Understanding Dietary Intervention-Mediated Epigenetic Modifications in Metabolic Diseases

The global prevalence of metabolic disorders, such as obesity, diabetes and fatty liver disease, is dramatically increasing. Both genetic and environmental factors are well-known contributors to the development of these diseases and therefore, the study of epigenetics can provide additional mechanistic insight. Dietary interventions, including caloric restriction, intermittent fasting or time-restricted feeding, have shown promising improvements in patients’ overall metabolic profiles (i.e., reduced body weight, improved glucose homeostasis), and an increasing number of studies have associated these beneficial effects with epigenetic alterations. In this article, we review epigenetic changes involved in both metabolic diseases and dietary interventions in primary metabolic tissues (i.e., adipose, liver, and pancreas) in hopes of elucidating potential biomarkers and therapeutic targets for disease prevention and treatment.


INTRODUCTION
The continuous rise in metabolic diseases, such as obesity, type 2 diabetes (T2D ; Table 1) and non-alcoholic fatty liver disease (NAFLD), is one of the leading causes of patient morbidity and mortality worldwide (Saklayen, 2018) and cannot be solely explained by the contribution of genetic and environmental factors. Indeed, epigenetics, which constitutes the reversible and heritable change in gene expression without modification of the underlying nucleotide sequence, serves as a mechanistic bridge. Epigenetic changes influenced by environmental cues can result in altered gene expression associated with metabolic function and dysfunction.
Overnutrition, especially of highly processed foods (Hall et al., 2019), accompanied by erratic diurnal eating patterns, constitute the major environmental contributors to the epidemic state of metabolic diseases today. As such, switching to a regular, nutritious diet can promote processes of maturation and restoration, and protect against the development of chronic metabolic disorders (Di Francesco et al., 2018). Since the applicability of pharmacological interventions in the treatment of metabolic disorders is limited by issues regarding off-target effects, patient compliance and tolerability, as well as lack of sufficiency in disease management (Longo and Panda, 2016); dietary interventions have become a promising, low-risk alternative or supplementary form of therapy. By adjusting meal timing and/or content, dietary interventions have shown continued success in reducing risk factors, inducing beneficial pleiotropic effects and ameliorating disease states (Longo and Panda, 2016). These dietary interventions involve limiting food intake of entire (i.e., fasting interventions) or selected nutrient compositions (i.e., nutritional interventions), without disturbing energy balance or inducing malnutrition. Specifically, fasting interventions can be categorized into intermittent fasting (IF) and periodic fasting (PF), where food intake is limited either on a daily/weekly basis or on a monthly basis, respectively (Anton et al., 2018;Yong-Quan Ng et al., 2019; Figure 1). IF cycles typically last 24 h and are separated by one or more days, whereas PF cycles last two or more days and are separated by at least a week (Longo and Mattson, 2014). Different forms of IF vary in their timing of meals and include the daily time-restricted feeding (TRF), and the weekly 5:2, 2:1, or 1:1 IF regimens. Moreover, nutritional interventions vary in their meal content and include caloric restriction (CR), dietary restriction (DR), ketogenic diet (KD), and fasting-mimicking diet (FMD).
These benefits conferred by dietary interventions involve cellular adaptations within various metabolic tissues, which are mediated by epigenetic modifications. Due to the plasticity of epigenetic factors, environmental changes, such as dietary interventions, which alter food intake and composition, have a significant impact on the epigenome. In this article, we will first review epigenetic changes in metabolic disease with a particular emphasis on adipose tissues, liver, and pancreas. We will primarily focus on DNA methylation and post-translational histone modifications (Figure 2), with the exception of non-coding RNAs reviewed elsewhere (Deiuliis, 2016;Green et al., 2017). Next, we will discuss how fasting as a component of most dietary interventions and caloric restriction modulate epigenetic regulation in these tissues. To conclude, we will also briefly review the epigenetics of gut microbiota and ketone body metabolism in the context of dietary interventions. Overall, the understanding of both metabolic diseases and dietary interventions from an epigenetic perspective will provide new insights for metabolic disease prevention, management and treatment.

Adipose Tissue in Health and Metabolic Disease
Obesity is a serious metabolic disease that has reached worldwide rates of over 27.5% for adults and 47.1% for children (Ng et al., 2014). It has been estimated that obesity is 40−70% inheritable, where genome-wide association studies (GWAS) account for 20% of the variation (Locke et al., 2015). As such, it is becoming increasingly clear that epigenetic modifications serve as a link between environmental and genetic causes of obesity.
Uncontrolled adipose tissue expansion and accompanying dysfunction drive obesity and associated metabolic pathogenesis (Choe et al., 2016). Adipocytes initially expand in size (i.e., hypertrophy) to accommodate increases in energy intake relative to energy expenditure. When adipocytes become lipid-engorged and can no longer store the excess energy, adipogenesis, FIGURE 2 | Description of the epigenetic change and its transcriptional modulators and markers in DNA methylation, histone methylation, and histone acetylation. Simplified diagrams show the forward and reverse reactions to each epigenetic mechanism.
Adipose tissue is classified into white adipose tissue (WAT) and brown adipose tissue (BAT). WAT stores excess energy in the form of TG and is localized to subcutaneous (i.e., "beneath the skin") and visceral depots (i.e., "surrounding internal organs"). BAT, by contrast, utilizes stored energy to produce heat in response to stimuli like cold stress, primarily via the uncoupling protein-1 (UCP-1), in a process known as non-shivering thermogenesis (Figure 3). BAT is also distributed subcutaneously (e.g., under the clavicles and in the interscapular region) and viscerally (e.g., perivascular, periviscus and around solid organs) (Sacks and Symonds, 2013;Jung et al., 2019); however, the initial distribution and size of BAT, as found in infants and young children, decreases with age. Moreover, as the "whitening" of brown adipocytes via dysregulated adipogenesis is associated with the development of obesity (Shimizu et al., 2014;Pellegrinelli et al., 2016;Longo et al., 2019), the "browning" of white adipocytes leading to increased thermogenesis may have therapeutic potential in the treatment of obesity (Choe et al., 2016).
Under feeding conditions, adipocytes take up and store circulating glucose and fatty acids (FA) through processes mediated by insulin-stimulated glucose transporter type 4 (GLUT4) translocation to the adipocyte cell surface and lipoprotein lipase (LPL) activity, respectively. Under fasting conditions, adipocyte lipolysis leads to FA release into circulation for use in other metabolic tissues, including muscle, kidney, gut and liver (Ahima and Flier, 2000;Choe et al., 2016; Figure 3). Adipose tissues also function as endocrine organs through secretion of cytokines (e.g., TNFα, IL-1, and IL-6) and FIGURE 3 | Epigenetic changes of adipose tissue in metabolic disease and dietary intervention. Adipose tissue in metabolic disease (i.e., obesity) predominantly consists of white adipocytes and presents with increased lipid content, inflammation and insulin resistance. Adipose tissue in dietary intervention is interspersed with both white and beige adipocytes and presents with reduced lipid content and inflammation as well as increased insulin sensitivity and thermogenesis. These physiological differences in adipose can be explained by epigenetic changes involving DNA methylation, histone methylation and histone and non-histone acetylation.
hormones (e.g., leptin, adiponectin), to mediate pathways of energy homeostasis, adipocyte differentiation, insulin sensitivity and inflammatory control (Wozniak et al., 2009). The reduction in insulin receptor density within the expanded adipose tissue and the subsequent development of insulin resistance both promote the uncontrollable release of FA, leading to aberrant lipid accumulation and lipotoxicity in peripheral tissues (Choe et al., 2016;Muir et al., 2016;Longo et al., 2019). Lipotoxicity, in combination with reduced anti-inflammatory adiponectin and increased pro-inflammatory cytokine secretion, promotes systemic inflammation associated with obesity, T2D and NAFLD.

DNA Methylation/Demethylation
DNA methylation is a key epigenetic modification involved in adipose development and function. Changes in DNA methylation in adipocytes have been associated with both the cause and effect of metabolic dysregulation in obesity, where hypomethylation appears to be the dominating change (Sonne et al., 2017).
A genome-wide screen has identified 625 significant differentially methylated regions (DMRs) associated with diet-induced obesity phenotypes, of which 232 DMRs correlate with high-fat diet (HFD) alone, and 249 regions are conserved in adipose tissue from obese subjects. Among these, 30 single-nucleotide polymorphisms (SNPs) are associated with T2D (Multhaup et al., 2015). As shown in Table 2, increased and decreased DNA methylations generally correspond to genes involved in positively (i.e., glucose homeostasis) and negatively (i.e., inflammation) regulating adipocyte metabolism, respectively. Additionally, DNA methylation profiles of diet-induced and genetically obese (i.e., ob/ob) mice revealed that methylation changes are more abundant in visceral than subcutaneous adipocytes (Sonne et al., 2017), with visceral fat being the greater contributor to obesity and its associated metabolic dysfunctions (Fox et al., 2007;Neeland et al., 2013).

Regulators of DNA methylation
DNA methyltransferases (DNMTs) catalyze DNA methylation. In contrast, ten-eleven translocation (TET) proteins catalyze the initial step of the reverse reaction (Figure 2; Jin and Robertson, 2013;Yong-Quan Ng et al., 2019). Altered expression of these DNA methylation modulators in adipose tissue can cause metabolic disease development and/or progression. High DNMT1 expression is found in adipocytes of obese humans (Kim et al., 2015). Similarly, increased Dnmt1 mRNA level is found in WAT of HFD-fed and genetically obese (i.e., db/db) mice compared to chow-fed and wild-type lean mice (Kim et al., 2015). Dnmt1 expression and activity in mouse 3T3-L1 adipocytes are induced by the pro-inflammatory cytokines TNFα and IL-1β (Kim et al., 2015). A known target of DNMT1 is Adipoq, which encodes the key anti-diabetic and anti-inflammatory adipokine, adiponectin (Stern et al., 2016). Over-expression of Dnmt1 in 3T3-L1 adipocytes increases methylation and decreases expression of Adipoq, while its knockdown results in the reverse (Kim et al., 2015), suggesting that direct hypermethylation and heterochromatin formation by DNMT1 at the Adipoq gene promoter is involved in obesity pathogenesis. In addition, the expression of Dnmt3a, not Dnmt3b, is increased in WAT of both diet-and genetically induced obese (i.e., ob/ob) mice (Parrillo et al., 2016;You et al., 2017) and the adipose-specific deletion of Dnmt3a in HFD-fed mice improves insulin sensitivity and glucose tolerance, independent of adiposity (You et al., 2017).
Using an unbiased screen, the authors have identified Fgf21 as a key target gene of DNMT3A. Correspondingly, adiposespecific Dnmt3a deletion leads to a decrease in the methylation and an increase in the expression of adipose Fgf21 in both WAT and BAT (You et al., 2017). However, experiments using adipose tissue-specific Dnmt3a and Fgf21 double knockout mice are still required to determine the mechanism of enhanced insulin sensitivity. Additionally, higher DNA methylation levels of FGF21 have been observed in the WAT of T2D patients, which negatively correlated to FGF21 mRNA expression, although DNMT3A levels were not measured (You et al., 2017).

Histone Methylation/Demethylation
Histone methylation, mediated by histone methyltransferases (HMTs) and reversed by histone demethylases (HDMs) (Figure 2), regulates adipogenesis through the addition and removal of activating and repressing histone methylation marks in adipocytes. Dysregulated adipogenesis through histone methylation impairs adipose tissue development and function and is associated with the maladaptive obesogenic condition.

H3K36 methylation (NSD2)
Dimethylation of histone H3 at lysine 36 (H3K36me2), an activation mark, is found to be protective against impaired adipose tissue function associated with obesity. The nuclear receptor binding SET domain protein 2 (NSD2) HMT mediates dimethylation at H3K36 for the activation of PPARγ-dependent gene programs, critical for mature brown and white adipocyte function. The depletion of the Nsd2-mediated H3K36me2 mark in adipocytes disrupts thermogenic function with "whitening" of BAT and increases insulin resistance of WAT (Zhuang et al., 2018).

Histone Acetylation/Deacetylation Class I/II HDACs
Histone acetylation and deacetylation involve the addition and removal of acetyl groups to lysine residues on histone tails, and are mediated by histone acetyltransferases (HATs) and histone deacetylases (HDACs), respectively (Figure 2). Total HDAC activity is decreased in the adipose tissue of obese individuals and HFD-fed mice (Bricambert et al., 2016), and mutations in HDAC4, a class II HDAC, have been associated with obesity (Williams et al., 2010). This reduced HDAC activity is mainly attributed to decreased HDAC5 and HDAC6 (class II HDAC) levels in WAT, which are accompanied by a decrease in inducible cAMP early repressor (Icer) function and an increase in its target activating transcription factor 3 (Atf3), associated with insulin resistance (Bricambert et al., 2016). Additionally, HDAC5 can interact with the GLUT4 enhancer factor (GEF) in adipocytes for the repression of Glut4 promoter activity (Sparling et al., 2008;Weems and Olson, 2011), suggesting a plausible mechanism by which Glut4 expression and insulin-mediated glucose uptake are dysregulated in obesity and T2D. Moreover, the expression of class I HDACs, HDAC1 and HDAC3, is reduced in the adipose tissue of obese female patients (Jannat Ali Pour et al., 2020), however, their role in adipose tissue is not yet well understood.

Class III HDAC (SIRT2)
Sirtuin 2 is another NAD + -dependent class III HDAC, which in contrast to the nuclear SIRT1, is primarily cytoplasmic, but can transiently shuttle to the nucleus for deacetylation of transcription factors (de Oliveira et al., 2012;Gomes et al., 2015). SIRT2 expression is found to be decreased in the adipose tissue of HFD-fed mice and obese patients (Krishnan et al., 2012;Perrini et al., 2020) and negatively correlates with their BMI, similar to SIRT1 (Perrini et al., 2020). In obesity, adipose SIRT2 expression is suppressed by adipose hypoxia-induced cellular hypoxia-inducible factor 1-α (HIF1α), which prevents SIRT2mediated post-translational deacetylation and activation of PGC-1α and its FAO transcriptional gene program Krishnan et al., 2012). Similarly, SIRT2 knockdown in isolated human adipose stem cells promotes adipogenesis and lipid accumulation through the induction of PPARG2, SREBF1C, FASN, ADIPOQ, and SLC2A4 gene expression (Perrini et al., 2020), whereas SIRT2 over-expression inhibits this process. Therefore, hypoxia-induced reductions of SIRT2 in obesity may contribute to adipocyte dysregulation by limiting oxidative capacity and increasing lipid mass.
Whether these functions of SIRT2 in adipocytes are mediated by its HDAC activity or cytoplasmic role remains unclear. Previous studies have demonstrated that SIRT2 regulates adipocyte differentiation through direct modulation of FOXO1 acetylation (Jing et al., 2007;Wang and Tong, 2009). On the other hand, SIRT2 controls mitosis by modulating histone H4K16 acetylation (Vaquero et al., 2006) and since mitotic clonal expansion is critical for adipocyte differentiation (Tang and Lane, 2012), this suggests that SIRT2 may regulate adipogenesis through histone modifications. Indeed, it has recently been shown that SIRT6, another class III HDAC, controls mitotic clonal expansion during adipogenesis by repressing kinesin family member 5C (KIF5C) expression with deacetylation of H3K9ac and H3K56ac at its promoter . Since the loss of Sirt6 blocks adipogenesis and Sirt6 mutant mice are extremely lean and die early with numerous severe metabolic abnormalities (Xiao et al., 2010), these results emphasize that proper development of adipocytes is critical for maintaining metabolic balance. Altogether, the lack of SIRT1-, SIRT2-and SIRT6-dependent deacetylation and activation of specific adipose gene programs can contribute to the development of metabolic conditions, including obesity and T2D.

Epigenetic Changes in Adipose Tissues in Dietary Intervention
Modulation of lipid compartmentalization and efficient utilization of excess energy in adipose tissues are critical targets for the treatment of obesity and related metabolic dysfunctions. Dietary interventions, including IF and CR, markedly reduce adipocyte size and depot weights in rodent models of obesity (Wheatley et al., 2011;Kim K.H. et al., 2017;Liu B. et al., 2019;Miyamoto et al., 2019), and confer improvements in adipose tissue inflammation and insulin sensitivity (Anson et al., 2003;Wheatley et al., 2011;Duncan et al., 2016;Gotthardt et al., 2016;Kim K.H. et al., 2017;Li et al., 2017;Kim Y.H. et al., 2019;Liu B. et al., 2019; Figure 3). Additionally, adipose thermogenesis via the induction of WAT "browning"(beige fat) and activation of BAT appear to be predominant pathways (Hatori et al., 2012;Hatting et al., 2017;Kim K.H. et al., 2017;Li et al., 2017), which elevate energy expenditure, mitochondrial biogenesis and energy dissipating capacity (Harms and Seale, 2013;Cypess et al., 2015). Here, we summarize evidence for epigenetic links between dietary interventions and resulting metabolic improvements.

DNA Methylation/Demethylation
Female obese patients subjected to bariatric surgery with significantly reduced body weight (∼27%) and food intake show reductions in global DNA methylation levels and differentially methylated genes associated with obesity and T2D in adipose tissues (Benton et al., 2015;Dahlman et al., 2015), thus providing context for weight loss and adipocyte reprogramming. These genes are associated with the regulation of body weight (LEPR, FTO), cholesterol homeostasis (CETP, LCAT), blood glucose (IRS1, INSR) (Benton et al., 2015), adipose tissue function (mTOR, RPTOR) (Macartney-Coxson et al., 2017), and epigenetics (FOXP2, HDAC4, DNMT3B) (Benton et al., 2015). Studies investigating DNA methylation changes in adipose tissue upon dietary interventions, however, are limited. In one study, obese women on a 6-month CR diet (1100−1800 kcal/day) who lost >3% of their body fat showed hypermethylation at three genomic loci in their subcutaneous adipose tissue. Genes at these loci were associated with lipid (e.g., PLCL4) and glucose (e.g., ENC1) homeostasis and epigenetic regulation (e.g., PRDM8) (Bouchard et al., 2010). In particular, the ectodermalneural cortex gene 1 (ENC1), previously associated with obesity (Zhao et al., 2000;Gerlini et al., 2018), was both differentially methylated (increased) and expressed (decreased) after CR treatment (Bouchard et al., 2010). In another study, 36-h of fasting in young, healthy men increased DNA methylation at the promoter site of LEP in subcutaneous adipose tissue, leading to a 3-fold decrease in plasma leptin levels (Hjort et al., 2017). Additionally, rosiglitazone, a PPARγ agonist, and a plausible CRmimetic, mediates TET2-dependent demethylation of promoter regions of PPARγ target genes, such as ADIPOQ and FABP4, and results in enhanced insulin-stimulated glucose uptake in 3T3-L1 adipocytes (Bian et al., 2018). Altogether, these CR-related DNA methylation changes in adipose tissue can potentially be used as biomarkers of improved adiposity.

Histone Methylation/Demethylation
Histone methylation changes in adipose tissue as a result of dietary interventions have not yet been studied. Thus, studies showing similar alleviation of the disease state can be used to suggest analogous epigenetic mechanisms. For example, histone demethylases LSD1 of H3K4 (Duteil et al., 2014), JMJD1A of H3K9 (Abe et al., 2018), and UTX (KDM6A) (Zha et al., 2015), and JMJD3 (KDM6B)  of H3K27 mediate the induction of BAT-selective genes (e.g., Ucp1, Pgc-1α, Ppara, Cidea) in WAT for the development of thermogenically active beige adipocytes. Consequently, whole-body Lsd1 overexpressing mice (Rosa26-Lsd1) present with reduced body weight gain and increased energy expenditure, associated with smaller adipocyte size and greater mitochondrial content in WAT (Duteil et al., 2014). These epigenetic mechanisms of thermogenesis are in response to cold stress. As adipose thermogenesis is a key adaptation seen with the implementation of dietary interventions, such as IF and TRF (Hatori et al., 2012;Hatting et al., 2017;Kim K.H. et al., 2017;Li et al., 2017), these epigenetic modulators may be involved in this process, but require further investigation to establish a causal link.

Class III HDAC (SIRT1)
Sirtuin 1, a class III HDAC that is upregulated in WAT of mice upon CR  and fasting (Picard et al., 2004), acts as a negative modulator of adipogenesis. SIRT1 complexes with NCoR/SMRT at the Pparg promoter to co-repress target genes involved in TG storage (Picard et al., 2004) and also posttranslationally deacetylates FOXO1 to increase the expression of its target gene Atgl in TG hydrolysis (Chakrabarti et al., 2011).

Class III HDAC (SIRT2)
Sirtuin 2, another class III HDAC, is also upregulated by CR and fasting in WAT (Wang et al., 2007) and can reduce cellular lipid stores and oxidative stress as seen with dietary interventions (Jing et al., 2007;Wang et al., 2007;Wang and Tong, 2009;Perrini et al., 2020). In adipocytes, the post-translational SIRT2mediated deacetylation of FOXO1 inhibits the transcriptional activation of PPARγ target genes involved in adipogenesis (Wang and Tong, 2009;Perrini et al., 2020). Additionally, SIRT2 can mediate the post-translational deacetylation of FOXO3a to promote the expression of FOXO target genes involved in the reduction of cellular reactive oxygen species (ROS) (MnSOD), the apoptotic clearance of damaged cells (Bim) and the inhibition of cell proliferation and propagation of mutations (p27 kip1 ) (Wang et al., 2007).

EPIGENETIC REGULATION IN LIVER DURING METABOLIC DISEASE AND DIETARY INTERVENTION The Liver in Health and Metabolic Disease
Fatty liver disease is closely associated with both obesity and T2D, with 82% of NAFLD patients presenting with obesity and 48% with T2D, in America (Younossi et al., 2016;Purnell et al., 2017). Intimately linked to systemic energy utilization and storage, the liver functions differently in the fed and fasted states, and its dysregulation can cause NAFLD. During feeding conditions, insulin promotes the storage of glucose into FA and TG or as glycogen through glycogenesis. During fasting, glucagon stimulates the mobilization of TG and glycogen stores for fuel delivery to extra-hepatic tissues, while simultaneously activating hepatic FAO and gluconeogenesis, fueled by adipocyte lipolysis and muscle proteolysis, respectively (Rui, 2014). Under conditions of metabolic dysregulation or disease, insulin resistance of the liver promotes inappropriate upregulation of gluconeogenesis while de novo lipogenesis (DNL) pathways remain insulin sensitive, contributing to hyperglycemia and hepatic lipid accumulation, respectively (Brown and Goldstein, 2008; Figure 4). Impaired FAO and very-low-density lipoprotein (VLDL) secretion as a result of insulin resistance further increase fat deposits in the liver (Bhatt and Smith, 2015). NAFLD develops when hepatic lipid stores exceed 5% of tissue mass, leading to increased inflammation, collagen deposition, fibrosis and cell death. If left untreated, NAFLD can progress to nonalcoholic steatohepatitis (NASH) and may continue to cirrhosis and hepatocellular carcinoma (HCC) (Byrne and Targher, 2015;Friedman et al., 2018). Altogether, the metabolic role of the liver in integrating these endogenous and exogenous fuel sources requires constant transcriptional modulation. Below we highlight some epigenetic changes regulating hepatic gene expression in both disease and dietary interventions.
FIGURE 4 | Epigenetic changes of liver in metabolic disease and dietary intervention. Liver in metabolic disease (i.e., NAFLD/NASH) presents with increased lipid content, inflammation, insulin resistance and fibrosis. The liver in dietary intervention has reduced lipid content and improved insulin sensitivity. These physiological differences of the liver can be explained by epigenetic changes involving DNA methylation, histone methylation and histone and non-histone acetylation.

DNA Methylation/Demethylation
Hepatic methylome and transcriptome studies have identified epigenetic links to the differentially expressed genes underlying the development of hepatic insulin resistance, T2D, and NAFLD (Ahrens et al., 2013;Nilsson et al., 2015;Kirchner et al., 2016;de Mello et al., 2017;Abderrahmani et al., 2018;Gerhard et al., 2018;Hotta et al., 2018). Notably, in the liver of T2D patients, the majority of significant differentially methylated CpG sites show reduced DNA methylation (Nilsson et al., 2015;Kirchner et al., 2016;Abderrahmani et al., 2018). Hypomethylation of a CpG site within or proximal to the activating transcription factor (ATF)-binding motif of hepatic genes accounts for increased expression of these genes involved in glycolysis (PFKL), DNL (ACACA, FASN), and insulin signaling (PRKCE) in obese and T2D patients (Kirchner et al., 2016). Moreover, the epigenetic induction of PDGFA, which encodes platelet-derived growth factor α (PDGF-AA), appears to be central to hepatic disease progression, as PDGF-AA causes insulin resistance by reducing hepatic insulin receptor density in a protein kinase C (PKC)-dependent manner (Thieringer et al., 2008). In patients with T2D, a CpG site (cg14496282) within PDGFA is found to be hypomethylated, leading to increased hepatic PDGFA expression and PDGF-AA secretion from insulin-resistant human hepatocytes (Abderrahmani et al., 2018). Additionally, PDGFA expression is positively correlated with hepatic fibrosis and NASH risk, and as such, the overexpression of PDGF-AA in mouse liver results in spontaneous liver fibrosis (Thieringer et al., 2008). Since the degree of hepatic fibrosis is an indicator of morbidity and mortality of liver diseases including NAFLD (Dulai et al., 2017), a DNA methylation study conducted with liver samples with NAFLD-related cirrhosis has identified genes enriched in ligand-activated nuclear receptor signaling pathways, involving farnesoid X receptor (FXR), liver X receptor (LXR), and retinoid X receptor (RXR), that play roles in fatty liver disease (Tanaka et al., 2017;Gerhard et al., 2018). Other DNA methylation studies (Ahrens et al., 2013;Nilsson et al., 2015;de Mello et al., 2017;Abderrahmani et al., 2018;Hotta et al., 2018) have also found a number of common differentially methylated sites, including the fibroblast growth factor receptor 2 (FGFR2) involved in liver fibrosis (Thieringer et al., 2008). In addition, hypomethylation of CpG sites within TGFβ1, Collagen 1A1, and PDGFα as well as hypermethylation of CpG sites within PPARα and PPARδ are frequently associated with the increased risk of fibrosis in NAFLD patients (Zeybel et al., 2015;Abderrahmani et al., 2018).
However, the question remains as to why hypomethylation is often found in metabolically dysfunctional livers as well as adipose tissues? Circulating folate levels are reduced in T2D patients compared with non-diabetic subjects (Nilsson et al., 2015). Since folate is a methyl donor in the methylation cycle, hypomethylation in the liver as well as in the pancreas of T2D patients  can be explained by a methyl donor supply consumption imbalance (Zhou et al., 2011;Nilsson et al., 2015). This is supported by a previous study using mice fed with a methyl-deficient diet (Pogribny et al., 2009). Lack of methyl donors accompanied by a loss of genomic cytosine methylation and a change in the expression of hepatic DNA methyltransferases, causes NAFLD and even NASH in mice, highlighting the role of hypomethylation in hepatic steatosis.
While hypomethylation is more common in hepatic tissues of metabolic disease, patients with NASH exhibit higher hepatic expression of DNMT1, which increases methylation and decreases expression of mitochondrial NADH dehydrogenase 6 (MT-ND6), leading to ultrastructural defects in mitochondrial morphology (Pirola et al., 2013). In addition, increased methylation at a CpG site (cg11669516) and reduced gene expression of insulin-like growth factor binding protein 2 (IGFPB2) are often found in mice and patients with NAFLD and NASH (Ahrens et al., 2013). Consistently, HFD feeding in young mice induces hypermethylation of Igfbp2 and reduces its expression prior to diet-induced obesity and hepatic steatosis development. This epigenetic inhibition of Igfbp2 becomes stable over time in adult mice, suggesting Igfbp2 methylation as a predictable risk indicator of liver disease development (Kammel et al., 2016). Together, genome-wide DNA methylation studies combined with ex vivo and in vitro analyses provide key epigenetic mechanisms underlying NAFLD development and progression in obese and T2D patients, by linking differential methylation states with the regulation of hepatic glucose and lipid metabolism, insulin resistance, and hepatic fibrosis.

Regulators of DNA methylation
In addition to changes in genomic DNA methylation, differential expression of DNA methylation regulators has been associated with the development of hepatic diseases, particularly the TET proteins involved in hydroxymethylation, the initial step in the reversal of DNA methylation. TET1 expression is reduced in both in vitro (HepG2 cells containing FA medium) and in vivo (HFD-fed mice) models of NAFLD . Loss of Tet1 (Tet −/− ) in mice further exacerbates HFD-induced NAFLD, indicated by increased intrahepatic TG levels. This study suggests that hepatic Tet1-mediated hydroxymethylation of the PPARα promoter enhances FAO and thereby prevents NAFLD development.
Furthermore, hepatic fibrosis is a critical pathological process that affects clinical management, as its advancement determines the therapeutic reversibility of NAFLD by leading to irreversible cirrhosis and even HCC (Stal, 2015;Thiele et al., 2017). Abnormal activation of the inflammatory transforming growth factor-beta (TGF-β) signaling pathway along with the transdifferentiation of hepatic stellate cells (HSCs) into proliferative, fibrogenic myofibroblasts, primarily drive hepatic fibrosis through the production of the extracellular matrix (ECM) (Tsuchida and Friedman, 2017). TET3 expression is increased in hepatocytes and HSCs of human fibrotic livers (Xu et al., 2020). In HSCs, TET3 mediates demethylation at specific CpG sites of genes involved in the TGF-β pathway, including TGFB1, to promote profibrotic gene expression and subsequent ECM production. In contrast, siRNA-mediated TET3 knockdown ameliorates liver fibrosis in mice, suggesting its crucial role in the pathological development and progression of hepatic fibrosis. This transdifferentiation of HSCs, largely regulated by a number of epigenetic processes including DNA methylation as described here, and post-translational modification of histones, is reviewed in detail elsewhere (Barcena-Varela et al., 2019).

Histone Methylation/Demethylation
The development of metabolic dysfunction in liver diseases accompanies global histone modifications including acetylation and methylation (Nie et al., 2017). Notably, alterations in global histone methylation patterns and expression of the regulators, such as HMTs and HDMs, during development and progression of NAFLD have been reported in a number of recent studies.

H3K9 methylation (EHMT2, JMJD1C, JMJD2B, PHF2)
The hepatic expression of G9a (EHMT2), a H3K9 HMT, is markedly reduced in genetically induced (i.e., db/db) and HFD-fed obese mice . The liver-specific loss of G9a is associated with a selective decrease in hepatic H3K9me2/me1 and an increase in serum cholesterol levels (Lu et al., 2019). Upon liver injury (e.g., lipopolysaccharide and acetaminophen overdose), G9a mutant mice exhibit severe liver phenotypes associated with increased immune cell infiltration, ROS production and cell death , suggesting an epigenetic protective role of G9a in the liver. On the other hand, the H3K9 demethylase, JMJD1C (KDM3), a candidate gene associated with T2D and plasma TG levels (Chasman et al., 2009;Teslovich et al., 2010;, regulates hepatic lipogenic gene expression (e.g., FAS, ACC, SREBF1) by demethylating the H3K9me2/me3 transcriptional repressor marks and leading to increased chromatin accessibility (Viscarra et al., 2020). Overexpression of JMJD1C in the liver increases DNL, whereas liver-specific deletion of Jmjd1c protects mice from diet-induced hepatic steatosis and insulin resistance. Similarly, the H3K9 diand tri-demethylase JMJD2B (KDM4B), involved in establishing the H3K9me activation mark, is upregulated in livers of dietinduced obese mice, resulting in increased hepatic PPARγ2 expression and induction of hepatic steatosis . Moreover, Phf2, another H3K9 HDM, specifically demethylates H3K9me2 on the promoter of carbohydrate-responsive elementbinding protein (ChREBP) (Bricambert et al., 2018), a major regulator of glycolytic and lipogenic genes (Ortega-Prieto and Postic, 2019). Interestingly, while glucose homeostasis remains preserved, liver-specific Phf2 over-expression results in hepatosteatosis, mediated by increased stearoyl-CoA desaturase (Scd1) expression and accumulation of monounsaturated fatty acids. Conversely, Phf2 silencing leads to liver fibrosis upon a high-fat, high-sucrose diet. With supporting human data, this study suggests Phf2 as a targetable epigenetic checkpoint to prevent NAFLD progression (Bricambert et al., 2018). Together, these studies demonstrate the critical and dynamic implications of H3K9 HMTs and HDMs in the development and progression of liver diseases.

H3K27 methylation (EZH1/EZH2)
Another HMT EZH2, which catalyzes trimethylation of H3K27 (H3K27me3) for transcriptional repression, also plays a key role in liver diseases. EZH2 expression is reduced in the liver of NAFLD rats and FA-treated HepG2 hepatocytes and is inversely correlated with lipid accumulation and inflammatory marker expression (Vella et al., 2013). The steatosis-related phenotypes are recapitulated when treated with 3-Deazaneplanocin A, an EZH2 inhibitor, suggesting a causal role of EZH2 in NAFLD development. In addition, EZH1, a homolog of EZH2, has H3K27 methyltransferase activity and can partially compensate for the loss of EZH2 (Ezhkova et al., 2011). Notably, when both Ezh1 and Ezh2 are deficient in the liver, the mutant mice develop liver fibrosis with increased fibrogenic gene expression (Fstl1, Fbn1 and Col1a1) (Grindheim et al., 2019).

H3K4 methylation (MLL4)
While PPARγ2 is a master transcriptional factor of adipogenesis in adipocytes, its expression is elevated in the fatty livers of obese animal models and NAFLD patients (Vidal-Puig et al., 1996;Westerbacka et al., 2007). Hepatic PPARγ2 stimulates the uptake and re-esterification of FA into lipid droplets by upregulating Cd36, Fabp4, Mag, Plin2, and Fsp27/Cidec, and thereby promoting steatosis . This upregulation of hepatic PPARγ2 can be epigenetically achieved by H3K4 methyltransferase MLL4 (KMT2D)  as well as through defective HDAC3, which normally associates with retinoic acid receptorrelated orphan receptor alpha (RORα) to repress PPARγ2 transcription . Together, these studies provide multimodal histone modulatory mechanisms of NAFLD via methyltransferase-and deacetylase-mediated transcriptional regulation of hepatic PPARγ2.

Histone Acetylation/Deacetylation
The modulation of hepatic histone acetylation, achieved through HATs and HDACs, contributes to the development of NAFLD. Among HATs, the transcriptional coactivator p300 is activated in mouse models of obesity and T2D, leading to post-translational hyperacetylation of ChREBP, involved in the transcriptional activation of lipogenic (e.g., Acc, Fas) and glycolytic (e.g., Pepck, G6Pase) genes (Bricambert et al., 2010). The activity of hepatic p300 is negatively regulated by salt-inducible kinase 2 (SIK2), thus, liver-specific SIK2 knockdown results in increased transcriptional activity of ChREBP via p300-mediated acetylation. In addition, hepatic p300 over-expression is sufficient to induce NAFLD and insulin resistance. Conversely, inhibition of HAT activity prevents NAFLD: the novel HAT inhibitor, tannic acid, binds to p300 and disrupts its occupancy on lipogenic genes (e.g., Fasn, Acly), leading to hypoacetylation of H3K9ac and H3K36ac (Chung et al., 2019). Together, these findings suggest a role for acetyltransferase p300 in NAFLD development.

Class I/II HDACs
Histone deacetylases also have key implications in the regulation and dysregulation of hepatic metabolism. As a significant metabolic hub, the functions of the liver in lipid, carbohydrate and amino acid metabolism as well as detoxification are regulated throughout the day by circadian clocks (Reinke and Asher, 2016). This rhythmic hepatic metabolism is orchestrated by epigenetic modulation, mainly through HDAC3, which is recruited by a key circadian clock component, Rev-ebrα (Feng et al., 2011). Upon hepatic Hdac3 depletion, mice develop hepatosteatosis with increased DNL, suggesting a critical role of histone acetylation-mediated circadian changes in the prevention of NAFLD. Additionally, HDAC8 has been identified as a commonly upregulated gene in dietary and genetic obesitypromoted HCC mouse models as well as in human HCC cells and tissues (Tian et al., 2015). HDAC8 promotes insulin resistance as well as cell proliferation, while its knockdown inhibits NAFLD-HCC tumorigenicity. Specifically, HDAC8, in association with the HMT EZH2, epigenetically regulates the Wnt pathway via decreased histone H4 acetylation and increased H3K27me3 methylation. This finding suggests an epigenetic mechanism involving HDAC8 in the progression of NAFLD-associated HCC.

Class III HDAC (SIRT1)
Sirtuin 1, a class III HDAC, acts on hepatic metabolic regulators in response to hormonal and nutritional signals. Sirt1 levels are reduced in a HFD-induced NAFLD rodent model (Deng et al., 2007) and are significantly lower in obese patients with severe steatosis (Wu et al., 2014;Mariani et al., 2015). Liver-specific deletion or knockdown of Sirt1 in mice leads to fatty liver disease even without a HFD challenge (Rodgers and Puigserver, 2007;Wang R.H. et al., 2010;Xu et al., 2010;Yin et al., 2014). Upon HFD or an alcoholic diet, these Sirt1 mutant mice develop severe liver injury and fibrosis (Purushotham et al., 2009;Yin et al., 2014;Ramirez et al., 2017). Moreover, these mice lacking hepatic Sirt1 exhibit elevated gluconeogenesis, leading to hyperglycemia and insulin resistance, which suggests that hepatic SIRT1 plays a role in not only protecting the liver from steatosis, but also in maintaining whole-body glucose metabolism.

Epigenetic Changes in the Liver in Dietary Intervention
Hepatic lipid accumulation is the primary characteristic and key contributor to NAFLD pathogenesis. Several dietary interventions, such as IF, CR, and KD, have shown protection against and improvement in hepatosteatosis with increased FAO, ketogenesis, and reduced lipogenesis in both obese or diabetic mice (Badman et al., 2009;Baumeier et (Horrillo et al., 2013;Marinho et al., 2019) and fibrosis (Horrillo et al., 2013;Johari et al., 2019), associated with the more severe NASH and cirrhosis condition of the liver (Figure 4). Taken together, the success of dietary interventions in halting and, in some cases, reversing NAFLD progression, makes it a promising alternative to current therapeutics. Thus, an improved understanding of both accompanying and causal epigenetic changes in the prevention and/or treatment of NAFLD will be necessary for determining novel molecular biomarkers and specific pharmaceutical targets for clinical translation.

DNA Methylation/Demethylation
Consistent with its anti-aging effect in humans and other model organisms (Fontana and Partridge, 2015;Redman et al., 2018), 40% CR can protect against the age-related increase in global hepatic DNA methylation (Miyamura et al., 1993) and reduce the epigenetic age of the mouse liver by approximately 1.7 years (Maegawa et al., 2017;Wang et al., 2017). In particular, a study comparing young and aged female mice showed that 40% CR delays the epigenetic aging of hepatic lipid metabolism by inducing hypermethylation and down-regulation of key enzymes involved in hepatic insulin resistance (e.g., Srebp1), lipid synthesis (e.g., Acly, Mel, Aacs2, Acac, Pklr, Gpam) and lipid elongation (e.g., Elov15, Elov16). These changes increase insulin sensitivity while reducing lipid content and chain length of TG-associated FA in the livers of old female mice (Hahn et al., 2017). More specifically, the reduction in the level of FA elongases and the subsequent shift in the hepatic TG pool from long-to medium-chain TG (Hahn et al., 2017) is associated with the prevention of diet-induced insulin resistance and liver disease (Ronis et al., 2013;Geng et al., 2016). Sterol regulatory elementbinding protein 1 (SREBP1) contributes to insulin resistance by interfering with the binding of FOXO1 to the insulin receptor substrate 2 (Irs2) promoter, thereby repressing Irs2 expression. CR has been shown to improve insulin resistance in mice through hypermethylation and downregulation of Srebp1 as well as through hypomethylation and upregulation of Irs2, the direct target of SREBP1 (Hahn et al., 2017). Moreover, in a separate study, CR-treated mice with a 25% reduction of body weight, exhibit decreased methylation and increased expression of hepatic Igfbp2 -an effect that is abolished by HFD re-feeding (Kammel et al., 2016). Since the reverse, increased methylation and decreased expression of IGFBP2, is found in both mice and humans with NAFLD and NASH (Ahrens et al., 2013), this CR-mediated change in Igfbp2 levels indicates metabolic improvements (Nam et al., 1997;Heald et al., 2006;Carter et al., 2014). CR also regulates the expression of DNA methylation modulators. In both young and old female mice, 40% CR increased Tet3 and Dnmt3a and decreased Tet2, Dnmt1 and Dnmt3b expression in the liver (Hahn et al., 2017). While altered hepatic expression of DNMT and TET enzymes in regulating DNA methylation is not well understood, this finding suggests that CR may modulate hepatic transcripts via dynamic regulation of DNA methylation machinery.

Histone Methylation/Demethylation
Dietary interventions, such as fasting, can modulate hepatic histone methylation and subsequent gene transcription. In particular, fasting-induced protein expression of JMJD3 (KDM6B), a H3K27me3 HDM Byun et al., 2020), stimulates the expression of β-oxidation (e.g., Fgf21, Cpt1a, Mcad)  and autophagy genes (e.g., Tfeb, Ulkl, Atgl) (Byun et al., 2020) in the liver, thereby promoting the removal of hepatic lipid stores via increased lipolysis and lipophagy. This histone modulation by hepatic JMJD3 in response to fasting is mediated by its association with two transcriptional activating complexes; JMJD3 in complex with PKA-phosphorylated SIRT1 (Ser434) and PPARα  or PKA phosphorylated JMJD3 (Thr1044) in complex with FGF21 (Byun et al., 2020). The fasting-induced JMJD3-SIRT1-PPARα complex additionally forms a feed-forward regulatory loop, which auto-induces the expression of its genes, including Fgf21, Jmjd3, Sirt1 and Ppara, to amplify the cellular responses under fasting conditions. Importantly, the downregulation of Jmjd3 and its associated factors (Sirt1, Fgf21, Ppara) in the mouse liver results in reduced hepatic β-oxidation and increased steatosis. These data suggest a critical role for histone methylation modulators in mediating the metabolic improvements associated with fasting-related dietary interventions against liver metabolic dysfunction.

Histone Acetylation/Deacetylation Class I/II HDACs
Hepatic histone acetylation in the context of dietary interventions mainly involve HDACs. Specifically, under fasting conditions, class IIa HDACs upregulate the hepatic gluconeogenic gene program. Fasting-induced glucagon-secretion in primary mouse hepatocytes promotes protein kinase A (PKA)-mediated phosphorylation and inactivation of SIK2, thereby allowing for the nuclear translocation of unphosphorylated class IIa HDACs . Nuclear HDAC4 and HDAC5 (class IIa HDACs) in association with HDAC3 (class I HDAC) deacetylate and activate FOXO transcription factors for the induction of gluconeogenic genes .
However, the necessity of SIRT1 in IF, TRF, and FMD has not been mechanistically tested yet. Interestingly, a study comparing transcriptomic changes by IF (i.e., ADF) and Sirt1 overexpression in mice concluded that despite functional similarities such as improved insulin sensitivity, Sirt1 gain-of-function does not mimic nor boost the metabolic effects of IF (Boutant et al., 2016). This suggests that Sirt1 may not be the only mediator of fasting-involved dietary interventions. Moreover, the current literature has conflicting results in regard to hepatic SIRT1 expression under fasting and nutritional interventions (i.e., CR). Although most studies show an upregulation of hepatic SIRT1 upon fasting or CR (Cohen et al., 2004;Rodgers et al., 2005;Deng et al., 2007;Rodgers and Puigserver, 2007;Hayashida et al., 2010;Ding et al., 2017), some studies have reported a decrease  or no change (Barger et al., 2008) in expression. In particular, it was reasoned that the decreased level and activity of Sirt1 in the livers of CR mice was due to low cellular NAD + /NADH levels (Hagopian et al., 2003;Lin et al., 2004;Chen et al., 2008;Kawakami et al., 2012). Furthermore, the liver-specific down-regulation of Sirt1 in these studies resulted in reduced hepatic fat synthesis and improved glucose homeostasis Erion et al., 2009) -an effect that contradicts other studies showing increased hepatic FA and cholesterol, impaired glucose tolerance, hepatic inflammation and steatosis (Rodgers and Puigserver, 2007;Purushotham et al., 2009) upon Sirt1 deficiency. Aside from variation in animal species, strain or age used in the studies, differences in hepatic SIRT1 expression may also be attributed to differences in length of fasting or extent of food restriction. Particularly, SIRT1 has been shown to be upregulated in the liver of mice upon longterm fasting (18−24 h), but not short-term fasting (6−8 h), and thus may play a role in the later stages of nutrient depletion (Liu et al., 2008). Some discrepancies might also originate from different experimental settings. Although metabolic phenotyping (e.g., indirect calorimetry) and tissue harvesting are commonly performed after an overnight fast post-feeding day, the metabolic benefits of dietary interventions can take place during fasting and/or refeeding. For example, both the elevation of energy expenditure via adipose thermogenesis by IF (Kim K.H. et al., 2017) and pancreatic β-cell regeneration by FMD (Cheng et al., 2017) occur during the refeeding period after fasting and nutritional interventions. Regardless, these conflicting studies do not undermine the importance of SIRT1 in the liver and the use of SIRT1 agonists or fasting-like mimetics as possible therapeutic options for patients with metabolic syndrome. Overall, the liverspecific epigenetic mechanisms described herein are important in understanding the pathophysiological development of NAFLD and its possible alleviation via dietary interventions.

EPIGENETIC REGULATION IN PANCREAS DURING METABOLIC DISEASE AND DIETARY INTERVENTION The Pancreas in Health and Metabolic Disease
The pancreas is a secretory organ with both exocrine (acinar) and endocrine (islet) function. Among the pancreatic islet cell types, the glucagon-secreting α-cells and the insulin-secreting β-cells are primarily involved in regulating the metabolic pathways of the fed and fasted states. Under fasting conditions, glucagon is released to increase blood glucose levels through the promotion of hepatic gluconeogenesis and glycogenolysis, whereas under feeding conditions, insulin is secreted to promote the uptake of glucose, amino acids and FA and to stimulate processes of glycogenesis, protein synthesis and DNL in insulin-sensitive metabolic tissues (Piciucchi et al., 2015;Roder et al., 2016;Weisbeck and Jansen, 2017). In the setting of insulin resistance, β-cells increase their insulin secretion (i.e., hyperinsulinemia) to maintain normal glucose levels (Johnson and Alejandro, 2008; Figure 5). However, when β-cells can no longer sustain the increased demand (i.e., hypoinsulinemia), glucose levels rise and initially present as impaired glucose tolerance (Kahn et al., 2014).
As β-cell dysfunction progresses, hyperglycaemia and diabetes arise (Weisbeck and Jansen, 2017). Eventually, the hyperactivity of β-cells and high levels of blood glucose and lipids contributing to glucotoxicity and lipotoxicity, can stimulate β-cell apoptosis and further propagate the pathogenesis of T2D . In addition to the genetic component, T2D is also largely influenced by environmental stressors, such as prolonged physical inactivity and an unhealthy diet (i.e., fatty foods high in dioxins) Ibrahim et al., 2020), which FIGURE 5 | Epigenetic changes of pancreas in metabolic disease and dietary intervention. In metabolic disease (i.e., T2D), the initial proliferation of pancreatic β-cells increase insulin secretion, but eventual β-cell failure leads to hypoinsulinemia and hyperglycemia. In dietary intervention, pancreatic β-cells are preserved through reduced inflammation and oxidative stress and present with increased function (i.e., GSIS) and regeneration. These physiological differences in the pancreas can be explained by epigenetic changes involving DNA methylation, histone methylation and histone and non-histone acetylation. result in changes in metabolic gene expression, mediated through epigenetics. Herein we discuss some of the epigenetic changes involved in both T2D development and alleviation via dietary interventions in the pancreas; the key organ regulating both plasma insulin and glucose levels.

DNA Methylation/Demethylation
As pancreatic islets are central to T2D development, the methylation status of the promoters of critical genes in islet function and development have been investigated. In islets of T2D patients, the increased DNA methylation of the insulin gene promoter (INS) at 4 CpG sites, correlates negatively with insulin mRNA levels and positively with glycated hemoglobin HbA(1c) levels, which reflect the cumulative blood glucose concentration (Yang et al., 2011). Similarly, in an in vitro model of a rat β-cell line (INS 832/13), 72-h of high glucose exposure (16.7 mmol/L), which recapitulates conditions of hyperglycemia in T2D, increases DNA methylation within the Ins promoter (Yang et al., 2011). Likewise, in another study, 48-h of high glucose exposure (19 mM) in isolated human islets results in differential methylation and expression of genes involved in islet function, including GLRA1, RASD1, VAC14, SLCO5A1, CHRNA5, and PDX1, and a decrease in glucose-stimulated insulin secretion (GSIS) (Hall et al., 2018). A 48-h exposure to palmitate (1 mM), a saturated fatty acid, also reduces GSIS in human islets while inducing methylation changes and differential expression of 290 genes, including the TCF7L2 and GLIS3, markers of T2D risk . These methylation changes in islets upon high glucose and lipid exposures may be analogous to the effects of glucotoxicity and lipotoxicity in pre-diabetic and diabetic conditions. Moreover, the expression of insulin promoter factor 1 (PDX1), a master transcriptional regulator of β-cell development and function, is reduced with increased methylation in pancreatic islets of T2D patients, which also corresponds to increased HbA1c levels (Yang et al., 2012). Similarly, the expression of the glucagon-like peptide-1 receptor (GLP1R), involved in enhanced GSIS upon GLP-1 peptide binding (Muller et al., 2019), is decreased in pancreatic islets of T2D patients and hyperglycemic rats (Xu et al., 2007;Shu et al., 2009;Taneera et al., 2012). In particular, the methylation level of a CpG site within the GLP1R promoter correlates negatively with its gene expression, but positively with BMI and HbA1c (Hall et al., 2013), suggesting that the obesogenic and diabetic conditions can impact the DNA methylation profile of GLP1R and possibly lead to changes in pancreatic β-cell insulin secretion (Hall et al., 2013).
GWAS have identified multiple loci associated with the T2D risk and accounting for ∼10% of heritable diabetes (Voight et al., 2010). These include loci related to impaired insulin secretion and insulin sensitivity (Ruchat et al., 2009). Additionally, among 40 SNPs previously associated with T2D in human pancreatic islets (McCarthy, 2010), 19 of them (48%) are involved in either introducing or removing a CpG site . Importantly, these CpG-SNP sites are differentially methylated and result in changes in gene expression, alternative splicing events and hormone secretion in human islets. Altered expression of genes associated with T2D risk include TCF7L2, HHEX, CDKN2A, SLC30A8, CDKAL1, ADCY5, and FS1 (Ruchat et al., 2009;McCarthy, 2010). Moreover, in a separate study, genomewide DNA methylation quantitative trait locus (mQTL) analysis in human pancreatic islets  has identified over 67,000 CpG-SNP pairs, with several mQTLs associated with differential expression of T2D-and insulin secretion-related genes (e.g., ADCY5, KCNJ11, INS, PDX1 and GRB10) in human islets. Together, these studies demonstrate that DNA methylation may provide a causal link between SNPs and pancreatic gene function, thereby contributing to T2D development and progression .
Men are at higher risk for T2D than women (Kautzky-Willer et al., 2016). In isolated islets, GSIS is greater in female versus male donors, independent of β-cell number (Sharp et al., 2011). Differences in the DNA methylome also exist in pancreatic islets between men and women; specifically, significant sex-specific differences have been observed in 61 X-chromosome genes and 18 autosomal genes, including NKAP, SPESP1 and APLN, which are expressed at lower levels in female islets. The methylation of NKAP and SPESP1 promoters decreases their expression, and the silencing of Nkap and Apln in clonal cells reduces GSIS (Sharp et al., 2011), suggesting that differential DNA methylation may explain sex differences in insulin secretion and T2D risk.

Histone Methylation/Demethylation
Compared to the implications of DNA methylation in pancreatic metabolic conditions, our understanding of histone modifications in the onset and progression of diseases is limited. Yet, as many studies focus on the regeneration of pancreatic β-cells for the treatment of T2D, histone methylation via transcriptional programming in pancreatic development and function have been studied. While insulin-secreting β-cells and glucagon-secreting α-cells have different physiological functions, a study revealed that human α, β, and exocrine cells share similar profiles of histone methylation of H3K4me3 and H3K27me3, suggesting an epigenomic plasticity of islet cells and their reprogrammable potentials to treat diabetes (Bramswig et al., 2013). For example, the activation of the S6K1 kinase promotes α to β-cell transition by activating β-cell genes while repressing α-cell genes through histone methylation of the activating H3K4me3 and repressing H3K27me3, respectively (Yi et al., 2018). Human pancreatic α-cells can also be reprogrammed into insulin-producing cells by PDX1, and when transplanted, can treat diabetic mice (Furuyama et al., 2019). Pancreatic β-cell proliferation and expansion are highly active early in life in humans and mice (Meier et al., 2008) and decays with maturation and aging (Teta et al., 2005). Interestingly, expression of EZH2, a HMT of H3K27me3 and a key regulator of cell differentiation and growth, also decreases with aging in mouse pancreatic β-cells, whereas its expression is increased with adaptive β-cell proliferation after streptozotocin-mediated β-cell destruction (Chen et al., 2009). EZH2 can regulate β-cell proliferation by epigenetically repressing the Ink4a/Anf locus, which encodes the cyclin-dependent kinase inhibitor p16INK4a and tumor suppressor p19Arf (Chen et al., 2009). Consequently, mice lacking Ezh2 in pancreatic β-cells exhibit mild diabetes, suggesting a role for pancreatic Ezh2 in β-cell function.

Histone Acetylation/Deacetylation
Altered pancreatic histone acetylation plays a part in the development of T2D. Specifically, the expression of HDAC7, a class IIa HDAC, is increased in pancreatic islets from patients with T2D (Daneshpajooh et al., 2017) and over-expression of Hdac7 in rat islets and clonal β-cells reduces insulin content and increases apoptosis, leading to impaired GSIS of β-cells (Daneshpajooh et al., 2017). Conversely, treatment with a class II HDAC inhibitor (MC1568) rescues the dysfunctional insulin release of Hdac7-over-expressed β-cells and human islets from T2D donors (Daneshpajooh et al., 2018). Interestingly, the overexpression of HDAC7 in islets from T2D patients is likely mediated by hypomethylation of HDAC7's CpG site , suggesting a cooperative epigenetic action.
The limited studies investigating the epigenetic effects of dietary interventions in the pancreas mainly pertain to histone acetylation changes. According to one study, 6 days of TRF (12-h fasting/feeding) in mice enhances GSIS in isolated islets without affecting body weight (Wortham et al., 2019) and is accompanied by histone acetylation of pancreatic islets during the re-feeding period. This epigenetic modulation takes place at sites occupied by the HDM LSD1 (KDM1A), which silences enhancers by removing mono-and dimethyl marks from H3K4 and is implicated in pancreatic endocrine cell development (Vinckier et al., 2020). The β-cell specific loss of Lsd1 results in histone hyperacetylation accompanied by insulin hypersecretion, indicating that the adaptive insulin secretory response to dietary interventions is regulated by the modulation of the epigenome in the pancreas (Rosen et al., 2018).
Another important epigenetic factor mediating the pancreatic response to dietary interventions is SIRT1. CR increases and activates SIRT1 in β-cells, thereby promoting pancreatic insulin secretion and ameliorating the T2D phenotype (Liang et al., 2009;Chen et al., 2010;Deng et al., 2010). Similar to the metabolic benefits conferred by dietary interventions (Chen et al., , 2013, Sirt1 over-expression in mice preserves glucose homeostasis through improvements in insulin secretion and glucose tolerance (Moynihan et al., 2005;Ramsey et al., 2008). Mechanistically, pancreatic SIRT1-mediated repression of Ucp2 increases cellular ATP to promote vesicular exocytosis and release of insulin from β-cells (Moynihan et al., 2005;Bordone et al., 2006;Chan and Kashemsant, 2006;Ramsey et al., 2008). Additionally, to accommodate CR-mediated increases in insulin secretion, β-cells proliferate and increase their mass, resulting in larger pancreas size (Chen et al., 2013) -an effect that is also seen with SIRT1 activation (Wu et al., 2019). CR also lowers pancreatic inflammation and oxidative stress Lanza-Jacoby et al., 2013;Harvey et al., 2014), which can otherwise lead to β-cell death and failure associated with T2D pathogenesis. This is likely mediated by SIRT1 activation since SIRT1 over-expression or resveratrol (CRmimetic) treatment post-translationally deacetylate both p65 for inhibition of the NF-kB inflammatory signaling pathway  and FOXO1 for defense against oxidative stress (Kitamura et al., 2005;Zhang T. et al., 2016). While limited in number, studies investigating epigenetic changes with dietary interventions in the pancreas highlight key factors and mechanisms involved in ameliorating the T2D condition, and warrant further research on this topic.

THE METABOLIC AND EPIGENETIC INTERPLAY BETWEEN GUT MICROBIOTA AND DIET
The gut is host to trillions of microorganisms, including bacteria, archaea, viruses, and eukaryotes. The combined genome (∼150 times larger than the human genome) and the function of these microorganisms make up the microbiome (Ley et al., 2006a;Qin et al., 2010). While the microbiome influences digestion, gut-hormone secretion, intestinal immunity and inflammation, it is also largely shaped by diet; specifically, diet alterations account for 57% of the changes in gut microbiota populations, whereas genetic mutations only account for 12% . Cumulating evidence places the gut microbiome and its metabolites at the origin of diet-induced metabolic dysregulation (Ley et al., 2006b), therefore, positive modulations of the gut microbiota by dietary interventions are of therapeutic interest.
Gut microbiota influences the host metabolism through various microbial-derived metabolites, which induce epigenetic alterations of key genes involved in the initiation and progression of diseases (Figure 6). Metabolites, such as folate, choline, vitamin B12, and betaine, can function as methyl donors and participate in DNA methylation processes via the generation of S-adenosylmethionine (SAM). In addition, the gut microbiota ferments complex carbohydrates from the diet to produce small organic acids, most of which are short-chain fatty acids (SCFA) (<95%), such as butyrate, propionate, acetate and lactate (den Besten et al., 2013). These SCFA, particularly butyrate and acetate, inhibit HDACs, leading to transcriptional activation via increased histone acetylation. Additional epigenetic roles of the gut microbiome are well discussed elsewhere (Sharma et al., 2019).
Importantly, fasting-feeding cycles can directly impact the gut microbiota (Thaiss et al., 2014). In particular, dietary intervention-mediated (i.e., IF, FMD, CR, KD) remodeling of gut microbial populations confers various health benefits Beli et al., 2018;Cignarella et al., 2018; FIGURE 6 | Epigenetic modulation by dietary intervention-induced ketogenesis and gut microbial metabolites. Dietary interventions stimulate ketone body production such as β-hydroxybutyrate, which can modulate gene expression through histone modification (bhb) and inhibition of histone deacetylases (HDACs). Dietary interventions also modulate the gut microbiota, through the release of short-chain fatty acids (SCFA) acetate and butyrate which inhibit HDACs, and folate which provides methyl donors for DNA methyltransferase (DNMT) activity. Fabbiano et al., 2018;Wang et al., 2018;Rangan et al., 2019;Ang et al., 2020;Liu et al., 2020). The causal and functional roles of these favorable microbes in dietary interventions are supported by microbiota transplantation and elimination (e.g., antibiotic treatment) studies. For example, IF-induced WAT browning and associated metabolic benefits are abolished in microbiotadepleted mice and subsequently restored with IF-microbiota transplantation . This is in part mediated by the expansion of the Lactobacillus bacteria population, resulting in increased levels of serum lactate (Vergnes and Reue, 2014) and acetate (Hernandez et al., 2019), which are taken up by the upregulated monocarboxylate transporter 1 (Mct1) in WAT (Iwanaga et al., 2009;Carriere et al., 2014;Li et al., 2017). Lactobacillus, populated upon IF, CR and FMD, is a probiotic bacterium with health-promoting properties against metabolic diseases (Sharma et al., 2019) and is involved in the production of folate and the fermentation of pyruvate to acetate and lactate. IF can also increase plasma butyrate levels through greater butyrateproducing Odoribacter (Liu et al., 2020). Notably, these health benefits of IF can be achieved through the direct administration of SCFA (Liu et al., 2020), suggesting that metabolites from a re-established microbiome mediate the health benefits of dietary interventions. Although studies of dietary intervention, such as CR and KD, have shown epigenetic-mediated mechanisms of improved gut stem cell homeostasis (Igarashi and Guarente, 2016;Cheng et al., 2019), the association with gut microbiota is still not fully understood and thus warrants further research.

KETONE BODIES AS EPIGENETIC REGULATORS IN DIETARY INTERVENTION
Many of the benefits related to fasting and nutritional interventions, such as IF, CR, FMD and KD, may stem from the activation of the ketogenic pathway and the increased production of ketone bodies (i.e., ketosis). This metabolic switch to ketosis, associated with improvements in lipid and glucose metabolism, contributes to a healthier metabolic state of the tissues discussed in this review. Dysregulation or insufficiency of ketogenesis, by contrast, can be associated with hepatic metabolic abnormalities and may contribute to NAFLD (Cotter et al., 2014;Mannisto et al., 2015;d'Avignon et al., 2018;Fletcher et al., 2019) and liver fibrosis (Puchalska et al., 2019).
Fasting stimulates hepatic ketogenesis to increase systemic ketone body levels. Mitochondrial 3-hydroxy-3-methylglutaryl-CoA synthase 2 (HMGCS2) is the rate-limiting enzyme of the ketogenic pathway and catalyzes the conversion of acetoacetyl-CoA to HMG-CoA -the first step in β-hydroxybutyrate (BHB) synthesis. HMGCS2 is a downstream target gene of FOXA2, a key transcription factor of hepatic lipid metabolism. Under feeding conditions, insulin/PI3K/Akt-mediated phosphorylation of FOXA2 reduces its transcriptional activity by nuclear exclusion (Wolfrum et al., 2004;von Meyenn et al., 2013;Newman and Verdin, 2014). Conversely, under fasting conditions, glucagon signaling inhibits SIK2 kinase and allows for p300mediated post-translational acetylation and activation of FOXA2, resulting in the induction of HMGCS2 transcription (von Meyenn et al., 2013;Newman and Verdin, 2014). In addition to the SIK2-p300-FOXA2 axis, SIRT3, the mitochondrial class III HDAC, directly increases the enzymatic activity of HMGCS2 under fasting conditions through post-translational deacetylation of lysine residues (Shimazu et al., 2010;Newman and Verdin, 2014). Mice lacking Sirt3 exhibit a reduction in fasting-induced BHB production, indicative of hypoketonemia, along with impaired IF-mediated neurological improvements .
Acetone, acetoacetate (AcAc) and BHB are the three ketone bodies produced by the liver, from which AcAc and BHB are transported primarily to skeletal muscle and the brain as carriers of additional energy, while acetone is mainly released through exhalation. Among the three ketone bodies, BHB is the most abundant in mammals. Notably, recent studies have demonstrated that ketone bodies play a pivotal role as direct or indirect signaling mediators of cellular and metabolic functions, including epigenetic gene regulation Verdin, 2014, 2017;Puchalska and Crawford, 2017). Similar to butyrate (Candido et al., 1978;Cousens et al., 1979;Louis and Flint, 2009), BHB inhibits HDACs, particularly HDAC 1, 3, and 4 (class I and IIa) (Shimazu et al., 2013; Figure 6). Both HDAC3 and HDAC4 are responsible for stimulating the expression of gluconeogenic genes , hence their inhibition by BHB would result in a reduction in plasma glucose levels, as seen in Hdac3-deficient mice (Knutson et al., 2008). In addition, ketosis stimulated by either BHB treatment, 40% CR or overnight fasting in cells and mice all result in histone hyperacetylation, particularly at H3K9 and H3K14, thereby promoting the expression of genes, such as Foxo3a -a core regulator of cellular homeostasis (i.e., cell cycle progression), stress response, and longevity induction (Shimazu et al., 2013). Since both butyrate treatment and HDAC inhibition improve hepatic steatosis (Endo et al., 2013;Zhou et al., 2017Zhou et al., , 2018 and glucose homeostasis (Gao et al., 2009;Xiao et al., 2011;Cho et al., 2018;Lee et al., 2018) in both mice and humans, enhancing BHB levels through dietary interventions may provide similar beneficial effects.
Furthermore, elevated BHB levels result in increased histone lysine β-hydroxybutyrylation (kbhb), a novel type of histone post-translational modification ; Figure 6). BHB produced from ketogenesis can be complexed with free molecules of Coenzyme A to form BHB-CoA, the donor for kbhb histone modifications (Sabari et al., 2017). A total of 44 histone kbhb sites have been determined in human embryonic kidney cells (HEK293) and mouse liver. These kbhb marks are found in the promoter sites of target genes and result in transcriptional activation. In particular, kbhb on H3K9 (H3K9bhb) is found in fasted liver . The H3K9bhb acylation mark targets and leads to the upregulation of genes involved in amino acid catabolism, redox balance and circadian rhythmmediating these specific processes in the switch from feeding to fasting state . It is thus tempting to speculate that the metabolic benefits by ketogenic dietary interventions, such as IF, CR, FMD and KD, involve histone kbhb-mediated epigenetic modulation.
Pharmaceutical therapeutics stimulating a state of ketosis similar to that with dietary interventions, such as sodiumglucose co-transporter 2 (SGLT2) inhibitors, are being clinically used in the treatment of NAFLD and T2D (Polidori et al., 2018). SGLT2 inhibitors have shown metabolic improvements in hyperglycemia, adiposity, oxidative stress and inflammation (Komiya et al., 2016;Scheen, 2019). While it has not been clearly understood, increased circulating ketone bodies by SGLT2 inhibitors has been considered one of the mechanisms of action mediating the metabolic benefits (Prattichizzo et al., 2018;Wojcik and Warden, 2019). In addition, in a recent study, loss of the G protein-coupled receptor 43 (Gpr43), activated specifically by AcAc in mice, abolishes IF-and KDmediated metabolic benefits, including those associated with lipid metabolism (i.e., body weight and fat mass reduction) (Miyamoto et al., 2019). This finding suggests that the improvements seen with dietary interventions are indeed mediated at least in part by ketone body metabolism. Further studies on the ketone bodymediated epigenetic changes in dietary interventions will provide additional mechanistic understanding.

CONCLUSION AND DISCUSSION OF FUTURE PERSPECTIVES
In this review, we summarized key tissue-specific epigenetic changes implicated in both metabolic diseases and dietary interventions, with a focus on adipose, liver and pancreas. While most of the epigenetic studies have been conducted separately in these metabolic tissues, it is important to recognize that the integrated tissue cross-talks can drive systemic changes in metabolic gene expression and function. For example, secretion of tissue-respective metabolites and hormones, such as adipokines (i.e., leptin, adiponectin), hepatokines (i.e., Fgf21) and pancreatic glucagon and insulin, which as we have discussed are all subject to epigenetic changes, are key mediators of tissue cross-talk and systemic homeostasis (Rosen, 2016;Stern et al., 2016). Therefore, the global metabolic changes seen in several tissues upon disease and dietary intervention make it difficult to not only characterize the adaptive or pathological role of these epigenetic events but, to also pinpoint the primary insult that triggers secondary, systemic aspects of these responses. Temporally identifying the whole-body, tissuespecific epigenetic changes throughout both disease progression and dietary intervention-mediated metabolic improvements will require further investigation. We further discuss some limitations, benefits and potentials to epigenetic modification by dietary intervention.

Sex Differences
One of the key limitations of dietary interventions arises from sex differences. Despite the successful outcomes of dietary interventions seen in humans and animal models (Di Francesco et al., 2018;de Cabo and Mattson, 2019), it is not clear whether and to what extent sex differences contribute to the impact of dietary interventions. For example, in the liver, males favor energy utilization by oxidizing FA, whereas females tend to prefer energy storage by converting FA into TG (Tramunt et al., 2020). Furthermore, females primarily store excess energy in subcutaneous fat, which in comparison to visceral fat, allows for greater and longer storage, prevents ectopic fat deposition in other tissues and resists the development of male-predominant metabolic diseases, such as diabetes and NAFLD (Tramunt et al., 2020). Thus, these sex differences in metabolism could lead to different outcomes when subject to dietary interventions. Indeed, unlike male mice showing reduced lipid accumulation, IF increases the hepatic lipid content of female mice (Piotrowska et al., 2016). This result can be further explained by the sex difference in the fasting response of the liver. Upon a shortterm fast (6 hr), male mice maintained steady-state metabolism with reductions in anabolic pathways, such as hepatic lipogenesis and gluconeogenesis, whereas females continued to use amino acids for the synthesis of hepatic TG (Della Torre et al., 2018). While it has been suggested that sex differences in the liver are established postnatally via testosterone-mediated DNA methylation (Reizel et al., 2015), this study with transcriptome and metabolomic analyses has demonstrated that the sexual differentiation of the liver exists when mice are born and is largely mediated by the sexually dimorphic hepatic estrogen receptor α (ERα) (Della Torre et al., 2018). Together, this evidence suggests that a comprehensive understanding of sex differences in metabolism is required for safe and efficacious utilization of fasting-involved dietary interventions in both males and females. Particularly, epigenetic changes and their regulatory roles in sexual differentiation in response to dietary interventions remain to be elucidated.

Age Differences
Aging is a key risk factor for metabolic disease development. Extension of lifespan and healthspan by dietary interventions has thus led to an interest in their application for the treatment of metabolic diseases in the elderly (Gensous et al., 2019). At the other extreme, pediatric obesity is also an emerging public health priority; dietary interventions are being considered for this population as well (Vidmar et al., 2019). However, as discussed in this review, most human and animal dietary intervention studies are conducted in young adult and middle-aged individuals, with very limited work in the elderly or pediatric population. Due to the stark differences in the metabolism and physiology among children, young adult and aged individuals, it is necessary to test the safety and efficacy of dietary interventions in these populations. For example, in contrast to young adult or middleaged individuals, mildly increased body weight and/or BMI in the elderly is often associated with a lower risk of mortality. This finding, known as the "obesity paradox, " suggests a protective role of body fat in the elderly against certain illnesses (i.e., T2D) (Hainer and Aldhoon-Hainerova, 2013). In one study, weight loss (>7.5% body weight) was strongly associated with a reduced survival outcome in T2D elderly patients with a mean age of 62 years (Doehner et al., 2012). Thus, dietary interventions resulting in body weight loss could be detrimental in the elderly (Thorpe and Ferraro, 2004;Locher et al., 2016). Moreover, a study has demonstrated that 4 weeks of TRF in juvenile mice (4-week-old) causes adverse effects including delayed puberty, fatty liver disease and an abnormal gut microbiota shift (Hu et al., 2019). Therefore, the overall risk-to-benefit ratio of dietary interventions in different age populations still remains uncertain and requires further research. Moreover, it would be interesting to see how epigenetic mechanisms govern the differential response to dietary interventions among different age populations with metabolic disease.

Benefits of Epigenetic Modification by Dietary Intervention
Despite the limitations of sex and age, the epigenetic modulation of dietary intervention is still a promising avenue. Due to the reversible nature of epigenetic modifications, the utilization of epigenetic therapies in the treatment of metabolic diseases is encouraging. Both in vitro and in vivo mechanistic studies of current epigenetic therapies, such as inhibitors of DNMT (e.g., 5-aza-2 -deoxycytidine) (Mann et al., 2007(Mann et al., , 2010, HDAC (e.g., Trichostatin A) ) and HAT (e.g., Tannic acid) (Chung et al., 2019) have shown promising preclinical results in targeting hepatic disease processes. However, the low specificity and broad range of outcomes associated with epigenometargeting agents could lead to side effects (Gius et al., 2004). Interest has also emerged in epigenetic dietary components including the DNMT/HAT inhibitor epigallocatechin-3-gallate (EGCG) found in green tea and the DNMT/HDAC inhibitor resveratrol found in peanuts, grapes and berries (Hardy and Tollefsbol, 2011). Resveratrol, commonly known as a SIRT1 activator, is used as a CR-mimetic in several studies, showing improvements in health and longevity (Baur et al., 2006). Although these natural compounds are associated with several benefits, they do not yet provide the same efficacy in achieving balanced and global effects as with dietary interventions against metabolic diseases. It will, however, be interesting to see whether combining these pharmacological or natural epigenetic modulators with dietary interventions could provide greater efficacy and reduced side effects -a perspective that can be explored in future studies. Overall, the numerous metabolic benefits associated with dietary interventions and their ability to reverse the disease state, makes them encouraging for clinical translation. Although current treatments are mainly associated with risk prevention and disease management, these dietary interventions, in combination with their epigenetic modulators, can serve to effectively improve the prognosis of metabolic diseases.

AUTHOR CONTRIBUTIONS
SA, NM, EM, and K-HK conceived and wrote the manuscript. SA and NM created the figures and table. All authors read and approved the final manuscript.