In this exploratory study, we examined DNA methylation profiles associated with very preterm birth (<31 weeks of gestation) using longitudinally collected blood sampled from newborns and 18-year-olds, both stored as dried blood spots. We identified methylation differences in birth samples at several gene loci previously reported to co-vary with gestational age. These results demonstrate the utility of archived birth blood spot DNA for methylation profiling, in accordance with other recent studies [48, 49, 52].
We found widespread differences in DNA methylation at birth in preterm infants compared to with term controls. In agreement with previous studies [23, 28, 58], our data revealed methylation changes in blood associated with age. Some of these methylation differences are likely to reflect cell composition or functional differences in blood between preterm and term neonates, and between birth and at 18 years of age. For example, preterm-associated methylation differences at birth coincide with hematological changes that are correlated with gestational age, such as leukocyte [59, 60] and nucleated reticulocyte  content. Further, gestational and age-related methylation changes may also reflect maturation of blood cells, including innate cytokine and adaptive immune induction [62–64]. Methylation change driven by these effects is evidenced by over-representation of birth DMPs and age DMPs in gene networks involved in hematological system development (see Additional file 4: Table S3; see Additional file 6: Table S5; see Additional file 7: Table S6).
Blood cell composition is well established as a predictive factor for inter-individual methylation variation in whole blood [65, 66]. However, this variable does not readily explain the significant overlap in differentially methylated CpGs between preterm and term subjects at birth and also between birth and 18 years of age. We found 159 sites that showed increased methylation and 141 that showed decreased methylation from preterm to term birth and subsequently up to adulthood, suggesting a continuum of methylation change from mid-gestation to young adulthood for some regions of the genome (Figure 2). Genes associated with these sites were over-represented by direct targets of the upstream transcription factors NLRC5 and CIITA, master regulators of the MHC I-dependent [67–69] and MHC II-dependent [70, 71] immune responses, respectively. These observations are consistent with gestational and post-natal changes to the immune system, during which time MHC responses are initiated . We also found that genes associated with a continuum of methylation change were over-represented in embryonic development/morphogenesis and nervous system development, consistent with previous reports [73, 74]. Teschendorff and co-workers have previously reported a correlation between age-associated loss of methylation and blood composition, but not with age-associated increases in methylation . However, our data cannot disentangle cell-type effects. Other contributing factors may include developmental processes such as age-related changes to the progenitor cell pool [75–77], as suggested previously , or age-related shifts in blood cell signaling and metabolism .
We identified 431 CpG sites that changed in the opposite direction from preterm to term birth compared with birth to 18 years of age (Figure 2). Previous studies in human prefrontal cortex illustrate non-linear and directional changes in DNA methylation  and gene expression  during aging, suggesting that such changes are part of normal development. Our data defining CpG methylation sites that change direction during gestation and subsequently during post-natal life may reflect a distinctive methylation signature in the blood of term neonates. We speculate that these observations may reflect cell composition or functional differences in blood cells that are evidently unique to neonates born at full term [62, 80, 81].
We found very similar genomic contexts enriched in birth DMPs and age DMPs. Both showed over-representation of rDMRs and under-representation of CGIs and promoter regions. These results are consistent with findings from cross-sectional studies in adult mice  and humans [25, 74], and with longitudinal studies of early post-natal life [23, 83, 84]. Taken together, these data indicate that similar regions of the genome are preferentially subject to epigenetic change during the second half of gestation, and during post-natal life in blood, and that these sites overlap rDMRs associated with in vitro pluripotency reprogramming. Although our genomic context enrichment data on differential methylation at 18 years used nominally significant methylation differences, our results suggest that inter-individual differences in methylation are more likely to occur in regions of the genome not associated with aging. These findings also suggest that gestation-related and age-related changes are unlikely to relate to ‘epigenetic noise’ . However, we cannot determine if these observations reflect inter-individual blood composition differences, inter-individual DNA methylation variation, or associations with preterm birth.
At 18 years of age, most methylation differences identified in preterm babies are resolved, as evidenced by the lack of genome-wide significance in differential methylation at this time point. This is consistent with our conclusion that developmental changes and cell composition are the main components of methylation variation associated with birth DMPs and age DMPs. However, comparing preterm and term group analysis of birth and 18-year samples identified 109 statistically significant DMPs. Interestingly, eight of these CpG sites were not significantly differentially methylated at birth, suggesting that a larger sample size may indeed detect a long-term epigenetic legacy of preterm birth at a single time point. We observed persistently altered CpG methylation at PCSK9, TRIM71, SLC44A4, GPC6, and NFYA gene loci and at two intergenic CpG sites flanking a binding site for the EGR1 transcription factor. Taken together, these observations raise the possibility that persistent DNA methylation differences reflect a long-term legacy of preterm birth.
Limitations of the study include confounding factors related to inter-individual variation in blood composition, which may restrict power to detect birth DMPs and age DMPs. Our exploratory study requires replication in a larger cohort. This is particularly important to confirm the persistent epigenetic legacy of preterm birth identified in this report. Use of term-equivalent samples from preterm subjects would be useful in this context. Furthermore, statistical methods for deconvoluting mixed cell types  or adjustment for age  have not been described in context of gestation or neonatal development. Therefore, further studies addressing methylation differences in sorted cells during gestation, at full-term birth, and later in life may provide empirical data necessary to account for these confounders, as suggested previously by Houseman and colleagues .