Figure 1.
Hierarchical organization of behavior and the cortico-BG circuit.
A, An example of a decision hierarchy for two alternative choices: drug vs. food. Each course of action is represented at different levels of abstraction, supposedly encoded at different cortico-BG loops. Seeking each of the two types of reward might follow a punishment of magnitude 16. B, Glutamatergic connections from different prefrontal areas project to striatal subregions and then project back to the PFC through the pallidum and thalamus, forming several parallel loops. Through the striato-nigro-striatal dopamine network, the ventral regions of the striatum influence the more dorsal regions. vmPFC, ventral medial prefrontal cortex; OFC, orbital frontal cortex; dACC, dorsal anterior cingulate cortex; SMC, sensory-motor cortex; VTA, ventral tegmental area; SNc, substantia nigra pars compacta. Figure 1B Modified from ref 21.
Figure 2.
Motivation for food vs. drug at different levels of abstraction (simulation results).
In the first 150 trials where no punishment follows the reward, the value of seeking natural rewards at all levels converge to 10 (A). For the case of drug, however, the direct pharmacological effect of drug (, set to
) results in the asymptotic value at each level to be
units higher than that of one higher level of abstraction (B). Thus, when followed by punishment, whereas cognitive loops correctly assign a negative value to drug-seeking choice, motor-level loops find drug-seeking desirable (positive value). The curves in this figure show the evolution of values in “one” simulated animal and thus, no statistical analysis was applicable.
Figure 3.
Dopamine efflux at different striatal subregions in response to drug-associated cues (simulation results).
In line with experimental data [17], the model shows (left column) that in response to drug-associated cues, there will be dopamine efflux in the ventral striatum, after limited and extensive training. In more dorsolateral subregions, however, cue-elicited DA efflux will develop gradually during the course of learning. The model predicts (second column from right) that this delayed development of cue-elicited DA efflux in dorsal striatum depends on the DA-dependent serial connectivity that links the ventral to the dorsal striatum. That is, as a result of disconnecting the DA spirals, whereas cue-elicited DA response remains intact in the ventral striatum, it significantly decreases in the dorsolateral striatum. Moreover, the model predicts (third column from right) similar results for cue-induced DA efflux in dorsolateral striatum for the case of lesioned ventral striatum. Finally, if after extensive drug-cue pairing in intact animals, a punishment follows drug, the model predicts (right column) that drug-related cue results in inhibition of the ventral leg of DA spirals, even after limited training. In more dorsal regions, however, DA efflux decreases slowly during learning, but will remain positive, even after extensive drug-punishment pairing. The data presented in this figure are obtained from “one” simulated animal and thus, no statistical analysis was applicable.
Figure 4.
Blocking effect for natural vs. drug rewards.
The model predicts that blocking occurs for natural rewards (A) and drugs (B), only if the initial training period is “extensive”, such that the first stimulus fully predicts the value of the outcome. After “moderate” training, cognitive levels that are more flexible fully predict the values and thus, block further learning. However, learning is still active in low-level processes when the second training phase (simultaneous presentation of both stimuli) starts. Thus, our model predicts that moderate initial training in a blocking experiment with natural rewards will also result in cognitive/behavioral inconsistency. The data presented in this figure are obtained from “one” simulated animal and thus, no statistical analysis was applicable.