Dopamine: prediction-error vs. incentive salience

Delay and Trace conditioning. CS = conditioned...

Image via Wikipedia

The exact role that dopamine plays in learning remains controversial; some think it acts as a prediction error signal, while Berrdige et al believe that dopamine codes for incentive salience.


A recent paper throws some light on the issue. It uses a  simple Pavlovian conditioning paradigm. To recap, US and CS are paired and after some time CS starts predicting the reward ; however the twist to usual Pavlovian conditioning is that when CS is presented before US; some rats become ‘sign trackers’ i.e. as soon as the CS comes start engaging with it; while other are ‘goal trackers’ i.e. as soon as CS comes start engaging with where the US would ultimately appear.

To elaborate,  both types of rats are able to learn that CS predicts US , but only sign tracker s attach importance to CS in itself. Also if they are given an option to indulge in instrumental behavior to bring forth the CS (in absence of US) , it seems only the sign trackers are more willing to do work to get the CS and are thus motivated enough by Cs in itself. In other words, while both goal trackers and sign trackers endow CS with predictive capabilities; only sign trackers also endow it with incentive salience.


If all this seems confusing , consider the fact that we are all conditioned to like food/sex; but a secondary reinforcer like money which may predict that food will follow, might become a reward in itself and motivate some of us. while for some money may be as good as it is an indicator of food/sex to come; for other money may acquire an importance/ motivational value in itself.


After that crude analogy, lets return to our sign trackers; these rats are found in wild populations also, but a selectively bred rat breed that has been bred for Novelty preference (bHR)  also displays these behaviors  prominently. On the other hand those selectively bred not to show novelty preference are goal trackers by large. (bLR)

What the authors of this study showed was a dissociation between the necessity of dopamine for learning and performance in Pavlovian conditioning. they showed that while dopamine is not required for learning the prediction part (i.e. even in absence of dopamine both goal trackers and sign trackers could learn that CS predicts US) , it is indeed required for performance (i.e. in absence of dopamine neither goal trackers or sign trackers would perform the task whereby their learning of CS/US association resulted in overt behavior. ) . Further , it was found that for sign tracker dopamine was required for the sign tracking behavior.

To me, and to the authors too,  the results seem to indicate that some individuals are more prone to associate incentive salience to CS and their primary mode of learning is via incentive salience mechanism of dopamine; these are also the one more susceptible to maladaptive behavior. However the learning that results in association of CS with US does not need dopamine; the association can happen without dopamine; but no behavior results if either CS/ US is not able to trigger dopamine release or able to tell the brain that this incentive/stimuli is salient.

To me this bodes victory for the Berridge et al camp of incentive salience theory of dopamine function, to whom I have always been more sympathetic ! do you agree?

Flagel, S., Clark, J., Robinson, T., Mayo, L., Czuj, A., Willuhn, I., Akers, C., Clinton, S., Phillips, P., & Akil, H. (2010). A selective role for dopamine in stimulus–reward learning Nature, 469 (7328), 53-57 DOI: 10.1038/nature09588

Enhanced by Zemanta

Effecient Related Posts:

2 thoughts on “Dopamine: prediction-error vs. incentive salience

  1. Rich

    nice post and an interesting finding. However not really a ‘nail in the coffin for prediction-error’. Incentive salience isn’t much of a theory, more like a redescription of results. While dopamine’s role in signalling prediction-error has a long history of convincing evidence behind it, some of which can only be predicted by associative learning theory, such as negative prediction-error and conditioned inhibitors. See

Comments are closed.