1. Overview of contributions to the special issue

The first topic addressed in this special issue is how we measure and define heterogeneity in autistic traits, and whether ways of measuring such individual differences are valid, reliable, and potentially subject to sampling bias. Wu et al., (Reference Wu, Hong, Kim, Adolphs, Paul and Charpentier2025) tackle three important questions relevant to characterizing heterogeneity in autistic traits using self-reported scores on the Social Responsiveness Scale (SRS), a widely used scale for assessing the two core domains of autistic functioning (Constantino et al., Reference Constantino, Davis, Todd, Schindler, Gross, Brophy, Metzger, Shoushtari, Splinter and Reich2003; Constantino & Gruber, Reference Constantino and Gruber2012). First, Wu et al., (Reference Wu, Hong, Kim, Adolphs, Paul and Charpentier2025) address the timely question of whether self-reported autistic traits differ between online and in-lab samples. This is of interest to the autism research community, but also potentially relevant to any researcher using self-reported questionnaire scores to measure various mental health symptoms and/or personality traits. Specifically, it helps characterize whether similar score distributions can be expected between traditional in-lab samples, and more recent but widespread online studies through platforms like Prolific Academic (https://www.prolific.com/). The findings reveal that scores on an autistic trait measure, the SRS, normed for individuals with and without an autism diagnosis, were substantially higher in online compared to in-lab samples, and that this effect was robust to controlling for other covariates, such as demographic characteristics and self-reported symptoms relevant to mental health (e.g. anxiety). The authors raise the intriguing possibility that “differences might come from a self-selection bias: people who are willing to volunteer for in-person research studies can be more outgoing, comfortable with socialization settings, and exhibit less autistic traits; whereas people who register at online crowdsourcing platforms may have difficulty finding on-site jobs, prefer less in-person social interactions, and show higher autistic traits (Bethlehem, Reference Bethlehem2010; Difallah et al., Reference Difallah, Filatova and Ipeirotis2018).”

Second, in an attempt to identify possible factors that could explain the discrepancy in autistic traits between in-lab and online samples, Wu et al., (Reference Wu, Hong, Kim, Adolphs, Paul and Charpentier2025) performed additional analyses that provided more general insights into the drivers of the variability in autistic traits. Specifically, they found that a regression model with only 7 predictors explained over 55% of the variability in SRS scores, highlighting that some of the individual differences observed in autistic traits – the aim of this special issue – share substantial variance with other individual difference measures. The strongest contributor to variability in autistic traits was social anxiety, followed by gender, depression and trait anxiety. Each predictor explained unique variance in autistic trait scores. These findings provide a detailed characterization of individual differences in autistic traits, particularly their association with internalizing symptoms, and gender differences, and suggest that these factors should be considered and accounted for in future work investigating the heterogeneity of autism. Finally, the authors contribute a comprehensive assessment of the stability of SRS scores via test–retest reliability analyses over longer intervals than typically studied (greater than 6 months). The study is also the first to provide estimates of test–retest reliability in online samples, indicating moderate-to-good reliability, and values comparable to in-lab samples. These findings help validate the use of this scale in online samples, and the stability of individual autistic traits over long periods, a key requirement for using such measures to characterize individual differences in autism.

This first article highlighting important considerations for measuring autistic traits is followed by a perspective that offers an important theoretical and computational organizing framework for understanding the heterogeneity of autism along mechanistic dimensions. Cahalan et al., (Reference Cahalan, Mitroff, Subiaul and Rosenblau2025) argue that autism, traditionally conceptualized as a unidimensional condition, is better understood as a multidimensional construct. Specifically, the article proposes refining established autism symptom dimensions through experimental and computational frameworks that provide more nuanced, mechanistic insights into cognitive and behavioral processes. Such an approach may better capture the heterogeneity of the autism spectrum than a single autistic-traits dimension, thereby allowing for a more precise characterization of both autistic traits in the general population and autism symptoms in diagnosed individuals. The paper focuses on one particularly important transdiagnostic dimension: the cognitive rigidity–flexibility dimension. Cognitive rigidity is strongly implicated in autism and plays a central role in the manifestation of restricted and repetitive behaviors, one of the two core autism symptom domains. More recent work, however, has also linked cognitive rigidity to social cognition (Kenworthy et al., Reference Kenworthy, Anthony, Naiman, Cannon, Wills, Luong‐Tran, Werner, Alexander, Strang, Bal, Sokoloff and Wallace2014), which underlies the second core autism symptom domain (American Psychiatric Association, 2022). Building on this perspective, the paper discusses how diminished cognitive flexibility may manifest across a broad range of cognitive functions associated with autistic traits and autism symptoms, including perception, attention, learning, and communication. It proposes investigating rigidity and flexibility within these domains using comprehensive task batteries combined with computational modeling approaches. This framework conceptualizes cognitive flexibility not as a single dimension, but as a set of related yet distinct dimensions.

Cahalan et al., (Reference Cahalan, Mitroff, Subiaul and Rosenblau2025) suggest that large, heterogeneous samples spanning variation in autistic traits and diagnostic status could complete these task batteries to capture individual variability and enable the identification of subgroups. Such an approach could help reconceptualize autism as a set of “autisms” with narrower, better-defined symptom profiles that may be more effectively targeted through tailored interventions. In particular, this framework could clarify whether cognitive rigidity in autism is domain-general or domain-specific, while also identifying subgroups with distinct profiles of cognitive flexibility across domains. More broadly, the authors propose that this framework could be extended to other higher-order dimensions relevant to autism and transdiagnostic symptom severity. By more precisely characterizing dimensions across cognitive domains and task settings, and by supplementing behavioral observations with computational models that infer latent mechanisms underlying observable behavior, the field may move closer to a more accurate understanding of the multifaceted structure of autism. Ultimately, this could improve diagnostic precision and support the development of more targeted interventions.

The next two manuscripts in this special issue are empirical studies that dive into specific neurocognitive mechanisms underlying differences in autistic traits in the general population. Beyond the broad rigidity-flexibility dimension proposed by Cahalan et al., (Reference Cahalan, Mitroff, Subiaul and Rosenblau2025), Haffey et al., (Reference Haffey, Hsu and Chakrabarti2025) focus on the more specific mechanistic domain of social reward and motivation. This study directly compares how adults with varying levels of autistic traits process social versus nonsocial stimuli representing emotionally rewarding real-world scenes matched on valence, arousal, and key low-level stimulus properties. The authors introduce a novel functional magnetic resonance imaging (fMRI) paradigm designed to compare social and nonsocial reward processing while also investigating whether functional connectivity between key brain regions involved in reward and social processing modulates the relationship between self-reported preference for social versus nonsocial rewards and autistic traits. The findings showed that individual differences in self-reported preference for social images were associated with functional connectivity between the left anterior insula (LAI) and the medial orbitofrontal cortex (mOFC), as well as between the left fusiform gyrus (LFG) and the anterior cingulate cortex (ACC). Importantly, autistic traits negatively moderated connectivity within both the LAI–mOFC and LFG–ACC networks. Reduced functional connectivity between these regions may therefore contribute to lower social reward responsiveness in individuals with higher autistic traits, which was also reflected in their lower valence ratings for social rewards. This study provides a novel experimental paradigm capable of assessing social reward processing at both behavioral and neural levels, which may in turn contribute to the identification of potential transdiagnostic biomarkers relevant to social cognitive processes. More broadly, the results point to a possible neural mechanism underlying the observation that individuals with higher autistic traits often experience social stimuli as less rewarding in laboratory-based tasks compared to individuals with lower autistic traits.

Next, Ip et al., (Reference Ip, Chiu, Ching, Law, Tang, Ng and Wong2026) also measured autistic traits dimensionally and linked them to functional connectivity from resting-state fMRI, but shifting from social reward to moral/self-conscious emotions. Adopting this dimensional approach, the authors assumed that mechanisms typically observed in autism may also be present, albeit to a lesser degree, in neurotypical individuals, and can therefore be systematically measured in non-clinical samples. Within this framework, the study focuses on self-conscious moral emotions – specifically shame and guilt – and explicitly seeks to uncover the neurocognitive mechanisms that may account for their differentiated experience. The authors proposed that individual differences in the experience of these emotions were partly shaped by social–cognitive abilities, particularly the capacity to understand the mental states of others. From a socio-cognitive perspective, difficulties in accurately predicting others’ thoughts and feelings may promote increased self-focused processing, a hallmark of shame, whereas guilt may be more strongly associated with concern for others and the interpersonal consequences of one’s actions. Empirically, this is the first study to investigate whether the profile of heightened shame-proneness combined with reduced guilt-proneness in individuals with higher autistic traits can be explained by the functional organization of the brain.

Replicating previous findings, Ip et al., (Reference Ip, Chiu, Ching, Law, Tang, Ng and Wong2026) showed that higher levels of autistic traits were associated with greater shame-proneness, while higher traits measured with the Broad Autism Phenotype Questionnaire were additionally linked to lower guilt-proneness. Using a hypothesis-free, data-driven analysis of resting-state functional connectivity, the study identified differential connectivity of the right frontal pole associated with autistic traits as well as with shame- and guilt-proneness. Follow-up seed-based analyses further revealed that several regions within the cortical midline structures (including the ACC, posterior cingulate cortex/precuneus, superior frontal gyrus) mediated the positive and negative associations between autistic traits and shame and guilt, respectively. Together, these findings provide initial evidence that functional connectivity within brain regions crucial for social cognition and self-referential processing constitutes a neurobiological mechanism underlying individual differences in the experience of negative self-conscious emotions across the spectrum of autistic traits.

While all articles thus far broadly conceptualized heterogeneity via autistic traits measured dimensionally in the general population, the last manuscript focused on an adult cohort with an autism diagnosis. Specifically, McQuaid et al., (Reference McQuaid, Lee and Wallace2025) leveraged the known association between depression and autism (i.e., elevated rates of depression in autistic adults compared to the general adult population), as well as challenges that can occur in emotion regulation both in depression and in autism, to assess whether emotion regulation challenges may be a risk factor for elevated depression in autism. While past research has used network analysis to better characterize the structure of the associations between emotion regulation and depressive symptomatology, it is likely that this structure could differ in an autistic population and/or that autistic individuals could exhibit unique associations relevant to the heterogeneity of autism.

McQuaid et al., (Reference McQuaid, Lee and Wallace2025) measured emotion regulation and depressive symptoms through self-report questionnaires in a large cohort of autistic adults (N = 775) and applied network analysis to identify the key components of emotion regulation, depressive symptoms, as well as their interaction. While the study did not include a non-autistic comparison sample, and therefore future work will be needed to statistically assess similarities and differences, the findings reveal that the structure of the three networks (emotion regulation only, depressive symptoms only, and interaction between the two) was highly consistent with previous findings of how these dimensions are structured and relate to one another in the general population. Of particular importance was the combined emotion regulation-depressive symptoms network, which clearly showed two distinct constructs but with some interrelatedness. Specifically, connections between the two constructs emerged between one’s ability to manage their emotions and troubles with movement (slowness/restlessness) and concentration as depression symptoms, as well as between cognitive reappraisal and feeling down. These associations, identified in autistic individuals, appear consistent with the broader literature linking emotion regulation and depression components in the general population, and were also not modulated by autistic traits. This study paves the way towards establishing emotion regulation challenges as a potential risk factor to depression in autism, as well as towards characterizing heterogeneity in emotional processes and internalizing symptoms in autism, though whether this heterogeneity is specific to autism or applies to the general population remains an open question. The authors conclude by setting several key future directions from this work, including examining “the potential associations with broader challenges in executive functioning that autistic people experience, and other cognitive constructs, such as alexithymia, which is likely to play a particularly important role in emotion regulation challenges among autistic people.”

Finally, McQuaid et al., (Reference McQuaid, Lee and Wallace2025) point towards the heterogeneity in the role of external factors, which are likely to further modulate the associations, and possibly differ from non-autistic individuals. Such external factors include social rejection, discrimination, negative social experiences and the burden of masking (Beck et al., Reference Beck, MacKenzie, Kumar, Breitenfeldt, Chang, Conner, Mandell, White and Mazefsky2024; Keenan et al., Reference Keenan, Gotham and Lerner2018). Overall, this study adds to the special issue by highlighting a clinically salient point: heterogeneity is not only about traits, the brain or cognition, it is also critical to mental health risks and lived experiences.

2. Thematic thread analysis

Five broad themes emerge from the manuscripts in this special issue, as well as touched upon in the introduction paper (Liu et al., Reference Liu, Lai, Le, Lan, Zhang, Huang, Xu, Jiang, Li, Kendrick and Zhao2025).

First, the manuscripts highlight a clear dimensional framing of autistic traits and heterogeneity. Treating autistic traits as continuously distributed, rather than purely categorical, can enable answering mechanistic questions in general-population samples (as seen in Haffey et al., Reference Haffey, Hsu and Chakrabarti2025; Ip et al., Reference Ip, Chiu, Ching, Law, Tang, Ng and Wong2026; Wu et al., Reference Wu, Hong, Kim, Adolphs, Paul and Charpentier2025). Ultimately, this could help with collecting more large-sample datasets, appropriately powered to link neurocognitive profiles to broader reconceptualization of autism as multidimensional (as proposed by Cahalan et al., Reference Cahalan, Mitroff, Subiaul and Rosenblau2025). Large samples of individuals diagnosed with autism (as seen in McQuaid et al., Reference McQuaid, Lee and Wallace2025) can also inform whether and how mental and physical health dimensions are related in autistic populations and if these relationships differ from those in the broader population.

A second theme is that transdiagnostic affective and internalizing processes are key sources of variability related to the heterogeneity of autism and autistic traits. Ip et al., (Reference Ip, Chiu, Ching, Law, Tang, Ng and Wong2026) provide insights into how shame and guilt profiles carry relevant variance to autistic traits, while Wu et al., (Reference Wu, Hong, Kim, Adolphs, Paul and Charpentier2025) show robust links between internalizing symptom dimensions (social anxiety, trait anxiety, depression) and autistic traits. Finally, McQuaid et al., (Reference McQuaid, Lee and Wallace2025) highlight the structure of depression-emotion regulation processes in autism. Mood and affective disorders carry a higher prevalence in autistic individuals than in the general population, underscoring the importance of studying their underlying processes and potential unique manifestations in autism.

A third theme stems from two manuscripts diving into the brain and suggesting mechanistic neurocognitive pathways linking traits to experience. Individual differences in autistic traits were found to relate to functional connectivity patterns that plausibly mediate differences in socio-affective experience: self-conscious moral emotions via cortical midline structures (Ip et al., Reference Ip, Chiu, Ching, Law, Tang, Ng and Wong2026) and social reward responsivity via insula/OFC and fusiform/ACC networks (Haffey et al., Reference Haffey, Hsu and Chakrabarti2025).

Already suggested in our introduction manuscript (Liu et al., Reference Liu, Lai, Le, Lan, Zhang, Huang, Xu, Jiang, Li, Kendrick and Zhao2025), this special issue generally advocates moving from unitary autism to profiles/subgroups via computational and multilevel approaches. Cahalan et al., (Reference Cahalan, Mitroff, Subiaul and Rosenblau2025) provides explicit insights for mapping relevant trait dimensions onto component processes (e.g., multiple facets of flexibility) using task batteries and computational modeling. This approach complements the other papers’ emphasis on decomposing heterogeneity into interpretable constructs, namely depression-emotion regulation networks in McQuaid et al., (Reference McQuaid, Lee and Wallace2025), mediators and moderators of brain’s functional connectivity in Haffey et al., (Reference Haffey, Hsu and Chakrabarti2025) and Ip et al., (Reference Ip, Chiu, Ching, Law, Tang, Ng and Wong2026), and predictors of trait variance in Wu et al., (Reference Wu, Hong, Kim, Adolphs, Paul and Charpentier2025).

Finally, the manuscripts support bridging social and non-social domains. Social processes are central to autism, with manuscripts focusing on directly socially-relevant constructs, such as social reward (Haffey et al., Reference Haffey, Hsu and Chakrabarti2025), mentalizing-related self-conscious emotions (Ip et al., Reference Ip, Chiu, Ching, Law, Tang, Ng and Wong2026), or social anxiety (Wu et al., Reference Wu, Hong, Kim, Adolphs, Paul and Charpentier2025). However, the issue also emphasizes that heterogeneity extends into broader cognitive and affective constructs – including rigidity–flexibility across domains (Cahalan et al., Reference Cahalan, Mitroff, Subiaul and Rosenblau2025) and depression/emotion regulation (McQuaid et al., Reference McQuaid, Lee and Wallace2025). Taken together, these insights support a more integrative model of autism-relevant individual differences.

3. New directions: naturalistic neuroscience

Research on autism has advanced considerably in recent years, spanning neural, cognitive, behavioral, and social levels of analysis. Yet several important directions remain underexplored. Here, we highlight three themes that we believe hold particular promise for shaping the next generation of autism research.

First, much of what we know about brain and behavioral differences in autism stems from paradigms in which autistic individuals are studied in isolation – observing stimuli, completing tasks, or being assessed outside of genuine social contexts. A promising shift is underway, however, toward studying autism through the lens of second-person neuroscience: an approach that emphasizes real-time, reciprocal social interaction as the proper unit of analysis (Redcay & Schilbach, Reference Redcay and Schilbach2019; Schilbach et al., Reference Schilbach, Timmermans, Reddy, Costall, Bente, Schlicht and Vogeley2013). From this perspective, social cognition is not a property of an individual’s brain but an emergent process shaped by dynamic coordination between partners. Some studies have begun to move in this direction, employing hyperscanning paradigms and behavioral coding during live interactions (e.g., Kruppa et al., Reference Kruppa, Reindl, Gerloff, Oberwelland Weiss, Prinz, Herpertz-Dahlmann, Konrad and Schulte-Rüther2021; Tang et al., Reference Tang, Wang, Liu, Li, Dai, Cui and Li2023; Wang et al., Reference Wang, Han, Hu, Feng, Wang, Liu and Yi2020). However, these efforts remain partial. Behavioral and biological indices of synchrony (the temporal alignment of behavior, physiology and neural activity between interactive partners) have largely been examined in isolation, and truly multimodal paradigms – capable of capturing simultaneous alignment across behavioral, autonomic, and neural levels simultaneously within a single interaction – are still rare in research on autism (Zaborowska et al., Reference Zaborowska, Pisula and Pluta2026). Such designs would allow researchers to examine whether coordination at one level corresponds to coupling at another, and whether discrepancies between levels carry specific clinical significance. These questions that are particularly pressing in autism, where the social difficulties that define the condition are inherently interactional and can only be fully understood in the context of real interpersonal exchange.

Second, and perhaps most importantly, the choice of interaction partner carries theoretical implications. The Double Empathy Problem (Milton, Reference Milton2012) proposes that social difficulties in autism arise not from a unidirectional deficit within the autistic individual, but from a bidirectional mismatch between people who differ in their predictive styles, communicative habits, and social expectations. Consistent with this view, emerging evidence suggests that autistic individuals interacting with other autistic individuals – rather than with non-autistic partners – show greater behavioral alignment and higher friendship quality (Bolis et al., Reference Bolis, Lahnakoski, Seidel, Tamm and Schilbach2021; Crompton et al., Reference Crompton, Ropar, Evans-Williams, Flynn and Fletcher-Watson2020), and that similarity in autistic traits may promote neural synchronization (Feng et al., Reference Feng, Ding, Wang, Zhang, Yuan, Zhang and Bai2024). However, this hypothesis has not yet been tested using multimodal paradigms that would allow for a comprehensive assessment of whether phenotypic similarity between partners promotes coherence across behavioral, physiological, and neural levels simultaneously. Interestingly, recent evidence suggests that this account should be narrowed to experiences during interaction as it does not generalize to higher accuracy when autistic individuals make judgments about autistic compared to non-autistic others (Cahalan et al., Reference Cahalan, Perla, Block, Loughlin, Korn and Rosenblau2026), likely due to higher heterogeneity in the autistic population.

Finally, a third important direction for future research concerns the neurobiological foundations and mechanisms underlying heterogeneity within autism. Although the clinical symptom profile has been the primary basis for characterizing autism, a growing body of research has begun to map potential neurobiological subtypes defined by patterns of brain functional organization rather than by behavioral presentation alone (e.g. Buch et al., Reference Buch, Vértes, Seidlitz, Kim, Grosenick and Liston2003). Individuals can differ substantially in functional connectivity profiles despite exhibiting comparable scores on standardized diagnostic measures such as the Autism Diagnostic Observation Schedule, the Autism Diagnostic Interview-Revised, and the SRS, underscoring the limitations of relying exclusively on behavioral indices to capture the full heterogeneity of the condition (Liu et al., Reference Liu, Lai, Le, Lan, Zhang, Huang, Xu, Jiang, Li, Kendrick and Zhao2025; Lombardo et al., Reference Lombardo, Lai and Baron-Cohen2019). This is particularly relevant in the context of targeted interventions, which may be more effective for some neurobiological subtypes than others – a possibility that remains largely unexplored and highlights the need for further research capable of reliably identifying and replicating such subtypes across diverse samples and methodological approaches.