Background
The CDC estimates that 1 in 68 8-year-old US children have an autism spectrum disorder (ASD) [
1]. The observed male-to-female prevalence ratio of approximately 4:1 in the CDC data is consistent with the vast majority of other epidemiologic studies around the world [
2]. Two of the hallmark characteristics of ASD, impaired social interaction and communication, have been recognized as traits that show sex differences in typical development [
3,
4], although debate continues around the exact magnitude, nature, and generalizability of these differences [
5]. While this striking ASD sex difference has long been acknowledged, the mechanisms underlying this difference remain largely unknown.
The nature of sex differences in cognition and behavior among typically developing children has led to the Extreme Male Brain theory of autism, which proposed that ASD is an extreme presentation of a typical male cognitive profile where the drive to “systemize” is stronger than the drive to empathize. A mechanism involving sex steroid exposure could be consistent with the Extreme Male Brain theory [
6]. Evidences from animal research indicate that altering prenatal sex-typical steroid levels can lead to permanent changes in the developing brain [
7]. Androgens, especially testosterone, are produced at higher levels in the male fetus compared to female fetus, and androgens have an established role in fetal brain development and are associated with sex-related differences in behavior [
3,
7,
8]. The predominant source of fetal testosterone exposure is endogenous production by the fetal-placental unit [
9] and, under normal circumstances, only small amounts of maternal testosterone passively diffuse across the placenta [
10,
11]. In typically developing male fetuses, prenatal testosterone levels are higher than in female fetuses, particularly between 8 and 24 weeks of gestation. There is also a neonatal testosterone surge after birth reaching pubertal levels during first 3 months, which then declines to pre-pubertal levels by 6 months of age [
12].
Girls with congenital adrenal hyperplasia (CAH) are exposed to elevated androgen concentrations and have been studied as a model of elevated prenatal androgen exposure in females. Girls with CAH show a more male-typical profile compared to unaffected females in gender-typed play across multiple studies [
5,
13‐
15]; however, findings regarding differences in cognitive abilities such as mental spatial rotation [
16‐
18] and language learning [
18,
19] have been less consistent.
The multiple-threshold liability model explains the sex difference in ASD prevalence as a result of females having a higher genetic threshold for attaining ASD case status than males and, therefore, female cases carry a higher mutational load [
20]. Two studies of high-risk children with ASD siblings did not observe differences in ASD recurrence by the older ASD sibling’s sex [
21,
22]. However, a large population-based dizygotic twin pair study observed that siblings of autistic females show greater autistic impairments than siblings of autistic males [
23], and a similar increase in average symptom severity on the Social Responsiveness Scale among ASD cases in families with female cases was observed in a high-risk study of children with ASD siblings [
24].
Researchers attempting to measure fetal testosterone have assayed amniotic fluid and umbilical cord blood. Amniotic fluid surrounding the fetus is collected via amniocentesis, which is generally performed between 15 and 20 weeks of gestation when the prenatal testosterone levels substantially differ across sexes. Amniocentesis samples are limited to pregnancies with indications for this procedure, however, which creates potential selection bias if the indications for amniocentesis are associated with outcomes under study. Moreover, the exact relationship between steroid levels in amniotic fluid and those in fetal circulation are not known. Umbilical cord blood is an easily collected biological sample reflecting late-term prenatal testosterone, available from any pregnancy. Testosterone levels measured in cord blood are thought to reflect prenatal levels in late gestation after the peak in males [
25].
Several previous ASD studies have used amniotic fluid in an attempt to determine fetal testosterone levels. Two studies found positive associations between amniotic fluid testosterone levels and increased autistic traits among 18–24 month olds and 4 year olds [
26,
27]; however, this was not replicated in a recent study among 3 to 5 year olds [
28]. Another recent birth cohort-based case control study of males found no difference in amniotic fluid testosterone between cases and controls but observed increased amniotic fluid steroidogenic activity, measured as principal components of cortisol and steroids levels along the testosterone synthesis pathway, across cases and controls [
29]. Two published population-based studies of cord blood testosterone level and ASD-related traits from one group found no association [
30,
31]. However, these studies were limited by assessment of the phenotype later in life (~20 years), and thus subjects were exposed to decades of postnatal environmental and developmental influences on behavior, which could influence the ability to detect an association.
The current study aims to investigate the association between umbilical cord blood levels of testosterone and other related androgens and autistic traits measured at 12 months of age and social impairment assessed at 36 months of age in the Early Autism Risk Longitudinal Investigation (EARLI), an enriched ASD risk pregnancy cohort. In addition, we aim to explore whether this association was modified by the sex of the older affected sibling.
Discussion
We investigated the association between prenatal androgen exposure and autism-related quantitative phenotypes among infant siblings of older children previously diagnosed with ASD. Androgens were not associated with autistic traits at 12 months of age within each sex. Male infants showed significantly higher umbilical cord testosterone levels and greater social deficits at 36 months of age; however, elevated umbilical cord testosterone level was not associated with social impairments after adjusting for confounders in the full sample or in infants of either sex. These findings are consistent with previous reports on early adulthood autistic traits and cord blood testosterone level and testosterone to estrogen ratio in the general population [
30,
31] but differed from some studies on early childhood autistic traits and testosterone in mid-gestation amniotic fluid samples [
26,
27], which was not replicated in a recent study [
51]. Our findings also suggest that the association between androgen levels and autistic traits may depend on sex of the older affected sibling with a positive association observed among infants with an older female affected sibling, after adjustment for infant sex as well as gestational and maternal age. This pattern of heterogeneity of effect by sex of the older affected sibling observed at 12 months persisted in analyses with 36-month outcomes.
While these findings of effect modification by older sibling sex must be considered as preliminary and need to be replicated, they are consistent with some prior findings suggesting that distinct etiologic mechanisms may be involved in multiplex families with female affected children. The empirical investigations published to date in multiplex ASD families have examined the influence of older affected sibling sex on recurrence risk, and two recent analysis reported higher recurrence rates in the next-born child when at least one of the previously affected children was female [
24,
52]. This is consistent with an earlier report on two large samples of siblings of ASD-affected children that found higher levels of quantitative ASD phenotype in younger siblings of female probands than in younger siblings of male probands [
23]. However, two large population-based studies [
22,
53] and two smaller prospective studies of infant siblings of ASD-affected children [
21,
54] found no differences in recurrence rates by older sibling sex.
Assuming multi-threshold liability, the model most commonly hypothesized to explain the sex difference in ASD prevalence, females have a higher genetic threshold for attaining ASD case status than males and, therefore, female cases carry a higher mutational load [
20]. If some of these mutations are also related to androgen pathways, then these androgen-dependent mechanisms might be more commonly involved in families with affected females. To date, there have been more than ten genes associated with both autism and autism-related phenotype that are also associated with some aspect of sex steroid function (i.e., synthesis [
55], transport [
55], metabolism [
55,
56], or feedback mechanism [
57]). In addition, genetic variants might also modulate the susceptibility of the developing brain to endogenous testosterone. For example, there are two known polymorphisms in the androgen receptor gene (AR), located on the X chromosome [
58], and, in one of these, the repeat sequence length has been inversely associated with receptor transcription activity [
59,
60]. Shorter variants in this polymorphism have been linked to higher in vitro androgen receptor transcription activity in a kidney fibroblast cell line [
59] and were more prevalent in ASD female cases compared to controls [
60]. A recent study also indicated that some of the genes involved in naturally occurring sexually dimorphic processes are upregulated in ASD postmortem brains [
61].
In further considering the results here, there are also a number of other issues and limitations. First, although we analyzed levels of three androgens separately, given the correlation between androgens and what is known about steroidogenesis, the specificity of individual hormone effects is not certain. Further, the current study relied on cord blood rather than amniotic fluid samples to evaluate androgen levels. These samples reflect late gestation androgen levels and do not capture exposure during the window where testosterone is higher in male fetuses than female fetuses, particularly from weeks 8 to 24 of gestation. Amniocentesis samples, however, are collected around the time of the large male-to-female prenatal testosterone level difference. If this sex difference in testosterone level was critical to brain changes influencing ASD-related behavior, then umbilical cord measurement of androgens would be weaker in testing hypotheses related to prenatal androgen exposure and ASD and could have contributed to our null result. While it has been assumed that the prenatal testosterone level difference is critical to testosterone’s influence on brain development, studies in nonhuman primates [
62] show that levels of fetal testosterone later in pregnancy still independently influence sex-associated behaviors, and research in sheep [
63] has identified late pregnancy windows where fetal testosterone influences structural and organizational changes in the brain [
64]. Consequently, there may be multiple periods when the developing brain is susceptible to the influence of testosterone [
64,
65].
Another potential limitation of the use of cord blood samples to measure androgens is that this sample commonly includes a mix of arterial and venous blood. While androgens in the umbilical artery primarily reflect circulating fetal androgens from fetal adrenal and gonad, those in the umbilical vein also reflect maternal and placental androgens [
66]. However, a recent study comparing umbilical artery versus umbilical vein androgen levels found that while there are absolute differences between them, they are highly correlated (
r = 0.67 ~ 0.83) [
67]. Further, a recent population-based study of children of mothers with polycystic ovarian syndrome, which has hyperandrogenism as a defining characteristic, showed an elevated risk of autism in the child, suggesting that maternal prenatal androgen may also play a role in ASD risk [
68].
In addition to androgen measurement issues, outcome measurement also deserves further reflection. Our findings were based on quantitative measures of ASD-related phenotype at 12 and 36 months, not categorical diagnostic classifications. However, twin studies have shown that the estimated magnitudes of genetic and non-shared environmental etiologic influences are quite similar regardless of whether outcome is a dichotomous or continuous ASD-related phenotype measure [
69,
70]. This supports the idea that continuous and dichotomous ASD outcome measures are caused by common mechanisms.
We also must be mindful that the sample size here was modest, limiting the precision of effect estimates. In addition, the sample included only 22 subjects who had female older affected siblings and, therefore, the interesting results pertaining to this group, though statistically significant, are quite imprecise and this imprecision increases the plausibility that these results could be a chance finding. Lastly, as with most observational analyses, we cannot rule out the possibility that our findings may be influenced by residual confounding despite adjusting for known factors associated with exposure and outcome.
Acknowledgements
JAK was supported by the Women and Infants Research Foundation of WA.
We would also like to thank Dr. Lucy Robinson, Drexel University School of Public Health, for providing the code for the multiple imputation analyses.