Background
Environmental exposures acquired early in life have been correlated with persistent modifications of the epigenome. The epigenetic information in human cells is stored via mitotically heritable DNA methylation, organization of the chromatin structure (for example, histone modification), and regulatory RNAs. Together, these mechanisms are responsible for regulating gene expression during cellular differentiation during embryonic development and throughout life 1. Our study focuses on the DNA methylation patterns of the imprinted Insulin-Like Growth Factor 2 (IGF2) gene, coding a well-characterized growth factor active throughout embryogenesis and fetal growth 23. In normal human tissues only the paternal IGF2 allele is transcribed; and its imprinting is regulated by at least two differentially methylated regions (DMRs): one is located upstream of the three IGF2 promoters that are subject to imprinting (IGF2 DMR), and the other is located upstream of the neighboring non-coding H19 gene (H19 DMR). The latter region is part of the imprinting control region (ICR) which harbors binding sites for the zinc finger protein CTCF. During early development, imprint marks are erased in the primordial germ cells and new methylation imprints are established according to the germ cell. This has particularly been demonstrated at the IGF2 DMR 4 and H19 DMR 45 in spermatogenic human cell stages. The progressive imprint re-establishment of the DNA methylation imprint marks throughout human spermatogenesis leads to fully methylated IGF2 and H19 DMRs. Consequently, methylation is only present on the paternally inherited allele in the offspring. Shifts in methylation established at these DMRs can lead to loss of imprinting and transcription of IGF2 may be altered 67. Hence, normal physiological mechanisms or homeostasis in the body may be skewed and lead to chronic diseases later in life. Until now, epidemiological studies have focused on maternal factors and especially in utero exposures to certain nutritional or environmental conditions as the potential explanation for such disruptions or shifts in methylation at the DMRs 8910. This can potentially contribute to a higher risk for obesity 11, chronic diseases at later age 12, including diabetes or cardiovascular diseases 131415, or even cancer 1617. Animal experiments show that modification of maternal diet during development can influence metabolism in adulthood 18. Although the underlying mechanism and crucial time points of exposure are not clear, changes in epigenetic regulation are now regarded as a highly plausible explanation for linking the associations between dietary exposures in early life to the onset of chronic diseases during adulthood. Several lines of evidence suggest that pre- or periconceptional obesity of the mother may affect metabolic programming 192021. Although paternal obesity is equally prevalent, the majority of the epidemiological studies to date suggest in utero exposures as the only possible origin of potential epigenomic modifications at birth. Obesity is associated with over-nutrition, unbalanced food intake (such as low vegetable consumption), and a sedentary lifestyle 22. Consequently, elucidating the epigenetic risks associated with the current "Western" lifestyle on the next generations is crucial. In the current report we determined whether preconceptional obesity, in the mother or the father, is associated with methylation patterns at the IGF2 DMRs in newborns using DNA from leukocytes isolated from umbilical cord blood at birth. By including paternal obesity in our study we were able to examine a potential preconceptional impact of the environment on imprint mark reprogramming during male gametogenesis. Consequently, we found that paternal obesity is associated with a decrease in DNA methylation at the IGF2 DMR.
Methods
Study participants
We studied a subgroup of the first 98 families enrolled in the Newborn Epigenetics Study cohort. Seventy-eight mothers provided detailed information about the biological fathers. One mother gave birth to twins, bringing the total number of newborns with corresponding paternal data to 79. The enrollment process was assisted by a trained interviewer. Detailed recruitment strategies for NEST have been described in a previous study 23. English speaking pregnant mothers delivered between July 2005 and November 2006 at Duke University Hospital, Durham, NC.
Data collection
The questionnaire included maternal socio-demographic data, such as age, marital status, race and education, as well as multiple lifestyle or health characteristics (for example, smoking, chronic diseases). One of the main items of the survey included a detailed list of questions about the mother's and the father's height, highest and lowest weight ever, current and usual weight. Medical records were used to verify medical conditions, and to abstract gestation time, birth weight and the newborn's gender. Body mass index was calculated from the data obtained from height and the mother's weight before pregnancy or the father's current weight. Obesity was defined as BMI ≥30 kg/m2.
Specimen collection
At delivery, umbilical cord blood samples were collected in a vacuum blood collection tube coated with K3EDTA. The tubes were centrifuged to isolate the buffy coat; this leukocyte-containing buffy coat was then stored at -80ºC. Genomic DNA was extracted using Gentra Puregene Reagents (Qiagen, Valencia, CA, USA).
DNA methylation analysis
IGF2 and H19 DMRs were analyzed by pyrosequencing. The IGF2 DMR includes three CpG dinucleotides upstream of IGF2 exon 3 (chr 11p15.5; CpG site 1: 2,169,518; CpG site 2: 2,169,515; and CpG site 3: 2,169,499; NCBI Human Genome Build 37/hg19). This DMR has been previously evaluated by Cui et al. 6 and Heijmans et al. 8. The region studied for the H19 DMR encompasses four dinucleotides located upstream of the H19 gene (chr 11p15.5; CpG site 1: 2,024,261, CpG site 2: 2,024,259, CpG site 3: 2,024,257, and CpG site 4: 2,024,254; NCBI Human Genome Build 37/hg19), which is the first of six known sequence motifs that bind the CTCF zinc finger protein 2425. The structural characteristics of these IGF2 and H19 loci and the genomic coordinates of the assays have been presented elsewhere 726. Genomic DNA (800 ng) was treated with sodium bisulfite 27, the IGF2/H19 regions were amplified by PCR, and pyrosequencing was performed in duplicate using a Pyromark Q96 MD pyrosequencing instrument (Qiagen). Control assays were also run to validate our methylation results. Detailed methodology, including assay conditions and validation studies, have been described previously 25.
Statistical methods
Variables were defined as follows: race (African American, Caucasian, and other), age (as a continuous variable), marital status (living with partner versus single), at least a college graduate (yes or no), pre-pregnancy maternal and paternal obesity (<30 versus ≥30 kg/m2), smoking status (never/quit smoking when pregnant/continued smoking during pregnancy), birth weight of the baby (<2.5 kg/between 2.5 kg and 3.5 kg/≥3.5 kg), gestation time (<37 weeks versus ≥37 weeks), and gender of the baby (male or female) (Table 1). Chi Square tests were used to compare obesity of the mother and the father within different sub-groups of pregnant women. If numbers were small (<5) the Fisher exact test was used. We also compared the characteristics for missing and non-missing experimental methylation results and for missing and non-missing BMI of the father. Methylation levels were distributed normally in the groups studied (verified by using the Kolmogorov-Smirnov test). Student's t-tests were computed to test for significant differences in the methylation means. We assessed the effect of maternal and paternal obesity (BMI ≥30 kg/m2) on the methylation levels of the IGF2 and H19 DMRs, which were analyzed separately by individual CpG site, as well as by the mean of the CpGs. To account for batch effects from the laboratory tests on different days or different plates we calculated the least square means (or estimated marginal means) of each CpG site. We evaluated potential confounding for paternal and maternal obesity by means of multivariable regression analyses. We used multiple regression analyses, separately for each exposure and characteristics from Table 1. These multiple regression analyses were used to predict DNA methylation at each DMR site. To further explore whether the association between obesity and DNA methylation varied by race, we repeated these analyses in African Americans and Caucasians. Graphs representing the associations between BMI and methylation outcomes were based on the predicted outcomes of the methylation means, after adjusting for maternal age, smoking status, birth weight and gender. Observations included in the regression analyses were determined by the availability of laboratory measures for the IGF2 and H19 DMRs. All statistical analyses were conducted in SAS v9.2 (SAS Institute Inc., Cary, NC, USA), and GraphPad Prism 5 was used to obtain Figures 1 and 2 (GraphPad Software Inc., La Jolla, CA, USA).
<p>Table 1</p>Parental and newborn characteristics
NEST - Newborn Epigenetics Study Cohort 2005 to 2006
n
%
BMI mother:
BMI <30 (not obese)
59
67.8
BMI ≥30 (obese)
28
32.2
BMI father:
BMI <30 (not obese)
63
79.7
BMI ≥30 (obese)
16
20.3
Marital status:
Living with partner
72
74.2
Single
25
25.8
Education:
Low (no college degree)
57
58.8
High (at least college degree)
40
41.2
Race:
African American
38
38.8
Caucasian
56
57.1
Other or not specified
4
4.08
Maternal age:
<30 years
56
57.1
≥30 years
42
42.9
Smoking:
Mother never smoked
45
48.4
Quit smoking when pregnant
26
27.9
Smoked during pregnancy
22
23.7
Gestation time:
Preterm (<37 weeks)
10
10.3
Normal (≥37 weeks)
87
86.7
Birth weight:
<2.5 kg
16
16.5
≥2.5 kg
81
83.5
Baby gender:
Male
48
49.5
Female
49
50.5
This sub-cohort includes all NEST families from whom babies were born at Duke University Hospital between July 2005 and November 2006. Characteristics of mothers, fathers and newborns are shown for the 98 participants when data were not missing.
<p>Figure 1</p>Methylation at the IGF2 and H19 DMRs in the offspring by parental obesity
Methylation at the IGF2 and H19 DMRs in the offspring by parental obesity. The graphs represent the mean estimated methylation values of 69 newborns at the IGF2 DMR, and 70 newborns at the H19 DMR. The IGF2 DMR results are based on 14 obese fathers and 25 obese mothers; the results at the H19 DMR are based on 15 obese fathers and 23 obese mothers. For each exposure the differences of the least square means of methylation percentages are shown at each CpG site (bold), as well as standard errors (SE) and P-values. Bars represent standard errors.
![]()
<p>Figure 2</p>Offspring's mean methylation % at the IGF2 DMR by BMI of the parents
Offspring's mean methylation % at the IGF2 DMR by BMI of the parents. The predicted methylation means at IGF2 DMR are plotted by BMI of the father (upper graph), and BMI of the mother (lower graph); adjusted for maternal age, smoking status, BMI of the other parent, the newborn's birth weight and gender.
![]()
Ethics
The experimental research in this report was performed with the approval of the Duke University Institutional Review Board, reference number: Pro00014548; this means that the research carried out is in compliance with the Helsinki Declaration. All participants signed the informed consent for the use of their medical record data, questionnaire data, their newborn's birth parameters and umbilical cord blood.
Results
Characteristics of study participants
The distributions of socio-demographic data are shown in Table 1. Seventy-four percent of the mothers were living with a partner, and nearly 26% were single. The majority of the study population were Caucasians (57.1%) or African Americans (38.8%), and the other races/ethnicities included Asians, Native Americans or non-specified race (4.1%). A quarter of the mothers were between 25 and 29 years old (26.5%). Nearly 59% never obtained a college degree, and were categorized as "low education". Approximately 24% of the mothers reported smoking during pregnancy, and 32% were obese before pregnancy. About 80% of the mothers provided data about the father's characteristics: 20% are obese. Approximately 16% of the newborns had a low birth weight, 10% was born preterm and gender was equally distributed. None of the included characteristics were associated with paternal obesity, with the exception of obesity in the mother (P = 0.01). For instance, potential associations between birth weight and parental obesity were excluded by Chi Square tests or Fisher exact tests; P-values were 0.7 for paternal and maternal obesity. Maternal obesity was positively associated with single motherhood (P = 0.002), being African American (P <0.001) and being a non-smoker (P = 0.06). Given that 19% of our subjects did not report detailed information about the fathers, we compared all measured characteristics and methylation end-results in newborns with missing anthropometric data about fathers versus those without missing data. The majority of the missing paternal data (78%) were from African American mothers with lower education. Single mothers were less likely to answer questions about the father's weight and height (32.0%), compared with married mothers or those living with a partner (13.9%), P = 0.07. Other characteristics and the methylation outcomes did not differ significantly between both groups.
Associations between paternal or maternal obesity and methylation profiles at the IGF2 and H19 DMRs in the offspring
Using DNA extracted from umbilical cord blood leukocytes we determined the levels of methylation at CpG sites within the IGF2 and H19 DMRs in newborns. From the 79 samples in our cohort, we obtained experimental data for 67 participants at the IGF2 DMR and for 70 participants at the H19 DMR. We further refer to these participants as our analytical cohort. No significant differences in characteristics were found when comparing the missing with the non-missing experimental methylation results for both DMRs. We analyzed methylation outcomes by the characteristics described in Table 1 and with the exception of obesity (BMI ≥30 kg/m2), none showed significant associations with methylation percentages at the IGF2 or H19 DMRs in our cohort. Figure 1 shows the estimated means of methylation percentages by obesity of the parents. Differences in methylation percentages among offspring from obese fathers compared to non-obese fathers at CpG1, CpG2, and CpG3 of the IGF2 DMR were: -3.36% (P = 0.08), -3.56% (P = 0.01) and -3.57% (P = 0.02), respectively; the mean methylation difference was: -3.50% (P = 0.02) (Figure 1). Methylation in offspring among obese fathers was significantly lower than when compared to fathers who were not obese. We observed no differences in methylation percentages at the IGF2 DMR between offspring of obese and non-obese mothers. Differences in methylation percentages at CpG1, CpG2 and CpG3 were +2.09% (P = 0.18), +0.66% (P = 0.57) and +0.90% (P = 0.44), respectively; the mean methylation difference was +1.19% (P = 0.32) (Figure 1).
At the H19 DMR we found no methylation differences in the offspring of obese compared to non-obese fathers. Differences in methylation percentages at CpG1, CpG2, CpG3 and CpG4 of the H19 DMR were +1.95% (P = 0.37), +1.68% (P = 0.42), +1.75% (P = 0.38) and +1.49% (P = 0.48), respectively; the mean methylation difference was +1.74% (P = 0.40) (Figure 1). However, the methylation percentages at the CpG sites of the H19 DMR among offspring from obese mothers were significantly higher at CpG1, CpG2, CpG3 and CpG4 compared to offspring from non-obese mothers. The differences were +3.56% (P = 0.04), +4.13% (P = 0.01), +4.08% (P = 0.01) and +4.55% (P = 0.007), respectively; the mean methylation difference was +4.10% (P = 0.01) (Figure 1). In brief, without adjusting for potential confounders we detected significantly lower methylation at the IGF2 DMR associated with paternal obesity, and significantly higher methylation at the H19 DMR when the mother was obese.
To assess the independent effects of preconceptional maternal or paternal obesity on DNA methylation at the IGF2 and H19 DMRs in newborns, we used multiple regression models, controlling for potential confounders, including maternal age, maternal smoking status, gender of the baby, birth weight and experimental batch effects. The results of the regression analyses, found in Table 2, show that paternal obesity is inversely associated with DNA methylation levels at the IGF2 DMR in newborns. Controlling for potential confounding did not change this relationship. After adjusting for several characteristics, we observed a significant decrease in methylation when fathers were obese, β-coefficient was -5.28 (P = 0.003) (Table 2, Model 3). Little difference for this association was observed when comparing the individual CpG sites. The β-coefficients for CpG1, CpG2 and CpG3 were -5.78 (P = 0.01), -5.18 (P = 0.002) and -4.76 (P = 0.007), respectively. In contrast, results for maternal obesity indicated an increase in methylation at the IGF2 DMR, the β-coefficient was +3.08 (P = 0.05) when controlling for the same characteristics (Table 2, Model 3). The β-coefficients at the individual CpG-sites were +4.23 (P = 0.04), +2.44 (P = 0.08), +2.42 (P = 0.12), for CpG1, CpG2 and CpG3; with only CpG1 reaching statistical significance.
<p>Table 2</p>Linear Regression Models: methylation at the IGF2 and H19 DMRs in relation to parental obesity
Linear regression models
IGF2 DMR
H19 DMR
Obesity of:
β
SE
P
β
SE
P
Model 1
Father
-3.83
1.48
0.01
+3.09
1.64
0.07
Model 2
Mother
+2.38
1.30
0.08
+2.80
1.38
0.05
Model 3
Father
Mother
-5.28
+3.08
1.62
1.48
0.003
0.05
+2.55
+1.05
1.82
1.70
0.17
0.54
Obesity was defined as BMI ≥30 kg/m2. All models were adjusted for maternal age and smoking status, as well as by the newborn's birth weight and gender. Models 1 and 2 include either maternal or paternal obesity. Model 3 includes both maternal and paternal obesity.
In our separate regression models by maternal and paternal obesity, we observed a positive trend in DNA methylation at the H19 DMR when adjusting for maternal and newborn characteristics (Table 2, Models 1 and 2, respectively). The β-coefficient was +3.09 when the fathers were obese, but this result was not significant (P = 0.07), and +2.80 when the mothers were obese (P = 0.05). The multivariate model with the mean methylation percent as a dependent variable and several measured maternal, paternal and newborn characteristics as independent variables (Model 3) showed a β-coefficient for offspring from obese fathers of +2.55 (P = 0.17), and a β-coefficient for offspring from obese mothers of +1.05 (P = 0.54). Similar results were seen when we evaluated the individual CpG sites (data not shown).
We further stratified our analytical cohort by race (Caucasian and African American) and repeated all regression analyses described above. At the IGF2 DMR, the outcome was similar as in our complete analytical cohort. We calculated a β-coefficient of -5.16 in offspring from obese fathers, among Caucasians (P = 0.01) (as in Model 3); and a β-coefficient of -3.65 in offspring from obese fathers, among African Americans. Although the estimates suggest an inverse relationship in Caucasian subjects, this effect was not significant among African Americans (P = 0.34). These observations were based on only 10 obese fathers in Caucasians, and only 4 obese fathers in African Americans. At the H19 DMR, results were also similar and in the same direction as those in our complete analytical cohort. However, we found one higher estimate in Model 1 for children from African American obese fathers: β-coefficient was +8.14 (P = 0.007). But, this result was based on a small sample size with only 6 obese African American fathers. Together with our statistical power calculations we conclude that the stratified analyses represent unstable estimates.
We further extended the regression analyses in our analytical cohort by replacing the dichotomous obesity variable by the continuous BMI variable. At the H19 DMR, we found a β-coefficient of +0.21 (P = 0.12) for paternal BMI and +0.16 (P = 0.09) for maternal BMI, controlling for the same variables as before; but these results were still not significant. Whereas at the IGF2 DMR, our data reached a significant positive association when both paternal and maternal BMI were used as independent variables in the multivariate model; β-coefficients were -0.38 (P = 0.009), and +0.22 (P = 0.02), respectively. These correlations between parental BMI and the predicted mean methylation levels at the IGF2 DMR in the offspring are depicted in Figure 2.
Discussion
We explored the potential effect of parental obesity on IGF2/H19 DMR methylation in newborns. We found a significant decrease in methylation among newborns of obese fathers at the IGF2 DMR in DNA extracted from cord blood leucocytes. This finding remained significant after controlling for potential confounders (β-coefficient = -5.28, P = 0.003). Hypomethylation at the IGF2 DMR has been associated with an increased risk of developing cancers, such as Wilms' tumor 28, colorectal cancer 6 and ovarian cancer 7. We found no significant changes in methylation levels associated with paternal obesity at the H19 DMR region.
Obesity is a metabolic condition that has paradoxically been associated with poverty, low quality of life, malnutrition and an imbalanced intake of nutrients 2122. Few epidemiological studies indicate potential correlations between obesity or food supplies in the paternal line and offspring's birth weight 29, body-fat in prepubertal girls 30, or mortality from chronic diseases 313233. Epidemiological data regarding associations between maternal obesity and the offspring's birth weight vary by study (reviewed by McDonald et al. 34). We examined possible associations between paternal or maternal obesity and birth weight but detected no associations. Epidemiological studies on maternal obesity-related exposures generally show a positive association between abnormally high BMI and congenital anomalies (reviewed by Stothard et al. 35). These harmful effects are mostly attributed to in utero exposures to malnutrition or overnutrition; while very often, data indicate the importance of exposures at the very early stages of development, even before conception. Research in animal models suggests that potential diet-dependent transgenerational effects may be explained by changes in the establishment of epigenetic gene regulatory marks 12363738. Analyses of adults born to mothers exposed to poor nutrition during the Dutch famine indicated a 5% decrease in methylation at the IGF2 DMR compared to the same sex non-exposed siblings. Interestingly, the magnitude of this effect is similar to the effect we observe in offspring from obese fathers at the same locus. In the Dutch famine cohort, exposure during the periconceptional period was more important than during the second or the third trimester of gestation 8. Furthermore, not only do exposures during early gestation cause harmful health outcomes, but famine prior to conception has also been associated with poor health 39. Similar results were found among Gambian children; they exhibit altered DNA methylation at several metastable epialleles according to the seasonal nutritional circumstances in which they were conceived 40. Both cohorts described above did not directly examine the relationship with the fathers' dietary patterns; although they were most likely exposed to the same famine or nutritional circumstances as the mothers. Analysis of the Framingham Heart Study indicates that early-onset paternal obesity, and not maternal obesity, increases the odds of aberrant serum levels of the metabolic biomarker ALT (alanine transaminase) in the offspring 41. Studies on animal models have shown similar associations. In rats, a high fat paternal diet results in offspring with early onset of impaired insulin secretion, altered expression of multiple genes related to normal pancreatic beta-cell function, and altered methylation at a putative regulatory region of the Interleukin 13 receptor, alpha 2 gene 42. Male mice whose mothers were exposed to a high-fat diet were not only obese, insulin insensitive and diabetic, they were also capable of passing part of this phenotype to the next generation, suggesting an underlying epigenetic mechanism transmitted through germ cells 43. To our knowledge, our study is the first epidemiological study that suggests a similar underlying epigenetic mechanism conferred by harmful paternal dietary patterns or obesity.
Obesity is associated with elevated IGF2 circulating levels 44 and increased estrogen levels 45. Although we did not include IGF2 protein levels in parents or offspring in our current analyses, we earlier showed that hypomethylation at the IGF2 DMR is associated with higher circulating IGF2 levels in the offspring 2546. This association was strongest in offspring from obese mothers, independent of race. In brief, a decrease of 5% at the IGF2 DMR methylation corresponded to an increase of at least 10% in serum concentration of IGF2 46. In addition, other studies have shown that small aberrant methylation changes at the IGF2 or H19 DMR is linked to increased expression of IGF2 67242547, as well as an increased susceptibility to chronic diseases 48495051. Similar small effects on DNA methylation have also been associated with the use of assisted reproductive technologies 52, the use of psychotropic drugs during pregnancy 53 and smoking 25. These subtle epigenetic changes have been described as adaptive responses to the environment, while major epigenetic shifts during development would cause more detrimental consequences 25. Furthermore, exploring the CpG sites at the IGF2 and H19 DMRs may represent only a fraction of changes occurring elsewhere in the genome. Environmental factors, among which is diet, have been associated with changes in DNA methylation and may have profound effects on genomic imprinting; accumulation of these effects may result in disturbed metabolic homeostasis 54. Follow-up studies on the anthropometric and other developmental factors of the NEST children are underway to further examine the influence of small changes in DNA methylations at several DMRs on childhood obesity or other adverse consequences. Evidence in animal studies indicates that DNA methylation at the IGF2/H19 locus in sperm might be under tight control of estrogen 55, produced by adipocytes. This suggests a mechanism by which increased exposure to estrogen could lead to inadequate establishment of methylation at the IGF2 DMR in sperm. Alternatively, obesity-related factors may also disrupt functioning of other components of the epigenetic machinery leading to an inability to appropriately establish imprint marks during spermatogenesis, which is ongoing through adult male life 5. Offspring of obese fathers may, therefore, demonstrate incomplete methylation. In order to further explore these hypotheses, more research on the epigenetic effects of obesity on human germ cells is needed.
Although our bivariate analysis did not indicate an association between maternal obesity and DNA methylation at the IGF2 DMR in newborns, the regression analysis showed that controlling for paternal obesity resulted in an opposite effect, meaning that while paternal obesity was associated with a decrease in methylation, maternal obesity tended to be associated with an increase in methylation, but this was only significant at one CpG site. Using pre-pregnancy BMI instead of obesity in our multivariate analysis strengthened this association. At the H19 DMR, the bivariate analysis showed an increase of 4.1% when mothers were obese (P = 0.01). An increase in DNA methylation by maternal pre-gestational BMI has also been reported earlier in cord blood samples, more particularly at the PPARG promoter 56. Hypermethylation at the IGF2 or H19 DMR has been associated with loss of imprinting of IGF2, and several disorders. For instance, IGF2 imprinting defects have been implicated in Silver-Russell syndrome 57, Wilms' tumor 2458, hepatoblastoma 59 and ovarian cancer 7. Our regression analyses at the H19 DMR showed that adjusting for potential confounders, including paternal obesity or BMI, diminished this positive association (Table 2, Model 3). We attribute this to the fact that maternal and paternal obesity are closely related and the fact that methylation outcomes for both parental exposures are in the same direction. A larger study is necessary to further explore the potential impact of parental obesity on DNA methylation at the H19 DMR. Furthermore, the exposure from oocyte stage till birth to maternal obesity or related lifestyles is complex. Hormonal factors that may influence DNA methylation cannot be ruled out. It has been shown that the rat H19 DMR has an estrogen responsive element, suggesting that estrogen can form a complex with Dnmt1, a DNA methyl transferase, leading to DNA methylation at a normally unmethylated maternal allele 55.
A potential limitation of our study includes the use of cord blood as a marker for the newborn's epigenetic status. However, we used isolated leucocytes and IGF2 is a well studied imprinted gene whose germline DMRs should be similarly methylated across all cell types, given the establishment of the epigenetic profile prior to conception. IGF2 and H19 DMR methylation profiles were verified in DNA from different cell fractions from umbilical cord blood and we found no differences across the cell types 26. Another possible limitation is proof of paternity, and the reliability of the paternal anthropometric data, which were reported by pregnant mothers. However, the questions regarding the anthropometric data were detailed and verified for consistency. We do not expect that methylation outcomes are differential with respect to the potential misclassification of exposures. Nineteen percent of our population had missing data about the father. The methylation outcomes of these newborns were not included in our final analytical study group. However, we compared methylation outcomes by missing and non-missing characteristics of the fathers and found no differences. We also compared all measured maternal and newborn characteristics in both groups and most characteristics were similarly distributed. We found no significant differences when mothers were single or not, but found significant differences by education and race. Most missing data were among the lower educated and African American mothers. As far as we could test, education was not associated with obesity in either of the parents, and race was only associated with obesity of the mother. It is very likely we missed a number of obese fathers from African Americans. We cannot verify if paternal obesity is equally distributed in all subgroups of missing and non-missing paternal data, and, therefore, selection bias cannot be excluded. However, when reanalyzing our regression models with race included as an independent variable, our results remained the same; only the effect of maternal obesity attenuated in Model 2 (Table 2) at H19 DMR, the β-coefficient for maternal obesity became +2.18 (SE = 1.65, P = 0.19) (data not shown). Possibly, race is on the same causal pathway as obesity, regarding the effect on methylation outcomes. As mentioned in the results section, race by itself was not associated with methylation outcomes at neither of the two DMRs studied in this cohort. Nevertheless, given our earlier analyses on maternal exposures showed a race dependent effect at the same imprinted loci 53, we further stratified our data by race. The outcomes were in the same direction as the complete analytical cohort, but given the smaller numbers, the results represented unstable estimates. In general, the small sample size remains a limitation in our study which may partly explain why our results at the H19 DMR do not reach statistical significance. Nevertheless, the data concerning associations between IGF2 DMR methylation and parental BMI or obesity reached sufficient power, especially when studying offspring from fathers with high BMI.