Following Robert Waterland’s and Andrea Baccarelli’s conceptual overview, workshop participants in Session 2, moderated by Karen Lillycrop of the University of Southampton, considered how the risk of childhood obesity can be affected by (1) maternal and paternal nutrition and other exposures before conception, (2) maternal and placental nutrition and health during pregnancy, and (3) postnatal maternal and infant nutrition and health. This chapter summarizes the Session 2 presentations and discussion.
“Obesity begets obesity,” began Jacob Friedman of the University of Colorado, Denver. Friedman discussed animal and human data demonstrating that both prenatal and postnatal exposure to maternal obesity predispose infants to early-onset metabolic disease and childhood obesity. For example, studies conducted on obese pregnant mothers and their newborn infants indicate that pre-pregnancy body mass index (BMI) of the mothers, but not infant fatness, is predictive of higher liver fat at 2 weeks of age, which continues to increase during lactation. Friedman explained that excess maternal fuels in obese mothers crossing the placenta have nowhere to go but into the fetal liver, and suggested that this fatty liver transgenerational effect of maternal obesity may be accompanied by epigenetic changes in offspring liver and other tissues at 1 year of age. Postnatally, evidence from breastfeeding mothers indicates that maternal diet and obesity may also affect immune function and by different mechanisms have an impact on infant behavior, weight gain, and obesity risk.
Linda Adair of the University of North Carolina described inequities among different socioeconomic groups that cut across a range of obesity-related prenatal and postnatal exposures and outcomes. Examples of significant prenatal disparities among different socioeconomic groups include differences in parental overweight and gestational weight gain. Examples of significant postnatal disparities include differences in overweight and obesity in children under the age of 5. Adair noted that the fastest-growing rates of childhood obesity worldwide are in low-income groups. Adair also introduced discussion of the “mismatch hypothesis,” that is, the notion that the risk of childhood obesity appears to be greatest among undernourished fetuses. The mismatch hypothesis was revisited several times later in the workshop.
While most of the workshop presentations and discussion focused on maternal exposure and its effect on the risk of childhood obesity, Stephen Krawetz of the Wayne State University School of Medicine shifted the focus to, in his words, what “Dad delivers.” He provided an overview of the role of sperm in early development; described the many different types of RNA molecules that sperm deliver to the oocyte, including several that have been implicated as having an early developmental role in obesity; and discussed evidence of epigenetic-mediated transgenerational inheritance through the paternal line.
Caroline Relton of Newcastle University emphasized that both over-nutrition and under-nutrition during pregnancy can impact childhood adiposity. In her opinion, the evidence is compelling. The question for her is: What is the role of epigenetics? Relton stated that while identifying associations between methylation patterns and phenotypes has become straightforward, inferring causality remains a challenge. Revisiting and expanding on ideas introduced by Andrea Baccarelli in the previous section, Relton laid out the steps necessary to infer causality, gave numerous examples, and considered ways to improve those steps. Among other suggestions, she called for more refined measures of maternal obesity, an increased awareness of the pitfalls of association studies, and the use of triangulation among multiple studies to infer causality.
Maternal obesity in the United States has reached a point that compels Jacob Friedman to ask, “How could this not impact infant outcomes?” An estimated 60 percent of women between the ages of 20 and 39 years are overweight or obese (BMI > 25), with 33 percent being obese (BMI > 30)
1 This section summarizes information and opinions presented by Jacob Friedman, Ph.D., University of Colorado, Denver.
and about 8.5 percent severely obese (BMI > 40) (Ogden et al., 2014). By race and ethnicity, almost 80 percent of Hispanic and black women in that age range are overweight or obese. There are known associations between maternal obesity and early-onset obesity, metabolic syndrome, and fatty liver and cardiovascular disease in children (Brumbaugh et al., 2013; Lawlor et al., 2011; Smith et al., 2009). Epidemiological data suggest that the maternal obesity effect on the offspring is not confined to neonatal life, affecting offspring across the life span, independent of lifestyle factors (Pirkola et al., 2010). However, surprisingly little is known about how maternal obesity may influence obesity risk in the human neonate.
Both animal and human studies have shown signs of aberrant methylation patterns—as well as mitochondrial dysfunction—in children born to mothers who are obese (Borengasser et al., 2013). Researchers have reported methylation changes in cord blood or in the placenta associated not just with intrauterine exposure to maternal obesity, but to maternal diabetes and famine as well (El Hajj et al., 2014). In Friedman’s opinion, the epigenetic changes associated with intrauterine exposure to maternal obesity are probably tissue-specific, although it is not clear to what extent. The greater question for him is whether the epigenetic associations are causal. Regardless of causality, in a 2011 editorial based on a study showing that the methylation status of specific genes in human cord blood predicted subsequent development of childhood obesity, Friedman and his coauthor considered whether epigenetic analysis at birth may be useful for identifying future risk of obesity (Choudhury and Friedman, 2011).
It is not “just mom” that is affecting the infant epigenome, Friedman said. The placenta matters as well. The placental transcriptome can be highly different between lean and obese patients, with many inflammatory pathway genes being turned on early in the placenta in mothers who are obese (Basu et al., 2011). Fetuses derive their signals from the placenta, with their epigenome being informed by whatever crosses the placental barrier and with outcomes ultimately playing out in the infant metabolome, epigenome, and proteome.
Friedman’s research group has focused many of its efforts on what obese women are eating when they become pregnant and how their diet affects fuel production, adipose tissue mass, and inflammation in the neonate. Using magnetic resonance imaging (MRI) data, he and his team are able to examine where the fat is accumulating in the infant. Additionally, by harvesting umbilical cord-derived mesenchymal stem cells from newborn infants exposed to maternal obesity, they are also able to detect epigenetic signatures associated with maternal obesity. Mesenchymal-derived stem cells (MSCs), Friedman explained, are a population of stem cells in the umbilical cord that can be programmed to develop into either myocytes or adipocytes, depending on what the stem cells are exposed to (Gang et al.,
2004; Janderová et al., 2003). MSC differentiation to either adipocyte or myocyte can be important for proper tissue development in utero, in particular because there is a large window where adipogenesis and myogenesis overlap during fetal development.
Kristen Boyle, a member of Friedman’s research team, has shown that obese women have higher pre-pregnancy BMI, higher homeostasis model assessment-estimated insulin resistance (HOMA-IR) indexes, and higher lipid levels. Because all of those features affect the placenta, Friedman said, his research team has hypothesized that infants born to obese mothers are predisposed to early-onset metabolic disease due to “metabolic programming” events before birth resulting from fuel overload, impaired mitochondrial energy metabolism pathways (i.e., fatty acid oxidation and amino acid metabolism), and, ultimately, changes in the epigenome. Boyle and colleagues (data presented at the American Diabetes Association meeting, 2015) showed that epigenetic signatures associated with myocyte versus adipocyte differentiation are affected by maternal BMI, with suppressed expression of both epigenetic regulators, DNMT1 and KDM6A, in the mesenchymal stem cells of infants born to obese women. Boyle and colleagues also showed that the stem cells from infants born to obese mothers had an increased adipogenic potential that correlated with percent fat of the infant. The pathways or mechanisms responsible for less lean mass and greater fat mass established during gestation are not well understood. Maternal diet and obesity impact fuels, hormones, and inflammation with powerful effects on fetal metabolic systems. Delving deeper into the mechanisms and molecules that differ in these cells on the basis of poor maternal health, including the potential epigenetic regulation of these differences, is an important area of future investigation and may hold the key to understanding how nutritional programming can lead to a susceptibility to obesity in adult life.
Metabolic Programming in the Fetus: Is It a Matter of Fat?
Friedman has been collaborating with Kevin Grove at the Oregon National Primate Research Center to develop a nonhuman primate model to study the effects of maternal diet, maternal obesity, and gestational diabetes mellitus on the development of metabolic systems in utero and on infant behavior and postnatal disease pathways.
A key finding from these studies is that the livers of fetuses from mothers fed a high-fat diet prior to conception are, Friedman said, “chock full of lipids” (McCurdy et al., 2009). The accumulation of lipids in the liver is not benign, Friedman explained. The livers are actually in a state of oxidative stress. The McCurdy et al. (2009) researchers fed a control chow diet to one pod of animals and a high-fat diet to the other pod, mated the animals, then performed a C-section and assayed the fetal livers. Because
they were also curious about how a change in diet would impact lipid accumulation in the liver, after the C-section they put the animals who had been on a high-fat diet on a healthy diet and conducted a second assay of fetal livers from subsequent pregnancies. In the second assay, they observed a reduction, but not a total reversal, of liver triglycerides. Friedman remarked that while the liver is one source of pathology related to a high-fat diet, in fact dietary exposure is affecting every single organ in the animal, including the brain.
To see if the same phenomenon occurs in humans, Friedman and his collaborators collected MRI data on infants born to mothers with gestational diabetes and estimated the amount of fat in the liver (i.e., using the ratio of water to fat in the liver). In a study of 13 infants born to mothers with normal weight and 12 infants born to mothers who were obese, Brumbaugh et al. (2013) found that hepatic lipids in infants born to obese mothers with gestational diabetes were 72 percent higher than in those born to normal-weight mothers with gestational diabetes.
The surprising finding in Brumbaugh et al. (2013), in Friedman’s opinion, was that it was maternal pre-pregnancy BMI, not infant fat mass, that predicted liver fat mass. Unlike adults, who if developing a fatty liver have a lot of central obesity, infants are not born with central obesity and have very little visceral fat. Friedman interpreted this finding to mean that fuels in a mother who is obese have nowhere to go but into the fetal liver and that it is her obesity, not the infant’s, that determines whether a child will be born with a fatty liver. Friedman observed that there is a high risk of fatty liver disease in children who are obese, with probably about 55 percent of children who are obese having fatty liver at the “first hit,” that is, because of genetics, perhaps epigenetics, and in utero exposures (Anstee and Day, 2013). The “second hit,” that is, when the hepatic lipid accumulation develops into a more severe form of nonalcoholic fatty liver disease (NAFLD), results from oxidative stress, hepatocyte injury, and inflammation (Vos et al., 2013). The second hit, Friedman explained, probably does not occur in utero.
The question for Friedman is, are the effects of maternal obesity on fat mass in offspring liver reversible? In a study with macaques, again using animals that were fed either a control diet or a high-fat diet, Friedman and his research team weaned offspring of mothers on a high-fat diet on to a healthy diet at 7 months. Then, at 1 year of age, they assayed the offspring livers. They found that juvenile livers from animals born to mothers on a high-fat diet but weaned on to a healthy diet had elevated de novo hepatic lipogenesis gene activation (Thorn et al., 2014). More importantly, when they looked more closely at who the mothers were, they found elevated activation of the liver lipid pathways only in offspring born to mothers on high-fat diets who were very insulin resistant.
On a molecular level, Friedman explained, the fuels from the high-fat diet are crossing the placenta and probably overwhelming the mitochondria, creating oxidative stress and a loss of control of the genes for de novo lipogenesis and triggering inflammation pathways that set the stage for high-risk NAFLD in obese teens and young adults.
Postnatal Influences of Maternal Diet and Breast Milk on the Infant Microbiome
It is widely known among scientists in the field that people who are obese have what is called a dysbiotic microbiome compared to people who are lean. Friedman asked, what does that mean? How does it occur? And when would it occur in an infant? The neonatal period immediately after birth is critical for programming the immune system. During breastfeeding, infants are exposed to novel nutrients, bioactive molecules, and bacteria; their brain neurocircuitry for gut-brain energy sensing systems is being established, and the gut and liver immune cells are receiving instruction from the diet in early-life aspects of immune protection (i.e., what things can get in and what things should be kept out). Part of this early programming has been shown to take place at the level of epigenetics, according to Friedman, and every aspect of it can be influenced by breast milk composition. If a mother is on a Western diet and has a dysbiotic microbiome, and if her infant is breastfeeding on that diet, many of the microbial products “setting up shop down in the infant gut” can readily cross into the liver because the gut is quite leaky in early development. As a result, what starts as simple steatosis can progress into the more severe fatty liver disease seen in obese adolescents, perhaps driven by early programming events in the infant immune system. While breastfeeding is generally associated with protection against rapid infant weight gain and later obesity, the mechanisms responsible are not known but likely involve the delivery of bioactive components that regulate infant appetite, metabolism, and weight gain and adiposity.
In what Friedman referred to as a “remarkable” paper, Koren et al. (2012) reported that when the normal microbiome of a woman in her first trimester of pregnancy is transplanted into a gnotobiotic mouse (i.e., a mouse with no microbiome), the mouse develops normally. But if the altered microbiota of a woman in her third trimester of pregnancy is transplanted into a gnotobiotic mouse, the mouse becomes fat and develops insulin desensitization. These findings raised a concern for Friedman, that is, what happens to their infants when women enter pregnancy with an already disordered microbiome?
As an example of what happens in primates, again with animals born to mothers fed a high-fat diet but then switched to a healthy diet at wean-
ing, Ma et al. (2014) reported that maternal diet during pregnancy altered the offspring microbiome such that, even when offspring were switched to a healthier diet, they retained the microbiome they received from their mothers one year later.
Friedman and his team are currently testing whether maternal obesity in humans directly affects the development of the infant microbiome and is associated with increased adiposity during the first 4 months of life. Specifically, they are evaluating four maternal phenotypes in the perinatal and gestational period: normal weight, overweight or obese, type 2 diabetes, and gestational diabetes. All of the infants are being delivered vaginally, with none of the mothers on antibiotics and all of them agreeing to breastfeed for at least the first 4 months. The researchers are assaying both the maternal and infant microbiomes and evaluating their associations with infant adiposity.
With respect to the relationship between breast milk from obese mothers and offspring inflammation, Friedman’s colleague Bridget Young has shown that while the amount of triglycerides in breast milk is not significantly different at 2 weeks between obese and normal-weight mothers, there is a striking increase in levels of leptin, insulin, and the pro-inflammatory fatty acids (i.e., the n-6/n-3 fatty acid ratio). Moreover, leptin and insulin levels and n-6/n-3 fatty acid ratios, in turn, are associated with the composition of the infant microbiome. Friedman interpreted these findings to mean that exposure to maternal diet through the breast milk appears to be patterning the infant microbiome and, therefore, might be either protecting the infant gut or making infants more prone to inflammation and weight gain, but that is not yet known.
To conclude, Friedman shared some final thoughts:
- Humans share a core microbiome, yet they differ by genes, species, ecology, and gene count or richness.
- The gut microbiome is dynamic, yet its timescales are largely unknown.
- While changes in diet can lead to short-term changes in the microbiome, it is not clear which of those changes are reversible. In the nonhuman primate model studied by Friedman and his colleagues, some of the changes are reversible, but most are not.
- While microbiome gene richness is a key stratifier for response to dietary intervention, causing obesity in mice, the mice revert back, so the response is not permanent.
- Some microbiome-derived metabolites can have a positive effect on anti-inflammatory activity or energy harvest, while others are toxic to the host.
- A key question for Friedman is, can specific species or patterns of the gut microbiome be identified that might be relevant as targets for obesity?
Linda Adair defined disparity as a great difference or a lack of equality. Part of the challenge to understanding childhood obesity is the substantial variation in prevalence of obesity among children of different ages and races or ethnicities (Ogden et al., 2014) (see Figure 3-1). Using data from a very large database across the United States, Ogden et al. (2014) reported variation in obesity prevalence even in the 11-month age range. In children between 2 and 5 years of age, the highest obesity rates are in American Indians and Alaskan Natives, and the lowest rates are in non-Hispanic whites.
The wide variation observed in the United States is being observed globally as well, according to Adair, with the Department of Health and Human Services data from 26 different low- and middle-income countries showing a range of weight-for-length/height Z-scores from under 2.0 to over 14.0, with most countries having Z-scores greater than 2.0.
Disparities with Implications for Child Obesity
Adair identified several types of disparities with important implications for child obesity, not the least of which are disparities in resources, particularly health-promoting resources, with a wide range of socioeconomic factors (e.g., wealth, income, education, social status) creating disparities in both nutritional exposures (i.e., food availability, food security, and diet quality) and physical activity opportunities.
Also having important implications for child obesity are the many pronounced disparities in exposure, particularly pathogenic exposures related to poor water quality, sanitation and hygiene issues, and close living quarters. Often, Adair observed, individuals who are the most disadvantaged are also more likely to be exposed to pesticides and toxic metals. Additionally, Adair mentioned disparities in stress and social support arising from financial challenges, variation in physical environment, emotional factors, and different types of life events.
2 This section summarizes information presented by Linda Adair, Ph.D., University of North Carolina at Chapel Hill.
FIGURE 3-1 Prevalence of obesity, by age and race/ethnicity. Variations in the prevalence of childhood obesity in the United States among different races and ethnicities.
SOURCE: Presented by Linda Adair on February 26, 2015; modified from Ogden et al., 2012.
In terms of which of these aspects of disparity the fetus or young infant actually perceives, nutritional exposures are arguably one of the most important factors to consider. Disparities in nutritional exposure are reflected in maternal stores, that is, how fat he mother is, or, in the case of limited resources, how thin she is; maternal dietary intake of specific macro- and micronutrients associated with child growth (e.g., B vitamins, methyl donors); and several aspects of maternal metabolism, such as gestational diabetes. Adair stated that all of these various manifestations of disparities in nutritional exposure are highly critical in the first 1,000 days (beginning at conception).
In addition to variations in nutritional exposure, variation in maternal toxic exposures are also perceived in some manner by the young infant and developing fetus. For example, smoking (tobacco) is well known to be a very important factor that is differentially distributed across race and ethnic groups. Young infants and developing fetuses are also affected by maternal exposure to toxic metals, Adair observed. She did not elaborate, but she mentioned that she has been doing some work in South Africa with populations living close to mining communities and being exposed to high levels of arsenic and lead. She remarked that there has been a lot of recent concern about exposure to endocrine disruptors, which, like smoking and heavy metals, are differentially distributed according to socioeconomic status. According to Adair, researchers have also been reporting differential
exposures to growth and metabolic hormones, such as insulin and leptin, and to stress hormones, such as cortisol, again with the exposures ultimately translating into something that the fetus perceives.
Maternal Nutrition and Risk for Child Obesity
Adair discussed how both maternal under-nutrition and maternal excess nutrition have implications for child obesity and noted that both have been shown to be highly disparate among different socioeconomic groups or races and ethnicities.
First, with respect to maternal under-nutrition, underweight (BMI < 18.5 kg/m2) and micronutrient deficiencies are more prevalent in low- and middle-income countries and in lower socioeconomic groups in high-income countries. Low pre-pregnancy BMI has, in turn, been associated with increased risk of low birth weight, small for gestational age, and preterm birth (Dean et al., 2014; Yu et al., 2013b). Micronutrient deficiencies—for example, iodine, zinc, and iron deficiencies—have been shown to increase risks of low birth weight, small for gestational age, and preterm birth (Ramakrishnan et al., 2012). More generally, Adair explained, when a mother is unable to supply nutrients to meet fetal demand, a cascade of metabolic events ensues involving the kidneys, liver, pancreas, bone and muscle tissue, the brain, and the hypothalamic-pituitary-adrenal (HPA) axis, with important implications for the development of obesity, particularly central obesity (Fall, 2011).
With respect to maternal excess nutrition, overweight (BMI > 25 kg/m2), excess gestational weight gain, and dietary excesses have all been associated with increased risk of large-for-gestational-age deliveries and infant macrosomia (Siega-Riz et al. 2009).
In terms of what these maternal pre-pregnancy weight status disparities look like, Adair cited 2010 U.S. Pregnancy and Perinatal Surveillance System (PPNSS) data showing substantial race and ethnicity disparities in both underweight and overweight and obesity. As shown in Figure 3-2, the highest rates of underweight are in Asian and Pacific Islanders. While Asian and Pacific Islanders have the lowest rate of overweight and obesity, at about 30 percent, close to 60 percent of American Indians/Alaska Natives and non-Hispanic blacks show a maternal pre-pregnancy BMI greater than 25. Pregnancy weight gain data from the same database show similar race and ethnicity disparities (see Figure 3-3). Adair remarked that a large literature suggests that it is not so much race and ethnicity that vary, but rather the underlying social disparities (i.e., in wealth, education, etc.) that race and ethnicity represent.
In addition to disparities in maternal under- and over-nutrition, PPNSS data also show racial and ethnic disparities in maternal anemia, pregnancy-
FIGURE 3-2 U.S. disparities in pre-pregnancy weight status among different races and ethnicities. Percentages of underweight (x-axis) by race/ethnicity (y-axis) are shown in the left panel, and percentages of overweight (x-axis) by race/ethnicity (y-axis) are shown in the right panel.
SOURCE: Presented by Linda Adair on February 26, 2015; modified from U.S. Pregnancy and Perinatal Surveillance Data (www.cdc.gov/pednss).
FIGURE 3-3 U.S. disparities in inadequate and excess weight gain during pregnancy among different races and ethnicities. Percentages of inadequate pregnancy weight gain (x-axis) by race/ethnicity (y-axis) are shown in the left panel, and percentages of excess pregnancy weight gain (x-axis) by race/ethnicity (y-axis) are shown in the right panel.
SOURCE: Presented by Linda Adair on February 26, 2015; modified from U.S. Pregnancy and Perinatal Surveillance Data (www.cdc.gov/pednss).
induced hypertension (PIH), and gestational diabetes, all with important implications for child obesity risk.
Globally, Black et al. (2013) reported that underweight has been declining, while overweight and obesity among women of childbearing age has increased fairly dramatically, with the Americas and the Caribbean, as well as Oceania, showing the largest increases worldwide.
Variation in Infant Outcomes
Data from PPNSS 2010 indicate that adverse birth outcomes in the United States, including preterm births, macrosomia, and low birth weight, are highly disparate across different races and ethnicities, with preterm births and low birth weight being highest in non-Hispanic black populations. Additionally, data from the U.S. Early Childhood Longitudinal Study Birth Cohort (2001–2007) showed dramatic differences in patterns of postnatal growth by race and ethnicity (Jones-Smith et al., 2014). Generally, according to Adair, the odds of being overweight or obese diverge among the different races or ethnicities at about 9 months of age. “So it’s happening in infancy,” she said, with breastfeeding and patterns of infant feeding having an effect and creating a divergence in obesity risk among race/ethnicity early in life.
With respect to breastfeeding, Adair referred to Jacob Friedman’s remarks about the quality and quantity of breast milk being critical for infant development. That there are differences in the composition of breast milk based on maternal weight status is of concern in her opinion. Estimates of the prevalence of breastfeeding among different races and ethnicities vary depending on the source of the data. For example, PPNSS data from 2010 show less disparity in initiated breastfeeding than shown by the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) data from the Urban Institute. While the PPNSS data show that more than 60 percent of both non-Hispanic blacks and non-Hispanic whites initiate breastfeeding, the WIC data show that only a little over half of low-income non-Hispanic white women and only about 42 percent of low-income African American mothers initiate breastfeeding.
Researchers have reported substantial variations in infant and toddler feeding as well, including, in Adair’s opinion, multiple aspects of infant and toddler feeding reflecting disparities in race, ethnicity, weight status, income, education, and age. Variation has been reported in initiation and duration of breastfeeding, timing of the introduction of complementary foods, and types of complementary foods fed to infants (e.g., fruits and vegetables, sweetened beverages, salty snacks) (Hendricks et al., 2006).
Global variation in infant outcomes is reflected in the very high rates of small for gestational age and low birth weight infants in South Asia,
whereas in sub-Saharan Africa the greater problem is large-for-gestational-age infants, or infant macrosomia (Black et al., 2013; Koyanagi et al., 2013). Macrosomia also occurs more frequently in China and Latin America. Additionally, Black et al. (2013) reported worldwide socioeconomic disparities in the prevalence of stunting, which, Adair noted, is still highly prevalent worldwide, and tracks with overweight and obesity in children under the age of 5 years. Specifically, they showed that rates of stunting are very different in low-income versus higher-income countries. While their reported disparities in obesity are not quite as great as they are for stunting, Adair believes, that trend is changing quite dramatically. In high-income countries, obesity prevalence is typically higher in disadvantaged groups. Historically, that has not been the case in low- and middle-income countries. It used to be that in low- and middle-income countries, obesity was found mainly in the upper socioeconomic groups. But today, even in low- and middle-income countries, the fastest rates of increase in both child and adult obesity are in lower-income groups (Jones-Smith et al., 2012). Adair emphasized the importance of this trend in the context of disparities because it is those low-income groups who are more exposed to all of the environmental risk factors she mentioned previously (e.g., low-quality diets, toxic substances).
Obesity Disparities and Epigenetic Mechanisms
According to Adair, there is evidence of an epigenetic association with almost every disparity with obesity implications. For example, disparities in maternal glycemia during pregnancy have been associated with altered methylation profiles of adiponectin genes in the placenta (Bouchard et al., 2012). Disparities in parental overweight, including both maternal and paternal overweight, have been associated with hypomethylation of IGF2 (Soubry et al., 2013). Disparities in maternal diet, including carbohydrate intake, famine exposure, and vitamin and mineral uptake, have been associated with altered methylation patterns (Godfrey et al., 2011; McKay et al., 2012; Tobi et al., 2009). Although Adair did not go into detail, she said the same is true of stress (Brockie et al., 2013), environmental exposures such as arsenic (Marsit, 2015), maternal depression (Oberlander et al., 2008), and breastfeeding (Verduci et al., 2014), all of which differ substantially according to socioeconomic status, race and ethnicity, and geography. Evidence suggesting that all of these disparities are influenced by mechanisms associated with epigenetics provides a way, in Adair’s opinion, to begin to understand the biology of how disparities translate into obesity risk for the child.
The Mismatch Hypothesis
Many of the developmental phenomena being discussed at the workshop can be viewed, in Adair’s opinion, as adaptations that enhance survival in the environment to which an organism is exposed. It has been hypothesized that a mother provides information about her environment to the developing fetus and that the fetus, in response, develops a set of adaptations to cope with that environment (Bateson et al., 2014). The adaptations are also hypothesized to be anticipatory, that is, they are well suited to the environment that an individual will experience in the future as well. However, if the environment that is experienced postnatally is not what was predicted prenatally, then an organism will be maladapted and at an increased disease risk. For example, poor maternal diet, inadequate nutrient stores, or placental factors that limit transmission of nutrients to the developing fetus can all lead to fetal nutritional insufficiency and the metabolic differences that characterize under-nutrition (e.g., altered cell numbers, altered regulatory processes, altered epigenetics). Postnatally, if exposed to dietary excesses or sedentary behaviors, an individual that is programmed for under-nutrition may be maladapted, hyperresponsive, and at increased risk of obesity and chronic disease.
Adair pointed to the current situation in India to illustrate this notion of mismatch. India is considered a classic case of mismatch where young undernourished infants are growing up in an environment with nutritional excesses and with offspring having deficits in lean body mass but not body fat (Yajnik, 2004). Compared to a Caucasian baby that weighs an average of 3,500 grams, with 10 percent fat stores and 20 percent muscle mass, an Indian baby of average birth weight, which is 2,700 grams, has 20 percent fat stores and 10 percent muscle mass (Yajnik, 2004). According to Adair, Yajnik calls this the “thin fat Indian baby,” that is, a baby who in utero developed a limited muscle mass and more adipose tissue and was well adapted to its maternal environment of constraint but who is at increased risk of obesity later in life.
According to data from the National Family Health Survey (NFHS), the World Health Organization (WHO), and UNICEF, conditions in India 30 years ago—that is, when today’s mothers were being born—were such that low birth weight was common (30 percent), child stunting was highly prevalent (47 percent), and mortality under the age of 5 years was high (118/1,000). Contrast that to today’s dramatic increase in maternal overweight and obesity among Indian women, with a 25 percent central obesity rate in women and 11 to 12 percent obesity in New Delhi among 14- to 17-year-olds (Garg et al., 2010). There have also been reports of increased rates of diabetes (Anjana et al., 2011).
Most studies of mismatch have focused on longer-term effects, Adair
said, with less focus on whether mismatch might help to explain short-term rapid infant weight gain. Adair mentioned having observed lower birth weights in infants born to first-time mothers, but with firstborns who are well fed experiencing rapid postnatal weight gain. That, in a way, Adair said, is mismatch. The infants experienced prenatal constraints and were prepared for under-nutrition, but experienced over-nutrition instead. She pointed to data from the Consortium of Health Orientated Research in Transitioning Societies (COHORTS), a collaborative among investigators conducting birth cohort studies in five low- and middle-income countries worldwide (Brazil, Guatemala, India, the Philippines, and South Africa). The data indicate that while low birth weight is associated with reduced risk of overweight at the age of 2 to 3 years, that association is modified in firstborn infants. Firstborn low birth weight infants gained more weight than higher-order low birth weight infants and were twice as likely as low birth weight infants of higher birth order to be overweight at 2 years of age. Again, that is mismatch, Adair said, with firstborns who were undernourished in utero experiencing very rapid postnatal growth.
Adair mentioned other studies of low birth weight or small-for-gestational-age infants who were deliberately being fed to catch up, with data indicating an increased risk of obesity associated with the catch-up growth. That situation may also represent mismatch.
In her summary, Adair made five key points: (1) wide disparities in socioeconomic status, physical environment, psychosocial factors, and stress contribute to substantial differences in fetal exposure to nutrients, toxins, hormones, and other regulatory substances; (2) these disparities may affect fetal and infant growth and susceptibility to later obesogenic factors through epigenetic and other pathways; (3) elevated risk of child obesity may result from prenatal under-nutrition as well as from nutritional excesses; (4) the risk may be greatest when the fetus is adapted to a maternal environment that differs from the environment faced as an infant and young child; and (5) understanding the exact nature of pathways of risk may lead to interventions to eliminate the adverse effects of health disparities.
Stephen Krawetz of the Wayne State University School of Medicine expressed surprise that no one had considered Dad as an initial driver, given that the paternal genome and, in fact, everything carried in the sperm as well as a father’s past can impact early development of a child.
In order to understand the paternal contribution to childhood early development, Krawetz suggested starting with spermatogenesis (Wykes et al., 1995, 1997). Spermatogenesis, he said, is unlike any other system in the body. From puberty onward, it is continually replenished from the stem cells toward the lumen tubule through a series of developmental stages to eventually yield mature spermatozoa. As spermatozoa differentiate, each forms a little bag of cytoplasm, known as the residual body or cytoplasmic droplet, which Krawetz said is the sperm cell’s way of getting rid of excess cytoplasm. Each mature spermatozoa is about 2 microns in diameter and, because of its large tail, up to 20 to 40 microns in length. The long tail is needed for motion and is what allows a sperm cell to reach the egg. It takes about 2 hours for the fastest sperm swimmer to make it to the egg. When the sperm reaches the egg, the entire sperm is taken in upon fertilization, including all components except the cytoplasmic droplet, followed by a rapid dissociation and integration.
The sperm genome is about 13 times more compact than the oocyte’s, even though it contains the exact same amount of information. Its compactness is caused by a unique set of proteins called the protamines, which are autosomal but expressed only in men and without which a man is infertile. Using radiolabeled antisense imaging, Krawetz and colleagues have shown that not all spermatozoa are equal in terms of the amount of labeled protamines mRNAs present (Wykes et al., 1997). However, it is still unclear, Krawetz said, whether that is a reflection of relative content, namely, whether spermatozoa contain unequal amounts, or a reflection of penetration of the probe.
In addition to the paternal genome, the sperm delivers to the egg an organelle called the centrosome, which Krawetz said is critical for subsequent cellular divisions; a sperm oocyte-activating factor, over which Krawetz noted there is now some controversy; and an RNA component (Ostermeier et al., 2002, 2004). On a per cell basis, sperm deliver about
3 This section summarizes information and opinions presented by Stephen Krawetz, Ph.D., Wayne State University, Detroit, Michigan.
50 to 100 femtograms of total RNA, including 0.3 femtograms of small noncoding RNA.
At the first cellular division, “Dad” plays a major role, in Krawetz’s opinion, with paternal microRNA-181c being essential for dictating which of the first two dividing cells actually maintains a stem cell likeness, with the other cell being set on a course for developing the trophectoderm. Additionally, unlike in other mammals, zygotic genome activation in humans appears somewhat delayed, occurring by the four- or eight-cell stage, which means, Krawetz explained, that all of the information needed for the two prior cell divisions must be housed by either “Mom” or “Dad” because it cannot be made. This is in contrast to the mouse, in which zygotic genome activation occurs before the first cell division.
Evaluating Sperm RNAs
To study the biological relevance of sperm RNA, Krawetz and colleagues pooled RNA from the testes of 19 individuals, pooled RNA extracts from the ejaculates of 9 individuals, and collected RNA from the ejaculate of a single individual, and synthesized from all of these different samples a series of cDNAs (Ostermeier et al., 2002). Using microarray hybridization, they showed, first, that the sperm had a rich population of RNAs that were similar to those of the testes and, second, that all but 4 of the approximately 2,500 transcripts present in the single individual were present in the pooled ejaculate sample. Krawetz interpreted these findings to mean that, basically, all sperm carry a very similar load of RNAs. The question is, are all of those RNAs actually delivered? Using a hamster sperm penetration assay, Krawetz and collaborators showed that, in fact, the RNA was delivered (into hamster eggs) and retained for at least 3 hours, which is as long as they followed the delivery (Ostermeier et al., 2004). Krawetz interpreted this finding to mean that the RNA did have a chance to actually be utilized by the egg cell.
Much of the work being done to evaluate sperm RNAs is technology and now sequence-driven, Krawetz remarked. In the early 2000s, researchers thought they were doing a great job when they were able to collect 0.0012 gigabytes of information. Today, researchers can generate a terabyte of information over the course of just a couple of days.
Normally, when testes RNA is isolated, it yields two peaks, one at 28S and the other at 18S (representing two different ribosomal, or rRNA, species). Those are the two peaks all researchers look for when evaluating the quality of RNA numbers, Krawetz explained. But RNA isolated from sperm yields peaks that look nothing like that 28S-18S two-peak pattern. Initially, when he and his team did their early microarray work, the 18S RNAs appeared absent, suggesting that sperm were void of any ribosomal RNA.
But when they sequenced the rRNA, Krawetz and his colleagues were hit by what he said was “the shock of our life,” which was that 89 percent of the RNA present was in fact rRNA, but fragmented rRNA (Sendler et al., 2013). Another 5 percent was mitochondrial RNA, another 5 percent other types of RNA, and about 1 percent was small noncoding RNA. Within the “other” category, approximately 50 percent of the transcripts had been described before as messenger RNA. One of the most abundant long noncoding RNAs present, MALAT1, is known to be involved in the regulation of chromatin structure. The functions of many of the “other” RNAs remain a mystery, Krawetz said, and they are “rife for discovery.” A few clues indicate that they may have interesting functions.
In terms of the stability of the transcripts identified in sperm RNA, Sendler et al. (2013) divided the 1,000 most abundant sperm transcripts into quintiles based on how much of the RNA was present or absent among their samples. For example, all 13 exons of a transcript called ACSBG2, which is encoded by chromosome 19 and extends about 60 kilobases, were found to be represented in equal amounts in 10 different individuals, leading the researchers to classify that particular transcript as intact. Its function in early development is unknown, Krawetz said.
In terms of spermatozoal small noncoding RNAs (i.e., the 1 percent of transcripts identified in Krawetz et al. ), the two major classes are the piRNAs and microRNAs. Four of the microRNAs identified (34c, 375, 184, 152) have been validated in other studies for their roles in controlling obesity or early development. There is also a series of tRNAs and tRNA fragments, the latter now known to be methylated and relatively stable.
Many functions have been proposed for spermatozoal RNAs, Krawetz continued, including confrontation and consolidation (i.e., in terms of how compatible the egg and the sperm genomes are), translation of intact paternal mRNAs, transcriptional regulation by paternal microRNAs, activation of paternal pre-microRNAs by maternal DICER (as shown from microRNA-181c), and transcriptional regulation by paternal microRNAs and RNA fragments (Jodar et al., 2013). Several studies have demonstrated that if microRNAs are delivered, they can have a fairly early and lasting effect. Rassoulzadegan et al. (2006) demonstrated that injecting either miR-221 or miR-222 microRNA into mice induced a mutated and heritable white-spotted phenotype. Subsequent work by Wagner et al. (2008) showed that injection of another microRNA, miR-1, which targets Cdk9, can induce heritable cardiac hypertrophy. Then, Grandjean et al. (2009) showed that injecting miR-124 microRNA, which targets Sox9, into mice yields a giant phenotype and twin pregnancies.
Transgenerational Epigenetic Inheritance
One role of epigenetics, in Krawetz’s opinion, is to respond to a changing environment. Rather than making a change permanent, which would be difficult to undo when the environment changes yet again, epigenetics provides a way to transmit what Krawetz called a “responsive state” without altering the primary structure of the DNA.
In terms of implicating paternal microRNA in transgenerational epigenetic inheritance, the best evidence, in Krawetz’s opinion, comes from studies on stress (Dias and Ressler, 2014; Gapp et al., 2014; Rodgers et al., 2013). Dias and Ressler (2014) showed that following intracytoplasmic sperm injection of microRNA-375 into the oocyte mice offspring exposed to the same odor that their parents had been exposed to before conception experienced the same averse response that their parents had. According to Krawetz, this has been demonstrated now with several different odorants.
In terms of transgenerational effects of paternal nutrition, in Krawetz’s opinion the Overkalix study by Kaati et al. (2002) is among the top studies. The Overkalix study was conducted on a series of Swedish cohorts born in 1890, 1905, and 1920 and followed until 1995. The researchers extrapolated food access from historical data and asked whether an abundance of food during a child’s slow growth period, that is, before the prepubertal peak, influenced descendants’ risk of death from cardiovascular disease and diabetes. They found that limited access to food during the father’s slow growth period limits a child’s cardiovascular disease risk, but that a paternal grandfather surrounded by a bounty of food during his slow growth period increases the grandchild’s risk of diabetes. The researchers concluded, “A nutrition-linked mechanism through the male line seems to have influenced the risk for cardiovascular and diabetes mellitus mortality” (Kaati et al., 2002, p. 682).
More recently, there has been a series of studies in mice on the intergenerational effects of paternal diet (Carone et al., 2010; Fullston et al., 2013; Lambrot et al., 2013). Carone et al. (2010) showed that offspring of male mice fed a low-protein, high-fat diet developed a fatty liver phenotype. In a study of diet-induced paternal obesity, Fullston et al. (2013) showed increases in adiposity and insulin resistance in both the F1 and F2 generations, with a heightened effect on female F1 offspring (67 percent increase in adiposity) and their F2 sons (24 percent increase in adiposity). The researchers isolated a series of microRNAs, including the paternally donated microRNA-205, which they suspect may play a mechanistic role.
Still to be resolved, Krawetz explained, is how paternal information gets relayed to the male gamete. He said, “It’s really incredible to me that you could eat something, or you could smell something, or experience
something, and it would go from the brain to the testes.” Also, is the change temporary or permanent? Does it cross the blood–testis barrier in order to be delivered to the sperm? Is it a modulated response, that is, does it occur by RNA production, stability, or acquisition? Using a mouse model xenografted with human cells expressing EGFP RNA, Cossetti et al. (2014) showed that even the murine sperm had traces of EGFRP RNA. Krawetz and collaborators have been looking at the structure of sperm and have observed that at high magnification the membrane is not a tight structure but looks almost like a series of vesicles. If in fact a series of vesicles, Krawetz believes, that structure may provide a mechanism for the exchange of genetic information. He and his team were very excited about that possibility until Chevillet et al. (2014) pointed out, if microRNAs are homogenously distributed across all exosomes, then each exosome would contain 0.01 copies of any particular microRNA, and microRNAs would have to be consolidated or concentrated in some way such that not all exosomes contain microRNAs. Krawetz and his team are currently exploring that possibility as a mechanism for the communication of genetic information.
In closing, Krawetz reminded the workshop audience that there are about 50–100 femtograms of RNA in a sperm cell, including 0.3 femtograms of small noncoding RNAs, compared to 40,000 femtograms of RNA in a somatic cell. “What this means,” he said, is that “Dad has his challenges cut out for him, but he does deliver a few very important things.”
The evidence is compelling, Caroline Relton said, that both maternal over- and under-nutrition are associated with adverse consequences for offspring through their effects on adiposity. The question for her is, what potential mechanisms underlie those observational associations, and do epigenetic mechanisms play an important role in explaining those associations (Lawlor et al., 2012)?
When thinking about epigenetic processes as a potential mediating mechanism linking maternal over- or under-nutrition with offspring adiposity, Relton reminded the workshop audience that although her talk would be very much framed around epigenetic modifications—DNA methylation in particular as the mediator—one could very easily substitute “epigenetic modification” with microRNA expression, metabolomics profiles, or the microbiome. The same principles exist for consideration of a number of different mediating mechanisms.
4 This section summarizes information and opinions presented by Caroline Relton, P.G.C.E., Ph.D., Newcastle University and the University of Bristol, United Kingdom.
While it has become increasingly straightforward to identify observational associations between a phenotype or an exposure and DNA methylation, Relton remarked, the challenge is to decipher whether that association is causal. There are several possible scenarios that could explain, for example, the observed association between maternal obesity (the exposure), offspring methylation (potential mediator), and offspring adiposity (the phenotype). Which is the right one? Is it indeed the case that maternal over-nutrition alters offspring methylation and subsequently has an impact on offspring phenotype? Or is it a case of reverse causation, where the exposure is altering the offspring phenotype, with the methylation change being a consequence of the altered phenotype? Or is it a situation of confounding, with exposure affecting both methylation and the phenotype?
Relton suggested a stepwise approach to addressing the challenge of determining whether epigenetic mechanisms are mediating maternal influences on childhood adiposity (see Box 3-1). First, establish an association between maternal factors (e.g., body weight and weight gain during pregnancy) and offspring adiposity. Relton reiterated that the observational evidence for such an association is very strong. Second, establish a relationship between the same exposure and a child’s DNA methylation. Third, establish an association between the child’s DNA methylation (i.e., the mediator) and the child’s adiposity (i.e., the outcome). Fourth, implement a variety of different methods to strengthen the causal inference.
Later during her presentation, Relton described how she and her colleagues applied these steps to data from the Avon Longitudinal Study of Parents and Children (ALSPAC), a Bristol, United Kingdom–based longi-
Epigenetics as a Potential Mechanism Mediating Maternal Influences on Childhood Body Composition
Caroline Relton identified four steps to determining whether epigenetics is a potential mechanism mediating influences on childhood body composition:
- Step 1: Establish an association between maternal factors (e.g., body weight, weight gain during pregnancy) and offspring adiposity.
- Step 2: Establish a relationship between the same exposure and a child’s DNA methylation.
- Step 3: Establish an association between the child’s DNA methylation and the child’s adiposity.
- Step 4: Apply methods to strengthen causal inference.
tudinal cohort study of 14,000 pregnant women who were recruited in the early 1990s (Sharp et al., 2015). Both the women, who are now between the ages of 23 and 34 years, and their children have been followed extensively since their recruitment. Relton noted that the data generated throughout the life span of this cohort are publicly available for any bona fide researcher to access and use (www.bristol.ac.uk/alspac). The data cover a range of health and demographic factors, environmental exposures, behavioral factors, development and education, and parental psychological well-being.
In addition to the wealth of ALSPAC exposure and outcome data, Relton and colleagues were recently awarded funding to profile genome-wide methylation data using the Illumina Infinium® HumanMethylation450 BeadChip. So far, they have collected methylation data on 1,000 mother–child pairs, with data collected on the mothers when they were pregnant and again 17 years later and on the children when they were born, at age 7 years, and between 15 and 17 years. Relton and her team wanted to know whether the effects of gestational weight gain and/or maternal pre-pregnancy body mass index on birth weight, childhood adiposity, and adolescent adiposity were being mediated through altered methylation. Data generated from an epigenome-wide association study based on cord blood methylation in children at birth indicate a number of hits in relation to maternal underweight, very few hits in relation to maternal overweight, and some hits in relation to maternal obesity (Sharp et al., 2015). But again, the question is, are the associations causal?
Tools for Strengthening Causal Inference
Researchers have several tools at their disposal to strengthen causal inference: randomized controlled trials, cross-cohort comparisons, negative controls, parental comparisons, sibling comparisons, appraising temporal relationships through longitudinal analysis, and Mendelian randomization.
Randomized controlled trials are considered the gold standard, Relton remarked, but they are often either implausible or extremely expensive. Basically, they involve randomizing exposure and then comparing the risk of outcome. She mentioned that while some randomized controlled trials of nutrition interventions with epigenetic components are under way, they are not at the present time the most obvious choice in terms of trying to interrogate causal pathways in epigenetic pathways.
A complementary approach, the cohort comparison, involves analyzing two independent cohorts with different confounding structures. Researchers compare the magnitude of association in the two cohorts. While this approach has not yet been used in epigenetic studies, according to Relton, it has been used in observational epidemiology to determine whether breastfeeding has an impact on childhood adiposity. For example,
it was used to compare breastfeeding in relation to childhood obesity and intelligence quotient (IQ) as part of both the ALSPAC study and the Pelotas (Brazil) Birth Cohort Study (Brion et al., 2011), with evidence from both studies showing that childhood IQ but not childhood adiposity is higher in children who were breastfed. In Relton’s opinion, the cohort comparison approach is potentially useful for epigenetic studies, but, again, it has yet to be applied.
The use of a negative control design, on the other hand, has been used for epigenetic studies, in particular in relation to maternal overweight and offspring methylation (Sharp et al., 2015). A negative control design involves using a second set of data with shared confounders and asking whether the same relationship is observed between exposure and offspring outcome by comparing the magnitudes of association. For example, Relton and colleagues postulated that maternal overweight would alter cord blood DNA methylation through an in utero biological event. If that truly was the case, then one would not expect to see the same association with paternal BMI and offspring phenotype. On the other hand, if the relationship was confounded by some other factor, then one would expect the two associations to be roughly the same magnitude. When Relton and colleagues compared their maternal overweight-offspring methylation association data with paternal overweight-offspring methylation association data, they observed robust associations between offspring methylation with maternal obesity (i.e., children born to mothers who were overweight were hyper- or hypo-methylated) but no significant associations with paternal obesity, suggesting that the effect of maternal obesity on offspring methylation is a biological in utero event.
Another helpful study design, one that Relton has not used but that has been reported in the literature, is sibling control. Sibling control involves exposure during the first pregnancy but no exposure during the second pregnancy, which overcomes the issue of familial confounding, and comparing the risk of exposure between siblings. An example is the comparison of adiposity or overweight in children born to mothers who were obese in their first pregnancy and then had bariatric surgery and were at a normalized weight in their second pregnancy (Kral et al., 2006).
Longitudinal modeling is yet another tool for strengthening causal inference with respect to DNA methylation. Of note, however, although a temporal association in a longitudinal study can help to overcome the issue of reverse causation, it may not overcome the issue of confounding. For example, Sharp et al. (2015) observed a difference in cord blood methylation between exposed and unexposed individuals (i.e., exposed to maternal factors) and were curious about how those differences tracked over time. They were able to track the differential methylation using multilevel modeling because they had DNA methylation data from the same
individuals at birth, at age 7, and between the ages of 15 and 17. They found one locus that was more highly methylated at birth among exposed individuals compared to unexposed individuals, but with the methylation attenuating over childhood and into adolescence in exposed individuals while remaining fairly static in unexposed individuals. In other situations—for example, when environmental influences come into play—unexposed individuals may also show changes over time. In Relton’s opinion, when both exposed and unexposed individuals are tracked, longitudinal modeling is a good method for interpreting methylation patterns observed at birth and their changes over time. This approach has also been used to understand the changes in DNA methylation in offspring born to women who smoke during pregnancy (Richmond et al., 2015).
Relton spent the remainder of her presentation discussing Mendelian randomization, an approach introduced in the previous session by Andrea Baccarelli (see Chapter 2 for a summary of Baccarelli’s presentation). The principle of Mendelian randomization is to use a proxy measure for an exposure that may otherwise be difficult to measure or that is subject to confounding or reverse causation. This measure is based on the application of instrumental variables analysis, used widely in the field of economics, and it allows for a better estimate of the effect of the exposure.
As an example of how Mendelian randomization could be applied to epigenetics, and how it was applied in the study that Baccarelli discussed (Dick et al., 2014), one could use as the proxy for maternal BMI an allele score generated from allelic variants known to influence maternal BMI and examine the relationship between this genetic proxy and a child’s DNA methylation. Additionally, if one wanted to know whether the child’s DNA methylation was causally related to the child’s BMI, one could identify genetic variants strongly correlated with site-specific DNA methylation and use those genetic variants as a proxy measure for DNA methylation. Because a genotype is not subject to reverse causation or much confounding, it provides a much better and more secure causal anchor and allows one to make inferences about exposure with regard to outcome.
As Baccarelli explained, Dick et al. (2014) reported an observed association between DNA methylation and BMI. Although the authors included in their paper the principle of using a genetic proxy as a causal anchor and went so far as to identify a genetic proxy for their methylation site of interest, they did not actually conduct a formal Mendelian randomization analysis. Relton and her colleagues have done so. Using genome-wide methylation data generated from their 1,000 ALSPAC mother–offspring
pairs, they focused on methylation sites in the HIF3A gene, which was the differentially methylated locus reported in Dick et al. (2014). They wanted to know whether maternal BMI was causally linked to DNA methylation at different stages throughout childhood.
Following the steps outlined in Box 3-1, the first step in the analysis was to observationally link maternal body mass index with methylation of HIF3A in the child. Relton and her coworkers did this using their maternal BMI allele score (i.e., the genetic proxy for maternal BMI), which was generated from 97 small nucleotide polymorphisms (SNPs) known to be robustly associated with BMI and which was recently published through the Genetic Investigation of Anthropometric Traits (GIANT) consortium. They used the weighted scores generated from those 97 SNPs to test for differences between the observed and expected estimates of the association between the generated allele score and the child’s BMI (i.e., expected if indeed the direction of effect was from BMI to DNA methylation change). Their results strengthened the causal inference that maternal BMI has a causal effect on childhood methylation.
The second step was to implement the same approach but using an allele score as a proxy for methylation and asking whether methylation was driving the child’s BMI. Again, they tested for differences between observed and expected (i.e., expected if methylation is indeed a determinant) estimates of the association, in this case, between the allele score proxy and the child’s BMI. Here, Relton and her colleagues concluded that there was no strong evidence that DNA methylation was having a causal effect on a child’s BMI.
The third step, again using the same approach, was to test whether the child’s own BMI had a causal effect on HIF3A methylation later in life. They found some evidence that, yes, a child’s BMI has a causal effect on DNA methylation at HIF3A.
In sum, in terms of addressing the questions about whether maternal BMI changes offspring methylation at birth and whether offspring methylation at birth subsequently has an effect on childhood adiposity, it was postulated at the outset that, yes, maternal BMI alters cord blood or child methylation at the HIF3A gene and that methylation at that locus subsequently alters the child’s BMI. However, having conducted the Mendelian randomization analysis, Relton and her team concluded that maternal BMI likely has a direct causal effect on the child’s BMI and that the effect is not mediated by early life changes in methylation.
The Relton team’s conclusion corroborates the conclusion made in the Dick et al. (2014) paper. That is, while there is an association between DNA methylation and BMI in adults, it is likely that the change in BMI is driving changes in methylation, not vice versa.
Strengthening Causal Inference: Improving the Evidence
To conclude, Relton considered ways to improve each of the four necessary steps toward strengthening causal inferences that epigenetics is mediating maternal influences on children’s body composition (see Box 3-1). To improve the first step, she called for better observational evidence using more refined measures of maternal exposure. To improve the second step, she called for improved technology for the assessment of genome-wide DNA methylation. All of the data that she presented were heavily reliant on the Illumina array, which she said has obvious limitations. Improving the third step will require both better observational evidence and improved technology for the assessment of genome-wide DNA methylation. Finally, to improve the fourth step, Relton called for an increased awareness of the pitfalls of the association studies and approaches being used and for a more widespread implementation of what she referred to as “triangulation of evidence.” She suggested not relying on one study design, but rather implementing a number of different tools to weigh the evidence.
Following Relton’s presentation, audience members asked several questions about the differences between animal and human studies, the impact of genetic variation on epigenetic variation, sex differences in offspring measures, and obesity-related outcomes other than BMI.
Animal Versus Human Studies
There were two questions posed by workshop participants on the differences between animal and human studies. First, an audience member speculated that epigenetic processes are likely to be different for short-lived species (e.g., rodents) versus long-lived species (e.g., humans). The questioner wondered whether short-lived rodents, for example, need to be immediately adaptable to their environments, compared to long-lived humans, who need to be more immediately malleable and adaptable over generations. None of the panelists directly answered the question, although Jacob Friedman replied that many organ-specific developmental changes are species-specific.
Second, Robert Waterland pointed out that studies in mice have shown that leptin has an early postnatal effect on hypothalamic development. He asked Jacob Friedman if any evidence suggests that the same may be true in humans, that is, that leptin affects brain development. Friedman replied
that he is unaware of any such evidence. His work with leptin has focused on the infant gut.
The Effect of Genetic Variation on Epigenetic Variation
The effect of genetic variation on epigenetic variation came up several times during the course of the 2-day workshop. Here, in reference to Caroline Relton’s demonstration of strong evidence indicating that maternal BMI is causally linked to child DNA methylation, Robert Waterland asked how Mendelian randomization rules out the alternative explanation that maternal genetics, which the child partially inherits, are causing the differences in the child’s DNA methylation. Relton mentioned an analysis that she and her team recently undertook that involved quantifying the variance in methylation explained by common genetic variation (SNP heritability). Initial results indicated that only a small portion of the known genetic contribution to BMI is explained by SNPs that alter methylation. In contrast, a much greater proportion of methylated SNPs are implicated in, for example, type 1 diabetes and rheumatoid arthritis. These findings, she said, suggest to her that there is an interesting difference across traits regarding the proportion of genetic influence on epigenetic patterning. Waterland pointed out that the influence of genetics is also likely to be tissue-specific.
Sex Differences in Offspring Methylation Patterns
The panelists were asked to comment on differences between male and female offspring. Caroline Relton replied that she and her team have observed a huge number of genome-wide methylation differences between the sexes, but they have not looked at those differences in relation to adiposity. They have detected approximately 17,000 statistically significant “hits,” all on autosomes, not on the sex chromosomes. They suspect that some of the hits are associated with sex hormones. In contrast, Stephen Krawetz added that he and his team have not found any statistical differences in the methylation patterns that they have directly targeted in sperm.
Beyond BMI: Other Outcome Measures of Obesity
An audience member expressed being in favor of seeking other measures of obesity besides BMI and asked the panelists if they had considered using metabolic disease correlates of BMI and not relying on size alone. Relton replied that she and her team have also used fat distribution data and other more subtle measures of adiposity at multiple time points. The beauty of BMI, she said, particularly in the context of Mendelian random-
ization, is that it is accompanied with a rich store of publicly available genotype data. One can gain a huge amount of leverage by querying data from large consortia such as GIANT with an SNP, or variant, that is known to be associated with methylation. The use of metabolic disease phenotypes as outcome measures instead of, or in addition to, BMI was an issue that was revisited several times throughout the workshop.