Importantly, under this interpretation, new state formation is intact; however, retrieval of appropriate states is disrupted or at least less selective. A second possible explanation (option B in Figure 1, bottom) is that the rats with disrupted cholinergic function might have been able to form a new state in extinction but not in the other challenges. Why would this happen? To answer this, it is useful to ask how the brain knows that a new state should be formed in the first place. One impetus for state

creation is significant differences between the current situation and past experience (Gershman et al., 2010). According to this idea, prediction errors—differences between what is expected (driving is on the right of the road, mass transportation is called “subway,” etc.) and what is currently experienced selleck chemicals llc (cars are on the left, the underground train is “the city circle”)—drive state formation. Importantly,

FG4592 these prediction errors include both errors in predicted value (the city circle is not cheap), and errors in predicted identity (would you expect “the city circle” to indicate an underground train system?). The former are typically termed reward prediction errors (though we use “value,” as changes in rewarding events can also induce identity prediction errors), and Bradfield et al. (2013) refer to the latter as “state prediction errors,” though we prefer “identity,” as any sort of error could lead to recognition of state change. Bradfield et al.’s first two until manipulations—contingency degradation and reversal learning—involved only identity prediction errors,

since the underlying value of the reward associated with lever pressing did not change. However, the last manipulation introduced value prediction errors since the reward was entirely omitted. If cholinergic transmission in the striatum is important for detecting, representing, or learning from identity prediction errors, one would expect to see no new state formation in the first two manipulations due to the cholinergic manipulation, but intact state formation during extinction learning. Thus, like a retrieval deficit, a selective effect on the formation of new states following identity prediction errors would also produce the observed pattern of results (Figure 1, bottom). Though relatively little is known about the function of cholinergic striatal interneurons, what we know so far relates nicely to these two interpretations. For example, one can easily imagine a key role for striatal acetylcholine (Ach) in retrieval: cholinergic interneurons are inhibitory, tonically active, and innervate (and receive input from) a large number of medium spiny neurons (Zhou et al., 2002). This places this local modulatory system in a prime position to provide network-wide inhibition, promoting retrieval of only the relevant state at each point in time (Apicella, 2007). By reducing cholinergic tone, Bradfield et al.