Multivoxel Pattern Analysis Reveals Dissociations Between Subjective Fear and Its Physiological Correlates

Abstract

I n studies of anxiety and other affective disorders, objectively measured physiological responses have commonly been used as a proxy for measuring subjective experiences associated with pathology. However, this commonly adopted 'biosignal' approach has recently been called into question on the grounds that subjective experiences and objective physiological responses may dissociate. We performed machine-learning based analysis on functional magnetic resonance imaging (fMRI) data to assess this issue in the case of fear. Participants were presented with pictures of commonly feared animals in an fMRI experiment. Multivoxel brain activity decoders were trained to predict participants' subjective fear ratings and their skin conductance reactivity, respectively. While subjective fear and objective physiological responses were correlated in general, the respective whole-brain multivoxel decoders for the two measures were not identical. Some key brain regions such as the amygdala and insula appear to be primarily involved in the prediction of physiological reactivity, while some regions previously associated with metacognition and conscious perception, including some areas in the prefrontal cortex, appear to be primarily predictive of the subjective experience of fear. The present findings are in support of the recent call for caution in assuming a one-to-one mapping between subjective sufferings and their putative biosignals, despite the clear advantages in the latter's being objectively and continuously measurable in physiological terms.

A neural signature for the subjective experience of threat anticipation under uncertainty

“…4d, see also Supplementary Results and Supplementary Fig. 7) in the combined validation and prospective generalization datasets (Study 9, n = 65), demonstrating that the SUITAS captured autonomic arousal to some extent but with a smaller effect size, which is in line with previous studies on the dissociation between subjective fear experience and its physiological correlates 66 . Further evidence arguing against the effect of nonspecific negative arousal can be found below ('Comparing the SUITAS with the predictive models of fear exposure and nonspecific negative affect').…”

“…Consistent model weights associated with subjective anxious arousal were observed in the aINS, thalamus, ACC, somatosensory cortices and IOG, while the within-individual models additionally revealed consistent contributions of aMCC, precuneus, PCC, putamen, precentral gyrus and SMA to predict momentary trial-bytrial variations. No single brain region or network was sufficient for accurately predicting anxious experience, underscoring that conscious emotional experiences require a distributed neural representation 10,21,33,52,75,76 2,78,79 . However, how varying levels of uncertainty impact the subjective experience of threat anticipation remained unknown.…”

“…Second, we asked if the SUITAS could capture the anticipation of non-shock related negative events (loss of money) or the anticipation irrespective of valence (anticipation of a positive event, i.e., gain of money). To this end, we tested the SUITAS on an independent dataset that used a monetary incentive delay (MID) task (https://zib.fudan.edu.cn) 64 Third, we accounted for unspecific negative emotional arousal which is inherently associated with uncertain threat anticipation 30,65 Given that physiological responses are expected to (partly) co-vary with subjective emotional experiences albeit with dissociable neural bases 66 , we finally explored whether the SUITAS predicts subjective anxious experience rather than its physiological correlates by applying the SUITAS to binned brain activation maps for anticipation period of five skin conductance levels (SCL) for each participant. We found that the SUITAS predicted SCL (r = 0.216, one-tailed permutation test P < 0.001) to a lesser degree than predicting subjective ratings (r = 0.556, P <0.001; difference in effect size: ∆r = 0.34, one-tailed permutation test P < 0.001, Fig.…”