Cross-modal representations of first-hand and vicarious pain, disgust and fairness in insular and cingulate cortex

Abstract

The anterior insula (AI) and mid-anterior cingulate cortex (mACC) have repeatedly been implicated in first-hand and vicarious experiences of pain, disgust and unfairness. However, it is debated whether these regions process different aversive events through a common modality-independent code, reflecting the shared unpleasantness of the experiences or through independent modality-specific representations. Using functional magnetic resonance imaging, we subjected 19 participants (and 19 confederates) to equally unpleasant painful and disgusting stimulations, as well as unfair monetary treatments. Multivoxel pattern analysis identified modality-independent activation maps in the left AI and mACC, pointing to common coding of affective unpleasantness, but also response patterns specific for the events' sensory properties and the person to whom it was addressed, particularly in the right AI. Our results provide evidence of both functional specialization and integration within AI and mACC, and support a comprehensive role of this network in processing aversive experiences for self and others.

Figure 1. Trial structure and behavioural results.

For (a) pain, (b) disgust and (c) ultimatum game (UG) tasks, trials were introduced by an arrow-shaped cue whose orientation and colour informed whether the upcoming event affected either the participants themselves (↓) or the confederate (↙). For the UG, cues were also associated with the photo of an unknown proposer. Cues (2 s) were followed by a stimulus event of varying unpleasantness (aversive versus neutral). For the pain task, this consisted of a 1-s-long electrical stimulation delivered to the left wrist. For the disgust task, 0.5 ml of liquid was delivered into the mouth, which had to be swallowed after 4 s when prompted by an instruction screen (3 s). To avoid transfer to the next trial, 2 ml of water were then delivered in the mouth (3 s), again followed by a 3-s swallowing instruction. For the UG, a 4-s screen was presented depicting the offer with 10 coins organized into 2 piles. At the bottom of the screen, the response options ‘accept/reject' were presented. For each task, stimuli were followed by a visual analogue scale, which participants used to rate the subjective unpleasantness evoked by the previous stimulation. Inter-trial intervals (ITI) in all three tasks varied from 2.5 to 7.5 s (average 5 s). For each task, average unpleasantness ratings are displayed in separate subplots (single experiment, sample size N=19). White bars refer to stimuli directed to the participants (self), whereas striped bars refer to stimuli directed to the confederates (other). For each subplot, aversive and neutral events are displayed separately along the horizontal axis. Error bars refer to bootstrap-based 95% confidence intervals.

Figure 2. ROIs and MVPA overview.

(a) Independent component analysis performed on participants' resting-state data, to map the AI–mACC network in an unbiased manner (surface rendering displayed at P<0.001). ROIs were based on significant clusters in the bilateral AI and mACC that were then used for MVPA on independent functional data obtained for the pain, disgust and UG tasks. (b) Processing steps in MVPA. For each ROI, data were extracted from every constitutive voxels and fed to a SVM classifier, to identify multivoxel patterns informative about an aversive event relative to its corresponding neutral control (for example, Ps versus nPs). The so trained classifier was then tested on independent data from the same task (within-task classification) or another task, for example, when aversive experiences occur in a different sensory modality (cross-modal classification). (c) Matrixes displaying the results of within-task, cross-target and cross-modal classifications for each of the three ROIs. In each matrix, row and column labels refer to classification of an aversive state, each relative to its corresponding neutral control: self-related pain (Ps), self-related disgust (Ds), other-related pain (Po), other-related disgust (Do) and unfairness (U). D' values from within-task classifications are displayed in the outer diagonal line of the matrixes. The remaining cells in the matrixes reflect cross-target and cross-modal classifications. White striped cells refer to d′ effects which are not significantly higher-than-chance, whereas green cells refer to effects significantly higher-than-chance. The luminance of green cells reflects the magnitude of the d′ values. Single experiment, sample size N=19.

Figure 3. Bar plots displaying d′ values from MVPA analyses in left AI, mACC and right AI.

(a) Self-pain and -disgust. Bar plots display d′ values representing the ability of a linear kernel SVM classifier to detect activity patterns characteristic for first-person experience of pain and disgust in ROIs in left AI, mACC and right AI. Ps and Ds refer to within-task classifications of self-related pain (Ps) and self-related disgust (Ds) compared with their respective neutral controls. Ps↔Ds refers to cross-modal classifications between both self-related aversive events. (b) Others' pain and disgust. D' values reflect within-task classification of other-related pain (Po) and disgust (Do) compared with their respective neutral controls, as well as cross-target classifications between aversive events directed at self and other for same modality (Ps↔Po; Ds↔Do) and different modality (Ds↔Po; Ps↔Do). (c) Ultimatum game. Bar plots display d′ values related to the within-task classification of unfair events (U) compared with midfair events; and the cross-modal classification between unfairness and first-person pain (Ps↔U) or first-person disgust (Ds↔U). Bars are coloured consistently to the corresponding cells in the classification matrixes displayed in each panel. The significance of permutation tests comparing d′ values against chance (or against values from other conditions) are also reported. **P<0.05 corrected for multiple comparisons for the three ROIs; *P<0.05 uncorrected. Error bars refer to bootstrap-based 95% confidence intervals. Single experiment, sample size N=19.