Nonshared environmental factors in the aetiology of autism and other neurodevelopmental conditions: a monozygotic co-twin control study

Neurodevelopmental conditions (NDCs) are childhood onset disabilities that persist into adulthood and are associated with altered architecture, maturation, and functioning of the developing central nervous system [1]. NDCs affect a wide range of mental functions, such as attention, executive control, top-down versus bottom-up processing, speech-language and communication, motivation, motor control, learning, and social awareness [2, 3]. In the Diagnostic and Statistical Manual of Mental Disorders 5th ed. (DSM-5) and the International Classification of Diseases 11th Revision (ICD-11), the umbrella concept of NDCs includes attention-deficit/hyperactivity disorder (ADHD), autism spectrum disorder (ASD), intellectual disabilities (ID), communication disorders, specific learning disorders, and motor disorders, such as tics and developmental coordination disorder (DCD) [4, 5]. Most NDCs are viewed as extreme manifestations of traits that are continuously distributed in the general population [6]. Even though different NDCs are associated with similar neurocognitive alterations, with an overlap of symptoms and frequent coexisting psychiatric and somatic problems, they differ with regard to their genetic vulnerabilities, clinical course, prognosis and response to treatment [1]. Therefore, the need for both grouping NDCs together while retaining diagnostic distinctions, as well as for including both categorical diagnoses and dimensional traits, has been emphasised [1].

NDCs are common paediatric challenges, with an estimated prevalence of 17.8% in the years 2015 to 2017 in the USA [7]. Although considered highly heritable, a proportion of NDC causality has been attributed to environmental factors [8], especially nonshared environmental (NSE) effects, i.e., exposures unique to different children within the same family that make them differ from each other. For example, twin–twin differences in autistic traits are substantially influenced by NSEs at a clinical severity level [9] and estimates of the NSE impact on ADHD and ASD range between 17 and 44% [10‐12]. Importantly, the identification of valid NSEs in the aetiology of NDCs might facilitate research on prevention of early adverse influences on the developing child.

NSE, or rather indicators of NSE, include prenatal factors such as gestational hypertension and diabetes, gestational bleeding, maternal use of medication and teratogens [13‐16]; perinatal factors, such as breech presentation, foetal distress, preterm birth, caesarean delivery, birth complications, or birth weight [8, 13, 17]; and postnatal factors such as infections and antibiotic exposure [18, 19]. Despite reported associations between NSE and NDCs, the evidence to support a causal relationship remains insufficient [16] and given the considerable overlap between different types of NDCs [1] it remains to be determined if the associations are generic to NDCs or if certain environmental factors are linked to specific conditions. For instance, the interaction between the same NSEs with different genetic vulnerabilities may result in an elevated likelihood for specific NDCs, or the association between NSEs and NDCs may be across diagnoses/conditions, where changes in brain development may mediate the association by means of shared neurocognitive alterations.

Thus, research designs are needed that control for the broad range of familial factors (genetics and shared environmental factors) indicated in NDCs [17]. The discordant monozygotic (MZ) co-twin design which automatically controls for both genetic factors (MZ twins, also called identical, are assumed to share essentially 100% of each other’s genes) and shared environment (including the prenatal environment) offers unique opportunities to investigate possible causal pathways to NDCs by focusing on the within-twin pair differences with respect to unique environmental experiences [20]. There is however a shortage of studies investigating MZ twin populations that are sufficiently phenotyped for concordance and discordance with regard to NDCs. In addition, most previous twin-studies were limited by small sample size, a lack of standardized assessments, and by overly focusing on single NDC diagnoses. In spite of these limitations, some support has been found for a within pairs association between lower birth weight and ASD traits [21]. Similarly, within pairs, the twin with lower birthweight had a higher risk of having ADHD [22] and traits of ADHD [23]. Birthweight discordance has also been associated with cognitive and verbal developmental delay [24] and lower IQ performance at 3 years of age [25] in the lower weight twin. Twin studies have also reported an association between perinatal hypoxia and respiratory distress and ASD [26] and DCD [27]. In addition, the cumulative load of early medical events has been linked to both ASD and autistic trait severity [28, 29].

In sum, while the general importance of NSE for NDCs is undisputed, it remains unclear which NSE factors may pose a specific risk for a certain NDC and whether exposure to several NSEs may further increase the likelihood of a neuroatypical development. Therefore, this study utilized a MZ co-twin design in order to (1) explore if a wide range of NSE are associated with dimensional (traits) operationalizations of NDCs within twin pairs, (2) test whether different NDCs are associated with different NSE factors and (3) investigate the role of cumulative exposure to several NSE factors for categorical (diagnosis) and dimensional operationalizations of NDCs within twin pairs.

Figure 1 depicts the distribution of and differences in the key dimensional constructs in terms of level of twin-twin discordance. In nine pairs twin-to-twin transfusion was reported. One participant had reported hydrocephalus, one cerebral haemorrhage, one craniosynostosis, one encephalitis and one had meningitis during childhood. Among the participants, 27.7% of the twins had one, 12.9% two and 30.9% three or more perinatal risk factors reported. The corresponding numbers for postnatal risk factors were 32.1% for one, 16.1% for two and 7.1% for three or more cumulative factors. Partial correlation, adjusting for sex and age, between the NSEs and NDCs are presented in Additional file 3: Table S2.

We investigated the significance of nonshared peri- and postnatal environmental factors in neurodevelopmental conditions in a monozygotic co-twin design. Cumulative perinatal and postnatal risks were associated with neurodevelopmental conditions, and especially autistic traits, within the pairs. For perinatal NSE, divergent twin weight was associated with traits of ASD and ADHD for the smaller twin, and among discordant pairs, only the twin with ASD or ADHD diagnosis had a birthweight ≥ 18% compared to their co-twin. In addition, the twin who received medication during infancy had more traits of ASD compared to their co-twin. For postnatal NSE, the twin with convulsions had more ASD traits, and at a diagnostic level only those with a diagnosis of ASD, ADHD or ID reported convulsions within discordant pairs. In addition, frequent ear infections and head injury were associated with ASD traits, whereas scarlet fever and jaundice were associated with lower IQ.

Since unmeasured familial confounders, such as genes and shared environment are adjusted for, the discordant MZ co-twin design offers unique opportunities to investigate possible causal NSE pathways to NDCs. In our study, twin birth weight discordance was the perinatal factor contributing most to the likelihood of emerging autistic and ADHD traits, as well as lower IQ when applying birthweight as a continuous measure, for the twin with less weight. In line with this, both ASD traits [21], ADHD [22], traits of ADHD [23] and IQ [25] have been previously associated with birth-weight difference, with the smaller twin showing increased severity of problems. Moreover, birth weight discordance may double the risk of other adverse perinatal outcomes among twins, including mortality, hypoxic-ischemic encephalopathy, respiratory distress, or sepsis [39].

Since twins have the same gestational age, a difference in their birth weights occurs due to the factors affecting the individual growth of each twin. Possible mechanisms for discordant growth include twin–twin transfusion syndrome or unequal placental sharing [39], although the association between weight discordance and adverse outcomes in the study by Breathnach and colleagues [39] remained even after excluding twins with transfusion syndrome [37]. Other factors that may influence birth weight discordance include maternal factors, such as older age and tobacco use, fetal risk factors, such as viral infections, and placental factors, such as velamentous cord insertion [38]. Hypothetically, a growth discordance might influence brain development in ways specific to ASD and ADHD, as it has been reported that variation in birth weight among MZ twins may be associated with alterations of brain anatomy persisting into late adolescence with lower birth weight being associated with reduced brain volume, and more specifically with altered cortical surface area [44], a cerebral area of significance to ASD [45]. The use of medication reported in our study, such as antibiotics, during infancy for the twin with higher autistic traits may increase microbiome depletion and bacterial dysbiosis, also reported in children with ASD [18].

For postnatal NSE factors, convulsions were associated with ASD traits, albeit no longer significant when correcting for multiple comparisons using Benjamini–Hochberg FDR corr. Even though the cause of convulsions was not investigated in this study, and only few participants had reported convulsions, previous research suggested that children with epileptic and non-epileptic convulsions in early childhood have a higher degree of neurodevelopmental and psychopathology scores, as compared to controls, at least nine years later [46, 47]. Epilepsy has also been more commonly reported among individuals with ASD, as compared to peers with ADHD [48]. A history of convulsions has also been linked to structural correlates, such as decreased hippocampal volume and white matter tract changes [46]. In our study, head injury during early childhood and frequent ear infections, was associated with more autistic traits, and scarlet fever and jaundice were associated with lower IQ. The association between scarlet fever and IQ remained significant even after using the FDR corr. Even though streptococcal infection has been associated with a wider range of NDCs [49] and exposure to infections in early childhood has been associated with lower IQ in adolescence [50], more research is needed to corroborate our findings. The association with head injury needs to be interpreted with caution because of the moderate degree of agreement between parental reports and medical records and the association lost its significance after the FDR corr.

Generally, our findings are consistent with the previous research on NSE in NDCs [8], indicating, an association between autism and perinatal and postnatal risks within twin pairs. Our finding thus suggests that autism may be particularly sensitive to cumulative early environmental risk load. Previous research has also reported a potential cumulative effect of the number of environmental adversities on the severity of ASD and autistic traits [28, 29, 51]. Accordingly, it has been suggested that ASD may be triggered by additive or cascading effects of high-level exposure to endogenous and environmental factors, hypothesised to affect the immune system and central nervous system functions during critical developmental stages [52]. In contrast with findings from previous twin and sibling studies [8], perinatal hypoxia and respiratory distress were not associated with neurodevelopmental outcomes in our study. This negative finding could reflect limited power in our model and that we used a proxy (oxygen therapy) for respiratory distress. We did however observe 11-raw score points more on the SRS-2 for the twin who received oxygen, although not significant.

In order to fully evaluate the validity of our findings, some limitations of the current study need to be discussed. First, data on NSE were based on retrospective parental inquiry, and may be subject to confirmation or recall bias. To maximize data validity, the self-reported medical information from home-stored medical reports and record-keepings was validated against medical registry data. While ratings on birthweight had an excellent agreement with the medical records, other NSEs had a lower agreement, e.g., good agreement for oxygen therapy (hypoxia and respiratory distress) and moderate agreement for breech position, which should be considered when interpreting the results. Second, the NSE factors should be understood as rather indirect indicators of environmental risks for NDCs, e.g., phototherapy, why future studies should seek to assess primary medical events as authentically as possible, such as hyperbilirubinemia. Third, several other potential indicators of NSEs that may be of importance, including Apgar score, were not included. Since twins share prenatal milieu, prenatal environmental factors such as maternal use of medication and teratogens could not be included in the within-pair analysis. Fourth, considering the design of the study, we were unable to draw any firm conclusions about the possible causality. Even though the MZ-twin design provides a unique opportunity to untangle the contribution of NSEs on development, reverse causality could not be excluded where NDCs or other factors related to these conditions may increase the risk for NSEs in early development. Fifth, although an MZ co-twin design implies a high degree of genetic and shared environment control, there may be post-twining de novo mutations that may contribute to a minority of cases of discordance. Sixth, concerns have been raised that findings in twin samples may not be generalizable to non-twin populations. However, previous research suggests that twins do not systematically differ from the general population of non-twins on measures of behavior and development [53, 54], including the prevalence of autism [55]. Our sample is also biased toward twins discordant for NDCs. Although this recruitment strategy makes it inappropriate to use bivariate twin models such as A (additive genetics) C (common/shared environment) and E (unique environment) ACE models, it is appropriate for investigating the association between environmental risk factors and outcomes while controlling for familial confounders. Seventh, when exploring the associations between each condition and specific indicators for NSEs the risk of being statistically underpowered increases. In addition, as the case with birth weight discordance and postnatal convulsions, no NSE was reported among the non-affected twins, why it is not feasible to obtain confidence intervals for the RRs in a small sample size using standard measures. Therefore, we only include descriptive data on the association between specific NSEs and NDCs. Lastly, the validity of applying a specific p-value could be discussed. In this study we conducted numerous statistical calculations which increases the risk for type I errors. At the same time, correction for multiple testing with e.g., Bonferroni correction, has been criticized since the interpretations depend on the number of tests that are performed and the likelihood of type II errors is also increased [56]. In this study, we therefore also present results after correcting for multiple comparisons using Benjamini–Hochberg procedure with FDR corr [43], which are deemed to be less conservative than the Bonferroni correction using familywise error rate.

On balance, the strengths of the study include a relatively large and rare sample of well phenotyped MZ twins, including a considerable number of MZ twins in Sweden concordant and discordant for NDCs (for a specification of number of concordant and discordant pairs from a nation-wide population-based study in Sweden see [57], for a more comprehensive review on previous twin research in ASD see [58]). Furthermore, the applied discordant MZ-twin design is a powerful method for investigating the contribution of NSE to neurodevelopmental conditions.

The current study offers insights into early risk factors associated with neurodevelopmental problems and provides additional perspectives on prognosis and prevention. Weight discordance at birth was an important risk factor for both autistic and ADHD traits. Medication early in life, as well the presence of convulsions and streptococcal infection in the first year of life were also associated with traits of NDCs in the within-pair analyses. Even though our finding needs to be interpreted with caution due to a small number of pairs being discordant for specific NSEs, the findings extend previous research by applying an MZ co-twin control design with a high degree of genetic and shared environment control. Our results support a relationship between early NSE and NDCs, where autism seems to be particularly sensitive to a cumulative impact of negative early environmental events. As previously suggested [9] and as supported by the present study, autistic traits tend to be continuously distributed even in discordant twins. It may be hypothesized that when the traits approximate a diagnostic threshold, multiple exposures to specific risk factors, such as those described in the present study may act as triggers and aggravate the phenotype to a clinical level. The findings highlight the need for interventions targeting factors contributing to fetal growth reduction including viral infections and maternal health problems. Improving prenatal care might influence a number of risk NSEs. Further identification of valid NSEs in the aetiology of NDCs may facilitate prevention efforts with regard to early adverse influences on the developing child.