Exposure to increased levels of prenatal testosterone has received increasing attention as a possible biological mechanism contributing to ASD. However, ASD is first recognized through behavioral symptoms during early childhood, and, therefore, data on the prenatal hormone environment are rare. This is the first report of the relationship between testosterone concentrations in umbilical cord blood and the ASD phenotype. Five children in the current sample had received a clinician-based diagnosis of ASD. Four of the five cases had TT and BioT levels lower than the sex-specific BioT means of the broader cohort, and all cases were within one standard deviation of these means. A measure of autistic-like traits, the AQ, was collected from 374 members of the cohort without ASD. Males scored higher than females on the three scales, but none of these were significantly correlated with early testosterone measurements (in males or females). Males were more likely than females to have ‘high’ scores on the Details and Patterns subscale of the AQ (but no other subscale). However, again, this scale was found to be unrelated to early testosterone concentrations. While the small number of individuals with a clinical diagnosis of ASD evokes caution in the interpretation of these data, we suggest that the current findings indicate that testosterone concentrations measured at the time of birth are not associated with autistic-like traits in the general population.
The Cambridge Fetal Testosterone Project has provided the most direct evidence linking prenatal testosterone exposure and ASD, reporting associations between testosterone levels in amniotic fluid and a range of autistic-like traits during early [21–23] and middle  childhood. Amniotic fluid samples from the Cambridge cohort were collected via amniocentesis during the second trimester of pregnancy (18 to 20 weeks gestation). One possible explanation for the null findings observed in the current study relates to the timing of testosterone exposure. Hormone concentrations are known to fluctuate throughout pregnancy , with considerable variability between measurements taken in the second and third trimesters . In the current study, DNA sequencing confirmed that the cord blood samples were free from maternal blood contamination, and the higher concentrations of BioT in male compared to female offspring suggest that these samples reflect fetal circulation during late gestation. The lack of association between cord blood concentrations of BioT and autistic-like traits suggests that either prenatal testosterone does not influence these behaviors, or that any effects are at an earlier gestation.
It is also possible that the effects of prenatal testosterone exposure are determined not only by individual differences in concentration, but also by individual differences in biological sensitivity to testosterone. The neuroactive effects of testosterone occur either through the activation of the X chromosome linked androgen receptor AR gene, located at Xq11-12, or after aromatization to estradiol, through the estrogen receptor . The cysteine, adenine, guanine (CAG) repeat sequence within exon 1 of the androgen receptor gene, is of particular interest because it is highly polymorphic. This CAG repeat codes for a polyglutamine tract of variable length in the N-terminal domain of the protein, and the number of repeats is inversely related to the transcriptional activity of androgen target genes [42, 43]. It remains possible that testosterone levels during late gestation are associated with the ASD phenotype, but these effects are modified by individual genotype. However, it is important to note that studies of testosterone concentrations in both amniotic fluid [21–24] and cord-blood  have reported associations with postnatal behavior when naïve to individual genotype. We suggest that future studies in this area may benefit from an understanding of the genetic as well as endocrine background of an individual.
A further explanation for the null findings relates to the pattern of AQ scores observed in the current sample. The primary hypothesis of the current study – that perinatal testosterone concentrations are associated with autistic-like traits - was formulated based on widely observed sex-differences in ASD . However, for the majority of AQ scales scored according to the original procedure , there were no differences in scores between males and females. This finding contrasts with studies from the Cambridge Fetal Testosterone Project, which have observed sex-differences in scores on other measures of autistic-like traits, including the Quantitate CHecklist for Autism in Toddlers among 18 to 24 month-old toddlers , and the Child Autism Spectrum Test and Autism-Spectrum Quotient-Child Version  among children between 6 and 10 years of age. It is possible that sample attrition may have contributed to the lack of sex differences in AQ scores, given that AQ scores were typically higher for socially disadvantaged participants, and that these individuals were less likely to take part in the current study. Importantly, however, sex differences were observed when an alternative scoring procedure of the AQ was applied , and these scales showed no association with early testosterone exposure in either sex.
Strengths of the current study include the relatively large sample size, the prospective longitudinal design spanning over two decades, and the highly sensitive and selective LC-MS/MS assay used to measure testosterone levels in umbilical cord blood. One potential concern is that degradation may have occurred in storage or with thawing, resulting in reduced concentrations of intact steroids. However, we believe such concerns are unfounded for several reasons. First, our sample set had been continually maintained at −80°C since collection and samples were thawed only once for aliquotting prior to shipping for assay. Studies of steroid stability (including testosterone) during long term storage confirm that serum samples are able to be stored for at least four decades at −80°C without loss or appreciable deterioration of steroid hormones [44, 45]. Second, freshly collected cord blood samples run as quality controls (n = 5 to 6) had values within one standard deviation of the means of the frozen samples, adding further weight to the view that no significant degradation had taken place; our stability studies showed no effects of several freeze-thaw cycles. Finally, our recovery estimates based on spiking of fresh cord serum were 93% to 111% indicating that the assays did not suffer from masking or interference effects.
A limitation of the study design was that the ascertainment of the five ASD cases in this study was dependent upon parent-report of a clinician-based diagnosis (by a pediatrician, psychologist and speech pathologist) according to DSM-IV guidelines at any of the 5-, 8-, 10-, 14- or 17-year follow-ups. While it is commonplace in ASD research to confirm clinical diagnoses using ASD-specific behavioral observation and/or parent interview assessments, this was not possible within the Raine cohort. However, it is important to note that a previous investigation of direct observation data obtained prior to five years of age , found that each ASD case in the current study demonstrated behaviors consistent with ASD (for example, poor eye contact, delayed language, absence of pretend play). Moreover, the overall rate of ASD within the current sample was 0.71%, which is highly similar to the most recent population-based prevalence estimates in Australia (0.625%) , and suggests that there was no systematic bias introduced to the study by our reliance on clinician diagnosis.
A further limitation is that sample attrition in the current investigation, as for many longitudinal studies, appeared to bias the loss of individuals from lower socioeconomic strata. Post hoc analyses found that AQ scores varied according to maternal income and education, which raises the concern that the attrition may have underestimated any effect of maternal umbilical cord blood testosterone concentrations on AQ scores. However, published studies reporting significant associations between amniotic fluid testosterone levels and postnatal behavior [23, 24], have included selected samples (that is, high-risk pregnancies undergoing amniocentesis) and experienced similar attrition effects. Furthermore, computer simulations using data from the Avon Longitudinal Study of Parents and Children (United Kingdom) have found that selective dropout in cohort studies only marginally affect regression coefficients, if participant selection occurs according to predictor variable(s) . For these reasons, we suggest that the sample attrition in the current study had minimal, if any, influence on the null finings observed.