Neural tracking of social navigation in autism spectrum disorder

Participants

Participants included 46 typically developing (TD) adults, 42 adults with ASD, and 34 adults with misophonia (see Table 1 for demographic characteristics of each group). ASD diagnoses were confirmed by licensed, research reliable clinicians using the Autism Diagnostic Observation Schedule (ADOS-2; Module 4; 30), developmental and clinical history, self and informant (e.g., parent, roommate) report of symptoms and adaptive behavior, cognitive functioning, and clinical judgment regarding whether the individual meets DSM-5 criteria for ASD. TD participants had no past or current psychiatric or neurodevelopmental diagnosis and no first-degree relative with autism; misophonia participants reported additional common psychiatric conditions (e.g., anxiety, depression) but no past, current, or suspected ASD. All participants had an IQ >70. To ensure active engagement with the task, responses on at least 75% of trials and above chance post-task memory scores were requisite for participant inclusion. For a full list of inclusion criteria and recruitment and attrition numbers, please see Supplemental Materials.

All participants provided written informed consent and were financially compensated for their participation. The study was approved by the Institutional Review Board at the Icahn School of Medicine at Mount Sinai. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.

Measures

Participants were administered surveys to self-report symptoms, including the Broad Autism Spectrum Questionnaire (BAPQ; 31), the Liebowitz Social Anxiety Scale (LSAS, Avoidance questions; 32), and the Amsterdam Misophonia Scale (A-MISO-S; 33). The ASD participants additionally completed a modified version of Bengston’s Positive Affect Index (34) measuring real-life social network size (see Supplemental Materials for more information), and were evaluated on the ADOS-2 (30).

Previous studies have reported discrepancies between self- and other-reports of ASD symptoms (35–38), indicating that self- and other-reported symptom measures may capture different aspects of autism. Self-reported symptoms may capture internal difficulties that may be externally invisible, while other-reported symptoms may capture external difficulties that can exist outside of the individual’s self-awareness. Discordances between symptom measures have been shown to predict outcomes better than independent reports alone (36), and are interpreted as reflecting meaningful variation in insight (35,36). For example, some autistic people may lack awareness into the ways that their actions differ from social expectations, while others may experience social anxiety that does is meaningfully reflected in their behavior. These dissociations may thus reflect individual differences in social and self-insight – domains that can strongly influence how someone perceives others’ social positioning relative to themselves. To investigate this, we calculated the discrepancy between self-reported symptoms and clinician-reported symptoms by z-scoring the BAPQ total score and subtracting it from the z-scored total algorithm raw scores from the ADOS.

Social Navigation Task

We used a validated social navigation task (1,2,39) that simulates dynamic social interaction. All participants completed the task inside of a 7 Tesla (7T) MRI scanner. Participants interacted with six characters across 63 decision trials embedded in a “choose-your-own-adventure” narrative format. Though the same sequence of slides was always shown, each choice appeared to prompt responsive feedback, giving the impression of an adaptive storyline with evolving relationships. Individualized decisions shaped distinct overall narratives, mimicking how individuals dynamically shape real-world interactions. Individuals’ behavior on this task has been shown to directly reflect real-life interpersonal outcomes in non-clinical populations (39). In the task, each decision implicitly alters either the affiliation or power dynamic between the participant and a character, allowing us to model that character’s location in a 2D social space framed by affiliation and power axes. Though participants were not shown this spatial structure, their choices unknowingly shifted character positions within this space. On each trial, we calculated the 3D (x: affiliation, y: power, z: trial number) Euclidean distance and angle of a vector drawn between the character’s location and the participant’s fixed point of view (maximum affiliation, neutral power), generating trial-by-trial indices of social distance and angle to be used as parametric modulators in the fMRI analysis. For full task details, see Supplementary Materials.

fMRI Analysis - General linear models

fMRI data were acquired at 7T, preprocessed using fMRIPrep, and modeled using general linear models in SPM12 (see Supplemental Materials for full details). Decision trials were modeled with a z-scored parametric modulator reflecting trial-by-trial social distance, defined as the vector length from a fixed point-of-view to the character’s location in social space. Two primary contrasts were defined: 1) decision trials vs. baseline and 2) the parametric distance modulator vs. baseline. At the second level, we tested whole-brain group-difference contrasts and whole-brain cross-diagnostic effects, with age, sex, and mean framewise displacement included as covariates. All group analyses were grey-matter-masked. Beta weights were extracted from the task-identified regions that survived statistical correction and were used for further analyses (i.e., associations with symptom scores), controlling for age, sex, race, and mean framewise-displacement.

Independent Atlas ROI Verification

To avoid circularity in group comparisons and validate our neural activation results, as well as better understand the functional correspondence of our detected clusters, we extracted beta weight values from independent ROIs. See Supplemental Materials for more information on atlas ROI definitions and visualizations of the overlap with task-identified ROIs.

fMRI Analysis – Functional Connectivity

We used the CONN toolbox to examine task-related functional connectivity, using clusters that showed group differences in activation during social decisions as seed regions for whole-brain seed-to-voxel analyses. See Supplement for full preprocessing and analysis details.

Group characteristics

The groups did not differ on IQ or MRI head motion (Table 1). To characterize each group’s symptom profiles and confirm increased self-reported autism symptoms in ASD, we compared their scores on the BAPQ (31). We identified a significant 3-group difference (F(2,119)=50.86, p<0.001), driven by higher BAPQ scores (indicating more symptoms) in the ASD group compared to both the TD group (mean difference = −1.31, 95% CI [−1.62, −1.00], p<0.001) and the misophonia group (mean difference = −0.76, 95% CI [−1.09, −0.43], p<0.001); the misophonia group also had higher BAPQ scores than the TD group (mean difference = −0.55, 95% CI [−0.88, −0.23], p<0.001), suggesting they may have higher autistic traits than individuals with no psychiatric concerns. Consistent with prior work (35–37), we did not detect any relationship between self-reported BAPQ scores and clinician-rated (ADOS-2; 30) scores in the ASD group (r(40)=0.005, p=0.98).

Diagnostic differences in social navigation behavior

In the social navigation task (Figure 1A; 1), the groups significantly differed in the average length (F(2,115)=15.74, p<0.001, Figure 1C), but not the angle (F(2,115)=1.36, p=0.26) of character vectors. Individuals with ASD showed larger social distances than both TD (mean difference = −0.62, 95% CI [−0.90, −0.35], p<0.001) and misophonia (mean difference = −0.51, 95% CI [−0.80, −0.21], p<0.001) groups; no difference was detected between the TD and misophonia groups (mean difference = −0.11, 95% CI [−0.41, 0.18], p=0.62).

To examine whether the increased social distances seen in ASD were driven by behavior in a specific domain, we performed exploratory group comparisons of affiliation and power tendencies. We detected a significant difference in affiliation (F(2,115)=13.06, p<0.001), such that the ASD group was less affiliative towards characters than the TD (mean difference = 0.28, 95% CI [0.15, 0.41], p<0.001) and misophonia groups (mean difference = 0.19, 95% CI [0.05, 0.33], p=0.005). No difference in affiliation was detected between the TD and misophonia groups (mean difference = 0.09, 95% CI [−0.05, 0.23], p=0.31), and no group differences were detected for power (F(2,115)=1.20, p=0.30).

We detected a significant group by self-reported ASD symptom interaction on social distance (F(2,112)=4.18, p=0.018, Figure 1D); individuals with ASD showed a positive correlation between distance and self-reported ASD symptoms (r(40)=0.467, p=0.002), while no relationship was detected in the TD (r(44)= −0.016, p=0.92) or misophonia groups (r(32)= 0.04, p=0.82). We also detected a significant group by social avoidance symptom interaction on social distance (F(2,112)=3.90, p=0.023, Figure 1E); individuals with ASD showed a positive correlation between distance and self-reported social avoidance (r(40)=0.424, p=0.005), while no relationship was detected in the TD (r(44)= −0.17, p=0.25) or misophonia groups (r(32)= 0.20, p=0.23).

Diagnostic differences in neural activity during social decisions

Given that the misophonia and TD groups showed comparable task behavior (Figure 1), we chose to combine them into one group, allowing us to investigate ASD-specific alterations in the neural processing of social decisions. For the MIS/TD > ASD contrast, whole-brain analysis revealed a cluster in the left temporal pole (L TP, peak-level p_FWE=0.045, K_E = 209, peak MNI coordinates x: −52 y: 12 z: −18, Figure 2A). To validate this finding, we extracted beta values from an independent, atlas-defined ROI, which corroborated that individuals with ASD show reduced L TP activation during social decisions compared to MIS/TD (F(1,115) = 9.26, p=0.003, Figure 2B). In addition, the whole-brain analysis for the MIS/TD > ASD contrast revealed a cluster in the left insula (peak-level p_FWE=0.038, K_E = 289, peak MNI coordinates x: −42 y: −10 z: 0). Beta values extracted from an independent atlas-defined ROI validated that the ASD group showed higher activation in the L insula during social decisions compared to the MIS/TD group (F(1,115) = 6.42, p=0.013).

We next tested for associations between task-identified activations and symptom and quality of life measures within ASD. We identified an association between reduced L TP activity and smaller real-life social networks (b=1.60, SE=0.51, t=3.13, p=0.003, Figure 2C). Reduced L TP activation was associated with higher ADOS compared to BAPQ scores (b=−1.44, SE=0.52, t=−2.74, p=0.009, Figure 2D). We further validated these relationships using betas extracted from the independent atlas L TP ROI (real-life social networks: b=2.32, SE=1.03, t=2.26, p=0.030; symptom discrepancy: b=−2.64, SE=1.01. t=−2.60, p=0.013). No relationship was detected between L TP activity and ADOS scores (b=−2.84, SE=1.77, t=−1.61, p=0.12). No relationships were detected for L insula activity with social network size (b=−0.48, SE=0.52, t=−0.92, p=0.36), symptom discrepancy (b=−0.051, SE=0.53, t=−0.10, p=0.92), or ADOS scores (b=−0.47, SE=1.67, t=−0.28, p=0.78). Social distance was not associated with L TP activity (b=0.046, SE=0.089, t=0.53, p=0.60) or the L insula (b=0.094, SE=0.097, t=0.97, p=0.34).

Diagnostic differences in functional connectivity during social decisions

We examined diagnostic differences in functional connectivity during social decisions, again using a combined MIS/TD comparison group given comparable task behavior. Using the L TP seed, we detected a cluster in the PCC that showed differential connectivity patterns between groups (cluster-level p_FWE <0.001, peak-level p-uncorrected <0.001, K_E = 333, peak MNI coordinates x: −06 y: −36 z: 26, Figure 3A). Compared to TD and misophonia, who showed negative connectivity between the L TP and PCC on average, autistic individuals showed connectivity averaging around zero (i.e., decreased decoupling; Figure 3B). No clusters showed group differences in functional connectivity with the L insula.

Within ASD, decreased decoupling between the L TP – PCC (i.e., less similar to MIS/TD) was associated with higher ADOS compared to BAPQ scores (b=2.86, SE=1.28, t=2.24, p=0.031, Figure 3C). We did not detect a relationship between L TP – PCC connectivity and social network size (b=0.96, SE=1.37, t=0.70 p=0.49) or ADOS scores independently (b=4.18, SE=4.34, t=0.96, p=0.34).

These results provide evidence in support of our first hypothesis that autistic adults differ in neural processing of information during social decision making. We next turned to testing the other hypothesis, that autistic adults would show reduced tracking of social distance – indicative of impairments in social map construction.

Cross-diagnostic neural systems for the neural tracking of social distance

We used a parametric modulation approach to identify brain regions that dynamically track trial-by-trial social locations. Given the identified group difference in the length but not the angle of the character vectors, we focused our neural investigation on social distance (see Supplemental Materials for angle analysis). Our whole brain analysis did not reveal any group differences (ASD vs MIS/TD) in neural social distance tracking, suggesting insufficient evidence to support impaired social map creation in ASD. Thus, we proceeded to testing cross-diagnostic effects.

Across all groups, participants showed significant neural tracking of social distance in 11 regions: the precuneus, posterior cingulate cortex (PCC), anterior cingulate cortex (ACC), left middle temporal gyrus (MTG), left middle frontal gyrus (MFG), and bilaterally in the temporoparietal junction (TPJ), dorsolateral prefrontal cortex (dlPFC), and a part of the posterior parahippocampal cortex that shows overlap with previously reported locations of the parahippocampal place area (40–43) (PPA, Table 2). Of these, only 6 regions showed a stronger effect for distance than for decisions: the right and left PPA, precuneus/right PCC, left PCC, ACC, and left MFG (Figure 4A). Our findings thus replicate prior work reporting vector length tracking in the PCC in neurotypical participants (1; see Supplemental Materials for a visualization of the ROI overlap).

We next tested for associations with cross-diagnostic symptoms, correcting for multiple comparisons. No regions showed associations with self-reported ASD symptoms. Of all six regions, only the right PPA showed a significant association with social avoidance, such that individuals with weaker tracking reported higher social avoidance (b=−0.001, SE=0.0005, t=−2.82, corrected p=0.034, un-corrected p=0.006, Figure 4B). To validate the claim for correspondence of our cluster with the place area, we calculated an independent ROI comprised of the overlap between the right posterior parahippocampal gyrus and a neurosynth (44) meta-analysis map of regions showing functional specificity for ‘place’ representation. Consistent with our interpretation, we found again strong evidence of distance tracking (t(121)=3.85, p<0.001) and correlation with avoidance (b=−0.001, SE=0.0005, t=−2.36, p=0.020).