Recent studies have discovered a number of associations between prenatal environmental exposures and children’s cognitive, behavioral, and mental health problems. For example, prenatal exposure to polybrominated diphenyl ether (PBDE) was found to be associated with decreased verbal performance and full-scale IQ (Herbstman et al., 2010), decreased cognitive capacity (Eskenazi et al., 2013; Chen et al., 2014), decreased language capacity (Eskenazi et al., 2013; Ding et al., 2015), decreased visual memory (Cowell et al., 2018), and decreased efficiency of the brain’s reading circuit at age 5 (Zhang et al., 2017); prenatal exposure to chlorpyrifos (CPF) has been associated with reduced full-scale IQ at ages three and 7 (Rauh et al., 2006; Rauh et al., 2011); prenatal polycyclic aromatic hydrocarbon (PAH) exposure has been linked with lower full-scale IQ, perceptual reasoning and working memory scores at age 7 (Vishnevetsky et al., 2015), and also linked with children’s lower mental development index measured by Child Behavior Checklist (CBCL) at age 3 (Perera et al., 2006), and attention-deficit/hyperactivity disorder (ADHD) behavior problems at age 9 (Perera et al., 2018), as well as alterations in the development of self-regulation capacity and social problems at age 11 (Margolis et al., 2016) and problems with inhibitory control in childhood that mediate problems with academic skills in adolescence (Margolis et al., 2021). Moreover, prenatal Bisphenol A (BPA) concentration was found to be associated with CBCL anxious/depressed and aggressive behavior (Perera et al., 2012), and CBCL internalizing and externalizing problems (Roen et al., 2015).

Environmental exposures have also been associated with epigenetic alterations including DNA methylation. Specifically, alterations in DNA methylation have been associated with exposure to BPA (Wolstenholme et al., 2011), PAH exposure and increased PAH–DNA adducts (Perera et al., 2011; Herbstman et al., 2012), exposure to phthalates (Kang and Lee, 2005), as well as exposure to high level of nitrogen dioxide (NO₂) and fine particulate matter (PM_2.5) (Prunicki et al., 2018). We recently developed and validated a pipeline method that predicts dichotomous high/low level of exposures such as NO₂, PM_2.5 and PAH using DNA methylation patterns in umbilical cord blood (Wang et al., 2021), which further demonstrated associations between DNA methylation and environmental exposures.

While previous research supports the hypotheses that prenatal exposures affect cord blood DNA methylation and also subsequent child health outcomes, it does not imply that DNA methylation is necessarily on the causal pathway between exposure and outcome. Few studies have explored the relationship among the three—environmental exposures, DNA methylation and children’s cognitive, behavioral, and mental health outcomes—simultaneously considering a spectrum of environmental exposures and a spectrum of developmental outcomes. Prior studies have focused on a single exposure or a single outcome. For example, a recent study showed that DNA methylation partly mediates the association between Bisphenol F (BPF) exposure and lower cognition in boys (Engdahl et al., 2021); another study showed that DNA methylation mediated the association between early-life lead (Pb) exposure and infant neurodevelopmental outcomes such as psychomotor development index and rating scale of emotional regulation (Rygiel et al., 2021). Similarly, another study considered a single outcome, body mass index (BMI), and a spectrum of exposures, with the goal of examining whether DNA methylation mediates the relationship between an array of environmental exposures and BMI (Cadiou et al., 2020). All of these studies were limited because they only consider a single exposure or a single outcome, because when considering many exposures or many outcomes and a large number of DNA methylation sites, the problem of multiple comparisons becomes the main limiting factor.

One study examined the effects of multiple exposures on a health outcome, BMI, via DNA methylation, by developing a “Meet-in-the-Middle” approach that attenuates the multiple comparisons problem by reducing DNA methylation dimensions a priori, identifying relevant genes and pathways (Cadiou et al., 2020).

In the present study, we aimed to examine the associations between a spectrum of environmental exposures and a range of children’s cognitive, behavioral, and mental health problems that might potentially be mediated through DNA methylation. We employ a similar dimension reduction approach as in (Cadiou et al., 2020) and only consider relevant biological pathways that potentially connect environmental exposures and children’s cognitive, behavioral, and mental health problems based on a priori knowledge obtained from the KEGG database (Kanehisa et al., 2000). We applied this approach in a longitudinal birth cohort from the Columbia Center for Children’s Environmental Health (CCCEH), seeking to identify methylation sites and corresponding genes that mediate the effect of environmental exposures on children’s cognitive, behavioral, and mental health problems.

Table 2 shows the numbers of pathways, genes and CpGs that are selected from KEGG database using keywords for different outcomes (from Step 1). Table 2 also shows the number of CpGs (from Step 2) that are significantly associated with each outcome in the CCCEH dataset after FDR adjustment for multiple testing. CBCL internalizing problems at age nine and CBCL social problems at age seven were found to be associated with two preselected CpGs after FDR adjustment for multiple comparisons, while no CpGs were found to be significantly associated with other outcomes. Full results of Step 2 were showed in Supplementary Figure S2; Supplementary Material SC.

For each outcome identified in Step 2 (Table 2) with at least one associated CpG site, Step 3 examines the relationship between each environmental exposure and the subset of CpGs identified in Step 2. Table 3 lists the exposures that are significantly associated with at least one CpG after FDR adjustment for multiple comparisons, representing the output of Step 3. For the two CpGs (cg27510182 and cg24713878) that were associated with CBCL social problems at age 7, only CpG cg27510182 is significantly associated with log-transformed PAH after accounting for multiple comparisons using FDR. No exposures were found to be significantly associated with these two CpGs that are associated with CBCL internalizing problems at age 9. Full results of Step 3 were included in Supplementary Material SD.

For the exposure and outcomes that show significant findings in Step 3, Step 4 evaluates the exposure-outcome relationships. The only association needs to be tested is the association between CBCL social problems at age seven and log-transformed PAH. Table 4 shows that the log-transformed PAH is positively and significantly associated with CBCL social problems at age 7. The direction is expected because higher scores of CBCL indicates more problems.

For the significant exposure-outcome relationship of Step 4, Step 5 evaluates the percentage of mediation effects in the total effect, when those potential CpGs are found to be mediators (Table 5). For the effect of log-transformed PAH on CBCL social problems at age 7, cg27510182 has a mediation effect about 46.7% of the total effect. In Table 6, we summarized our overall finding in the discovery dataset with 240 samples with cg27510182, including regression coefficients and raw p-values for each step that involve this CpG from CpG-outcome, exposure-CpG and exposure-outcome relationships. The corresponding results in the validation dataset with 101 samples are included in Supplementary Table S1, where the associations were not significant, possibly due to the small sample size, but the directions of CpG-outcome and exposure-CpG relationships were replicated as those in the discovery dataset.

We further summarized the identified CpG cg27510182 and the corresponding gene together with their related keywords and pathways from the KEGG database in Table 7.

In the CCCEH birth cohort, we have explored the associations between a spectrum of environmental exposures, DNA methylation, and a range of children’s cognitive, behavioral, and mental health problems using a “Meet-in-the-Middle” approach. As previously noted, this approach takes advantage of relevant biological pathways to initially reduce the number of DNA methylation CpG sites tested. Using sequential steps that further reduce the number of comparisons, this methodology can be used to test exposure-DNA methylation-outcome relationships where a priori information supports the biological plausibility of findings. To compare with our method, we also conducted a standard EWAS (epigenome-wide association study) for each considered outcome, without using relevant biological pathways to preselect CpGs. Although we found CpGs that are significantly associated with some outcomes (Supplementary Material SE) in EWAS after FDR adjustment, there was no significant finding from the following steps based on these significant CpGs. This emphasizes the benefit of our method.

Among all the exposure-DNA methylation-outcome relationships we examined, we found one CpG site and one gene that potentially mediates the effect of exposure to PAH on CBCL social problems at age 7. Specifically, we identified the methylation CpG cg27510182 and corresponding gene DAB1 (DAB Adaptor Protein 1), that potentially mediate the effect of exposure to PAH on CBCL social problems. Close investigation of the identified gene and its relevant pathways suggests that they are biologically relevant. The identified gene DAB1 has been reported to be associated with many neurodevelopmental and psychiatric disorders, such as schizophrenia (SCZ) and autism spectrum disorders (ASD) (Wang et al., 2014; Li et al., 2015; Chen et al., 2017; Stessman et al., 2017; Sánchez-Sánchez et al., 2018; Nawa et al., 2020), because DAB1 is involved in the Reelin signaling pathway which plays a critical role in the central nervous system such as regulating neuronal position in the developing brain (Howell et al., 1997; Rice et al., 2001; Trotter et al., 2013). In addition, the identified gene DAB1 is in the KEGG pathway of a group of progressive neurodegenerative diseases “Spinocerebellar ataxia” (hsa05017), which are usually due to the dysfunction of the cerebellum (Paulson 2009; Matilla-Dueñas et al., 2010), and it has been reported that cerebellar damage is associated with an increased risk of the ASD (Becker et al., 2013; D'Mello et al., 2015). Thus, prenatal exposure to PAH may associate with the malfunction of cerebellum and the Reelin signaling pathway via epigenetic processes, which then associate with children’s neurodevelopmental problems, such as the ASD. In conclusion, the Meet-in-the-Middle approach has revealed that DNA methylation may mediate the effect of prenatal exposure of PAH on neurodevelopmental problems, by affecting the potentially relevant portion of the brain and neurological pathways.

There have been many studies focusing on the relationships between environmental exposures and neurodevelopmental outcomes through DNA methylation. Some studies considered one single exposure and one single outcome (Engdahl et al., 2021); some studies considered one single exposure and multiple outcomes (Rygiel et al., 2021); while some studies considered multiple exposures and one single outcome (Cadiou et al., 2020). Our research is distinct from these existing studies in that we simultaneously consider a spectrum of exposures and a spectrum of correlated cognitive, behavioral, and mental health problems outcomes. To overcome the multiple comparison problem in our work with many exposures and many outcomes as well as high dimensional DNA methylation, we identified a subset of genes in relevant biological pathways with neurodevelopmental outcomes through keywords search.

The step of keywords search is also a limitation of our study, as different keywords will identify different subsets of genes, which may lead to different methylation mediation effects. A more rigorous and more specific method to choose the keywords is necessary. Another limitation is that the identified biological pathways and genes are obtained from a single database, KEGG, which may only contain limited knowledge from existing literature. Besides, the relatively small sample size and lack of adjusting other potential confounders are also limitations of our study. In this study, we randomly selected 70% of our samples as the discovery set and the other 30% as the validation set, which further reduces the sample sizes in each set. We acknowledge the randomness in sample selection, which may affect the findings in both discovery set and the validation set. However, we want to emphasize that we our method is promising and can overcome small sample size problems when large prospective birth cohort studies with multiple cognitive, behavioral, and mental health outcomes of children and multiple prenatal environmental exposures, together with epigenetics are relatively hard to find.

In summary, the methodology outlined in this analysis provides a roadmap for analyses that reserve hypothesis testing to relationships along the exposure-DNA methylation-outcome pathway that have enhanced biological plausibility and thus increased the potential to provide meaningful results. In this study, we detected one CpG cg27510182 on the gene DAB1 that partially mediates the association between prenatal exposure to PAH and CBCL social problems at ages 7. The results are consistent with most recent literature, where PAH could produce a long-lasting effect on self-regulatory capacities and PAH-DNA adduct had a positive association with the CBCL social competence problem (Margolis et al., 2016). PAH exposure during pregnancy has been reported to be positively correlated with CBCL social scores and also positively correlated with the Autism Behavior Checklist (ABC) total scores, which suggest that PAH could be one of the risk factors of ASD-related behaviors for children (Liu et al., 2019). Future epidemiologic studies that can replicate/confirm these associations and mechanistic studies that can evaluate these pathways will enhance our understanding of how prenatal exposure of PAH may lead to neurodevelopmental problems in children.