Dopamine is probably the most studied neurotransmitter, and yet the neuroscience literature contains a huge variety of perspectives on its functional role. This post summarizes a systems-level perspective on the function of dopamine that has motivated several successful drug studies and informed the construction of artificial neural network models. The details of this perspective are maddeningly complex (at least for me), which is why I thought it would be useful to summarize it here, in the simplest terms possible.
There are at least two ways to talk about dopamine release. We can talk about the average amount of dopamine available over a relatively long period of time (known as “tonic dopamine levels”), or we can talk about how much is being released in a given burst of neuronal firing (known as “phasic release”). Phasic release of dopamine is often thought to encode “reward prediction error” – some very compelling experiments show that dopamine is released in bursts when animals underestimate the reward of an event, but sharply dips in activity when animals overestimate or do not receive an expected reward from a given event.
But it’s also important to distinguish between phasic and tonic dopamine because they are often argued to have different (and even contradictory) effects. For example, some argue that increases in tonic dopamine cause inappropriately frequent updating of working memory, leading to distractibility. In contrast, increases in phasic dopamine may cause more effective updating of task-relevant information, leading to improved working memory.
So, how exactly do tonic and phasic dopamine interact? At first blush, tonic dopamine seems like the temporal derivative of phasic dopamine: as more dopamine is released in a given burst, the tonic level of dopamine increases. But there are very important nuances here. For example, This 2002 article by Cohen Braver & Brown argues that higher tonic dopamine levels are likely to be associated with relatively lower effects of phasic release, perhaps due to a kind of nonlinear “saturation” effect: if more dopamine is already available in the synaptic cleft, then the additional phasic release of DA is likely to have a smaller effect.
More recent analysis by Frank & O’Reilly gives a better idea of how complex the effects of dopamine may truly be. This model extends earlier work by specifying exactly how two distinct dopaminergic pathways may control working memory and learning.
Here’s the cartoon version of this model: imagine “working memory” as a loop in prefrontal cortex through which information will cycle indefinitely. When dopamine is phasically released this loop is temporarily opened, to allow different information to be cycled within working memory. Since phasic dopamine release is related to rewards, this ensures that working memory will be filled with the information that is most relevant for the organism to encounter rewards in the environment.
But clearly there is a balance that must be achieved here: if the loop is opened too frequently, it leads to distraction and “hyperactive updating” of memory. In contrast, if the loop is not opened up frequently enough, working memory maintenance is good but there is a flexibility cost: memory is not often updated when it should be. This tradeoff is known as the stability-flexibility dilemma, and is closely related to the exploration-exploitation dilemma reviewed earlier this week.
More technically speaking, this model portrays prefrontal cortex as basically a “storage space” for information that needs to be actively maintained. Entry to this storage space is provided by bursting activation within the “Go” pathway of the striatum (a part of the basal ganglia), which has the end result of updating working memory by gating information currently in the thalamus into prefrontal cortex. However, this “go” pathway directly competes with a default “NoGo” pathway, which itself has the opposite effect – in other words, the NoGo pathway suppresses information in the thalamus and thereby allows the prefrontal “storage space” to continue to maintain its current information.
These two pathways are both affected by dopamine, but in opposite directions: the striatal Go pathway predominantly contains excitatory D1 receptors, whereas the striatal NoGo pathway predominantly contains inhibitory D2 receptors.
Therefore, responses may be more likely when the efficacy of D2 receptors is increased over the long term (through D2 agonists): this inhibits the No-Go pathway, and ultimately disinhibits the thalamus. The opposite effect could be expected of D2 antagonists, which would increase response thresholds by disinhibiting No-Go neurons and therefore ultimately inhibiting the thalamus.
Importantly, these effects are caused by how these drugs act on postsynaptic D2 receptors – in other words, how they affect the neurons that are receiving the dopamine. These drugs act even more strongly on the neurons that are sending dopamine, because those neurons have presynaptic D2 receptors. Presynaptic receptors seem to act as a kind of negative feedback signal: if they become active, it’s likely that too much dopamine was released, and so phasic responses are then curtailed.
So, if you increase the efficacy of D2 receptors, you may increase tonic levels of dopamine (by upregulating the efficacy of dopamine on postsynaptic receptors) but you will also decrease phasic levels of dopamine (by upregulating the efficacy of dopamine on presynaptic receptors, which then tells the sending neurons to release less or stop releasing dopamine).
Thus the overall picture is that D2 agonists increase tonic dopamine levels (lowering response thresholds) but decrease phasic dopamine release (decreasing positive feedback learning which requires the “Go” pathway, and also limiting the amount of task-relevant updating of working memory). In contrast, D2 antagonists decrease tonic dopamine levels (raising response thresholds) but increase phasic dopamine release (increasing positive feedback learning whichi is reliant on the “Go” pathway, and increasing the task-relevant updating of working memory).
This theory becomes even more complicated in practice, where individuals differ in terms of working memory (and thus, presumably, dopamine). In fact, certain genes have been associated with working memory via their effects on dopamine: individuals with the met allele of the COMT gene have higher levels of tonic dopamine, which means it is harder to update working memory; they also show the expected deficits in task switching. Dopamine agonists decrease the working memory performance of individuals with this version of COMT, but often help those with the val allele of COMT (reviewed previously).
Consistent with the idea that drugs targeting the dopamine system may have different effects depending on subject’s working memory capacity, Frank & O’Reilly found that low working memory span subjects experienced greater effects of D2 drugs.
So far we have discussed only D2 receptors, which are much less prevalent in PFC than D1 receptors. Unfortunately it can be difficult to disentangle the roles of D1 and D2 activity. Nonetheless, it appears that D1 receptors are important for recurrent connectivity in prefrontal cortex – D1 activity seems to uphold the “stability” half of the stability-flexibility dilemma.
Eyes, Window to the Soul – And to Dopamine Levels?
Enhancement of Working Memory
Models of Dopamine in Prefrontal Cortex
Exploration & Exploitation Balanced by Norepinephrine and Dopamine