.
How did they do it?
The authors completed two similar studies for replication purposes. In the first study, 33 participants completed a behavioural and fMRI experiment, while in the second study, 31 participants completed a behaviour-only experiment. In both studies, the participants' task was to learn whether the best response to a stimulus (scenes presented on a screen) was either ‘Yes’ or ‘No’ (the wording was arbitrary). The stimuli had been randomly assigned by the experimenter as either reward-associated or loss-associated. Reward-associated stimuli resulted in the participant winning $0.35 for ‘Yes’ (on average) and losing $0.05 for ‘No’. Loss-associated stimuli resulted in the participant losing $0.25 when ‘Yes’ was selected and gaining $0.0 when ‘No’ was selected. Feedback was given after each trial, and feedback was probabilistic, meaning there was an 80% chance that the best response would result in the best outcome/payment. In an initial learning session in the lab, participants learned about 8 'spaced' stimuli. Next, three learning sessions for the ‘spaced’ stimuli were done online (over ~ two weeks). In a lab session about two weeks later, participants learned about 8 new 'massed' stimuli for the same number of times as the previously seen 'spaced' stimuli.
After learning, participants also rated whether they thought the stimuli were reward-associated or loss-associated. In the first study, fMRI data were also collected after learning was completed. Finally, three weeks later, participants tried to remember and rate whether each stimulus was reward-associated or loss-associated. This was completed online in the first study and in the lab in the second study.
What did they find?
In both the ‘spaced’ and ‘massed’ conditions, participants learned the best 'Yes' or 'No' response quickly, and performance was equivalent at the end of learning for both the ‘massed’ and ‘spaced’ training sessions. However, three weeks after learning, participants remembered the value of the ‘spaced’ stimuli much better. Additionally, during learning, working memory capacity was associated with learning after participants got used to the task (in the ‘massed’ session). These results indicate that the spacing out of learning sessions results in better long-term memory for whether stimuli are reward-associated or loss-associated, and that working memory is used during shorter learning paradigms (the ‘massed’ paradigm). During fMRI, using a searchlight (multivariate pattern analysis) approach, the authors discovered that there were very different patterns of activity for reward-associated versus loss-associated stimuli for ‘spaced’ stimuli but not ‘massed’ stimuli within the medial temporal cortex and prefrontal cortex. Patterns of brain activity were also different between the ‘spaced’ and ‘massed’ conditions in the striatum, a region of the brain known to be involved in reward learning. These results suggest that the neural mechanisms underlying ‘spaced’ and ‘massed’ learning may be partially different.
What’s the impact?
This study is the first to clearly demonstrate the effect of learning over a period of several weeks versus minutes on the maintenance of learning and the neural underpinnings of learning. The results have implications for studies of disorders that may involve changes in reward learning, such as addiction and mood disorders.
Wimmer et al., Reward learning over weeks versus minutes increases the neural representation of value in the human brain. Journal of Neuroscience (2018). Access the original scientific publication here.