Epigenetic Mechanisms of Transmission of Metabolic Disease Across Generations

Abstract

Both human and animal studies indicate that environmental exposures experienced during early life can robustly influence risk for adult disease. Moreover, environmental exposures experienced by parents during either intrauterine or postnatal life can also influence the health of their offspring, thus initiating a cycle of disease risk across generations. In this perspective, we focus on epigenetic mechanisms in germ cells, including DNA methylation, histone modification, and non-coding RNAs, which collectively may provide a non-genetic molecular legacy of prior environmental exposures and influence transcriptional regulation, developmental trajectories, and adult disease risk in offspring.

eTOC

Sales et al. provide perspective on how environmental exposures, particularly those experienced during intrauterine life, can influence metabolic disease risk in adulthood. Moreover, both early-life exposures and metabolic disease in adult parents can modulate epigenetic regulation in germ cells, thus providing a non-genetic molecular legacy to influence health of subsequent generations.

Introduction

Metabolic disease arises at the confluence of genetics and environmental factors, such as nutrition, exercise, chemical exposure, behavior, and the microbiome (Baccarelli and Bollati, 2009; Barres and Zierath, 2016; Gallou-Kabani and Junien, 2005; McGowan et al., 2009; Miska and Ferguson-Smith, 2016; Skinner et al., 2010; Song et al., 2013). Interestingly, studies in both humans and animal models indicate that environmental exposures experienced by parents during either intrauterine or postnatal life can also influence the health of their offspring, thus initiating a cycle of disease risk across generations. In this review, we briefly discuss the potential mechanisms by which epigenetic mechanisms may contribute to environmentally-induced metabolic disease after direct exposure and in subsequent generations, with a particular focus on paternal lineage effects mediated by germ cells.

Evidence for non-genetic impact on metabolic disease risk in subsequent generations

Metabolic disease, including obesity and T2D, arises at the interface of genetic and environmental factors. While family history has long been recognized as an important determinant, only ~5–10% of T2D risk can be attributed to genetic factors (Voight et al., 2010). This so-called “missing heritability” (Maher, 2008) may result from environmental exposures shared by family members, including those experienced during early life.

The concept of developmental origins of adult disease proposes that during intrauterine life, the fetus adapts to the environment to which it was exposed, potentially conferring resistance to similar exposures experienced postnatally (Barker, 2000). Unfortunately, if the postnatal environment differs from the prenatal exposures, such responses may be maladaptive. For example, exposure to undernutrition during pregnancy may “program” the fetus to have more efficient energy metabolism, thus reducing weight loss and enhancing survival during future periods of undernutrition (Beauchamp et al., 2015). However, if the postnatal environment is instead marked by nutrient excess, offspring will be more susceptible to weight gain and obesity.

Unfortunate times of humanitarian catastrophes, such as wars and political/economic crisis, have provided the possibility to study the influence of poor nutrition on human populations. Retrospective studies of human cohorts exposed to undernutrition during prenatal and early postnatal life provide evidence to support intergenerational effects. Studies of the Dutch “hunger winter,” a period of abrupt onset of severe famine during German blockade of the Netherlands at the end of World War II (1944–1945), have been very informative. Poor nutrition experienced by pregnant mothers during the famine was associated with increased fat mass, hypertension, glucose tolerance, and psychiatric disorders emerging in their children during adult life (Lumey et al., 2007; Ravelli et al., 1998). Similar findings have been demonstrated for populations around the world. The Ukraine famine (1932–1933) also revealed associations between early gestational nutrient restriction and development of type 2 diabetes (T2D) in adulthood (Lumey et al., 2015). Similarly, individuals born immediately after periods of famine in Austria (1918–1919–1938; 1946–1947) (Thurner et al., 2013) or China (1959–1961) have higher rates of T2D or hyperglycemia (Li et al., 2010). Scarcity of food during the Biafran famine during the Nigerian Civil War (1967–70) and resulting undernutrition during gestation and childhood has also been associated with impaired glucose tolerance in adulthood (Hult et al., 2010).

Independent of nutritional factors, low birth weight itself increases the risk of metabolic disease in adult life (Barker, 2000; Mericq et al., 2016). David Barker was the first to identify relationships between birth weight and the development of cardiovascular and other metabolic disease during adult life (Barker, 2000). Subsequent studies in distinct populations worldwide have confirmed these relationships, including the Helsinki birth cohort study (Eriksson, 2006) and the China birth weight study (Li et al., 2010). While birth weight is a phenotype easily discerned in epidemiologic studies, birth weight itself is unlikely to play a causal role in adult disease risk; rather, birth weight is a biomarker arising from reduced fetal growth, in turn resulting from placental dysfunction, poor nutrition, or other pregnancy stressors. In a recent review, Mericq summarized the extensive body of literature supporting the association between children born small for gestational age or premature and the subsequent development of metabolic disease (i.e. higher glucose levels, T2D, cardiovascular disease) (Mericq et al., 2016). These associations are further magnified when low birth weight individuals experience postnatal “catch-up growth” (Dulloo, 2009). Conversely, overnutrition experienced during gestation also predisposes the fetus to develop disease later in life (Boney et al., 2005; Dabelea and Crume, 2011; Yu et al., 2013)

Exposures experienced by parents either prior to or at the time of conception can also impact the health of future generations (Alfaradhi and Ozanne, 2011; Patti, 2013). For example, either maternal or paternal prenatal exposure to famine (Chinese famine of 1959–1961) has been associated with increased hyperglycemia in offspring; these effects are greater when both parents had prenatal exposure (Li et al., 2016). Moreover, studies from a population born in the Overkalix region of northern Sweden from 1890 to 1920 analyzed food availability during specific phases of childhood. Interestingly, overnutrition during the childhood slow growth period in males was associated with increased risk of cardiovascular disease and diabetes in his grandsons– thus a transgenerationally inherited phenotype (Kaati et al., 2002).

Experimental models of multigenerational disease risk

Multiple animal models have demonstrated that early life exposures experienced by the parent can potently impact phenotypes in the developing offspring and in subsequent generations, even without further environmental stressors (Figure 1). For example, undernutrition experienced by the mother during gestation can cause not only increased adiposity and glucose intolerance/diabetes in her offspring (F1) (Isganaitis et al., 2009; Jimenez-Chillaron et al., 2005) but also increased adiposity and glucose intolerance in the next (F2) generation (Hanafi et al., 2016; Jimenez-Chillaron et al., 2009). Moreover, if an undernutrition insult is sustained, there can be further propagation of metabolic phenotypes across many generations. This was demonstrated in a recent study in Wistar rats subjected to continuous 50% caloric restriction over 50 generations. Offspring animals had fasting hyperinsulinemia, glucose intolerance, and increased adiposity with aging (Hardikar et al., 2015). Neither metabolic nor epigenetic phenotypes were reversed by restoration of nutrition for two generations, indicating striking durability of ancestral effects. Conversely, overnutrition and obesity in the F0 dam can also yield phenotypes in the F2 generation (Dunn and Bale, 2009; Gniuli et al., 2008). Environmental exposure to chemical agents also leads to increased metabolic disease susceptibility in subsequent generations (Baccarelli and Bollati, 2009; Holliday, 1998; Skinner et al., 2010). Interestingly, acquired behaviors can also be transmitted across multiple generations through non-genetic factors, as recently reviewed (Bohacek and Mansuy, 2015).

Figure 1. The vicious cycle of intergenerational paternal disease risk.

Adverse intrauterine exposure can impact the development and implantation of the embryo, ultimately increasing disease risk in the offspring. In addition, lifestyle also contributes to epigenetic changes, including alterations in the germ cells, both sperm and oocytes.

In summary, many influential factors may contribute to non-genetic inheritance of disease risk. Hence, consideration of potential mechanisms contributing to such phenotypes across generations requires analysis of several key components:

• 1Do phenotypes in offspring result from alterations in parental health or behavior existing at the time of conception? Alternatively, can these offspring phenotypes result from parental germ cell exposures, either during prenatal or postnatal life of the parent?

Parental health pre-conception, at conception, and post-conception have all been shown to impact offspring health. Specifically, nutritional status (both over- and undernutrition), insulin resistance, exercise, chemical exposure, and behavior in the parent before and during pregnancy can all affect both paternal and offspring health (Anderson et al., 2006; Anway et al., 2006; Chambers et al., 2016; Hanafi et al., 2016; Isganaitis et al., 2014; Jimenez-Chillaron et al., 2009; Murashov et al., 2016; Rodgers et al., 2013; Rodgers et al., 2015; Skinner et al., 2010; Stanford et al., 2015).

These patterns of parent-to-child transmission of disease risk could result from the impact of parental health per se on the pregnancy environment (particularly prominent in the case of maternal exposures) or on either maternal or paternal germ cells. Evidence that the isolated germ cell can mediate offspring disease was recently described by Huypens and collaborators, who utilized in vitro fertilization to demonstrate that germ cells harvested from mice exposed to nutritional factors (low-fat diet, normal diet and high-fat diet) are able to transmit metabolic phenotypes to offspring. Moreover, both offspring and parental sex influence the severity of offspring adiposity and glucose intolerance (Huypens et al., 2016).

Thus, both parental health as well as direct exposures experienced by germ cells may contribute to offspring developmental programming and disease risk.

• 2Are phenotypes transmitted via maternal or paternal lineage, or both?

Mechanisms contributing to transmission of disease risk likely depend on whether phenotypes are transmitted through the maternal or paternal lineage, or both (Table 1). Both parents contribute to offspring genetics as well as shared postnatal environments (e.g. family diet, activity, behavior, socioeconomic factors). Unique maternal factors include the well-recognized effects of maternal diabetes or obesity, altered structure/function of reproductive organs, alterations in vaginal or gut microbiome, mitochondrial DNA inheritance, placental function, or epigenetic phenotypes in female germ cells. By contrast, unique paternally-mediated effects on offspring implicate indirect effects on the fetal component of the placenta, seminal fluid proteins, or sperm epigenetic mechanisms (refer to Table 1 for detailed citations). Due to space restriction, we will only briefly review maternal effects on intergenerational disease transmission, and focus on paternal lineage mediated phenotypes in greater detail; for deeper information on the maternal lineage please refer to other reviews (Alfaradhi and Ozanne, 2011; Clarke and Vieux, 2015; Ferguson-Smith and Patti, 2011; Rando and Simmons, 2015).

Table 1. Intergenerational vs. transgenerational inheritance.

	Maternal Lineage	Paternal Lineage
Nuclear DNA inheritance	Yes a	Yes b
Mitochondrial DNA inheritance	Yes c,d	No e
Epigenetic modification:
- Germ cells	Yes f,g	Yes h, i, j, k, l, m, n, o, p, q, r, s, t, u, v, w, x, y, z, aa, ab, ac
Maternal environment during pregnancy:
- Hormones	Yes ad
- Nutrients	Yes ae, af, ag, ah, ai, aj, ak, al, am, an, ao
- Metabolism	Yes ap, aq, ar, as
- Uterine structure/function	Yes at, au, av, aw	Not applicable
- Behavior	Yes ax, ay, ao, az
- Chemical exposure	Yes ba, bb
- Milk composition	Yes bc
Placental structure/function	Yes bd, ay, be, bf	Yes*, bg, bh
Shared postnatal environment:
- Diet	Yes bi, bj	Yes bj
- Behavior	Yes bk	Yes bk
- Enviromental chemicals	Yes bl	Yes bl
- Activity	Yes bm	Yes bm
- Microbiome	Yes bn	Yes bn

^*Paternal contribution to expression of placental development genes

^aFerguson-Smith et al., 2006;

^bSchagdarsurengin and Steger, 2016;

^cSaben et al., 2016;

^dWu et al., 2015;

^eWang et al., 2016;

^fBranco et al., 2016;

^gHuypens et al., 2016;

^hChen et al., 2016;

ⁱDaxinger et al., 2016;

^jde Castro Barbosa et al., 2016;

^kDenham et al., 2015;

^lDonkin et al., 2016;

^mHammoud et al., 2009;

ⁿHuypens et al., 2016;

^oJodar et al., 2013;

^pLane et al., 2014a;

^qMcPherson et al., 2015;

^rOkada et al., 2010;

^sOst et al., 2014;

^tPalmer et al., 2012;

^uPeng et al., 2012b;

^vRadford et al., 2014;

^wRodgers et al., 2015;

^xSchuster et al., 2016;

^ySharma et al., 2016;

^zShea et al., 2015;

^aaSiklenka et al., 2015;

^abTerashima et al., 2015;

^acvan de Werken et al., 2014;

^adSeckl and Meaney, 2004;

^aeAlfaradhi and Ozanne, 2011;

^afBenz and Amann, 2010;

^agCarone et al., 2010;

^ahChambers et al., 2016;

^{ai, aj}Dunn and Bale, 2009, 2011;

^akJimenez-Chillaron et al., 2005;

^alJimenez-Chillaron et al., 2009;

^amMcCurdy et al., 2009;

^anO’Neill et al., 2014;

^aoStanford et al., 2015;

^apDabelea and Crume, 2011;

^aqGuenard et al., 2013;

^arIsganaitis et al., 2014;

^asYu et al., 2013;

^atBromfield et al., 2014;

^auChan et al., 2015;

^avIbanez et al., 2011;

^awvan Dijk et al., 2015;

^baHolliday, 1998;

^bbSkinner, 2010;

^bcSilverman, 1992,

^bdGatford et al., 2010;

^ayHowerton et al., 2013;

^beMaclennan et al., 2004;

^bfNelissen et al., 2011;

^bgFerguson-Smith et al., 2006;

^bhWang et al., 2013;

^biJimenez-Chillaron et al., 2006;

^bjSavage et al., 2007;

^bkMcGowan et al., 2009;

^blBaccarelli and Bollati, 2009;

^bmBarres and Zierath, 2016;

^bnSong et al., 2013.

• 3Is the inheritance pattern intergenerational or transgenerational?

Understanding the distinction between these two terms is important, as they reflect differences in pathophysiology and lead to consideration of distinct mechanistic possibilities. Parental (F0) exposures before conception or during pregnancy can influence the health of offspring (F1) and grandoffspring (F2); these intergenerational effects could be mediated by direct effects on the somatic cells of the developing F1 embryo/fetus, potentially affecting proliferation, differentiation, and organ development. Such exposures also directly impact the developing embryonic germ cells (which will become F1 adult gametes) and thus can affect F2 offspring. Thus, a single exposure during pregnancy in the F0 generation female can have intergenerational effects (Deans and Maggert, 2015; Heard and Martienssen, 2014; Patti, 2013).

By contrast, a transgenerational phenotype requires persistence of the phenotype, even after cessation of the insult, i.e. even in cells not directly exposed to the insult. Persistence of phenotypes in cells not directly exposed to the original insult might indicate that a stable modification in DNA or chromatin regulation may have occurred which was passed on via unexposed gametes to the next generation or alternatively, reconstructed anew in each generation in response to a transmitted mechanistic cue. Transgenerational inheritance can be modeled by exposures occurring during adult life in the F0 generation which impact the F2 generation, as the F1 germ cells have not been directly exposed. Conversely, in the case of exposures during pregnancy, demonstration of transgenerational effects requires phenotypes to be observed in the F3 generation, in which no cells were directly exposed (Rando and Simmons, 2015).

What is the evidence that epigenetic mechanisms contribute to these phenotypes?

Epigenetic regulation of transcription is a strong candidate for mediating environmentally-induced risk for metabolic disease, as it is a highly dynamic process occurring throughout the lifespan of an individual, including during intrauterine life. A wide range of environmental stressors, including global or protein malnutrition, high fat diet, maternal obesity or excessive gestational weight gain, maternal insulin resistance, placental insufficiency, behavior, glucocorticoids, and hypoxia, if occurring during critical phases in development, can induce lasting changes in tissue structure and function associated with increased disease risk during later life and in subsequent generations (Jimenez-Chillaron et al., 2005; Seckl and Meaney, 2004; Simmons et al., 2001). As shown in Table 1, many factors contribute to non-genetic mechanisms from the maternal lineage. However, studying paternal lineage transmission in controlled experimental models is more likely to isolate epigenetic mechanisms, since the impact of maternal metabolism on the pregnancy milieu can be excluded.

Epigenetics: definitions and general mechanisms

The term epigenetics, or “above genetics,” was initially used by Conrad Waddington in 1942 to refer to “the branch of biology that studies the causal interactions between genes and their products which bring the phenotype in being” (Waddington, 2012). Waddington’s definition was used to describe how cells with the same genetic material can give rise to different cell types during the course of development. Around 1975, Riggs, and Holliday and Pugh, suggested that heritable non-genetic modulation of gene transcription, previously demonstrated in yeast, Caenorhabditis elegans and Drosophila melanogaster, could also be present in mammals (Holliday and Pugh, 1975; Riggs, 1975). These articles reshaped the definition of epigenetics; for more information on this topic, please refer to (Felsenfeld, 2014). We consider a broader definition, encompassing changes in regulation of gene expression which occur without DNA base pair sequence changes (Daxinger and Whitelaw, 2012).

DNA methylation, traditionally considered a relatively stable modification, is actually a highly dynamic modification regulated by both methyltransferases and iterative demethylating enzymes. DNA methylation can be efficiently replicated on daughter strands by maintenance methyltransferase enzymes such as DNMT1. DNMT1 is guided by palindromic domains within cytosine- and guanine-rich regions on the parental template DNA strand (Bestor, 1992). Additional DNA methyltransferases, DNMT3A and DNMT3B, are responsible for de novo DNA methylation (Okano et al., 1998). Recently, DNMT3C was identified as a mediator of transposon methylation in male germ cells (Barau et al., 2016). The cofactor DNMT3L, which recruits DNMT3A/B, is also required for de novo methylation in the germ line (Bourc’his et al., 2001).

DNA demethylation can be either active or passive (Messerschmidt et al., 2014). Passive demethylation refers to dilution of methyl marks during replication cycles, if methylation is not reestablished after DNA synthesis (Jones, 2012). Active demethylation is mediated in a regulated fashion by enzymes such as the TET enzymes that convert 5-methylcytosine to 5-hydroxy-methylcytosine (Tahiliani et al., 2009).

Modification of histone tails at multiple sites provides a potent method for the regulation of gene expression. Specific marks, such as H3K4me1, characterize regions of active recruitment of transcriptional complexes and initiation of transcription, typically on gene promoters. By contrast, trimethylation of lysine 27 on histone 3 (H3K27me3) conferred by the polycomb repressor complexes (PRC) recruited to DNA promotes gene silencing by increasing DNA compaction, inhibiting transcription by Polymerase II, and excluding the SWI/SNF family of nucleosome remodelers (Di Croce and Helin, 2013; Jenuwein and Allis, 2001). H3K27ac occupancy is characteristic of active enhancer regions (Creyghton et al., 2010). These and other histone tail modifications are in turn regulated by site-specific histone modifying enzymes: methyltransferases/demethylases, acetylases/deacetylases, and protein phosphatases/kinases (Feng et al., 2010; Jones, 2012).

Non-coding RNAs also participate in post-transcriptional regulation of gene expression, and hence fit within Waddington’s definition. MicroRNAs (miRNAs), small non-coding RNA of ~22 nucleotides, bind to mRNA, ultimately inducing its degradation or inhibiting its translation (Esteller, 2011). PIWI-associated RNA (piRNAs) and other small non-coding RNAs regulate gene expression by modulating DNMT3 activity in germ cells (Kuramochi-Miyagawa et al., 2008).

Hence, the central dogma of biology, with a linear flow of DNA → RNA → protein is overly simplistic. Gene and protein expression is regulated by several layers of epigenetic mechanisms, resulting in an intricate combination of chromatin accessibility and transcription factor recruitment to coding sequences (Esteller, 2011; Feng et al., 2010; Jenuwein and Allis, 2001; Jones, 2012; Messerschmidt et al., 2014).

Epigenetic modifications during development

Epigenetic mechanisms are particularly strong candidates for developmental effects as epigenetic phenotypes are modified during normal development. Waves of demethylation and de novo methylation occur during embryonic development (Heard and Martienssen, 2014). For example, demethylation occurs as early as E3.5–4.5, consistent with the need for activation of expression of those genes critical for early development which are typically silenced during adult life. DNMT3s restore methylation in embryonic somatic cells at the epiblast stage (E6.5). After this methylation phase, the embryo undergoes accelerated growth and differentiation, with initiation of organogenesis. At day ≈E13.5–16.5 progenitor germ cells are mature and migrate to the genital ridge. Before this time, they undergo profound loss of epigenetic marks which are then regained during germline development. In male fetuses, those cells undergo re-methylation of sex-specific genes as they differentiate in the gonads during the prenatal period. In female fetuses, those cells remain in a quiescent state, largely unmethylated, with subsequent re-methylation occurring after birth (Messerschmidt et al., 2014).

Consideration of this epigenetic dynamic that occurs during normal embryonic development is important when analyzing the impact of environmental exposures during prenatal life. Exposures during embryonic life can not only directly affect the somatic cells of the developing fetus, but also their developing progenitor germ cells, thus potentially impacting future generations (Daxinger and Whitelaw, 2012; Ferguson-Smith and Patti, 2011; Lane et al., 2014b).

Potential epigenetic mediators of intergenerational metabolic disease

Maternal mediators

Maternal metabolism during pregnancy, including over- or undernutrition, altered macronutrient composition, obesity, insulin resistance, or diabetes, behavior and chemical exposure, promotes alterations in the metabolism and health of her offspring (Anway et al., 2005; Anway et al., 2006; Beauchamp et al.; Dunn and Bale, 2009, 2011; Isganaitis et al., 2014; Jimenez-Chillaron and Patti; McCurdy et al., 2009). For example, preexisting or gestational diabetes results in excessive circulating glucose crossing the placental barrier, stimulating insulin secretion by the fetus. This has been associated with offspring excessive growth, increased adiposity and hypothalamic dysregulation (Dabelea and Crume, 2011). Children born after maternal weight loss resulting from bariatric surgery have reduced risk of developing metabolic disease than their siblings born before maternal weight loss (Guenard et al., 2013). Interestingly, these phenotypic differences between siblings are associated with differential methylation in genes that regulate insulin resistance (Guenard et al., 2013).

Early life exposure to stressors also directly affect the developing female reproductive system; from size of the uterus (Ibanez et al., 2011), to the health of the oocytes (Wu et al., 2015), trophoblast implantation (Branco et al., 2016) and placentation (reviewed by (van Dijk et al., 2015)). In fact, oocytes from obese female mice have dysfunctional mitochondria, which is transmitted transgenerationally (Saben et al., 2016; Wu et al., 2015). Placental dysfunction can impact offspring development, potentially via altered delivery of nutrients and oxygen to the developing fetus (Reik et al., 2003). Experimentally, impaired placental function can be modeled by bilateral uterine artery ligation, which induces intrauterine growth restriction of the fetus (Simmons et al., 2001), in parallel with decreased expression of DNA-methylating enzymes, altered one-carbon metabolism, increased S-adenosyl-homocysteine levels, silencing of Pdx-1 expression in islets, and increased offspring diabetes risk (Gatford et al., 2010; Maclennan et al., 2004). Imprinted genes are particularly important in placental development, size, and function. Paternally expressed genes generally increase fetal and placental size, while several maternally expressed genes limit placental growth, as corroborated by mouse models (reviewed in (Coan et al., 2005).

Methyl donors implicated in epigenetic control, such as folate and betaine, are important for offspring health (O’Neill et al., 2014). For example, a remarkable study by Padmanabhan et al., identified non-genetic multigenerational effects in wild-type mice derived from maternal grandparents deficient in methionine synthase reductase (Mtrr) (Padmanabhan et al., 2013). Furthermore, in a series of classical studies by Waterland and colleagues, maternal supplementation with methyl donors altered methylation at IAP elements upstream of the agouti (Avy) locus in offspring, darkening offspring coat color (Waterland et al., 2008). Whether broad supplementation with methyl donors would have beneficial effects remains uncertain, as other studies have associated those metabolites with induction of epimutations which may not benefit the offspring (O’Neill et al., 2014). Although human studies in offspring can be confounded by genetic differences, these data suggest that epigenetic regulation resulting from maternal exposures can contribute to phenotypes in her children.

Paternal mediators

An increasing body of literature derived from several experimental paradigms indicates that metabolic phenotypes can be also be transmitted via the paternal lineage (Schagdarsurengin and Steger, 2016), independent of genetics (Figure 2). For example, HFD feeding of male rats induces glucose intolerance and β-cell dysfunction in female offspring (Ng et al., 2010). Similarly, F1 offspring of male mice fed a low-protein/high-carbohydrate diet have increased hepatic expression of genes involved in lipid and cholesterol biosynthesis, and reduced liver cholesterol. This may be linked to reduced Ppara expression levels and parallel increases in methylation in a Ppara enhancer region (Carone et al., 2010). Likewise, Wei and colleagues demonstrated that paternal prediabetes, induced by combination of streptozotocin and HFD, also resulted in glucose intolerance and insulin resistance in both F1 and F2 offspring mice (Wei et al., 2014). Impressively, even a single bout of fasting in male mice prior to breeding reduces plasma glucose in his offspring. This effect can be amplified in female offspring if the fathers are repeatedly fasted prior to breeding (Anderson et al., 2006).

Figure 2. Intergenerational vs. transgenerational inheritance.

Exposures during pregnancy (left) are termed intergenerational when impacting F1 and F2 generation offspring, but are transgenerational when affecting the F3 generation. By contrast, exposures occurring during adult life in either males or females (right) can yield intergenerational effects on the F1 generation and transgenerational effects in F2 offspring. Blue arrows – intergenerational effects; red arrows – transgenerational effects.

Even prior environmental exposures experienced by a male during critical periods of development (and not persistent at the time of breeding) can influence the health of his offspring. We and others have demonstrated that prenatal and early postnatal exposures can reduce the number and function of pancreatic islets (Jimenez-Chillaron et al., 2005; Wei et al., 2014), nephrons (Benz and Amann, 2010), alter reproductive organs (Chan et al., 2015) and even stem cells (Woo et al., 2011). In our model of low-birth weight, F1 male mice (with prior exposure to maternal undernutrition during prenatal life) produce F2 male offspring with increased adiposity and impaired glucose tolerance (Jimenez-Chillaron 2009; Radford, 2014). These phenotypes even extend to F3 male offspring (unpublished data). Remarkably, effects in offspring occur even when F1 fathers are phenotypically healthy at the time of breeding. Collectively these data indicate that even previous exposures can influence paternal effects on his offspring.

Impressive examples of transgenerational transmission of non-genetic traits is provided by exposure to obesogens and endocrine dysruptors (Baccarelli and Bollati, 2009; Janesick and Blumberg, 2016; Skinner et al., 2010). Treatment of pregnant females with vinclozolin, an antiandrogenic conmpound, or methoxychlor, an estrogenic compound, increased male infertility and reduced spermatogenic capacity in male offspring. Those effects were persistent and reported to be transmitted through the male germline transgenerationally from F2 to F3 and F4 (Anway et al., 2005). Rodent models of maternal exposure to obesogens - dichlorodiphenyltrichloroethane (DDT) and tributyltin (TBT), bisphenol-A (BPA) and others, showed that grandoffspring of those exposed-females have increased risk of obesity and T2D; these effects are transmitted through the male lineage (Janesick and Blumberg, 2016).

Collectively, these findings demonstrate that paternal exposures, either those experienced during adult life or during his intrauterine development, can both potently modify offspring metabolism, potentially via sperm-mediated epigenetic transmission.

Impact of sperm phenotypes on offspring metabolism

Direct DNA damage to sperm can also impact offspring phenotypes. For example, treatment of sperm with hydrogen peroxide is associated with low birth weight, glucose intolerance, and increased adiposity in adult female offspring (Lane et al., 2014a). HFD-induced obesity can also reduce sperm number, motility, and alter morphology. High fat feeding also reduces sperm capacitation and oocyte binding, increases spontaneous acrosome reaction, oxidative stress, mitochondrial membrane potential and DNA fragmentation; these effects can be reversed by control diet and swim exercise for 6 weeks (Palmer et al., 2012).

To understand potential mechanisms by which sperm may transmit epigenetic signals from father to offspring, we need to consider how both genetic and epigenetic information are packaged in sperm. During spermatogenesis, in the haploid spermatids, the majority of histones are exchanged for protamines, which permit DNA compaction needed for formation of mature sperm (reviewed by (Schagdarsurengin and Steger, 2016). These sperm-derived protamines are replaced by maternal histones post-fertilization, at the zygote stage (Jones, 2012). Interestingly, not all DNA in sperm is bound by protamines. The small subset of loci retaining histones (5–10%) in mature sperm are enriched for RNA transcripts that may regulate early development and potentially contribute to epigenetic inheritance (Jodar et al., 2013).

Which epigenetic marks are responsible for intergenerational transmission of information via the sperm? Candidate mediators include changes in DNA methylation, chromatin-mediated alterations in transcriptional control, and non-coding RNA. DNA methylation has been traditionally assumed to be a more stable epigenetic mark and thus more broadly investigated as a candidate mediator of sperm-mediated phenotypes. For example, paternal haploinsufficiency for the histone 3 lysine 9 methyltransferase Setdb1 influences coat color of wild type offspring of Avy mutant mice. Paternal haploinsufficiency of Setdb1 does not alter methylation at the Setdb1 locus itself, but results in hypomethylation of transposable elements of the ERVK class that potentially act in trans to regulate expression of Avy gene on the maternally-inherited chromosome, and thus influence coat color (Daxinger et al., 2016). Interestingly, this mechanism depends on the co-repressor Trim28; prior studies demonstrated that Trim28 haploinsufficiency causes a bi-stable obesity phenotype in both mice and humans. This obesity phenotype occurs despite identical genotypes, and is dependent on reduced expression of the imprinted genes Peg3 and Nnat and reduced expression of DNA methyltransferase enzymes in the obese mice (Dalgaard et al., 2016).

Studies from our group using methylation-dependent immunoprecipitation demonstrated that intrauterine exposure to caloric restriction during the last week of gestation can alter the sperm methylome in those directly-exposed F1 offspring males; reduction in methylation at specific differentially methylated regions was confirmed by pyrosequencing (Radford et al., 2014). Since DNA methylation is not altered at the same loci in tissues of F2 embryo offspring of these F1 males, direct “transmission” of differential methylation is an unlikely mechanism for intergenerational inheritance in this model. However, gene expression is altered in F2 embryos at these loci, suggesting the possibility that other mechanisms could alter chromatin states and contribute to the observed transcriptional dysregulation in F2 offspring. For example, differential methylation in F1 sperm at the KCNJ11 locus, immediately downstream of the sulfonylurea receptor SUR1, is associated with differential expression of SUR1; alterations in transcription of this key beta-cell regulatory gene could contribute to observed reductions in insulin secretion, reduced efficacy of the sulfonylurea tolbutamide, and impaired glucose tolerance in F2 offspring.

Other studies have also identified differential methylation in sperm at loci relevant for offspring metabolism. In a paternal prediabetes model, methylation of genes linked to glucose metabolism and insulin signaling was also altered in sperm (Wei et al., 2014). Donkin and colleagues recently reported that sperm methylation differed in obese vs. lean humans, and that these patterns were altered in response to bariatric surgery and associated improvements in metabolism (Donkin et al., 2016). Interestingly, genes adjacent to differentially methylated loci in both obesity and in response to bariatric surgery were enriched for those regulating control of food intake in the central nervous system. By contrast, Carone and colleagues did not detect changes in DNA methylation in the sperm of fathers fed a low-protein/high-carbohydrate diet, despite effects on offspring metabolism (Carone et al., 2010).

Chemical exposures have also been demonstrated to have impact on sperm methylation (Anway et al., 2005; Anway et al., 2006). Sperm analysis of F3 descendents of rats treated with a mixture of plastic derived endocrine disruptor compounds bisphenol-A (BPA), bis (2-ethylhexyl) phthalate (DEHP) and dibutyl phthalate (DBP) showed differentially methylated regions in the promoter of genes associated with metabolic disease, such as Gdnf, Fgf19 and Esrra (Manikkam et al., 2013).

How can we interpret these diverse results for sperm DNA methylation? Firstly, differences between studies may reflect varied methodologies used to assess DNA methylation, such as reduced representation bisulfite sequencing (RRBS), MeDIP-Seq, or whole-genome bisulfite sequencing(Hisano et al., 2013; Martinez et al., 2014; Radford et al., 2014; Shea et al., 2015). Secondly, the timing of the paternal exposure differs between models. Paternal exposures occurring during intrauterine life would be expected to have particularly strong impact on developing germ cells, given that the exposure is occurring during a period of genome-wide modulation of methylation. By contrast, exposures occurring during postnatal life might have lower impact, influencing only spermatogonia or mature spermatozoa (Schagdarsurengin and Steger, 2016). Nevertheless, these data collectively indicate that paternal exposures can modulate sperm DNA methylation, with variable extent potentially depending on the timing and type of insult. Genetic sequence variation between experimental groups may also contribute to differences in methylation patterns, potentially linked to rDNA copy number variation and promoter sequence (Holland et al., 2016; Shea et al., 2015). rDNA copy number has been hypothesized to contribute to the impact of nutrition on offspring phenotypes. For example, adult Drosophilae fed high concentrations of dietary yeast (overnourished) have reduced ribosomal DNA (rDNA) copy number in somatic cells and alterations in insulin signaling, with fewer copies of these rDNA in offspring; thus, alterations in rDNA can occur in the germline, and can be heritable (Aldrich and Maggert, 2015; Ost et al., 2014).

Histone modifications have also been implicated as a mediator of paternal lineage phenotypes. Interestingly, transgenic mice overexpressing lysine-specific histone demethylase 1 (LSD1) in male gonads have reduced H3K4me2 histone methylation in sperm, without changes in DNA methylation. The wild type offspring of LSD1-overexpressing fathers have several developmental phenotypes e.g. skeletal abnormalities; these can persist transgenerationally, even in the absence of the original transgene. Paternal histone H3K4 demethylation in sperm also alters sperm RNA content over generations (Siklenka et al., 2015).

Histone modifications can also be perturbed in response to nutritional modifications. HFD-fed male mice have increased retention of histone H3 at genes involved with embryonic development (Terashima et al., 2015). Nutrition also influences chromatin regulation in flies; acute provision of a high sucrose diet (as short as 24 hours) yielded loss of the repressive H3K9me3 histone mark and polycomb complex in the sperm of exposed fathers. These same patterns were observed in somatic tissues of progeny over several generations. Comparison of these findings to publicly available studies in mice and humans revealed similar depletion of ortholog pathways to the Drosophila H3K9 histone methyltransferase (Su (var) 3–9) in obese individuals (Ost et al., 2014). In humans, H3K9 heterochromatin domains can be paternally transmitted to the zygote via sperm (van de Werken et al., 2014). However, no differences sperm histone DNA positioning, as evaluated by MNase-seq, were observed between lean and obese men (Donkin et al., 2016). Regulation of protamine levels could also alter chromatin regulation, as seen for the KCNQ1OT1, MEST, SNRPN, PLAGL1, PEG3, H19, and IGF2 loci in sperm of oligozoospermic men (Hammoud et al., 2010; Schagdarsurengin and Steger, 2016).

Alterations in small noncoding RNAs (sncRNAs) in sperm, including miRNA, piRNA, and tRNA, have been proposed to affect epigenetic modulation of gene expression in many models (Jodar et al., 2013). For example, stress-induced dysregulation of the hypothalamic-pituitary-adrenal axis and its transgenerational transmission can be recapitulated by microinjection of 9 distinct miRNAs in the developing mouse embryo (Rodgers et al., 2015). Similarly, males subjected to restraint stress generate offspring with hyperglycemia due to increased hepatic gluconeogenesis. The stressed fathers exhibit glucocorticoid-dependent hypermethylation of the Smbt2 gene in sperm, a pattern recapitulated in liver of their F1 offspring. This pattern, in turn, results in decreased expression of miR-466-3p, encoded by the Smbt2 gene; this alters post-transcriptional regulation of the PEPCK gene and increases PEPCK protein, a rate-limiting step for gluconeogenesis (Wu et al., 2016). Similar to what has been demonstrated in other models, in the transgenerational model of male vinclozin exposure, sperm sncRNAs are also altered, the latter being a potential mechanism of transgenerational transmission of epigenetic markers (Schuster et al., 2016).

Piwi-associated RNAs, or piRNAs, are another class of sperm-abundant small RNA, also linked to epigenetic regulation (Jodar et al., 2013). Heat-induced stress in C. elegans is propagated over generations via alterations in piRNA processing in the germline (Ni et al., 2016). Interestingly, recent human data indicate that piRNA content is altered in obese as compared with lean men (Donkin et al., 2016).

Dietary manipulation can also alter additional classes of small RNA in sperm (Chen et al., 2016). Male mice fed a low protein diet have increased expression of tRNA fragments (tRFs) in the sperm; these tRFs are transferred to the sperm via epididymal vesicles/exosomes. tRFs, particularly of the tRNA-Gly-CGG species, can repress the transcription of embryonic genes regulated by the long terminal repeat (LTR) of the endogenous retroelement MERVL (Sharma et al., 2016). Similarly, HFD-fed obese fathers as well as their F1 male offspring have differential expression of the miRNA let-7c in sperm; altered expression of let-7c was also observed in liver, muscle and adipose tissue of their female offspring. In adipose tissue, expression of let-7c inversely correlates with that of multiple metabolic target genes, such as uncoupling protein 2 (Ucp2), in both F1 and F2 female offspring (de Castro Barbosa et al., 2016).

Alterations in small RNAs have also been implicated in metabolic phenotypes in offspring and grandoffspring of obese Avy/a males, which display impaired glucose tolerance, higher serum insulin, and hepatic lipid accumulation. miRNAs and tRNAs, such as miR-10 and tRF-5-GluCTC, which are altered in sperm of grandoffspring of obese Avy/a males, could be interacting with the Ago2 enzyme in the oocyte cytoplasm to alter embryonic development (Cropley et al., 2016). Thus, small RNAs are likely to be important for epigenetically-mediated intergenerational and transgenerational phenotypes in multiple mammalian models.

It’s Not Just the Sperm

Not only the sperm, but also other components of semen may contribute to offspring phenotypes via alterations in female tract response. This was shown dramatically in studies by Bromfield and colleagues, in which mice without seminal fluid due to excision of the seminal vesicles, have reduced fertility, placental hypertrophy, and obesity, impaired glucose tolerance, and hypertension in male offspring. Lack of seminal fluid resulted in downregulation of embryotrophic factors Lif and Egf and up-regulation of the apoptosis-inducing factor Trail in the oviduct (Bromfield et al., 2014). Additional components of semen, including proteins (Bromfield et al., 2014) and exosomes (Vojtech et al., 2014), may also be important contributors to paternal intergenerational transmission. Both humans and mice have exosomes containing small non-coding RNAs in seminal fluid (Peng et al., 2012; Vojtech et al., 2014). As previously noted, tRNA-derived fragments may be important for sperm maturation within the epididymis, and appear to be altered in response to paternal diet (Sharma et al., 2016). Interestingly, injection of sperm tRNA fragments from HFD males into normal oocytes induces metabolic disease in their offspring, suggesting a potential pathogenic role for exosomes in paternal transmission (Chen et al., 2016).

Can epigenetic mechanisms proposed to contribute to phenotypic transmission from father to offspring be modulated?

While epigenetic marks were previously considered to be relatively fixed, it is now clear that epigenetic regulation of transcription is a dynamic process. For example, epigenetic states can be acutely modified in response to exercise (Barres et al., 2009). Emerging data indicate similar plasticity of epigenetic marks even in sperm. For example, weight loss and exercise can affect both epigenetic marks and offspring metabolism. Exercise training in humans (Denham et al., 2015) can modulate sperm methylation. Bariatric surgery in obese males is not only associated with improved metabolism, but also with alterations in methylation at loci adjacent to genes regulating food intake (Donkin et al., 2016). Whether these changes have any impact on offspring phenotype is difficult to assess in humans and will require long-term follow up studies.

Offspring studies in mouse models yield divergent data. Some studies indicate that exercise training in fathers can improve offspring phenotypes. Diet and exercise over 8 weeks in HFD-fed male mice normalized the abundance of X-linked sperm miRNAs and prevented metabolic impairments in female offspring (McPherson et al., 2015). By contrast, one study indicated that long term-exercise in male mice (F0) predisposes their male offspring to glucose intolerance and insulin resistance, potentially due to exercise-induced increases in ROS leading to epimutations and changes in sperm miRNA content (Murashov et al., 2016). Dietary modification, including paternal vitamin and antioxidant supplements during caloric restriction in fathers, can modulate both sperm phenotypes and reduce risk in their offspring (McPherson et al., 2016). An important research imperative is to determine whether interventions which modulate paternal health, metabolism, and sperm epigenetic states will indeed be accompanied by improved health of offspring. If successful, interventions might interrupt vicious cycles of intergenerational metabolic disease risk.

Summary and Perspectives

Collectively, multiple lines of evidence now indicate that history of environmental exposures, as well as current metabolic or environmental exposures, can influence offspring metabolism, and that epigenetic marks in sperm can be altered in response to these same exposures. If we are to disentangle mechanisms responsible for intergenerational transmission of phenotypes, we need to understand how and when epigenetic perturbations in germ cells influence offspring. Are germ cell epigenetics influencing early preimplantation embryonic development, later-stage embryonic growth, tissue-specific stem cell pools, or developmental trajectories which modulate adult function? Might they influence susceptibility to adult environmental insults such as inactivity or suboptimal diet? Available data, though limited, provides support for each of these potential mechanisms. Additional studies will be required to fully assess the impact of germ cell epigenetics across the lifespan.

In summary, it is clear that the paternal lineage is responsible for more than just its genetically-encoded information. A variety of distinct epigenetic mechanisms, such as DNA methylation, histone modification and small RNAs, may collectively reflect prior and current environmentally-determined phenotypes in the father, thus providing a robust legacy of critical molecular information for his descendants which contribute to paternally-mediated inheritance of phenotypes and initiation of a vicious cycle of disease risk across generations. Thus, identification of interventions which improve health in men of reproductive age and potentially interrupt cycles of disease risk in their descendants is a critical public health and research imperative for the 21^st century.