The anterior insula processes a time-resolved subjective risk prediction error.

Abstract

The insula processes errors in the prediction of risky, motivationally relevant outcomes, and thereby is thought to respond similarly to better than predicted and worse than predicted outcomes. However, the nature of the encoded risk prediction error signals remained unclear. Moreover, the insula was proposed to preferentially process events and stimuli in the aversive domain, rather than in a domain-general fashion. Here we aimed to illuminate these issues. Within a Pavlovian task, participants (n=41; 19 women) rated both cues and outcomes, allowing us to quantify not only objective but also trial-specific subjective risk prediction errors. We found preferential coding of subjective risk prediction errors in the anterior insula and adjacent frontal cortex. This contrasted with preferential coding of objective risk prediction errors in the mid-insula. The anterior insula encoded the subjective risk prediction errors not only at the time of outcomes, but also at the time of cues, in-line with a temporally fine-grained computation of these prediction errors. Cue-induced subjective risk prediction error signals occurred predominantly in the aversive domain, while outcome-induced subjective risk prediction error signals occurred also in the appetitive domain. Domain-specific analyses of risk prediction errors elicited by the preceding outcome at the time of the next cue indicated that the anterior insula updates risk predictions more strongly in the aversive than the appetitive domain. Together, our findings specify the nature of risk prediction errors processed by the anterior insula as subjective, time-resolved, and partly domain-general (outcome), partly domain-preferential (cue), thereby reconciling apparently disparate lines of research.Significant Statement The anterior insula is a hub of the salience network, thought to process both good and bad surprises. However, salience needs defining and the nature of salience signals in anterior insula remained unclear. Here, we define salience as risk and use a Pavlovian task with subjectively evaluated cues and liquid outcomes, allowing us to determine unsigned surprise, i.e., risk prediction errors, at both time points. In a double dissociation, subjective risk prediction error signals occurred preferentially in anterior insula and objective signals in mid-insula. We also show that subjective risk prediction errors are preferentially encoded for aversive rather than appetitive cues. These findings unify divergent frameworks of insula function, specifying the insula's nuanced role in salience processing for motivated behavior.