Periaqueductal gray matter and medial prefrontal cortex reflect negative prediction errors during differential conditioning

Abstract

Computational models of associative learning posit that negative prediction errors (PEs) arising from the omission of aversive outcomes weaken aversive Pavlovian associations during differential conditioning and extinction. It is possible that negative PEs may underlie exaggerated conditioned responses to the conditioned stimulus not paired with an aversitve outcome (CS-) during differential conditioning and to the conditioned stimulus originally paired with a aversive outcome (CS+) during extinction in patients with clinical anxiety disorders. Although previous research has demonstrated that manipulations of the periaqueductal gray matter (PAG) interfere with extinction learning in animals, the role of the PAG in processing negative PEs within the human brain is presently unclear. We set out to investigate how PAG responses and connectivity are impacted by negative PEs using ultra-high-field (7 T) functional magnetic resonance imaging and hierarchical Bayesian analysis. During differential conditioning, negative PEs were associated with larger responses within the lateral and dorsolateral PAG and increased connectivity between the dorsolateral PAG and medial areas of Brodmann area 9. Collectively, these results shed light on the association between activity within the PAG and medial prefrontal cortex and the omission of aversive outcomes during Pavlovian learning.

Keywords: computational models; medial prefrontal cortex; periaqueductal gray matter; prediction error; safety learning.