Anterior Insula Processes a Subjective Salience Prediction Error

766 

Abstract Submission 

Jae-Chang Kim¹, Lydia Hellrung¹, Stephan Nebe¹, Philippe Tobler¹

¹University of Zurich, Zurich Switzerland

Jae-Chang Kim  
University of Zurich
Zurich Switzerland

Lydia Hellrung  
University of Zurich
Zurich Switzerland

Stephan Nebe  
University of Zurich
Zurich Switzerland

Philippe Tobler  
University of Zurich
Zurich Switzerland

The process of learning is not solely affected by the degree to which outcomes are better or worse than predicted (signed value prediction error; Rescorla and Wagner, 1972; Sutton and Barto, 1998), but also by the level of surprise they elicit, irrespective of their direction (unsigned value prediction error; Mackintosh, 1975; Pearce and Hall, 1980). Surprise is salient and the insula has been associated with salience processing. However, it remains unclear how exactly the insula processes a formal salience prediction error. We studied the insulas role in processing salience prediction errors using a novel Pavlovian conditioning paradigm involving appetitive, aversive, and neutral stimuli as reinforcers.

We studied 41 participants (22.4 ± 0.43 years, mean ± SEM; 19 women) across three lab sessions, including a taste screening session and two main task sessions within an MRI scanner.
During screening, we matched two appetitive and two aversive tastes in unsigned subjective value. Distilled water with the main ionic components of saliva served as neutral liquid. Participants evaluated these liquids by bidding (Becker-DeGroot-Marschak auction; Becker, DeGroot, & Marschak, 1964) and rating (general labelled magnitude scale; Barrett & Simmons, 2015; Bartoshuk et al., 2004).
In the main task, we used Pavlovian associations (pre-trained during the screening session; Figure a) between visual cues and one of the three outcomes (appetitive, aversive, or neutral). Outcome probability varied across cues (p=0, 0.5, 1).
We estimated general linear models that assessed both objective (defined by cues and outcomes) and subjective (based on individual cue and outcome ratings) salience prediction errors. Furthermore, we differentiated the insula's response to appetitive, aversive, and neutral modalities, investigating whether subjective salience prediction error activity reflected general processing of surprise or was specific to outcome kind.

A non-parametric analysis revealed a significant objective salience prediction error signal for valenced outcomes (i.e., appetitive and aversive; Figure b) but not for non-valenced liquids (i.e., neutral) within the insula at the time of the outcome (p<0.05, FWE-whole brain voxel-level corrected). A parametric approach replicated these results and revealed both objective and subjective salience prediction error signals in the anterior insula (Figure c). Direct comparisons showed a significantly stronger association of insula activity with subjective than objective salience prediction errors at the time of the outcome (p<0.05, FWE-whole brain voxel-level corrected; Figure c, top, right side). Moreover, subjective salience prediction errors activated the anterior insula also at the time of the cue (Figure c, bottom, left side). As one would expect based on discounting, this signal was significantly weaker than the signal at the time of the outcome (Figure c, bottom, right side). Finally, separate analyses for the different modalities (i.e., appetitive, aversive, neutral, and no outcome) revealed subjective salience prediction error signals for all modalities in the anterior insula at the time of the outcome (p<0.05, FWE-whole brain voxel-level corrected; Figure d, top). By contrast, at the time of the cue, a subjective salience prediction error occurred only for cues predicting aversive outcomes (Figure d, bottom), compatible with weaker discounting of aversive than appetitive or neutral outcomes.

Our findings demonstrate that the insula actively processes subjective salience prediction errors. These subjective salience prediction error signals appear to be more modality-general at the time of the outcome than at the time of the cue. By extension, the anterior insula appears to compute salience prediction error for aversive stimuli with a different time constant than for non-aversive stimuli, which could explain why some studies view it as a region primarily interested in the aversive domain.

Reward and Punishment ¹

Emotion and Motivation Other

Perception and Attention Other ²

Computational Neuroscience

FUNCTIONAL MRI

Taste

Other - Insula