Abstract
DNA methylation acts at the interface of genetic and environmental factors relevant for autism spectrum disorder (ASD). Placenta, normally discarded at birth, is a potentially rich source of DNA methylation patterns predictive of ASD in the child. Here, we performed whole methylome analyses of placentas from a prospective study of high-risk pregnancies. 400 differentially methylated regions (DMRs) discriminated placentas stored from children later diagnosed with ASD compared to typical controls. These ASD DMRs were significantly enriched at promoters, mapped to 596 genes functionally enriched in neuronal development, and overlapped genetic ASD risk. ASD DMRs at CYP2E1 and IRS2 reached genome-wide significance, replicated by pyrosequencing, and correlated with expression. Methylation at CYP2E1 associated with both ASD diagnosis and cis genotype, while methylation at IRS2 was unaffected by cis genotype but modified by preconceptional maternal prenatal vitamin use. This study therefore identified two potentially useful early epigenetic markers for ASD in placenta.
Introduction
Autism spectrum disorder (ASD) is a heterogeneous neurodevelopmental disorder diagnosed by a combination of behavioral features including restricted interests, repetitive behaviors, language deficits, and impairments in social communication (Baio et al., 2018). 1 in 59 children in the United States are diagnosed with ASD, at a mean age of 4.2 years (Baio et al., 2018). ASD is currently diagnosed by clinicians trained on the Autism Diagnostic Observation Schedule (ADOS) and the Autism Diagnostic Interview - Revised (ADI-R) according to the Statistical Manual of Mental Disorders (DSM-5) which is most accurate at or after 36 months (Baio et al., 2018). However, an early assessment of ASD risk could identify infants and toddlers who would benefit from behavioral interventions that improve cognitive, social, and language skills.
Monozygotic versus dizygotic twin and sibling studies suggest a strong genetic basis for ASD (Hannon et al., 2018; M. B. Jones & Szatmari, 1988; Ritvo et al., 1989; Tsai & Bell, 2015; Wessels & Pompe van Meerdervoort, 1979). However, mutations in any individual gene account for less than 1% of ASD cases (Bourgeron, 2015; Tsai & Bell, 2015). Genetic sequencing analyses can only identify a potentially causative genetic abnormality in ~25% of clinical ASD diagnoses (Bourgeron, 2015; Tsai & Bell, 2015; Turner et al., 2016). While genome-wide association studies (GWAS) also support common genetic variants in ASD, the complexity and heterogeneity of ASD has been a major challenge (Abrahams et al., 2013; Grove et al., 2017; Iossifov et al., 2014; Sanders et al., 2015). Evidence for environmental risk factors in ASD point to in utero maternal exposures such as air pollution, fever, or asthma and nutrients specifically the absence of pre-conceptional prenatal vitamin intake (Raz et al., 2015; Schmidt et al., 2011, 2012; Zerbo et al., 2013). Maternal prenatal vitamins, which contain high levels of folate and other additional B vitamins, protect offspring by up to 70% for neural tube defects (Caramaschi et al., 2017; Howsmon, Kruger, Melnyk, James, & Hahn, 2017; Kalkbrenner, Schmidt, & Penlesky, 2014; Relton et al., 2004; Rush, Katre, & Yajnik, 2014; Zeisel, 2009), and correlate with an overall 40% reduction in ASD risk if taken during the first month of pregnancy (P1) (Schmidt et al., 2011, 2012; Suren et al., 2013). This finding was replicated with a large prospective study in Norway including over 80,000 pregnancies (Suren et al., 2013).
DNA methylation shows dynamic changes during fetal development (Vogel Ciernia & LaSalle, 2016; Crawley, Heyer, & LaSalle, 2016; Smallwood & Kelsey, 2012b) and contains the molecular memory of in utero experiences such as maternal nutrition (Howsmon et al., 2017; Teh et al., 2014). Term placenta is an accessible fetal tissue that maintains the distinctive embryonic bimodal DNA methylation pattern, in which expressed genes are marked by higher methylation levels (Schmidt et al., 2016; Schroeder et al., 2013, 2015, 2016). Placenta therefore offers a unique window to study DNA methylation patterns that may reflect altered fetal development relevant to ASD genetic risk (Schroeder et al., 2015; Smallwood & Kelsey, 2012b, 2012a; Watson & Cross, 2005). Specifically, a recent study of polygenic risk scores for schizophrenia demonstrated a significant interaction of genetic risk with maternal perinatal environmental factors that affected placental gene expression (Ursini et al., 2018). Previous analyses of DNA methylation patterns in placenta samples from a high-risk ASD cohort also identified an association between self-reported use of lawn and garden pesticides and large-scale changes in DNA methylation patterns, and identified a putative enhancer of the DLL1 gene as differentially methylated in ASD (Schmidt et al., 2016; Schroeder et al., 2016).
Here, we continue the epigenetic investigation of ASD risk through the novel approach of identifying differentially methylated regions (DMRs) in whole methylomes from placenta samples from male children later diagnosed with ASD compared to children with typical development (TD) controls. Two genome-wide significant ASD-associated DMRs at CYP2E1 and IRS2 were further validated and investigated for effects of genotype, RNA expression, and protein levels as well as interactions with preconception prenatal vitamin use. Understanding the epigenetic patterns of ASD associated with maternal prenatal vitamin use in placenta could lead to the development of preventative and therapeutic early interventions for high-risk children with ASD.
Results
Placenta ASD DMRs discriminate ASD from TD samples
To identify novel differentially methylated gene loci between ASD and TD, a differentially methylated region (DMR) bioinformatic analysis was performed on placenta whole genome bisulfite sequencing (WGBS) data (Schmidt et al., 2016; Schroeder et al., 2016). 400 DMRs were identified with a threshold of > 10% methylation difference between ASD and TD groups, and these were associated with 596 genes (Fig. 1A, Supplementary Table 1). There was no bias for gene length in the ASD DMR associated genes compared to all human genes (Supplementary Fig. 1). 296 DMRs were hypermethylated, while 104 DMRs were hypomethylated in ASD compared to TD placenta (Fig. 1A). Principal component analysis (PCA) using methylation levels for each sample over the 400 DMRs demonstrated a clear separation of placental samples by child outcome of ASD versus TD (Fig. 1B). In addition, most ASD DMRs showed a highly significant association with Mullen Scales of Early Learning (MSEL) scores and autism severity score from Autism Diagnostic Observation Schedule (ADOS), but not with potential confounding variables (Supplementary Table 2, Supplementary Fig. 2).
Placenta ASD DMRs were enriched for transcription start sites and genes that function in transcriptional regulation and neuronal fate
To further study the location and function of ASD DMRs in placenta, we calculated the location of each ASD DMR relative to the assigned gene’s transcription start site (TSS) (Supplementary Table 3). Both hyper- and hypomethylated ASD DMRs were enriched within 5kb on either side of TSS compared with background regions (Fig. 1C). Gene ontology (GO) analysis of ASD DMRs genes revealed significant enrichment for functions in transcription, protein modification, embryonic organ development, and neuron fate commitment by Fisher’s exact test after false discovery rate (FDR) multiple test correction (Fig. 1D, Supplementary Table 4).
Placenta DMR genes were enriched in ASD but not ID risk genes
To test a hypothesized overlap between epigenetic and genetic ASD risk loci observed previously in ASD and neurodevelopmental disorder brain tissues (Vogel Ciernia et al., 2018), we investigated the possible overlap of placenta ASD DMR genes with identified genetic risk factors for ASD and other types of intellectual disability (ID). First, the curated Simons Foundation Autism Research Initiative (SFARI) gene list was separated into six categories based on SFARI ASD gene scores (Abrahams et al., 2013). The entire list of SFARI genes as well as the high confidence gene list both showed significant overlap with placenta ASD DMR genes (Fig. 2A, Supplementary Table 5). The 39 genes in common between the SFARI gene list and placenta ASD DMRs were significantly enriched for functions in positive regulation of histone H3K4 methylation, multicellular organ development, and system development. Second, high risk ASD genes from Sanders et al. (Sanders et al., 2015) and likely gene-disrupting (LGD) recurrent mutations and missense mutation on de novo mutations to ASD gene lists from whole genome exome sequencing (Iossifov et al., 2014) were also significantly enriched for placental ASD DMRs. In contrast, no significant enrichment was observed for placental ASD DMRs with ID, Alzheimer’s disease, or lung cancer genetic risk (Gilissen et al., 2014) or a random set of 400 genomic regions mapped to 600 genes (Fig. 2A, Supplementary Table 5). When placental ASD DMRs were separated by direction of change, hypomethylated ASD DMRs exhibited more categories of significant enrichment with ASD genetic risk compared with hypermethylated ASD DMRs (Supplementary Fig. 3)
Methylation data from human postmortem brain was obtained from previous published datasets, GSE8154 (ASD and TD) (Vogel Ciernia et al., 2018). C. Gene ontology and pathway analysis on the 36 genes in common between placenta ASD DMRs and brain ASD DMRs associated genes. Enrichment tests were done on Fisher’s exact test with FDR 0.05 correction. Genes in each gene ontology term are shown within each bar.
Placenta ASD DMR genes significantly overlapped with brain ASD DMRs that were enriched for Wnt and cadherin signaling pathways
From a prior methylation analysis in brain frontal cortex, 210 ASD discriminating DMRs from brain (10% methylation difference between ASD and TD) were identified, which mapped to 371 genes (Vogel Ciernia et al., 2018). A significant overlap of ASD DMR genes was observed between placenta and brain (Fisher’s exact test, p-value < 0.001), with 36 genes in commons (Fig. 2B, Supplementary Table 5). Those 36 genes were significantly enriched for functions in the Wnt signaling and cadherin signaling pathways (Fig. 2C). Of these shared genes four overlapped with SFARI genetic risk: GADD45B, MC4R, PCDH9, and TBL1XR1.
Placenta ASD DMRs were enriched for placental and brain active promoter H3K4me3 peaks, promoter flanking regions, and CpG shores
To functionally annotate the ASD DMRs identified by placenta WGBS, multiple histone modification ChIP-seq peaks and chromatin state predictions from multiple tissue types in the Roadmap Epigenomics Projects were compared to ASD DMR chromosomal locations for enrichment compared to background regions (Kundaje et al., 2015). Placental ASD DMRs were significantly enriched for H3K4me3 and H3K4me1 marks of promoters and enhancers across multiple tissues, although placental H3K4me3 marks showed the strongest (odds ratio = 17.08, FDR q < 1.8E-42) and brain H3K4me3 marks showed the second strongest enrichment (odds ratio = 13.75, FDR q < 3.55E-31) (Fig. 3A). Next, we overlapped ASD DMRs with published chromatin state predictions that use histone modification ChIP-seq data to annotate the genome into 15 functional states (chromHMM) (Ernst & Kellis, 2012). Placental ASD DMRs showed significant enrichment in regions flanking transcription start site (TssAFlnk) and transcription start site (TssA) compared to background over multiple tissues (Fig. 3B). When separated by directional change in ASD, both hyper- and hypomethylated ASD DMRs were significantly enriched for H3K4me3 peaks, transcription start sites and their flanking regions, as well as enhancers (Supplementary Fig. 4).
We also separated the genome into four parts relative to CpG island location (Aryee et al., 2014; Sandoval et al., 2011; Timp et al., 2014). CpG shores were defined as the region within 2 kb on both sides of CpG islands, while another 2 kb extension from the shores were defined as CpG shelves. The remaining regions were defined as “open sea”. Placental ASD DMRs showed significant enrichment at CpG shores, and hypermethylated ASD DMRs more significantly overlapped CpG islands compared with hypomethylated DMRs (Fig. 3C).
Two genome-wide significant placental ASD DMRs at CYP2E1 and IRS2 validate by pyrosequencing and correlated with gene expression
Two of the 400 ASD DMRs identified in ASD placenta reached genome-wide significance by family-wide error rate (FWER), including chr10: 133527713–133529507, located inside CYP2E1 (cytochrome P450 2E1), and chr13: 109781111–109782389 located inside IRS2 (insulin receptor substrate 2) (Fig. 4). The CYP2E1 DMR was located after the first exon, included the first intron and part of the second exon, and was hypomethylated in ASD versus TD (Fig. 4A). The IRS2 DMR, spanning the end of the first exon and the beginning of first intron and was hypermethylated in ASD versus TD (Fig. 4B). Both CYP2E1 and IRS2 were also present in the gene lists overlapping with brain ASD DMR related genes and high risk ASD genes (Fig. 2A, Supplementary Table 5) (Sanders et al., 2015).
Pyrosequencing was performed as an independent method to verify methylation differences between ASD and TD placental samples at CYP2E1 and IRS2 DMRs (Supplementary Table 6). For the CYP2E1 DMR, there was a significant difference in average percent methylation detected by pyrosequencing between ASD and TD samples (Fig. 4C). 13 CpG sites were included in the CYP2E1 DMR pyrosequencing test, and all but two also showed individually significant differences between ASD and TD after false discovery rate (FDR) correction (Supplementary Table 6, Supplementary Fig. 5A). Pyrosequencing also confirmed a significant difference between ASD and TD average percent methylation at the IRS2 DMR (Fig. 4D) and all of the 11 CpG sites individually assayed at IRS2 (Supplementary Table 6, Supplementary Fig. 5B).
While MARBLES placenta samples were not collected in a manner conducive to RNA stability for gene expression analyses, we were able to examine expression level of both CYP2E1 and IRS2 in MARBLES umbilical cord blood from an Affymetrix Human Gene 2.0 array analysis in a related study (Mordaunt, Park, et al., 2018). A trend for lower CYP2E1 transcript levels was observed in ASD compared with TD cord blood samples (Fig. 5A) consistent with the direction of the placental methylation for this locus (Fig. 4C). Similarly, a trend for higher IRS2 expression in ASD versus TD cord blood was observed along with higher ASD methylation in placenta (Fig. 4D, Fig. 5B). To investigate the direction of expression changes for these genes in ASD brain, we utilized the dbMDEGA database (Zhang et al., n.d.). CYP2E1 showed a significantly downregulated in ASD compared with TD in human male cortex, while IRS2 trended for higher levels in ASD compared with TD. These were both in the same direction as ASD cord blood expression and ASD placental methylation compared to controls. Furthermore, a trend for higher IRS2 protein in ASD placenta samples compared with TD placenta samples was observed by Western blot (Fig. 5C, Supplementary Fig. 6), as expected based on transcript and methylation levels. Because of the distinctive methylation landscape in placenta, positive correlations between methylation and expression were expected for gene body locations outside of CpG islands (Schroeder et al., 2013).
CYP2E1 but not IRS2 DMR methylation levels were associated with cis genotypes
We performed Sanger sequencing within the CYP2E1 and IRS2 ASD DMRs to identify single nucleotide polymorphisms (SNPs) that could explain some of the methylation differences. Two SNPs (rs943975, rs1536828) were identified within the boundaries of the CYP2E1 DMR in the 41 placenta samples (Supplementary Table 7). A significant association between rs1536828 (but not rs943975) genotype and CYP2E1 DMR percent methylation was observed, with samples homozygous for the minor allele (G/G) showing the lowest methylation (Fig. 6A). A single informative SNP (rs9301411) was also identified within the IRS2 DMR (Supplementary Table 7) but was not significantly associated with methylation level (Fig. 6B).
Preconception prenatal vitamin use corresponded to protective placental DNA methylation patterns at CYP2E1, IRS2, and genome-wide
Placental samples from mothers who took prenatal vitamins during the first month of pregnancy showed a trend for higher CYP2E1 DMR methylation that was not significant, but in the same direction expected for protection from ASD (Fig. 7A). At the IRS2 DMR, however, there was a significant association with maternal prenatal vitamin use and lower methylation, also in ASD-protective direction (Fig. 7B).
To further explore the relationship between placental methylation patterns influenced by prenatal vitamin use in the first month of pregnancy, placental methylomes were analyzed for DMRs by prenatal vitamin use in the first month of pregnancy (PreVitM1) with more than 10% methylation difference, and 376 DMRs were identified over 587 genes (Supplementary Table 8). 60 genes overlapped between PreVitM1 DMRs and ASD DMRs in placenta (Supplementary Table 8, Fig. 7C). Gene ontology analysis showed that genes common to PreVitM1 and ASD DMRs were significantly enriched for functions in neuron fate commitment, transcription regulation, central nervous system development, and regulation of phosphatidylinositol 3-kinase activity (Fig. 7D). We also separated placental samples based on when mothers started taking prenatal vitamins during pregnancy into three periods (Supplementary Table 9). For the CYP2E1 DMR, we found that during all three periods, ASD placental samples showed lower percent methylation on the CYP2E1 DMR compared to TD (Supplementary Fig. 7A). The expected opposite finding of higher methylation levels in ASD compared with TD placental samples was observed at the IRS2 DMR (Supplementary Fig. 7B).
To further investigate the potential inter-relatedness of diagnosis, prenatal vitamin use, and cis genotype on methylation at CYP2E1 and IRS2 DMRs, we calculated associations between each factor and methylation separately by two-tailed t-test or ANOVA, as well as two-way diagnosis and PreVitM1; diagnosis and genotype; genotype and PreVitM1 by Pearson’s chi-squared test. These analyses illustrate that CYP2E1 genotype and diagnosis significantly contributed to CYP2E1 DMR methylation, while PreVitM1 and diagnosis were significantly associated with IRS2 DMR methylation (Fig. 8A, Fig. 8B). No significant association was found between two-way interactions among each of the three factors and each DMR methylation level by ANOVA (Supplementary Table 10). When combining CYP2E1 genotype, PreVitM1, and diagnosis to predict methylation at the CYP2E1 DMR, a significant association was observed on CYP2E1 DMR methylation with diagnosis and rs1536828 genotype (Fig. 8C, Supplementary Fig. 8A). At the IRS2 DMR, PreVitM1 and diagnosis significantly contributed to IRS2 DMR methylation (Fig. 8D, Supplementary Fig. 8B).
Discussion
This study utilized whole methylome analysis of prospectively stored placenta samples in a high risk ASD cohort to bioinformatically identify novel gene loci that were able to discriminate child outcome at age three. This unbiased analysis of ASD differentially methylated regions in placenta tissue resulted in several novel findings. First, the 596 genes identified from 400 placental ASD DMRs significantly overlapped with genetic risk for ASD from curated databases and gene functions in neurons. Second, two genome-wide significant placental ASD DMRs at CYP2E1 and IRS2 were discovered that were validated by pyrosequencing and also overlapped with ASD-associated genetic variation and gene expression changes. Lastly, we investigated genotype and nutrient factors correlating with methylation at CYP2E1 and IRS2, demonstrating specific effects for cis genotype and diagnosis at CYP2E1 and prenatal vitamin use at IRS2. These results therefore suggest that DNA methylation patterns in placenta provide a direct link between genetics, environment, and fetal epigenetic programming, which can reflect early development relevant to the complex etiology of ASD. The epigenomic signature of ASD in placenta also provides important insights into gene functions, pathways, gene-environment interactions, and potential biomarkers that may be useful in improving early detection of ASD.
This study is the first to identify 400 potential ASD DMRs that distinguish between ASD and TD placenta samples and highlights specific locations and gene functions of differentially methylation in placental samples from children with ASD. First, these placental ASD DMRs were highly enriched around transcription start sites and H3K4me3 marks that are clear marks of gene regulatory functions (Carninci et al., 2006; Koudritsky & Domany, 2008; Portales-Casamar et al., 2007; Yang, Bolotin, Jiang, Sladek, & Martinez, 2007). Furthermore, gene ontology analysis of the 596 genes mapped to placental ASD DMRs pointed to enriched gene functions in transcription, neuron fate, and embryonic development, which were expected based on previous studies (Dapretto et al., 2006; Geschwind & Levitt, 2007; Schroeder et al., 2016). Genes with ASD DMRs in both placenta and brain were enriched for Wnt and cadherin signaling pathways. Wnt signaling is important in embryogenesis, tissue regeneration, and neurodevelopment (Katoh & Katoh, 2006; Logan & Nusse, 2004; Nusse & Clevers, 2017), while cadherin signaling plays a vital role in connecting major intracellular signaling pathways with adhesion protein complexes (Klezovitch & Vasioukhin, 2015; Yap & Kovacs, 2003). Our results therefore complement previous studies that have shown the importance of Wnt and cadherin pathways in the etiology of ASD (Betancur, Sakurai, & Buxbaum, 2009; Kalkman, 2012; Krey & Dolmetsch, 2007). We also replicated our previous finding of differential methylation at DLL1 in ASD placentas (Schroeder et al., 2016) (Supplementary Table 2). DLL1 encodes a ligand of Notch, activated by Wnt signaling (Hofmann et al., 2004).
When overlapped with datasets of genetic risk for neurodevelopmental disorders including ASD and intellectual disability (Abrahams et al., 2013; Iossifov et al., 2014; Sanders et al., 2015), placenta ASD DMRs were significantly enriched for ASD but not for intellectual disability genetic risk, illustrating the specificity of the ASD DMRs identified in our study. The highest overlap of ASD DMR associated genes was with the SFARI high confidence genes, including KMT2A, MYT1L, and TBR1 (Abrahams et al., 2013). KMT2A is expressed in brain and placenta and encodes for a transcriptional coactivator (lysine methyltransferase 2A) that modulates H3K4 methyltransferase activity, specifically the transfer of methyl groups from S-adenosylmethionine to lysine residues on histones (Allis et al., 2007; Shilatifard, 2008) and has been previously implicated in brain development (E. Shen, Shulha, Weng, & Akbarian, 2014; Vallianatos & Iwase, 2015). MYT1L encodes for zinc finger transcription factor that functions in the developing mammalian central nervous system and is associated with neurodevelopmental disorders (Blanchet et al., 2017; Wang et al., 2010). TBR1 (T-box, brain, 1) is a transcription factor which is vital for vertebrate embryo development, neuron migration and differentiation (Bedogni et al., 2010; Englund et al., 2005). In addition, our two genome-wide significant ASD DMR associated genes, CYP2E1 and IRS2, were both on the list of “high confidence” genetic risk for ASD (Sanders et al., 2015).
Four genes were found to be in common between placenta ASD DMRs, brain ASD DMRs and SFARI genetic risk, specifically GADD45B, MC4R, PCDH9 and TBL1XR1 (Betancur et al., 2009; Garbett et al., 2008; Orlik & Halawa, 2016; Tabet et al., 2014). GADD45B (growth arrest and DNA-damage-inducible, beta) responds to environmental stress through JNK pathway induced DNA demethylation of neurogenesis and synaptic plasticity at gene promoters (Garbett et al., 2008; Ma et al., 2009; Sultan, Wang, Tront, Liebermann, & Sweatt, 2012). MC4R encodes the membrane-bound melanocortin 4 receptor, implicated in hormone and cell growth pathways in obesity and insulin resistance (Chambers et al., 2008; Orlik & Halawa, 2016; Yeo et al., 1998). PCDH9 (protocadherin 9) is a cadherin signaling pathway gene with specific signaling function at neuronal synaptic junctions (Betancur et al., 2009; Bruining et al., 2015). TBL1XR1 encodes the nuclear receptor corepressor transducing beta like 1 X-linked receptor 1 that binds to histone deacetylase 3 (HDAC 3) complexes in neuron development (Gonzalez-Aguilar et al., 2012; Pons et al., 2015; Tabet et al., 2014). A GWAS noise reduction (GWAS-NR) method to correct for false-positive association with ASD identified cadherin and signaling transduction pathways that included PCDH9 and IRS2 as high confidence ASD genes (Hussman et al., 2011).
Our study identified novel methylation differences at CYP2E1 and IRS2, which exhibited genome-wide significant differences between ASD and TD. Both CYP2E1 and IRS2 are identified as ASD genetic risk genes in multiple databases related to ASD genetic risk across different tissues and populations (Vogel Ciernia et al., 2018; Sanders et al., 2015). Both CYP2E1 and IRS2 DMRs are located close to the TSS site at CpG shore intragenic regions, which is also consistent with the enrichment for TSS flanking regions and H3K4me3 promoter marks in the 400 ASD DMRs. These results are consistent with the developmental dynamics of H3K4me3 marks in human prefrontal cortex, which have been observed to be altered in ASD (Cheung et al., 2010; Shulha et al., 2012). Structural variants and SNPs in cis-regulatory elements also showed significant contribution to ASD (Brandler et al., 2018; W. Sun et al., 2016).
CYP2E1 encodes a member of the cytochrome P450 superfamily that is involved in the metabolism of drugs, including analgesics like acetaminophen, fatty acids such as arachidonic acid, and a range of chemical toxins, including halogenated hydrocarbons, benzene, and its activity is inducible by drugs, alcohol, and xenobiotics. It thus has an important role in drug bioavailablity (Gonzalez, 1988; Koop, 1992; Rasheed, Hines, & McCarver-May, 1996; Traglia et al., 2017). Previous studies showed those proteins essential for embryonic development in human, rat and zebrafish (Jukka Hakkola et al., 1996; S. M. Jones, Boobis, Moore, & Stanier, 1992; Kishida & Callard, 2001; Ko, Choi, Green, Simmen, & Simmen, 1994; Majdic, Sharpe, O’Shaughnessy, & Saunders, 1996). We observed a significant association between methylation and cis genotype at the CYP2E1 DMR, a finding which is consistent with the identification of this locus in a screen for human metastable epialleles variability between individuals (Silver et al., 2015). In addition, in immune models of ASD, maternal interleukin-6 (IL6) crosses the placenta, disrupting development of hippocampal spatial learning (Boksa, 2010; Jonakait, 2007; Krakowiak et al., 2012). Previous studies showed that IL6 inhibits CYP1A1, CYP1A2 and CYP2E1 expression (Abdel-Razzak et al., 1993; J. Hakkola, Hu, & Ingelman-Sundberg, 2002; Jover, Bort, Gómez-Lechón, & Castell, 2002; Patel et al., 2014), consistent with the lower methylation and expression levels in ASD versus TD observed in our study. In addition, CYP2E1 expression is transcriptionally regulated by the JAK2/STAT3 pathway, providing a potential convergent pathway with IRS2 (Fig. 9) (Patel et al., 2014).
IRS2 encodes for insulin receptor substrate 2, a cytoplasmic signaling molecule that mediates the effects of insulin and insulin-like growth factor 1 (IGF1) (Park et al., 2016; Withers et al., 1998) and cytokine receptors (Ihle, 1995; Rui, Yuan, Frantz, Shoelson, & White, 2002; X. J. Sun et al., 1995). IRS2 has a phosphotyrosine-binding domain which contributes to the intracellular affinity to cell membrane receptors (Fig. 9) (Eck, Dhe-Paganon, Trüb, Nolte, & Shoelson, 1996; Machado-Neto, de Melo Campos, & Traina, 2018; Schlessinger & Lemmon, 2003; White, 1998). PI3K/AKT/mTOR and MAPK signaling pathways are linked with IRS2 in the regulation of protein synthesis and cell proliferation (Archuleta et al., 2009; Machado-Neto et al., 2018; Patti et al., 1995; Tamburini et al., 2008; Velloso et al., 1996). When activated by cytokine and hormone receptors, IRS2 stimulates JAK2, leading to STAT and MAPK signaling activation (Carvalheira, Ribeiro, Folli, Velloso, & Saad, 2003; Machado-Neto et al., 2018; Saad, Carvalho, Thirone, & Velloso, 1996; Saad, Velloso, & Carvalho, 1995; Velloso et al., 1996). The insulin-like growth factors (IGF1) pathway, which includes IRS2, also mediates de novo DNA methylation by DNA methyltransferase (DNMT) through AKT (Fang et al., 2015; Lin & Wang, 2014). This pathway may explain why methylation at the IRS2 DMR was sensitive to maternal prenatal vitamin intake, since IRS2 stimulates the mTOR (mechanistic target of rapamycin) pathway, which responds to nutrients and growth factors signaling to regulate protein synthesis (Gulati & Thomas, 2007; M. Laplante & Sabatini, 2013; Mathieu Laplante & Sabatini, 2012; Shimobayashi & Hall, 2014). Previous studies have shown that maternal folate alters amino acid transport activity in the placenta, resulting in affecting fetal growth by the mTOR signaling pathway (Huang & Fingar, 2014; Shimobayashi & Hall, 2014). Our analysis shown that placental differentially methylated gene loci associated with prenatal vitamin intake were also highly enriched for functions in fetal growth and development. We showed that high IRS2 methylation is significantly associated with ASD and mothers who did not take prenatal vitamins before conception, suggesting that sufficient folate levels around placental implantation may be protective for ASD through IRS2-mTOR signal transduction. The link between epigenetic alterations in IRS2 and risk for ASD is particularly intriguing given a growing body of epidemiologic evidence demonstrating higher ASD risk in offspring born to mothers who experienced diabetes during pregnancy, including some very large and methodologically sound studies with clinical diagnoses of both maternal diabetes and child ASD (Li et al., 2016; Xiang et al., 2018; G. Xu, Jing, Bowers, Liu, & Bao, 2014).
We did not observe any significant associations between other potential cofounders such as maternal age, pregnancy BMI, or gestational age at birth and ASD diagnosis in the MARBLES study (Schmidt et al., 2016). Cell type heterogeneity in the placenta may complicate the interpretation of our results, however, our previous study did not detect differences in methylation levels by placental region at specific gene loci (Schroeder et al., 2016). Other potential limitations of our study include the relatively small sample size and sequencing depth. This study serves as a proof-of-principle that placenta methylation patterns detected by WGBS may be informative in ASD. Replication with additional samples, other similar prospective cohorts, and improved sequencing and bioinformatic strategies will be important in future studies.
In conclusion, we identified two high confidence genes differentially methylated in ASD from an unbiased analysis of DNA methylation in placenta from high-risk pregnancies and investigated possible genetic and environmental modifiers of methylation at both loci. Methylation levels at the CYP2E1 DMR were associated with genotype, while the methylation levels at the IRS2 DMR were associated with prenatal vitamin use. Our results are consistent with a previous study using the Illumina 450K array, which showed that both genetic and environmental effects influence DNA methylation levels (Hannon et al., 2018). Placenta reflects the essential interface between the fetus and mother, mediating the impacts of endocrine and growth factors in the maternal environment on fetal development (Koukoura, Sifakis, & Spandidos, 2012; Zeltser & Leibel, 2011). Both CYP2E1 and IRS2 are related to protein synthesis, cell proliferation, and cell metabolism, consistent with previous studies of convergent gene pathways in ASD (Vogel Ciernia et al., 2018; Sanders et al., 2015; Voineagu et al., 2011; Xu et al., 2012; Zhang et al., n.d.). These results therefore provide evidence that placental methylation levels reflect the intersection of genetic and environmental risk and protective factors that are expected to be useful for early intervention and prevention of ASD.
Materials and Methods
MARBLES study design, sample selection, and DNA isolation
The Markers of Autism Risk in Babies: Learning Early Signs (MARBLES) study design was described in a previous publication (Hertz-Picciotto et al., 2018). In MARBLES, mothers of at least one child with confirmed ASD who were pregnant or planning a pregnancy were recruited in the Northern California area. Inclusion criteria for the study were: 1) mother or father has one or more biological child(ren) with ASD; 2) mother is 18 years or older; 3) mother is pregnant; 4) mother speaks, reads, and understands English sufficiently to complete the protocol and the younger sibling will be taught to speak English; 5) mother lives within 2.5 hours of the Davis/Sacramento, California region at time of enrollment. With shared genetics, the next child has a 15-fold higher risk for developing ASD compared to the general population (Hertz-Picciotto et al., 2018). Demographic, diet, lifestyle, environmental, and medical information were prospectively collected through telephone-assisted interviews and mailed questionnaires throughout pregnancy and the postnatal period. Infants received standardized neurodevelopmental assessments beginning at 6 months and concluding at 3 years of age (Hertz-Picciotto et al., 2018). Diagnostic assessments at 3 years included the gold standard Autism Diagnostic Observation Schedule (ADOS) (Lord, Risi, Lambrecht, Cook Jr, et al., 2000; Lord, Risi, Lambrecht, Cook, et al., 2000), the Autism Diagnostic Interview-Revised (ADI-R) (Lord, Rutter, & Le Couteur, 1994), and the Mullen Scales of Early Learning (MSEL) (Mullen, 1995). Participants were classified into outcome groups including ASD and Typical Development (TD), based on a previously published algorithm that uses ADOS and MSEL scores (Chawarska et al., 2014; Ozonoff et al., 2014). Children with ASD outcomes have scores over the ADOS cutoff and meet DSM-5 criteria for ASD. Children with TD outcomes have all MSEL scores within 2.0 SD and no more than one MSEL subscale that is 1.5 SD below the normative mean and scores on the ADOS at least three more points below the ASD cutoff. This study utilized 41 male MARBLES placenta samples, including 20 samples from children later diagnosed with ASD and 21 children determined to have TD, matched for enrollment time frame and date of birth. DNA was isolated from 50–100 mg frozen placental tissues (20 ASD and 21 TD) using the Gentra Puregene tissue kit (Qiagen).
Whole Genome Bisulfite Sequencing (WGBS)
Raw sequencing data (fastq files) were published previously (Schroeder et al., 2016). Briefly, WGBS libraries were made with the sonicated genomic DNA (around 300 bp) and ligated with methylated Illumina adapters using NEB’s NEBNext DNA library prep kit (Schroeder et al., 2013, 2016). The library was bisulfite converted using EZ DNA Methylation lighting kit (Zymo), amplified for 12 cycles using PfuTurbo Cx Hotstart DNA Polymerase (Agilent) and purified with Agencourt AMPure XP Beads (Beckman Coulter). The quality and quantity of libraries were measured on a Bioanalyzer (Agilent) and sequenced on Illumina HiSeq 2000 with each sample per single lane. Reads after trimming were uniquely mapped to human reference genome (hg38) as described previously using BS-Seeker2 on average 1.6X genome converge with 99.3% bisulfite conversion efficiency (measured through the percentage of non-CpG cytosines that were unconverted) (Dunaway et al., 2016; Guo et al., 2013; Schroeder et al., 2013).
ASD Differentially Methylated Regions (DMRs) and genome-wide significant DMRs
DMRs were called as described in previous publications (Coulson et al., 2018; Dunaway et al., 2016; Mordaunt, Shibata, et al., 2018) using the default settings. In this case, each ASD DMR contained greater than 10% methylation difference between ASD and TD samples at least three CpGs within 300 base pairs (bp) and a p-value < 0.05. Background regions were defined using the same conditions as DMRs but without any percent methylation filters to identify all possible DMR locations based on CpG density and sample sequencing coverage. Hypermethylated ASD DMRs were defined as higher percent methylation in ASD versus TD, while hypomethylated ASD DMRs was were defined as lower percent methylation in ASD versus TD samples. Genome-wide, significant DMRs were identified based on a family-wide error rate (FWER) < 0.05, determined by permuting the samples 1000 times by chromosome, and counting the number of null permutations with equal or better DMRs ranked by number of CpGs and areaStat (Box, 1980).
Hierarchical clustering and principal component analysis (PCA)
Methylation was extracted at each ASD DMR for every sample. Percent methylation of each sample was normalized to the mean methylation of each ASD DMR. ASD DMRs were grouped by Ward’s Method of hierarchical clustering (Wilks, 2011). Principal component analysis was performed on methylation at all ASD DMRs across all samples using the prcomp function and ggbiplot package in R. The ellipses for each group were illustrated as the 95% confidence interval. The lack of overlapping ellipses for ASD and TD samples indicated significant methylation difference in ASD DMRs between groups (p < 0.05).
Assignment of DMRs to genes and relative location to TSS
Genes were assigned to DMRs using the Genomics Regions Enrichment of Annotations Tool (GREAT) on the default association settings (5.0 kilo-base (kb) upstream and 1.0 kb downstream, up to 1000.0 kb max extension) (McLean et al., 2010). The distance (kb) was calculated from the ASD DMRs, hypermethylated ASD DMRs, hypomethylated ASD DMRs and background regions to transcription start site (TSS) of the GREAT assigned gene. The gene length was calculated for both placental ASD DMR genes and all genes in human genome and tested for potential distribution differences by Pearson’s chi-squared test.
Gene Ontology Term and Pathway Enrichment Analysis
Gene ontology (GO) analysis was done using PANTHER (Protein Analysis Through Evolutionary Relationships) overrepresentation test, with the GO Ontology database (Ashburner et al., 2000; The Gene Ontology Consortium, 2017) and Fisher’s exact test with false discovery rate (FDR) multiple test correction. GO term enrichments were presented by the hierarchical terms rather than specific subclass functional classes, as described previously (Mi, Muruganujan, & Thomas, 2012; Thomas et al., 2003).
Tests for ASD DMR Enrichments
All tests of enrichment for ASD DMRs were compared to a set of all possible background regions that are calculated in the DMR analysis pipeline. Enrichment tests for placenta ASD DMRs associated genes and published gene lists were done using the GeneOverlap R package which implements Fisher’s exact test and adjusted for FDR correction (L. Shen et al., 2013). *p < 0.05, **p < 0.01, ***p < 0.001 by Fisher’s exact test with FDR corrected. Brain cortex (BA9) ASD DMRs were defined as either a 5% or 10% methylation difference between ASD and TD and were described previously using the same method as placenta ASD DMRs (Vogel Ciernia et al., 2018). The SFARI (Simons Foundation Autism Research Initiative) database was used for the five categories of ASD risk genes (https:gene.sfari.org/database/gene-scoring/) (Abrahams et al., 2013). High effect ASD risk gene lists were also identified from Sanders et al. (Sanders et al., 2015). Likely gene-disrupting (LGD) recurrent ASD mutations and missense mutation on de novo mutations were obtained from Iossifov et al. (Iossifov et al., 2014). Gene lists on intellectual disability (ID) were obtained from Gilissen et al. (Gilissen et al., 2014). Alzheimer’s disease GWAS gene lists were extracted from SNPs showing association with Alzheimer’s disease (P ≤ 1×10−3) (Harold et al., 2009). Lung cancer GWAS gene lists were acquired from Landi et al. (Landi et al., 2009). The random genes category contains the same number of regions as the placenta ASD DMRs to serve as a specificity control. ASD DMRs were examined for enrichment with known chromatin marks compared to the background using LOLA R package with two-tailed Fisher’s exact test after FDR correction (Sheffield & Bock, 2016). Placenta histone marks H3K4me1, H3K4me3, H4K9me3, H3K36me3, H3K27me3 and H3K27ac were extracted from ENCODE (Encyclopedia of DNA Elements) placenta ChIP-seq dataset (ENCODE Project Consortium, 2012; Sloan et al., 2016). ASD DMRs were also analyzed for overlap with chromatin states predicted by chromHMM, which use histone modification ChIP-seq data to separate the genome into 15 functional states in the Roadmap Epigenomics Project using a Hidden Markov Model (Ernst & Kellis, 2017; Kundaje et al., 2015). For promoters, chromHMM separates active transcription start site (TssA), TSS flank (TssAFlnk), bivalent TSS (TssBiv), and bivalent TSS flank (BivFlnk) states. For enhancers, genic enhancer (EnhG), enhancer (Enh), and bivalent enhancer (EnhBiv) are the different states. Human CpG island locations were extracted from UCSC genome browser (Kent et al., 2002). CpG island shores were defined as 2 kb flanking regions on both sides of CpG island. CpG island shelf was characterized as 2 kb flanking regions on both sides of CpG island shore, not including CpG island or CpG island shore. CpG island “open sea” includes all genomic regions except CpG island, CpG island shore and CpG island shelf. A custom R script was used to generate the locations of CpG islands (https:github.com/Yihui-Zhu/AutismPlacentaEpigenome).
Pyrosequencing
Genomic DNA (500 ng) was bisulfite converted using the EZ DNA Methylation kit (Zymo). Amplification and sequencing primers were designed using the PyroMark Assay Design Software 2.0 (Qiagen). DMRs were amplified using the PyroMark PCR kit (Qiagen). Pyrosequencing of 13 CpG sites at CYP2E1 gene, and 11 CpG sites in human IRS2 gene was performed in triplicate. Pyrosequencing was performed on a Pyromark Q24 Pyrosequencer (Qiagen) with the manufacturers recommended protocol. Enzyme, substrate, and dNTPs were from the Pyromark Gold Q24 Reagents (Qiagen) and the methylation levels were analyzed using Pyromark Q24 software.
CYP2E1 related DMR pyrosequencing primers:
Forward: GGTGTTTTGTTTTGGGGTTGA
Reverse: ACCCATTCAATATTCACAACAATC (5’ Biotin)
Sequencing: GGTTGATGATGGGGA
Amplification region: chr10: 133527817 – 133527938 (hg38)
IRS2 related DMR pyrosequencing primers:
Forward: TTAGGAATATAGGGAAAGGTGAAAGT
Reverse: CCACCCATTCACCCATTCTA (5’ Biotin)
Sequencing: GGGAAAGGTGAAAGTT
Amplification region: chr13: 109781623 – 109781794 (hg38)
Gene Expression in Umbilical Cord Blood
Data for gene expression assessed by Affymetrix Human Gene 2.0 array were extracted a previous publication on umbilical cord blood from subjects in the MARBLES study (GEO ID: GSE123302) (Mordaunt, Park, et al., 2018). Placenta and cord blood were collected at the same time period in the same study. Raw intensity values from cord blood samples were normalized by RMA and data from 70 male samples were extracted, including 30 ASD and 40 TD samples. Normalized expression was examined at the only probe annotated to CYP2E1 (16711001) and the only probe annotated to IRS2 (16780917). Analysis was done on those two probes with 70 samples on the normalized matrix data.
Western Blot
In Western blot experiments, placental proteins were isolated with RIPA buffer containing 10mM Tris-Cl (pH 8.0), 1mM EDTA, 1% Triton X-100, 0.1% sodium deozydholate, 0.1% SDS, 140mM NaCl, 1mM PMSF and complete protease inhibitors (ThermoFisher), incubated at 37°C for 30 minutes, sonicated and heated at 95°C for 5 min. BCA (Bicinchoninic Acid) protein assay (ThermoFisher) was used to determine protein concentration. Protein samples (20–30 ug) were resolved on 4–20% tris-gylcine polyacrylamide gels (Biorad). Proteins were separated and transferred to nitrocellulose membranes for 60 minutes at a constant voltage of 100. The membranes were blocked in Odyssey Blocking Buffer (PBS) (Licor, 927–40000) for 40 min. Anti-IRS2 (1:5,000, Cell Signaling, 3089S) and anti-GAPDH (1:10,000, Advanced Immunochemical, Inc., 2-RGM2) were incubated with the membrane with Odyssey Blocking Buffer containing 0.2% Tween overnight at 4°C. Membranes were washed with 1 X PBS (Phosphate-buffered saline) containing 0.2% Tween and then incubated with secondary antibodies, IRDye 800CW Donkey anti-Mouse IgG (1:50,000, Licor, 926–32212) and IRDye 680RD Donkey anti-Rabbit IgG (1:50,000, Licor, 926–68073) for 1 hour. Membranes were scanned using a Licor Odyssey infrared imaging system based on the manufacturer’s guidance (with resolution: 84; quality: medium, 600-channel: 6; 800-channel: 5). Relative protein quantification was done using the ImageJ software program (Rueden et al., 2017; Schneider, Rasband, & Eliceiri, 2012) in densitometry mode. IRS2 signals were normalized to GAPDH (Glyceraldehyde 3-phosphate dehydrogenase) for each sample.
Sanger Sequencing
PCR amplification was performed on each sample using PCR 10x buffer, 25 mM MgCl2, 5 M betaine, 10 mM dNTPs, DMSO, and HotStart Taq (Qiagen). Each PCR program was unique to the region being amplified with specific primers. The PCR product was then resolved by gel electrophoresis using a 1% Agarose gel in 1 X TE to later be extracted using the gel extraction kit (Qiagen) based on the default protocol. After DNA quantitation by NanoDrop, the samples were sent to the UC Davis Sequencing Facility for sequencing on the 3730 Genetic Analyzer (Applied Biosystems Prism) with DNA sequencing Analysis software v.5.2 (Applied Biosystems Prism). The sequencing results were assembled and analyzed using CodonCode Aligner version 7.0 (CodonCode).
CYP2E1 related SNP (rs943975, rs1536828) primers:
Forward: CTACAAGGCGGTGAAGGAAG
Reverse: CCCATCCCCATAAACTCTCC
IRS2 related SNP (rs943975) primers:
Forward: TTAGGAATATAGGGAAAGGTGAAAGT
Reverse: CCACCCATTCACCCATTCTA
Maternal Prenatal Vitamin Use and Timing
Maternal prenatal vitamin information and timing of maternal intake for 6 months before and each month during the pregnancy were recorded though telephone interviews and/or questionnaires as previously describes (Hertz-Picciotto et al., 2018). Mothers who took prenatal vitamins in the first month pregnancy or not were grouped into PreVitM1 Yes/No. Mothers who took prenatal vitamins beginning from 6 months to 2 months before pregnancy were grouped as “Before Pregnancy”. Mothers beginning prenatal vitamins one month before pregnancy through the second month of pregnancy were grouped as “Near Conception”. Mothers beginning prenatal vitamins from 3 months to 9 months of pregnancy were grouped as “During Pregnancy”.
Code availability
Custom scripts for WGBS analysis are available at https:github.com/kwdunaway/WGBS_Tools with the instructions. Custom Scripts for DMR finder are available at https:github.com/cemordaunt/DMRfinder with the instructions. The rest of code and scripts for each figure and tables are available at https:github.com/Yihui-Zhu/AutismPlacentaEpigenome.
Data availability
WGBS data were previously published, Gene Expression Omnibus (GEO) accession number GSE67615 (Schroeder et al., 2016). The rest of the relevant data and information are included in supplementary tables.
Competing interests
The authors declared they do not have anything to disclose regarding funding or conflict of interest with respect to this manuscript.
Acknowledgements
We would like to thank members of the LaSalle lab and UCD Children’s Center for Environmental Health for helpful discussions, and the MARBLES study participants. This work used the Vincent J. Coates Genomics Sequencing Laboratory at UC Berkeley, supported by NIH S10 Instrumentation Grants S10RR029668 and S10RR027303. This work was supported by R01 ES025574, NIH R01ES021707, NIH P01ES011269 (CCEH), EPA 83543201 (CCEH), DOD AR110194, R01ES020392 (MARBLES), U54HD079125 (IDDRC), P30ES023513 (EHSC) and NIH-UL1-TR000002 (CTSC).
Footnotes
yhzhu{at}ucdavis.edu, cemordaunt{at}ucdavis.edu, dhyasui{at}ucdavis.edu, rmarathe{at}ucdavis.edu, rlcoulson{at}ucdavis.edu, kwdunaway{at}ucdavis.edu, ckwalker{at}phs.ucdavis.edu, sozonoff{at}ucdavis.edu, iher{at}ucdavis.edu, rjschmidt{at}ucdavis.edu, jmlasalle{at}ucdavis.edu