Editorial Perspective: Furthering research on temperament in autism spectrum disorder

Heterogeneity is a long-recognised feature of autism spectrum disorder (ASD), observed in the presentation and severity of core/associated symptoms as well as trajectories of symptom onset and life span course. Identifying the sources of variation in ASD symptomatology and clinical/life outcomes is critical for the development of more targeted and individually tailored recommendations and interventions that will, in turn, improve life outcomes for individuals with ASD. It is becoming increasingly clear that maintaining the search for disorder-specific sources of heterogeneity may not be an effective means of achieving this goal. Although philosophically different, the framework of developmental psychopathology and National Institute of Mental Health's (NIMH) Research Domain Criteria (RDoC; Insel et al., 2010) initiative both emphasise the need to go beyond current symptom-based categorical demarcations of mental health/illness to explain clinical phenomenology at the fundamental biobehavioural level. We propose that captilising on both traditions holds promise for providing a richer mechanistic understanding of heterogeneity in ASD, and specifically nominate temperament as a relevant construct for future such work. Temperament reflects early emerging emotional and behavioural traits that result from interactions among genetic, biological and environmental influences (Shiner et al., 2012). While the structure of temperament remains a topic of continuing debate, we consider temperament to be a composite of three high-order dimensions: (a) negative emotionality, the tendency to experience negative emotions, (b) sociability, the tendency to actively and surgently engage with others and (c) self-regulation, the capacity to regulate cognitions, emotions and action. We illustrate in Table 1 how these high-order dimensions map onto different domains of functioning proposed within the RDoC framework, have distinct neurological substrates, and are encompassed within the dominant theoretical models of temperament (for a review, see Shiner et al., 2012). These high-order temperament dimensions interact in complex ways to affect positive and negative developmental outcomes, such as academic achievement and social-emotional competence (Shiner & Caspi, 2003). Extensive literature also suggests that temperamental variation may be associated with differences in susceptibility and resilience towards psychopathological outcomes. For instance, in two-independent samples of children and adolescents, Hankin et al. (2017) found that high negative emotionality and low self-regulation conferred broad-based, transdiagnostic risk towards general psychopathology (i.e. p factor). Moreover, low self-regulation was uniquely related to externalising liability while high negative emotionality and low sociability were linked more specifically to internalising liability. Considerably less is known, however, about how temperament relates to the variance of core and associated ASD features. We have recently reviewed 40 studies on temperament in the context of ASD and found that much existing research has been descriptive in nature – comparing the temperamental attributes of ASD-diagnosed individuals to those with normative development and/or other clinical/developmental conditions (Uljarević et al., 2018). This conventional strategy of between-group comparison is predicated on the assumption that all individuals with ASD diagnoses share the same temperamental attributes. However, such a possibility seems highly unlikely given that the aetiology and consequent phenotypic expression of ASD symptoms is highly heterogeneous. We therefore argue that there is a need to reorient research efforts towards exploring temperament as a predictor of individual differences within the ASD-diagnosed population. As an illustration, we identified a small number of studies reporting concurrent associations between temperament and the severity of core and noncore/associated ASD features (Uljarević et al., 2018) – that is, an emerging evidence base broadly supporting our position that temperament constitutes a transdiagnostic factor contributing to heterogeneity in outcomes for individuals with and without ASD alike. Nevertheless, this line of research remains in its infancy. Here, we highlight how the RDoC approach may further our understanding of temperament in ASD. We begin by identifying key conceptual and methodological avenues for improvement and conclude by raising some unanswered questions for the field. There is a need to acknowledge and address confounding in the conceptualisation and measurement of temperament and core/associated ASD symptoms. For example, most measures of temperament and internalising psychopathology tap behaviours related to social withdrawal and inhibition, which form part of the ASD diagnostic criteria. A failure to address such overlap might artificially inflate the strength of associations. While one straightforward solution is to eliminate overlapping items from temperament and outcome measures, we propose that estimates of item similarity should be empirically derived (e.g. through joint confirmatory factor analyses) rather than based solely on researcher judgment. Next, it is crucial to select temperament measures that tap the biologically based, high-order traits of negative emotionality, sociability and self-regulation. Parent-report questionnaires represent a long-standing measurement tradition, and those based on Mary K. Rothbart's conceptualisation are one appropriate option. While this set of measures includes slightly different fine-grained subscales for specific temperament traits, there is convergence around three overarching dimensions – surgency, negative affectivity and effortful control (referred to as orienting/regulation in infancy) – that are grounded in biology and align closely with our proposed three-factor taxonomy (Table 1). Nevertheless, we advocate the need to supplement questionnaire information with other, more objective data. In the context of ASD research, most studies of temperament have relied soley on the parent-report of temperament, offering numerous practical advantages over observational/laboratory-based indices, and drawing on parents’ extensive knowledge of their children to provide a rich picture of behaviour across contexts. However, parental response biases may well be at play. For instance, parent ratings on temperament scales may be influenced by dispositional characteristics, transient mood state, mental health and perceptions of the parent–child relationship. Research comprising multiplex ASD families – including studies of high-risk infant siblings – should also consider the potential of parents to either inflate (i.e. assimilation effect) or underestimate (i.e. contrasting effect) the degree of temperamental similarity between siblings by evaluating them relative to one another. While this bias likely operates across all parent-rated temperament measures, questionnaires that call for global judgements (e.g. ‘child cries easily’) may be more susceptible than those seeking reports on behaviour within specified contexts (e.g. ‘child cries before going to sleep’; Saudino, Wertz, Gagne, & Chawla, 2004). The principal alternative approach is direct observation of child behaviour in the home or during laboratory-based assessments. While this method may afford greater objectivity than parent-report questionnaires, this nevertheless carries limitations in terms of test–retest reliability and ecological validity. Hence, we advocate a multimethod/multi-informant approach to measuring temperament in the context of ASD diagnosis, combining self/other reports and behavioural observations. We also encourage the continuing development of ecologically valid indices tapping these traits. For example, experience-sampling methods could provide a way of measuring temperament on multiple occasions while ‘in-the-moment’, circumventing retrospective recall biases and behavioural artefacts created by the laboratory environment. Moreover, it will be important to incorporate temperament measures across different units of analysis – from observable behaviour to underlying neurobiological systems (i.e. genes, cells, molecules and circuits; Insel et al., 2010) – to gain a more mechanistic understanding of individual differences in ASD-diagnosed samples. Different neuroimaging modalities capture different structural and functional properties of the brain, and multimodal neuroimaging indices are necessary to achieve a comprehensive understanding of how individual differences in temperament domains map onto variation in the structural and functional integrity of the specific features and circuits. More specifically, while structural magnetic resonance imaging (MRI) and diffusion tensor imaging (DTI) capture structural properties of brain features and connectivity, task-evoked and resting-state functional MRI capture different aspects of functional dynamics and integrity of particular circuits. Hence, utilising these methods in isolation to explore, for example, the neurobiological underpinnings of cognitive control will not provide comprehensive mapping across the unit of analysis. Going beyond measurement-related issues, the question of precisely how temperament and ASD are associated is an important one for the field. Several theoretical models seem plausible: that temperament (a) represents a predisposition towards the development of ASD (i.e. vulnerability association), (b) alters the manifestation of ASD symptoms after their onset (i.e. pathoplastic association) or even (c) exists on the same continuum as ASD, such that ASD is an extreme variant of continuously distributed temperament traits (i.e. spectrum association). Shiner and Caspi (2003) provide a detailed review of these competing accounts. Characterising the nature of the temperament-psychopathology relationship can only be addressed through prospective longitudinal research design. For instance, empirical evidence providing support for the possibility of a spectrum – rather than vulnerability or pathoplastic – association would come from evidence of corresponding longitudinal changes in temperament and ASD features. We also emphasise that research seeking to understand associations between temperament and features of ASD need not be restricted to the examination of diagnosed individuals who have ‘clinical’ levels of ASD symptoms. There is growing recognition that the ASD phenotype has a spectrum of expression in the neurotypical population and across samples of people with many different neurodevelopmental/psychiatric disorders. That is, the diagnostic features of ASD are themselves transdiagnostic. Hence, research on the temperament-ASD relation is well-suited to an RDoC-informed design, involving the recruitment of study participants spanning multiple disorder categories. Even while maintaining a core objective of understanding heterogeneity in the context of ASD diagnosis, researchers need not recruit solely on the basis of DSM-defined diagnostic categories. Rather, recruiting transdiagnostically, the ASD diagnostic label would effectively be invisible, allowing the examination of how neural circuits and systems (e.g. temperament) contribute to individual differences in social-communicative skills and/or restricted/repetitive behavioural features. While this research design deviates from the between-group comparison approach familiar to researchers with an interest in ASD, it holds the exciting potential to achieve true mechanistic understanding of development that cuts across diagnostic boundaries. Future research should also employ person-centred statistical techniques to identify ‘natural’ subgroups of individuals who share similar temperamental attributes, rather than maintaining the traditional focus on variable-centred analyses. Person-centred methods – such as cluster and profile analysis – are well-suited to the study of temperament because, unlike variable-centred regression analysis, they take into account the nonorthogonal nature of temperament traits which may be critical in seeking to draw conclusions about predictive associations with other factors within a given sample. The value of this approach has recently been exemplified by Karalunas et al. (2014), who used a community detection clustering technique and identified three temperament subtypes among children with ADHD diagnoses. Mild, Surgent and Irritable subtypes were distinguished by unique patterns of cardiac physiological response and resting-state functional brain connectivity, stable over time and predictive of clinical outcomes 1 year later. Notably, these subtypes were also independent of clinical demarcations of ADHD symptom severity and presentation. Finally, it will be important to consider the role of the environment on associations among temperament and core/associated features of ASD. In the broader literature, child temperament and parenting behaviours have been shown to shape one another over time and to interact in predicting child outcomes. While the association between temperament and parenting has received little empirical attention in the context of ASD, emerging evidence suggests that dyadic parent-child interaction may be less synchronous when children with ASD have higher negative emotionality and lower self-regulation (e.g. Hirschler-Guttenberg, Golan, Ostfeld-Etzion, & Feldman, 2015). Further longitudinal research is needed to clarify how the interplay of child temperament and the parenting environment contributes to heterogeneity in ASD; specifically, whether parenting practices may attenuate or intensify the effects of child temperament on outcomes, and/or whether the effects of parenting may vary as a function of child temperament. In conclusion, it is our view that research in the context of ASD lags well behind that being conducted in other clinical fields in maintaining an almost exclusive focus on describing group-level differences in temperament across samples of participants with/without a diagnosis. We call into question the utility of this conventional approach and recommend that future research efforts be directed towards more comprehensive exploration of temperament as a predictor of individual differences. This will necessitate a shift towards a multimethod/multi-informant measurement approach, informed selection of instruments, more person-centred statistical methods, and the application of more rigorous research designs informed by the transdiagnostic RDoC framework. For this editorial perspective, the authors have declared that they have no competing or potential conflicts of interest.