Introduction
Autism is a neurodevelopmental condition characterised by difficulties in social interaction and communication, together with restricted interests and a tendency to engage in repetitive behaviours [
1]. Aetiology is complex, with an interaction of genetic, environmental, and neurodevelopmental pathways thought to lead to clinical manifestation [
2]. The first behavioural signs typically emerge in early childhood [
3] and are accompanied by atypical development of brain structure, function, and connectivity, which are hypothesised to play a role in behaviours across the lifespan [
4]. Characterising the neural correlates of autism has therefore remained a focus of the field [
5].
Advances in neuroimaging technology in the last two decades have allowed the development of novel in vivo imaging methods of the human brain. Of these, structural magnetic resonance imaging (sMRI) has been most widely used to characterise the neuroanatomy of autism [
6,
7]. The identification of a neural endophenotype could help inform clinical care, such as earlier diagnosis and intervention, and be used in the subtyping of individuals within the heterogeneous autism umbrella [
8]. An additional motivating factor has been that a robust biomarker could help distinguish autism from other neurodevelopment disorders that have overlapping clinical features, enabling individuals to access targeted treatments. For example, a recent study investigated whether distinct morphological differences could be detected when comparing autistic participants to those with obsessive–compulsive disorder (OCD) and attention-deficit/hyperactivity disorder (ADHD) [
9]. However, whilst differences were identified, including thicker cortical grey matter in frontal regions of autistic participants compared to the other clinical groups, significant overlap between groups was observed. Finally, if associations were identified to be causal, knowledge of the specific regions implicated could provide mechanistic insights and help inform novel therapeutic strategies [
10].
The neuroimaging literature demonstrates considerable heterogeneity regarding direction and effect size of brain morphology differences in autism [
5,
11‐
13]. Whilst this may reflect the high level of aetiological and neurobiological heterogeneity across the autistic spectrum [
14], methodological factors are thought to be a contributing factor. Firstly, existing studies tend to be in relatively small samples, leading to overestimation of effect sizes and low reproducibility [
15]. Secondly, there is substantial heterogeneity in study design, including differences in participant characteristics such as age and symptom severity, covariates controlled for, and neuroimaging outcomes assessed. Finally, analytic differences, such as variation in MRI acquisition, processing, and analytic pipelines, impact results derived from individual studies [
5]. Two mega-analyses (
n ~ 3000) from the Enhancing Neuro-Imaging Genetics through Meta-Analysis consortium aimed to characterise the neuroanatomical correlates of autism, whilst addressing these issues [
16]. Global measures were found to be higher in autistic participants, including intracranial volume (ICV), total grey matter, and mean cortical thickness. Regional differences in cortical thickness, including increases in the frontal regions and decreases in the temporal regions, were also observed. In contrast, no differences in cortical surface area were found [
17]. Altered lateralised neurodevelopment was also revealed, with reduced regional asymmetry in cortical thickness and area, and increased asymmetry in the putamen [
18]. In addition, a recent landmark study of autism identified early differences in brain morphology associated with a diagnosis. In a cohort of infants at high likelihood of being autistic,
Shen et al. [
19] identified that those who went on to be diagnosed with autism showed a faster amygdala growth between 6 and 24 months, and larger volumes at 12 months, when compared to typically developing controls.
An additional limitation of the existing literature is the use of categorical diagnoses when assessing neural correlates of autism [
6]. There has been little focus on identifying the brain morphology correlates associated with subclinical autism in typically developing populations. Subclinical autistic traits include social communication differences alongside restricted behaviour patterns, which does not cause difficulties for everyday functioning [
20]. The behaviours associated with autism can therefore be considered continuous traits, which extend into the general population [
21,
22]. Genetic studies have demonstrated that liability to autism influences typical variation in the population of social–emotional interaction and communication ability, providing further evidence for the importance of studying autism-related phenotypes in a quantitative manner [
23]. With this wealth of evidence that autistic traits fall along a continuum in the general population, it can therefore be hypothesised that the neuroanatomical differences associated with autism also extend into the general population.
Two longitudinal studies have examined cortical correlates of autistic traits in community-based samples, using measures from the Social Responsiveness Scale (SRS) [
24]. Higher SRS scores were correlated with reduced regional cortical thickness including the right superior temporal sulcus [
25] and the middle temporal gyri, ventral precentral and postcentral gyri, anterior cingulate, and right frontopolar cortex [
26], which remained stable from childhood to adolescence.
Associations between autistic traits and brain morphology have also been examined in the Generation R Study, a population-based longitudinal cohort. Cross-sectional vertex-wise modelling was used to demonstrate autistic traits measured at age 6 years were associated with properties of cortical morphology, including surface area, thickness, and gyrification in late childhood [
27], and adolescence [
28]. Importantly, differences persisted after exclusion of autism cases, providing further evidence for the extension of autistic traits into the general population. Differences in brain structure associated with varying levels of autistic traits may therefore reflect alternate trajectories of brain development, which in turn are associated with behavioural differences across this continuum.
Whilst this work has begun to reveal the neurobiological differences associated with autistic traits, there remains a gap in the literature regarding differences in subcortical morphology. Given that previous studies show differences in subcortical structures when comparing autistic participants with typically developing controls, which has included reports of both increases [
19,
29] and decreases [
17] in volumes of specific ROIs, and the plausible role of these structures in the socio-motivational, cognitive, and motor symptoms seen in autism, it will be important to explore whether differences in these structures are observed in non-clinical samples.
In addition, children with autism are at increased risk of mental health issues and frequently present with problems in emotion, attention, and behaviour [
30]. Whilst prevalence varies greatly, anxiety disorders, depression, OCD, ADHD, and specific phobias are most consistently reported as secondary psychiatric disorders co-occurring with autism [
30‐
32]. Autistic traits have also been identified as a risk factor for poorer mental health, with associations appearing stronger in childhood than adulthood [
33]. Whilst co-occurring psychopathology will confound behavioural-brain associations, such traits are not routinely controlled for in the existing autism neuroimaging literature [
6]. The fact that the majority of studies are in clinical or community-based samples will bias towards a high occurrence of multiple diagnoses, and therefore, there is a gap in the literature for the application of methods in epidemiological cohorts. The identification of brain morphology features that remain associated with autistic traits beyond correction for co-morbidities will help delineate the biological underpinning of autism from other neurodevelopment disorders with overlapping clinical features.
In our study, we aim to expand on existing literature by exploring whether autistic traits are associated with differences in subcortical morphology, and whether any observed differences are explained by co-morbid psychopathology. We present the first population-based analysis of subcortical morphology associated with autistic traits, in an epidemiological sample of 9-to-10-year-old children participating in the Adolescent Brain Cognitive Development (ABCD) Study (n = 7005). Firstly, we explored whether a quantitative measure of autistic traits was associated with differences in child subcortical morphology (Aim 1). Secondly, to understand if any identified neural endophenotypes were specific to autistic traits, we tested whether associations persisted after controlling for co-occurring internalising and externalising symptoms (Aim 2).
Discussion
There is emerging evidence that the neuroanatomy of autism falls along a continuum within the general population. Whilst several studies have assessed cortical phenotypes of autistic traits [
26,
47], there remains a distinct gap in the literature regarding subcortical morphology. Thus, the primary aim of the present study was to investigate the association of autistic traits in childhood, measured by parent reported SRS score, with subcortical brain morphology. Our second aim was to test whether any observed differences were robust to adjustment for co-occurring psychopathology, measured as total scores of externalising and internalising symptoms. To our knowledge, this is the first such study to examine this association within the general population and therefore represents a novel contribution to the current body of literature.
To summarise, in this study of school aged children in the ABCD cohort, we did not find strong evidence for an association of autistic traits with differences in the subcortical volumes assessed, with results compatible with the null hypothesis and generally wide confidence intervals throughout.
Whilst we observed lower absolute volumes of the NAcc and putamen in those scoring higher on the SRS, this attenuated towards the null once overall brain size was accounted for. As univariate analyses had demonstrated children in the upper group of SRS scores had on average a lower ICV, this suggests the observed differences were not beyond that of proportional differences in brain size of the children in our sample. This finding of a reduced global measure of brain volume is in line with other studies assessing the neural correlates of autistic traits in epidemiological samples [
27,
47]; however, it is important to note this is not consistent with findings from clinical populations [
5,
13].
In the ABCD sample, being male or of white ethnicity was associated with higher SRS scores. It has been previously reported that in samples from the general population, male children tend to have higher SRS scores than female [
27,
38,
48]. In contrast, there is little published literature regarding distribution across ethnic groups, and therefore, this is an area which requires further investigation.
The existing literature is composed predominantly of studies using a case-cohort design. Most notably, findings from the ENGIMA consortium identified lower volumes of the pallidum, putamen, and NAcc in participants with autism compared to controls [
9].
Post hoc analyses demonstrated these differences were related to the degree of autism symptom severity, measured by scores extracted from the Autism Diagnostic Observation Schedule (ADOS) [
49]. Although it must be noted that there are qualitative differences between the SRS and ADOS [
50] and that these findings from the ENGIMA consortium have not yet been replicated, we had hypothesised we may see similar effects of a smaller magnitude focussed on these specific ROIs when examining the correlates of SRS scores in our sample. One possible explanation for our null results is that the differences in subcortical morphology observed in autism cases may represent neurobiology associated with a higher degree of autistic symptoms that meet the criteria for a clinical diagnosis.
A further study from the Generation R neuroimaging cohort, whilst predominantly focussed on cortical morphology, examined one subcortical ROI in relation to autistic traits [
26]. The authors utilised a sample from 9- to 12-year-olds in the Netherlands (
n = 2400), examining amygdala volume in relation to SRS scores. In line with our findings, amygdala volume was found to not differ significantly with SRS score when covariates were accounted for. In contrast, strong evidence was found for differences in metrics of cortical morphology, including lower gyrification, thickness, and surface area, suggesting that autistic traits in this sample are primarily associated with cortical, rather than subcortical ROI, differences.
Our second aim was to explore the role of co-occurring psychopathology, to understand if neural phenotypes were specific to autistic traits or simply a reflection of generalised psychopathology. Inclusion of these covariates had little impact on effect estimates; however, given that we found little association with SRS scores alone and that univariate analyses did not demonstrate strong associations of these covariates with our outcomes of interest, this is unsurprising.
It is important to note that whilst we did not detect significant group differences in subcortical ROIs, it is possible these volumetric measures are not sensitive to what may be more subtle differences exerted by autistic traits in the general population. Aggregate measures such as volume do not fully capture the complexity of subcortical structures and may be insensitive to specific local effects, or obscure heterogeneous local effects by averaging out subtle differences in shape [
51]. This is particularly true for phenotypes which are likely characterised by specific associations with functionally distinct subfields of subcortical structures, such as traits of autism. Therefore, our lack of detectable volumetric differences in subcortical ROIs may be due to analytic methods, which do not allow for these subtler differences to be assessed. Whilst no studies have specifically used shape-based methods when assessing the subcortical correlates of autism, it has been demonstrated that for other neurobehavioral phenotypes, these methods provide more information than volumetric methods alone. For example, a recent study examining the subcortical alterations associated with major depressive disorder found little difference in subcortical volumes, beyond that of lower hippocampal volume [
52]. In contrast, subsequent analyses using shaped-based methods identified specific effects localised to regions of the amygdala and hippocampus associated with patients in comparison to controls [
53]. Complementary analyses, using shape-based analytic methods, will therefore be necessary to understand if autistic traits are associated with more sensitive markers of difference in subcortical morphology.
Limitations
When interpreting our findings, several limitations must be considered. Firstly, as the ABCD cohort excluded participants with a moderate or severe autism diagnosis (based on whether a child’s caregiver reported they did not attend mainstream school), the average severity of autistic traits will be artificially lower than in the general population, and therefore, findings may be biased towards the null. Secondly, as information regarding whether children had received a clinical diagnosis of autism was not available, it was not possible to conduct sensitivity analyses excluding these participants. Thirdly, neuroimaging measures and SRS scores were not contemporaneous; however, given the relatively short time period between clinics, and that autistic traits have been shown to remain stable over time [
54,
55], this will likely have had limited impact. Fourthly, it is important to note that the SRS is contaminated by general behavioural problems [
50] and therefore may not be wholly indicative of autism-specific symptoms. For example, SRS scores have been shown to be higher when co-occurring conditions are present, such as mood disorders [
56], and child behaviour problems account for a significant proportion of the variance in SRS scores [
57]
. In addition, the 11-item SRS, rather than full 65-item SRS, was used in the ABCD cohort to reduce participant burden. Whilst the brief measure has been used previously [
39], it is possible it may be a less sensitive marker of autistic traits than the full scale. This point, alongside the exclusion of participants with moderate/severe autism, may have reduced power to detect brain morphology correlates of autistic traits in this sample. It will be important to replicate this analysis in samples fully representative of the general population to enrich the higher end of score distribution for these traits.
Fifthly, our analyses were based on sMRI data obtained at a single time point, limiting our analyses to a cross-sectional design. Currently, there is limited longitudinal analysis of brain morphology outcomes associated with autistic traits in the general population, with a single study finding cortical morphology differences associated with autistic traits in the general population remain relatively stable over time [
28]. Interestingly, this finding is not consistent with studies examining brain morphology associated with an autism diagnosis, with differences in developmental trajectories of total brain volume and subcortical morphology identified [
19,
58]. It will therefore be important to replicate these previous epidemiological findings, and test associations in the context of subcortical morphology. As the ABCD cohort is an ongoing, longitudinal study, it will provide the ideal sample to continue examining these trends as further data are released, to understand if autistic traits are associated with individual or group differences in trajectories of subcortical volumes [
59]. Finally, it is also important to note that neuroimaging phenotypes were derived using FreeSurfer 5.3, as described in the ABCD Release Notes for Data Release 3.0 (
https://nda.nih.gov/abcd/). As newer versions are now available, this must be considered as a source of heterogeneity if comparing study findings to those using updated software.
These limitations must be also contrasted against the multiple strengths of our study. Firstly, data were drawn from a large population-based cohort with autistic traits measured continuously. The use of a dimensional approach, rather than a case-cohort design, is better suited to the idea of an autism spectrum and allowed us to test whether the underlying subcortical neurobiology of these traits extends into the general population. In addition, the ABCD cohort is socioeconomically, ethnically, and racially diverse, whilst being relatively homogenous regarding the age of participants. This allowed the generation of a representative estimate of the association of autistic traits with subcortical morphology, minimising the selection bias that has hindered previous studies in clinical samples. In addition, the wealth of phenotypic data available allowed us to control for all identified potential confounders of the exposure-outcome relationship, a significant source of bias in existing studies. In addition, utilising data from the ABCD cohort allowed a large sample size, with a total of 7005 included participants, twofold greater than that of the largest published study in this area.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.