Liver‐specific deletion of de novo DNA methyltransferases protects against glucose intolerance in high‐fat diet‐fed male mice

Alterations to gene transcription and DNA methylation are a feature of many liver diseases including fatty liver disease and liver cancer. However, it is unclear whether the DNA methylation changes are a cause or a consequence of the transcriptional changes. It is even possible that the methylation changes are not required for the transcriptional changes. If DNA methylation is just a minor player in, or a consequence of liver transcriptional change, then future studies in this area should focus on other systems such as histone tail modifications. To interrogate the importance of de novo DNA methylation, we generated mice that are homozygous mutants for both Dnmt3a and Dnmt3b in post‐natal liver. These mice are viable and fertile with normal sized livers. Males, but not females, showed increased adipose depots, yet paradoxically, improved glucose tolerance on both control diet and high‐fat diets (HFD). Comparison of the transcriptome and methylome with RNA sequencing and whole‐genome bisulfite sequencing in adult hepatocytes revealed that widespread loss of methylation in CpG‐rich regions in the mutant did not induce loss of homeostatic transcriptional regulation. Similarly, extensive transcriptional changes induced by HFD did not require de novo DNA methylation. The improved metabolic phenotype of the Dnmt3a/3b mutant mice may be mediated through the dysregulation of a subset of glucose and fat metabolism genes which increase both glucose uptake and lipid export by the liver. However, further work is needed to confirm this.


| INTRODUCTION
The addition of a methyl group to a cytosine base at the carbon 5 position to make 5-methylcytosine is the most characterized epigenetic modification.It is an important component of transcriptional regulation and genome stability.In mammalian genomes, between 60% and 90% of CpG dinucleotides (cytosine nucleotide-phosphateguanine nucleotide) have methylated cytosines, with the variation depending on cell type and organism.In general, DNA methylation of gene promoters and repetitive elements is transcriptionally repressive, whereas gene body methylation is associated with transcriptional activity.In mammals, DNA methylation is achieved by the DNA methyltransferase family of proteins.De novo methylation of cytosine residues is catalyzed by DNA methyltransferases 3a (Dnmt3a) and 3b (Dnmt3b), while existing methylation is maintained through cell division by Dnmt1.In mice, null mutations cause severe developmental abnormalities that result in death in utero (Dnmt1 and Dnmt3b) or low survival beyond 4 weeks (Dnmt3a). 1,2Reduction of DNA methylation in cell culture and mice using tissue-specific gene deletion, gene knockdown, or drug treatments also can drastically alter gene expression. 3,4Thus in mammals, the genomic DNA methylation patterns that are set during development by the de novo methyltransferases Dnmt3a and Dnmt3b can persist for the rest of the individual's life 5 and are predominantly maintained by Dnmt1.
][8] DNA methylation changes occur throughout the genome in this period, particularly in males due to the involvement of testosterone, with enhancers the most commonly altered genomic element. 9In addition to developmental regulation, DNA methylation changes have been associated with homeostatic mechanisms in mature hepatocytes such as circadian rhythm 10,11 as well as environmental stimulation of hepatocyte functions such as diet-induced carbohydrate 12 and lipid 13,14 metabolism and drug metabolism. 15enome-wide and locus-specific DNA methylation changes in humans and mice have also been widely reported in liver diseases such as non-alcoholic fatty liver disease (NAFLD) also known as metabolic dysfunctionassociated fatty liver disease (MAFLD), [16][17][18] alcoholrelated liver disease (ARLD), 19 hepatitis C and B, and hepatocellular carcinoma. 20,21However, the causal importance of these DNA methylation changes is often unclear.DNA methylation changes may be a consequence of gene transcription levels 22,23 or even may not actually be required for gene expression changes if other epigenetic modifications are sufficient. 24These issues have clinical relevance also as therapies which aim to reverse pathology by inhibition or activation of epigenetic mechanisms are being explored for NAFLD, 25 hepatitis B, 26 and HCC. 27,28n cell culture, deletion or inhibition of DNMTs has shown a role of DNA methylation in regulation of Creactive protein 29 and highly expressed liver genes such as Albumin and Cytochrome P450. 30,31However, it is unclear whether these mechanisms are part of normal gene regulation or a mechanism specific to cancer cells.Furthermore, such studies give little functional insight into how liver-specific methylation changes integrate into multi-organ or body-wide biology.
To investigate the role of DNA methylation in postnatal liver function and disease, we generated a mouse line with liver-specific deletions of the de novo methyltransferases Dnmt3a and Dnmt3b and exposed them to a steatosis-inducing HFD.In our choice of experimental model, we wished to inhibit de novo methylation in mature hepatocytes in order to limit the risk of inducing developmental abnormalities or genomic instability which can be caused by deletions in early embryonic stages and by liver-specific deletion of the maintenance methyltransferase Dnmt1. 32Therefore, we chose to inhibit only the de novo methyltransferases Dnmt3a and Dnmt3b in hepatocytes using a cre-recombinase which is driven by the albumin gene promoter.Deletion of both Dnmt3a and Dnmt3b was required as in some contexts these genes act cooperatively, and deletion of only one does not always alter DNA methylation state at a locus. 33

| MATERIALS AND METHODS
All authors had access to the data and have reviewed and approved the final manuscript.

| Animals
All animal experimental protocols were approved by the Animal Ethics Committee, UNSW, and carried out according to their guidelines and regulations (ethics number 16/32B).Generation of Dnmt3a floxed/floxed , Dnmt3b floxed/floxed , Albumin-Cre mice, referred to as double knockout (DKO) mice (Supplementary Figure S1).Floxed Dnmt3a 34 and Dnmt3b 35 mice were purchased from the RIKEN BioResource Center (mouse lines RBRC03731 and RBRC03733).These were interbred with each other and C57BL/6 mouse line that has a Cre recombinase transgene under the control of the rat albumin promoter/ enhancer (Alb-cre) for hepatocyte-specific expression (B6.Cg-Tg(Alb-cre)21Mgn/JAusb). 36The original deletions of Dnmt3a and Dnmt3b were on 129/Sv background, although have been backcrossed to C57BL/6.Genome-wide analysis of 564 SNPs by the Australian Genome Research Facility Ltd confirmed that the DKO mice were >99% C57BL/6J strain and the only 129/Sv regions were at the Dnmt3a and Dnmt3b regions (Supplementary File S1).The 'wild-type' control mice used in experiments were the alb-cre line which were confirmed to be C57BL/6J strain at every SNP (i.e.100% C57BL/6J).
Primers spanning the deletion or loxP sites were used to quantify the effectiveness of Cre-driven deletion in genomic DNA.The primers Dnmt3aDelF GTGGACCGCTACATTGCCTC and Dnmt3aDelR TCC CTCTTGTCACTAACGCC-3 detected a double-floxed undeleted Dnmt3a locus of ~1283 bp or the deleted locus ~673 bp.For Dnmt3b, a semi-nested PCR was performed.Dnmt3bDelF1 GGATGTTCGAGAATGTTGTGGCC paired with Dnmt3bDelR CAGGTCAGACCTCTCTGGTGACAAG to amplify a ~883 bp region of the undeleted Dnmt3b locus.Dnmt3bDelF2 CGCAGGAAAGATTGGAACAT paired with Dnmt3bDelR to amplify a ~504 bp region of the deleted locus.Relative intensities of bands corresponding to the deleted or undeleted locus were quantified with Quantity One software (Bio-Rad).
Mice were bred and maintained at the Australian BioResources (ABR), Moss Vale facility.At four weeks old, DKO or WT mice were transferred to the Biological Resources Centre facility, UNSW, Australia, at 21°C ± 2, with a 12 h/12 h light cycle.Mice were acclimatized for a week on standard chow diet (11 kJ/g, 4% total fat, 13% digestible energy from lipids, Gordon's Stock Feeds, Yanderra, NSW, Australia).Then, mice were split into cages of the same genotype, with siblings assigned to either standard chow or HFD.Care was taken to ensure the average starting weights of each group was similar.Higher numbers of mice were allocated to the DKO HFD group as this group had increased variation in body weight.The HFD pellets were a semi-pure formulation for laboratory rats and mice based on Research Diets D12451 from Specialty Feeds, Glen Forrest, WA, Australia (23.5% of total weight is fat, 43% digestible energy from lipids, and 19 MJ/kg digestible energy, SF 04-001).Mice remained on their respective diets until the end of the experiment (21 weeks old).Livers from embryonic day E18.5 wild-type C57BL6 mice were dissected after timed matings and the embryo sex determined with qPCR 37 on DNA extracted with a DNeasy Blood and Tissue Kit (Qiagen 69504).

| EchoMRI and glucose tolerance test
At 15 weeks old, all mice underwent a whole-body composition analysis by nuclear magnetic resonance imaging (EchoMRI) assessing their fat mass and lean mass.The next day mice were fasted for 5 h prior to a glucose tolerance test (GTT).Baseline blood glucose was measured by tail nick at time 0. Mice were then challenged with an i.p. glucose bolus (2 g/kg lean weight as determined by echoMRI).Blood glucose was measured using Accu-ChekH glucose meter (Roche Diagnostics).

| Insulin ELISA
5 μL of whole blood from the time 0 and 30 min after glucose injection in the GTT was measured by Ultra-Sensitive Mouse Insulin ELISA kit (Crystal Chem Inc. 90 080).

| Columbus respirometry (CLAMS)
At 17 weeks old, respiratory quotient, resting energy expenditure, and activity were determined using comprehensive laboratory animal monitoring systems (CLAMS; Columbus Instruments) over a 24-hour period after acclimatization in the cages for 20 h.The CLAMS equipment allowed a maximum of 10 mice to be studied at a time.Therefore, the final data are compiled from 4 experiments.Representatives of all diets and genotypes were included in each experiment.The first of the four CLAMS experiments had only 2 h of acclimatization; however, statistical significance was equivalent with or without the inclusion of these mice.

| Primary hepatocyte isolation
Primary hepatocytes were isolated by collagenase perfusion.Briefly, while the mouse was anesthetized with ketamine/xylazine mix (200/20 mg/kg/BW i.p.), the liver was perfused through the inferior vena cava with EGTA buffer (HBSS buffer +0.5 mmol/L EGTA) for 15 min, followed by collagenase digestion (Collagenase H, Roche 11074032001) in calcium buffer (HBSS buffer +2 mmol/L CaCl 2 ) for 9 min.Liver cells were sieved through a cell strainer.Ketamine and xylazine anesthetic mixes have been reported to cause DNA damage to various organs in rodents. 38The damage in liver, which is the main site of its metabolism, is much lower than in the brain.However, the use of ketamine and xylazine in anesthesia and euthanasia in this study may have influenced the DNA damage and genotoxic transcriptional responses observed in the RNA sequencing (RNA-seq).

| Liver histological analysis
After all experimental protocols, mice received an overdose of anesthesia.Liver samples were collected and fixed in buffered formalin, dehydrated, embedded in histological paraffin, and then sectioned (5 μm), followed by deparaffinization, rehydration, and hematoxylin and eosin staining (H&E).To determine the degree of steatosis, 5 livers were selected randomly and scored from 1 to 4 by an operator blinded to treatment.Tissue sections were examined using a camera (QColor 3, Olympus®) attached to a light microscope.Groups were compared with Mann-Whitney's U test (nonparametric).

| Liver triglyceride quantitation
With the Abcam Triglyceride Assay Kit (Ab65336), 17-29 mg of powdered liver was added to 800ul lysate buffer and the experiment performed according to the manufacturer's instructions.

| Liver methylated DNA immunoprecipitation (MeDIP) and qPCR of LINE retrotransposons
Genomic DNA was extracted from ~20 mg of powdered liver tissue with Qiagen DNeasy Blood and Tissue Kits according to the manufacturer's instructions.1.5 μg of DNA was used as starting material for each immunoprecipitation with Abcam Methylated DNA Immunoprecipitation (MeDIP) Kit (ab117133) according to the manufacturer's instructions.Sonication was performed with a VirTis Virsonic 100 Sonicator, 2 × 10 s pulses at setting 5. Quantitative PCR was performed on a QuantStudio 5 (Thermo Fisher Scientific) with PowerTrack SYBR Green Master Mix (Applied Biosystems A46012) using mouse LINE-1 ORF1 primers 37 comparing input and immunoprecipitated DNA levels with a delta Ct calculation.

| RNA sequencing
Seventeen male mice were chosen for next-generation RNA sequencing of isolated hepatocytes: 4 WT chow group average body weight 28.0 g SD ± 2.5; 4 DKO chow group 31.1 g SD ± 2.3; 4 WT HFD group 37.2 g SD ± 3.4; and 5 DKO HFD group 44.9 g SD ± 3.1.An extra mouse in the DKO HFD group was added to compensate for the greater body weight variation in this group.RNA was TRI reagent extracted from isolated hepatocytes and treated with DNase I Amplification Grade (AMPD1-1KT; Sigma-Aldrich), rRNA depleted with Ribo-Zero and TruSeq HT libraries sequenced 2 × 150 bp on an Illumina NovaSeq S1 at ~45M reads per sample.Initial QC on the reads was conducted via fastp 39 at default settings for quality-based filtering and to trim overlapping read adapters and polyG tails.Trimmed paired reads were then quantified using the Salmon 40 aligner in quasi-mapping mode with the addition of non-default flags 'validateMappings', 'seqBias', and 'gcBias', mapping to GENCODE M21 transcript sequences indexed over k-mer length 31.The gene counts as quantified by Salmon were then carried on to DESeq2 41 in R to estimate differential expression between treatments via a negative binomial GLM with a Wald significance test over a maximum of 1000 iterations, testing the following four comparisons separately: WT chow vs. DKO chow, WT chow vs. WT HFD, DKO chow vs. DKO HFD, and WT HFD vs. DKO HFD.Genes were deemed significantly differentially expressed if they had a p-value of <.05 following Benjamini-Hochberg adjustment.
For estimation of the relative proportion of liver cell types in after primary hepatocyte isolation, we used CIBERSORTx. 42This digital cytometry method for deconvolution provides an estimation of the abundances of member cell types in a mixed cell population, using gene expression data from purified and validated individual cell types.We used previously identified liver cell-type count matrices of 900 genes 43 and compared them to the count matrices of the same genes in the 8 chow-fed mouse RNA-seq files from this study.CIBERSORTx Impute Cell Fractions function was performed to 500 permutations for significance analysis.

| Whole-genome bisulfite sequencing
Sixteen male mice were chosen for whole-genome bisulfite sequencing (WGBS): 4 WT chow group average body weight; 4 WT chow group average body weight 28.0 g SD ± 2.5; 4 DKO chow group 31.1 g SD ± 2.3; 4 WT HFD group 37.2 g SD ± 3.4; and 5 DKO HFD group 45.3 g SD ± 3.4.DNA was extracted with a Qiagen DNeasy Blood & Tissue Kit (Cat.No. 69504) and eluted in AE buffer (10 mM Tris-HCl, 0.5 mM EDTA). 10 μg of DNA was sent to Novogene and underwent sodium bisulfite conversion, library preparation, and Illumina PE150 sequencing at 90Gb raw data per sample.Sequence analyses were performed on the Galaxy Australia Bioinformatics Platform (https:// usega laxy.org.au/ ).Sequence files were aligned to the GRC39 (mm39) mouse genome with BWA-meth.Per-base CpG methylation metrics were extracted with MethylDackel with the output limited to CpG sites with a minimum of 5 reads coverage with output presented as CpG methylation fractions.Differentially methylated CpG-rich regions (DMRs) between the groups identified with Metilene with settings of a minimum of 10 CpGs per DMR and a minimum of 10% methylation difference between groups.Statistically significant DMRs which were below the threshold of Bonferroni adjusted p value of q < .05by were examined further.DMR regions were visualized and nearby genes identified in the UCSC Genome Browser (https:// genome.ucsc.edu/ , RefGenes extracted via Table Browser), and gene enrichment analysis was performed on significantly differential DMRs at (https:// david.ncifc rf.gov/ ).The locations of DMRs and RNA-seq reads were displayed in relation to each other and mouse genes using the IGV Genome Browser. 44

| Statistical analysis
To explore effects of genotype and/or diet, data were analyzed either by two-way ANOVA or repeated measures ANOVA with a Bonferroni adjustment for multiple comparisons using SPSS Statistics 24 software (IBM).Data were presented as mean ± SEM.Liver score in WT HFD and DKO HFD was analyzed with Mann-Whitney's U test.

Dnmt3b-deficient livers have altered body weight and fat deposition
To investigate the role of de novo DNA methylation in postnatal liver, we bred Dnmt3a floxed/floxed , Dnmt3b floxed/floxed , and Albumin-Cre mice (Supplementary Figure S1).These widely used mutant and transgenic lines have not been used in combination before.The Dnmt3a mutant deletes an exon which encodes the highly conserved PC motif of the methyltransferase catalytic domain, 34 whereas the Dnmt3b mutant deletes four exons which encode the PC and ENV motifs of the methyltransferase catalytic domain. 35PCR on genomic DNA from liver, heart, quadriceps, prefrontal cortex, retroperitoneal fat, testes, and ovaries from DKO mice suggested that the average effectiveness of Cre-driven deletion was greater than 86% at Dnmt3a and Dnmt3b in females and males.All other tissues had less than 7% of deletion (Supplementary Table S1).Examination of the transcript levels in the RNA-seq from male primary hepatocytes confirmed the reduction of the single exon at Dnmt3a and four deleted exons in Dnmt3b (Supplementary Figure S1B,C respectively).
6][47] We chose to evaluate the effects on liver development and metabolism with liver weight and histology, liver triglyceride measurement, RNA-seq and WGBS.Additionally, due to the liver's major role in regulating body-wide glucose and lipid metabolism, we measured body weight, body length, systemic glucose tolerance, circulating insulin levels, adiposity (fat pad weight and echoMRI of lean and fat mass), and respirometry (oxygen consumption, carbon dioxide output, and physical activity over 20 h).As expected, both males and females on HFD had higher body weight and percentage body fat than chow-fed mice, regardless of genotype (Table 1, Figure 1A,B,E,F).Body weight and percentage body fat were not different between female WT Cre and DKO mice.However, male DKO mice had higher body weight and percentage body fat than WT Cre.Lean mass was unaffected by diet or genotype in female mice, but males fed HFD had increased lean mass compared to chow-fed mice, with no genotype effect (Figure 1C,D).Statistically significant genotype diet effects were observed at sacrifice (22 weeks old) on naso-anal length, liver weight, retroperitoneal white fat weight in male but not female mice (Table 1).

Dnmt3b-deficient livers have higher glucose tolerance and energy expenditure than control mice
To investigate the effect of liver-specific impairment of de novo DNA methylation, we performed i.p.GTT on mice after 11 weeks on chow or HFD.At this stage, the DKO mice had higher body weight and adiposity than controls.Female mice displayed the expected increases in blood glucose levels and area under the curve (AUC) in HFDfed mice compared to chow-fed mice, but no differences due to genotype (Figure 2A,C).However, male mice, in addition to the expected diet effects, also had improved glucose tolerance as indicated by lower blood glucose levels and AUC (Figure 2B,D).This result seemed paradoxical in the HFD groups as the DKO mice had improved glucose tolerance yet were more obese than WT Cre mice.Measurement of blood insulin levels during the GTT in male mice indicated higher insulin levels in mice fed a HFD but no genotype effects (Figure 2E).
Further insight into the metabolic and adiposity differences was gained through close monitoring of male mice in a CLAMS metabolic cage system which showed that the daily energy intake was similar across diet and genotype groups (Supplementary Figure S2A).Energy expenditure, as derived by assessment of the exchange of oxygen for carbon dioxide, was significantly increased by HFD and DKO genotype (Supplementary Figure S2B).There was no genotype or diet effect on activity as measured by total horizontal movement (Supplementary Figure S2C).

Dnmt3b-deficient livers may be less prone to HFD-induced steatosis
Histological analysis revealed no major anatomical differences between WT Cre and DKO mice on either diet.However, there was a trend for HFD DKO mice to have less steatosis than HFD WT Cre mice (Figure 3A,B).This trend was also present in the level of triglycerides in male mouse liver (Figure 3C).

| Genome-wide LINE-1 retrotransposon DNA methylation is reduced in male but not female DKO mice
To examine genome-wide DNA methylation levels in female and male mice, we performed MeDIP on whole liver and qPCR for the ORF1 region of the most abundant autonomous retrotransposon in the mouse genome, LINE-1 (Figure 3D).LINE-1 retrotransposons typically have high levels of DNA methylation in order to reduce the genome-destabilizing effects of insertional mutation and recombination. 48This analysis revealed another sex difference in the effect of the deletions.As expected for inhibition of a DNA methyltransferase, in males the DKO deletions caused a significant reduction in LINE-1 methylation compared to WT.However, in females there was no difference.Examination of wild-type embryonic day 18.5 (E18.5)liver methylation levels also revealed that at the timepoint of the initiation of albumin-cre expression, males have significantly higher levels of LINE-1 methylation than females.
Due to the absence of significant genotype effects in females on body weight, echoMRI, GTT and genome-wide LINE-1 methylation we focused the subsequent transcriptomic and methylomic analyses to males only.

Main effect
Body weight (g)

| Transcriptomic analysis of hepatocytes
To assess the impact of the DKO deletions on liver transcriptional regulation and epigenetic state, we performed bulk RNA-seq and WGBS on hepatocytes purified from whole liver of male WT and DKO mice (main observations summarized in Table 2).Examination of the RNAseq data revealed that the primary hepatocyte purification procedures enriched for hepatocytes but did not deplete all other cell types (Supplementary Figure S3).All samples were between 72 and 80% hepatocytes, and there were no significant differences in the proportions of cell types between WT and DKO samples (Supplementary File S3).
Comparison of liver transcriptomes from chow-fed WT and DKO mice revealed relatively few differences.Only 8 genes were significantly downregulated and 30 significantly increased (Figure 4C, Supplementary File S2).Dnmt3a was reduced approximately 5-fold but Dnmt3b was not significantly different, as expected as the mutation alters protein stability but not transcriptional changes apart from the deleted exons (Supplementary Figure S1).There were no significantly enriched functional-related gene groups detected with the Database for Annotation, Visualization and Integrated Discovery (DAVID).However, the downregulation of the genes acyl-coenzyme A amino acid N-acyltransferase 2 (Acnat2), glycerol-3phosphate acyltransferase 3 (Gpat3), and orosomucoid 2 (Orm2) suggested that lipid metabolism may be affected in all DKO mice, even those on chow diet.Additionally, the reduction of S-adenosylhomocysteine hydrolase (Ahcy) hints at compensatory mechanisms at play to maintain the cellular methylation potential.
HFD had extensive effects on liver transcriptomes in WT and DKO mice (Figure 4E,G,I, Supplementary File S2).224 and 180 genes were significantly up-and downregulated, respectively, in WT mice in response to HFD. 297 and 290 were significantly up-and downregulated, respectively, in DKO in response to HFD.DAVID Gene Ontology analysis revealed similarities and differences in the response to the challenge of a HFD between WT and DKO (Supplementary File S2).Lipid metabolism pathways were significantly upregulated in response to HFD in WT and DKO with increases in lipid metabolism genes such as ApoA4, Acnat2, Acot11, and Cidec.Both genotypes also had increased immune and viral response processes in response to the HFD.However, many DNA methylation and epigenetic regulation processes were increased in DKO mice fed a HFD, but not in WT.Increases in genes such as Uhrf1, Mgmt and several histone proteins may suggest that compensatory epigenetic mechanisms are induced in the DKO mice when fed a HFD.
Alternatively, these epigenetic changes may be linked to the significant enrichment in cell stress, apoptosis, and DNA damage which were induced by HFD in DKO but not WT mice.With regard to processes reduced by HFD, in both WT and DKO, HFD reduced cellular glucuronidation, but glutathione metabolic process was only reduced in WT.
Direct comparison of the HFD-fed WT and HFD-fed DKO mice identified 69 genes with significantly higher levels and 47 with significantly lower levels in DKO (Supplementary File S2).DAVID gene ontology analysis showed that cellular response to xenobiotic stimulus, glutathione metabolism, lipid biosynthesis and catabolism, and glucose transmembrane transport were higher in DKO mice fed a HFD, while developmental processes were lower in DKO compared to WT mice.

| Whole-genome methylation analysis of hepatocytes
DKO mice had small but significant reductions in CpG methylation across the genome, regardless of diet.DKO chow-fed mice had an average 3.0% less CpG methylation than WT chow-fed mice (Figure 4A).Similarly, DKO HFD-fed mice had an average of 2.7% less CpG methylation than WT HFD mice.There was also a small significant overall effect of diet.HFD-fed WT mice had 1.0% less CpG methylation than chow-fed WT mice.Similarly, DKO HFD mice had 0.7% less CpG methylation than chow-fed DKO mice.
Locus-specific CpG analyses revealed strong effects of genotype, but very few due to diet (Figure 4B,D,F, Supplementary File S2).603 statistically significant DMRs were identified between chow-fed WT and DKO hepatocytes.Of these, 594 had lower methylation in DKO (average difference of 46% methylation, average DMR size 533 bp), and 9 had higher methylation in DKO (average difference of 16% methylation, average DMR size 217 bp).2017 statistically significant DMRs were identified between HFD-fed WT and DKO hepatocytes (Figure 4H, Supplementary File S2).Of these, 2013 had lower methylation in DKO (average difference of 44% methylation, average DMR size 533 bp), and 4 had higher methylation in DKO (average difference of 24% methylation, average DMR size 165 bp).13 statistically significant DMRs were identified between chow and HFD WT mouse hepatocytes.Of these, 11 had lower methylation in response to HFD (average difference of 15% methylation, average DMR size 300 bp), and 2 had higher methylation in HFD (average difference of 27% methylation, average DMR size 324 bp).7 statistically significant DMRs were identified between chow and HFD DKO mouse hepatocytes.Of these, 5 had lower methylation in HFD (average difference of 17% methylation, average DMR size 211 bp), and 2 had higher methylation in HFD (average difference of 24% methylation, average DMR size 121 bp).
4 Gene Ontology Biological Processes were significantly enriched in an analysis with the genes at or near the 603 DMRs between WT and DKO chow-fed mice (DMRs shown in Figure 4B, listed in Supplementary File S2). 3 contained overlapping gene sets and were linked to regulation of transcription (e.g.positive regulation of transcription from RNA polymerase II promoter), and the other was cell fate commitment.One pathway of note which was significantly enriched, but only prior to Benjamini correction multiple testing, was negative regulation of glucose import, due to DMRs at Appl2, Esr1, Grb10, and Prkca (examples in Supplementary Figure S4).
24 Gene Ontology Biological Processes were significantly enriched in an analysis between WT and DKO HFD mice of the genes at or near the 2017 DMRs (DMRs shown in Figure 4H, listed in Supplementary File S2).Positive regulation of transcription from RNA polymerase II promoter was again highly significant, but other processes such as phosphorylation, glucuronidation, development, and morphogenesis were also enriched.As for the WT versus DKO chow-fed mice comparison, some notable processes were significantly enriched but did not remain after Benjamini correction.These included lipid  into the role of novo methylation in post-natal development of the liver, environmental response, pathology, and homeostatic function.
At the outset of the project, it was unclear whether the deletions would lead to lethality, as is seen in the individual constitutive deletions of Dnmt3a or Dnmt3b. 34lternatively, potentially no phenotype at all could be the outcome if there is sufficient degeneracy in epigenetic regulation of genes (e.g.histone modifications are sufficient for liver homeostasis and response to a HFD).Conditional deletions of both Dnmt3a and Dnmt3b have been performed in a variety of other mouse post-natal cell and tissue types such as epidermis, 49 B cells, 50,51 kidney, 52 neurones, 53 hematopoietic stem cells, 54 lens, 55 and chondrocyte. 56,57All mouse mutants are viable.0][51][52] However, effects in unstimulated cells have been reported if the deletions are initiated at very early stages of cell lineage development, 54,58 or aging/degenerative phenotypes are examined. 57The observations of the hepatocyte-specific deletions in this study are in keeping with these other studies in that the deletions did not appear to reduce organism health, yet functional differences were observed upon stimulation (glucose injection or HFD).For consideration of the implications of the RNA-seq and WGBS data in this study, it should be noted that they represent the hepatocytes in the fasting state, as all mice were fasted for 6 h before sacrifice.Additionally, Kupffer cells, endothelial cells, and T cells were still detected in the hepatocyteenriched samples sent for RNA-seq and WGBS, and these cells will not have undergone deletion as albumin is not expressed in them (Supplementary Figure S3).
The total global methylation differences induced by the DKO were relatively small with both WT and mutant mice being within the range of 72%-77% of all CpGs being methylated (Figure 4A).A small reduction in global methylation in DKO mice compared to WT was expected as the majority of developmentally-set DNA methylation should be unaffected as Dnmt1 is present and able to maintain pre-established methylation through cell division.However, it is possible that some Dnmt3a and Dnmt3b activity is also present in DKO hepatocytes due to incomplete Cre recombination or compensatory increased transcription from unmutated alleles.Nonetheless, the DKO mutations did induce major reductions in methylation throughout the male genome, at LINE-1 retrotransposons in the MeDIP, and in the WGBS with 603 DMRs identified in chow-fed DKO compared to WT (an average of 46% less methylation, up to an 80% reduction), and 2017 identified in HFD DKO compared to WT (an average of 44% less methylation, up to a maximum 78% reduction).
Comparison of the number of DMRs between groups with the number of DEGs between the same groups indicated a striking separation between de novo methylation and transcriptional responses to a HFD.As expected for the deletion of de novo methyltransferases, the DMRs were nearly all a reduction in CpG methylation in DKO hepatocytes compared to WT.However, despite these 603 DMRs of considerable percentage methylation change, only 38 genes were differentially expressed.Importantly, this apparent disconnect between DNA methylation and transcription could be misleading if DMR-associated transcriptional changes require a particular stimulus to cause abnormal transcription.Thus, future work is needed to apply a variety of stimuli to the mice to confirm whether the DMRs really do not influence hepatocyte transcription or do influence transcription after certain physiological triggers.Nonetheless, the data do show that reduction of DNA methylation at hundreds of DMRs across the genome is not sufficient in itself to induce widespread transcriptional dysregulation.
Conversely, the observation that the extensive HFDinduced transcriptional changes are not associated with widespread DNA methylation changes suggests that de novo methylation is not essential for the regulation of most genes involved in hepatocyte macronutrient metabolism.In support of this, few DNA methylation changes were observed in another study which compared transcriptional and DNA methylation responses to a HFD. 59On the other hand, the differences in physiological response to HFD between WT and DKO mice (i.e.differences in body-wide glucose metabolism and gene transcription in hepatocytes) show that post-natal de novo methylation does influence some HFD responses.As discussed below, this may be due to de novo methylation being essential for the post-natal stages of hepatocyte development; thus, the differences in the physiological response to a HFD in WT and DKO mice may be due to methylation changes which occurred prior to the HFD exposure, rather than in response to the diet.
Four other hepatocyte-specific DNA methylation mutants are particularly relevant for this study-deletions of the maintenance methyltransferase Dnmt1, 32 Dnmt1 and Dnmt3a, 60 the DNA demethylases Tet2 and Tet3, 8 and liver-specific Dnmt3b. 61The phenotypic effects of these hepatocyte-specific deletions of methyltransferases and demethylases differed from Dnmt3a/3b deletion (DKO).The Dnmt1 deletion caused a nodular liver structure due to the activation of repetitive elements, resulting in DNA damage, cell death, and fibrosis. 32The transcriptional changes were fewer in the DKO mouse compared to the Dnmt1 mutant, with only 38 significantly differentially expressed genes in chow-fed mice compared to 231 in the Dnmt1 KO.Crygn was the only differentially expressed in both However, as expression of Igsf9, Slc1a1, Slc13a2, Phlda1, and Phlda3 in liver-specific Dnmt1 KO, and Ifsf23, Slc1a2, Slc13a3, and Phlda2 in DKO were differentially expressed, some gene families may be especially prone to regulation through DNA methyltransferases in post-natal liver.Phlda1 may be of particular importance for explaining the liver phenotype observed in this study as it has previously been shown to be regulated by DNA methylation and to be part of a network of genes that influence steatosis in mouse HFD models. 18Liver-specific knockdown of Phlda1 promoted steatosis, which fits with the increased expression and reduced steatosis seen in the DKO mice.
The AAV-cre-driven deletions of Dnmt1 and Dnmt3a reduced liver steatosis in HFD-fed mice. 60The authors attributed this to the deletions reducing DNA methylation at the Beta-klotho (Klb) promoter, thus increasing expression and the downstream increases of lipid oxidation in the liver.However, in our study we found no DNA methylation changes at the Klb promoter due to diet nor the combination of Dnmt3a and Dnmt3b deletions, and Klb transcript levels were actually decreased by HFD in WT mice (Supplementary File S2).
Unlike Dnmt1 mutants, Tet2/3 mutants have stable genomes, and phenotypes can be attributed to gene expression changes.The Tet2/3 deletion increased liver size and more than doubled blood glucose levels in males, which points to differences in the roles of de novo methylation and demethylation in the regulation of post-natal liver function. 8However, the major difference between the genomic regions which undergo demethylation in post-natal liver and those which gain methylation is that the former induce widespread transcriptional increases in the postnatal stages of hepatocyte development, particularly genes involved in liver lipid metabolism. 8The DMRs which had reduced methylation in the DKO mice and therefore normally gain methylation in the peri/post-natal period were not associated with widespread transcriptional changes.In DKO mice, the DMRs were found to be enriched at genes involved in lipid metabolism; however, the greatest enrichment was at genes involved in transcriptional regulation and cell fate commitment.Additionally, while 8% (4500/52000) of the post-natally demethylated regions were present in previously described super-enhancers, only 3% (18/594) of the post-natally de novo methylated regions were in liver super-enhancers. 62These differences in the relationship between transcription and DNA methylation change in the Tet2/3 mutants compared to the DKO mutant therefore support the model that DNA demethylation is more critical for gene activation than de novo methylation is for gene repression in development. 63onetheless, in wild-type mice there are roughly equivalent numbers of genomic regions which gain methylation as which lose methylation in post-natal liver, 6 so de novo DNA methylation does occur; it just may be nonessential for normal regulation of hepatocyte development and homeostasis.
A study which performed albumin-cre-driven deletion of only Dnmt3b gives insight into the relative contributions of Dnmt3a and Dnmt3b in the DKO mice. 61While that study did not include transcriptomics, it did have WGBS of WT and Dnmt3b-KO mice.Many of the DMRs identified were also seen in the DKO mouse.For example, the locations of the DKO and Dnmt3b-KO DMRs at Sirt4 and Sox9 overlapped (Supplementary Figure S4A).Comparing the 458 genes with nearby DMRs in the WTC vs. DKOC mice with the 424 genes with nearby DMRs in the WT vs. Dnmt3b-KO mice revealed only 79 genes in common, which suggests that methylation in the postnatal liver at many regions is regulated by either Dnmt3a alone or the combination of Dnmt3a and Dnmt3b.
The phenotypic changes in the DKO mice in this study indicate that de novo methylation is essential for normal regulation of some processes.The most striking physiological feature of the DKO mouse was the paradoxical increased adiposity yet improved glucose tolerance in males.Common explanations for increased adiposity include increased food intake, decreased activity, or decreased energy expenditure.However, no sign of these was seen in the DKO metabolic cage studies.In fact, the DKO mice displayed increased energy expenditure.Liver histological and triglyceride quantitation examination suggested that the Dnmt3a/3b deletions may reduce liver lipid levels.One explanation that links these observations is that the DKO liver has increased uptake of glucose which is then converted to lipid and exported from the liver to be stored in peripheral adipose tissue.The WGBS and transcriptome analyses provided evidence to support this physiological change in DKO mice.The WTC versus DKOC comparisons can be considered to represent the hepatocyte epigenetic and transcriptional state prior to HFD.Therefore, prior to HFD, the DKO mice will have had reduced methylation at genes involved in negative regulation of glucose import (WTC vs. DKOC DMRs at Appl2, Esr1, Grb10, Prkca-Supplementary Figure S4).The WTC versus DKOC DEGs also support the idea of altered lipid metabolism before exposure to HFD.Examples of these are increased Phlda1 (described above), the imprinted gene Rian (described below), the adipogenic hepatokine orosomucoid 2 (Orm2) 64,65 and lipogenic gene glycerol-3phosphate acyltransferase 3 (Gpat3). 66,67These alterations of macronutrient metabolism appeared to be exacerbated by HFD, and the WTH versus DKOH transcriptome and WGBS comparisons again supporting the model of increased carbohydrate uptake and lipid export in the liver.Reduced DNA methylation was found at 61 genes in the significantly enriched gene ontology 'Lipid metabolism genes', with the same term significantly enriched in the genes with higher transcript levels in DKOH than WTH (Supplementary File S2).Of those highly expressed genes, ApoA4 (chylomicron protein) (Supplementary Figure S4) is particularly linked to increased lipid export.Higher transcript levels of the main glucose and fructose transporters (Slc2a4 and Slc2a5, respectively) were also observed in HFD-fed DKO compared to WTH.
Previous work has shown that the male, but not the female liver genome undergoes extensive testosteronedependent demethylation in the post-natal period. 9This raises the possibility that the male-specific effects of DKO (e.g.adiposity, glucose tolerance) are due to testosteronedependent de novo methylation.However, the MeDIP experiment illustrated that natural sex differences in genome-wide methylation in liver development exist even before the onset of DKO mutation-LINE-1 retrotransposons had lower methylation in female E18.5 WT liver than males (Figure 3C).In adulthood, LINE-1 methylation levels were only affected in male DKO mice.Therefore, the absence of female phenotypes in DKO mice may be connected to female hepatocytes simply having less DNA methylation to lose or having the ability to function normally with lower DNA methylation levels.Future comparison of male and female hepatocyte methylation patterns in DKO mice, particularly through perinatal development, is required to investigate this further.
Finally, our initial aim was to impair de novo DNA methylation after the completion of hepatocyte development which would allow us to investigate DNA methylation which is only environmentally induced (e.g. in response to a HFD).However, the peri-natal activation of albumin-cre may mean that the Dnmt deletions occur at a critical timepoint for both the transition from carbohydrate to lipid energy metabolism (in utero glucose metabolism to milk) 68 and for the terminal stages of hepatocyte development. 6,69Alpha-fetoprotein (Afp) the major fetal plasma protein is produced by the liver until its repression around post-natal day 15. 6We observed that in adult WT mice, the 1st exon of Afp was 77% methylated but only 12% methylated in DKO suggesting that hepatocytes in DKO have not completely repressed fetal epigenetic programs.It is therefore possible that the metabolic changes seen in the DKO mice are due to the persistence of juvenile metabolism into adulthood. 70In support of this, it is noteworthy that a key regulator of the neonatal transition of liver energy metabolism, RNA imprinted and accumulated in nucleus (Rian), 71 was significantly upregulated in the liver of chow-fed DKO mice compared to WT chow-fed mice (Supplementary Figure S4Bi).Rian transcription is regulated by a genomic imprint differentially methylated region (IG-DMR). 72Reduced methylation was seen at this gene in the mir-341, mir-1188 region (36% reduction) as well as at the IG-DMR (28% reduction in R2 sub-region 73 ).Furthermore, the imprint control region at another gene which is influential in liver energy metabolism, Grb10, 74 had reduced methylation (40% reduction in DKO at CGI2 74 ).These observations also indicate that Dnmt3a and/or Dnmt3b is needed for maintenance of some genomic imprints in hepatocytes.
In conclusion, our study shows that post-natal de novo CpG methylation influences body-wide glucose metabolism and steatosis induced by high-fat feeding.This potentially occurs through the dysregulation of a subset of glucose metabolism regulatory genes and imprinted genes to enhance the uptake of carbohydrates and export of lipids by hepatocytes.Inhibition of de novo DNA methylation in the peri-to post-natal period does cause considerable locus-specific reduction of DNA methylation in adulthood, but this does not cause genome-wide loss of transcriptional repression.Additionally, most transcriptional responses to a HFD are not dependent on de novo methylation.Future work on this model is needed to investigate to what extent other hepatocyte roles such as detoxification, regeneration, biliary, and protein secretion functions are reliant on de novo DNA methylation.

T A B L E 1
Effects of diet and genotype on anthropometric measures at sacrifice (22 weeks old).

F I G U R E 1
Diet and genotype effects on body weight (A, B) and composition (C-F) in female and male mice.Data are shown and mean ± SEM.Body weight over time (A, B) was analyzed by repeated measures 2-way ANOVA with Bonferroni correction.Female BW n = 5-13 per group, male BW n = 13-25 per group.Percentage of body weight composed of lean mass (C, D) and fat mass (E, F) in female and male mice at 15 weeks was analyzed by 2-away ANOVA, main effects presented.Female echoMRI n = 5-12 per group, male echoMRI n = 12-22 per group.*p < .05diet effect, # p < .05genotype effect.

F I G U R E 2
Diet and genotype effects on glucose tolerance and insulin release.Data expressed as mean ± SEM for ipGTT curve, area under the curve (AUC), and insulin in blood during ipGTT.Blood glucose levels in ipGTT in female (A) and male (B) mice at 15 weeks old were analyzed with repeated measures 2-way ANOVA with Bonferroni correction.Female GTT n = 5-13 per group, male GTT n = 13-23 per group.AUC of the ipGTT (C, D) and insulin in blood during ipGTT (E) were analyzed by 2-way ANOVA n = 6-15 per group, main effects presented.Male insulin levels examined in.*p < .05diet effect, # p < .05genotype effect.

F I G U R E 3
Diet and genotype effects on liver steatosis and LINE-1 retrotransposon methylation.(A) H&E stain of 21-weekold male mice livers.Representative liver sections from the 4 diet and genotype groups.(B) Five livers from each group were scored for steatosis with '5' being the highest level of steatosis.Mann-Whitney's U test indicated a non-significant trend for a difference in score between WT HFD and DKO HFD (p = .095).(C) Liver triglyceride levels of 21-week-old male mice livers.No statistically significant group or diet effects.WT Chow n = 5, WT HFD n = 4, DKO chow n = 5, DKO HFD n = 8.No significant genotype or diet differences in 2-way ANOVA, WT chow compared to WT HFD in Student's t-test p = .16. (D) MeDIP-qPCR of relative LINE-1 retrotransposon DNA methylation levels in 21-week-old male and female chow-fed WT and DKO livers and WT E18.5 embryonic livers.Group comparisons performed with Student's t-test, *p < .05,***p < .001,ns no significant difference.Male E18.5 WT n = 5, female E18.5 WT n = 4, male adult WT n = 5, male adult DKO n = 5, female adult WT n = 3, and female adult DKO n = 6.All data expressed as mean ± SEM.

4 |
DISCUSSIONHepatocyte-specific deletion of both of the de novo methyltransferases Dnmt3a and Dnmt3b has not been attempted previously.As such, our observations give insightEffect of DKO Effect of HFDTranscriptome • Minor effects on transcription due to deletions • Major effects on transcription in both WT and DKO mice • Alteration of transcriptional response to HFD.Epigenetic processes, apoptosis, DNA damage, glutathione metabolism altered in DKO HFD, not WT HFD • Lipid metabolism and glucuronidation processes altered in both genotypes Methylome • Major effect on methylome • Minor effects on methylation observed • 99% of all DMRs had lower methylation in DKO than WT • DMRs enriched near genes involved in transcriptional regulation and cell fate commitment • No Gene Ontology Biological Processes were significantly enriched in the HFD-induced DMRs • Potential enrichment of DMRs near genes involved in glucose and lipid metabolism T A B L E 2 Summary of transcriptomic and methylomic data on chow and HFDfed WT and DKO male mouse livers.F I G U R E 4 (A) Average percentage of methylation at all genomic CpGs in hepatocytes of WT and DKO mice on chow or high-fat diet (HFD).Analyzed by 2-way ANOVA, main effects presented.(B-I) Volcano plots of RNA-seq differentially expressed genes (C, E, G, I) and WGBS differentially methylated regions (B, D, F, H) in hepatocytes from WT and DKO mice fed a chow (C) or HFD (H).Only differentially expressed genes (DEGs) or differentially methylated regions (DMRs) passing multiple correction are displayed.All comparisons of primary hepatocytes are n = 4-5 mice per group.
/group in males and 5-13 in females.Low n for liver values as liver perfusion for hepatocyte isolation prevented liver weighing.High n in HFD groups to compensate for greater body size variation seen with HFD.The significant effects (simple main effects) from a 2-way ANOVA are presented as: *p Abbreviation: BW, body weight.| 7f 17YAO et al.