Distinct neurocognitive bases for social trait judgments of faces in autism spectrum disorder

The study investigates why people with autism spectrum disorder (ASD) experience difficulties in social cognition and affect despite substantial individual variability. It asks two main questions: where autistic traits lie within a comprehensive social-affective personality space, and which personality dimensions explain atypical social trait judgments of faces in ASD. Prior work often targeted one or two constructs (e.g., alexithymia or empathy), leading to incomplete or biased views. Using a dimensional (trans-diagnostic) approach, the authors aim to clarify the position of autistic traits relative to other social-affective traits (empathy, prosociality, antisociality, anxiety) and to test whether personality differences translate into differences in social trait judgments of faces, focusing on trustworthiness and warmth, traits central to social approach tendencies.

Previous research links ASD-related social difficulties to alexithymia (difficulty identifying/describing emotions) and to empathy deficits, but findings are mixed and often confounded by overlapping constructs and unrepresentative sampling. Behavioral work has emphasized emotion recognition from static facial expressions and atypical gaze patterns, showing impairments in ASD, but social trait judgments from faces (e.g., warmth, trustworthiness, competence) remain underexplored with small samples and mixed results. Some studies report abnormal trustworthiness judgments in ASD, while others find largely typical trait judgments. The authors argue for a comprehensive, dimensional assessment to situate autistic traits among broader social-affective personalities and to link these dimensions to face-based social trait judgments.

Participants: Four groups were recruited: (1) 412 general population online participants (after exclusions, controls N = 307), (2) 113 self-identified ASD online participants (after exclusions, ASD N = 89), (3) 8 neurosurgical epilepsy patients with depth electrodes (667 recorded neurons; 106 face-responsive), and (4) 16 high-functioning ASD participants tested in-lab (used for validation; primary analyses report groups 1 and 2). Demographics and exclusions followed preregistered criteria. IRB approval obtained. Self-reported personality: Participants completed 10 questionnaires yielding 33 subscales across four content domains: affective deficits (Social Anxiety, Apathy, Alexithymia, Moral Scrupulosity), antisocial traits (Dark Factors, Utilitarianism), Big Five (short form), and other-oriented/empathic traits (QCAE, Perceived Social Support, Prosocialness). Autistic traits were measured with AQ and SRS. Demographics (age, sex at birth, education, SES) were collected. Face judgment task: Stimuli were 500 celebrity face images (50 identities × 10 images) from CelebA with varied viewpoints, expressions, accessories, and contexts. Participants rated each image on 10 social traits (warm, critical, competent, practical, feminine, strong, youthful, charismatic, trustworthy, dominant) using a 7-point Likert scale, administered in 10 modules (50 images/module), rating traits in blocks. Exclusions: trial RT <100 ms or >5000 ms; block exclusion if >30% trials excluded or <3 distinct ratings; participant exclusion if >3 blocks excluded. Less than 5% of trials were excluded per participant. Recognition of celebrities was queried for control analyses. Single-neuron recordings: Depth electrodes in amygdala and hippocampus (anterior/posterior) recorded via microwires in 8 neurosurgical patients while viewing the same 500 images. Firing rates normalized to baseline; analyses focused on mean activity 250–1250 ms post-onset, with primary RSA using face-responsive neurons (N=106), and comparable results with all neurons. Network visualization of traits: Correlation-based network plots (R corrr::network_plot) illustrated relative positions of autistic traits among grouped personality domains for exploratory visualization. Factor analysis: Exploratory factor analysis on 33 subscales (ASD and controls combined) determined number of factors via Cattell-Nelson-Gorsuch test (R nFactors), estimated with factanal and oblimin rotation. Factor scores used in group comparisons (linear models controlling for age, sex, SES) and in subsequent analyses. A subset matched to typical ASD gender ratio was used to confirm factor structure. Representational similarity analysis (RSA) with neuronal activity: For each social trait, consensus image ratings were averaged within groups, then identity-by-identity dissimilarity matrices (DMs) were computed using Pearson correlation across the 10 images per identity (z-scored). Neural DMs were computed from averaged normalized firing rates across neurons for each image, then identity-by-identity dissimilarities. Correspondence between rating and neural DMs assessed with Spearman correlation. Group differences (ASD vs controls) in DM correspondence tested via 1000-label permutation. RSA linking personality dimensions to social trait judgments: For each pair of participants, Euclidean distances were computed for trait ratings (Trustworthy, Warm) and for each factor score (4 dimensions). Linear regression predicted trait DM structure from the four factor DMs, including group (ASD vs control) and group-by-factor interactions; covariates of no interest included as specified.

- Factor structure: A 4-factor model explained 42.1% of variance. Factors were labeled: Factor 1 Autistic traits and social avoidance (AQ, SRS, alexithymia, social anxiety, apathy), Factor 2 Empathy and prosociality (QCAE, prosocialness, negative loadings of apathy), Factor 3 Antisociality (Dark Triad/Core), Factor 4 Social agreeableness (other-regarding tendency, perceived social support, agreeableness). - Group differences in factors: ASD > controls on Factor 1 (B = 0.59 ± 0.12, t = 5.15, p < 0.001). No significant group differences on Factor 2 (B = 0.06 ± 0.12, p = 0.641), Factor 3 (B = -0.06 ± 0.12, p = 0.649), or Factor 4 (B ≈ -0.19 ± 0.13, p = 0.148). - Social trait ratings: ASD participants showed marginally or significantly higher ratings on some traits: warm (t(640) = 2.75, p = 0.006), practical (t(640) = 3.09, p = 0.002), youthful (t(640) = 3.43, p = 0.0006), with trends for trustworthiness (t(640) = 1.80, p = 0.07) and strong (t(640) = 1.71, p = 0.09). Recognition rates of celebrities were similar across groups (ASD 29.28% ± 19.02%; controls 27.40% ± 12.77%; t(394) = 1.09, p = 0.28). - Neural-rating correspondence (RSA): The correspondence between trustworthiness rating DM and neural DM (amygdala/hippocampus) was weaker for ASD than controls; permutation test showed a significant group difference (p = 0.041), with within-group correspondences significant (ASD p = 0.037, controls p = 0.026). For warmth, controls > ASD marginally (p = 0.084). - Personality-trait judgment associations (individual-level regressions): • Trustworthiness: Significant group × Factor 3 (Antisociality) interaction (B = 0.15 ± 0.08, t = 1.96, p = 0.05). In ASD, higher antisociality predicted lower trustworthiness ratings (B = -0.11 ± 0.06, p = 0.079), not observed in controls (B = 0.04 ± 0.04, p = 0.418). Group × Factor 2 (Empathy/prosociality) was marginal (B = -0.13 ± 0.07, p = 0.08): in controls, higher Factor 2 predicted higher trustworthiness (B = 0.13 ± 0.05, p = 0.004), absent in ASD (B ≈ 0.00 ± 0.06, p = 0.989). • Warmth: Main effect of Factor 3 negative across groups (B = -0.12 ± 0.06, t = -2.16, p = 0.032). Group × Factor 4 (Social agreeableness) marginal (B = 0.12 ± 0.07, p = 0.068): controls showed positive association (B = 0.11 ± 0.04, p = 0.002), not present in ASD (B = -0.01 ± 0.06, p = 0.850). - Similarity-based RSA between personality and trait-judgment spaces: For trustworthiness, Factor 1 showed a significant positive main effect across groups (B = 0.05 ± 0.01, t = 4.11, p < 0.001), and Factors 2–4 exhibited significant group × factor interactions (ps < 0.001), indicating stronger positive associations for controls with empathy/prosociality and social agreeableness, and a distinct pattern for antisociality in ASD. For warmth, Factors 2 and 3 were positively associated overall, with a stronger Factor 2 association in controls than ASD.

Findings show that autistic traits cluster with social avoidance, social anxiety, alexithymia, and apathy, but are orthogonal to empathy/prosociality and antisociality. Despite comparable mean levels of empathy/prosociality and social agreeableness between ASD and controls, the translation of these self-reported dispositions into face-based social trait judgments differs. Controls with higher empathy/prosociality and social agreeableness tend to perceive faces as more trustworthy and warm, respectively, whereas these associations are attenuated or absent in ASD. Conversely, antisociality relates more strongly to reduced trustworthiness judgments in ASD. Neural RSA reveals weaker alignment between behavioral trait spaces (trustworthiness, warmth) and amygdala/hippocampus representations in ASD, suggesting distinct neural encoding contributing to altered person perception. Together, the results address the research questions by situating autistic traits within a dimensional social-affective space and identifying specific personality-to-perception links that differ in ASD, offering a neurocognitive account of social approach-related difficulties.

The study integrates a dimensional personality framework, face-based social trait judgments, and single-neuron recordings to delineate neurocognitive mechanisms of social perception in ASD. Autistic traits align with social avoidance/anxiety and alexithymia but are independent of empathy, prosociality, antisociality, and moral preferences. Although average warmth and trustworthiness judgments are not reduced in ASD, their neural encoding and their associations with prosocial and agreeableness dimensions differ from controls. These atypical links may underlie social approach challenges in ASD. Future research should replicate in larger, diagnostically verified ASD samples, disentangle contributions of perceptual cues versus semantic knowledge about familiar faces, and combine self-report with naturalistic behavioral assessments and first-person accounts to clarify how empathic dispositions translate into social perception and interaction.

- ASD sample size (N = 89) is modest, though larger than many prior studies; replication with larger, multi-center cohorts is needed. - Online ASD participants were self-identified; diagnoses could not be independently verified, though comparisons with a small in-lab diagnosed ASD sample were consistent. - Use of celebrity stimuli introduces potential confounds from prior knowledge; recognition rates were similar across groups, but semantic knowledge was not explicitly measured, limiting precise estimation of perceptual versus semantic contributions. - The study focuses on trustworthiness and warmth; generalization to other social traits requires further investigation.