Autism is a complex neurodevelopmental disorder with a largely unknown etiology. It is characterized by impaired language, disrupted reciprocal social interactions, and stereotyped behaviors and interests (
1). Both genetic and environmental factors have been associated with the disorder (
2–
4). To date, few studies have examined an association between prenatal exposure to toxins and autism, and among these, most have been based on ecologically, rather than serologically, documented exposures. For example, specific associations were reported for residential proximity to sites contaminated with pesticides (
5,
6) and traffic-related air pollution (
7). However, associations between prenatal exposure to air pollution and risk of autism spectrum disorder were not shown in a Swedish twin study (
8) or in a large collaborative European study (
9).
Persistent organic pollutants are lipophilic halogenated organic compounds and include the insecticide dichlorodiphenyltrichloroethane (DDT), as well as its metabolite p,p′-dichlorodiphenyl dichloroethylene (p,p′-DDE), and polychlorinated biphenyls (PCBs) (
10). Before the late 1970s, these compounds were in widespread use in developed countries (mainly as insecticides [i.e., DDT] and in transformers and electrical equipment [i.e., PCBs]). Although these chemicals were widely banned in many countries more than 30 years ago, they became ubiquitous in these countries, including in the United States (
11) and Finland (
12). Because of their lipophilic nature and chemical half-lives of as long as several decades (
13), these compounds persist in the food chain, particularly through fatty food sources, resulting in continuous exposure among populations. Persistent organic pollutants are transferred across the placenta, resulting in lipid-adjusted cord blood concentrations that range from 30% to 50% of levels found in maternal plasma (
14). Thus, there is ongoing prenatal exposure potential among nearly all children because of existing maternal body burdens (
11), as demonstrated in a nationally representative sample of U.S. women in 2003–2004 (
11).
Maternal exposure to persistent organic pollutants has been associated with aberrant perinatal outcomes and childhood neurocognitive outcomes. Increasing maternal levels of p,p′-DDE were associated with an elevated risk of preterm birth in a study with a large sample size (
15), and reductions in indices of psychomotor development and other cognitive functions (
16,
17) as well as delayed processing speed (
18) have been observed in exposed offspring. Furthermore, maternal PCBs have been associated with aberrant neurocognitive outcomes, although findings are inconsistent (for a review, see reference
19).
Although associations with autism-related behaviors have been reported in several studies that used interview and ecologic data on maternal exposure to persistent organic pollutants (
5,
6,
20,
21), few studies have examined biomarker-based measures of maternal exposure to these pollutants and autism spectrum disorder in offspring (
22,
23). In a small pilot study from the Finnish Prenatal Study of Autism, we previously reported that maternal levels of p,p′-DDE and PCBs were associated with autism in childhood, although the findings were not statistically significant (
22). In the Early Markers of Autism study, Lyall et al. (
23) observed increased mean levels of several PCB congeners, including PCB 138/158 and PCB 153, in mothers of children with autism spectrum disorder.
We therefore hypothesized that maternal p,p′-DDE levels and total PCB levels, each in the highest quartile of the distribution, would be related to risk of autism among offspring. Supplementary analyses were conducted to investigate whether offspring sex and comorbid intellectual disability modified the relationship between maternal exposure to persistent organic pollutants and autism.
Method
Study Population
The study is derived from a large national population-based birth cohort. The Finnish Prenatal Study of Autism is based on a nested case-control design. The sampling frame was defined such that all members of the national birth cohort were within the age of risk of autism. Toward this end, the study subjects comprised all offspring born in Finland from 1987 to 2005, and they were followed up until 2007. The study methods are described in further detail by Lampi et al. (
24).
Description of the Birth Cohort, Biobank, and National Registries
All offspring in the Finnish Prenatal Study of Autism were derived from the Finnish Maternity Cohort, which consists of more than 1 million pregnancies with archived prenatal serum specimens drawn since 1983 (
24). Sera were obtained during the first trimester and early second trimester (months 2–4 of pregnancy) from more than 98% of pregnant women in Finland. One maternal serum sample was acquired for each pregnancy. After the screening, serum samples were stored as one aliquot at –25°C in a single biorepository at the National Institute for Health and Welfare in Oulu, Finland. All samples in the Finnish Maternity Cohort can be linked to offspring by a unique personal identification number, assigned to all residents of Finland since 1971.
Identification of Case and Comparison Subjects
The Finnish Hospital and Outpatient Discharge Register was used to identify all recorded diagnoses from psychiatric hospital admissions and outpatient visits for childhood autism (ICD-10 code F84.0) among individuals registered with the Finnish Maternity Cohort. Registry diagnoses of childhood autism were validated with the Autism Diagnostic Interview–Revised (
24). Computerized data are available from January 1, 1987, to the present. Only singleton births were included. Cases diagnosed over the sampling frame were identified from registry linkages between the Finnish Maternity Cohort and the Finnish Hospital and Outpatient Discharge Register from January 1, 1987, to December 31, 2007. The total number of childhood autism cases in the Finnish Prenatal Study of Autism was 1,132.
Case subjects were matched 1:1 to comparison subjects (singleton births only) on date of birth, sex, birthplace, and residence in Finland. Comparison subjects were drawn from the birth cohort and were without a diagnosis of autism spectrum disorder (no ICD-10 code F84.0 diagnosis). The analytic sample comprised 778 case subjects (from the 1,132 total cases of autism spectrum disorder mentioned above) and 778 matched control subjects.
Laboratory Analyses
All assays were performed blind to case-control status. Matched case and control subjects were analyzed in the same run to minimize variation between runs. In the analysis of persistent organic pollutants, we tested the following two primary hypotheses: autism in offspring would be associated with increased maternal concentration of p,p′-DDE and autism in offspring would be associated with increased maternal concentration of total PCBs. The analytical method used is described in detail elsewhere (
25). Briefly, ethanol and [
13C]-labeled internal standards of each persistent organic pollutant compound were added to the serum samples. Dichloromethane-hexane was added for extraction of persistent organic pollutants, followed by activated silica to bind the sample water, ethanol, and protein precipitate. The upper dichloromethane-hexane layer was poured into a multilayer silica column for the removal of coextracted compounds that interfere with the gas chromatography-mass spectrometry quantitation. Persistent organic pollutants were eluted from the cleanup column with additional dichloromethane-hexane. The recovery standard [
13C]-PCB-128 was added, and the eluate was concentrated for gas chromatography with tandem mass spectrometry analysis for quantification of persistent organic pollutants.
For each batch of samples, a control serum sample from the National Institute of Standards and Technology, Standard Reference Material 1958, was included. Recoveries of p,p′-DDE and PCB levels varied from 86% to 106% (coefficient of variation, 1.6%−6.5%) of the certified concentrations for Standard Reference Material 1589 and from 83% to 101% (coefficient of variation, 2.0%−6.6%) of the calculated concentrations for diluted Standard Reference Material 1589, respectively. The limits of quantification were 5 pg/mL for each PCB congener and 40 pg/mL for p,p′-DDE. Fresh weight serum concentrations of persistent organic pollutants, which demonstrated high correlation with lipid-based concentrations in a previous study (overall, r=0.95) (
26), are reported here.
Classification of Persistent Organic Pollutant Variables
In order to limit the number of analyses of the compounds, we focused on two hypothesized primary measures of maternal exposure to persistent organic pollutants: maternal p,p′-DDE levels in the highest 75th percentile of the distribution and maternal total PCBs, quantified as the sum of concentrations of the 10 measured congeners (PCB 74, PCB 99, PCB 118, PCB 138, PCB 153, PCB 156, PCB 170, PCB 180, PCB 183, and PCB 187), in the highest 75th percentile of the distribution. These PCBs were selected because they represent approximately 85%−90% of all PCBs on a mass basis.
The study was approved by the ethical committees of the hospital district of Southwest Finland and the Finland National Institute for Health and Welfare as well as the institutional review board of the New York State Psychiatric Institute. At the time all maternal serum specimens were obtained, mothers provided informed consent after receiving a description of the nature and possible consequences of the procedure and the data derived from serum analyses.
Statistical Analysis
We computed descriptive statistics and correlations between levels of persistent organic pollutants. These were analyzed separately for case and control subjects. Potential confounders were selected on the basis of previous relationships with exposure to persistent organic pollutants or autism from other studies (
27) and compared between the two study groups by using chi-square and t tests. Potential confounders were maternal age, number of previous births (0 or ≥1), socioeconomic status (upper white collar, lower white collar, blue collar, or other), maternal and parental history of psychiatric disorders, and gestational week of the blood draw (
Table 1). Data on maternal age, maternal socioeconomic status, and previous births were acquired from the Finnish Medical Birth Register. Data on maternal and paternal history of psychiatric disorders were acquired from the Finnish Hospital and Outpatient Discharge Register. Data on paternal age were obtained from the Finnish Population Register. Data on gestational week of the blood draw were obtained from the Finnish Maternity Cohort.
Appropriate to the case-control study design, point and interval estimates of odds ratios for the association of maternal levels of p,p′-DDE and total PCBs with autism were obtained by fitting conditional logistic regression models for matched sets. Statistical significance was set at a p value <0.05. Covariates were included in the adjusted models on the basis of associations with the outcome. We did not match on these covariates because of the disadvantages of overmatching (
28) and because they could be controlled effectively in the multivariable analyses.
For the primary analyses of persistent organic pollutants (p,p′-DDE and total PCBs), exposures were analyzed as dichotomous variables, with cutoff points at the 75th percentile. Exploratory analyses were conducted after stratification by sex and intellectual disability of the case subjects, given well-known sex differences in autism (
29) as well as extensive evidence of comorbid intellectual disability (
30) (ICD-9 codes for intellectual disability: F317, F318.0, F318.1, F318.2, and F319; ICD-10 codes for intellectual disability: F70, F71, F72, F73, F78, and F79). Moreover, previous studies have indicated that some risk factors may be distinct for autism with intellectual disability (
31) relative to autism without intellectual disability (
32), including our previous finding that accelerated growth velocity of head circumference at 3 months of age was associated with autism with intellectual disability but not autism without intellectual disability (
31). We examined effect modification of p,p′-DDE levels and total levels of PCBs by adding product terms to models for each variable by p,p′-DDE or by PCB levels higher than the 75th percentile. The evidence for heterogeneity of the odds ratios between strata for each potential effect modifier was assessed on the basis of the p values for the product terms. In order to evaluate whether maternal levels of PCB 138/158 and PCB 153 were associated with autism, we conducted supplementary analyses of these maternal PCBs and autism.
Statistical analyses were performed with SAS 9.4 (SAS Institute, Cary, N.C.). Bonferroni correction was not performed given that only two primary hypothesized variables were tested (maternal p,p′-DDE levels higher than the 75th percentile and maternal total PCB levels higher than the 75th percentile), as mentioned above.
Discussion
This large national birth cohort study of maternal levels of persistent organic pollutants and autism among offspring produced two principal findings. First, maternal levels of p,p′-DDE were significantly increased in mothers of case subjects with autism compared with mothers of control subjects without autism. The findings persisted after adjustment for covariates related to autism. To our knowledge, this is the first biomarker-based evidence of this association. Second, maternal levels of total PCBs, and other PCBs variously defined, were unrelated to autism risk, and thus we did not replicate the associations between maternal levels of PCBs and autism in our previous pilot study from the Finnish Prenatal Study of Autism birth cohort, despite the similar source population and method. However, the sample size in the pilot study was small, and none of the findings reached statistical significance (
22).
We propose two reasons for the observation that maternal exposure to p,p′-DDE was related to autism while maternal exposure to PCB was not. First, maternal exposure to DDT and DDE is associated with both premature birth and small gestational age status. Exposures to both of these compounds have been well replicated as risk factors for autism spectrum disorder (
15,
33). In contrast, maternal PCB exposure has not been related to prematurity or small gestational age status (
34). Second, p,p′-DDE inhibits androgen receptor binding, androgen-induced transcriptional activity, and androgen action, including in developing rats (
35). Offspring of rats injected with valproic acid, an in utero risk factor for autism (
36), exhibited reduced androgen receptor expression in most cerebellar lobules, in both male and female offspring (
37); cerebellar abnormalities, including Purkinje cell numbers, have been observed in the brains of individuals with autism (
38) and in rat offspring exposed prenatally to valproic acid (
39). In contrast, PCBs increase androgen receptor transcription (
40).
One possible reason for inconsistent findings of an association between maternal exposure to persistent organic pollutants and autism in offspring across studies is differences in the chemical mixtures and contexts of exposure between populations. A study that examined PCB 153 and p,p′-DDE levels in adults from four different geographic populations found that the correlation between the levels of these two persistent organic pollutants varied considerably, as did the associated covariates, likely reflecting differences in primary routes of exposure (
41). Animal studies have demonstrated interactive effects on behavior and learning between different neurotoxicant chemicals, including PCBs and methylmercury (
42). Therefore, it is possible that differences in these findings for p,p′-DDE and PCBs in studies of persistent organic pollutants and autism were due to differences between populations in the exposure profiles for other chemicals that interact with these pollutants. Moreover, Schmidt et al. (
43) observed that folic acid intake during pregnancy attenuated the relationship between maternal insecticide exposure (determined on the basis of interviews and ecologically defined exposures to pesticides); conceivably, substances that may protect against developmental pathology from insecticides and other persistent organic pollutants differ between populations (
43).
Our findings are not in agreement with those of Lyall et al. (
23), who demonstrated that maternal levels of PCB 138/158 and 153 in the highest 75th percentile of the distribution were significantly associated with offspring with autism spectrum disorder. In another study of maternal levels of persistent organic pollutants associated with autism in offspring, based on a different cohort, a relatively small sample was utilized, and autistic behaviors rather than clinical diagnoses of autism spectrum disorder were included, and therefore the results may not be comparable with the results of the present study (
44). The authors of that study found that maternal PCB 178 levels were associated with fewer autistic behaviors among offspring, and maternal levels of p,p′-DDE and dichlorodiphenyl trichloroethane did not show associations with the outcome.
In our study, the association between maternal levels of p,p′-DDE and autism in offspring was isolated to offspring with comorbid intellectual disability. This may suggest that the relationship between maternal exposure to p,p′-DDE and autism is related to intellectual disability in general and not to autism specifically. Previous studies have shown associations between maternal levels of p,p′-DDE and cognitive dysfunction, including reduced psychomotor development (
16), general cognitive function, verbal and memory ability (
17), and processing speed and verbal comprehension (
18) among offspring; however, other studies have not shown these associations. For example, transplacental exposure to p,p′-DDE was associated with higher scores on the Bayley Scales of Infant Development at 6 months; the relationship disappeared at 12 months (
45); and no associations were observed for maternal p,p′-DDE concentrations and scores on the Bayley Scales of Infant Development at age 8 months and on IQ at age 7 (
46). If maternal exposure to DDT or p,p′-DDE has no effect on childhood neurocognition in general population samples, this may suggest that this exposure is related to a subgroup of autism cases characterized by comorbid intellectual disability, rather than to intellectual disability itself.
Although the association between maternal p,p′-DDE levels and autism was significant among male but not female offspring, the estimates of association did not differ significantly between males and females. It is possible that the lower number of female subjects hindered our ability to detect differences, if any were present, between the sexes.
Strengths and Limitations
Strengths of this study include a larger sample size than in some previous studies (
22), high detection rates, and a national population-based sample. Our study had several limitations as well. First, we did not examine a comparison group of individuals with intellectual disability but without autism; hence, we cannot rule out the possibility that our finding of an association between maternal p,p′-DDE exposure and autism in offspring was accounted for by intellectual disability. However, in addition to the preceding discussion on maternal levels of p,p′-DDE and neurocognition in offspring, we note that Lyall et al. (
23) reported similar associations in subgroup analyses of autism with and without comorbid intellectual disability. In their analysis of intellectual disability without autism spectrum disorder, they found a numerically increased risk for intellectual disability among offspring in the second and fourth quartiles of maternal levels of PCB 138/158 and in the third quartile of maternal levels of p,p′-DDE, although these associations were not statistically significant. Second, although the majority of mothers of case subjects in our birth cohort had serum samples tested, a significant proportion were not included. The included case subjects were born after 1993 (p<0.0001) and were more likely to be positive for a maternal (p=0.01) or parental (p=0.04) history of psychiatric disorders than those who were not included (for further details, see Table S2 in the
online supplement). However, this should not have biased our results given that we accounted for these characteristics in the design and analyses. Third, we presented fresh weight serum concentrations of persistent organic pollutants, rather than lipid-adjusted concentrations. However, the unadjusted measures were found to have a low degree of bias under a range of causal scenarios (
47) and to be highly correlated with lipid-adjusted concentrations (r=0.95) (
26). Nonetheless, we cannot entirely rule out bias due to uncontrolled confounding by serum lipids. Fourth, we did not adjust for multiple comparisons; if we had, the association of maternal levels of p,p′-DDE with autism would have narrowly missed the Bonferroni-corrected traditional threshold for statistical significance (alpha=0.025), given that two persistent organic pollutants were tested. However, the Bonferroni method focuses on situations in which multiple statistical tests are conducted without a priori hypotheses or when testing for whether all null hypotheses are true simultaneously (
48). These situations did not apply to our primary statistical tests, which were restricted to a priori hypotheses, and these were evaluated separately. Finally, although potential confounders were adjusted in the analyses, there is always the possibility, as in any observational study, of residual confounding. However, given the selectivity of our finding for maternal levels of p,p′-DDE, this does not appear to be likely, unless the potential for residual confounding is greater for p,p′-DDE compared with PCBs.