A model of amygdalar modulations and interactions in pavlovian conditioning
Résumé
The amygdala is a brain structure known for its implication in pavlovian conditioning, and in its different phases of learning(acquisition, extinction, reversal, habituation). Recent advances in the amygdala comprehension (Lee et al., 2013), and in itsrelations to other regions of the medial temporal lobe (Paz and Pare, 2013), give us a better idea of the neural basis of pavlovianconditioning. Our model of pavlovian conditioning, by modeling the amygdala and its interactions with surrounding structures,including neuromodulation (dopamine, noradrenaline, acetylcholine), gives a possible interpretation about the way pavlovian mod-els like Rescorla-Wagner, Pearce-Hall and Mackintosh (Le Pelley, 2004), could be implemented in cerebral architecture. Such acomputational model also proposes how various phases of learning can occur during the lifespan.From an architectural point of view our computational model of the amygdala underlies three main information flows (Price, 2003).(i) Both the thalamus and the sensory cortex project sensory information to the lateral amygdala nucleus (LA). Such projectionsare modulated by noradrenaline, which helps to favor one projection over the other one, depending on the learning phase (Johnsonet al., 2011). In LA, sensory stimuli (CS, conditioned stimuli) are associated with the arrival of reward, or unconditioned stimulus(US). (ii) In the basal amygdala (BA), theses associations are subsequently modulated by input from the hippocampus, the prelimbic and the infralimbic cortex, in fear-neurons and extinction-neurons (Herry et al., 2008). (iii) Both BA and LA have projectionsto the central amygdala (CeA), into which competition determines the pavlovian answer (Ciocchi et al., 2010). CeA triggers motorpavlovian responses by projecting to the periaqueductal gray and to the hypothalamus, and hormonal pavlovian response, likedopaminergic and cholinergic release, which influence pavlovian learning by respectively enhancing signal-to-noise ratio in LA andBA (Lee et al., 2013), thus enhancing salient predictions learning as in Mackintosh rule, and enhancing CS associability in BA, inthe sense of Pearce-Hall rule (Johnson et al., 2011).From a temporal point of view, acquisition of fear conditioning first takes place in the basolateral nucleus of the amygdala, basedon thalamic inputs. Categorisation occuring in the sensory cortex during the developpement will then provide more adequate information to LA, helping forming more adequate predicting rules in LA, which will take over US predictions in BA. In an oppositeway, for context-based rules, or in the case of extinction or reversal, US prediction from the basal nucleus, which receives contextualinformation from the hippocampus and high-level information from the prefrontal, will predominate (Pauli, 2011).