Aberrant uncertainty processing is linked to psychotic-like experiences, autistic traits, and is reflected in pupil dilation during probabilistic learning

Abstract

Aberrant belief updating due to misestimation of uncertainty and an increased perception of the world as volatile (i.e., unstable) has been found in autism and psychotic disorders. Pupil dilation tracks events that warrant belief updating, likely reflecting the adjustment of neural gain. However, whether subclinical autistic or psychotic symptoms affect this adjustment and how they relate to learning in volatile environments remains to be unraveled. We investigated the relationship between behavioral and pupillometric markers of subjective volatility (i.e., experience of the world as unstable), autistic traits, and psychotic-like experiences in 52 neurotypical adults with a probabilistic reversal learning task. Computational modeling revealed that participants with higher psychotic-like experience scores overestimated volatility in low-volatile task periods. This was not the case for participants scoring high on autistic-like traits, who instead showed a diminished adaptation of choice-switching behavior in response to risk. Pupillometric data indicated that individuals with higher autistic- or psychotic-like trait and experience scores differentiated less between events that warrant belief updating and those that do not when volatility was high. These findings are in line with misestimation of uncertainty accounts of psychosis and autism spectrum disorders and indicate that aberrancies are already present at the subclinical level.

Keywords: Autism; Belief updating; Hidden Markov model; Psychosis; Pupillometry; Uncertainty; Volatility.

Fig. 1

Probabilistic prediction task. Notes. Figure adapted from Kreis et al. (2021). (a) Example trials with a change of stimulus probabilities on trial 21. In the second, cued task block, this change was preceded by a “change” message on screen. In response, participants had to press “enter” before they could continue with the task. (b) Task structure: probabilities for the left- (p_left) and the right-tilted (1-p_left) stimulus in the first task block (volatile block; solid line) and the second task block (cued block; dashed line). Time points of changes were identical in both blocks (lines are slightly jittered for better readability), as was the order of the different risk conditions. The identity of the majority stimulus in the different risk conditions was inverted in the second as opposed to the first task block. (c) Boxplots displaying the proportion of trials where the majority stimulus was predicted (accuracy) and where choices differed from those on the preceding trial (switches) for the different task blocks and risk conditions, respectively. Means are displayed as crosses

Fig. 2

Relationship between trait and experience scores and proportion of choice switches. Notes. Proportion of choice switches is presented separately for the different task blocks (columns) and risk conditions (color). Trait and experience scores are average scores of AQ (top row) and CAPE-P (bottom row). Points represent values per participant and task condition; lines are regression lines (linear model) to demonstrate trends. Proportion of choice switches was higher under high risk than low risk conditions. This association was moderated by AQ scores, with decreasing differentiation between high and low risk trials as AQ scores increased (see top two panels; interaction effect risk*AQ: β = −0.19, p = 0.03)

Fig. 3

Relationship between trait and experience scores and subjective volatility (transition probability). Notes. Subjective volatility estimates (transition probability) are plotted against trait and experience scores of AQ (top row) and CAPE-P (bottom row), separately for the different task blocks (columns). In accordance with the non-normal distributions of those variables, Spearman correlations are used and ranked values are presented. Statistics of the Spearman correlations are displayed in the top-right corner of each panel, and regression lines (linear model) are added to demonstrate trends

Fig 4

Pupil responses to choice uncertainty (entropy) and Bayesian surprise, moderated by AQ scores. Notes. Pupil responses during outcome presentation to choice uncertainty (entropy; top row) and Bayesian surprise (bottom row) are presented separately for the two task blocks (columns). Colors differentiate between responses for trials with high or low entropy/Bayesian surprise (defined as values within participant-specific upper and lower quartile) and participants scoring high or low on the AQ (defined as values above or below the sample-based median). These quartile- and median-based categorizations of high versus low entropy/Bayesian surprise trials and high versus low AQ scores, respectively, were not used in any of the statistical models and only applied here for illustration purposes. Reddish colors indicate a high, blueish colors a low AQ score, darker shades represent high, brighter shades low entropy/Bayesian surprise values. Mean (solid line) and standard error of the mean (shaded area) were calculated for each sample of the z-scored and baseline-corrected pupil signal during outcome presentation

Fig. 5

Pupil responses to choice uncertainty (entropy) and Bayesian surprise, moderated by CAPE-P scores (C-P). Notes. Pupil responses during outcome presentation to choice uncertainty (entropy; top row) and Bayesian surprise (bottom row) are presented separately for the two task blocks (columns). Colors differentiate between responses for trials with high or low entropy/Bayesian surprise (defined as values within participant-specific upper and lower quartile) and participants scoring high or low on the CAPE-P (C-P; defined as values above or below the sample-based median). These quartile- and median-based categorizations of high versus low entropy/Bayesian surprise trials and high versus low CAPE-P scores, respectively, were not used in any of the statistical models and only applied here for illustration purposes. Reddish colors indicate a high, blueish colors a low CAPE-P score, darker shades represent high, brighter shades low entropy/Bayesian surprise values. Mean (solid line) and standard error of the mean (shaded area) were calculated for each sample of the z-scored and baseline-corrected pupil signal during outcome presentation