Epigenetic Regulation of Hepatic Lipid Metabolism by DNA Methylation

Abstract While extensive investigations have been devoted to the study of genetic pathways related to fatty liver diseases, much less is known about epigenetic mechanisms underlying these disorders. DNA methylation is an epigenetic link between environmental factors (e.g., diets) and complex diseases (e.g., non‐alcoholic fatty liver disease). Here, it is aimed to study the role of DNA methylation in the regulation of hepatic lipid metabolism. A dynamic change in the DNA methylome in the liver of high‐fat diet (HFD)‐fed mice is discovered, including a marked increase in DNA methylation at the promoter of Beta‐klotho (Klb), a co‐receptor for the biological functions of fibroblast growth factor (FGF)15/19 and FGF21. DNA methyltransferases (DNMT) 1 and 3A mediate HFD‐induced methylation at the Klb promoter. Notably, HFD enhances DNMT1 protein stability via a ubiquitination‐mediated mechanism. Liver‐specific deletion of Dnmt1 or 3a increases Klb expression and ameliorates HFD‐induced hepatic steatosis. Single‐nucleus RNA sequencing analysis reveals pathways involved in fatty acid oxidation in Dnmt1‐deficient hepatocytes. Targeted demethylation at the Klb promoter increases Klb expression and fatty acid oxidation, resulting in decreased hepatic lipid accumulation. Up‐regulation of methyltransferases by HFD may induce hypermethylation of the Klb promoter and subsequent down‐regulation of Klb expression, resulting in the development of hepatic steatosis.


Introduction
Nonalcoholic fatty liver disease (NAFLD) is a growing metabolic disorder that has reached a high prevalence in US, with ≈10%-25% of adults being diagnosed in the general population. [1] The prevalence of NAFLD increases up to 75% in patients with obesity, and therefore the disease has been proposed to be renamed as metabolicassociated fatty liver disease. [2] The hallmark of NAFLD is the excessive deposition of triglycerides (TG) in the liver. [1] The disease begins with hepatic steatosis, a mild pathological change, and may progress to severe abnormalities such as nonalcoholic steatohepatitis (NASH), cirrhosis, and cancer. [1] While extensive investigations have been devoted to the study of genetic pathways related to fatty liver diseases such as hepatic steatosis, fibrosis, and NASH, much less is known about epigenetic mechanisms underlying these disorders. Epigenetic regulation, including DNA methylation, links common environmental factors (e.g., diets) to complex diseases (e.g., metabolic disorders). [3] DNA methylation of cytosines primarily at the CpG dinucleotides is one of the most common epigenetic modifications. CpG methylation frequently occurs in the promoters and 5' ends of genes, thereby critically regulating gene transcription. [4] De novo methylation of DNA is mainly handled by DNMT3A and 3B, while maintaining methylation patterns relies on another enzyme DNMT1 that catalyzes methylation on hemimethylated DNA strands during mitosis. [4] However, substantial evidence also supports the role for DNMT1 in de novo methylation in non-dividing cells. [5] Meanwhile, DNA demethylation can be achieved by the ten-eleven translocation (TET) dioxygenases that catalyze the hydroxylation of 5-methylcytosine to form 5-hydroxymethylcytosine and subsequent generation of 5-formylcytosine and 5-carboxylcytosine, which are then converted into unmodified cysteines by replication-related dilution or glycosylation-mediated base-excision repair. [6] DNA hypermethylation on the gene promoters is often associated with gene silencing, whereas DNA hypo-methylation typically represents a transcriptionally active state. [4] Metabolic disorders including obesity and NAFLD are complex diseases resulting from the interplay between genes and environmental factors, in which epigenetic mechanism serves as a link between the two. [3] Increasing lines of evidence indicate that epigenetic regulation plays a key role in the development of metabolic diseases including NAFLD. [7,8] This is an evolving research area and yet much remains to be discovered on how DNA methylation regulates hepatic lipid metabolism and the development of fatty liver diseases. In the present study, we employed a comprehensive approach integrating a genome-wide profiling of DNA methylation and gene expression, genetic models, and single-nucleus RNA sequencing (snRNA-Seq) analysis to delineate the role of DNA methylation in the regulation of hepatic lipid metabolism and to identify the key gene(s) whose DNA methylation status is epigenetically altered by the HFD, thereby contributing to the development of hepatic steatosis in male C57BL/6J mice.

A Dynamic Change of the DNA Methylome in the Liver of HFD-Fed Mice
To profile the DNA methylome during the development of hepatic steatosis, we performed a genome-wide DNA methylation analysis in the liver of C57BL/6J mice fed with either an HFD or low-fat diet (LFD) using the Reduced Representation Bisulfite Sequencing (RRBS) approach. We first confirmed the establishment of the mouse model of hepatic steatosis by characterizing lipid accumulation in the liver. Biochemical analysis showed a marked increase in liver TG contents in HFD-fed mice ( Figure  S1A, Supporting Information). Histological analysis by oil red O staining further confirmed lipid accumulation in hepatocytes, a feature of hepatic steatosis ( Figure S1B, Supporting Information). The RRBS analysis revealed that there are up to 686 differentially methylated regions (DMRs) in HFD-versus LFD-fed mice ( Figure 1A). These DMRs cover 380 genes, including 298 genes whose methylation rates are up-regulated by HFD (Table  S1, Supporting Information, pages 1-7) and 82 genes whose methylation rates are down-regulated by HFD (Table S1, Support-ing Information, pages [8][9]. The data suggest that HFD feeding mainly increases the DNA methylation on genes, which accounts for 78% of all genes with altered methylation rates (298/380). The methylation changes occur in the gene body spanning from 5'-end, coding sequence (CDS), intron, to 3'-end ( Figure 1B). It is noteworthy that there are 119 genes with the methylation changes on the 5'-end ( Figure 1B), a known gene region whose methylation may influence gene transcription. Genes with methylation changes are involved in various pathways including biological process, cellular component, and molecular function ( Figure 1C).
To determine which DNA methylation modifying enzymes mediate the HFD-induced alterations on DNA methylation in the liver, we performed a comprehensive analysis of Assay for Transposase-Accessible Chromatin Using Sequencing (ATACseq) and RNA-seq, which surveys genome-wide chromatin accessibility and gene expression respectively, in the hopes that a gene with a concerted change of open chromatin accessibility and increased expression can be converged. Using the liver samples from HFD-and LFD-fed mice, we compared the genome-wide changes in chromatin landscape assessed by ATAC-seq with the corresponding gene expression assessed by RNA-seq and discovered a strong correlation between the chromatin accessibility status and the gene expression in Dnmt1 and Dnmt3a. As shown in Figure 1D, ATAC-seq analysis revealed an enhanced peak at the Dnmt1 promoter (shown in green-highlighted box) in the liver of HFD-fed mice relative to LFD-fed mice, indicative of increased chromatin accessibility at the Dnmt1 promoter. This was associated with an up-regulation of Dnmt1 mRNA reads in the liver of HFD-fed mice revealed by the RNA-seq data ( Figure 1D, lower panel). A similar trend of an open chromatin structure at the Dnmt3a promoter with increased gene expression reads was observed in HFD-fed animals ( Figure 1D, bottom panel). In support of this observation, we confirmed the enhanced expression of Dnmt1 and Dnmt3a at both mRNA ( Figure 1E) and protein ( Figure 1F) levels in HFD-fed mice by quantitative RT PCR and immunoblotting respectively. However, there were no changes of Dnmt3b mRNA and protein levels in the liver of HFDfed mice ( Figure S2A,B, Supporting Information), nor were there any changes of chromatin accessibility and RNA-seq reads at the Dnmt3b gene ( Figure S2C, Supporting Information). These data suggest that DNA methylation may be important for the development of hepatic steatosis, and that DNMT1 and DNMT3A may be the key enzymes in this process.

DNA Methylation Regulates Hepatic Lipid Accumulation
To further determine the role of DNA methylation in the regulation of hepatic lipid metabolism, we employed a genetic approach by generating the mice with liver-specific deletion of Dnmt1 or Dnmt3a (LD1KO or LD3aKO) by an intravenous injection of AAV8-TBG-Cre virus, which has been successfully used in hepatocyte-specific deletion of genes of interest. [9,10] We found that mRNAs and protein levels of Dnmt1 and Dnmt3a decreased by more than 50% in the liver of LD1KO and LD3aKO mice respectively ( Figure S3A,B, Supporting Information). To confirm the knockout efficiency of Dnmt1 and Dnmt3a in hepatocytes, we isolated primary hepatocytes from LD1KO and LD3aKO mice and measured respective Dnmt1 and Dnmt3a at mRNA and protein levels. We found that hepatocytes isolated from LD1KO mice had a reduction of Dnmt1 mRNA by 60% and DNMT1 protein by 78% ( Figure S4A, Supporting Information). A similar decrease of Dnmt3a mRNA and protein levels was also observed in primary hepatocytes isolated from LD3aKO mice ( Figure S4B, Supporting Information).
We then put LD1KO mice on HFD for 10 weeks and conducted metabolic characterization. LD1KO mice had a slightly decreased body weight ( Figure S5A, Supporting Information) without change in fat pad mass ( Figure S5B, Supporting Information). Interestingly, LD1KO mice had significantly lower liver weight compared to the control flox/flox (fl/fl) mice receiving the AAV8-TBG-GFP virus (Figure 2A), which was consistent with a reduced liver TG contents ( Figure 2B). Histological analysis by H&E and oil red O staining further confirmed less hepatic lipid accumulation in LD1KO mice ( Figure 2C). However, there was no difference in the circulating lipid profile including TG, total cholesterol (TC), and free cholesterol (FC) ( Figure S5C, Supporting Information) between the two genotypes. We next conducted a metabolic characterization on LD3aKO mice fed HFD for 12 weeks. Similarly, there was a slight decrease in body weight ( Figure S6A, Supporting Information) of LD3aKO mice without change in fat pad mass ( Figure S6B, Supporting Information). Moreover, Dnmt3a deficiency in the liver reduced liver weight ( Figure 2D) and TG contents ( Figure 2E). Histological examination revealed a decrease in hepatic steatosis in LD3aKO mice ( Figure 2F). Despite no change in circulating TG ( Figure S6C, Supporting Information), LD3aKO mice exhibited reduced TC and FC contents in circulation ( Figure S6C, Supporting Information). In sum, the data indicate that inhibiting DNA methyltransferases by the genetic approach ameliorates HFD-induced hepatic steatosis in mice.
DNA demethylation can be made by a family of enzymes called the TETs including members TET1, TET2, and TET3 that are capable of removing the methyl group from 5-methylcytosine. [6] To study the physiological significance of TETs in the development of HFD-induced hepatic steatosis, we first determined whether HFD feeding alters Tets' mRNA expression in the liver of mice. Interestingly, 4-week HFD feeding consistently inhibited the expression of Tet1, Tet2, and Tet3 respectively ( Figure S7, Supporting Information).
Using the liver samples from HFD-and LFD-fed mice, we compared the genome-wide changes in chromatin landscape assessed by ATAC-seq with the corresponding gene expression assessed by RNA-seq. While Tet1 mRNA was barely detectable by the RNA-seq analysis ( Figure S8A, Supporting Information), we discovered a tendency of decreased expression of Te2 and Tet3 mRNA in the liver of HFD-fed mice relative to LFD-fed mice ( Figure S8B,C, Supporting Information). Unlike the ATAC-seq analysis that revealed increased chromatin accessibility at the promoters of Dnmt1 and Dnmt3a, there was no significant change of chromatin landscapes at the genes of Tets, suggesting that mechanisms other than chromatin structural changes might be responsible for the down-regulation of Tets' expression.
To further study the role of TETs in the development of hepatic steatosis, we knocked down all three Tets in the liver of mice. Intravenous injection of AAV Tet1-3 shRNA achieved a reduction of Tet1 mRNA by 75%, Tet2 mRNA by 50%, and Tet3 mRNA by 50% respectively without interfering nontargeted Tet expression ( Figure S9A, Supporting Information). The animals were then challenged with HFD for 5 weeks. Among the three Tets, knocking down Tet2 exhibited the most significant effect on the liver phenotype evident by the most increased liver weight ( Figure  S9B, Supporting Information). This was consistent with a dramatic increase in the hepatic TG contents ( Figure S9C, Supporting Information) and steatosis ( Figure S9D, Supporting Information) in Tet2 knockdown liver. To confirm the importance of TET2 in hepatic lipid metabolism, we further generated the mice with liver-specific deletion of Tet2 by intravenously injecting AAV8-TBG-Cre virus into Tet2 fl/fl mice (LT2KO). Challenged with HFD for 8 weeks, LT2KO mice had a slightly increased body weight ( Figure S10A, Supporting Information) and subcutaneous (SQ) fat pad mass ( Figure S10B, Supporting Information). Further characterization of LT2KO mice discovered increased liver weight ( Figure 2G) and TG contents ( Figure 2H), which was consistent with the histological examination showing more lipid accumulation in LT2KO mice ( Figure 2I). The hepatic steatosis observed in LT2KO mice was associated with increased circulating TC ( Figure  S10C, Supporting Information) and a trend of increase in FC and TG ( Figure S10C, Supporting Information). In sum, these data indicate that LT2KO mice largely exhibit an opposite phenotype to those of LD1KO and LD3aKO mice, further underscoring the importance of DNA methylation in hepatic lipid metabolism.

DNA Methylation Regulates Fatty Acid Oxidation
Liver is a heterogenous tissue that comprises of hepatocytes and non-parenchymal cells. To delineate the cell type-specific pathways underlying DNA methylation's effect on hepatic lipid metabolism, we performed a snRNA-seq analysis for the frozen liver tissues from LD1KO mice and their fl/fl controls using the 10X genomics platform. After sequencing a total of 19 063 nuclei (6645 fl/fl and 12 451 LD1KO) and conducting an unbiased clustering with the Seurat R package, we identified 12 main cell populations in the liver including hepatocyte, endothelial cell (EC), hepatic stellate cell, macrophage/Kupffer cell (KC), plasmacytoid dendritic cell (PDC), dendritic cell (DC), T/NK cell, B cell, cholangiocyte, plasma cell, mesothelial cell and dividing cell ( Figure 3A), based on the known cell type markers [11] (Figure 3B). We observed a shift in the relative composition of liver cell populations between LD1KO mice and fl/fl controls. Macrophages/Kupffer cells decreased from 25% in fl/fl liver to 9.4% in LD1KO mice, which was associated with a reduction of other immune cells, including T cells (from 5.3% to 3.2%) and DCs (from 2% to 1%), and ECs (from 37.5% to 28.8%) ( Figure 3C). By contrast, hepatocytes accounted for more proportion of LD1KO liver cells, increasing up to 39% from 22% in fl/fl mouse liver cells ( Figure 3C). These data suggest that Dnmt1 deletion prevents HFD-induced remodeling of the liver cell compositions by diminishing immune cell infiltration into the liver, thereby maintaining a healthy liver. Since the Dnmt1 knockout was largely restricted to hepatocytes, we performed a bioinformatic analysis of gene expression patterns in hepatocytes. A volcano plot analysis disclosed 5552 genes that were differentially regulated (fold change ≥1.5), among which 1055 genes, including a panel of genes involved in oxidative phosphorylation such as Ndufa1, Cox5a, and Cox6c, and fatty acid oxidation such as Pgc1 and Cpt1a, were up-regulated, and 4497 genes, including genes involved in fibrosis (e.g., Tgfbr1) and inflammation (e.g., Eif2ak2), were down-regulated in Dnmt1-deficient hepatocytes ( Figure 3D). Further KEGG pathway analysis of the differentially expressed genes revealed 10 significantly up-regulated pathways, among which oxidative phosphorylation was top-ranked, and a dozen down-regulated pathways, including the inflammatory pathway NF-B and the hepatic fibrosis signaling pathway ( Figure 3E), two hallmarks of NASH development. The results of the pathway analysis were in line with a hierarchical cluster analysis as shown in heatmaps indicating a marked up-regulation of genes responsible for oxidative phosphorylation ( Figure 3F) and fatty acid oxidation ( Figure 3G) and a broad down-regulation of genes involved in the NF-B pathway ( Figure 3H), phagosome formation pathway ( Figure S11A, Supporting Information) and hepatic fibrosis signaling pathway ( Figure S11B, Supporting Information) in Dnmt1-knockout hepatocytes.
Hepatocytes that align along the lobule axis display a zonation pattern with a spatial heterogeneity in both gene expression and metabolic functions. [12,13] We therefore sought to examine the hepatic gene expression in a spatial resolution. We identified 3 sub-clusters of hepatocytes ( Figure 4A) based on the expression of hepatocyte zonation markers ( Figure 4B) [14] and named them as periportal zone, midzonal zone (midzone), and pericentral zone as previously defined. [12] Unlike the zonation marker genes Sds and Glul that display a distinct expression in the periportal and pericentral zone respectively, no such specific landmark genes can be found in our study for the midzone presumably due to its feature as a transition region with gradients of hepatocyte gene expression across all three zones ( Figure 4B). [15] Interestingly, the periportal hepatocytes, featured by their superb abilities to utilize fatty acids due to possession of oxida-tively active mitochondria, [15] increased dramatically in their proportion accounting for the total LD1KO liver cells, compared to that of fl/fl mice ( Figure 4C). Further hierarchical cluster analysis showed a marked up-regulation of genes responsible for oxidative phosphorylation ( Figure 4D left panel) and fatty acid oxidation ( Figure 4D right panel) across the three zones in the LD1KO hepatocytes. Interestingly, the midzonal hepatocytes in LD1KO mice appeared to have a conspicuous increase in the expression of oxidative phosphorylation genes ( Figure 4D left panel), suggesting a functional transition to the periportal hepatocytes featured by a strong capacity for oxidative metabolism. In sum, these data indicate that enhanced fatty acid oxidation may be responsible for the reduced hepatic steatosis in LD1KO mice.
Indeed, quantitative PCR analysis further confirmed that inhibiting DNMT1 in hepatocytes significantly promoted the expression of genes involved in fatty acid oxidation such as carnitine palmitoyl transferase 1(Cpt1), acyl-coa oxidase 1 (Acox1), peroxisome proliferator-activated rexeptor-gamma coactivator 1 alpha (Pgc1 ), and peroxisome proliferator-activated receptor alpha (Ppar ) (Figure 5A), without change of lipogenic gene expression Figure S12A, Supporting Information). Similarly, LD3aKO mice exhibited increased expression of fatty acid oxidative genes ( Figure 5B) without change of most lipogenic gene expression except a decreased expression of acetyl-coa carboxylase 1 (Acc1) and fatty acid synthase ( Figure S12B, Supporting Information). In contrast, LT2KO mice with Tet2 deletion in hepatocytes had a down-regulation of fatty acid oxidative gene expression ( Figure 5C) with a reciprocal up-regulation of lipogenic gene expression such as Acc1 and stearoyl-coa desaturase 1 and fatty acid transporter gene cluster of differentiation 36 in the liver ( Figure S12C, Supporting Information). In support of altered DNA methylation that regulates the fatty acid oxidative program, seahorse analysis revealed an upregulation of oxygen consumption rate (OCR) in Dnmt1-deficient hepatocytes isolated from LD1KO mice ( Figure 5D). In sum, these data indicate that regulation of fatty acid oxidation by DNA methylation might be important for the development of HFD-induced hepatic steatosis.

Methylation at the Klb Promoter is Enhanced by HFD Feeding
To narrow down the molecules whose down-regulations are direct targets of promoter hypermethylation due to HFD feeding, we analyzed our RRBS data and discovered 83 genes with hypermethylation at 5'-end, a known gene region whose altered methylation most likely influence gene transcription. We then prioritized the genes that have been known to be involved in signal transduction and transcriptional regulation in metabolism, which led to 16 genes (Klb, Ctsb, Dusp26, Lgals3, Gdf10, One-cut1, Mgat1, Agap2, Aph1a, Lhcgr, Prom2, Mamstr, Crb2, Ub15, Actg, and Palm3). We reasoned that if the genes are epigenetic targets for the development of hepatic steatosis, HFD feeding ought to alter their DNA methylation, resulting in changes of gene expression. We therefore measured the expression of the 16 genes with changes of DNA methylation at the 5'-end, which are also involved in metabolism, in the hope that a gene with a concerted change of increased DNA methylation at the 5'-end and decreased gene expression can be converged. Among the 16 genes we screened with quantitative RT PCR, the expression of some genes such as Dusp26, Mgat1, Lhcgr, Prom2, and Ub15 was hardly detectable in the liver (data not shown), while there was no difference in the expression of Ctsb, Gdf10, Onecut1, and others between HFD-fed and LFD-fed mice (representative data of gene expression was shown in Figure S13, Supporting Information). Klb was left as the only hit that fits the pattern with increased DNA methylation at the 5'-end and decreased gene expression. Specifically, we discovered that the 5'-end methylation rate of the Klb promoter was significantly increased in the liver of HFD-fed mice as indicated in the UCSC Genome Browser ( Figure 6A). KLB has been identified as a coreceptor necessary for the physiological functions of FGF15/19 and FGF21, two hormones involved in various metabolic pathways including hepatic lipid metabolism. [16,17] The proximal promoter and 5' region of Klb is enriched with CpG islands ( Figure S14, Supporting Information), suggesting a possibility that the Klb promoter is subject to the regulation of DNA methylation. We therefore further conducted pyrosequencing to determine the methylation status of the CpG sites, which are located within the downstream proximity of the TATA box at the Klb promoter and at the beginning of the first exon. Indeed, our pyrosequencing analysis confirmed that HFD feeding significantly increased DNA methylation at the CpG sites at the Klb promoter ( Figure 6B). This was associated with downregulation of Klb mRNA expression in the liver of HFD-fed mice analyzed by quantitative PCR (Figure 6C), suggesting that enhanced DNA methylation at the Klb promoter may inhibit its gene expression. Changes in DNA methylation have been shown to modulate histone modifications, which may act cooperatively to influence chromatin structure and thereby regulate gene expression. [18] Indeed, ATAC-seq analysis revealed a reduced chromatin accessibility at the Klb promoter in the liver of HFD-fed mice, which was associated with a downregulation of Klb mRNA reads revealed by RNA-seq data ( Figure 6D). To determine whether the Klb promoter is indeed regulated by methylation, we cloned a 700 bp Klb proximal promoter including the CpG-enriched region into pGL3-luciferase expression vector. We then examined the fully methylated versus unmethylated Klb promoter activity. Our luciferase assays showed that the luciferase activity of the unmethylated promoter was fivefold higher than that of the fully methylated promoter ( Figure 6E). We next determined whether the Klb promoter may serve as a target for DNMT1 or DNMT3A to regulate its methylation and expression.
We conducted ChIP assays followed by SYBR green quantitative PCR to examine DNMTs' binding to the Klb promoter. We found that HFD feeding significantly increased DNMT1 ( Figure 6F) or DNMT3A ( Figure 6G) binding to the Klb promoter in the liver of mice. Unbiased snRNA-seq analysis of hepatocyte gene expression discovered an up-regulation of the Klb mRNA in Dnmt1deficient hepatocytes across the three zones as shown in the heatmap with down-regulated Dnmt1 expression as a control ( Figure 6H). Indeed, Klb mRNA was up-regulated in Dnmt1-or Dnmt3a-deficient liver while down-regulated in Tet2 knockout liver ( Figure 6I). Deletion of Dnmt1 reduced the average DNA methylation rates in almost all CpG sites at the Klb promoter ( Figure 6J), whereas deletion of Tet2 increased the average DNA methylation in most CpG sites at the Klb promoter ( Figure 6K). Signal network analysis using the differentially expressed genes of the Dnmt1-deficient hepatocytes from the snRNA-seq dataset revealed KLB as a metabolic nexus that governs the signals integrating mitochondrial oxidative phosphorylation and hepatic fatty acid oxidation ( Figure S15, Supporting Information). For instance, the FGF15/FGF21/KLB axis may regulate mitochondrial oxidative phosphorylation through the upregulation of PGC1 , a master regulator of mitochondrial biogenesis. Meanwhile, the FGF15/FGF21/KLB signal may also promote fatty acid oxidation by cooperating with the nuclear receptor PPAR that activates the transcriptional program of the fatty acid oxidative genes.
To test whether FGF15 or FGF21 has a direct impact on lipid metabolism in hepatocytes with Dnmt1 or Dnmt3a deficiency, we generated Dnmt1 or Dnmt3a deficient hepatocytes by infecting primary hepatocytes from Dnmts-floxed mice with AAV8-TBG-Cre virus and treated them with FGF15 or FGF21. Quantitative RT PCR analysis revealed that Dnmt1 or Dnmt3a deficiency increased FGF15 or FGF21-stimulated fatty acid oxidative gene expression (Figure 7A,B). These data strongly suggest that DNMT1 and DNMT3A may act on the Klb promoter to promote DNA methylation in response to HFD, resulting in HFD-induced hepatic steatosis.

Specific Demethylation at the Klb Promoter Ameliorates Hepatic Steatosis
Although we have determined the roles of DNMT1, DNMT3a, and TET2 in the regulation of Klb promoter methylation and lipid metabolism, it is not clear whether specific methylation at the Klb promoter mediates these metabolic changes in the liver of LD1KO, LD3aKO, and LT2KO mice. Besides, DNMT1 and 3A potentially catalyze DNA methylation in many genes other than Klb and deletion of the DNA methylation enzymes inevitably has an impact on the global DNA methylation status, which may  potentially confound the metabolic phenotypes. We therefore adopted a modified clustered regularly interspaced short palindromic repeats (CRISPR)/RNA-guided system to induce genespecific demethylation at the Klb promoter, guided by sequencespecific single guide RNAs (sgRNAs). [19] Male C57BL/6J mice were injected with lentiviral mixtures expressing deactivated CRISPR-associated protein 9 (dCas9)-Tet1 and sgRNA or scramble non-targeting sgRNA as a control intravenously, and mice were put on HFD one week after lentiviral injection. Mice infected with dCas9-Tet1 and sgRNA lentivirus did not show any difference in body weight and fat pad mass compared to the controls ( Figure S16A,B, Supporting Information). Lentiviral dCas9-Tet1 and sgRNA effectively reduced the average DNA methylation rate in almost all CpG sites at the Klb promoter ( Figure 8A), which was associated with an up-regulation of Klb mRNA and protein expression ( Figure 8B,C). Moreover, injection of dCas9-Tet1/sgRNA lentivirus significantly reduced liver weight and TG content in mice fed HFD ( Figure 8D,E). As a result, circulating TG levels were also decreased in mice infected with dCas9-Tet1 lentivirus ( Figure 8F). Further histological examination revealed reduced hepatic steatosis in the mice injected with dCas9-Tet1/sgRNA lentivirus compared to those infected with the control virus ( Figure 8G). Since Klb is a coreceptor required for a proper FGF21 signaling, the mice infected with dCas9-Tet1/sgRNA or the control lentivirus were injected with FGF21 to examine its signaling. dCase9-Tet1/sgRNA infected mice displayed an enhanced phosphorylation of extracellular signal-regulated kinase 1 and early growth response 1 induced by FGF21 compared to the control mice ( Figure 8H). This was consistent with an up-regulation of fatty acid oxidative gene expression including Cpt1 , Acox1, and cytochrome c oxidase 1 ( Figure 8I), although there was no difference in lipogenic gene expression ( Figure S16C, Supporting Information). Since FGF15/19 has been shown to repress bile acid synthesis, we therefore measured the expression of genes involved in bile acid synthesis. Quantitative RT-PCR analysis did not show a difference in the expression of genes including cytochrome P450 (Cyp)7a1, Cyp8b1, and Cyp27a1 between dCas9-Tet1 virusinfected mice and their controls ( Figure S16D, Supporting Information). In addition, dCas9-Tet1 virus-infected mice did not exhibit any change in the expression of FGF21 in the liver ( Figure  S16E, Supporting Information).  Using the same modified CRISPR/RNA-guided system, we conducted a gain-of-function experiment to induce specific methylation at the Klb promoter with lentiviral mixtures expressing dCas9-Dnmt3a and sgRNA. Male C57BL/6J mice were injected with lentiviral mixtures expressing dCas9-Dnmt3a and sgRNA or scramble non-targeting sgRNA as a control intravenously, and mice were put on HFD 1 week after lentiviral injection. Mice infected with dCas9-Dnmt3a and sgRNA lentivirus had a slight increase in body weight ( Figure S17A, Supporting Information). Lentiviral dCas9-Dnmt3a and sgRNA effectively increased the DNA methylation rates at the Klb promoter ( Figure  S17B, Supporting Information), which was associated with a down-regulation of Klb mRNA expression ( Figure S17C, Supporting Information). Moreover, injection of dCas9-Tet1/sgRNA lentivirus increased liver weight and TG content in mice fed HFD ( Figure S17D,E, Supporting Information), which was associated with increased TG levels in circulation ( Figure S17F, Supporting Information). Further histological examination revealed a marked increase in hepatic steatosis in the mice injected with dCas9-Dnmt3a/sgRNA lentivirus compared to those infected with the control virus ( Figure S17G, Supporting Information).
Taken together, our data indicate that inhibiting methylation directly at the Klb promoter promotes hepatic

Regulation of Hepatic DNMT1 Protein Stability by HFD
Since HFD appeared to exert a more potent effect on DNMT1 and DNMT3A protein levels than their respective mRNA levels, we reasoned that DNMT1 and DNMT3A protein might be regulated by protein stability. Recently the E3 ligase ubiquitin-protein ligase E3A (UBE3A) has been reported to prevent HFD-induced hepatic steatosis in mice. [20] Interestingly, ATAC-seq analysis revealed reduced peak reads at the Ube3a promoter in the liver of HFD-fed mice, which was associated with decreased Ube3a mRNA reads revealed by RNA-seq data ( Figure 9A). This was consistent with a down-regulation of UBE3A protein levels in the liver of HFD-fed mice ( Figure 9B). We then assessed the role of UBE3A in the regulation of DNMT1 protein stability. We found that overexpressing UBE3A in the liver of the C57Bl/6J male mice infected with AAV8 virus carrying Ube3a expression constructs significantly reduced DNMT1 protein levels while without effect on DNMT3A protein levels ( Figure 9C). We next examined DNMT1 ubiquitination with gain-or loss-of Ube3a in HEK293 cells. HEK293 cells with silenced or forced expression of Ube3a were treated with proteasome inhibitor MG132 to accumulate ubiquitinated proteins. We found that knocking down Ube3a in HEK293 cells abolished the ubiquitination of DNMT1 protein, which was largely restored by re-introducing the Ube3a expression vector into the knockdown cells ( Figure 9D). In contrast, overexpressing Ube3a significantly increased DNMT1 ubiquitination ( Figure 9D). We further confirmed UBE3A regulation of DNMT1 protein stability in a cycloheximide (CHX) chase assay. Cells were pretreated with CHX, a ribosome inhibitor, to suppress DNMT1 protein synthesis. Inactivating Ube3a by shRNA knockdown maintained high levels of DNMT protein at the late time point of 16 hours while overexpression Ube3a markedly reduced DNMT1 protein at the early time point of 6 h ( Figure 9E). These data suggest that DNMT1 protein stability was regulated by an UBE3A-mediated ubiquitination and that HFD feeding may enhance DNMT1 protein levels in the liver via inhibiting Ube3a expression.

Discussion
NAFLD is a serious metabolic disorder that has reached a high prevalence in US. [1] While numerous studies have been devoted to the investigation of genetic pathways related to the fatty liver disease, much is unknown about the role of epigenetic regulation in this disease. It is believed that epigenetic mechanisms mediate the interplay between environmental factors (e.g., diets) and the genome, leading to the development of various metabolic disorders including NAFLD. [3] Therefore, we employed a comprehensive approach that integrates multi-OMIC analysis and functional studies 1) to examine whether HFD, which has been believed to cause obesity and its associated disorders, modulates DNA methylome during the development of fatty liver and 2) to identify key gene(s) whose DNA methylation status is epigenetically altered by the HFD, thereby contributing to the development of hepatic steatosis. Using a genome-wide RRBS analysis, we demonstrated that DNA methylome in the liver is dynamically altered by HFD feeding, which presents a stark contrast to the notion that DNA methylation is a relatively stable epigenetic mark compared to other epigenetic regulations such as histone methylation and acetylation. Notably, we discovered that the DNA methylation rate at the Klb promoter is enhanced by HFD feeding through RRBS analysis. Given the fact that the proximal promoter and 5' region of Klb are enriched with CpG sites, we posited that Klb is subject to DNA methylation by HFD. Indeed, this was further confirmed by pyrosequencing analysis and luciferase assays. The enhanced DNA methylation at the Klb promoter may bear biological consequences, evident by the observation that Klb mRNA expression is down-regulated by HFD feeding, which may increase DNMT1 expression at mRNA levels or via decreasing UBE3A-mediated protein ubiquitination and degradation as a result of lower levels of UBE3A in the liver induced by HFD. Epigenetic programming of the Klb promoter by HFD down-regulates Klb expression, leading to hepatic lipid accumulation in obesity.
Several lines of evidence support the importance of DNA methylation in the development of NAFLD. For instance, aberrant DNA methylation patterns have been associated with the development of NAFLD. [21] This is congruent with the clinical studies also revealing altered CpG methylation on genes involved in the development of steatohepatitis, fibrosis, and carcinogenesis in patients with advanced NAFLD. [22] The global change in DNA methylation might be partially explained by the observation on the increased expression of the DNA methylation enzyme DNMT1 in NAFLD. [23] Despite the overexpression of DNMT1, which may potentially have a pleiotropic effect, the targets of the enzyme could still be pathway-and gene-specific in a given tissue or a physiological scenario. For example, abnormal DNA methylation has been identified in genes responsible for  A) The association of the peaks of the chromatin accessibility at the Ube3a promoter analyzed by ATAC-seq and the reads of the Ube3a mRNA expression analyzed by RNA-seq in the liver of HFD-fed or LFD-fed mice. B) Immunoblotting analysis of UBE3A in the liver of mice fed LFD or HFD. All data are expressed as mean ± SEM. n = 5; *p < 0.05 versus LFD. C) Immunoblotting analysis of DNMT1 and DNMT3A protein in the liver of mice infected with AAV Ube3a expression virus. All data are expressed as mean ± SEM. n = 5; *p < 0.05 versus control. D) UBE3A regulation of DNMT1 ubiquitination. Ube3a was knocked down by shRNA or re-expressed into the knockdown cells with Ube3a expression vectors or overexpressed with Ube3a expression vectors in HEK293 cells. DNMT1 protein was immunoprecipitated with an anti-DNMT1 antibody and followed by immunoblotting with an anti-ubiquitin antibody. E) UBE3A promotes DNMT1 protein degradation in a cycloheximide (CHX) chase assay. The CHX chase assay was conducted in the HEK293 cells with Ube3a knockdown or overexpression. The cells were pre-treated with CHX and then harvested in a time course ranging from 0 to 16 h. www.advancedsciencenews.com www.advancedscience.com specific pathways such as hepatic lipid metabolism. Compared with healthy subjects, patients with NAFLD exhibit increased CpG methylation in fatty acid oxidative genes and decreased methylation in fibrogenic genes, which may collectively contribute to the initiation and progression of NAFLD. [24] Moreover, analysis of liver biopsy samples from a cohort of NAFLD patients showed that methylation levels at the promoter of the fatty acid oxidative gene PGC1 was negatively associated with its gene expression and positively associated with insulin resistance. [25] We also determined the physiological impact of the altered methylome on hepatic steatosis. Since our RRBS data show that HFD tends to enhance DNA methylation in the genome, we reasoned that inhibiting DNA methylation through genetic approaches may correct the phenotype of hepatic steatosis. Indeed, we found that inhibiting DNA methylation by genetic deletion of Dnmt1 or Dnmt3a dramatically ameliorates hepatic steatosis in diet-induced obese mice. To examine the pathways mediating lipid metabolism, we found that Dnmts' deficiency suppresses the expression of genes involved in fatty acid oxidation and has some effects on the expression of genes involved in lipogenesis. This is associated with demethylation of the Klb promoter and upregulation of its mRNA expression, suggesting that Klb may serve as an epigenetic target that mediates the beneficial effect of Dnmt deficiency on hepatic steatosis. Klb is a coreceptor required for normal physiological functions of FGF15/19 and FGF21, which play an integral role in the regulation of key metabolic pathways. [16,17] For hepatic lipid metabolism, both FGF15/19 and FGF21 have been shown to antagonize hepatic steatosis via a coordinated control over stimulation of fatty acid oxidation and suppression of lipogenesis. [16,17] For instance, suppression of FGF21 in liver promotes hepatic steatosis via downregulation of hepatic fatty acid oxidation, [26] whereas over-expression of FGF21 in hepatocytes inhibits de novo lipogenesis via suppression of key lipogenic gene expression. [27] Similarly, increasing FGF19, the human counterpart of FGF15, via pharmacological or genetic approaches, can promote fatty acid oxidation [28,29] while suppressing de novo lipogenesis in the liver. [30] Since the main metabolic pathways that exhibits a dominant change in Dnmt-deficient liver involve fatty acid oxidation and oxidative phosphorylation, it is conceivable that other genes with changes of DNA methylation at 5'-end promoters in our RRBS data may also contribute the protective effects of Dnmt1 or Dnmt3a deficiency on hepatic steatosis. Further studies are required to examine the methylation status on these candidate genes and unravel the molecular events underlying the metabolic dysfunctions in the liver.
Our snRNA-seq analysis has shown a broad down-regulation of genes involved in inflammatory pathways and fibrosis in Dnmt1-deficient hepatocytes, suggesting a role of DNMT1 in hepatic inflammation. NASH is characterized by the presence of chronic inflammation and fibrosis in the setting of hepatic steatosis. [31] The activation of inflammatory response in the liver is a major driving force of the pathological progression towards hepatic fibrogenesis, [32] activation of which has also been discovered in our snRNA-seq analysis. The key question centers on the identification of the mechanism responsible for the down-regulated inflammation in Dnmt1-deficient hepatocytes. Several plausible pathways have been advanced to unravel the mechanisms underlying the activation of inflammatory response in hepatic steatosis, including immune cell infiltration (e.g., macrophages), lipotoxicity, endoplasmic reticulum stress, oxidative stress, mitochondrial dysfunction, etc. [31] However, future studies are warranted to uncover exactly how DNA methylation regulates the hepatic inflammatory program.
It is noteworthy that hepatic Dnmt1 deficiency preferentially increases the proportion of periportal hepatocytes in the total number of liver cells. Because hepatocytes residing in the periportal zone harbor larger mitochondria with a strong ability to oxidize fatty acids due to their unique anatomic milieu exposed to the polarized blood flow rich in oxygen and nutrients, [15] the disproportional increase of periportal hepatocytes is physiologically significant for Dnmt1-deficient hepatocytes to maintain a healthy lipid homeostasis through mobilizing lipid oxidation against excess fatty acid influx during HFD feeding. In light of the periportal hepatocytes being adept at performing oxidative metabolism, our hierarchical cluster analysis showed a marked upregulation of genes responsible for oxidative phosphorylation and fatty acid oxidation in Dnmt1-deficient hepatocytes. We posit that Dnmt1 deficiency ameliorates HFD-induced hepatic steatosis, at least in part, via increasing the proportion of the periportal hepatocytes. Future studies are required to address how hepatic Dnmt1 deficiency promotes fatty acid oxidation in a zonation-specific manner.
Along the course of our study, Kim et al showed that FGF19 ameliorates hepatic steatosis via DNMT3a-mediated repression of lipogenesis. [33] The role of DNMT3A in this study appears to be contradictory to the inhibitory effect of DNMT3A on Klb and fatty oxidation we observed in our study. The exact reason is not clear, but the two studies employed two different approaches to inhibit hepatic DNMT3A. While their study utilized AAVmediated shRNA to knock down Dnmt3a in the liver where other cells other than hepatocytes presumably underwent the knockdown regimen, we used a more hepatocyte-specific Cre-lox model where Dnmt3a-floxed mice were injected with AAV virus carrying hepatocyte-specific TBG Cre. [9,10] Further studies are required to explore this discrepancy.
Our data indicate that UBE3A regulates DNMT1 protein stability via ubiquitination-mediated proteasomal degradation and that HFD feeding may enhance hepatic DNMT1 protein contents via inhibiting Ube3a expression. Although HFD feeding also appeared to augment DNMT3A protein contents in the liver, we did not observe any effect of UBE3A on DNMT3A protein stability. We posited that other ubiquitination pathways might mediate the effect of HFD on the DNMT3A protein stability. Future studies will be warranted to address this question. Along the course of our current study, there are two studies showing that HFD treatment may down-regulate Dnmt1 expression in the liver. [34,35] Remely et al reported that a 16-week HFD feeding increased hepatic Dnmt1 mRNA expression, [34] while Li et al showed that treatment of C57BL/6 mice for 22 weeks promoted DNMT1 protein levels in the liver without change of Dnmt1 mRNA. [35] The exact reason for the discrepancy between their results and ours is not clear. The two studies employed a 22-week and 16-week HFD feeding regimen respectively, which are longer than that of our study. We speculate that the difference in HFD feeding duration may cause the difference in the effect of HFD on hepatic DN-MTs' expression. In addition, the fat content in the HFD used by Remely's study is slightly different from ours, which may have an impact on the expression of Dnmt1 in the liver, since different www.advancedsciencenews.com www.advancedscience.com HFDs with varied fat contents may cause the variations of hepatic steatosis. [36,37] Apart from the observations that different fat contents in HFDs may cause the varying degrees of hepatic steatosis and subsequently the DNA methylome, the mouse strain, and sex may also have impacts on the hepatic DNA methylation. [38] For example, mouse strains and sex that are prone to develop NAFLD are associated with a genome-wide DNA hypermethylation. [38] Along this line, the discovery of our study that employed the male C57BL/6J mice may not be extrapolated to other mouse strains, female mice, and humans.
In summary, our study demonstrates that the DNA methylome is under dynamic regulation during the development of diet-induced hepatic steatosis in male C57BL/6J mice, in which Klb may be an epigenetically regulated target of DNA methylation by HFD. We conclude that up-regulation of DNA methyltransferases by nutrient-rich diets may result in hypermethylation of the Klb promoter and subsequent down-regulation of Klb expression, thereby impairing fatty acid oxidation; this may in turn contribute to the development of NAFLD through the suppression of oxidative phosphorylation and fatty acid oxidation in male C57BL/6J mice.

Experimental Section
Mice: All animal procedures were conducted in accordance with the Institutional Animal Care and Use Committee guideline at Georgia State University (A22004). For diet-induced obesity studies, 6-week-old male C57BL/6J mice were purchased from the Jackson Laboratory (Bar Harbor, ME) and were fed either a LFD (D12450B, 10% kcal from fat, Research Diets Inc., New Brunswick, NJ) or HFD (D12492, 60% calorie from fat, Research Diets Inc.) up to 24 weeks. At the end of the study, liver and various fat pads were dissected, weighed, and snap-frozen in liquid nitrogen for further analyses.
We have generated mice with liver-specific deletion of Dnmt1(LD1KO), Dnmt3a (LD3aKO), or Tet2 (LT2KO) by intravenously injecting AAV8human thyroid hormone binding globulin (TBG)-Cre virus (produced by UPenn Viral Vector Core) [9,10] into Dnmt1-floxed mice from the NIHsupported Mutant Mouse Regional Resource Centers (MMRRC, No. 01 4114); Dnmt3a-floxed mice (No. 02 9885, MMRRC), or Tet2-floxed mice (No. 01 7573; the Jackson Laborary, Bar Harbor, ME). The Dnmt1-floxed mouse was created by inserting two loxP sites flanking exons 4 and 5, which causes frameshift and lacks the motifs for the catalytic domain. [39] The Dnmt3a-floxed mouse was created by inserting two loxP sites flanking exon 19, which encodes the catalytic motif. [40] The Tet2-floxed mouse was generated with the third exon flanked by two loxP sites. [41] All three mouse models have been backcrossed to B6 background for multiple generations in our lab. Although both LD1KO and LD3aKO lines (and their respective fl/fl controls) have been backcrossed to C57/BL6J background for more than 8 generations, the two lines may still have subtle differences in genetic background, which may affect their degrees of hepatic steatosis and time required for such development in responses to the same HFD feeding. During the HFD feeding study, we therefore monitored major metabolic phenotypes including body weight and blood lipid profile to ensure a comparability of the metabolic phenotypes between LD1KO and LD3aKO lines. We observed a slightly slower development of body weight gain and increased blood TG levels in LD3aKO mice. We therefore put LD1KO mice on the HFD for 10 weeks while LD3aKO mice for 12 weeks.