Electrocortical signatures of prediction error processing during threat avoidance and monetary reward learning
Loading...
Date
Authors
Publisher
Supervisors
Abstract
Approaching reward and avoiding punishment or threat are fundamental ways in which
organisms adapt to changing environments. Achieving these motivational goals demands
cognitive and affective control, broadly defined as the ability of neurocognitive systems to
flexibly organize cognition and behaviour in non-affective and affective situations. In affective
and non-affective reinforcement learning, expectations and their violations, i.e., prediction
errors (PEs) are central for successful reward and threat avoidance learning. PEs can either be
defined as unsigned, reflecting any surprising outcome, or signed, indicating better-thanexpected (positive, PE+) or worse-than-expected (negative, PE−) outcomes. In the
electroencephalogram (EEG), PE processing and other control mechanisms often have often
been linked to frontal midline theta (FMθ, 4–8 Hz). According to the adaptive control
hypothesis (TACH), FMθ plays a key role in indexing control processes during reward
maximization and the regulation of negative affective states. Thus, feedback signalling
monetary non-reward or threat, and high levels of reward and threat PEs are hypothesized to
increase FMθ. Moreover, TACH also makes assumptions about FMθ functioning in individuals
high vs. low in negative affective personality traits, especially trait neuroticism/anxiety.
Specifically, FMθ is expected to be amplified in highly neurotic/anxious individuals. Other
relevant EEG markers in feedback and PE processing include slow wave delta-band (1–4 Hz)
oscillations and the phase-locked Reward Positivity (RewP), both linked to PE+ processing,
and the salience-sensitive P300. However, most EEG research on reinforcement learning
focused involved monetary reward rather than primary threat reinforcer. Thus, the processes
involved in threat avoidance learning and its links to individual differences in trait
neuroticism/anxiety are poorly understood. The present thesis addresses this gap by examining
electrocortical responses to expected and unexpected feedback in reward gain vs. threat
avoidance learning, and their associations with individual differences in trait
neuroticism/anxiety.
In both studies of the present thesis, we measured EEG activity in two reinforcement
learning tasks with different reinforcers following positive vs. negative feedback. In the reward
learning task, feedback signalled monetary reward vs. monetary non-reward, whereas in the
threat avoidance learning task, feedback signalled the successful avoidance of a loud noise burst
vs. its presentation. To my knowledge, these are the first studies in which the intensity of the monetary reward and primary threat reinforcer were titrated to ensure meaningful taskcomparisons. PEs were comprehensively measured via computational modelling (estimated
PEs, ePEs) as well as self-reported trial-by-trial expectations (self-reported PEs, sPEs).
In Study I, we replicated the often-reported feedback valence effect in the reward task,
with amplified FMθ responses to monetary non-reward vs. monetary reward. There was also a
feedback valence effect in the threat avoidance learning task, indicating increased FMθ
responses to feedback signalling threat vs. successful avoidance/safety, although this effect was
substantially smaller than in the reward task. Moreover, we found no evidence for a link
between FMθ and reward and threat prediction errors (ePE & sPE), and explorative analyses
revealed no significant influence of PEs on the feedback valence effects.
In Study II, we strongly focused on reward and threat PE+ since comparisons of positive
feedback trials between reward and threat avoidance learning were less confounded by potential
defensive preparation responses before threat onset. We found that FMθ positively scaled with
the computationally derived ePE+ in both, the reward learning and threat avoidance learning
task. There were no significant correlations between trait neuroticism/anxiety and
cognitive/affective control except of a marginal and selective positive moderation of the link
between FMθ and reward ePE+. Moreover, reward ePE+ were stronger positively correlated to
the RewP and exclusively to delta-band activity, while threat ePE+ were stronger linked to the
P300. The importance of the P300 in reinforcement learning was further supported by its role
in predicting subsequent behavioural adaptation and by exploratory analyses of ePE−.
The present thesis provides first evidence on affective and cognitive control in a direct
comparison of reward and threat avoidance learning. Our studies suggest that while FMθ can
be considered an index for cognitive and affective control, it may not represent a simple on-off
signal but instead shows activity patterns that may depend on the learning context and specific
task demands. The mixed findings on the relationship between trait neuroticism/anxiety and
FMθ, including null results for the conceptually closer threat avoidance learning, were
discussed in relation to present challenges in personality neuroscience. The dissociable
sensitivity of the RewP in reward-related, and the P300 in threat-related reinforcement learning
may point to different representations of the neural reward and threat system. Overall, the
results of the present thesis particularly contribute to a better understanding of the
electrocortical mechanisms underlying threat avoidance learning and provide a helpful starting
point for future research on threat-related reinforcement learning.
Review
Metadata
Contributors
Supervisor:
Dates
Issued: 2025-12-10
Faculty
FB04:Psychologie
Language
en
Keywords
Threat avoidanceReward learningEEGPrediction Error
License
show more
Stolz, Christopher: Electrocortical signatures of prediction error processing during threat avoidance and monetary reward learning. : 2025-12-10.
License
Except where otherwised noted, this item's license is described as CC BY-NC-SA 4.0