The primary aim of this prospective study was to examine the developmental pathways that contribute to the co-occurrence between anxiety and ASD traits. We investigated how infant reactivity and regulation related to later anxiety in a cohort enriched for later ASD symptoms. The models showed that (1) higher levels of BI at 24 months were associated with anxiety traits at 36 months and this association was not shared with other aspects of infant negativity; (2) more BI at 24 months was associated with more anxiety traits at 36 months, although this was reduced by having better EC at 15 months. (3) there was a similar association between reduced EC, higher BI and ASD; (4) mediation analyses showed that BI was related to ASD traits through emerging anxiety traits whereas EC was related to anxiety through ASD traits, which indicates separate developmental pathways for anxiety and ASD.
Infant BI and Anxiety
Model 1 confirmed that heightened reactivity in infancy was associated with higher levels of anxiety in our cohort, consistent with the existing literature in the general population (Bufferd et al. 2016
; Buss 2011
). Specifically, heightened BI at 24 months was related to heightened levels of anxiety scores at 36 months. This is consistent with a previous study showing that shyness in toddlerhood was significantly correlated with anxiety symptoms in a cohort of infant siblings followed up to age 7 (Shephard et al. 2018
Prior evidence has suggested increased negative affect (Macari et al. 2017
) and BI in children with an ASD diagnosis (Brock et al. 2012
), and temperament traits have been investigated in relation to ASD traits at 36 months in infants at risk of ASD (Garon et al. 2016
). However, few studies have examined associations between these temperament traits and emerging internalising problems (though see Burrows et al. 2016
; De Pauw et al. 2011
; Shephard et al. 2018
). Thus, Model 2 assessed whether the observed associations between BI and anxiety in the present sample were specific, or whether they may extend to other aspects of negative affect, specifically sadness. Planalp et al. (2017
) argued that BI and sadness share similar reactions such as withdrawal, but the context in which we observe them differs. Specifically, BI is characterised by reactions to novelty or over-arousing stimuli whereas sadness is characterised by broader reactions to goal blockage or loss without an approach/withdrawal orientation. Our results showed a significant concurrent association between BI and sadness at each time point and higher BI in infancy was related to increased sadness in toddlerhood (Model 2). However, infant/toddler sadness was not related to anxiety outcome. In line with Planalp et al. (2017
) concurrent association at each time point may indicate that the association between BI and sadness share similar underlying features (e.g., withdrawal) in the current cohort but the relationship between infant BI and anxiety confirms that prediction of anxiety is specific to BI.
Following Planap et al. (2017
), infants’ intolerance of uncertainty in a novel context may explain why infants’ anxiety is specifically related to BI. Intolerance to uncertainty is the perception that an uncertain, unpredictable context is stressful and threatening, and this perception results in avoidance (Anderson et al. 2012
). Indeed, in children and young adults with ASD, intolerance to uncertainty may be the mechanism that explains the variation between anxiety and ASD traits (Boulter et al. 2014
). Thus, elevated worry in ambiguous or over-arousing situations may be an early form of intolerance to uncertainty, which results in maladaptive emotionality in infant siblings.
Longitudinal Association Between BI, EC and Anxiety
Model 3 shows how EC interrelates with BI in the prediction of later anxiety. EC has been proposed to be a protective factor against a wide range of conditions including both ASD (Johnson 2012
) and anxiety (White et al. 2011
). Thus, EC would be expected to interrelate with other risk factors in shaping later psychopathology. The cross-lagged approach we employed allowed us to test this possibility directly. Specifically, we explored the longitudinal directionality of the association between BI and EC over the time window in which behavioural symptoms of ASD and anxiety emerge. As expected, better EC at 15 months was associated with less BI at 24 months. Furthermore, successfully employed EC in infancy reduced the levels of dysregulated BI in toddlerhood, in turn, lower levels of BI in toddlerhood reduced the likelihood of presenting anxiety traits at 36 months. This is consistent with the proposal that EC and related constructs may be broad protective factors against later psychopathology (Johnson 2012
Longitudinal Association between BI, EC, Anxiety and ASD
Since symptoms of anxiety and ASD intertwine, it is also crucial to further investigate whether infant predictors related similarly or differently to ASD and anxiety outcomes. Model 4 showed that for some paths, BI and EC related similarly to later anxiety and ASD traits. Specifically, higher EC at 15 months related to lower levels of BI at 24 months; and lower levels of BI at 24 months related to both fewer anxiety and ASD traits (36 months). This may suggest shared developmental pathways to both anxiety and ASD traits. Alternatively, infant temperamental features may have apparently similar relations to both ASD and anxiety because phenotypic causality drives stronger associations between ASD and anxiety symptoms once they have emerged. To dissect this possibility, we used mediation analyses to explore whether BI and EC at 24 months relate to ASD traits through co-occurring anxiety symptoms, or whether BI and EC relate to anxiety symptoms through ASD traits. This could tell us, for example, whether more fearful infants develop anxiety symptoms that then exacerbate their ASD symptoms, or whether infants who have poorer EC develop ASD symptoms that then lead to anxiety symptoms.
Our results suggest differences between EC and BI in terms of their relationship with ASD and anxiety traits. We found that early BI predicted later anxiety symptoms, and this was only partially mediated by ASD symptoms. In contrast, the relation between early BI and later ASD symptoms were fully mediated by anxiety. This means that the degree to which early reactivity predicts later ASD is likely attributable to more anxious children tending to have higher levels of ASD symptoms in our cohort. This association could be due to phenotypic causality such that anxiety exacerbates social difficulties (Duvekot et al. 2018
). These results suggest that apparent infant predictors of later ASD in sibling studies are actually related to risk for co-occurring conditions like anxiety or ADHD.
Our results diverge from those observed in a slightly smaller cohort in which Shephard et al. (2018
) showed that shyness at 24 months was related to 7-year ASD but not anxiety symptoms. Our observation of associations between 24 months BI and anxiety may have been due to more power associated with our larger sample (N
= 143 in our study, compared to N
= 104 in Shephard et al. (2018
)). Further, our statistical approach examined the longitudinal bidirectional effects between BI and EC (see also, Eggum et al. 2009
; Wolfe et al. 2014
), whereas Shephard et al. (2018
) specifically tested the relation between shyness at 24 months and anxiety scores at 7 years. Incorporating longitudinal data and taking into account the bidirectional relation between EC and BI may reveal patterns of association that are not detectable when focusing on simple associations. Alternatively, there may be stronger associations between toddler BI and early-emerging anxiety traits at age 3 than in later development, where other cascading effects and phenotypic interactions further complicate the relationship. Some studies suggest that anxiety symptoms increase from childhood to young adulthood (Gotham et al. 2015
; van Steensel et al. 2011
), less is known about continuity or change from infancy to mid-childhood. There may be a change in symptom presentation over the course of development and there may not be a gold-standard age to examine anxiety so, to better understand the general presentation of anxiety, assessments should be carried out at different time points.
In contrast, strong early regulation was related to later ASD symptoms and this was only partially mediated by anxiety. Further, the relation between early regulation and later anxiety traits were fully mediated by ASD symptoms. This means that the degree to which early regulation predicts later anxiety is likely accounted for by the fact that children on the path to developing ASD often subsequently develop more symptoms of anxiety. Further, in Model 4 higher EC at 24 months was associated with both reduced anxiety symptoms and ASD traits at 36 months. However, the standardised beta coefficients were over twice as large for the relation between EC and ASD (Fig. 3
). Again, this suggests that lower EC is more strongly associated with later ASD than anxiety. One way to interpret this finding is that infants within our cohort with more genetic/early risk factors for ASD have lower levels of EC. This means infants are less likely to compensate if they also happen to be high in BI (due to genetic or other risk factors), leading to a raised risk of subsequently developing anxiety symptoms. So, anxiety may commonly co-occur with ASD because generally lower levels of EC leave children vulnerable to the effects of other risk factors that may vary in the general population.
Model 4 showed a positive association between better EC at 9 months and more ASD symptoms at 36 months. Many of the items that comprise the EC construct at this age ask about how long the child can pay attention to things in their environment. Thus, this significant association may reflect prolonged visual fixations in infants (“sticky attention”;Elsabbagh et al. 2009a
). This result also corroborates the findings from eye-tracking studies that indicated difficulties in switching attentional focus from peripheral stimuli and longer duration of attention that consistent with processing speed of the exposed stimuli in infants who diagnosed with ASD (Elsabbagh et al. 2009a
; Richard and Lajiness-O’Neill 2015
). From 15 months and on, this association was not significant. Further, by 24 months we saw the expected association between EC at 24 months, and more symptoms of anxiety and ASD at 36 months. By age 2, the EC construct is less heavily influenced by questions about visual attention, engagement, and contains more items asking about inhibitory control, sustained attention. This change in emphasis of the construct may explain this pattern. Alternatively, it may be that atypically strong EC early in development indicates an unusually paced developmental trajectory, which may be a sign of emerging vulnerability. The limited skills of infants may be developmentally beneficial because they facilitate learning (Bjorklund 1997
; Elman 1993
) and so developing strong EC too early may be a risk factor.
Limitations and Future Directions
One limitation of the present study was the modest sample size, particularly within the LR group. Due to this, we were unable to examine the multi-group models which would inform us whether such associations are consistent for both HR and the LR groups. Second, the exploratory mediation analyses should be interpreted with caution since some of the mediator and outcome variables (anxiety and ASD) were measured cross-sectionally. Maxwell et al. (2011
) stressed that a cross-sectional investigation of mediation might lead to substantial bias. A third limitation arises due to relying on parent-reported data, which may increase the risk of informant bias. Having an older child with an ASD diagnosis might affect the parental judgement for the younger child. Moreover, parental anxiety problems have been found to mediate the relationship between temperament and anxiety problems (Gartstein et al. 2010
). This might raise attributional bias and future research should take into account the possible effects of parental psychopathology on childhood anxiety problems. Fourth, another limitation may arise from the instruments used to measure BI. Since the Fear subscales of the IBQ-R includes reactions to novel social and non-social stimuli, it captures unitary constructs of BI well. However, evidence in toddlers and older children indicates that the social and non-social aspects of BI may be separable and differentially related to anxiety (Brooker et al. 2016
; Dyson et al. 2011
). Indeed, this is why the ECBQ contains separable Fear (non-social) and Shyness (social) subdomains. In the present study, results were consistent if we used the ECBQ Shyness and Fear subdomains as a composite (Supplementary Material Section 7) or the Shyness domain alone (which presented greater developmental stability with infant Fear and a better model fit). Future work should examine whether social and non-social components of infant BI can be better dissociated with multiple methods (e.g., Dyson et al. 2011
; Eggum et al. 2009
) and whether these social and non-social components are differentially related to later anxiety and ASD from infancy (or whether this only emerges later in development).
Lastly, employing a single method approach may increase the possibility of yielding results that are specific to a single method or due to shared method variance between predictors and outcome. However, our observation that results were specific to BI and not sadness (extracted from the same questionnaire) mitigates this possibility. Further research should employ a multi-method approach encompassing both parent report and observational measurements to boost ecological validity.