Pain-Related Expectation and Prediction Error Signals in the Anterior Insula Are Not Related to Aversiveness

Abstract

The anterior insula has repeatedly been linked to the experience of aversive stimuli, such as pain. Previously, we showed that the anterior insula is involved in the integration of pain intensity and its prior expectation. However, it is unclear whether this integration occurs by a pain-specific expectation or a more general expectation of an aversive event. To dissociate these possibilities, we conducted an experiment using painful stimuli and aversive pictures with three levels of aversiveness on human male volunteers. Stimuli were preceded by a probabilistic, combined modality and intensity cue in a full factorial design. Subjective ratings of pain intensity and skin conductance responses were best explained by a combination of actual pain intensity and expected pain intensity. In addition, using fMRI, we investigated the neuronal implementation of the integration of prior expectation and pain intensity. Similar to subjective ratings and autonomic responses, the dorsal anterior insula represented pain intensity and expectations. The ventral anterior insula additionally represented the absolute difference of the two terms (i.e., the prediction error). The posterior insula only represented pain intensity. Importantly, the pattern observed in the anterior insula was only present if the cued modality was correct (i.e., expect pain); in case of an incorrect modality cue (i.e., expect aversive picture), the ventral anterior insula simply represented pain intensity. The stimulus expectation and prediction error specificity in the ventral anterior insula indicates the integration of expectation with painful stimuli in this area. Importantly, this pattern cannot be explained by aversiveness.SIGNIFICANCE STATEMENT The anterior insula has been shown to integrate pain intensity and their expectation. However, it is unclear whether this integration is pain-specific or related more generally to an aversive event. To address this, we combined painful stimuli and aversive pictures with three levels of aversiveness. The ventral anterior insula represented pain intensity, expectation, and their absolute difference (i.e., the prediction error). Importantly, this pattern was only observed if the cued modality was correct. In case of an incorrect modality cue, this area simply represented as pain intensity. The stimulus expectation and prediction error specificity in the ventral anterior insula indicates the integration of expectation with painful stimuli in this area. Importantly, this pattern cannot be explained by aversiveness.

Keywords: expectation; human fMRI; pain; predictive coding; somatosensory perception.

Copyright © 2018 the authors 0270-6474/18/386461-14$15.00/0.

Figures

Figure 1.

Experimental design and predicted response patterns for INT, INT+EXP, INT+PE, and INT+EXP+PE models. A, In the INT model, the response to the stimulus is only driven by its intensity. B, In the INT+EXP model, the stimulus response is the weighted sum of the stimulus intensity (left) and the stimulus expectation (middle). C, In the INT+PE model, the stimulus response is the weighted sum of the stimulus intensity (left) and the stimulus prediction error (middle). D, For the INT+EXP+PE model, the stimulus response is the weighted sum of the stimulus intensity (left), the stimulus expectation (second from left), and the prediction error (third from left). Right, Examples shown here assume an equal contribution of the expectation and the prediction error signals to the stimulus response. Different contributions of the expectation or prediction error can lead to a different pattern for the stimulus response. E, Trial structure: At the beginning of each trial, a central cue indicated the probability of the intensity and the modality of the upcoming stimulus. Then after a blank screen with variable duration, volunteers received the stimulus (2 s), followed by the rating (2 s). The intertrial intervals were randomly jittered between 2 and 3 s. Volunteers were asked to fixate a central-fixation-dot (except for the rating duration). F, Table illustrating the cue-stimulus contingencies.

Figure 2.

Flowchart of the model selection procedure. A, The EXP+PE+INT model will be considered the winning model if the associated p values of the likelihood ratio tests are <0.05. B, The EXP+INT is considered the best model, if the p value for the likelihood ratio test of EXP+PE+INT against PE+INT is <0.05, but the p value for EXP+PE+INT versus EXP+INT is >0.05. C, The PE +INT model is considered the best model, if the p value for the likelihood ratio test of EXP+PE+INT versus EXP+INT is <0.05, and the p value for EXP+PE+INT versus EXP+INT is >0.05. D, If both p values of the likelihood ratio tests are >0.05, the INT model is considered the best model.

Figure 3.

Behavioral, autonomous responses and NPS to thermal pain and aversive pictures. A, Color bars represent the subjective ratings of warm (38°), medium-painful (46°), and highly painful (48°) stimuli (left) and subjective ratings for aversiveness of IAPS pictures (right). B, SCRs to the pain and aversive pictures. C, NPS responses to painful stimuli and aversive pictures. Error bars indicate SEM across volunteers.

Figure 4.

Voxelwise representation of the best model explaining responses to thermal pain within insular cortex. A, Contrast estimates were fitted to four alternative models of INT, INT+EXP, INT+PE, and INT+EXP+PE (valid modality cue). The INT+EXP+PE model (red) best explains the data in the anterior division of the circular insula (vAI). Color coding is based on likelihood-ratio test model comparison (FDR-corrected, q < 0.05). q values are the adjusted p values found using an optimized FDR approach in voxels within the insula. The INT+EXP model (yellow) best explains the fMRI data in the superior division of the circular insula (dAI). Within the inferior division of circular insula, the INT model (blue) is the winning model. B, When the modality cue was invalid, the contrast estimates were best explained by the INT model (blue) in all regions. Error bars indicate SEM across volunteers.

Figure 5.

Voxelwise map of the respective weights of the individual components of the INT+EXP+PE and INT+EXP models within insular cortex for the valid modality cue conditions. A, Weights for the expectation term are plotted for all INT+EXP+PE representative voxels in insular cortex (i.e., Fig. 4, red voxels). Bar graph represents the average of the expectation weight within voxels best explained by the INT+EXP+PE model. Error bars indicate SEM across volunteers. B, Voxelwise map for the weight of the prediction error term as well as its average for all voxels best explained by the INT+EXP+PE model. C, Map of the expectation term and its average across voxels best explained by the INT+EXP model (i.e., Fig. 4, yellow voxels).

Figure 6.

Color map representing the best model for each voxel at an uncorrected threshold of 0.01 projected on flattened cortex and inflated brain surfaces. A, In addition to the insular clusters depicted in Figure 4, an INT+EXP cluster is also evident within the ACC and visual cortex. There is also a small cluster of the INT+PE model within the cingulate cortex (green). B, For the invalid modality cue, the INT model is the best model across all pain-responsive areas. C, Color-coded model selection for subcortical areas. In parts of the PAG, the INT+EXP model best described the data. For the invalid modality cue, the INT model best explains PAG responses.

Figure 7.

Parameter estimates of ROIs. Mean of the parameter estimates (± SE) at three pain intensities of high, medium, and low in different intensity cue conditions for valid modality cue trials are plotted for left (LH) and right (RH) hemispheres (except for the midline PAG) in each ROI. The fit of the parameter estimates to the best model among the four nested models is also depicted. In the ACC and PAG, the best model was determined as the INT+EXP model. Otherwise, the INT model was the best explanatory model. Bayesian model weights (BICw) for all different models are illustrated in each ROI. Error bars indicate SEM across volunteers.