16 October 2007. Pavlov may not have known it, but his bell got more than saliva flowing. His experiments undoubtedly spurred dopamine release in his dogs’ brains, and probably in his own as well. We now know that the neurotransmitter mediates positive reinforcement associated with reward. More controversial is whether dopamine plays any role in negative reinforcement, as in learning to avoid displeasurable, or non-rewarding stimuli. A paper in the October 8 PNAS suggests that it does.
Researchers led by Michael Frank at the University of Arizona, Tucson, have correlated genetic polymorphisms with reward learning in young, healthy adults. Their findings indicate that dopamine plays a role in three independent reward pathways—short-term adaptability, and positive and negative long-term reinforcement. The findings are of interest to schizophrenia researchers given that compromised decision-making is one of the most debilitating facets of the disease.
Frank and colleagues studied the DARPP-32, DRD2, and COMT genes, all three linked to dopamine function: DRD2 is the gene for the D2 dopamine receptor; DARPP-32 codes for dopamine and cAMP regulated phosphoprotein of 32 kDa, which mediates the effects of dopamine D1 activation on synaptic plasticity; and COMT codes for catechol-O-methyl transferase, an enzyme that degrades dopamine. All three are candidate risk genes for schizophrenia (see SRF related news story and SRF news story).
The authors focused on an A/G DARPP-32 polymorphism that modulates striatal function (see SRF related news story). They predicted that this polymorphism might affect reward-based learning, since the DARPP-32 is highly abundant in the striatum, where D1 activation has been linked to decision-making based on positive outcomes. They focused on a C957T polymorphism in the DRD2 gene that affects post-synaptic D2 density, predicting that it might influence decisions associated with negative outcomes. And they looked at the Val158Met polymorphism in the COMT gene, which is associated with changes in dopamine level in the prefrontal cortex (see SRF related news story). The authors write that “this genetic marker of prefrontal DA function would predict the extent to which participants maintain negative outcomes in working memory to quickly adjust their behavior on a trial-to-trial basis.” The prefrontal cortex is an area of particular interest to schizophrenia researchers.
The authors correlated the three polymorphisms with reinforcement learning. Frank had healthy, young undergraduate students take part in a computerized test that simultaneously measures how well positive and negative feedback is learned. Briefly, the volunteers learn that on a probabilistic basis one possibility, “A,” is best chosen, while another, “B,” is best avoided.
The results supported the predictions. Averaging over a number of trials, DARPP-32 AA homozygotes were better than G carriers in the “choose-A” positive reinforcement scenario. In contrast, compared to those carrying the C allele, DRD2 TT homozygotes were much better in the “avoid-B” scenario, suggesting that they learn better from negative feedback. The Val/Met COMT polymorphism had no effect on positive reinforcement learning or avoidance learning over the long term. However, it did affect behavior on a trial-to-trial basis. Volunteers homozygous for the Val allele (and also the lowest prefrontal cortex dopamine) were less likely to alter their response based on a prior negative outcome.
“One of the surprising findings was that we saw such a large effect with individual genes,” said Frank in an interview with SRF. He noted that the results do not suggest 100 percent predictability; in other words, you cannot look at any one person’s genotype and know exactly how he or she is learning, “but nevertheless, across the samples the effect sizes we found were relatively large,” said Frank.
The other surprise was the D2 dopamine receptor role in negative reward learning, which has been a highly debated topic. Frank explained that work in primates has shown there is a pause in dopaminergic firing when the animals fail to get an expected reward. This has led to the suggestion that a dip in dopamine release may be involved in negative feedback learning. “The reason that is controversial is because the pause in these dopamine cells that happens during negative feedback is relatively small and because the baseline firing rate is already pretty low, so when they pause the change in firing rate is not nearly as big as during reward,” he explained. “In our specific model we can account for that because this negative-feedback learning depends on the D2 receptor, which is really sensitive to dopamine levels: essentially it is more sensitive to these small changes than the D1 receptor, which requires greater activation,” he said. He also stressed that this latest finding does not rule out the involvement of other neurotransmitters in avoidance learning.
Do these findings have any significance for schizophrenia research? The disease is certainly linked to dopamine dysfunction (see SRF Current Hypothesis) and also to deficits in the prefrontal cortex. Frank has already done a study in schizophrenic patients, in collaboration with Jim Gold’s lab at the University of Maryland, Baltimore. They found that people with schizophrenia were indeed impaired in that same measure of rapid learning from negative feedback, but they were just fine on long-term integration of negative feedback over many trials (see Waltz et al., 2007).
But though the dopamine hypothesis for schizophrenia is a long-standing one, “there are many other factors that need to be considered in schizophrenia,” said Frank. “However, I do think that studying schizophrenia from a motivational standpoint, looking at reward processing, can be very fruitful. If there is a fundamental dysfunction in reward learning circuitry, that can lead to compounding effects on all sorts of behaviors," he said.—Tom Fagan.
Frank MJ, Moustafa AA, Haughey HM, Curran T, Hutchison K. Genetic triple dissociation reveals multiple roles for dopamine in reinforcement learning. PNAS. 2007 Oct 8;104:16311-16316. Abstract