Neighborhood and Family Environment of Expectant Mothers May Influence Prenatal Programming of Adult Cancer Risk: Discussion and an Illustrative Biomarker Example

Abstract

Objectives

Childhood stressors including physical abuse predict adult cancer risk. Prior research portrays this finding as indirect through coping behaviors including adult smoking or through increased toxic exposures during childhood. Little is known about potential direct causal mechanisms between early-life stressors and adult cancer. Because prenatal conditions can affect gene expression by altering DNA methylation with implications for adult health, we hypothesize that maternal stress may program methylation of cancer-linked genes during gametogenesis.

Methods

To illustrate, we relate maternal social resources to methylation at the imprinted MEG3 differentially methylated regulatory region linked to multiple cancer types. Mothers (n=489) in umbilical cord blood of diverse birth cohort (Durham, North Carolina) provided newborn’s cord blood and completed a questionnaire.

Results

Newborns of currently-married mothers show lower (−0.321 SD, p<0.05) methylation vs. newborns of never-married mothers, who did not differ from those whose mothers are cohabiting and others (adjusted for demographics). MEG3 DNA methylation levels are also lower when maternal grandmothers co-reside before pregnancy (−0.314 SD, p<0.05). A 1-SD increase in prenatal neighborhood disadvantage also predicts higher methylation (−0.137 SD, p<0.05).

Conclusions

Maternal social resources may result in differential methylation of MEG3, which demonstrates a potential partial mechanism priming socially disadvantaged newborns for later risk of some cancers.

Introduction

Adverse social conditions during early life are known to associate with increased cancer incidence and mortality in adulthood, but the mechanisms are poorly understood. For instance, childhood stressors including physical and sexual abuse (Brown et al., 2010; Fuller-Thomson & Brennenstuhl, 2009; Goldsmith et al., 2010; Kelly-Irving et al., 2013; Morton, Schafer, & Ferraro, 2012), father absence or violence (Sobrinho et al., 2012), and a large sibship (Smedby et al., 2007) all predict increased cancer incidence. Excess cancer cases among those with adverse childhood experiences have been explained as either due to an increased risk of toxic or infectious exposures during childhood (Montgomery, Ehlin, Sparen, Bjorksten, & Ekbom, 2002; Sandler, Everson, Wilcox, & Browder, 1985), or to carcinogenic health behaviors (Maynard, Gunnell, Emmett, Frankel, & Davey Smith, 2003) and emotional coping mechanisms throughout the lifecourse (Boynton-Jarrett, Rich-Edwards, Jun, Hibert, & Wright, 2011; Clark, Beesley, Holcombe, & Salmon, 2011; Williams et al., 2012) resulting from problematic conditions during childhood (Colditz & Wei, 2012). However, the link between adverse childhood experiences and adult cancer risk can only partly be explained by health and psychosocial behaviors in epidemiological models (Brown et al., 2010).

Another possibility is that the prenatal environment has long-lasting physiological effects on adult cancer risk, as the fetal origins hypothesis (Barker, 1995) proposes for cardiovascular disease risk. The last several years have seen an evolution in understanding of the etiology of cancer, of epigenomics more broadly, and more specifically the role of epigenetic mechanisms in regulating carcinogenesis (Dawson & Kouzarides, 2012). Still, the role of the prenatal social environment highlighted by the fetal origins literature (Barker, 2004) and others (Copper, Goldenberg, Das, & al., 1996; Farley et al., 2006; Lobel, Dunkerl-Schetter, & Scrimshaw, 1992) has heretofore been underappreciated in cancer epidemiology (e.g. (Hiatt & Breen, 2008)). Our view is that the role of epigenetics in guiding both normal and abnormal growth provides a mechanism for social differences in toxic exposures which influence perinatal epigenetic development to play a role in the social gradient in cancer risk in adulthood.

Meaney, Szyf, and colleagues explain the epigenetic mechanism by which stress response is transmitted in early care as due to a cascade of effects beginning with the release of serotonin in offspring’s’ brains during grooming (Szyf et al., 2007). This is consistent with prenatal transmission in humans in at least three ways. First, the relationship between maternal social resources and methylation in our study could occur if the newborn largely inherited the mother’s epigenome, and the mother’s epigenome influenced the mother’s family composition. A good relationship between mother and grandmother during the mother’s childhood could lead both to lower methylation through the already proposed mechanism, and also to coresidence with mother (and putting marriage before childbirth) later on. Second, given that human gestation is long and ‘bonding’ between mother and child begins in utero, the release of serotonin during interaction with the mother (and perhaps caresses from father, grandmother, and others) could already have begun during gestation. Third, the methylation of MEG3 is not known to be similar to or differ from the mechanism described by Meaney and colleagues, except that accrual of additional methylation normally occurs on both the paternally- and maternally-derived chromosome post-fertilization. Finally, social resources appear to differentially relate to DNA methylation at different promoter sites (King et al., 2015), but the reasons why are not yet known.

Family members’ roles in encouraging healthy behaviors such as prenatal vitamin use, maintaining a healthy weight, avoiding smoking and other toxic exposures and consuming a healthy diet may also be crucial (Sear & Mace, 2008). Husbands contribute to early initiation of prenatal care (Albrecht & Miller, 1996; Kimbro, 2008; Redshaw & Henderson, 2013), and early prenatal care initiation is generally considered to be protective against poor perinatal health (Goldenberg, Culhane, Iams, & Romero, 2008). Folate is a key dietary source of methyl groups for methylation reactions. Methylation varies by folate levels (Haggarty et al., 2013; Hoyo, Murtha, Schildkraut, Forman, et al., 2011), and folic acid supplementation varies substantially within a population (Hoyo, Murtha, Schildkraut, Jirtle, et al., 2011). Nutrition can also remedy epigenetic irregularities in adulthood, at least in mice (Weaver et al., 2005), so the extent to which parental nutrition that favors optimizing the epigenome is passed to offspring may be better in families. In a bivariate model, maternal folate level predicted MEG3 methylation, but this association was not significant when added to the full model (not shown).

However, links between family composition and perinatal health can be complex. For instance, unmarried women and those in high-conflict/abusive relationships are at greater risk of prenatal smoking (Kimbro, 2008). In turn, smoking during pregnancy partially accounts for an effect of never-married status on birth weight (Kane, 2012), supporting the notion that social support mechanisms can influence growth of a fetus. Prenatal smoking also influences the epigenome (Breton et al., 2009; Joubert et al., 2012; Murphy, Adigun, et al., 2012). Meanwhile, obese people are less likely to have a partner (Averett, Corman, & Reichman, 2013; Kane & Frisco, 2013; Metwally, Li, & Ledger, 2007), and (both maternal and paternal) obesity is associated with altered methylation (Soubry et al., 2013). Stress (Hoffmann & Spengler, 2012) and mental health (Galea, Uddin, & Koenen, 2011; A. Soubry et al., 2011) also have epigenetic facets which may relate to family conditions.

The epigenome has been prominently endorsed as a key link between neighborhood disadvantage and health disparities (Olden, Olden, & Lin, 2015). Neighborhood physical and social conditions can causally influence prenatal health, and neighborhoods also serve as a powerful proxy for household wealth, household composition, and for selection processes which sort mothers and children into other sorts of spatially-variant individual-level health risks and resources. The present analysis may be the first to demonstrate epigenetic variation by neighborhood.

In this study, we hypothesize that a potential additional and more proximal pathway between prenatal social conditions and adult cancer may involve DNA methylation. Below we discuss how perinatal maternal social context may also alter the epigenome in ways which may influence cancer risk in adulthood, giving illustrative empirical evidence of such a mechanism. Recent research has revealed that childhood adverse social circumstances are often accompanied by changes in epigenetic regulation and gene expression (Mehta et al., 2013). Stress, depression, and social isolation can alter neurochemical balance (Lutgendorf et al., 2011), hormone status (Kalantaridou, Makrigiannakis, Zoumakis, & Chrousos, 2004), immune response (Salim, Chugh, & Asghar, 2012; Sanna et al., 2013), gene expression patterns (Cole et al., 2010; Volden et al., 2013), and cancer aggressiveness (Moreno-Smith, Lutgendorf, & Sood, 2010). DNA methylation status of the glucocorticoid receptor gene differs in adolescent children by their mother’s experience of intimate partner violence during pregnancy (Radtke et al., 2011) and also differs in suicide victims by history of childhood abuse (McGowan et al., 2009). Glucocorticoid receptor genes are often hypermethylated in breast tumors (Nesset, Perri, & Mueller, 2014). A case-control study of abused/neglected children found significantly different methylation at 2868 CpG sites, including genes involved in lung, colon, breast, prostate and ovarian cancers (Yang et al., 2013). Studies in mice show an increased risk of breast cancer in pups separated for extended periods from their mothers postnatally (Schuler & Auger, 2010), indicating that psychosocial stressors may influence cancer risk during multiple developmental windows.

Some of the strongest evidence for social factors in epigenetic programming relates to effects of prenatal stress and early maternal care on stress responses (Champagne & Curley, 2011; Szyf, Weaver, & Meaney, 2007), documented by numerous animal and human studies, in vitro, and for over 900 genes (Weaver, Meaney, & Szyf, 2006). Prenatal stress is important here as it is often found to be inversely related to mother’s level of prenatal emotional support{Landale, 2001 #6829}. Prenatal stress can also inhibit neurological development (McEwen, Eiland, Hunter, & Miller, 2012), specifically related to the HPA axis (Kapoor, Dunn, Kostaki, Andrews, & Matthews, 2006; Welberg & Seckl, 2001) and hippocampus (Maccari & Morley-Fletcher, 2007), both of which are linked with stress responsiveness. Also, male mouse pups prenatally exposed to stress differed in epigenetic markers of stress response, specifically methylation of central corticotropin-releasing factor and glucocorticoid receptor genes (Mueller & Bale, 2008).

These and similar data now constitute an emerging model where variations in one’s social and physical environment may affect health outcomes through changes in the epigenome at multiple life stages, changes which likely endure across the lifecourse and may be passed to future generations. Maternal smoking (Joubert et al., 2012; Murphy, Adigun, et al., 2012) and depression also predict methylation patterns (A Soubry et al., 2011) in numerous regions including an association between maternal severe depressed mood and a 2.4% higher methylation of MEG3 (the present outcome) (Liu et al., 2012). Another study found differential methylation of MEG3 by maternal smoking in utero, but did not adjust for predictors of smoking, including stress or social support (Markunas et al., 2014). Depression and tobacco smoke exposure are risk factors for cancer (Sandler et al., 1985; Zonderman, Costa, & McCrae, 1989). Moreover, there is also a longer history of research linking early life family climate and socioeconomic status (SES) with health – through what are deemed to be epigenetic processes – but without identifying specific epigenetic markers (e.g. Gregory E. Miller & Chen, 2010; G. E. Miller, Chen, & Parker, 2011). No known prior research has linked maternal social resources with epigenetic transmission of risk factors for cancer.

While understanding of social-epigenetic links is emerging, genetic instability clearly both causes and contributes to the development of cancer, and epigenetics appear to as well. Cancer is a process of abnormal cell growth and death, with DNA altered at various genes in tumor cells and microenvironments. MEG3 (maternally expressed gene 3) is a genomically imprinted gene from which is transcribed a long non-coding RNA (Benetatos, Vartholomatos, & Hatzimichael, 2011). MEG3 can activate one of the most well-known tumor suppressor genes, p53 (Zhou, Zhang, & Klibanski, 2012; Zhou et al., 2007), and lower levels of the p53 inhibitor, MDM2 (Benetatos et al., 2011), as well as regulating senescence genes (Gritz & Bhandari, 2015). MEG3 gene expression levels are inversely associated with incidence of pituitary adenomas, renal carcinoma, multiple myelomas, meningiomas, bladder cancer, and hepatocellular carcinoma (Anwar et al., 2012; Benetatos et al., 2008; Benetatos et al., 2011; Kawakami et al., 2006; Ying et al., 2013; Zhang et al., 2010; Zhou et al., 2012). In many human cancer cell lines, MEG3 RNAs are not detectable (Benetatos et al., 2011). In ovarian cancer tissues, loss of MEG3 expression is linked to high MEG3 promoter methylation (Sheng et al., 2014). Alteration of the genome in cancer cells occurs through multiple mechanisms (loss- and gain-of-function mutations, amplifications, deletions, epigenetic deregulation, etc.) Researchers are still learning more about how loss of the integrity of epigenetic mechanisms, such as methylation of MEG3, alters the function of other tumor suppressor genes.

MEG3 is maternally expressed, meaning that only the maternal allele is actively transcribed. Methylation present on the paternally-derived allele contributes to silencing of this gene. Theoretically imprinted gene regulatory regions have methylation levels that are close to 50%, since only one of the parental alleles carries the methylation marks. Since methylation averaged 72.3% for MEG3 in umbilical cord blood, and MEG3 methylation is approximately normally distributed, the maternally-derived chromosome also carries partial methylation in this tissue, suggesting that the increased methylation in umbilical cord blood may function to repress MEG3 expression in leukocytes.

This paper proposes a potential additional and direct mechanism between early-life social conditions and adult cancer and demonstrates that involves maternal prenatal social stress/support is associated with MEG3 tumor suppressor DNA methylation. The present analysis examines prenatal maternal social (e.g. relationship status and coresidence with her mother) and neighborhood sociodemographic resources in relation to methylation of the MEG3 differentially methylated region in cord blood in a sizeable (n=489) birth cohort in Durham, NC. The central contribution of this study is that we assess, for the first time, the role of maternal prenatal social resources and neighborhood disadvantage in the epigenetic programming of the child related to cancer risk.

Materials and Methods

Study Design

The Newborn Epigenetic STudy (NEST) recruited pregnant women from six prenatal clinics in Durham, North Carolina (Liu et al., 2012; Vidal et al., 2013). Participants were enrolled during their first prenatal clinic (mean gestational age ~13 weeks). Questionnaire and peripheral blood were collected at enrollment, and offspring and maternal blood specimens and parturition data were collected at birth. Eligible expectant mothers were ≥18, literate, without known HIV infection, intending to deliver in one of the participating obstetric facilities (Duke and Durham Regional Hospitals), and to retain custody of the child locally for ≥3 years. NEST Wave II (2009–2011), approached 2548 pregnant women and 67% (n= 1700) agreed to participate and were consented. Umbilical cord blood was successfully collected in 1304. DNA methylation of umbilical cord blood leukocytes was evaluated for nine DMRs of imprinted genes in the first 619 newborns early in Wave II, the only wave in which household composition was reported. Among remaining mothers, methylation of the promoter region of MEG3 was evaluated by bisulfite pyrosequencing of umbilical cord blood DNA for 518 newborns, as some data were lost at random due to laboratory processing. Home addresses were geocoded, yielding sufficient data for 489 newborns. The study protocol was approved by the Duke University Institutional Review Board.

Umbilical cord blood is a readily available fetal tissue and has the advantage of being naïve to the external environment; thus any epigenetic alterations that occur are a direct result of the context of the in utero environment. Epigenetic marks do vary by cell and tissue type, but studies of buccal cells and umbilical cord blood mononuclear and polymorphonuclear cells in NEST (Joubert et al., 2012; Murphy, Huang, & Hoyo, 2012) did not detect cell type differences in DNA methylation marks in this region.

Variables and Measurement

Maternal relationship status was coded as: never married, currently married (reference category), cohabiting with partner, and other (e.g. divorced/separated, widowed, other). Coresidence of mother’s mother in household was coded as: present or not present. We adjust for individual SES of the mother as it may predict household composition and residential disadvantage, and also DNA methylation. Maternal race was categorized as Black, Asian, Other, and White (reference category), and ethnicity as Hispanic or non-Hispanic. At present, it is not clear how ancestral DNA may relate to DNA methylation, and we lack DNA methylation markers of race/ethnicity, so we treat race/ethnicity as a social measure. Maternal household income was categorized as ranging from less than $25,000; $25,000–50,000 (reference category); or more than $50,000. Mother’s years of schooling ranged from 1–20. Newborn gender was coded as: female or male. DNA methylation can vary by sex even outside sex chromosomes, but the reasons why are not yet fully understood. Maternal antibiotic use and smoking periconceptionally or during pregnancy were each coded as: yes or no. Maternal pre-pregnancy body mass index was coded as ranging from less than 18.49 (underweight), 18.5–24.9 (normal), or over 25 (overweight). Missing data was represented by categorical variables.

Mothers’ residential addresses were geocoded to match 2010 Census tracts. Adapting a respected procedure (King, Morenoff, & House, 2011; Sampson, Raudenbush, & Earls, 1997), neighborhood disadvantage was assessed as a principal components factor of six measures of tract social composition: % non-Hispanic Black, % of families with income below the poverty level, % of households on public assistance, % households with an unmarried female head, % population under age 18, and % of the civilian labor force over age 16 unemployed. Using national data to facilitate comparability with other research, the factor analysis included 73,097 Census tracts with non-missing data in the 50 states, the District of Columbia, and Puerto Rico. One factor with an eigenvalue of 3.2 (above the standard cutoff value of 1) was retained and standardized to the mean. Factor loadings and descriptive statistics are given in Supplementary Table 1.

DNA Methylation Analysis

Measurement and analysis of MEG3 methylation was performed as described previously (Murphy, Huang, et al., 2012). Briefly, bisulfite pyrosequencing was used to quantitatively assay the level of methylation at CpG sites within a differentially methylated region (DMR) of the DLK1/MEG3 imprinted domain on chromosome 14q32.2.

Reported values represent the mean methylation for the CpG sites contained within the sequence analyzed. Validation of the pyrosequencing assays was performed using predetermined mixtures of fully methylated/unmethylated Epitect control genomic DNAs (Qiagen) (Murphy, Adigun, et al., 2012).

Statistical Analysis

We reported frequencies and percentages of sociodemographic variables, along with summary statistics and analysis of variance (ANOVA) statistics on how MEG3 methylation varies by sociodemographic group. We used a multilevel model (Snijders & Bosker, 1999) to estimate associations of social circumstances with MEG3 methylation, adjusting for clustering of residences within Census tracts.

Results

The sample was diverse in terms of mother’s race and ethnicity, relationship status, and education, and coresidence of maternal grandmother (Table 1). Methylation of the MEG3 promoter region DMR averaged 72.3% and showed significant (p<0.05) unadjusted differences in means across maternal race, ethnicity, relationship status and coresidence of mother’s mother. Tract mean levels of disadvantage for NEST mothers were 0.55 SD higher than for the nation overall.

Table 1.

Summary Statistics of Participants and Associations with MEG3 Methylation

	Sample		MEG3 Methylation
	N	%	Mean	SD	Min	Max
Relationship Status
Never Married	136	27.8	73.4	5.3	56.6	85.3***
Cohabiting	133	27.2	72.7	5.6	57.1	85.0
Married	189	38.7	70.9	5.3	56.0	84.3
Other	31	6.3	74.5	6.2	60.2	82.5
Mother’s Mother Present
Yes	61	12.5	72.0	5.9	57.1	85.2
No	428	87.5	72.3	5.5	56.0	85.3
Mother’s Race
White	240	49.1	71.7	5.6	56.0	85.0+
Black	193	39.5	73.1	5.4	56.6	85.3
Asian	10	2.0	73.3	4.7	64.5	79.3
Other	36	7.4	71.9	5.2	62.1	85.2
Mother’s Hispanic Origin
Hispanic	139	28.4	73.3	5.3	61.0	85.2*
Non-Hispanic	350	71.6	71.9	5.6	56.0	85.3
Mother’s Household Income
$0–24.9K	198	40.5	72.8	5.2	57.1	85.3***
$25–49.9K	56	11.5	70.8	6.3	56.0	82.2
≥$50K	129	26.4	71.2	5.3	57.9	84.3
Missing	106	21.7	73.6	5.6	56.6	85.2
Mother Smoked During Pregnancy?
Yes	77	15.8	73.3	0.6	72.1	74.5+
No	412	84.3	72.1	5.6	56.0	85.2
Antibiotics Used During Pregnancy?
Yes	123	25.2	72.1	5.5	56.6	85.3
No	366	74.8	72.4	5.6	56.0	85.2
Mother’s Pre-Pregnancy BMI
Underweight	13	2.7	69.9	6.0	56.0	78.2
Normal	168	34.4	72.1	5.7	57.9	85.3
Overweight	227	46.4	72.2	5.5	56.6	85.0
Missing	81	16.6	73.4	5.4	61.1	85.2
Newborn Sex
Female	244	49.9	72.8	5.3	57.9	85.3
Male	245	50.1	71.8	5.7	56.0	85.0

	Mean	SD
% MEG3 Methylated	72.3	0.3
Mother’s Years of Education	12.9	0.2
Tract Disadvantage (Std. to National Mean)	0.550	0.051

^***p<0.001;

^**p<0.01;

^*p<0.05;

⁺p<0.1;

Newborn Epigenetic Study, 2010–2011 (N=489)

Table 2 presents the results of a regression model predicting methylation of MEG3. The model considers individual-level maternal social circumstances in a hierarchical linear (multilevel) framework (the xtmixed command in Stata 13.0 (StataCorp, 2011)) which includes neighborhood variables while considering the potential that residents of the same tract are more similar than the overall study population. Specifically, we examine how neighborhood disadvantage and maternal household composition, as indicators of maternal social resources, might relate to MEG3 methylation. The model adjusts for factors previously associated with DNA methylation: maternal race/ethnicity, sex of newborn, mother’s years of schooling, maternal household income, pre-pregnancy body mass index, and cigarette smoking and use of antibiotics during pregnancy.

Table 2.

Maternal and Neighborhood Predictors of MEG3 Methylation, Hierarchical Linear Model with Neighborhood Random Effects

	Coef.	SE	p-value
Neighborhood Disadvantage	0.76	0.24	0.002**
Relationship Status (ref=Married)
Never Married	1.79	0.84	0.033*
Cohabiting	0.65	0.75	0.387
Other	2.41	1.18	0.041*
Mother’s Mother Present	−1.75	0.81	0.030*
Covariates
Mother’s Race (ref=White)
Black	1.64	0.76	0.031*
Asian	2.93	1.76	0.096+
Other	−0.09	0.97	0.040
Mother is Hispanic	2.44	0.90	0.007**
Mother’s Years of Schooling	0.03	0.09	0.752
Mother’s Household Income (ref=$25–49.9K)
$0–24.9K	0.82	0.87	0.345
≥$50K	1.75	0.92	0.057+
Missing	1.41	0.99	0.157
Mother Smoked During Pregnancy	0.72	0.73	0.326
Antibiotics Used During Pregnancy	0.15	0.58	0.789
Mother’s Pre-Pregnancy BMI (ref=Normal)
Underweight	−1.69	1.55	0.277
Overweight	−0.56	0.57	0.326
Missing	−0.20	0.83	0.805
Newborn is a Girl	0.87	0.50	0.079+
Constant	68.11	1.85	0.000***

^***p<0.001;

^**p<0.01;

^*p<0.05;

⁺p<0.1;

Newborn Epigenetic Study, 2009–2011 (N=489)

Neighborhood disadvantage was associated with significantly higher methylation (β= 0.76 SD, p<0.01). Newborns with never-married (β= 1.79 SD, p<0.05) and divorced/widowed/other (β= 2.41 SD, p<0.05) (vs. married) mothers and with coresiding maternal grandmothers (β= −1.75 SD, p<0.05) have significantly different MEG3 methylation levels. Newborns with Black (β= 1.64 SD, p<0.05) (vs. White) and Hispanic (β= 2.44 SD, p<0.01) mothers also had significantly higher methylation levels, as did Asian (vs. White) mothers, with marginal significance (β= 2.93, p<0.1). These results are consistent with a pattern of lower methylation of MEG3 in socially advantaged mothers (e.g. non-Hispanic White, married, living with mother, advantaged neighborhoods). ⁱ

Discussion

Understanding how social factors can mold risk of disease will move forward by identifying biological intermediaries of a downstream outcome (e.g. “diagnosis”) which are influenced by the same social factors as the downstream outcome (G. Miller, Chen, & Cole, 2009). In this paper we present evidence relevant to the hypothesis that a potential additional and more proximal pathway between prenatal social conditions and adult cancer may involve DNA methylation. DNA methylation varies in cancer. We demonstrate that early life social conditions can be related to cancer-relevant DNA methylation as early as birth, before the child’s own behavior can come into play. These findings lend credence to a role of maternal prenatal social resources in the epigenetic programming of the child related to cancer risk.

This is the first known finding of differential methylation in socioeconomically disadvantaged neighborhoods, and joins extensive evidence linking neighborhoods with health. Indeed, neighborhood SES accumulates over the life course and childhood neighborhood SES remains predictive of adult health even after adjustment for extensive controls (Clarke et al., 2013). Several studies link neighborhood features, including SES, population density, and air pollution, with cancer incidence in adults and children (Borugian, Spinelli, Abanto, Xu, & Wilkins, 2011; Ghosh et al., 2013; Meijer, Bloomfield, & Engholm, 2013; Reynolds et al., 2002; Schootman et al., 2010). Air pollution exposure during gestation predicts birth outcomes (Stieb, Chen, Eshoul, & Judek, 2012), perhaps by influencing the epigenome (Janssen et al., 2013). Neighborhood SES also predict birth outcomes (Buka, Brennan, Rich-Edwards, Raudenbush, & Earls, 2003; Culhane & Elo, 2005; Messer, Kaufman, Dole, Savitz, & Laraia, 2006; Reichman, Teitler, & Hamilton, 2009; Schempf, Kaufman, Messer, & Mendola, 2011), likely both as an additional indicator of wealth, and also because of risks, resources, and norms which assort by neighborhood and influence fetal development. Given “remarkable continuity in neighborhood economic status from one generation to the next” (Sharkey, 2008), how neighborhood resources/risks may influence epigenetic programming in early life deserves additional attention.

We interpret race/ethnic differences as likely social in origin rather than linked to genetic ancestry, although we do not adjust for genetic population structure. A prior analysis of NEST data (King, Murphy, & Hoyo, 2015) found that race/ethnic differences in MEG3 DMR methylation lost significance after adjustment for maternal education, household income, and paternal race/ethnicity, but in the present framework adjusting for father’s race/ethnicity did not affect the pattern of results; father’s race/ethnicity was not significant, even when mother’s race/ethnicity was removed from the model (not shown). Some other studies have found differences between populations in epigenome-wide DNA methylation (e.g. (Fraser, Lam, Neumann, & Kobor, 2012)), but these studies have not always considered the very different physical and social contexts these populations live in and how the environment may influence the epigenome. Evidence of group differences is not enough, as both parents and grandparents may have experienced differential toxic and social exposures by social group. In addition, there should be limited heritability of DNA methylation between mothers and offspring given the complete erasure and re-establishment of DNA methylation (Kile et al., 2010; McRae et al., 2014) marks in the gametes and early embryo. Thus, additional evidence is needed in evaluating ancestry-linked epigenetic differences between “race” groups, including specific genes or gene variants that influence methylation. Any ancestry-linked epigenetic differences may not follow the race/ethnic lines as socially constructed in the United States. In addition, these racial lines are blurred in our cohort in that 25% of our newborns and 7% of our mothers have parents for whom different race groups were reported or for whom father’s race/ethnicity was left blank. Furthermore, White mothers in our sample lived in neighborhoods which averaged .28SD above the national overall mean social disadvantage, while Black mothers’ neighborhoods averaged .95SD above the overall national mean.

This is a noteworthy model of special concern to etiologic investigation and subsequent public health policy, but there is also reason for caution. Although DNA methylation at the DLK1/MEG3 imprinted domain has been linked to both changes in allele-specific gene expression and hepatocellular carcinoma, it is a single region and unlikely, by itself to causally alter cancer risk. This study did not directly link cancer outcomes with either prenatal social circumstances or MEG3 methylation, and we do not claim that MEG3 methylation is a central cancer concern. Rather, we chose MEG3 as one of a few maternally expressed genes whose expression and methylation status have been linked with cancer and which have been evaluated by pyrosequencing with social variables available in a sizeable diverse geographically-clustered sample. While a growing body of prior research links early life social circumstances with epigenetic programming of stress responses (Gudsnuk & Champagne, 2012; Heijmans et al., 2008; Isles, Davies, & Wilkinson, 2006; Kaati, Bygren, Pembrey, & Sjostrom, 2007; McGowan & Szyf, 2010), this is the first study we know of which links prenatal social circumstances and epigenetic programming of cancer risk, and both replication and further theoretical elaboration are needed.

In addition, the finding that household structure may predict epigenetics should not necessarily be interpreted causally. While father presence can be beneficial (Cunningham, Elo, Herbst, & Hosegood, 2010), expectant mothers have likely sought the best possible circumstances available to them, and if they have not married the father of their child these findings are not an argument that they should do so. There are also social factors, such as mother’s race, education, family background (i.e., parental SES, childhood environment, family instability), health behaviors, and risk exposures which can simultaneously affect her relationship status (and thus indirectly birth outcomes) but also directly affect birth outcomes. Studies accounting for this type of confounding have shown that these underlying mechanisms account for about half of the association between marital status and birth weight, for example (Buckles & Price, 2013; Kane, 2012). We account for some of these social resources in our study (race, education), but it is plausible that at least a portion of the association we observe between marital status and methylation reflects the effect of earlier life social resources on methylation, such as family background. Also, given the sample and association sizes, the specific findings need confirmation in other datasets. We believe no suitable replication dataset exists at present, so the associations cannot be independently confirmed. How DNA methylation relates to maternal social conditions might vary by temporal or spatial context, for instance if another difference between the study sites (e.g. toxic exposure, folate consumption) altered these associations or if the implications of social conditions changed. However, our goal is not to prove a link between maternal social conditions and either MEG3 methylation or future cancer risk. Rather, we wanted to draw attention to DNA methylation in the relationship between social conditions in utero and later life health, and give an example of how conditions in utero can affect a biomarker associated with cancer. Future studies that collect an extensive social history of mothers would be better able to examine how social relationships relate to children’s DNA methylation. More broadly, future research can build on this study by delving into the complexities of the causal processes involved, some of which may even involve macro-level processes, or paternal epigenetic factors as well as maternal epigenetic factors (Richardson et al., 2014).

Even though we did not link DNA methylation with health outcomes and genetic expression in this study, previous research has shown that MEG3 expression can be strongly affected by its methylation status and is downregulated in a number of different histological subtypes of cancer (Anwar et al., 2012; Benetatos et al., 2008; Benetatos et al., 2011; Ying et al., 2013; Zhang et al., 2010; Zhou et al., 2012). It is important to note though, that methylation is not the only established mechanism by which MEG3 expression may be regulated (Braconi et al., 2011; Zhao, Zhang, Zhou, Ansell, & Klibanski, 2006). This is just one of untold pathways to cancer. Still, these findings when taken together with existing literature support a model where maternal prenatal social resources potentially cause changes to the epigenetic landscape and thus to future health outcomes such as cancer. Further research is needed to replicate these findings and establish both causality and the range of genes for which perinatal social stressors predict methylation. This would suggest one concrete biological mechanism for observed enduring effects of poverty and social disorder in childhood on adult health and across generations.

This research takes place at a time when understanding of the role of epigenetic factors in disease etiology is still evolving, but links between MEG3 and cancer outcomes are beginning to crystallize. Few studies have measured methylation of the MEG3 DMR, and even fewer have included social measures in a sizeable sample. This makes the present study unique, but also makes it difficult to seek to replicate or extend at present. Future research should consider that mothers often move or change household composition before, during, or shortly after pregnancy, so that living conditions are dynamic and evolving. Family relationships are complex and relatives tend to offer both support and stress, so that measures of relationship quality would also add to the picture.

Despite the noted shortcomings of the present study, these data are consistent with a burgeoning literature that could have profound implications in health disparities research. Specifically, maternal social support during gestation appears related to DNA methylation of MEG3 in a way that opens a line of investigation into how early life conditions influence later cancer risk. Such a model would effectively aid in substantiating one mechanism for the disturbing observation that social environmental and inequality tends to result in health effects that may persist for generations. Therefore, further research into this potentially paradigm-shifting model should be an important concern to future sociological and epidemiological studies.