Dopaminergic Systems Can Regulate the Decision Threshold during Reinforcement Learning in Humans

Dopamine is a neurotransmitter that plays several vital roles in the brain. Dopamine is necessary for several cognitive processes, including cognitive control, reinforcement learning, and decision-making. It primarily contributes to learning reinforcement by encoding prediction errors. These errors are deviations from expected outcomes. Various studies have replicated the prediction error codings in dopaminergic regions using functional neuroimaging. However, there is limited research on the behavioral replications and the neural effects of pharmacological modulations of the dopamine neurotransmitter in learning. Dopamine also contributes to motor functions and action selection. In this new study, scientists examined the mechanism of dopamine underlying human reinforcement learning in a within-subjects pharmacological approach.

Dopamine

Prediction errors are dependent on dopamine neuron responses.

The study was carried out on a group of male volunteers using pharmacological combinations of Haloperidol, L-dopa, and a placebo. The study also applied stationary reinforcement learning tasks and functional magnetic resonance imaging. The Researchers aimed to replicate previous results that report the beneficial effects of pharmacological interventions on reinforcement learning. They also hoped to determine if dopamine plays a role in action selection by regulating decision thresholds.

This replication was a failure. The researchers found no difference in performance between L-Dopa and Haloperidol and no evidence of changes in neural prediction error signaling. Bayesian analyses, on the other hand, provided moderate evidence in favor of the null hypothesis. This failed replication is most likely due to a variety of differences in experimental design.

The model comparison revealed that a reinforcement learning drift-diffusion model with different learning rates for positive and negative prediction errors best accounted for the data.

In formal reinforcement learning theory, prediction error signals are crucial. Positive prediction errors are thought to be signaled on a neural level by phasic burst firing of DA neurons, which primarily activate low affinity striatal D1 receptors in the direct pathway that facilitates go learning. Negative prediction errors, on the other hand, are thought to be signaled by phasic dips in dopamine neuron firing rates below baseline, primarily affecting high affinity striatal D2 receptors in the indirect pathway that promotes no-go learning. Summarily, the difference between the obtained and expected reinforcement outcome is encoded by phasic responses of midbrain dopamine neurons.

Clinical significance

Prediction error signals are important in learning. It shows the direction formal reinforcement learning can go. The findings are clinically significant in determining the role of dopamine and its receptors in learning. The result of reduced thresholds following pharmacologically-induced dopamine increases is beneficial in knowing the clinical effects of L-dopa and Haloperidol administration.

Conclusion

The present study did not replicate any beneficial effect of medications on reinforcement learning. However, these medications have an effect on decision thresholds. These findings show the role of dopamine in action selection. The potential of a threshold modulation can bridge circuit-level responses in action selection in the brain. Dopamine is also considered functional in the regulation of response intensity.