In this study, we have demonstrated that genome-wide methylation patterns remain stable between cord blood at birth and saliva in childhood at the age of 6–12 years. Furthermore, some individuals that are identified to display an OMP at birth continue to display it in childhood. This raises the possibility that certain childhood outcomes may be predictable based on DNA methylation analysis at birth.
Prior studies have identified persistent differences in DNA methylation over time after an environmental exposure. For example, prenatal exposure to maternal smoking can be identified in differentially methylated regions (DMRs) at birth, and these findings persist into childhood and adulthood [16,17,18]. In an analysis of twenty children with respiratory allergies compared to 20 control children using the Illumina Methylation 450 K BeadChip array, Boever et al. reported that there are specific DMRs in the saliva and blood of children that are relevant to respiratory allergies and further found that those methylation patterns were already present in the children’s bio-banked cord blood [19]. Specifically, they identified 83 DMRs in the blood of the 11-year-old children, 26 DMRs in their cord blood, and 5 DMRs in their saliva; two of these DMRs were in common between all three samples and were noted to be involved in pathways previously implicated in respiratory allergy disease.
Another study performed an epigenome-wide association study (EWAS) using the Illumina Methylation 450 K BeadChip array (and target validation using the Sequenom MassArray EpiTYPER) to compare methylation differences between twelve extreme preterm birth infants (born at < 31 weeks of gestational age) and twelve term infants at birth and 18 years of age [20]. Although the authors noted that the majority of significant methylation differences identified at birth were largely resolved by 18 years of age, they identified 10 probes (of 1555) that persisted to be significantly different over time, suggesting an ‘epigenetic legacy of preterm birth.’ Furthermore, they noted a significant overlap between CpGs that were differentially methylated at birth and those that changed by age 18 years. This is concordant with our study where we noted that the R2 values of the outlier CpGs over time were lower than those of CpGs that did not meet outlier criteria. This demonstration of increased variation over time is consistent with the hypothesis that these sites are likely more susceptible to environmental influences, including during the periconception period.
In contrast, other studies have not found that epigenetic differences identified at birth persist into childhood, suggesting some mechanism of self-correction. Specifically investigating the association of preterm birth with altered DNA methylation in genes encoding insulin-like growth factor 2 (IGF2) and FK506-binding protein 5 (FKBP5), Piyasena et al. [21] noted that the DMRs in these two genes were hypomethylated in the saliva of 50 preterm infants at birth compared to 40 term infants, but these differences were no longer identified at 1 year of age. Looking at the effects of maternal (i.e., prenatal) socioeconomic status on DNA methylation using the Illumina Methylation 450 K BeadChip in 609 children, Laubach et al. found that only one of 29 DMRs identified at birth in cord blood remained apparent in peripheral leukocytes in early childhood (age 3 years) and none persisted to mid-childhood (age 7 years) [22]. Similarly, another study exploring the effect of maternal mood disorders and the prescription of antidepressant medications during pregnancy found a single CpG site (using Illumina Methylation 450 K BeadChip) that was differentially methylated in two cohorts at birth (n = 479 and n = 999); in the smaller cohort, this difference persisted in early childhood (n = 120) but attenuated by mid-childhood (n = 460) [23].
Given these attenuations, it is prudent to evaluate whether this apparent loss of DNA methylation differences over time is simply due to age-related changes. Although one can calculate or predict an individual’s age based on epigenetic clocks derived from DNA methylation data, age-related DNA methylation changes are small. Prior studies have observed incremental, 0.1% per year, age-related DNA methylation changes at ~ 15–20% of CpG sites [24,25,26,27,28]. For example, in an EWAS of infant saliva between 6 and 52 weeks of age, Wilkenius et al. [29] demonstrated that only 42 genes in a total of 101 out of 423,315 probes demonstrated statistically significant DNA methylation changes over that time frame. This is consistent with our findings of a high correlation of DNA methylation over time across all CpG sites, age-related CpG sites (Table 1), and outlier CpG sites (Fig. 2).
This persistence of significant methylation differences in our study suggests that a methylome disrupted during the periconceptual period is like to continue to be disrupted in childhood, and potentially into adulthood. As we have previously reviewed [30], multiple studies have found that IVF-related procedures (e.g., superovulation and intra-cytoplasmic sperm injection) and manipulation of laboratory conditions (e.g., oxygen tension, temperature, humidity, and pH) can lead to alterations in DNA methylation in offspring. Furthermore, some, but not all prior studies, have identified DNA methylation differences between ART-conceived and non-ART-conceived children. We suspect that these inconsistent results may be due to both significant heterogeneity between the studies and the variable presence of individuals who are ‘epigenetic outliers’ similar to those identified in this study.
Of additional interest, we noted that families clustered together except in the case of the one member of a dizygotic twin-pair who demonstrated on OMP (Child Jb). As these were dizygotic twins, we expect them to behave more likely siblings than monozygotic twins, which suggests that there is something inherent to that individual that contributed to them being an outlier. Furthermore, an increasing number of studies have demonstrated that multiple factors, including genetic, environments, stochastic, and age/time, affect epigenetic variation and the resulting phenotypes, including in twins [31].
The major strength of this study is the prospective collection and longitudinal analysis of cord blood at birth and saliva in early childhood. With regards to identifying outliers, we used external cord blood datasets to confirm that individuals consistently demonstrated an OMP. We also reconfirmed the finding that DNA methylation patterns remain stable over time using externally identified age-related CpGs.
A potential limitation is that different tissues were used for analysis at birth and in childhood. It has been well characterized that ~ 15% of CpG sites exhibit tissue or cell-type-specific DNA methylation patterns. As such, differences in the cell composition of cord blood and saliva may contribute to differences in methylation patterns seen. Nevertheless, prior studies have shown that there is a high correlation in DNA methylation profiles between saliva and peripheral blood in adults and adolescents [12, 32, 33]. In the current study, deconvolution analyses using cell-type-specific differentially methylated regions to infer cell-type proportions (based on 8 CpG sites as described in [34]) demonstrated that the majority of the isolated cells from both saliva and blood were leukocytes (Additional file 4: Fig. S2). Despite this, we cannot completely exclude the possibility that ‘outlier’ patterns may be due to cell-composition differences.
Other limitations are as follows. First, the overall sample size remains small. The COVID-19 pandemic limited our ability to bring children from the cohort back in for follow-up visits in childhood. Future work will focus on the inclusion and analysis of more children from the original (and ongoing) large cohort.
Second, all but two of the children analyzed in this study were conceived via ART. Although we cannot comment from this study whether or not it was any specific ART exposure that led to the outlier phenotypes, our prior analyses have demonstrated that children conceived via ART are more likely to demonstrate an OMP than children conceived without assistance [11]. We included the unassisted conceptions in our analyses for several reasons; 1) to see if they demonstrate similar stability over time to the ART-conceived children (which they do), and 2) to maximize the number of children for whom paired samples were available. Furthermore, our sensitivity analyses identified that even after excluding the children not conceived via ART, the same three individuals demonstrate an OMP.
Third, while the majority of our study population had parents who identify as White Non-Hispanic, two out of the three children who demonstrated an OMP at birth had parents who identified as Black/African-American. We can thus not exclude that there may be racial/ethnic differences that contribute to DNA methylation differences. Furthermore, several children were unable to be included in the paired analyses due to lack of either their cord blood or their salivary sample. However, all available samples were used for calculating outlier status. Given the high correlations seen in all of the pair analyses, this is unlikely to have affected the results, and the additional data only strengthens the OMP status of the identified individuals. Finally, we lack a functional assessment of the differential methylation patterns on gene expression. Therefore, we cannot definitively conclude that the gene pathways identified are responsible for any differences in childhood diseases or other outcomes.
