The neurotransmitter dopamine is vital to reinforcement learning. The phasic activity of dopamine neurons, which in part encodes reward prediction error, is thought to reinforce the behaviour of animals to maximise chances of receiving reward in the future. The cellular mechanism of this reinforcement learning is believed to be dopamine-dependent long-term plasticity, or more specifically, potentiation of the efficacy of corticostriatal synapses on spiny projection neurons (SPNs) due to phasic dopamine release into the striatum.
However, previous theories have been unable to explain heterogeneous effects on corticostriatal synapses of widespread dopamine activity seen in animal studies (in vivo). Not all events that drive phasic activity of dopamine neurons potentiate all corticostriatal synapses on SPNs. Firstly, phasic activity in dopamine neurons encodes information other than reward prediction error (RPE), such as motivation. Consequently, long-term plasticity should only be induced by the RPE dopamine activity, but not the others. Secondly, out of thousands of active synapses, the dopamine signal should only potentiate or depress the synapses involved in the behaviour that leads to the reward. These issues strike at the very heart of fundamental principles of reinforcement learning.
Striatal cholinergic interneurons (ChIs) develop an excitation-pause-rebound multiphasic response during learning. This pause phase coincides with phasic activity in dopamine neurons. ChIs have therefore been speculated to provide a time window for differentiating ‘rewarding’ phasic dopamine from the others. However, this hypothesis has never been appropriately tested due to technical challenges. Simulating pause in ChIs requires sufficient tonic background firing activity in ChIs. This only happens in intact brains. It is almost technically impossible to manipulate the sparse distributed ChIs to form a synchronised pause in vivo, and it is even harder to trigger phasic activity in dopamine neurons simultaneously.
In a 2018 study published in Neuron and first authored by Dr Yanfeng Zhang, researchers identified the local field potential could be used as the proxy of the firing pattern of striatal ChIs after cortical stimulations, or during the slow oscillation in urethane anaesthetised rats. This finding finally enables the researchers to trigger phasic activity in dopamine neurons to coincide with either the pause or excitation of ChIs.
In a new study published today in Nature Communications, researchers revealed that long-term potentiation (LTP) was induced only when the ChI pause coincides with phasic activity of dopamine neurons, regardless of whether the dopamine neurons were activated by a physiologically meaningful visual stimulation or a train of electrical stimulation. This was the first evidence that the ChI pause is required for dopamine-dependent LTP.
In addition to identifying ChIs as involved in dopamine-dependent LTP, Dr Zhang and colleagues further tested the hypothesis that depolarisation of postsynaptic SPNs is required. Although it has been proposed for decades with ex vivo evidence, this hypothesis has not been confirmed in vivo for dopamine-dependent LTP. The team then performed control experiments and found LTD and not LTP was induced if the SPNs were not depolarised during the period of phasic dopamine release. Therefore, they have provided the first in vivo evidence that LTP is only induced at synapses with depolarised SPNs in dopamine-dependent plasticity.
By using two distinct in vivo preparations, Dr Zhang and colleagues have now demonstrated that long-term potentiation of corticostriatal synapses on SPNs requires the coincidence of phasic activity of dopamine neurons, the pause of striatal cholinergic interneurons, and the depolarisation of SPNs.
The full paper "Coincidence of cholinergic pauses, dopaminergic activation and depolarisation of spiny projection neurons drives synaptic plasticity in the striatum" is available to read in Nature Communications
Text credit to Dr Yanfeng Zhang