Keywords

1 The Developmental Origins of Health and Disease

The search for the origins of chronic disease has shifted the focus toward the earliest phases of the life course. While classic epidemiology has targeted lifestyle patterns of adults at various ages, in recent decades the importance of early life for determining lifelong health has been increasingly recognized. Following the seminal work of Rose [1], Forsdahl [2], and Barker [3, 4], the period from conception to birth and the first few years of life are considered critical in influencing disease susceptibility throughout life. This shift in thinking and research gave birth to the “Developmental Origins of Health and Disease” (DOHaD) hypothesis.

Epidemiologic studies support the hypothesis that chronic diseases have their roots in early life. Barker’s work linked low birthweight to a number of cardiovascular diseases (including ischemic heart disease), hypertension, cholesterol levels, stroke, and impaired glucose tolerance [4,5,6,7,8]. His findings have been confirmed by other groups in different populations [9,10,11]. Data from the Dutch Famine in 1944/45, when food rations dropped below 1000 kcal/day for six months, suggest an increased risk of obesity among offspring of mothers exposed to the famine during the first and second trimester [12], glucose intolerance if exposure peaked during late gestation [13], and schizophrenia if conception occurred during the famine [14]. Other maternal characteristics such as maternal weight and malnutrition also increase the risk of coronary heart disease in the offspring [15].

However, there is trouble at both ends of the birthweight spectrum. Like low birthweight, high birthweight is also associated with adult obesity [16]. Similarly, women are more likely to become obese in adulthood if their mother was obese prior to pregnancy and/or had very high or very low gestational weight gain [17]. Furthermore, gestational diabetes (associated with fetal macrosomia) increases the risk of childhood and adult obesity in the offspring [18]. High birthweight is also associated with an elevated risk of several cancers. Numerous epidemiologic studies support the association between high birthweight and increased risk of premenopausal breast cancer [19, 20]. In addition, high birthweight has been linked to childhood leukemia [21], childhood brain tumors [22], and testicular cancer [23].

2 DOHaD Mechanisms

The fetal origins hypothesis suggests that perturbations at a critical period of development induce persistent alterations with potentially lifelong consequences. These epidemiologic observations led Hales and Barker to suggest the “thrifty phenotype” hypothesis, which proposes that poor fetal nutrition and growth lead to metabolic reprogramming of glycemic metabolism [24]. This adaptive developmental plasticity allows the fetus to adjust to and survive adverse environments. According to this model, a limited supply of transplacental nutrients compels the fetus to channel nutrients to the most vital organs, namely brain and heart, at the expense of other organs, which may remain underdeveloped and compromised in growth and function [25]. Moreover, permanent insulin resistance may be induced during development and reduce basal metabolic requirements; this permits survival under suboptimal prenatal and predicted postnatal conditions [26]. Indeed, environmental perturbations may have a long-lasting impact at times of greatest plasticity during growth and development, while decreasing plasticity with increasing age allows less adaptation.

The potential benefits of such so-called predictive adaptive responses depend on the accuracy of the prediction; the cost of inaccurate predictions is high [26]. If developmental conditions that induce intrauterine growth retardation are followed by a resource-rich postnatal environment, high plasma glucose levels will coincide with insulin resistance, greatly increasing the risk for metabolic disease in later life [26]. This “mismatch” between predicted and actual postnatal environment may explain profound long-lasting implications for chronic disease among individuals prenatally exposed to the Dutch Hunger Winter, which lasted only nine months and was followed by normal nutritional availability [27]. Individual variation in sensitivity to mismatch and consequent disease susceptibility is likely due to a variety of factors including genetic variation and the degree of developmental plasticity [25].

3 Potential Critical Periods for Developmental Epigenetics

The biologic mechanisms underlying the long-term persistence of DOHaD phenomena are not well understood. Developmental plasticity allows a specific genotype to create alternative phenotypes depending on embryonic, intrauterine, and early postnatal conditions, which may induce lasting changes in chronic disease susceptibility. Among the various potential biologic mechanisms underlying developmental plasticity [28], environmental influences on developmental epigenetics are receiving increasing attention [29]. Epigenetics describes the study of mitotically heritable alterations in gene expression potential that are not mediated by DNA sequence alterations [30]. Essentially, epigenetic regulation involves a repertoire of cell-autonomous molecular modifications that govern selective access to the genetic information; because these are mitotically heritable, they are perpetuated in differentiated tissues. The specific molecular mechanisms that function interactively to heritably regulate chromatin conformation include DNA methylation (which occurs predominantly at cytosines within cytosine-guanine dinucleotides, i.e., CpG sites), various modifications of the histone proteins that package DNA in the nucleus, and autoregulatory DNA binding proteins [31]. The ontogenic periods during which these mechanisms undergo establishment and maturation suggest potential critical periods of environmental sensitivity (Fig. 6.1).

Fig. 6.1
figure 1

Sources of interindividual variation in the epigenome. Environmental influences on the epigenome are likely most important during establishment of the epigenetic marks in prenatal and early postnatal development. [Reprinted with permission from R. A. Waterland and K. B. Michels: Annu Rev Nutr 27:363–388, 2007 [31]]

Full size image

Many studies of epigenetics in DOHaD have focused on genomically imprinted genes. Genomic imprinting is the epigenetic silencing of either the maternal or paternal allele of specific genes by DNA methylation, leading to parent-of-origin-specific expression. Loss of imprinting results in the aberrant biallelic expression of an imprinted gene. Loss of imprinting of fetal growth genes, in particular that encoding insulin-like growth factor 2 (IGF2), has been associated with childhood disorders such as Beckwith–Wiedemann syndrome [32, 33], Silver–Russell syndrome [34], and Wilms’ tumor [35, 36], as well as with adult-onset diseases [37, 38]. In humans, approximately 100 imprinted genes have been identified. Since most imprinted genes play a role in intrauterine and early life growth, they have long been proposed as good candidates to translate early nutritional and environmental influences into fetal development [29]. Whether epigenetic regulation at imprinted genes is particularly susceptible to early developmental influences remains unresolved [39, 40].

Importantly, the epigenome is established at crucial developmental time points that coincide with vulnerable periods of adaptive plasticity. In the mouse model, each generation undergoes two waves of epigenomic reprogramming. As part of gametogenesis during mid-gestation development, primordial germ cells differentiate into oocyte and sperm [41], assuming distinct epigenomic profiles markedly different from those of somatic tissues. Then, after fertilization, the non-imprinted gene regions in the zygotic genome undergo another round of epigenetic reprogramming that restores totipotency. Genome-wide de novo methylation in the preimplantation embryo [42,43,44,45] permits cell fate commitment of the first cell lineages (discussed in more detail in Chap. 5). These dramatic waves of epigenetic reprogramming make mid-gestation and early embryonic development likely critical periods during which nutritional, environmental, and metabolic factors may affect the developmental establishment of epigenetic regulation in the gametes and somatic tissues, respectively.

As a first step toward understanding the role of epigenetic mechanisms in DOHaD, defining the window of susceptibility is crucial. In the mouse, for example, de novo methylation occurs at different times for imprinted and non-imprinted genes and in the developing female and male germline [43] (discussed in more detail in Chap. 5). The DNA methylation signature of non-imprinted genes may be most amenable to environmental stimuli just prior to implantation, when the totipotent blastocyst, largely stripped of genomic methylation, undergoes lineage-specific remethylation during cellular differentiation. Nutritional and metabolic factors affecting the blastocyst during the early part of the first trimester therefore have great potential to augment or impair the introduction of cytosine methylation. The timing of remethylation of imprinted genes is less clear. By extrapolation from the mouse model, cytosine methylation of the differentially methylated regions (DMRs) of one of the two parental chromosomes is established at different time points for different imprinted genes [46,47,48]. In the mouse, maternal imprints are established at some point between oocyte development and ovulation [48], and paternal imprints are completed by the time spermatocytes enter meiosis [49]. Whether the establishment of imprinting marks is similar in humans remains to be established.

Periconceptional environmental stressors may yield downstream epigenetic effects in multiple tissues if induced epimutations are maintained during subsequent differentiation; perturbations during late gestation, on the other hand, are more likely to induce cell type-specific epigenetic changes [31]. Further, epigenetic development is not limited to prenatal life; for example, the early postnatal period appears to be a critical period for establishment of DNA methylation in the brain [50].

We have previously proposed two mechanisms to explain environmental influences on the developmental establishment of DNA methylation [31]. First, an imbalance in dietary methyl donors and/or activity of DNA methyltransferases may induce hyper- or hypomethylation. While most transposable elements in the mammalian genome are silenced by CpG methylation [51, 52], some are metastable and can also affect expression of neighboring genes [53]. Such metastable epialleles show large interindividual differences in DNA methylation and gene regulation—even among isogenic individuals—and appear particularly labile in response to environmental stimuli during developmental establishment of the epigenome.

Second, nutritional or environmental stimuli may alter transcriptional activity during periods of de novo DNA methylation, which may permanently alter epigenetic regulation and corresponding phenotypes. Genes actively transcribed during de novo methylation are protected from methylation and remain hypomethylated [54]. Interference with active transcription renders these promoters susceptible to de novo hypermethylation and alters their function [55].

The placenta’s critical role in nutrient transfer from mother to fetus makes it particularly vulnerable to adverse intrauterine conditions. Whereas induced epigenetic changes in the soma persist to influence later phenotype, maternal nutrition may also induce epigenetic changes in the placenta, affecting nutrient transport and fetal growth [25]. Imprinted genes are highly expressed in the placenta, which may make them vulnerable to variation in maternal nutrition [56, 57].

4 First Clues from the Agouti Mouse

Seminal experiments in the agouti viable yellow (Avy) mouse model support the idea that maternal nutrition can induce developmental programming via epigenetic mechanisms [58]. The agouti gene codes for yellow pigment in fur. Transposition of an IAP retrotransposon upstream of agouti resulted in the Avy metastable epiallele. DNA methylation of the retrotransposon exhibits spontaneous interindividual variation, controlling expression of the agouti gene and therefore the coat color of the animal. Moreover, supplementation of mouse dams during pregnancy with the dietary methyl donors and cofactors folic acid, vitamin B12, betaine, and choline shifts the coat color distribution of the offspring from yellow to brown [29, 59]. This was shown to occur by induced hypermethylation at the Avy locus [58] systemically and permanently reducing expression of agouti.

Similarly, supplementation of the dams with the phytoestrogen genistein results in an analogous coat color shift also mediated through Avy hypermethylation [60]. Maternal methyl donor supplementation studies in another murine metastable epiallele model, the axin fused mouse, corroborated the findings in the Avy model [61], indicating that epigenetic regulation at metastable epialleles is generally susceptible to early environmental influences. Putative metastable epialleles are now being identified in humans [62]; as in the mouse models, these human loci show dramatic and systemic interindividual epigenetic variation that is influenced by maternal nutrition around the time of conception [63].

5 Epigenetic Epidemiology of DOHaD

Over the past two decades, numerous epidemiologic studies have been performed to explore the role of epigenetics in DOHaD. Within the epigenetic toolbox, DNA methylation is the most likely candidate to explain DOHaD observations due to its relative stability over time; furthermore, it is easiest to study due to its persistence within stored samples. Within the framework of DOHaD, epigenetic studies have addressed either the link between perinatal exposures and DNA methylation at various timepoints throughout life or the link between DNA methylation in early life and later health outcomes [64]. Studies exploring DNA methylation as a mediator connecting early life exposures and later life disease have been sparse [64]. In the following, we highlight some of the studies of particular interest.

Associations between prenatal and early life exposures and DNA methylation are being extensively examined by the Pregnancy and Childhood Epigenetics (PACE) consortium [65] as well as other groups. The best-established association is between maternal smoking during pregnancy and DNA methylation in the offspring cord blood, with a consistent change across most studies found in a CpG in the AHRR gene [66]. Four other CpGs that were changed in cord blood also showed changes in the placenta in a subset of these cohorts that had also collected placenta tissue [67]. In another PACE meta-analysis including 9340 mother–newborn dyads, both very high and very low maternal pre-pregnancy body mass index were linked to several small DNA methylation differences (<0.2% per BMI unit) in cord blood [68]. Interestingly, maternal alcohol consumption was not found to be associated with offspring cord blood methylation [69]. The PACE consortium and other cohorts have also linked preeclampsia and gestational diabetes to cord blood and placental DNA methylation [70,71,72,73].

Maternal self-reported high folic acid supplementation (defined as an average 1200 μg/day or more) was associated with 2.4% lower methylation at the H19 differentially methylated region (DMR) in umbilical cord blood leukocytes if initiated before pregnancy and 3.7% less methylation at that locus if initiated during pregnancy, compared to mothers not reporting supplementation; no difference was found for the IGF2 DMR0 [74]. A study from the Netherlands reported 4% higher methylation at one CpG of IGF2 DMR0 in the blood of 17-month-old children whose mothers reported taking 400 μg folic acid during pregnancy compared to those whose mothers took no folic acid supplements; however, IGF2 expression levels were not examined [75]. Maternal plasma folate levels during pregnancy were associated with DNA methylation in the cord blood of 1988 newborns [76].

In addition to smoking and folic acid supplementation, studies have linked other intrauterine exposures to DNA methylation in adulthood. Data on survivors of the Dutch Famine suggest that, compared to their unexposed siblings, individuals prenatally exposed to famine had somewhat lower DNA methylation at IGF2 six decades later [77]. In this study of 60 same-sex sibling pairs, the authors examined five CpGs in the IGF2 DMR0 and found, on average, 2.7% lower methylation among individuals exposed to famine in utero. Whether this small difference in methylation has any functional consequence remains unclear, in particular since the authors did not examine IGF2 expression levels. Methylation differences of even smaller magnitude were observed for some other genes including IL10, GNASAS, INSIGF, LEP, and MEG3 [78]. The association between an epigenetic difference assessed in adulthood and a prenatal exposure does not allow causal inference about the induction of that change by the prenatal factor, unless the change is already present directly after the exposure period [28, 31]. Of course, collecting appropriate samples in humans to test such causal pathways is logistically challenging.

Studies on DNA methylation in cord blood or placenta have considered several aspects of weight. A number of studies linked DNA methylation with birthweight with varying results [79,80,81,82]; in any event, changes observed did not persist to adulthood. The PACE meta-analyses of EWAS including 8825 neonates from 24 birth cohorts found birthweight associated with DNA methylation in neonatal blood at 914 sites, with a difference in birthweight ranging from −183 to +178 grams per 10% increase in methylation levels [83]. Some studies specifically explored the epigenetic profile of newborns with low birthweight or intrauterine growth retardation (IUGR). Einstein and colleagues compared cord blood samples from five IUGR and five normal pregnancies and identified methylation differences at a restricted number of loci [84]. A few small studies identified differences in methylation or expression of selected imprinted genes in the placenta and cord blood of IUGR or low birthweight compared to normal-weight infants [85,86,87,88] and in selected non-imprinted genes [89,90,91]. Conversely, high birthweight has been associated with increased promoter methylation of the glucocorticoid receptor gene in human placenta [84,85,86,87,88, 92]. Overall, differences in methylation in these studies were small, and it remains unclear whether DNA methylation changes are a cause or consequence of aberrant birthweight. Studies on childhood weight suggested an association between changes in newborn methylation of the RXRA gene and the promoter or the long noncoding RNA ANRIL with childhood adiposity [93, 94]. CpG methylation of 68 CpGs in five candidate genes was assessed in umbilical cord tissue from healthy neonates in two prospective cohorts [93]; DNA methylation of one CpG was consistently associated with adiposity at the age of 9 in both cohorts.

In another childhood obesity study applying an array-based genome-scale screen to neonatal blood screening cards, although no statistically significant site emerged comparing the lowest and the highest BMI quartile at age 5, 13 CpG sites showed a > 5% difference in in DNA methylation levels [95]. All 13 were located in close proximity to the nc886 gene. This gene, which encodes a small non-coding RNA, shows polymorphic imprinting in neonatal blood which appears to be modifiable by maternal age and nutrition status during pregnancy [40]. Methylation of the differentially methylated region nc886 may operate as a mediator between maternal characteristics and childhood outcomes, although a study demonstrating this link directly remains to be conducted.

Few studies have directly evaluated DNA methylation as a mediator between perinatal exposures and subsequent health outcomes. Cardenas et al. examined whether DNA methylation changes may mediate the association between intrauterine exposure to mercury and lower cognitive performance in childhood [96]. In newborn cord blood of 321 children, they found prenatal mercury levels were associated with lower DNA methylation at the paraoxonase 1 gene, which predicted lower regional cognitive test scores during early childhood. DNA methylation levels at this site, however, were attenuated in blood samples collected in mid-childhood, arguing against direct mediation.

Recently, focus has shifted to studies at the interface between epigenetics and the microbiome to explain DOHaD effect persistence [97]. Similar to epigenetic marks, the gut microbiome is established at birth, but remains malleable to a certain extent by lifestyle factors. The gut microbiome can influence DNA methylation and the activity of DNA methyltransferases and histone deacetylases, although the direction of this crosstalk is not always clear [98,99,100,101]. Bacterial metabolites, in particular short chain fatty acids, can function as HDAC inhibitors [102] and correlate with DNA methylation [103].

In summary, epigenetic mechanisms are likely candidates to explain at least some DOHaD phenomena. Nevertheless, despite a recent proliferation of studies in this area, it remains unclear whether the mostly small differences in DNA methylation at birth associated with intrauterine exposures have functional relevance and are maintained into adulthood. Whether the embryonic, intrauterine, and early postnatal environments affect adult disease susceptibility in humans via induced epigenetic alterations remains to be established. Although challenging, longitudinal cohorts assessing links between the periconceptional, pregnancy, and infant environment with adult health and disease status (including measurements of DNA methylation and potentially the gut microbiome at birth and throughout life) are needed to shed more light on these questions. Due to both their malleability by early nutrition and other lifestyle factors, and their noted long-term stability once established, DNA methylation and the gut microbiome hold promise for life-course prevention efforts.

6 Challenges for Epigenetic Epidemiology in DOHaD

In 2003 the International HapMap Project set out to identify common sequence variants in the human genome [104]. This “toolbox” enabled large-scale studies to test for associations between these variants and human diseases and phenotypes, heralding the dawn of the genome-wide association study (GWAS) era. In the past two decades, GWASs have identified an impressive and growing number of disease risk-associated genetic variants. Despite this success, however, the majority of individual variance in disease risk remains unexplained, contributing to increased interest in the idea that epigenetic variation could influence the etiology of disease [105,106,107,108] and leading to the development of so-called epigenome-wide association studies (EWAS) [109].

But the epigeneticists skipped a crucial step: no “epiHapMap” project was conducted. Rather, the overwhelming majority of the hundreds of “EWAS” studies in the literature employ DNA methylation arrays produced by Illumina (most notably the HM450 and more recently the EPIC850 array). Inexplicably, interindividual variation in DNA methylation was never considered in the design of the Illumina arrays [110, 111]. In fact, most of the probes on these arrays show negligible interindividual variation [112, 113]. A study evaluating the HM450 array in blood, using 256 technical replicates from 130 participants, showed that fewer than half of the CpG sites demonstrated greater interindividual variation than the variation due to technical errors [114]. Another study showed that the power of EWASs could be improved by focusing on the minority of CpG sites with substantial interindividual variation in DNA methylation [115]. A more recent study reported that in peripheral blood DNA, the greatest source of variation at most HM450 probes is intra-individual variability (most likely from variation in leukocyte composition) rather than interindividual variation [116]. The upgrade from the HM450 to the EPIC array in 2016 has not substantially improved the situation. Between HM450 and EPIC arrays, about 55% of the CpG sites show a correlation <0.20, due to low interindividual variability [117]. A recent study that used the EPIC array to examine test-retest reproducibility of peripheral blood DNA methylation of the same women over a one-year period [113] found extremely poor performance (average intraclass correlation coefficient of 0.22), and attributed this to the fact that “99.9% of CpG sites (covered by the array) in the non-sex chromosomes had similar methylation profiles between individuals.” These data underscore the unfortunate fact that, over the last decade, over 1000 studies attempting to associate individual epigenetic variation with risk of disease have focused on genomic regions in which DNA methylation is largely invariant.

Another major factor overlooked by the HM450 and EPIC platforms is the cell type specificity of DNA methylation. Generally, we cannot “epigenotype” an individual using peripheral tissues such as blood; epigenetic variation detected in the blood may not be relevant for a disease involving the brain, for example. Reverse causality is another major confounding factor for epigenetic epidemiological studies. Even if the tissue of interest is obtained [118, 119], the disease process itself can cause epigenetic differences, making it difficult to infer causality. Based on these observations, the designs of HM450 and EPIC arrays are far from ideal.

7 The Field Needs to Focus on Systemic Interindividual Epigenetic Variation

A logical approach to overcome these challenges is to focus on genomic regions displaying systemic interindividual variation (SIV) in DNA methylation [62, 120, 121]. A recent study, which could be viewed as a “mini-epiHapMap” project, conducted the largest screening of SIV regions in the human genome [122]. A computational algorithm was developed to analyze deep whole-genome bisulfite-sequencing data on tissues representing all three embryonic germ layers (thyroid, heart, and brain) from each of ten donors from the NIH Genotype-Tissue Expression project [123]. The authors identified 9926 correlated regions of systemic interindividual variation (CoRSIVs). Each CoRSIV is statistically significant (P < 0.05), includes at least 5 CpGs, and exhibits an interindividual methylation range of at least 20%. The multiple-tissue interindividual screening approach to identify SIV is similar to that previously used to identify candidate metastable epialleles [62, 120, 124], but unlike metastable epialleles, CoRSIVs are defined without regard to potential genetic influences on their interindividual variation.

Although only <1% of HM450 or EPIC probes are within CoRSIVs, these regions are often associated with a wide range of diseases. For example, the SIV region encompassing nc886 (also known as VTRNA2–1) is a confirmed metastable epiallele; DNA methylation at this locus is influenced by maternal nutrition during periconceptional development [120]. More recently, evidence has emerged demonstrating this region is influenced by maternal alcohol use prior to pregnancy [40]. Additional studies found that methylation in this region is associated with risk of cancer [124, 125], type 2 diabetes [126], and preterm birth [127]. As mentioned above, a prospective study in infants found that nc886 methylation in peripheral blood at birth predicts BMI at the age of 5 [95]. Consistent with the DOHaD hypothesis, hypermethylation at the DUSP22 promoter (another CoRSIV) shows an association between in utero famine exposure and schizophrenia [128]. Methylation at a CoRSIV located in the promotor of the PM20D1 gene has been linked with Alzheimer’s disease [129]. More studies have found associations between CoRSIV gene methylation and Parkinson’s disease [130], autism [131, 132], major depression and suicide [133], rheumatoid arthritis [134], multiple sclerosis [135], and obesity [136]. Methylation in SIV regions near the OR2L13 promoter and gene body of CYP2E1 [124] is associated with maternal gestational diabetes mellitus [71]. Hence, despite their under-representation on the Illumina arrays, CoRSIVs are often among top hits in HM450 and EPIC profiling studies screening for associations with disease and associated phenotypes, indicating immense potential for these regions to contribute to disease prediction, diagnosis, and prognosis. From a DOHaD perspective, a focus on CoRSIVs is particularly warranted, given their well-documented plasticity to periconceptional environment [28, 62, 63, 121, 122, 137].

In addition to a focus on CoRSIVs, we believe the field will benefit from development of novel analytical approaches. Most studies of DNA methylation and disease have utilized univariate regression methodologies and focused on detecting associations rather than making predictions [138, 139]. It is increasingly recognized, however, that individual CpG sites do not provide as much information as coordinated interactions among multiple CpGs. Multivariate approaches can harness crucial synergistic biological effects [140], motivating increased interest in using machine learning to analyze DNA methylation. Target-capture approaches to study DNA methylation across the entire set of known CoRSIVs are under development. Meanwhile, there are many publicly available HM450 and EPIC datasets, in which ~10% of known CoRSIVs are covered by at least one probe. A recent study [141] took advantage of a publicly available HM450 data set on peripheral blood of schizophrenia (SZ) cases and controls [142] to develop a CoRSIV-focused machine learning classifier based on sparse partial least squares discriminant analysis (SPLS-DA). The model calculated an epigenetic risk score which was able to identify SZ cases with 80% positive predictive value, far surpassing the performance of an analogous SPLS-DA classifier based on polygenic risk score. Additional analyses indicated that these associations were not due to reverse causality, as might be caused by the tendency for SZ patients to smoke heavily and/or take psychotropic medications. Together these findings indicate that the systemic interindividual variants distinguishing SZ cases from controls were present prior to diagnosis; prospective studies will be required to confirm this. Nonetheless, this study provides compelling evidence that a focus on SIV, combined with sophisticated machine learning approaches, may ultimately enable blood-based disease risk prediction for a wide range of complex human diseases, with obvious implications for DOHaD.

8 Outlook

An epigenetic basis for DOHaD involves two steps: (1) early environmental influences during critical ontogenic periods can induce lasting epigenetic changes, and (2) these individual epigenetic differences must influence risk of disease later in life. There is now extensive evidence supporting the first step. Particularly in the context of human metastable epialleles and CoRSIVs, it is clear that periconceptional environment affects establishment of DNA methylation states that persist for years [62, 120, 122, 124]. The focus now must be on the second step of the pathway, i.e., establishing causal links between individual epigenetic variation and risk of disease. Despite the “failed start” due to the problems with the Illumina platforms, we believe that an increasing focus on CoRSIVs heralds great potential in the field of epigenetic epidemiology. The systemic nature of interindividual epigenetic variation means that CoRSIVs are essentially epigenetic polymorphisms, facilitating the use of DNA samples from blood, saliva, or buccal cells in large-scale epigenetic epidemiologic studies. Development of commercial platforms focused on CoRSIVs, coupled with the establishment of prospective longitudinal cohorts, will allow epigenetic epidemiologists to probe causal links between early environment, DNA methylation, and disease.

9 Conclusions

The concept of DOHaD arose from epidemiologic studies. Developmental plasticity implies that fetal development adapts to transient nutritional and environmental experiences, resulting in lasting changes in chronic disease susceptibility. While our understanding of the underlying mechanisms is rudimentary, alterations in epigenetic regulation are likely contributors. Although CoRSIVs provide a promising avenue for future DOHaD-centered epigenetic studies, we emphasize that epigenetics is only one of several potential mechanisms explaining developmental plasticity. A better understanding of the mechanisms underlying DOHaD should someday make it possible to reduce individual risk of disease by both preventive strategies targeted to early life and corrective approaches designed to normalize malleable cellular and molecular mechanisms set askew by adverse early exposures.