A profile and review of findings from the Early Markers for Autism study: unique contributions from a population-based case–control study in California

Autism spectrum disorder (ASD) is a complex neurodevelopmental condition characterized by deficits in social communication and the presence of restricted behaviors or interests [1]. Decades of research have firmly established the importance of genetics in the etiology of ASD [2, 3], although that role is a complex one that is still not fully understood. There is also evidence of substantial involvement of environmental factors in ASD etiology, particularly those acting during the prenatal and early neonatal periods as critical windows of neurodevelopment and susceptibility to environmental influence. Though the pathways to ASD are not fully understood, a growing body of evidence supports immune system dysregulation in risk of ASD [4‐7]. Furthermore, fetal exposure to chemicals may also impact neurodevelopment, perhaps through immune pathways [8‐10]. However, the mechanisms by which environmental agents interact with the developing immune and nervous systems to cause neurodevelopmental disorders in genetically susceptible individuals are poorly understood. Motivated by these gaps in our knowledge, and the desire to learn more about early biologic processes involved in ASD, the Early Markers for Autism (EMA) study was created in 2004. This review describes the motivation for the EMA study, the methods of this novel population-based case control project, and summarizes major findings published (or accepted) as of October 2020 that serve as unique contributions to the field of ASD epidemiology.

In the early part of the decade from 2000 to 2010, few studies had directly investigated the role of prenatal immune profiles (e.g., cytokine and chemokine profiles, antibodies to fetal brain proteins, antibodies to specific infectious agents) on neurodevelopmental outcomes in humans [11, 12]. However, findings of immune conditions in the family members of affected individuals, and aberrations in immune markers in both cases and mothers were motivating mechanistic work in this area [13‐26]. Also during this time frame, no ASD studies had been conducted measuring biomarker levels of environmental contaminants in samples collected during the prenatal or early neonatal susceptibility periods. Research in related fields suggested developmental concerns for a number of environmental chemicals, including polybrominated diphenyl ethers (PBDEs), polychlorinated biphenyls (PCBs), and certain pesticides [27], but work examining these exposures using prenatal biomarkers for associations with neurodevelopmental outcomes more specifically was truly in its infancy. There was also a gap in knowledge regarding the potential mechanistic pathways of such exposures. When created in 2004, EMA was the first large, population-based, nested case–control study of ASD to utilize archived prenatal and newborn specimens mother-baby pairs. Key features of the EMA study are provided in Table 1.

The EMA Study sought to advance our knowledge of the effects of immunologic, genetic, and environmental factors, individually and in combination, on ASD etiology, utilizing biomarkers of exposures and examining immune and endocrine pathways. With a large, essentially population-based sample of children with confirmed ASD, children with other developmental disabilities, as well as unaffected children, we had an unprecedented opportunity to explore the interplay between genetic susceptibility and markers of immune function and environmental exposures that contribute to ASD risk. Furthermore, because banked biospecimens were being used, the design enabled us to prospectively investigate measured levels of these factors during critical periods of fetal brain development in a novel way. We focused on exposures and markers that were known or suspected to play a role in early brain development. These included antibodies to viral infections, cytokines, chemokines, antibodies to fetal brain proteins, neurotrophins, hormones, metals, and endocrine disrupting chemicals (EDCs) that could be measured in these archived maternal blood (sera and cell pellets) collected in mid-pregnancy and neonatal blood spots collected in the first few days of life. We anticipated that findings from this study would define areas for further investigations of physiologic mechanism, provide a better understanding of the underlying biology involved in ASD etiology, and provide normative data on unaffected controls that may be useful as benchmarks for future studies utilizing prenatal and neonatal specimens.

The EMA study framework is depicted in Fig. 1. The EMA study population was drawn from a 2000–2003 cohort of births in Southern California in which mother’s prenatal screening specimens were archived as part of Project Baby’s Breath (M Kharrazi and GN DeLorenze, Co-PIs) [28]. Women who participated in the prenatal expanded alphafetoprotein screening program (XAFP), [29] offered to all pregnant women in California in the second trimester of pregnancy, during 1999–2002 in Orange, Imperial, and San Diego Counties were eligible for participation in Project Baby’s Breath; in addition to this participation in screening, for inclusion in EMA, women who had a live born infant for whom a California birth certificate match could be found (approximately 90% of screening participants), and for whom both a maternal prenatal screening blood specimen and a newborn screening blood specimen were available in the State archives, were eligible. EMA was conducted in two phases; the first phase was drawn from women who were pregnant in Orange County, California, and who delivered a live born infant from July 2000 to September 2001. The second phase was drawn from women who were pregnant in Orange, San Diego, or Imperial Counties in Southern California and delivered a live born infant from January 2000 to June 2003, after excluding those in the first phase, thereby serving to expand the sample size, geographic region, and birth year range of the first phase. This geographic region included urban, suburban and rural areas in the southern part of California closest to the Mexico border with diverse geographies and race/ethnicities. Study activities were approved by the Institutional Review Board at Kaiser Permanente and the State Committee for the Protection of Human Subjects. Informed consent was not required for the EMA study, given that consent forms for screening programs were distributed at the time of collection, which stipulated that specimens and results from prenatal and neonatal testing could be used for research purposes given IRB approval.

Representativeness: Approximately 70% of pregnant women participated in the XAFP in CA during this time period [29, 30]. Screened women were more likely to be under age 35, but were similar to the total population of women delivering in study counties for these birth years with respect to maternal education, race/ethnicity, parity, and health insurance status (Additional file 1: Table S1). Basic characteristics of the EMA study population, by phase and study group, are provided in Table 2.

Following a protocol initially developed by the CDC, trained medical record abstractors reviewed and abstracted detailed diagnostic and clinical data from RC records for all children identified as ASD and ID for whom archived blood specimens were found. Final case status was determined by expert clinical review of the abstracted information, according to DSM-IV-TR criteria. The overall EMA participant selection process is shown in Fig. 1. Most of those ascertained as ASD cases were confirmed (72% for phase 1 and 98% for phase 2), while 22% (phase 1) and 38% (phase 2) of those originally identified as ID were reclassified as ASD cases according to expert clinician review of records. After this expert review, final analytic samples consisted of 84 children with ASD and 49 with ID for phase 1 and 545 children with ASD and 181 children with ID for phase 2. Certain analyses across phases also considered ASD subtypes, including early onset or regressive (phase 1), and presence or absence of co-occurring ID (phase 1 and 2). Determination of ID (and co-occurring ID within the presence of ASD) was based on composite scores < 70 on all standardized cognitive and functional tests recorded in RC records (with no ID defined according to all or some scores > 70, and unknown ID defined according to the absence of standardized scores in chart).

Data were obtained through California Vital Statistics, the GDSP, and the DDS. Parental demographics, including parental age, dates of birth, education level, race, ethnicity, country of origin, and insurance status at delivery were obtained through birth certificates. Additional maternal information, including weight, height, and pregnancy-related characteristics were obtained at time of prenatal sample collection. Infant death and neonatal information, including birthweight and gestational age, was obtained through Vital Statistics. While covariates included in adjusted models varied across studies, study matching factors, maternal age, weight at prenatal specimen collection, and race/ethnicity were typically adjusted for in analyses; additional adjustment factors were considered as appropriate for analyses (see also Table 4).

Maternal mid-pregnancy (range 15–20 weeks gestation) blood specimens were obtained from Project Baby’s Breath and dried newborn bloodspots were obtained from the CA Genetic Disease Screening program (GDSP) of the California Biobank Program (https://​www.​cdph.​ca.​gov/​Programs/​CFH/​DGDS/​Pages/​cbp/​default.​aspx). All biospecimens were retrieved from -20 C freezer archives, barcoded with a unique study identification number, and shipped to the participating laboratories. For all subjects, one whole dried bloodspot, 1–2 ml (mean 1.8 ml, SD 0.3 ml) of maternal serum, and 1-2 ml of maternal blood cell pellet were retrieved.

Biomarkers measured in specimens from EMA participants are shown in Table 3 (see also Additional file 1: Appendix). Laboratory procedures for analytes have been previously published. We refer the reader to individual papers for more detailed information on methods for a given biomarker; those examined in multiple EMA studies are briefly described here. Cytokines and chemokines were measured in maternal serum and newborn samples using commercially available multiplex bead-based kits, with assays carried out following manufacturer protocols [31, 32]; TSH was measured in regional laboratories as part of routine screening in maternal serum using an I-labeled immunoradio-metric assay and newborn bloodspots using solid-phase, time-resolved sandwich fluoroimmunoassay [33, 34]. Vitamin D was measured as total 25-hydroxyvitamin D (sum of 25(OH)D2 and 25(OH)D3, hereinafter referred to as 25(OH)D) by a sensitive isotope dilution liquid chromatography-tandem mass spectrometry method (LC/MS/MS) in both maternal serum and dried newborn blood spots (converted to serum equivalents) [35‐37]. Endocrine disrupting chemicals (PCBs, OCPs, PBDEs, and PFASs) were measured in maternal serum samples using gas chromatography isotope dilution high resolution mass spectrometry, [38‐40].

Air pollution in EMA was assigned as averaged monthly exposure for the month prior to maternal serum collection based on the Environmental Protection Agency’s Air Quality System data (http://​www.​epa.​gov/​ttn/​airs/​airsaqs) for nitrogen dioxide (NO₂), ozone, and particulate matter less than 10 and 2.5 microns in diameter (PM₁₀, PM_2.5) using the maternal residential address reported during the prenatal screening visit.

DNA was extracted from a subset of 790 maternal cell pellets and 764 neonatal bloodspots and 675,000 SNPs were genotyped across the genome using standard protocols and quality assessment, as previously described [41]. The genotyped samples included 366 ASD mother–child pairs and 369 control pairs [42, 43]. The ID cases and ASD cases identified through reclassification of those originally enrolled as ID were not included in the genome-wide genotyping due to budget limitations, but were used for analyzing specific candidate genetic factors [44].

Given that EMA is comprised of mother–child pairs of cases and unrelated controls, we were able to take advantage of some features of family-based analyses. We considered models where the maternal genetic make-up could act as a determinant of the fetal environment, impacting the early risk for ASD. We tested for maternal main effects influencing risk for ASD, as well as maternal-child genetic combinations or interactions that might impact risk [41]. With genotyping data, we also estimated copy number variants (CNVs) and tested for maternal effects that might be mediated through CNVs [45]. Combining EMA genetic and biomarker data we conducted G × E analyses. Previous studies using genome-wide association study (GWAS) data to identify G × E interactions encountered statistical power issues compared with the detection of main effects [46, 47]. We hypothesized that the most likely genetic candidates for G × E would be the variants associated with the E biomarker itself. Our strategy was thus to perform a quantitative trait locus (QTL) analysis to identify candidate genetic factors that contribute to the inter-individual variation of the E biomarker and then test for interaction with case–control status among those top candidates [42, 43]. Given the variety of genetic ancestries in the EMA study population, methods appropriate for mixed-ancestry data were used to avoid ancestry-related false positives, such as adjustment for genome-wide principal components [48]. In addition, because mothers and their affected children are genetically correlated with each other, we performed separate GWAS to analyze distinct maternal and neonatal effects. When we observed both maternal and fetal genetic contribution for the same biomarker, we identified the offspring alleles not inherited from the maternal lineage (fetal-specific alleles) and the maternal alleles not transmitted to the offspring (maternal-specific alleles) in order to distinguish whether both maternal and fetal genomes independently affected a specific biomarker [43].

We also applied the principles of Mendelian randomization to assess the possible causal relationship of a risk trait with ASD outcome. If a trait (i.e. immune biomarker or vitamin D) is causally linked to ASD, the genetic variants influencing the trait should also influence ASD risk. We thus tested the association between the mapped genetic determinants of the trait in the EMA dataset and ASD outcome [42, 43, 49].

EMA has contributed 7 studies focused on the role of immune markers in ASD, many of these among the first to examine such markers in prospectively collected prenatal and neonatal biospecimens. Summaries of adjusted associations between ASD and immune markers in maternal and neonatal specimens are provided in Fig. 2a, b, respectively. Immune markers measured in EMA, by phase, are shown in these figures; those measured in phase 2 (which expanded phase 1 measures) are also listed in the Appendix. Results generally supported associations in the same direction across Phases (with the exception of neonatal eotaxin and IL-10), although the samller sample size in Phase 1 yielded less precise estimates. Across these studies, there was a general pattern of moderate increases in ASD risk with inflammatory cytokines; additional patterns are highlighted further below.

One of EMA’s first publications demonstrated a relationship between maternal antibodies to fetal brain proteins measured in mid-pregnancy samples and ASD [50], a result consistent with earlier work suggesting similar findings in maternal samples collected after the birth of the child, as well as in animal models [51‐53]. Another investigation in phase 1 suggested lower levels of the immunoglobulin IgG to Toxoplasma gondii in both maternal and neonatal samples were associated with increased risk of ASD, but no differences were noted in the ID group [54]. In addition, work conducted in the phase 2 sample suggested lower levels of C-reactive protein (CRP, acute phase protein produced in response to inflammation) in maternal samples of ASD cases, or significantly reduced risk of ASD with higher levels of CRP [49]. Together, these results could suggest an impaired response to immune activation in mothers of ASD cases.

Examination of cytokine profiles from phase 1 [31] demonstrated differences potentially suggestive of an asthma/allergy phenotype in mothers of ASD children, with elevated concentrations of IFN-γ, IL-4 and IL-5. A more inflammatory phenotype was found in mothers of ID children, characterized by elevated concentrations of IL-2, IL-4, and IL-6. This work was expanded to examine a larger panel of cytokines and chemokines levels in a greater number of maternal samples from phase 2, which demonstrated elevated maternal levels of several proinflammatory cytokines and chemokines in association with ASD with ID [55]. This differentiated the ASD with ID group from all other groups including ASD without co-occurring ID, ID without ASD, and GP controls. When neonatal samples were similarly examined, higher levels of inflammatory cytokines were found in ASD cases relative to GP controls, and other inflammatory T cell cytokines and certain chemokines were elevated in ASD relative to ID [56].

Results across these studies of immune markers therefore suggest immune dysregulation, in some cases consistent with pro-inflammatory processes, during the prenatal and neonatal periods may be involved in the development of ASD. In addition, a number of these studies suggested these relationships may differ based on the presence or absence of co-occurring ID (Additional file 1: Table S2).

Thyroid hormones: Thyroid hormones play a critical role in brain development in regulating neurogenesis, dendrite proliferation, synaptogenesis, and myelination [57‐59]. Furthermore, hypothyroidism early in life can lead to intellectual disability, and maternal thyroid conditions have been linked with ASD in prior work [60‐64]. In order to address the role of thyroid hormones in ASD, thyroid stimulating hormone (TSH) was examined in maternal samples in phase 1, and neonatal bloodspot TSH in phases 1 and 2 [33, 34]. In these studies we observed inverse relationships between ASD and maternal TSH, as well as early onset ASD specifically [34]. Relationships between ASD and neonatal TSH were not significant, though there was a suggestive inverse relationship with ASD with developmental regression in the phase 2 sample [33].

Vitamin D: Vitamin D is a steroid molecule whose primary source is sunlight. Neonatal levels depend on maternal levels, and deficiency has been linked with a range of maternal and child health outcomes [65], including neurodevelopmental outcomes such as cognitive function, language and motor development [66, 67]. Vitamin D plays a well-documented role in immune functioning, including the ability to influence growth and differentiation of innate and acquired immune cells, influence cytokine production, act as an anti-inflammatory agent, and regulate placental inflammation during pregnancy [68, 69]. Vitamin D also plays a role in protecting against oxidative stress, and can act as a neurosteroid in neurodevelopment [70], providing a basis for potential associations with ASD. Thus, for both maternal serum and neonatal vitamin D levels, we hypothesized potential protective effects with higher 25(OH)D levels, or increased risk with vitamin D deficiency [65]. However, primary analyses considering vitamin D deficiency (defined as < 50 nmol/L total 25(OH)D; 14% of newborns and 9% of mothers), insufficiency (defined as 50–74 nmol/L; 26% newborns and 26% of mothers), or according to continuous 25(OH)D levels revealed no significant associations overall at either time point [37, 71]. We also did not observe different associations with ID, or ASD with or without comorbid ID (Additional file 1: Table S2). However, for both maternal and neonatal 25(OH)D levels, there were interactions with child sex and race/ethnicity, such that males and non-Hispanic whites had lower odds of ASD with higher vitamin D levels. In addition, a non-linear, inverted j-shape pattern was observed overall, and among males, between maternal levels and ASD, with the peak around 100 nmol/L 25(OH)D. This pattern was not observed for ID. Of note, 25(OH)D levels were relatively high in EMA (median ~ 85 nmol/L), as may be expected for a study based in relatively lower latitudes (in an area with higher sunlight) [72].

EDCs: Endocrine disrupting chemicals have been hypothesized to influence risk of ASD given evidence for associations with broader adverse neurodevelopmental outcomes. Potential links between these factors are supported by evidence of EDC disruption of thyroid-hormones [73, 74], influence on immune markers, or direct influence on developing neurons [75‐78]. Three published analyses have examined classes of organohalogen endocrine disrupting chemicals in association with ASD and ID in EMA [38‐40], and one study has examined these associations considering all chemicals simultaneously [79]. Results from individual chemical analyses suggested significant associations with 2 PCB congeners, and a general trend of increases in odds of ASD with levels in the highest quartiles of most PCBs, but no association between ASD and the two organochlorine pesticides (OCP) detected (p,p-D’DE and trans-nonachlor) [40]. No sex differences or differences by ID comorbidity were noted, and the general pattern of associations with PCBs was similar when examining ID as the outcome rather than ASD, though a significant, non-monotonic association was noted between the OCP p’p’-DDE and ID. In contrast, analyses of PBDEs demonstrated null or inverse associations with ASD and ID [39]. When stratified by sex, associations with higher levels of PBDEs were below the null for boys but above for girls, suggesting potential sexual dimorphism (Additional file 1: Table S2). Analyses investigating more mechanistically-defined subgroupings of PCBs and PBDEs (such as dioxin-like or cytokine-P450 inducing PCBs, or highly brominated PBDEs), did not yield differing findings, and thus results did not directly support a specific pathway. For PFASs, associations were generally null, though inverse associations were found for the highest quartile of perfluorooctanoate (PFOA) perfluorooctane sulfonate (PFOS) and ASD [38]. Results were similar for ID and ASD with and without ID, and stratified by sex. Assessing all these chemicals together in single Bayesian model for associations with ASD, no significant associations were found [79]. ORs for the 25 included chemicals were all close to the null and attenuated when compared to analyses based on individual chemicals (Additional file 1: Fig. S1).

Air pollution: Air pollution may impact neurodevelopment via inflammatory or oxidative stress pathways, influences on methylation, or via direct effects of specific components on the developing brain [80, 81]. We had two goals in assessing air pollution in EMA: 1) to examine the relationship between maternal cytokine and chemokine levels measured during mid-pregnancy and previous-month (short term) air pollution exposure; and 2) to determine if immune markers measured mid-pregnancy mediate an association between mid-pregnancy air pollutant exposure and child ASD and ID. We found that previous month air pollution exposure and mid-gestational maternal cytokine and chemokine levels were significantly, though weakly, correlated [82]. Some differences by ASD with and without ID, and for ID only, were noted for certain immune marker and pollutant combinations, though overall conclusions were similar across groups. While mid-pregnancy air pollution was not associated with any neurodevelopmental outcome in EMA, we did find that IL-6 remained associated with ASD with ID even after adjusting for air pollution exposure. These results further support a role of maternal immune activation in risk for neurodevelopmental outcomes, and suggest that prenatal air pollution exposure may result in small effects on the immune system with biological relevance.

Other environmental toxicants: One study, conducted within the first phase of EMA, examined the association between total mercury, a known neurotoxicant, and ASD [83]. While mean mercury levels were higher in the maternal samples for ASD cases than for GP controls in the unadjusted analysis, differences did not persist after adjustment for covariates, and no case–control differences were noted in neonatal samples. Thus, mercury was not examined in phase 2.

Genetic factors are known to play a strong role in ASD, with heritability estimates ranging from 50–95%, and evidence of contributions of both rare and common genetic variants [84‐86]. Genetic studies in EMA have investigated several hypotheses. First, we applied a novel approach considering maternal genetic effects both alone and in combination with offspring genetic effects. We identified suggestive transgenerational interactions that cannot be explained by maternal or child main effects only [41]; Fig. 3a). We also examined maternal autosomal burden of large, rare copy number variants (CNVs) for transgenerational effects contributing to ASD in children, addressing the hypothesis of a maternal-specific effect (during gestation, rather than interaction between maternal and offspring genotype). We observed that the mothers of children with ASD showed a higher burden of deletions and duplications (> 1 kb) compared with (1) control mothers of unaffected children, (2) both the ASD or control children’s CNV burden, and (3) the enrichment observed in females in the general population [45]; Fig. 3b). Thus, a maternal-specific genetic contribution of excess CNVs in proband was demonstrated.

Second, in order to ultimately assess potential G × E effects, we first applied SNP-based heritability and genome-wide association studies to several E biomarkers. Our biomarker results (unrelated to ASD risk) suggested that a number of environmental prenatal and postnatal biomarkers were highly heritable and/or strongly associated with specific loci [42‐44, 49] (Fig. 3c). Results suggested that maternal PBDEs (5/7), maternal (2/22) and neonatal (8/42) cytokines/chemokines and vitamin D circulating during pregnancy and/or at birth are not solely determined by external stimuli (e.g. pollution or infections), but show significant contribution of specific maternal genetic loci relevant for xenobiotic and lipid metabolism [42], immune function [43], and vitamin D metabolism [44], respectively.

Third, we examined the relationships between biomarkers and genetic markers across mothers and neonates, independent of ASD outcome. We found evidence that maternal genetics was associated with neonatal immune biomarkers (3/42), and that fetal genetics had a strong association with mid-gestational maternal immune biomarkers (2/22) [43]. In most cases, we observed independent genetic contribution from the maternal non-transmitted genome and fetal non-inherited genomes and/or a different set of regulatory loci independently associated with maternal or fetal genetic influences. An example is the maternal soluble IL-2 receptor (sIL2-Ra) regulated by maternal genetics through the cytokine-encoding locus (IL2RA) and by fetal genetics through a non-cytokine encoding locus (ADCYAP1). The nearby ADCYAP1 gene exerts a transcriptional activating role not previously related to immune function [43]. Additionally, suggestive cross-associated loci influencing maternal and neonatal vitamin D were observed near immune genes. Relevant to pregnancy outcomes, we observed two distinct suggestive interactions between a maternal genetic variant, neonatal vitamin D and ID outcome, and between a second independent maternal genetic variant, neonatal IL-8 and ASD outcome [43, 44].

When we assessed whether the genetic factors controlling the measured biomarkers were associated with ASD outcome, we found none that were statistically convincing (although our power for case–control status was low due to the sample size) [42‐44, 49]. However, these studies help to elucidate how genetics strongly influences maternal physiology during gestation with intriguing potential consequences for consideration of the environmental response and its complex interplay with genetics.

Despite the unique strengths and contributions of the EMA study, certain limitations and considerations for placing findings from this work into the larger context of our understanding of ASD should be noted. The age at ascertainment of Phase 1 cases was on average younger than that for Phase 2, presenting the possibility that Phase 1 cases could represent greater severity. EMA relied on existing diagnostic records from regional centers of the DDS; while these records were reviewed by expert clinicians for consistency with ASD diagnosis, that determination was reliant on available data. Available information from a study that examined the completeness of DDS diagnoses suggests that DDS captures approximately 80% of ASD cases in the state, with milder cases more likely to be missed [117, 118]. We cannot rule out potential misclassification in controls, though any such misclassification would not be expected to influence findings here given the low prevalence of ASD in the general population. Likewise, information on comorbid ID status was reliant on DDS records, and it is possible that some ID cases, or comorbid ID status, may have gone undiagnosed. We do not have data to quantify the extent to which this may occur. We have not been able to examine associations with other comorbidities highly prevalent in ASD, such as anxiety and ADHD, as may be useful to address questions related to associations with other phenotypic subgroups and comorbidites. We had a relatively small number of female ASD cases, limiting power to address sex differences in some analyses. We also lacked in-depth clinical data across the child’s lifespan, preventing the study of developmental trajectories. Certain subgroup analyses (e.g., by sex) may have had more limited statistical power and require replication in larger samples. Many factors examined in phase 1 were not examined in phase 2 participants, precluding ability to pool data across phases. Prenatal confounder information was limited to birth certificate and prenatal screening data, and information on some potential confounders and modifiers, such as parental psychiatric history, or maternal diet, was not available. The EMA design did not include any direct contact with the participants; thus, additional research measures (such as parent-report questionnaires) that may be conducted in ASD case–control studies are not present in our database. These have been the trade-offs enabling a large population-based study with diagnostic information and biospecimens collected during critical developmental windows in mother–child pairs.

As a genetic study, the EMA design and dataset had several relatively unusual features. Though we were able to utilize mother–child pairs to assess some family-based analyses, we did not have paternal genetic contribution to maximize the understanding of which effects might be specific to pregnancy. On the other hand, the availability of unrelated controls allowed for exploration of G × E effects that are challenging in family-based designs. Although the genetic heterogeneity of the dataset created the risk of population stratification, we were able to adjust the analyses for genetic ancestry using the principal components derived from genome-wide maternal and fetal SNPs to account for population stratification, facilitating increased generalizability to the broader US population. The study contributed novel information regarding the relationship between genetic variation and biomarker levels, supporting strong patterns of genetic influence in such levels.

Additional considerations for EMA’s environmental exposure findings, and for placing this study’s results within the larger context of related work, include its geographic location and setting. Our overall null findings for certain factors may have been related to reduced exposure variability; specifically, vitamin D deficiency was low, while air pollution exposure was relatively high, for the Southern California counties included in EMA. EMA’s decision to match cases and controls on month and year of birth (to address concerns related to seasonality in infections, given a key focus of the study was on the immune system’s role in ASD) may also impact exposures that are seasonally associated and reduce ability to detect such associations. For exposures with decreasing (e.g., PBDEs, OC pesticides) or increasing (e.g., PFASs) use, the time period of EMA births (early 2000′s) must also be considered, as varying exposure levels may impact the ability to detect associations, and their observed effect sizes.

Finally, in placing EMA environmental findings in context, combined effects of exposures must also be considered. Environmental exposures vary across populations; these include not only air pollution levels and EDCs, but also other environmental factors not measured here but likely to either influence neurodevelopmentally-relevant pathways directly, or interact with these exposures, such as maternal diet. Such variation means that even in the face of true causal associations, results across studies of environmental risk factors may not be consistent. Future replication of findings should consider such population-level variability and seek to address exposures from a comprehensive standpoint. In compiling a dataset with over 70 biomarkers, many of which are measured in both maternal and child samples, as well as genotypes, the EMA study has been an important step in ASD research towards this overarching aim of considering multiple exposures. Future studies in the -omics and consortium era are likely to build on these findings.

The next phases of the EMA study focus on tying together the above exposures to consider pathways and mechanisms, and additional extensions of our combined exposures work. Animal model evidence, and some emerging human studies, suggest that both thyroid hormones and vitamin D may lie along the hypothesized pathway between the environmental chemicals considered in EMA and neurodevelopmental outcomes such as ASD and ID. Emerging work has demonstrated air pollution levels may lead to lower vitamin D levels [119]. Though evidence to date has not provided strong and consistent support for direct associations between thyroid hormones and ASD, thyroid hormones are known to play an important role in neurodevelopment, and links between thyroid hormone levels and EDC exposure have been demonstrated [120]. Ongoing analyses in EMA will examine the potential interaction or mediating effects of vitamin D, as well as TSH, on the associations between chemical exposures and ASD. To our knowledge, EMA will be the first study to examine such pathways in pregnancy for certain classes of chemicals.

In addressing genetic contributions, EMA will also be able to apply our novel G × E and maternal–fetal genetics approaches to additional pre- and post-natal risk factors, including maternal antibodies to fetal brain, to potentially identify novel genetic involvement. In addition to further G × E considerations, EMA has the capacity to examine novel E × E interactions, including not just extensions of existing EDC mixture analyses that may incorporate air pollutant exposures, but also similar strategies to address combined effects of biomarkers available and previously examined only as single exposures (such as, for example, cytokines).