UCL Logo

Computing prediction errors in the fear conditioning circuit 

Tad Blair

Department of Psychology, UCLA , USA

Rewards and punishments generate instructive signals in the brain that modify circuits for controlling adaptive behaviors.  These instructive signals are hypothesized to be represented by prediction errors that encode the difference between actual versus expected rewards or punishments. Convergent evidence suggests that prediction errors for reward may be encoded by midbrain dopamine neurons, but the neural substrate for encoding aversive prediction error signals is less well understood.  Our laboratory has investigated this question by recording single-unit neural activity from the lateral/basal (LA/B) amygdala and midbrain periacqueductal gray (PAG) during Pavlovian fear conditioning in freely behaving rats.  Fear conditioning is a simple aversive learning task in which rats are trained to fear a neutral conditioned stimulus (CS) by pairing it with an aversive unconditioned stimulus ( US ). Convergent evidence indicates that LA/B is an important site of neural plasticity where memories of the CS-US association are stored, making it a likely target where aversive prediction error signals could regulate associative plasticity. The amygdala sends outputs to PAG which are thought to mediate expression of conditioned fear responses, but PAG is also a critical locus of pain modulation and may thus play a role in computing aversive prediction errors. Our studies show that during fear conditioning, many LA/B and PAG neurons respond to the US during early training trials when it is unexpected, but not during later trials when it is well predicted by the CS. Many of these same neurons also exhibit conditioned enhancement of their responses to the CS. After conditioning, LA/B and PAG neurons respond preferentially to the US when it is delivered unexpectedly (in the absence of the CS) rather than expectedly (in the presence of the CS). Inactivation of the amygdala abolishes the tendency of PAG neurons to respond preferentially to unexpected presentations of the US . These and other findings suggest that LA/B and PAG may be critical components of a neural circuit for generating an aversive prediction error signal that could regulate a broad range of aversively motivated learning processes.

Workshop HOME