Introduction

Alcohol consumption in pregnancy is the most common preventable cause of neurodevelopmental impairments in children (1). Alcohol can pass through the placenta acting as a teratogen in fetal tissues causing physical, cognitive, behavioural, and neurodevelopmental impairment in children at high doses with lifelong consequences for health. Fetal Alcohol Spectrum Disorder (FASD) and Fetal Alcohol Syndrome (FAS) can arise at binge levels of exposure, although not always at lower levels of exposure. Whether PAE is sufficient to induce overt physiological abnormalities depends on multiple environmental and genetic factors including the dose and timing of alcohol use during pregnancy, maternal diet, smoking, stress, and potentially other factors that collectively influence fetal outcomes (2–4).

Patterns of alcohol consumption in pregnancy vary, but epidemiological surveys suggest most women in Western countries consume moderate to high levels between conception until recognition of pregnancy (5), after which time consumption largely ceases, apart from occasional use (6). While the effects of binge-levels of exposure are well documented to cause FASD, more subtle effects that reflect the more common patterns of drinking are unclear and more research is needed to support public health initiatives to reduce alcohol consumption in pregnancy.

Studies suggest alcohol can disrupt fetal gene regulation through epigenetic mechanisms. DNA methylation is one epigenetic mechanism involving the catalytic addition of methyl groups to cytosines bases within cytosine-guanine (CpG) dinucleotide motifs during one-carbon metabolism. Methylation of DNA can alter chromatin density and influence patterns of gene expression in a tissue-specific and developmentally appropriate manner and disruption to this process may cause of some of the difficulties experienced by people with FASD (7, 8). Previous studies on human participants (9–11) and animals (12–14) report alcohol can disrupt DNA methylation either globally (11–13), and/or at specific gene regions (10, 13, 14). Our recent systematic review however found limited replication of effects between studies suggesting the effects of alcohol on DNA methylation may be stochastic and influenced by numerous confounding factors (15). PAE can either directly inhibit DNA methyltransferase enzymes or disrupt one-carbon metabolism via inhibition of bioavailability of dietary methyl donors, such as folate and choline to the fetus (16) (17). Choline in particular has been explored in several clinical trials to reduce cognitive deficits caused by PAE in affected individuals (18, 19), or when administered during pregnancy (20, 21), with encouraging results.

Given the lack of clarity around the effects of typical patterns of alcohol consumption, which often do not cause observable phenotypes, we conducted an epigenome-wide association study of early moderate PAE in mice with replication in human cohorts. The study was a controlled intervention investigating the impact of PAE on offspring DNA methylation comparing exposed and unexposed mice, with an additional arm comparing the effect of alcohol exposure in the context of a high methyl donor maternal diet. The exposure period covers the equivalent of pre-conception up until the first trimester in humans when neurulation occurs, reflecting a typical situation in which women may consume alcohol up until pregnancy recognition (22). The primary outcome of the study was changes in offspring DNA methylation and secondary outcomes of behavioural deficits across neurodevelopmental domains relevant to FASD were also examined. We employed whole genome bisulfite sequencing (WGBS) for unbiased assessment of the epigenome in newborn brain and liver, two target organs affected by ethanol (23), to explore tissue-specificity of effects and to determine any ‘tissue agnostic’ effects which must have arisen prior to the germ-layers separating in early gastrulation.

Results

Comparison of prenatal characteristics across treatment groups

To investigate the effects of PAE and HMD on offspring DNA methylation and behavioural outcomes, we employed a murine model with four treatment groups (Figure 1). The trajectory of weight gain during pregnancy was similar across all treatment groups with some evidence of more rapid weight gain in the HMD groups in the last 2 days (Linear mixed effects regression model; H₂0-HMD: –2.282 ± 0.918 g, P = 0.0177; PAE-HMD: –1.656 ± 0.814 g, P = 0.0493; figure S1a), although the total amount of weight gained between days 1 and 17-19 was not significantly different between treatment groups by linear mixed effects regression (Figures S1a-b). The total amount of liquid consumed over the course of pregnancy was significantly lower in HMD dams by unpaired t-test (Figures S1b and S2c). There was no significant difference in the average litter size (6.525 ± 0.297 pups, figure S1c) and pup sex ratios (Figure S1d) between treatment groups by unpaired t-test.

Figure 1.

Effects of PAE and HMD on behavioural outcomes in adult mice

Behavioural testing was carried out on adult mice from each study group. There was no evidence that PAE had a significant effect on any of behavioural outcomes tested (Figure S3). Mice exposed to HMD exhibited greater locomotor activity, in terms of distance travelled (Figure 2).

Figure 2.

Effects of PAE on DNA methylation in newborn mice tissue

Whole-genome bisulfite sequencing (WGBS) of brain and liver samples from 16 newborn pups was performed to determine the effects of PAE on fetal DNA methylation. Global levels of DNA methylation stratified across different genomic features were preserved across treatment conditions, with no major differences in average methylation content between groups (Figure S4). To investigate region-specific effects of PAE on newborn DNA methylation, we conducted genome-wide testing comparing exposed and un-exposed mice on the normal chow. We identified 78 differentially methylated regions (DMRs) in the brain and 759 DMRs in the liver (P < 0.05 and delta > 0.05) from ∼19,000,000 CpG sites tested after coverage filtering (Figure 3a-b). These regions were annotated to nearby genes using annotatr and are provided in Tables S1-2. Two of the DMRs overlapped in mouse brain and liver (tissue agnostic), but the remainder were tissue specific. Among these tissue agnostic regions was the Impact gene on chromosome 18, which was hypomethylated in PAE+NC mice compared to H₂0+NC mice in both the brain and liver (Figure 3c-d). The other tissue agnostic region was within 5kb downstream of Bmf and was hypermethylated in brain and liver tissue of PAE+NC mice.

Figure 3.

In both tissues, hypomethylation with PAE in NC mice was more frequently observed in DMRs (52.6% of DMRs in brain, 93.5% of DMRs in liver hypomethylated with PAE) compared to hypermethylation. Some DMRs localised to the same genes in both brain and liver, although they were different regions. The three genes affected by PAE in both brain and the liver tissues were the Autism Susceptibility Gene 2 (Auts2), Androglobin (Adgb), and RNA Binding Protein Fox 1 (Rbfox1) genes (Table 1). In both brain and liver tissues, DMRs were enriched in non-coding intergenic and open sea regions and relatively underrepresented in coding and CpG island regions (Figure 3e-f). Using open chromatin assay and histone modification datasets from the ENCODE project, we found overwhelming enrichment (p < 0.05) of DMRs in open chromatin regions (ATAC-seq), enhancer regions (H3K4me1), and active gene promoter regions (H3K27ac), in mouse fetal forebrain tissue and fetal liver (Table 2). Gene ontology enrichment analysis of liver DMRs that did localise to genes showed enrichment in ten predominantly neuronal pathways, with neuron projection being the most significant (Figure 3g, Tables S3-4).

Table 1:

Table 2.

HMD mitigates the effects of PAE on DNA methylation

To determine whether administration of a HMD throughout pregnancy could mitigate the effects of PAE on offspring DNA methylation, we examined alcohol sensitive DMRs in the HMD mice. Compared to control mice, PAE+HMD mice exhibited methylation differences in only 12/78 (15%) brain (Table S7), and 124/759 (16%) liver (Table S8) DMRs, suggesting the effects were predominantly mitigated. Effect sizes compared to mice on the normal chow were substantially lower, in some cases more than 25% reduced in mice on the high methyl donor diet (Figure 4).

Figure 4.

Replication studies in Human PAE and FASD case-control cohorts

We undertook validation studies in human cohorts to address the generalizability of our murine model of PAE. Only 36 of the 78 (46.2%) brain DMRs, and 294 of the 759 (38.8%) liver DMRs, had homologous regions in the human genome and were able to be tested. DNA methylation array data from 147 newborns buccal swabs from the Asking Questions About Alcohol in Pregnancy (AQUA) cohort (40) was available for analysis. We tested a total of 1,898 CpG sites that corresponded to mouse DMRs, comparing ‘never exposed’ newborns to ‘any exposure’ and found no evidence of differential methylation at these CpG (data not shown). We also accessed publicly available DNA methylation array measurements from buccal swabs taken from a Canadian clinical cohort of children with diagnosed FASD, and controls. To avoid confounding due to ancestry we analysed the 118 Caucasian individuals (30 FASD and 88 controls). Testing a total of 2,316 CpG sites that were homologous to mouse DMRs we statistically replicated 7 DMR associations with FASD status (FDR P < 0.05) after adjusting for participant age, sex, array number, and estimated cell counts (Table 3). Visual comparison of methylation changes across these seven DMRs revealed striking differences in effect sizes between people with FASD and mice (Figure 5). Genes associated with these DMRs are linked to clinically relevant traits in the GWAS catalogue including facial morphology (GADD45A (41)), educational attainment (AP2B1 (42), intelligence (RP9 (43)), autism and schizophrenia (ZNF823 (44)).

Figure 5.

Table 3.

Candidate Gene Analysis of previously defined alcohol sensitive regions

In candidate gene studies there were 21 CpG sites (FDR<0.05) identified in the brain from 15,132 CpG sites tested, including two sites in the Mest (Peg1) gene and 19 sites in Kcnq1 (KvDMR1) (Table S9). There were nine FDR-significant CpG sites identified in the liver out of 15,382 CpG sites tested, all of which were in Peg3 (Table S10). All FDR-significant CpG sites identified in both tissues were hypomethylated in mice with PAE.

Discussion

In this study, found that moderate early (first trimester) PAE was sufficient to induce site-specific differences to DNA methylation in newborn pups without causing overt behavioural outcomes in adult mice. Global levels of DNA methylation were not significantly different with PAE, and effects were characterized predominantly by a loss of methylation (hypomethylation), mostly at non-coding regions of the genome. In our model, alcohol’s effects on DNA methylation were predominantly tissue-specific, with only two genomic regions and four genes that were similarly affected in both tissues. These perturbations must have arisen before the germ layers separated suggesting alcohol can perturb methylation events as early as gastrulation. In general, DMRs were enriched in non-coding regions of the genome with regulatory potential suggesting alcohol may have broad effects on genome regulation.

Both the human validation studies and the candidate gene analysis provide validity to our model for recapitulating some of the genomic disturbances reported in patients with clinical FASD. It is remarkable that associations were reproduced despite differences in biosamples and species and suggests that at least some methylation changes are stable over time. We replicated associations from published reports of hypomethylation within Peg3 and KvDMR1 from South African children with fetal alcohol syndrome (10). Both these genes are methylated in a parent-of-origin specific manner, suggesting that alcohol may affects imprinting processes, although results are not entirely consistent (45, 46). On the balance of this, we speculate duration of exposure, dose, and other tissue-related factors all likely influence the extent to which genome-regulation is perturbed and manifests as differences in DNA methylation.

Our results are encouraging for biomarker studies and aid in the prioritisation of associations for future follow-up, particularly in relation to diagnosis of FASD. For example, three genes are zinc finger proteins (RP9, PEX12, and ZNF823) that play an important role in fetal gene regulation. Notably, PEX12 is associated with Zellweger syndrome, which is a rare peroxisome biogenesis disorder (the most severe variant of Peroxisome biogenesis disorder spectrum), characterized by neuronal migration defects in the brain, dysmorphic craniofacial features, profound hypotonia, neonatal seizures, and liver dysfunction (47).

Another key finding from this study was that HMD mitigated some of the effects of PAE on DNA methylation. Preclinical studies of choline supplementation in rodent models have similarly reported attenuation of memory and behavioural deficits associated with PAE (48, 49), and mitigating effects on DNA methylation (50). These data have been largely consistent and collectively support the notion that alcohol induced perturbation of epigenetic regulation may occur, at least in part, through disruption of the one-carbon metabolism. The most encouraging aspect of this relates to the potential utility for evidence-informed recommendations for dietary advice or supplementation, particularly in population groups with limited access to antenatal care or healthy food choices.

Strengths of this study include the use of controlled interventions coupled with comprehensive assessment of the effects of PAE on multiple tissues. Caveats include a limited ability to determine the contribution of specific cell types within tissues to the methylation differences observed, and we did not assess markers of brain or liver physiology.

In conclusion, this study demonstrates that early moderate PAE can disturb fetal gene regulation and supports current public health advice that alcohol consumption during pregnancy, even at low doses, may be harmful.

Materials and Methods

Murine subjects and housing

To study the effects of PAE on offspring DNA methylation processes, we adapted a murine model study design that has previously reported DNA methylation changes at the A^vy locus in Agouti mice (24) (Figure 1). This study received animal ethics approval from the Telethon Kids Institute Animal Ethics Committee (Approval Number: 344). Sixty nulliparous C57BL/6J female mice aged ∼8 weeks were mated with equivalent stud male mice. Pregnant dams were randomly assigned to one of four treatment groups (n = 15 dams per group) that varied based on composition of the drinking water and chow given to the dams:

1. PAE-NC (Prenatal Alcohol Exposure-Normal Chow): 10% (v/v) ethanol in non-acidified water ad libitum from 10 days prior to mating until gestational days (GD) 8-10. This is intended to replicate typical patterns of drinking during the first trimester of pregnancy in humans. Dams received non-acidified reverse osmosis water for the remainder of pregnancy and normal chow (Rat and Mouse Cubes, Speciality Feeds, Glen Forrest, Australia) throughout pregnancy.

2. PAE-HMD (Prenatal Alcohol Exposure-High Methyl Donor diet): 10% (v/v) ethanol in non-acidified water ad libitum from 10 days prior to mating until GD8-10 and non-acidified reverse osmosis water for remainder of pregnancy. Isocaloric high methyl donor (HMD) chow consisting of 20 mg/kg folate and 4,970 mg/kg choline throughout pregnancy (Speciality Feeds, Glen Forrest, Australia).

3. H₂0-NC (Water-Normal Chow): non-acidified water and normal chow throughout pregnancy.

4. H₂0-HMD (Water-High Methyl Donor diet): non-acidified water and HMD chow throughout pregnancy.

Whole-genome bisulfite sequencing of newborn mouse tissues

Pups selected for WGBS in each intervention group were matched on sex and litter size to minimize variability in exposure. Two male and two female pups per treatment group (n = 16 total) were euthanised by intraperitoneal injection with ketamine and xylazine on the day of birth for WGBS of their brain and liver tissues. Mouse tissue samples were stored at –80°C. Remaining littermates grew until adulthood for behavioural testing. Ten milligrams of tissue were collected from each liver and brain. Total nucleic acid was extracted from the tissues using the Chemagic 360 instrument (PerkinElmer) and quantified with Qubit DNA High Sensitivity Kit (Catalogue Number: Q32854, Thermo Scientific). 100 ng of genomic DNA was spiked with 0.5 ng of unmethylated lambda DNA (Catalogue Number: D1521, Promega) to assess the bisulfite conversion efficiency. Each sample was digested with 2 µl RNase A (Invitrogen) at 37°C for 20 minutes to remove RNA. 100 ng of genomic DNA from each sample was sheared using a Covaris M220 (300bp settings, Covaris). Libraries were prepared using the Lucigen NxSeq AmpFREE Low DNA Library Kit (Catalogue Number: 14000-1, Lucigen), according to the manufacturer’s instructions. Nextflex bisulfite-seq barcodes (Catalogue Number: Nova-511913, PerkinElmer) were used as the adapters with incubation at 25°C for 30 minutes. The libraries were bisulfite converted using the Zymo EZ DNA Methylation-Gold Kit (Catalogue Number: D5005, Zymo Research) and PCR amplified using the KAPA HiFi Uracil PCR Kit (Catalogue Number: ROC-07959052001, Kapa Biosystems). The final libraries were assessed with the Agilent 2200 Tapestation System using D1000 Kit (Catalogue Number:5067-5582). WGBS was performed by Genomics WA sequencing core on a NovaSeq 6000 (Illumina) using 2×150bp chemistry on an S4 flow cell. The bisulfite conversion rate in each tissue sample was at least 99%. The overall mean coverage in each sample was 9.69× (range: 6.51-12.12×).

Behavioural testing in adult mice

Littermates who were not sacrificed at birth were reared on normal chow and drinking water ad libitum until adulthood (∼8 weeks after birth) when they underwent behavioural tests assessing five neurodevelopmental domains that can be affected by PAE including locomotor activity, anxiety, spatial recognition, memory, motor coordination and balance. These tests included the open field test (locomotor activity, anxiety) (25), object recognition test (locomotor activity, spatial recognition) (26), object in place test (locomotor activity, spatial recognition) (27), elevated plus maze test (locomotor activity, anxiety) (28), and two trials of the rotarod test (motor coordination, balance) (29). Between mouse subjects, behavioural testing equipment was cleaned with 70% ethanol. Video recording was employed for all behavioural tests, except for the rotarod, and the assessment process was semi-automated using ANY-maze software (Stoelting Co., Wood Dale, Illinois, U.S.A.).

Statistical analysis

Dam characteristics and pup behavioural testing results were generally assessed using unpaired t-tests comparing each treatment group to the baseline control group that was given non-acidified reverse osmosis water and normal chow throughout pregnancy. Trajectories of liquid consumption and weight gain across pregnancy, which were assessed using a quadratic mixed effects model and the trajectory of chow consumption across pregnancy which was assessed using a linear mixed effects model. Raw fastq files were mapped to the mm10 mouse reference genome with BSseeker 2 (version 2.1.8) (30) and CG-maptools (version number 0.1.2) (31) using a custom bioinformatics pipeline. CGmap output files were combined as a bsseq object in the R statistical environment (32). We filtered the X chromosomal reads and then combined reads from mice in the same treatment group using the collapseBSseq function, to maximise coverage prior to differential methylation analysis. CpG sites with an aggregated coverage below 10× in each tissue type were removed prior to modelling to ensure there was sufficient coverage in all assessed CpG sites. This retained 94.9% of CpG sites in the brain and 93.8% of CpG sites in the liver. Differentially methylated regions (DMRs) were identified within each tissue using a Bayesian hierarchical model comparing exposed and unexposed groups using the Wald test with smoothing, implemented in the R package DSS (33). We declared DMRs as those with P-value < 0.05 based on the p-values of each individual CpG site in the DMR, and an effect size (delta) > 0.05. Gene ontology analysis was performed on the brain and liver DMRs using the Gene Set Enrichment Analysis computational method (34) to determine if the DMRs were associated with any transcription start sites or biological processes. Brain and liver DMRs were tested for enrichment within ENCODE Project data sets (35) by an overlap permutation test with 100 permutations using the regioneR package. The ENCODE Project data sets that were assessed included ENCFF845WSI, ENCFF764NTQ, ENCFF937JHP, ENCFF269TLO, ENCFF676TSV and ENCFF290MLR. DMRs were then tested for enrichment within specific genic and CpG regions of the mouse genome, compared to a randomly generated set of regions in the mouse genome generated with resampleRegions in regioneR, with equivalent means and standard deviations. We compiled a set of key genes and genomic regions identified in previous mammalian PAE studies for site-specific testing based on our prior systematic review of the literature (15), which identified 37 candidate genes (Tables S5-6). The brain and liver datasets were filtered to candidate gene regions and differential testing was then performed across the entire coding sequence, separately in the brain and liver of the mice on a normal diet using the callDML feature in DSS.

Validation studies in human cohorts

Validation studies in human cohorts with existing genome-wide DNA methylation data sets and matching PAE data are described in the Supplementary Material. Briefly, Illumina Human Methylation array. iDAT files were pre-processed using the minfi package (36) from the Bioconductor project (http://www.bioconductor.org) in the R statistical environment (http://cran.r-project.org/, version 4.2.2). Sample quality was assessed using control probes on the array. Between-array normalization was performed using the stratified quantile method to correct for Type 1 and Type 2 probe bias. Probes exhibiting a P-detection call rate of >0.01 in one or more samples were removed prior to analysis. Probes containing SNPs at the single base extension site, or at the CpG assay site were removed, as were probes measuring non-CpG loci (32,445 probes). Probes reported to have off-target effects in McCartney et al. (37) were also removed. Mouse DMRs were converted into human equivalent regions using an mm10 to hg19 genome conversion with the liftover tool in the UCSC Genome Browser (38). A minimum 0.1 ratio of bases that must remap was specified as recommended for liftover between regions from different species and multiple output regions were allowed. Differential testing of candidate mouse DMRs was carried out using the R package DMRcate (39) for each dataset and DMRs were declared as minimum smoothed false-discovery rate (FDR) < 0.05.

Data Availability Statement

The mouse whole-genome bisulfite sequencing data will be available on the GEO repository at http://ncbi.nlm.nih.gov/geo. The human array data used in the validation study is not publicly available due to privacy and ethical restrictions.

Conflict of Interest Statement

The authors declare no conflicts of interest.

Author Contributions

D. Martino, M. Symons, A. Larcombe, R. Lister, D. Hutchinson, E. Muggli, J. Craig, J. Halliday, J. Fitzpatrick, S. Buckberry, M. Bestry and E. Elliott contributed to the study design and funding application. E. Chivers, A. Larcombe performed the mouse work including administering the mating, treatments, measurements, monitoring and extracting biological samples. E. Chivers, A. Larcombe and M. Bestry performed the mouse behavioural testing. C. Bakker analysed the videos from the mouse behavioural testing. M. Bestry prepared the whole-genome bisulfite sequencing libraries and performed the data analyses. N. Kresoje assisted with preparing whole-genome bisulfite sequencing libraries. S. Buckberry and D. Martino provided advice and support on the data analysis. J. Halliday and E. Muggli contributed human datasets for reproducibility analysis. M. Bestry and D. Martino drafted the manuscript. M. Symons, A. Larcombe, R. Lister, D. Hutchinson, E. Muggli, J. Craig, J. Halliday, J. Fitzpatrick, S. Buckberry, E. Elliott and N. Kresoje contributed to the development and editing of the manuscript.

Acknowledgements

We wish to acknowledge the assistance of Dr Jahnvi Pflueger who provided training on preparation of whole-genome bisulfite sequencing libraries. We wish to acknowledge the financial contribution of the Centre for Research Excellence in FASD who supported the murine experiments.