Genetic and epigenetic factors determining NAFLD risk.

BACKGROUND: Hepatic steatosis is a common chronic liver disease that can progress into more severe stages of NAFLD or promote the development of life-threatening secondary diseases for some of those affected. These include the liver itself (nonalcoholic steatohepatitis or NASH; fibrosis and cirrhosis, and hepatocellular carcinoma) or other organs such as the vessels and the heart (cardiovascular disease) or the islets of Langerhans (type 2 diabetes). In addition to elevated caloric intake and a sedentary lifestyle, genetic and epigenetic predisposition contribute to the development of NAFLD and the secondary diseases. SCOPE OF REVIEW: We present data from genome-wide association studies (GWAS) and functional studies in rodents which describe polymorphisms identified in genes relevant for the disease as well as changes caused by altered DNA methylation and gene regulation via specific miRNAs. The review also provides information on the current status of the use of genetic and epigenetic factors as risk markers. MAJOR
CONCLUSION: With our overview we provide an insight into the genetic and epigenetic landscape of NAFLD and argue about the applicability of currently defined risk scores for risk stratification and conclude that further efforts are needed to make the scores more usable and meaningful.

Introduction

Non-alcoholic fatty liver disease (NAFLD) represents a disease spectrum ranging from simple benign steatosis, which can develop further, to steatohepatitis characterized by inflammation and fibrosis. Disease progression eventually leads to cirrhosis or liver cell carcinoma. Epidemiological and clinical studies imply that NAFLD is strongly associated with other metabolic disorders such as obesity [1], insulin resistance [2], and type 2 diabetes (T2D) [3]. In fact, NAFLD is diagnosed in >70% of T2D patients [4]. Furthermore, NAFLD increases the risk of cardiovascular disease, including heart failure [5].

Genetic and environmental factors such as nutrition and physical activity interact and modulate individual risk of NAFLD development and the severity of progression. Several genetic variants associated with NAFLD and/or NASH were identified by genome-wide association studies (GWAS) and candidate gene approaches. Among these, a few genetic variants were proposed in a genetic risk score (GRS) for predicting individual risks [6] and might contribute to the early diagnosis and development of precision treatments. Genetics also helps to understand the NAFLD prevalence in different ethnic groups. For example, as shown by a meta-analysis, the prevalence in Hispanics is particularly high, whereas that in Blacks is the lowest [7]. This can partly be explained by the different occurrence of single nucleotide polymorphisms (SNPs) in risk variants in these cohorts.

In addition to genetic predisposition, epigenetic changes that occur in response to environmental factors such as nutrition contribute to disease risk. Epigenetic changes include modifications that alter gene expression and ultimately the phenotype. In various mouse [[8], [9], [10]] and human studies [[11], [12], [13]], epigenetic modifications have been associated with pathomechanisms of NAFLD. These include altered DNA methylation patterns [11], expression of miRNAs [14], and histone modifications [15]. Epigenetic alterations can be transferred to the next generation and thus transgenerationally modify disease risk of offspring [16]. Development of metabolic diseases later in life partly depends on the metabolic phenotype of the mother and the intrauterine environment [[17], [18], [19]]. Establishing an epigenetic profile that reflects disease status could improve individual NAFLD risk assessment. As epigenetic alterations are not only inheritable but also reversible, this could offer new approaches for individualized prevention and therapy.

Our review presents major genetic and epigenetic alterations on the level of DNA methylation and miRNA changes that have been observed in relation to the risk of NAFLD and provides information on their potential for risk assessment.

Genetic contribution to NAFLD

Frequent genetic variants that mediate NAFLD risk

A GWAS by Romeo et al. [20] was the first to identify the most prominent fatty liver gene patatin-like phospholipase domain-containing 3 (PNPLA3), also known as adiponutrin. The association of the SNP rs738409 (C > G) in PNPLA3 with NAFLD was replicated in several subsequent GWAS [21,22]. PNPLA3, which exhibits a 46% sequence homology with the lipase PNPLA2, also designated adipose triglyceride lipase (ATGL), acts as triglyceride lipase, exhibits acylglycerol transacylase activities, and appears to play a role in lipid remodeling of hepatic triglycerides [23,24]. The SNP rs738409 causes the missense sequence variation I148M that disrupts the enzyme's phospholipase activity, thereby interfering with lipid catabolism (Figure 1). PNPLA3 is associated with increased hepatic fat content, elevated liver enzymes, hepatic fibrosis, and cirrhosis [20,25,26]. Lipidomic analyses of five lipid fractions from liver tissue samples of control and PNPLA3 carriers revealed unaltered levels of palmitic acid, oleic acid, and linoleic acid in the triglyceride fraction. However, the concentration of trans-palmitoleic acid was increased and that of stearic, arachidic, and lignoceric acid saturated fatty acid was decreased in this fraction. The fatty acid composition of the other lipid fractions (phospholipids, diacylglycerols, and cholesteryl esters) was not affected [27]. Similar effects were detected in mice overexpressing the PNPLA3 variant in the liver [28]. The effect on hepatic fat accumulation was strongest in Hispanics, who also displayed the highest allele frequency (0.49) compared to that in European Americans (0.23) and African Americans (0.17) [20]. Therefore, the SNP rs738409 may partly explain the variable prevalence of hepatic steatosis among different ethnic groups in the US population [7], with individuals of Hispanic descent displaying more NAFLD (45%) than those of European descent (33%), who have a higher prevalence of NAFLD than those of African descent (24%) [29].

Figure 1

Frequent gene variants associated with NAFL and/or NASH and their major effects. CPT1, carnitine palmitoyl transferase-1; LPI, lysophosphatidylinositol; PI, phosphatidylinositol.

Most interesting, the effect of PNPLA3 was independent of insulin resistance and lipid concentration, because PNPLA3 allele carriers had significantly higher levels of liver fat but no difference in glucose tolerance, C-reactive protein, lipids, and liver enzymes compared to controls [20,30]. The PNPLA3 effect is modulated by dietary conditions in mice [11,31] and humans, as its expression is regulated by the transcription factors sterol regulatory binding protein 1c (SREBP1c) [32,33] and carbohydrate response element binding protein (ChREBP) [34]. High carbohydrate levels cause transcriptional upregulation of Pnpla3 and indirect inhibition of protein degradation [33], whereas Pnpla3 expression was reduced in mice by fasting [33,35]. In a human study, Davis et al. showed that in Hispanic children carrying the risk allele, the effect on hepatic fat storage was amplified by carbohydrate-rich diets. The authors proposed that specific dietary interventions based on genetic predisposition may lead to more effective therapeutic outcomes for fatty liver [36].

Although the importance of PNPLA3 for NAFLD based on GWAS has been repeatedly confirmed, the underlying pathogenic mechanism is still not fully understood. As PNPLA3 acts as a lipase intracellularly in hepatocytes and hepatic stellate cells (Table 1) [37], the lack of triglyceride hydrolase activity was postulated to cause hepatic triglyceride accumulation. However, the deletion of Pnpla3 in mice did not cause hepatic steatosis [38]. Thus, not a loss of function of PNPLA3 but probably a change in its function could lead to elevated hepatic fat storage induced by PNLPA3. Two independent studies demonstrated that overexpression of PNPLA3 in the liver of mice induced hepatic steatosis [31,39], whereas mice overexpressing the wild-type PNPLA3 had normal hepatic triglyceride levels [40]. Basuray et al. showed in a series of in vivo experiments that the degradation of the PNPLA3 variant was prevented by inhibiting autophagy or proteasomal degradation. As a result, the protein accumulated in the lipid droplets, which limited their mobilization and promoted hepatic steatosis [41]. In another set of in vitro and in vivo studies, Wang et al. [42] demonstrated that PNPLA3 recruitment to lipid droplets depends on cofactor comparative gene identification-58 (CGI-58), a cofactor for ATGL. Co-expression of ATGL and PNPLA3 (either wild-type or I148M) in hepatoma cells prevented the depletion of lipid droplets, suggesting the inhibition of ATGL-mediated lipid hydrolysis by PNPLA3 (Table 1). The authors hypothesized that PNPLA3 sequestrates CGI-58 to the surface of lipid droplets, which limits its availability for activation of ATGL [42]. In addition, PNPLA3 appears to interfere with retinol production and release of hepatic stellate cells by affecting retinyl-palmitate lipase activity (Figure 1 and Table 1), thereby promoting fibrosis development [43]. The lack of enzymatic activity leads to a reduced secretion of matrix-modulating enzymes, resulting in the deposition of extracellular matrix [44]. This potential mechanism is supported by data on NAFLD patients, who have reduced circulating retinol concentrations and concurrent intrahepatic retinol increases [43].

Table 1

Expression patterns (mRNA and/or protein) and function of genes associated with NAFLD risk. The information is based on the Human Protein Atlas (http://www.proteinatlas.org [145,146]; shown in italics) or the indicated references.

Gene	Tissue expression [145,146]	Liver cell type	Function
PNPLA3	Liver and kidney	Hepatocytes [43]	Lipid droplet remodeling [40,42]
		Stellate cells [43]	Modulation of retinol production and release [43,147]
GCKR	Liver, smooth muscle, and stomach	Hepatocytes	Increasing glycolytic flux [51], regulation of de novo lipogenesis [48,148]
TM6SF2	RNA enriched in intestine and liver [57]	Hepatocytes [60]	VLDL secretion [53,57,58]
HSD17B13	Ubiquitous (RNA enriched in liver) and liver [64]	Hepatocytes [149]	Lipid droplet remodeling [63,64,149], involved in retinol metabolism [64]
MBOAT7	Ubiquitous and RNA low tissue specificity [69]	Hepatocytes, hepatic sinusoidal cells, and stellate cells [69]	Remodeling of phosphatidylinositol [69,150]
PPP1R3B	Ubiquitous and low tissue specificity	Hepatocytes [79]	Hepatic glycogen storage [77,79]
IRGM	Liver [82]	Hepatocytes [82]	Modulation of lipophagy [39] via interaction with lipase ATGL [82]
LPIN1	Ubiquitous, adipose tissue, and liver [84,151]	Hepatocytes [152]	Regulation of fatty acid metabolism [153,154]

ATGL, adipose triglyceride lipase; VLDL, very low-density lipoproteins.

In contrast to Pnpla3 knockout mice, which did not store ectopic fat in the liver, a shRNA-mediated reduction in Pnpla3 expression in mice after development of high-fructose diet-induced steatosis resulted in decreased hepatic triglyceride levels, supporting the assumption that PNPLA3 accumulation per se causes steatosis [41]. Overall, different mouse studies showed that PNPLA3-associated hepatic steatosis requires the presence of the catalytically inactive protein and not simply the absence PNPLA3 activity [31,41]. The action of PNPLA3 is not restricted to the liver. It was recently shown that PNPLA3 is also expressed in adipose tissue where the protein itself is more abundant than in the liver. PNPLA3 carriers exhibited increased levels of PUFA triglycerides than controls. However, adipocyte lipolysis was not altered [45].

A second robust NAFLD gene is glucokinase regulatory protein (GCKR), which is involved in the control of glucose metabolism by regulating hepatic glucose uptake and hepatic glucokinase activity. The intronic SNP rs780094 (G > A), which is associated with hepatic lipid content, has been identified in various GWAS [22,46]. A meta-analysis by Zain et al. confirmed the association of rs780094 with increased NAFLD risk and demonstrated different allele frequencies for Africans (0.13), Europeans (0.41), and East Asians (0.48) [47]. In addition, association of rs780094 with other metabolic traits such as decreased fasting blood glucose [48] and decreased risk of T2D [49,50] were reported. The variant rs780094 is in strong linkage disequilibrium with the non-synonymous SNP rs1260326 (C > T; P446L). GCKR is supposed to exhibit a reduced ability to inhibit GCK (glucokinase) and thereby increasing glycolytic flux and glucose uptake by the liver (Figure 1 and Table 1). GCK serves as a metabolic switch that controls glucose metabolism. In the postprandial state when more glucose is taken up by the liver, GCK phosphorylates glucose to glucose-6-phosphate, which is converted into glycogen for storage. However, excess dietary glucose that cannot be stored as glycogen is converted into fat by de novo lipogenesis using acetyl-CoA that is generated from glycolysis-driven pyruvate and NADPH [51]. Fructose-6-phosphate (F6P) enhances the GCKR-mediated inhibition and this effect was shown to be significantly attenuated in the GCKR variant, which indirectly enhances GCK activity and glycolysis. Consequently, the production of metabolites such as malonyl-CoA increases and elevates triglyceride storage in the liver theoretically via two mechanisms [48]. On the one hand, malonyl-CoA serves as a substrate for de novo lipogenesis; on the other hand, it inhibits the import of fatty acids into the mitochondria by blocking carnitine palmitoyl transferase-1 (CPT1), thus disrupting fatty acid oxidation (Figure 1).

Santoro et al. explored the combined effect of the two genetic risk variants PNPLA3 and GCKR. In a study cohort of 455 obese children and adolescents, an additive effect of both variants on liver fat content was reported. Furthermore, this additive effect explained approximately 32% of the liver fat variance in Caucasians, 39% in African Americans, and 15% in Hispanics [52].

In a human exome-wide association study [53], the rs58542926 (G > A; E167K) variant transmembrane 6 superfamily member 2 (TM6SF2) was associated with increased hepatic triglyceride content and higher risk of advanced fibrosis in NAFLD patients [[53], [54], [55]], but paradoxically associated with a lower concentration of hepatic-derived triglyceride-rich lipoproteins [54,56]. Therefore, despite the increased risk of NAFLD, carriers of TM6SF2 have a lower risk of cardiovascular disease [54]. TM6SF2 is the causal gene variant that explains the association of the NCAN locus with hepatic triglyceride content and lipid levels already described in a GWAS by Speliotes et al. [22]. TM6SF2 is a transmembrane protein located in the endoplasmic reticulum (ER) and the ER-Golgi intermediate compartment. In vitro silencing of TM6FS2 in hepatocytes reduced the secretion of VLDL and thus promoted the retention of triglycerides [57] (Figure 1 and Table 1). In vivo knockdown of Tm6sf2 or transient overexpression of human TM6SF2 in mice altered serum lipids and hepatic fat content [56], which again illustrates that TM6SF2 is relevant for aberrant hepatic lipid storage. To ascertain the effect of TM6SF2 on hepatic lipids and subsequent NAFLD risk, Luukkonen et al. analyzed hepatic lipid profiles between carriers and non-carriers of the TM6SF2 gene variant [58]. TM6SF2 carriers were characterized by impaired incorporation of polyunsaturated fatty acids into hepatic triglycerides, phospholipids, and cholesterol esters. The deficiency of polyunsaturated phosphatidylcholines that are required for VLDL assembly was described to cause increased degradation of intrahepatic VLDL, thereby reducing their secretion [59].

The previously described observations were verified using an in vitro approach. Prill et al. generated and characterized a 3D spheroid model from primary human hepatocytes obtained from individual donors, either wild-type or heterozygous for the TM6SF2 allele, and demonstrated that the genetic variant induced elevated fat storage in hepatocytes by reducing secretion of APOB particles. In addition, mRNA expression of genes related to cholesterol biosynthesis (FDPS, HMGCS1, FDFT1, DHCR7, and SC5D), de novo lipogenesis (FASN and ACSS2), phospholipid dephosphorylation (PLPP3), and gluconeogenesis (FBP1) was higher in hepatic TM6SF2 spheroids than in those of wild-type donors [60].

A further gene variant that describes a link between hepatic phospholipids and the risk of advanced NAFLD is the splicing variant rs72613567 (T > TA) with an adenine insertion in HSD17B13 that encodes for the hepatic lipid droplet protein hydroxysteroid 17-beta dehydrogenase 13. The HSD17B13 rs72613567 variant leads to the synthesis of a truncated loss-of-function enzyme [61] that protects against advanced NAFLD, NASH, ballooning degeneration, lobular inflammation, and fibrosis [62] (Figure 1). Surprisingly, the gene variant does not influence the development of steatosis, as several studies showed no difference in the degree of steatosis between rs72613567 carriers and noncarriers; however, it decreases the risk of chronic liver damage in NAFLD patients [[61], [62], [63]]. Interestingly, the loss-of-function HSD17B13 rs72613567 allele is sufficient to mitigate the risk of liver injury in PNPLA3 allele carriers who are genetically predisposed to NAFLD. This effect was associated with a decrease in PNPLA3 mRNA in an allele dose-dependent manner [61]. Ma et al. investigated two other SNPs of HSD17B13, rs683413 (T > G/C), which links with the splice variant rs72613567, and rs62305723 (G > A; P260S), which encodes an HSD17B13 variant. Both were associated with increased steatosis but decreased ballooning and inflammation [64]. Discrepant results exist regarding the protein levels of HSD17B13 in liver samples from NAFLD and controls. Ma et al. detected a higher expression of HSD17B13 in NASH compared to controls, but with no differences between wild-type, rs683413, or rs72613567 allele carriers [64]. In contrast, Pirola et al. reported lower or absent HSD17B13 levels in NAFLD patients hetero- or homozygous for rs72613567 in an allele-dependent manner [62]. Overexpression or deletion of HSD17B13 in HepG2 cells did not affect lipid content, demonstrating an indirect function [64]. Luukkonen et al. suggested that the splicing variant rs72613567 might protect from progressive liver disease by increasing the synthesis and/or decreasing the degradation of phospholipids. Lipidomics revealed a general increase in hepatic phospholipids in rs72613567 carriers, and transcriptomics showed a downregulation of inflammation-related genes. These effects were accompanied by lower plasma concentrations of the proinflammatory cytokine interleukin-6 [63].

The family of 17β-hydroxysteroid dehydrogenases (HSD17Bs) consists of 15 members that are mainly involved in sex hormone metabolism. Some members also play key roles in cholesterol and fatty acid metabolism. The substrate and enzymatic function of HSD17B13 is not entirely known [65]. However, there is indication that similar to PNPLA3, it also acts as a retinyl-palmitate lipase, and the loss-of-function variant HSD17B13 rs72613567 affects retinol metabolism (Figure 1). Ma et al. discovered a retinol-dehydrogenase enzymatic activity of HSD17B13 that requires its binding to lipid droplets (Table 1). The enzymatic activity to catalyze the oxidation of retinol is reduced or absent in gene variants mediating anti-fibrotic/anti-inflammatory effects [64]. Based on several studies, a number of proteins involved in retinol metabolism, including retinol-binding protein 4 (RBP4) and aldehyde dehydrogenase 1A1 (ALDH1A1), have also been implicated in metabolic diseases including NAFLD and NASH [66].

A recent study combining animal models and human data challenged the dogma that lobular inflammation precedes hepatic fibrosis by mechanistically linking membrane-bound O-acyltransferase domain-containing 7 (MBOAT7) to lipid-driven inflammation-independent development of fibrosis [67]. A genetic variant rs641738 (C > T) located near two genes encoding MBOAT7 and the transmembrane channel-like 4 (TMC4) was first reported to increase the risk of alcoholic cirrhosis [68], but a subsequent candidate gene study also linked rs641738 to NAFLD and disease progression [69]. In particular, eQTL analysis and the characterization of Mboat7 and Tmc4 knockout mice demonstrated that Mboat7 loss of function mediates the progression of NAFLD [70]. Interestingly, MBOAT7 expression is reduced in livers of obese mice and humans, independent of the rs641738 risk allele [70]. In the Liver Biopsy Cross-Sectional Cohort of Individuals of European descent, the MBOAT7 rs641738 SNP was associated with the spectrum of liver damage related to NAFLD (Figure 1), including a higher degree of steatosis, more severe necroinflammation, and more advanced fibrosis. MBOAT7 belongs to the family of lysophospholipid acyltransferases with a specificity for arachidonoyl-CoA (Table1) [71]. Multiple phosphatidylinositol species showed differences in the plasma of MBOAT7 rs641738 allele carriers who also have lower hepatic MBOAT7 protein levels. Other lipid classes such as triglycerides, ceramides, or phospholipids were not affected by the genotype [69]. This again highlights the role of phospholipid remodeling in NAFLD pathogenesis. Mice with an hepatocyte-specific knockout of Mboat7 had increased hepatic fibrosis on a NASH-inducing diet without induction of inflammation as shown by a decrease in monocytes and unchanged levels of inflammatory mediators [67]. Thus, fibrosis development might occur independent of the inflammatory state. Similarly, in those with a BMI ≤35 in a cross-sectional NAFLD liver biopsy cohort, the MBOAT7 rs641738 allele was significantly associated with the presence of fibrosis in the absence of lobular inflammation. Helsley et al. also detected a general decrease in M2 macrophages in Mboat7 knockout mice. However, the increase in M1 macrophages and higher hepatic expression of the pro-inflammatory markers Tnfα and Il1b are an indication of inflammation in this model of Mboat7 deletion mediated by antisense nucleotides that was not restricted to the liver but also affected the fat tissue [70]. In accordance with Mancina et al. [69], remodeling of the lipidomic pattern was detected that was similar between humans carrying the rs641738 risk genotype and mice with hepatocyte-specific deletion of Mboat7 [67]. In particular, phosphatidylinositol (PI) species with arachidonoyl side chains were reduced and PI species with monounsaturated fatty acids were increased. Lysophosphatidylinositol (LPI) species serving as MBOAT7 substrates were also elevated (Figure 1), which may cause the development of fibrosis, as plasma LPI levels are elevated in fibrosis patients. Furthermore, treating Mboat7 loss-of-function mice with LPI resulted in the induction of pro-fibrotic genes. Based on these findings, the authors concluded that healthy subjects are protected from obesity-linked progression of NAFLD by the MBOAT7-mediated acylation of LPI lipids [70]. Thus, the risk genotype MBOAT7 rs641738 appears to mediate its pro-fibrotic effect in patients via LPI remodeling without significant induction of liver inflammation.

Polymorphisms of two fatty liver genes (TM6SF2 and PNPLA3) have been shown to associate with one specific diabetes cluster. Based on several pathophysiological parameters, patients with adult-onset diabetes are allocated to five clusters [72,73]. Patients in the severe autoimmune diabetes (SAID) cluster exhibit a T1D/LADA-like phenotype, severe insulin-deficient diabetes (SIDD) and severe insulin-resistant (SIRD) patients display the most severe T2D forms with a high risk of developing secondary complications [73], whereas mild age-related diabetes (MARD) and mild obese diabetes (MOD) patients exhibited only minor metabolic abnormalities. Among these groups, SIRD patients exhibit the highest hepatic fat content and the lowest whole-body insulin sensitivity and this cluster showed a significant association with the rs10401969 (T > C) variant of TM6SF2 [73]. Furthermore, patients in the SIRD cluster were shown to more frequently carry the risk variants rs738409 (CG and GG) of PNPLA3 and exhibit higher circulating free fatty acid concentrations and a more pronounced adipose tissue insulin resistance than non-carriers [74].

Rare genetic determinants of NAFLD and NASH

In addition to the most reliable fatty liver genes PNPLA3, TM6SF2, HSD17B13, and MBOAT7, several other genetic determinants of NAFLD and NASH have been identified that appear to be specific for only one ethnic population or have been confirmed by few studies, presumably due to their small effect size. In this section, we will introduce some of these candidates.

The contribution of protein phosphatase 1 regulatory subunit 3B (PPP1R3B) to the genetic risk of NAFLD is rather controversial. The noncoding SNP rs4240624 (G > A/C) near PPP1R3B that was associated with hepatic steatosis diagnosed by computed tomography but not histologically defined NAFLD was identified in the same GWAS that linked PNPLA3 and TM6SF2 (the NCAN locus) to NAFLD [22]. Interestingly, the risk allele was also associated with altered serum lipids, increased HDL and LDL cholesterol, and decreased fasting glucose. Histological assessment of hepatic steatosis in bariatric patients could not confirm the initial association of PPP1R3B rs4240624 with steatosis [75]. Speliotes et al. already questioned this association, which could instead be linked to hepatic glycogen storage (Table 1) than increased hepatic fat content [22]. Accordingly, PPP1R3B promotes hepatic glycogen storage by dephosphorylation and activation of glycogen synthase and decreases glycogen breakdown by inactivation of glycogen phosphorylase, which is the rate-limiting enzyme in glycogenolysis [76]. Thus, Stender et al. [77] and Seidelin et al. [78] attempted to clarify if the higher computed tomography attenuation associated with PPP1R3B rs4240624 is caused by differences in hepatic glycogen or hepatic triglyceride content. The minor allele was shown to promote hepatic glycogen synthesis in the postprandial state [78] and was linked to a mild form of liver glycogenosis leading to hepatic injury [77]. The SNP rs4841132 (A > G) is in complete linkage disequilibrium with rs4240624 and is associated with increased hepatic X-ray attenuation and serum liver enzyme alanine aminotransaminase (ALT) but not with hepatic triglycerides [22]. Interestingly, rs4841132 was associated with an increased hepatic expression of PPP1R3B [77]. Experiments in mice confirmed this relationship, as Ppp1r3b knockout mice displayed lower hepatic glycogen [77,79] and liver-specific overexpression of PPP1R3B and increased glycogen content [77]; however, hepatic triglycerides were not altered in either model.

In a genome-wide approach conducted in obese children and adolescents from a predominantly Han Chinese population, Lin et al. detected a variant in the immunity-related GTPase M (IRGM) gene (rs10065172, C > T) in addition to PNPLA3, GCKR, and TM6SF2 polymorphisms associated with NAFLD detected by ultrasonography [39]. Similar results were reported in an Italian study showing that the risk allele of IRGM rs10065172 was significantly associated with elevated plasma aminotransferase levels and mild to severe steatosis in children. However, in adults, no link with the IRGM risk allele was observed in liver disease progression diagnosed by histological evaluation of ballooning, inflammation, and fibrosis [80]. The role of IRGM in the development of hepatosteatosis became questionable when another variant of IRGM rs13361189 (C > T), which is in linkage disequilibrium with the SNP rs1006517, was analyzed in Framingham Heart Study participants who underwent computed tomography scans. No association of rs13361189 with NAFLD was established [81]. Nevertheless, there were indications that IRGM plays a role in the regulation of autophagy [39] and hepatic lipid storage. We have discovered that the mouse orthologues of IRGM are the immunity-related GTPases 2 and 4 (Ifgga2 and Ifgga4), which are located in close proximity on mouse chromosome 18. Their expression was markedly reduced in mice with NAFLD, and accordingly, suppression of their expression in hepatocytes or mouse liver increased fat accumulation, whereas the overexpression of Ifgga2 in hepatocytes decreased fat storage. IFGGA2 appears to induce lipophagy via interacting with the lipase ATGL and increasing the association of the autophagy protein LC3B with lipid droplets. Interestingly, we also showed that the human IRGM protein interacts with ATGL (Table 1) and that the expression of IRGM was significantly reduced in livers of NAFLD patients [82].

Similar to Ifgga2 and Ifgga4, Lpin1 (Lipin 1) mRNA levels are reduced in a rat model of NAFLD [83]. As LPIN1 expression in the liver and adipose tissue is inversely correlated with adiposity and positively associated with insulin resistance, Valenti et al. evaluated the association of an LPIN1 SNP (rs13412852, C > T) with the susceptibility to and progression of NAFLD [84]. The authors conducted a study of Italian children and adults and tested the rs13412852 SNP that was earlier linked to lower body weight [85]. Only in children but not in adults was the homozygous rs13412852-T allele associated with protection from NAFLD. However, pediatric and adult patients homozygous for the minor allele of LPIN1 exhibited a significantly reduced risk of histological fibrosis and less severe liver damage [84]. Lipin 1 plays a major role in adipose tissue and influences its development and function. Deletion of Lpin1 in mice results in a marked reduction in adipose tissue depots (lipodystrophic phenotype) and insulin resistance, whereas the adipocyte-specific overexpression of Lpin1 causes diet-induced obesity [86].

Genetic risk scores for predicting steatohepatitis

As approximately 10–30% of patients with a simple hepatosteatosis develop NASH [87,88] that is associated with liver-related morbidity and mortality, a specific focus has been placed on developing a genetic risk score that allows the prediction or early diagnosis when fibrosis is still at an early stage.

Nobili et al. developed a genetic risk score that in combination with clinical risk markers such as aminotransferases significantly predicts NASH in obese children and adolescents [89]. In 152 study participants with biopsy-proven NAFLD and increased liver enzyme, polymorphisms of PNPLA3 rs738409 (C > G), SOD2 rs4880 (C > T), KLF6 rs3750861 (G > A), and LPIN1 rs13412852 (C > T) were tested. Polymorphisms of SOD2 and KLF6 were included in the analysis because they were associated with progressed liver disease in pediatric and adult NAFLD patients, whereby SOD2 is supposed to be involved in the induction of oxidative stress and KLF6 in determining fibrogenesis [90,91]. The prediction of fibrosis based on the four genetic determinants was less accurate (ROC-AUC of 0.60) than by combining this information with three clinical risk factors (age, diastolic blood pressure, and AST [aspartate aminotransferase]; ROC-AUC of 0.80) [89]. Similarly, a NASH risk score was established in a Korean cohort. The scoring system (NASH PNPLA3-TM6SF2 score, NASH-PT) was based on risk alleles of PNPLA3 rs738409 (C > G) and TM6SF2 rs58542926 (G > A), diabetes status, insulin resistance, and levels of AST and C-reactive protein. NASH-PT scores identified patients with NASH with a ROC-AUC between 0.787 and 0.859 [92].

In a cohort of 514 obese children and adolescents in which almost 70% of the participants were diagnosed with NAFLD by ultrasonography, Zusi et al. tested 11 genes and detected highly significant associations with risk alleles of PNPLA3 rs738409 (C > G), TM6SF2 rs58542926 (G > A), and GCKR rs1260326 (C > T) and a weaker association with a polymorphism of ELOVL2 rs2236212 (G > C) with a higher risk of NAFLD [93].

Di Costanzo et al. conducted exon sequencing of fatty liver genes discovered by GWAS and determined a polygenic risk score for NAFLD by applying logistic regression analysis. The authors confirmed PNPLA3 rs738409 (C > G), GCKR rs1260326 (C > T), TM6SF2 rs58542926 (G > A), and MBOAT7 rs641738 (C > T) as genetic contributors of hepatosteatosis, whereby the PNPLA3 SNP exhibited the strongest association, followed by GCKR, TM6SF2, and MBOAT7. The probability of NAFLD was highest (5-fold), when a risk score of all four SNPs was used [6]. EASL-EASD-EASO Clinical Practice Guidelines [94] already suggest genotyping for PNPLA3 rs738409 and TM6SF2 rs58542926 to identify individuals with a higher risk of hepatic steatosis.

In summary, a genetic risk score should include SNPs with three to four genes that showed the strongest and most robust association with hepatosteatosis and hepatosteatitis. These are PNPLA3, TM6SF2, GCKR plus MBOAT7, SOD2, and KLF6. Risk scores that combine the detection of these genes’ genotypes with clinical parameters might help clinicians to more effectively identify NAFLD patients at risk of NASH without taking liver biopsies.

Altered DNA methylation and miRNA expression in NAFLD

In general, there is increasing interest in elucidating disease-relevant epigenetic alterations, as they also have potential for therapeutic approaches. Lifestyle changes such as calorie restriction [95] and exercise [96] as well as more invasive interventions such as bariatric surgery [97,98] have already proven to cause changes in DNA methylation that positively affect the metabolic status of obese and diabetic patients or mice. Whether this also applies to NAFLD is subject of current and future research. Much less is known about specific histone modifications that lead to NAFLD and liver fibrosis. A review by Moran-Savador and Mann [15] provided information on the modifications of the histone code in liver disease.

Alterations of DNA methylation linked to NAFLD

The causal relationship between differential DNA methylation and disease has been extensively studied for various types of cancer [99]. The amount of data on the effects of aberrant DNA methylation on the development of metabolic diseases such as NAFLD has also increased enormously over the past decade. The relationship of methylome-transcriptome was analyzed in a histologically characterized NAFLD cohort to investigate whether differences between mild and advanced NAFLD are detectable [100]. Overall, hypomethylation occurred in NAFLD compared to controls regardless of disease severity. Furthermore, the genes whose transcription correlated with DNA methylation status were different in mild and advanced NAFLD. In advanced NAFLD, genes involved in wound-healing responses such as fibrogenesis were hypomethylated and their expression was upregulated, which distinguished them from the mild form. Murphy et al. suggested that the data might help to establish non-invasive markers to identify NAFLD patients at a high risk of liver disease progression [100]. Other genes with an altered DNA methylation pattern were identified in liver biopsies from mild and severe NAFLD cohorts using a candidate gene approach analyzing pro- and anti-fibrogenic genes. Zeybel et al. detected a higher methylation of specific CpGs within TGFβ1 and PDGFα, whereas specific CpGs exhibited a lower degree of methylation in the anti-fibrogenic PPARα and PPARδ genes in patients with mild fibrosis [101]. The same group later detected hypermethylation at the PPARγ promoter by analyzing plasma cell-free circulating DNA methylation [102].

In 2010, Sookoian et al. showed in a case-control study that epigenetic changes occur in hepatic insulin resistance in NAFLD patients [103]. Specifically, the decreased expression of PGC-1α, a key regulator of mitochondrial biogenesis and fatty acid oxidation in NAFLD, compared to controls was inversely correlated with its promotor methylation. Moreover, PGC1-α methylation positively correlated with peripheral insulin resistance and negatively correlated with mitochondrial biogenesis, two features that contribute to the pathogenesis of fatty liver. Further data describing a positive correlation between maternal BMI on methylation of the PGC1-α promoter in neonatal cord blood support the concept of intrauterine overnutrition and fetal epigenetic programming [104].

To investigate whether methylation changes can be detected even before disease onset, Kammel et al. conducted an experiment with the inbred mouse strain C57BL/6J, whereby a genetic effect on the phenotype could be precluded [9]. The experimental design allowed the identification of obesity-prone animals at a very young age when they were still in a metabolically stable state and did not show ectopic fat accumulation in the liver. As impaired hepatic insulin sensitivity contributes to the development of metabolic diseases, insulin growth factor 2 (IGF-2) and thus the IGF axis are crucial for regulating body weight. Disorders such as reduced plasma levels of IGF-1-binding protein 2 (IGFBP-2), which controls the bioavailability of IGF-1 and contributes to the development of obesity, have been observed in obese adults [105,106]. In young obesity-prone mice, Igfbp2 was hypermethylated at specific CpG sites and transcriptionally repressed even before the development of fatty liver and impaired glucose homeostasis in adolescence [9]. The results were not restricted to the mouse model. Interestingly, hypermethylation occurred at CpG sites that are homologous to humans with NAFLD [11], and IGFBP2 promoter hypermethylation correlates with type-2 diabetes risk [107]. IGFBP-2 levels are sensitive to weight changes as weight loss induced by bariatric surgery normalized IGFBP-2 levels and reduced liver fat content [108]. Whether this could partly be attributed to changes in methylation was not reported.

Hyperglycemia precedes ectopic fat accumulation in the liver, thus laying the grounds for further progression of fatty liver disease. Dipeptidyl peptidase 4 (DPP4), an exopeptidase that cleaves and inactivates numerous peptides including the incretin hormones glucagon-like peptide 1 (GLP-1) and gastric inhibitory peptide (GIP) [109], has been studied as a novel adipokine elevated in obesity [110]. Because of these facts, DPP4 inhibitors are widely used in clinical practice to improve glycemic control. DPP4 expression was elevated in NAFLD livers compared to healthy controls. In addition, expression levels negatively correlated with HOMA-IR, suggesting an association with DPP4 and insulin resistance and further supporting the relationship for disease progression and poor glycemic control [111]. Our own studies support the role of DPP4 in hepatosteatosis as well as broaden the information on this enzyme. In vitro studies have shown that DPP4 circulating in the blood is secreted primarily from the liver, so DPP4 can be referred to as a hepatokine [8]. Furthermore, the expression of Dpp4 in mice and humans negatively correlated with the methylation status, and similar to Igfbp2, these changes preceded the manifestation of the phenotype in mice [8]. A detailed analysis of human DPP4 expression in the ABOS cohort in combination with histology showed the inverse correlation of DPP4 expression and degree of steatosis, which was also valid for a second cohort (KOBS) comparing NASH patients and controls [8]. To determine whether the hepatic expression and methylation of Dpp4 could be influenced by diet, we used the New Zealand obese (NZO) mouse model, which is known for its early onset of hyperglycemia and T2D that can be postponed by low-protein diets [112]. Indeed, methylation of the Dpp4 gene was higher in mice fed a low-protein diet, which was associated with lower expression and reduced circulating DPP4 concentration. Furthermore, we reported a positive correlation of liver triglyceride content and DPP4 activity in the previously described setting [113].

Taking liver biopsies to reliably confirm NASH and measuring gene expression and the corresponding DNA methylation is highly invasive and poses risks to patients. Therefore, several groups are attempting to use less invasive procedures to screen for changes in DNA methylation in peripheral blood cells in epigenome-wide association studies (EWAS) to test whether specific changes allow a stratification of NAFLD patients with a higher risk of liver fibrosis. This approach with a total of more than 4,500 participants from four population-based cohort studies including European, Hispanic, and African participants was used to link elevated liver fat content measured by computed tomography or ultrasound imaging to altered DNA methylation levels. In the European participants, 22 CpGs were identified as associated with hepatic fat; of these, 19 CpGs were annotated to 18 unique genes such as DHCR24, SLC43A1, CPT1A, SREBF1, SC4MOL, and SLC9A3R1, which are involved in liver function. Epigenetic changes in ABCG1 and SREBF1 were linked to cholesterol biosynthesis. Thus, most affected CpGs were located in genes regulating key biological processes relevant to developing steatosis and explained approximately 10% of interindividual variations [114]. A smaller study including 18 histologically confirmed NAFLD and 17 NASH patients from a Han Chinese population identified 6 CpG sites located in the ACSL4, CRLS1, CTP1A, SIGIRR, SSBP1, and ZNF622 genes, which are differentially methylated in peripheral blood leukocytes of patients with NASH compared with those exhibiting simple steatosis. However, only differences in DNA methylation of ACSL4 were confirmed by pyrosequencing [115]. Thus, in particular for an early distinction between people at risk of simple hepatosteatosis and NASH will require further and larger studies to identify and verify robust epigenetic biomarkers.

Role of miRNAs in the development of NAFLD

Micro-RNAs (miRNAs) have been implicated in a number of diseases including metabolic diseases such as T2D and obesity [116]. Each miRNA acts post-transcriptionally to control the expression of multiple genes by translational repression or interference with RNA stability [117,118]. Thus, miRNAs are involved in regulating liver development, metabolic functions, and regeneration. Some alterations in intrahepatic miRNA networks have been associated with hepatosteatosis and NASH [119]. In addition, as miRNAs are also released into the circulation, specific miRNA signatures in blood can serve as noninvasive biomarkers for disease state and progression [14] and might perform similar or slightly superior compared to established biomarkers such as cytokeratin 18 (CK-18) or ALT and aspartate aminotransferase (AST). A review by Torres et al. provided a list of most miRNAs and their main targets identified in animal models, hepatocytes, human liver biopsies, and plasma [120]. We will focus on those candidates (miR-122, miR-33a/b, miR-34a, and miR-192) that are important for the pathogenesis of NAFLD.

MiR-122 stands out as it accounts for approximately 70% of all miRNAs expressed in the liver [121]. By targeting important transcription factors (for example, HNF6), miR-122 is implicated in liver development and physiology [122] and plays a fundamental role in lipid metabolism by targeting ACC2 [123] and SREBP [124]. In liver biopsies of obese patients with or without NAFLD, a reduced miR-122 expression was shown to be associated with fatty liver due to decreased fatty acid metabolism and altered expression of the transcription factors ChREBP, PPARγ, PPARα, and LXRα [124]. Whereas in serum of NAFLD patients, miR-122 levels were higher than in healthy controls and further increased in the state of NASH [125]. In contrast to humans, who exhibit a reduced hepatic miR-122 expression under circumstances of liver disease, high-fat diet-fed mice in which miR-122 was transiently inhibited by an antisense approach were protected from hepatosteatosis. They displayed reduced plasma cholesterol levels, increased hepatic fatty acid oxidation, and decreased rates of fatty acids and cholesterol synthesis in the liver [123]. Conversely, whole-body and liver-specific miR-122 knockout mice showed the expected phenotype; they developed steatohepatitis, fibrosis, and hepatocellular carcinoma [10,126]. Interestingly, fibrosis appeared to be mediated via targeting Klf6 [10], a known liver disease gene (see [89]).

MiR-192 is mainly expressed in the liver, especially in hepatocytes; lower amounts are detected in most other tissues [127]. Several members of the hepatocyte nuclear factor family play important roles in liver metabolism. One, HNF4α, regulates miR-122 and the expression of miR-192 (Figure 2). The deletion of this transcription factor in mice resulted in a marked reduction in miR-192 expression and a subsequent upregulation of miR-192 targets (activated leukocyte cell adhesion molecule [Alcam], epiregulin [Ereg], and moesin [Msn]) [128]. In contrast, TGF-β1 downregulates the expression of miR-192 by decreasing the binding of HNF family members to the promoter [129]. MiR-192 is specifically downregulated in hepatocytes upon liver injury in response to the induction of ischemic liver damage. As miR-192 knockdown experiments in Hepa1-6 cells resulted in an increased cell survival, it was speculated that the downregulation of miR-192 caused by liver damage might represent a protective mechanism against hepatocyte cell death. This effect appears to be mediated via the miR-192 target zinc finger E-box binding homeobox 2 (Zeb2), as the effect of miR-192 inhibition was reverted by co-transfecting Zeb2-siRNA [127]. Zeb2 is best known for its impact on epithelial to mesenchymal transitions [130] and its role as an anti-apoptotic protein that was mainly investigated in cancer. Through its interaction with SMAD, it mediates survival of tumor cells, particularly by enhancing cell proliferation and promoting cell migration and invasion [131].

Figure 2

Summary of the most relevant miRNAs and their targets in the liver and those detected in the plasma of NAFL and NASH patients.

In isolated exosomes of diet-induced obese (DIO) mice, Castano et al. detected elevated levels of miR-122 and miR-192 compared to those of lean mice. In livers of DIO mice, the amount of both miRNAs was also increased, whereas opposite effects were observed in white adipose tissue, and miR-122 and miR-192 were decreased compared to lean mice [132]. Similar to mice, miR-122 and miR-192 were upregulated 2-fold in serum of people with simple steatosis compared to controls. The difference was higher (7.2- and 4.4-fold, respectively) between control and NASH [14].

MiR-33a and miR-33b are two additional important miRNAs involved in fatty liver disease, lipid metabolism, and energy homeostasis of mice and humans. They differ only in two nucleotides in their mature form but are identical in their seed sequence. Both miRNAs are located in introns of SREBP-2 and SREBP-1, respectively, and are co-transcribed with these genes [133]. SEREBP-1 regulates genes required for fatty acid biosynthesis and SREBP-2 is involved in cholesterol metabolism [134]. Silencing miR-33a in mice increased the hepatic expression of the cholesterol transporter ABCA1 and increased HDL synthesis and circulating HDL levels [135]. Interesting results were obtained by inhibiting both miR-33 forms in non-human primates (African green monkeys). Again, ABCA1 was the most affected target and showed an elevated expression after miR-33 inhibition. Other upregulated genes were carnitine O-octanoyltransferase (CROT) and hydroxyacyl-coenzyme A-dehydrogenase (HADHB) encoding two enzymes involved in fatty acid oxidation, and insulin receptor substrate 2 (IRS2), which plays a key role in insulin signaling. When monkeys were fed a carbohydrate-enriched, moderate cholesterol diet, the authors detected an increased expression of carnitine palmitoyltransferase 1A (CPT1A), which is involved in fatty acid oxidation (Figure 2). Presumably via upregulating genes active in fatty acid synthesis (SREBF1, FASN, ACLY, and ACACA), miR-33 inhibition reduced the monkeys’ plasma VLDL triglyceride levels. Therefore, it was speculated that therapeutic inhibition of miR-33 could be a sufficient strategy for treating dyslipidemias [133]. Koyama et al. discovered that, at least in mice, miR-33b exhibits a higher atherogenic potential than miR-33a. Characterizing specific mouse lines that either carry a knockout of miR-33a or miR-33b on the genetic background of ApoE-deficient mice, an appropriate model for atherosclerosis progression, revealed that those mice that lacked miR-33a but expressed miR-33b developed increased atherosclerotic plaques. These results correlated with the expression levels of both miR-33 in wild-type mice, wherein miR-33b was much higher in the liver than miR-33a [136]. Martino et al. detected elevated circulating miR-33a and miR-33b levels in familial hypercholesterolemic children compared to aged-matched controls. For both miRNAs, the authors observed a positive correlation with total cholesterol, LDL cholesterol, the LDL cholesterol/HDL cholesterol ratio, APOB, C-reactive protein, and glycemia [137].

MiR-34 plays a fundamental role in the dysregulation of lipid metabolism associated with NAFLD [120]. The expression of miR-34 was induced by liver X receptor-α (LXRα), which itself was 4 and 7 times higher in NAFLD and NASH than in controls, respectively [138]. A recent study provided evidence that miR-34 plays a role in regulating autophagy/lipophagy by targeting ATG4B and Rab-8B, which are responsible for autophagosome and autolysosome formation [139]. Thus, after activation of LXRα, lipids accumulate in the liver due to the induction of miR-34, which suppresses autophagy (Figure 2). Another important target of miR-34 is sirtuin 1 (SIRT1), which is downregulated in the liver of NAFLD patients. SIRT1 is a regulator of energy homeostasis, which itself activates PPARα and LXR and inhibits PGC1-α expression [140]. Accordingly, Ding et al. observed a suppression of PPARα and SIRT1 in hepatocytes and livers in response to an upregulation of miR-34a, whereas silencing of miR-34 lead to an elevated expression of both regulatory proteins [141].

The impact of HNF4α on the liver function was mentioned previously; it exhibits a 2-fold higher expression in livers of NASH patients. Mice lacking this central transcription factor develop fatty liver. In humans it was shown that the expression of HNF4α is markedly reduced in NAFLD and NASH. This appears to be mediated by miR-34 binding to the 3′-UTR of HNF4α. The adenovirus-mediated application of miR-34a in mice reduced HNF4α expression by 40%, increased ectopic fat storage in the liver, and reduced plasma triglyceride concentrations. Opposite effects were observed in miR-34−/− mice, which showed a more than 3-fold increased HNF4α protein level [142].

MiRNAs detected in serum as diagnostic biomarkers

Several miRNAs are released by the cells packed in exosomes or circulating in a complex with argonaute2, which protects miRNAs from degradation via plasma RNases [116]. A review by Newman et al. summarized the miRNAs that were shown to exhibit significant alterations in liver disease and indicated the potential of their individual or combined levels as noninvasive diagnostic biomarkers [143]. Lopez-Rira et al. re-evaluated all 18 previously described serum miRNAs detected in clinical studies. Among these miR-122, -192, -34a, -16, and -21 were recognized in more than one study and Lopez-Rira et al. confirmed that they were affected in NAFLD. In addition, miR-27b, miR-22, miR-197, and miR-30c were significantly altered in more severe NAFLD patients. It was also confirmed that serum levels of miR-192, miR-34a, and miR-22 increased and miR-197 decreased in NASH patients (Figure 2). The authors also tested the diagnostic potential of miRNAs and observed similar classification performances of miRNAs vs conventional serum markers such as transaminases. However, when different ratios between induced and repressed miRNAs were considered (for miR-34a, miR-122, miR-192, miR-375, and miR-21), NASH prediction was better than using serum markers with ROC area under the curve values between 0.68 and 0.81 [144]. Pirola et al. previously reported that the detection of miR-122, miR-192, and miR-375 in serum has the potential to distinguish NASH from simple steatosis [14].

Conclusion

Predicting the individual risk of NAFLD and determining the probability of disease progression is the basis for further developing prevention and treatment strategies. Among other parameters, this requires knowledge of the genetic and epigenetic modifiers of NAFLD for genotype-guided risk stratification. In the near future, the current gold standard of liver biopsy could be circumvented by using the specified non-invasive risk scores that include plasma parameters, relevant clinical variables and a list of genetic and epigenetic changes. In the next step, risk scores need to be translated into clinical settings to benefit patients.

Financial support

This study was supported by a grant from the German and the state of Brandenburg (82DZD00302, A.S).

Conflict of interest

None declared.