0%

Acetylcholine

Acetylcholine (ACh) is an important neuromodulator throughout the brain, and it plays an important role in the Rubicon goal-driven model, along with dopamine. The basic distinction between these two neuromodulators is that dopamine reflects a reward prediction error (RPE), with the emphasis on error, i.e., relative to a prediction, whereas ACh is more of a raw reward signal without the error-signaling component, and it also fires robustly to novel, unexpected stimuli.

In the Rubicon framework, ACh is critical for modulating the excitability and learning of circuits that should only be engaged at the time of goal selection, and then again at the goal outcome, but not in between. Consistent with this function, ACh fires for CSs that signal future rewards, and for USs that are the rewards themselves. Thus, the onset of a CS provides a goal-selection opportunity, and the corresponding ACh activation then makes the goal-selection circuitry more excitable, so that it can actually fire and engage a goal. Once a goal is engaged, it inhibits the activity of ACh, so the system does not get too distracted during goal pursuit.

Unlike dopamine which is released by two anatomically localized nuclei in the midbrain, there are a number of different ACh release sites throughout the brain, including in the basal ganglia. However, it appears that most of these different sites are coordinated together via the activity of two midbrain nuclei: the laterodorsal tegmental nucleus (LDT) and the pedunculopontine (PPN) nucleus (Huerta-Ocampo et al., 2021). Thus, from a computational perspective, we can treat them as one coordinated signal, at least as a starting point.

BFCN = basal forebrain cholinergic neurons

Biological data

In general, the ACh system is dauntingly complex at a biological level, with multiple different types of muscarinic and nicotinic receptors having opposing responses to ACh binding, and many ACh-releasing neurons also co-releasing other neurotransmitters (Ananth et al., 2023; Abudukeyoumu et al., 2019). This is consistent with an evolutionary process that is constantly tweaking existing neural pathways to optimize behavior in various ways. Given our interest in more generalized mechanisms that support learning and modulation to accomplish well-defined computational functions, we are necessarily oversimplifying and omitting many of these details.

Volume transmission

Like other neuromodulators such as dopamine ACh is typically released via volume transmission over broad neural areas, delivering the ACh signal across all neurons within volumes of neural tissue, with huge numbers of release sites per neuron. Furthermore, individual neurons project to diverse, functionally related targets, effectively binding together distal neural areas under a common neuromodulatory signal (Ananth et al., 2023).

Rapid phasic responding

As with most neuromodulators, there are both tonic and phasic aspects of ACh release. There is clear evidence for rapid, phasic, behaviorally correlated firing in BFCNs (Laszlovszky et al., 2020) consistent with the Rubicon model.

Salience encoding