A system-level neural model of the brain mechanisms underlying instrumental devaluation in rats

Abstract

AbstractGoal-directed behaviours are defined by the presence of two kinds of effect on instrumental learning. First, degrading the contingencies between produced actions and desired outcomes diminishes the number of instrumental responses; second, devaluing a reward results in a lower production of instrumental actions to obtain it. We present a computational model of the neural processes underlying instrumental devaluation in rats. The model reproduces the interaction between the basolateral complex of the amygdala (BLA) and the limbic, associative and somatosensory striato-cortical loops. Firing-rate units are used to abstract the activity features of neural populations. Learning is reproduced through the use of dopamine-dependent simple and differential hebbian rules. Constraints from anatomy of the projections between neural systems are taken into account. The central hypothesis implemented in the model is that pavlovian associations learned within the BLA between manipulanda and rewards modulate goal selection through the activation of the nucleus accumbens core (NaccCo). Selection processes happening in the limbic basal ganglia with the activation of the NaccCo decide which outcome is choosen as a goal within the prelimbic cortex (PL). Connections between the BLA and the NaccCo are learned through hebbian associations mediated by feedbacks from the PL to the NaccCo. Information about selected goals from the limbic striato-cortical loop influences action selection in the sensorimotor loop both through cortico-cortical projections and through a striato-nigro-striatal dopaminergic pathway passing through the associative striato-cortical loop. The model is tested as part of the control system of a simulated rat. Instrumental devaluation tasks are reproduced. Simulated lesions of the BLA, the NaccCo, the PL and the dorsomedial striatum (DMS) both before and after training reproduce the behavioural effect of lesions in real rats. The model provides predictions about the effects of still undocumented lesions.