Individuals in the NZ VLBW cohort are now in their fourth decade and at increased risk of cardiovascular disease and other health sequelae arising from their difficult start in life. Within NZ there is no systematic monitoring of the ongoing health and wellbeing of this population group, nor any mechanism to trigger interventions to prevent the onset of adverse health outcomes. The goal of the current study was to identify methylation markers in neonatal DNA that may be predictive of health outcomes in VLBW adults. This is, to our knowledge, the first report where altered methylation at birth in VLBW infants has been associated with health outcomes as adults, particularly their cardiovascular health. This was consistent with finding that the Cardiac Hypertrophy Signaling pathway was the canonical pathway most highly enriched for CpGs with altered methylation in VLBW neonates. This study is also the first genome-wide screen of DNA methylation sites in adulthood with sufficient statistical power to detect differences between VLBW individuals and controls, highlighting differentially methylated clusters within the EBF4 and HIF3A genes. Moreover, the observation that methylation at EBF4 CpGs reversed direction between birth and 28 years and was associated with adult outcomes leads us to propose that dynamic methylation of EBF4 may be associated with the trajectory of compensatory growth and development from birth to adulthood.
We noted many commonalities between our data and that of previous studies of birthweight and methylation, despite differing study designs, such as comparing pre-term versus full-term birth rather than birthweight per se, sampling adipose tissue, saliva or cord blood versus peripheral blood, assaying methylation at various time points between birth and middle age, or using differing methylation assay platforms. The finding of analogous methylation sites to our study was particularly striking in a report from Gillberg et al. [59], who examined adult adipose tissue from low- versus normal- birthweight men after 5 days of high-fat diet versus a control diet. They observed 53 CpGs located on 40 genes where methylation differed between low-birthweight men and controls, and of these 40 genes, 21 also featured within our long list of differentially methylated CpGs in neonates (C17orf97, KAZALD1, SORBS2, DPP10, CPLX1, CACNA2D2, FADS2, LOC100271832, CUGBP2, TTYH3, HCCA2, PTPRN2, ACAT1, C7orf50, CASZ1, IGF2R, PDLIM4, CREB3L2, ARHGAP23, ARID1B, and UPK3A).
Findings convergent with the current study can also be seen in the report from Simpkin et al [51], who performed a longitudinal study of DNA methylation profiles from cord blood and peripheral blood at birth and at ages 7 and 17 years in over 900 children from the Avon Longitudinal Study of Parents and Children (ARIES). They identified 224 CpG sites associated with gestational age (predominantly a negative association, as we also found), and 23 CpG sites associated with birthweight. Their supplementary list of probes where cord blood methylation was negatively associated with gestational age in ARIES showed a remarkable overlap with our most significant CpGs in VLBW neonates, including cg02001279 in ARID3A; cg07835443in C16orf55 (SPATA33); cg11932158 in PLCH1; cg04347477 in NCOR2; cg18623216 in PLCH1; cg12713583 in ARID3A. Some of the same CpGs were also identified by Cruikshank and co-workers [49], who compared 12 extremely pre-term cases and 12 matched controls with DNA from archived blood spots collected from neonates and the same individuals at age 18 years. Although in 18-year-olds no sites achieved genome-wide significance, there was extensive overlap between those CpGs reported by Cruikshank[49] to be different between pre-term and term birth and our top CpGs in neonates, including in C16Orf55 (SPATA33), ARID3A, DOCK6, NCOR2, SLC9A3R2, PLCH1 and FBLN7.
A meta-analysis of epigenome-wide association studies with birthweight by Küpers et al.[35], included 8,825 neonates from 24 birth cohorts in the Pregnancy And Childhood Epigenetics Consortium and demonstrated that lower birthweight, even within the normal range, is related to altered DNA methylation at 914 sites. In additional analyses in 7,278 participants, < 1.3% of birthweight-associated differential methylation was also observed in childhood and adolescence, and this study could find no associations in adulthood (30–45 years). The top hits for association with birthweight included some minor overlap with gene families we observed associated with VLBW in neonates, such as ARID5B ARHGAP20, ARHGAP29, and ARHGAP45. However, Küpers and colleagues [35] also identified 147 birthweight-related methylation quantitative trait loci (mQTL, where genetic variants are associated with methylation), which included 23 CpGs from our neonate long list, most notably cg06870470 in DOCK6, one of our most significant CpGs at birth.
Other prior studies that did not replicate the current findings include a report from Tan et al.,[60] who carried out DNA methylation profiling of pre-term birth in 144 adult twins with a median age of 33 years (with 26 twin pairs of premature birth). They found three genomic regions associated with pre-term birth annotated to the SDHAP3, TAGLN3 (both hypomethylated) and GSTT1 (hypermethylated) genes. While none of these CpGs reached genome-wide significance individually, combining multiple CpGs in each locus led to the clusters attaining significance. These differentially methylated regions replicated in an older, independent set of 175 twin pairs (median age 66 years) with 40 twins classified as preterm birth (at least 3 weeks before term). Finally, Wheater et al. [61] investigated the impact of low gestational age at birth on methylation patterns in neonatal saliva samples and also documented associations of methylation with brain white matter structure by diffusion magnetic resonance imaging of these infants. The most significant probes associated with gestation in the Wheater study showed no apparent overlap with CpGs identified in neonates in this current study, possibly related to their use of saliva samples versus peripheral blood.
In this current study, we observed age-related changes in methylation at EBF4. Between birth and 28 years, the CpG that showed the greatest reversal in direction of altered methylation in VLBW cases compared to controls (EBF4 cg16426670) was associated with cardiovascular traits in adulthood. The EBF gene belongs to a family of transcription factors associated with B-lymphocyte maturation and neuronal development, and also has a role in governing the differentiation of cardio-pharyngeal mesoderm into heart versus pharyngeal muscle fate [62]. Methylation of members of this gene family have been associated with a number of phenotypes, including EBF4 methylation associated with hematopoiesis and neuronal development in persons with Trisomy 21 [63], and EBF1 and EBF3 methylation to neurobehavioral development in very pre-term infants [64]. Our observation of dynamic methylation of EBF4 between birth and adulthood extends the findings of Simpkin and colleagues [51], who reported dynamic EBF4 methylation in children up to 7 years of age. Their analysis of serial methylation from birth to adolescence suggested that methylation differences did not persist beyond early childhood, with the authors suggesting that methylation levels had largely stabilized by age 7. Notably, however, that study found that among 36 probes with increased methylation per week of gestation, the probe with the maximum increase was cg16426670 in EBF4, showing 6.7% increase per year between birth and 7 years, the same CpG probe that we found to switch from hypomethylation in VLBW neonates to hypermethylation in VLBW adults compared with controls.
A recent study by Long et al. [65] investigated associations between methylation at copper-related CpGs and risk of acute coronary syndromes. They found higher methylation at cg05825244 in EBF4 (among our 12 CpGs differentially methylated in VLBW adults), which was associated with a 23% increased risk of acute coronary syndromes. Further, mass spectrometry of sera from patients about to undergo coronary artery bypass graft (CABG) surgery identified that presence of EBF4 protein was more prevalent and helped distinguish patients with T2DM [66]. Our gene network analysis identified that in VLBW adults EBF4 is a key gene in a network centering on the serine/threonine protein kinase, Akt, which has key roles in many signalling pathways [67]. Together, these findings suggest that the reversal from EBF4 hypomethylation in VLBW neonates to hypermethylation in adulthood may be involved in the compensatory catch-up of growth and development of the cardiovascular system. It is plausible that overcompensation of the EBF4 pathway with age may even contribute to adult cardiovascular risk in VLBW cases, with lower EBF4 methylation levels at birth potentially associated with higher adult systolic blood pressure, smaller cardiac dimensions and lower cardiac output. Future functional studies in animal models may be needed to clarify the relationship between EBF4 methylation and phenotypic traits of adult VLBW survivors.
Methylation patterns have been used to generate epigenetic age estimates, which have now been established to predict chronic disease burden and time to death [47, 48]. Using the DNAm GrimAge calculator, we confirmed previous reports that VLBW survivors have significantly greater methylation than age-matched controls. Van Lieshout et al.[52] compared epigenetic age estimates in adults aged 30 to 35 years born extremely low birthweight (ELBW) (< 1000 g), with a sample of age- and sex-matched controls. This study analysed DNA from buccal cells and generated a methylation score consisting of 353 CpGs in Horvath’s epigenetic-clock algorithm. They demonstrated that ELBW men had a significantly older epigenetic age (4.6 years) than normal birthweight men, although women born ELBW were not found to be epigenetically older than their normal birthweight peers. A further study of 143 ELBW infants born 1991–1992 in Victoria, Australia, used DNA extracted from neonatal blood spots collected after birth to generate an algorithm to estimate DNAm-based gestational age[50]. The residual of DNAm gestational age and clinically estimated gestational age (referred to as “gestational age acceleration”) was used to assess developmental maturity. Infants with higher gestational age acceleration were less likely to have received surfactant or postnatal corticosteroids, had fewer days of assisted ventilation, and less frequently had bronchopulmonary dysplasia.
One limitation of our study is that our DNA methylation analysis in neonates was performed on dried blood spots that had been archived for 30 years, while in the young adults we were able to assay fresh whole blood. The differences in storage time and DNA sample preparation method meant that it would be inappropriate to directly compare raw methylation data files between ages; rather we analyzed the samples separately and compared the resultant lists of candidate CpGs with differing methylation between VBLBW cases and controls.
In future work, it would be of interest to relate our findings to current cardiovascular and respiratory health outcomes of the Victoria cohort [49, 50], and to examine the associations with their neonatal DNAm gestational age. Those infants were born 1991–1992 and would now be aged 29–30 years, similar to the age of the NZ VLBW cohort in the present study.