Atypical audiovisual temporal function in autism and schizophrenia: similar phenotype, different cause

Abstract

Binding across sensory modalities yields substantial perceptual benefits, including enhanced speech intelligibility. The coincidence of sensory inputs across time is a fundamental cue for this integration process. Recent work has suggested that individuals with diagnoses of schizophrenia (SZ) and autism spectrum disorder (ASD) will characterize auditory and visual events as synchronous over larger temporal disparities than their neurotypical counterparts. Namely, these clinical populations possess an enlarged temporal binding window (TBW). Although patients with SZ and ASD share aspects of their symptomatology, phenotypic similarities may result from distinct etiologies. To examine similarities and variances in audiovisual temporal function in these two populations, individuals diagnosed with ASD (n = 46; controls n = 40) and SZ (n = 16, controls = 16) completed an audiovisual simultaneity judgment task. In addition to standard psychometric analyses, synchrony judgments were assessed using Bayesian causal inference modeling. This approach permits distinguishing between distinct causes of an enlarged TBW: an a priori bias to bind sensory information and poor fidelity in the sensory representation. Findings indicate that both ASD and SZ populations show deficits in multisensory temporal acuity. Importantly, results suggest that while the wider TBWs in ASD most prominently results from atypical priors, the wider TBWs in SZ results from a trend toward changes in prior and weaknesses in the sensory representations. Results are discussed in light of current ASD and SZ theories and highlight that different perceptual training paradigms focused on improving multisensory integration may be most effective in these two clinical populations and emphasize that similar phenotypes may emanate from distinct mechanistic causes.

Keywords: autism; causal inference; multisensory integration; schizophrenia; speech.

Figure 1

Causal Inference Model of Audiovisual Speech. Participants possess a prior tendency to bind sensory information together across time (P(c=1), at the top) and sample the sensory world with a certain degree of noisiness (σ_{sensory noise}, at the bottom). When auditory and visual stimuli are taken to emanate from the same cause, the likelihood distribution where these signals stem from in the world has a relatively narrow variance (σ_C=1) and a mean (μ_C=1) equal to zero. In contrast, when these signals emanate from different causes, the likelihood from which they stem has a relatively large variance (σ_C=2), as they are independent from one another, and a negative mean (μ_C=2) in comparison to μ_C=1. These parameters are all related via Bayes’ rule (Eq. 2), and define the Bayes’ Optimal Synchrony Window (right).

Figure 2

Temporal binding window and Bayes’ optimal window of synchrony judgments for ASD and TD participants. Top panel: Averaged proportion of synchronous responses (y-axis) as a function of stimuli onset asynchrony (SOA, negative values indicate audio-leading SOAs and positive values indicate visual-leading conditions) and fit to the average (for visualization only). Temporal binding windows for audiovisual speech stimuli are larger for ASD (orange) than TD (blue) participants. Error bars represent +/− 1 SEM. Bottom panel: Top, C=1 (continuous) and C=2 (dashed) distributions as a function of SOA for a representative ASD (orange) and TD (blue) participant. Bottom, individual data for all ASD (orange) and TD (blue) participants. Each participant is represented by 2 dots indicating the location in time where their C=1 and C=2 curves intersect, one at position x < 0 and one at position x > 0 along the x-axis. The median for each group is represented via a box, the length of which illustrates the average Bayes’ optimal window of synchrony judgment for each group.

Figure 3

Comparison of Bayesian Causal Inference parameters inferred from fitting reports of audiovisual synchrony judgments in ASD and TD participants. A) Probability of common cause. Participants with ASD had on average a higher general probability of ascribing audiovisual events to a common cause. B): Sensory noise. There was no difference between ASD and TD groups with regard the sensory noise parameter. C) Mean of the source likelihood when two causes are ascribed. No difference in μ_C=2 between ASD and TD groups. Individuals dots are single participants (ASD=orange; TD=blue). Black horizontal lines represent the group medians.

Figure 4

Temporal binding window and Bayes’ optimal window of synchrony judgments for SZ and TD participants. Top panel: Averaged proportion of synchronous responses (y-axis) as a function of stimuli onset asynchrony (SOA, negative values indicate audio-leading SOAs and positive values indicate visual-leading conditions) and fit to the average (for visualization only). Temporal binding windows – indexed via Gaussian fitting – for audiovisual speech stimuli are larger for SZ (red) than TD (green) participants. Error bars represent +/− 1 SEM. Bottom panel: Top, C=1 (continuous) and C=2 (dashed) distributions as a function of SOA for a representative SZ (red) and TD (green) participant. Bottom panel: population data for SZ (red) and TD (green) participants. Each participant is represented by 2 dots, one at position x<0 and one at position x > 0 along the x-axis. The median for each group is represented via a box, the length of which illustrates the Bayes’ optimal window of synchrony judgment for each group.

Figure 5

Comparison of Bayesian Causal Inference parameters inferred from fitting reports of audiovisual synchrony in SZ and TD participants. A) Probability of common cause; There was a strong statistical trend suggesting that participants with SZ may have on average a higher general tendency to bind auditory and visual information together across time. B): Sensory noise: SZ participants exhibited a greater degree of sensory noise than their TD counterparts. C) Mean of the source likelihood when two causes are ascribed: No difference in μ_C=2 between SZ and TD groups. Individuals dots are single participants (SZ=red; TD=green). Black horizontal lines represent the group’s mean.

Figure 6

Direct comparison of Bayesian Causal Inference model parameter values across ASD and SZ groups. The distributions depicted (ASD = orange; SZ = red) are bootstrapped differences between the individuals composing the psychopathological group and their respective control groups. That is, the distributions are centered on the true difference between the clinical group and their control, and have a variance that corresponds to subject-to-subject contrasts. Top panel: distributions of P_(C=1) differences between patients and controls. Solid vertical line indicates the median of the distribution and ‘ns’ = non-significant. Middle panel: distributions of σ_{sensory noise} differences between patients and controls. In this case the distributions of σ_{sensory noise} differences were statistical different from one another (* indicates p < 0.05). Bottom panel: distributions of μ_(C=2) differences between patients and controls.