DNMT1 maintains metabolic fitness of adipocytes through acting as an epigenetic safeguard of mitochondrial dynamics

Significance

Adipose tissue dysfunction is a major pathogenic determinant of obesity and related metabolic disorders. Although DNA methylation is one of the crucial epigenetic modifications in adipocytes, the molecular determinant and its roles in maintaining adipocyte function remain elusive. Here, we show that DNA methylation directs distal enhancer-mediated transcriptomic features of adipocytes. Particularly, deletion of adipocyte Dnmt1, the major methylation writer, provokes adipocyte hypertrophy and defective glucose homeostasis. Mechanistically, adipocyte Dnmt1 is crucial for preventing aberrant CTCF binding to sustain the proper chromosome architecture required for interactions between enhancer and Dnm1l and, in turn, facilitates mitochondrial fission and lipid homeostasis. Thus, we identify Dnmt1 as a key epigenetic safeguard of adipocyte integrity, which could have therapeutic implications for metabolic diseases.

Abstract

White adipose tissue (WAT) is a key regulator of systemic energy metabolism, and impaired WAT plasticity characterized by enlargement of preexisting adipocytes associates with WAT dysfunction, obesity, and metabolic complications. However, the mechanisms that retain proper adipose tissue plasticity required for metabolic fitness are unclear. Here, we comprehensively showed that adipocyte-specific DNA methylation, manifested in enhancers and CTCF sites, directs distal enhancer-mediated transcriptomic features required to conserve metabolic functions of white adipocytes. Particularly, genetic ablation of adipocyte Dnmt1, the major methylation writer, led to increased adiposity characterized by increased adipocyte hypertrophy along with reduced expansion of adipocyte precursors (APs). These effects of Dnmt1 deficiency provoked systemic hyperlipidemia and impaired energy metabolism both in lean and obese mice. Mechanistically, Dnmt1 deficiency abrogated mitochondrial bioenergetics by inhibiting mitochondrial fission and promoted aberrant lipid metabolism in adipocytes, rendering adipocyte hypertrophy and WAT dysfunction. Dnmt1-dependent DNA methylation prevented aberrant CTCF binding and, in turn, sustained the proper chromosome architecture to permit interactions between enhancer and dynamin-1–like protein gene Dnm1l (Drp1) in adipocytes. Also, adipose DNMT1 expression inversely correlated with adiposity and markers of metabolic health but positively correlated with AP-specific markers in obese human subjects. Thus, these findings support strategies utilizing Dnmt1 action on mitochondrial bioenergetics in adipocytes to combat obesity and related metabolic pathology.

Obesity is a worldwide epidemic that is closely linked to metabolic diseases such as cardiovascular diseases, type 2 diabetes (T2D), and certain types of cancer. Among various metabolic tissues, disturbance in white adipose tissue (WAT) function, including proinflammation and lipid spillover, are strongly associated with the development of obesity-related chronic diseases such as insulin resistance, dyslipidemia, and atherosclerosis (1, 2). Recent studies have also underscored the clinical importance of balanced adipose tissue plasticity with respect to adipose tissue function and related metabolic health regardless of body fat mass (3, 4). Adipose tissue plasticity is primarily modulated by the formation of new adipocytes from adipocyte precursors (APs) (hyperplasia) or increasing the size of preexisting adipocytes (hypertrophy). In particular, in lean as well as obese subjects, WAT expansion characterized by fewer but larger fat cells is associated with WAT dysfunction, hyperinsulinemia, insulin resistance, and dyslipidemia, whereas WAT expansion characterized by many small adipocytes is associated with improved metabolic parameters (5). Additionally, the findings that WAT hypertrophy is linked to the reduced adipocyte turnover indicate the importance of the interconnected mechanisms of interplay between adipocyte hypertrophy and hyperplasia in WAT plasticity (6). However, little is known about the underlying mechanisms promoting differences in WAT plasticity.

DNA methylation is a heritable epigenetic mark which frequently occurs at a cytosine immediately 5′ to a guanine (CpG sites) as catalyzed by DNA methyltransferases (DNMTs) in mammals. DNA methylation is implicated in regulation of various cellular processes and human diseases through modulating gene expression (7). Generally, DNA methylation has been associated with repression of cis-regulatory elements either by blocking the binding of transcription factors (TF) or by promoting recruitment of repressors (7). Additionally, recent progress in the methods [e.g., Hi-C (8)] to map regulatory interactions from genome-wide epigenetic datasets has provided insights into the broader functions of DNA methylation in the modulation of three-dimensional (3D) genomic architecture (9–11). Particularly, alteration of DNA methylation at binding sites of CCCTC-binding factor (CTCF), which is critical in consolidating the 3D genomic architecture, promotes aberrant CTCF occupancy at methylation-sensitive sites and disrupts the cell type–specific CTCF landscape (9–11). These changes in CTCF occupancy have been linked to pathogenicity by causing aberrant chromosome looping between distal cis-regulatory elements and their target promoters that promote altered gene expression (10, 12, 13). In the context of adipose biology, large-scale epigenetic studies have identified a large number of DNA methylation loci implicated in human obesity and related metabolic traits, many of which are intergenic and likely regulatory. So far, most studies on adipocyte biology have primarily focused on the genomic effects of DNA methylation at gene-proximal regulatory regions (14–16). However, recent findings have highlighted that, in addition to proximal regulatory regions, the regulatory circuit mediated by long-range genomic interactions contributes to adipocyte gene expression (8, 17) and that disruption of chromosomal looping in adipocytes is closely associated with obesity-related metabolic traits in humans (18). While these discoveries indicate the importance of transcriptional regulation mediated by long-range genomic interactions for adipocyte biology and metabolic pathophysiologies, the biological role of DNA methylation in this context remains poorly understood.

Here, using multilayer genomic analyses, we found that white adipocyte–specific DNA methylation, which is manifested in enhancers and CTCF sites, directs distal enhancer-mediated transcriptomic features required to maintain metabolic fitness of white adipocytes. Specifically, using Cre-Lox to generate mouse models in which Dnmt1 was selectively deleted in WAT or brown adipose tissue (BAT), we showed that Dnmt1 deficiency in white adipocytes promotes metabolically unfavorable adipose tissue expansion, systemic hyperlipidemia, defective systemic glucose metabolism, and decreased energy expenditure. Mechanistically, we identified Dnmt1-dependent DNA methylation and its downstream regulation of chromosome architecture as one of the key epigenetic mechanisms required for appropriate mitochondrial fission to retain mitochondrial bioenergetics and lipid metabolism in white adipocytes.

Results

Genome Analyses Reveal Enrichment of Adipocyte-Specific DNA Methylome in Distal Enhancers and CTCF.

To define a comprehensive landscape of adipocyte-specific DNA methylome, we measured the levels of CpG methylation in adipocytes at base pair resolution using whole-genome bisulfite sequencing (WGBS), which was intersected with publicly available WGBS data of the liver, which has distinct metabolic functions from adipocytes. As expected, adipocytes displayed a significantly different DNA methylome pattern from the liver, with 20,745 adipocyte hypermethylated differentially methylated regions (Hyper-DMRs) and 36,939 adipocyte hypomethylated differentially methylated regions (Hypo-DMRs) (Fig. 1A and SI Appendix, Fig. S1A). The majority of both adipocyte Hyper- and Hypo-DMRs were found at noncoding genetic segments and, to a lesser extent, within exons and promoters (Fig. 1B). Noncoding segments are enriched for enhancers that are crucial in the acquisition of the cell type–specific transcriptome profile (19), and up to 50% of enhancers cross over their nearest gene and regulate a more distal one (20). However, in contrast to proximal enhancers, it has not been comprehensively defined what functions DNA methylation may carry out in adipocyte distal enhancers. Thus, we mapped putative adipocyte distal enhancer-target gene(s) by integrating recently published adipocyte promoter capture Hi-C (PCHi-C) data (17) and histone chromatin immunoprecipitation sequencing (ChIP-seq) data. Among 12,522 genes interacting with adipocyte enhancer(s), 50% of genes were linked to ≥3 PCHi-C–assigned enhancers (SI Appendix, Fig. S1B). The median distance between PCHi-C–assigned enhancers and their target genes was 245 kb, while the closest genes were located 43 kb away from proximal enhancers (Fig. 1C). Of note, differentially expressed genes (DEGs, defined as |fold change| > 2) linked to enhancers (n = 3,449) were up-regulated in adipocytes, whereas DEGs without interacting enhancers (n = 5,747) were significantly down-regulated in adipocytes, which attests to the importance of distal enhancers in robust gene expression in adipocytes (Fig. 1D). Integration of WGBS data with the enhancer-target gene map revealed that 45.2% of enhancers harbored DMRs with a predominance of hypomethylation (61.6%) over hypermethylation (38.4%) (Fig. 1E), and the majority of these DMR enhancers lie distal (∼207 kb) to their target genes (SI Appendix, Fig. S1C). To assess the relationship of enhancer DMR classifications to target gene expression, DEGs were categorized based on the DMR types of their linked enhancers. While DEGs linked to Hypo-DMR enhancers were most up-regulated in adipocytes, there was no substantial difference in fold changes between DEGs linked to Hyper-DMR enhancers and DEGs linked to no DMR enhancers (Fig. 1F). Hypergeometric Optimization of Motif EnRichment (HOMER) motif analysis revealed that Hypo-DMR enhancers were preferentially coincided with adipocyte-related TF binding motifs (e.g., ETS1 and EBF), whereas enhancers with Hyper-DMRs harbored motifs for TFs of liver function (e.g., HNF and FOXA3) (Fig. 1G). Gene ontology (GO) analysis showed that not only did Hypo-DMR enhancers harbor motifs for adipocyte-specific TFs but their target genes were also enriched for adipocyte-specific function (e.g., adipogenesis and peroxisome proliferator-activated receptor [PPAR] signaling) (Fig. 1H), indicating a potential role of DNA methylation in the maintenance of adipocyte-specific distal enhancer activity and consequent gene expression.

Fig. 1.

DNA methylation participates in the regulation of transcriptomic profiles related to metabolic features of adipocytes (ADs). (A) Volcano plot showing the DMRs (liver versus ADs). (B) Genomic distribution of AD Hyper- (Liver < ADs, Left) and AD Hypo (Liver > ADs, Right)-DMRs. (C) Cumulative distribution of genomic distances between AD enhancers and nearest promoters (black) or promoter targets assigned by PCHi-C data (blue). (D) mRNA levels of DEGs with or without AD enhancers in ADs. The central lines indicate the median, the box limits are the 25th and 75th percentiles, the whiskers are the minimum to maximum values, and discrete points represent outliers. (E) Total number of AD enhancers with Hyper- and Hypo-DMRs. (F) mRNA levels of DEGs associated with different types of AD enhancers in ADs. (G) The enrichment ratios (observed/expected ratio) of TF motifs in Hyper- and Hypo-DMR enhancers. Red and blue dots represent TF motifs overrepresented in Hyper- and Hypo-DMR enhancers, respectively. The top eight enriched TF motifs in the respective groups are labeled. (H) Pathway enrichment of DEGs linked to enhancers with Hypo-DMR (database: WikiPathways 2016). (I) The percent of CTCF sites associated with differential CpG methylation as a proportion of CTCF sites harboring CpGs. (J) Boxplot of Δ CpG methylation for indicated CTCF sites, arranged by ascending AD CpG methylation decile. (K) An example of constitutive (Right) and AD-specific (Right) CTCF sites. (L) Boxplot of expression difference of genes in interaction power group 3 and 4, arranged by descending DMR type in ADs. (M) Pathway enrichment of DEGs in interaction power group 4 (database: WikiPathways 2016). *P < 0.05; **P < 0.01; ***P < 0.001; and ****P < 0.0001. Data are mean ± SEM.

Given that the majority of adipocyte enhancers appear to potentiate target gene expression through long-range genomic interactions (Fig. 1 C and D), we sought to validate whether, in addition to enhancer activity, adipocyte-specific DNA methylation was also associated with chromosomal looping, which would enable distal enhancers to control their target genes. Among the architectural proteins that consolidate the 3D genomic structure required for chromosomal looping, DNA methylome is associated with cell type–specific CTCF landscape (11, 21). To determine the functional relationship of adipocyte-specific DNA methylation with the CTCF binding landscape, we established the adipocyte selectivity of CTCF binding by comparing the combination of CTCF ChIP-seq data from 29 mouse diverse cell types, including hepatocyte and adipocytes. Next, we identified CTCF sites that were occupied in all analyzed cell types (constitutive sites; n = 2,661) and adipocyte-specific CTCF sites (n = 9,468). Intriguingly, the integration of WGBS data revealed that 46.8% of adipocyte-specific CTCF sites exhibited differential methylation in adipocytes (Fig. 1I). This proportion was significantly greater than the proportion of constitutive CTCF sites exhibiting differential methylation (9.2%) (Fig. 1I). In particular, adipocyte-specific CTCF sites were predominantly hypomethylated in adipocytes, whereas DNA methylation at constitutive CTCF sites was retained relatively low regardless of cell type (SI Appendix, Fig. S1 D and E). Parsing adipocyte-specific CTCF sites into adipocyte DNA methylation deciles further revealed that hypomethylation was not proportionally represented in all adipocyte-specific CTCF sites, but the extent of hypomethylation was greater in the lower and middle deciles than in the top deciles (Fig. 1 J and K). In contrast, constitutive CTCF sites showed little or no hypomethylation in all deciles (Fig. 1 J and K). These data suggested a close association of DNA methylation with adipocyte-specific CTCF binding landscape and consequent consolidation of chromosomal looping in adipocytes, implicating a new layer of complexity in DNA methylation–transcription control in adipocytes.

Enhancer–Promoter Interaction and DNA Methylation Function Coordinately to Establish Distinct Gene Expression Profiles Required for Metabolic Features of Adipocytes.

In adipocytes, the majority of genes were linked to more than one distal enhancer (SI Appendix, Fig. S1B), suggesting that adipocyte gene expression would not be explicable by altered activity of a single enhancer alone. Instead, it is likely that transcription would be modulated by cooperation between enhancer activity and interaction power, with the latter being determined by the number of interacting enhancers and the intensity of enhancer–gene looping interaction. Therefore, we next investigated the synergistic effect of adipocyte-specific DNA methylation on the coordinated action of enhancer activity and interaction power for gene expression. To avoid pleiotropic effects on transcription, we limited our analysis to DEGs that lacked a DMR promoter and were linked to DMR enhancers harboring only one type of DMR. The filtering process identified 2,264 DEGs, of which 50% were linked to ≥2 DMR enhancers (SI Appendix, Fig. S1F). These DEGs were further partitioned into four groups according to their interaction power, and the degree to which DMR enhancers affected target gene expression was assessed. In adipocytes, the impact of DMR types on target gene expression varied depending on the interaction power. While there were comparable fold changes in the expression of DEGs in all DMR types of enhancers in the lower interaction power groups (SI Appendix, Fig. S1G), DEGs with hypomethylated enhancers showed higher levels of up-regulation than DEGs with hypermethylated enhancers in the strong interaction power groups (Fig. 1L). Particularly, the combinational effect of strong interaction power and strong Hyper-DMR enhancers resulted in significant down-regulation of target gene expression in adipocytes (Fig. 1L). GO analysis revealed that these genes (e.g., ACAT3 and APOA4) were enriched in metabolic pathways, such as the statin pathway, which are retained relatively inactive in adipocytes (Fig. 1M). In contrast, strong interaction power and hypomethylated enhancers jointly drove the robust expression of key adipocyte genes (e.g., PPAR-γ and CD36) that were related to fat cell differentiation and fatty acid biosynthesis (Fig. 1 L and M). Together, these findings support the significance of an interplay between distal enhancer activity and long-range genomic interaction in the transcriptomic features of adipocytes and implicate the adipocyte DNA methylome signature as a key epigenetic modification mediating this interplay.

Adipocyte Dnmt1 Is Required for the Maintenance of Adiposity and Systemic Energy Homeostasis.

The cell type–specific DNA methylome pattern is orchestrated by DNMTs and the ten-eleven translocation enzymes that are DNA methylation writer and eraser, respectively. To identify the key mediator(s) of adipose DNA methylation, we systemically screened for genes expressed in various metabolic tissues from Genotype-Tissue Expression (GTEx) (22). Among various metabolic tissues, adipose tissue showed markedly higher enrichment of DNA methylome modifiers, and DNMT1, one of the major DNMT isoforms in mammals, was the most abundant DNA methylome modifier in adipose tissue (SI Appendix, Fig. S2A). Therefore, we decided to generate adipocyte-specific Dnmt1 knockout (Dnmt1 AKO) mice to dissect the causal role of DNA methylation in adipose tissue (SI Appendix, Fig. S2 B–E). In body weight chasing experiment, comparable body weight between wild type (WT) and Dnmt1 AKO mice until four postnatal weeks of age suggested that Dnmt1 depletion would not impact early development (Fig. 2A). In contrast, at six postnatal weeks of age, Dnmt1 AKO mice started to exhibit increased body weight concurrent with a substantially higher total fat mass, which was attributable to enlarged major fat depots, including inguinal WAT, epididymal WAT, and intrascapular BAT (Fig. 2 A and B and SI Appendix, Fig. S2F). Such changes in body weight and fat mass were observed without change in food intake and locomotor activity (SI Appendix, Fig. S2G). Therefore, these data suggest that adipocyte Dnmt1 deficiency would play a critical role in promoting fat mass gain, consequently increasing body mass.

Fig. 2.

White adipose Dnmt1 is an epigenetic modifier that regulates adiposity and systemic energy metabolism. (A) Body weight curves of WT and Dnmt1 AKO mice. (B) Total fat content and mass of different adipose tissue deposits. (C) Systemic glucose (Left) and insulin (Right) concentrations. (D) Serum levels of TG (Left) and NEFA (Right). (E) Plasma glucose profile between 0 and 120 min after glucose loading of WT and Dnmt1 AKO mice. (F and G) Absolute (Left) and average (Right) heat production (F) and rate of O₂ consumption (G) under dark and light cycles. *P < 0.05 and **P < 0.01. Data are mean ± SEM.

To validate whether increases in body weight and fat mass impact systemic energy metabolism, we analyzed serum profiles and whole-body metabolism. In Dnmt1 AKO mice, enlargement of adipose tissue was associated with elevation of circulating insulin and increased levels of serum lipid metabolites, including triglycerides (TG) and nonesterified fatty acids (NEFAs) (Fig. 2 C and D). Moreover, a glucose tolerance test revealed that Dnmt1 AKO mice were more glucose intolerant (Fig. 2E). Also, adipocyte Dnmt1 deficiency led to significant decreases in O₂ consumption (VO₂) and energy expenditure (Fig. 2 F and G). These data suggest that increased fat mass in Dnmt1 AKO mice would contribute to dysregulation of glucose homeostasis as well as whole-body energy expenditure. Previous studies have shown that both WAT and BAT play a critical role in regulating systemic energy homeostasis. Given that Adiponectin-cre in Dnmt1 AKO mice mediates gene deletion in both WAT and BAT, we further generated brown adipocyte–specific Dnmt1 KO (Dnmt1 BKO) mice where Dnmt1 was selectively depleted in BAT (SI Appendix, Fig. S2 H and I) to determine whether BAT was the primary tissue that contributes to metabolic dysfunction in Dnmt1 AKO mice. Similar to Dnmt1 AKO mice, there appeared to be no significant effects of brown adipocyte Dnmt1 depletion on early development (SI Appendix, Fig. S2J). However, by contrast with Dnmt1 AKO mice, the difference in body weight and fat mass disappeared between adult WT and Dnmt1 BKO mice (SI Appendix, Fig. S2 J and K), which coincided with the lack of differences in systemic glucose tolerance between the two genotypes (SI Appendix, Fig. S2L). These results argue in favor of a primary role of WAT in metabolic abnormalities in Dnmt1 AKO mice. To provide a clinical support for the findings in Dnmt1 AKO mice, we further examined the relationships between white adipose DNMT1 and surrogate markers of metabolic health in humans. As shown in SI Appendix, Fig. S2M, white adipose DNMT1 was lower in obese humans (body mass index [BMI] > 30, n = 14) than nonobese females (BMI < 30, n = 14). In this cohort, white adipose DNMT1 expression inversely correlated with BMI, suggesting a close association of white adipose DNMT1 with adiposity in humans (SI Appendix, Fig. S2M). Moreover, among various serum markers of metabolic complications, white adipose DNMT1 expression displayed a strong inverse correlation with serum insulin, Homeostatic Model Assessment for Insulin Resistance score, and serum TG (SI Appendix, Fig. S2 N and O), reinforcing the potential link between white adipose DNMT1 and metabolic health.

Adipocyte Dnmt1 Suppression Promotes Excess Lipid Accumulation in Preexisting Adipocytes and Leads to Hypertrophic Adipose Tissue Expansion.

Because adiposity was significantly increased by adipocyte Dnmt1 deficiency, we next determined the major contributing factor(s) to increased adiposity in Dnmt1 AKO mice. A histological analysis revealed that enlargement of adipose tissue was associated with marked adipocyte hypertrophy in all major fat depots including WAT and BAT in Dnmt1 AKO mice (Fig. 3 A and B). Quantitation of histological data further revealed decreased frequency of small adipocytes with pronounced increase in the frequency of large adipocytes in Dnmt1 AKO WAT (Fig. 3C). On the other hand, the difference in adipocyte size disappeared between WT and Dnmt1 BKO mice (Fig. 3 D and E), suggesting that white adipocyte Dnmt1 would be a major mediator of adipocyte hypertrophy in Dnmt1 AKO mice, and adipocyte hypertrophy in Dnmt1 AKO BAT would be a secondary effect stemming from the direct consequence of dysfunctional WAT exhibiting increased lipid spillover (Fig. 3F). In addition to increased adipocyte hypertrophy, Dnmt1 AKO mice exhibited a significant decrease in the total number of adipocytes in WAT, suggesting decreased in vivo adipocyte hyperplasia (Fig. 3G). To test whether decreased adipocyte hyperplasia was due to reduced AP proliferation or differentiation, we analyzed adipogenic potential of APs ex vivo. After 11 d of incubation with adipogenic induction medium, APs derived from WT and Dnmt1 AKO WAT exhibited comparable adipogenic capacity, as measured with Oil Red O staining and expression profiling of adipogenic marker genes (SI Appendix, Fig. S3 A and B). In contrast, flow cytometric analysis revealed a significant decrease in the number of highly enriched Cd31⁻; Cd45⁻; PDGFRα⁺ (Lin⁻; PDGFRα⁺) APs in Dnmt1 AKO WAT (Fig. 3 H and I). Such decreased AP abundance was accompanied by a decreased rate of AP proliferation (Ki67⁺ APs) in Dnmt1 AKO WAT (Fig. 3 H and I). Thus, it is likely that Dnmt1 deficiency in mature adipocytes has no effect on differentiation of APs into mature adipocytes but affected hyperplastic potential of APs. Next, we further validated whether hypertrophic adipocytes induced by Dnmt1 depletion contribute to decreased hyperplastic potential of APs (Fig. 3J). Consistent with the in vivo observation, coculture of ex vivo isolated Dnmt1 AKO adipocytes significantly reduced proliferation of APs as compared to ex vivo isolated WT adipocytes (Fig. 3 K–M). Together, these findings suggest that adipocyte Dnmt1 ablation would potentiate metabolically unfavorable adipocyte hypertrophic expansion in conjunction with inhibition of healthy hyperplastic growth of WAT. Additionally, because in obese humans, a phenotype characterized by adipocyte hypertrophy is associated with WAT dysfunction and more adverse metabolic profiles, whereas a phenotype characterized by adipocyte hyperplasia is protective (6, 23), we next investigated clinical relationships between white adipose DNMT1 and two metabolically distinct types of human obese subjects. A comparison of microarray data revealed that patients with metabolically unhealthy obesity (MUO; BMI = 49 ± 7, HOMA2 = 3.6 ± 5.3, n = 10) displayed reduced expression of PDGFRA, a representative marker of APs, and DNMT1 compared to BMI-matched patients with metabolically healthy obesity (MHO; BMI = 48 ± 3, HOMA2 = 1.3 ± 0.5, n = 10) (Fig. 3N). Moreover, we found that white adipose DNMT1 expression exhibits strong positive correlations with surrogate makers of APs whose abundance was positively correlated with insulin sensitivity in the previous study (23) (Fig. 3O). Reduced expression of these AP markers and DNMT1 was significantly associated with increased insulin resistance in the same cohort of obese patients (Fig. 3O). Collectively, these data obtained from human patients further suggest the potential association of adipose DNMT1 with hyperplastic capacity of WAT and related metabolic function.

Fig. 3.

Adipocyte (AD) Dnmt1 deficiency promotes impaired adipose tissue expansion by inducing AD hypertrophy. (A) Hematoxylin and eosin sections of distinct adipose depots of WT and Dnmt1 AKO mice. (Scale bar, 25 μm.) (B) Quantification of individual lipid droplets in epididymal WAT (eWAT) and intrascapular BAT (iBAT). (C) AD distribution and frequency with respect to the mean diameter of ADs of eWAT. (D) Hematoxylin and eosin sections of AT of WT and Dnmt1 BKO mice. (Scale bar, 50 μm.) (E) Quantification of individual lipid droplets in eWAT and iBAT. (F) Levels of glycerol released from eWAT explants. (G) Total number of ADs in eWAT. (H and I) The percent of APs in live singlet cells (Left) and percent of proliferating APs (Right) in WAT of 4-wk-old WT and Dnmt1 AKO mice as assessed by intracellular staining of Ki67. (J) Experimental design for ex vivo AP proliferation analysis. Stromal vascular fractions (SVFs) isolated from WT were cocultured with AD either from WT or Dnmt1 AKO WAT for 5 d, followed by colony formation analysis. (K) Photograph of crystal violet staining. (L and M) Comparisons of the number (L) and diameter (M, Left) of colonies. Cells from each genotype form colonies with different sizes as shown in the magnifications (100×) (M, Right). (Scale bar, 200 μm.) (N) WAT PDGFRA (Left) and DNMT1 (Right) expression in BMI-matched MHO or MUO. (O) WAT DNMT1 expression in relation to human AP–specific markers that are negatively correlated with insulin sensitivity. *P < 0.05; **P < 0.01; ***P < 0.001; and ****P < 0.0001. Data are mean ± SEM.

Hypomethylation Is Promoted by Dnmt1 Deficiency in Adipocytes.

Given that Dnmt1 functions as a DNA methylation writer, we next investigated whether these phenotypic changes in Dnmt1 AKO mice were associated with alterations of DNA methylation in adipocytes. WGBS profiling revealed that the changes in fat mass and systemic energy metabolism coincided with locus-specific hypomethylation rather than a global loss of methylation in Dnmt1 AKO adipocytes (SI Appendix, Fig. S4A). The majority of DMRs resided in noncoding segments, including intronic, intergenic, and enhancer regions, that were enriched for Hypo-DMRs rather than Hyper-DMRs (SI Appendix, Fig. S4 B and C). Particularly, these Hypo-DMR enhancers exhibited a preponderance of motifs for TFs involved in the maintenance (e.g., CTCF) and modification (e.g., RAR-γ and GATA3) of chromosome architecture (SI Appendix, Fig. S4D). GO analysis of target genes of Hypo-DMR enhancers further demonstrated the preferential association of these regions with biological processes regulating chromosome structure and lipid metabolism (SI Appendix, Fig. S4E). Collectively, these observations indicate that Dnmt1 would function as a fine-tuner of adipocyte-specific DNA methylation signature and that Dnmt1-dependent DNA methylation may be linked to adipose tissue plasticity.

Adipocyte Dnmt1 Deficiency Aggravates Obesity-Induced Adipose Tissue Remodeling and Defects in Energy Metabolism.

Even in the absence of nutritional stress, phenotypic changes of Dnmt1 AKO mice were reminiscent of obesity-induced adipose tissue remodeling and subsequent metabolic consequences. To ascertain whether Dnmt1 AKO mice would be more susceptible to high-fat diet (HFD)-induced aberrant alteration, we challenged Dnmt1 AKO mice and WT mice with HFD and measured adiposity and whole-body metabolism. Upon HFD, Dnmt1 AKO mice gained more body weight than WT mice in conjunction with increased adipocyte hypertrophy in WAT (SI Appendix, Fig. S5 A–C). Additionally, Dnmt1 AKO WAT displayed elevated lipid leakage and inflammation (SI Appendix, Fig. S5 D and E). These changes in Dnmt1 AKO WAT were associated with aggravation of metabolic parameters, including increased serum blood glucose, insulin and NEFAs, marked glucose intolerance, impaired insulin sensitivity, and severe fatty liver (SI Appendix, Fig. S5 F–J). Together, these data suggest that adipocyte Dnmt1 ablation would exacerbate HFD-induced pathological adipose tissue expansion and metabolic abnormalities.

Adipocyte Dnmt1 Regulates Mitochondrial Homeostasis and Lipid Metabolism by Modulating Dnm1l Expression.

To identify key metabolic pathways that led to adipocyte hypertrophy in Dnmt1 AKO WAT, we performed several assays, including import, synthesis, and utilization of energy sources. 2-Deoxyglucose uptake and fatty acid uptake analyses revealed that glucose and lipid uptake was not altered in adipocytes of Dnmt1 AKO mice (Fig. 4 A and B). Also, Dnmt1 depletion did not influence the expression of key lipogenic genes, including PPAR-γ and FAS in adipocytes (SI Appendix, Fig. S6A). By contrast, Dnmt1 AKO WAT displayed a significantly decreased rate of fatty acid oxidation that was attributable to the reduction of β-oxidation in adipocytes but not in the SVF (Fig. 4 C and D). Similarly, seahorse analysis revealed that oxygen consumption rate was reduced in Dnmt1 AKO adipocytes (Fig. 4E and SI Appendix, Fig. S6B). JC-1 staining with laser-scanning confocal fluorescence microscopy in combination with flow cytometry displayed a significant decrease in the red-to-green ratio of the JC-1 dye, indicating compromised mitochondrial membrane potential in Dnmt1 AKO adipocytes (Fig. 4 F and G). Moreover, Dnmt1 AKO adipocytes showed a reduction of mitochondria quantity, evident as the decreased abundance of mitochondrial DNA (mtDNA) and mtDNA-derived transcripts, including D-loop, ND1, ND4, and ND6 (Fig. 4H and SI Appendix, Fig. S6C). These results suggest that Dnmt1 depletion would promote defects in mitochondrial bioenergetics and mitochondrial mass control in adipocytes.

Fig. 4.

Diminished mitochondrial fission by Dnmt1 deficiency abrogates mitochondrial bioenergetics and lipid metabolism in white adipocytes (ADs). (A) [¹⁴C] Deoxyglucose uptake in epididymal WAT (eWAT) with or without 100 nM insulin treatment. (B) Boron-dipyrromethene-fatty acid uptake in ADs isolated from eWAT. (C) Ex vivo β-oxidation in eWAT. (D) Ex vivo β-oxidation in AD and SVF from eWAT. (E) Basal (Left) and maximal (Right) oxygen consumption rate of SVF-derived ADs. (F) Representative JC-1 staining of WAT. (Scale bar, 25 μm.) (G) Representative JC-1 aggregate fluorescence-activated cell sorting fluorescence distributions (Left) and ratio of red/green fluorescence (Right) in AD isolated from eWAT. (H) Relative level of mtDNA/nuclear DNA in eWAT. (I) Relative mRNA levels of genes involved in mitochondrial fission. (J) Representative eWAT immunoblot (Left) of DNMT1, DRP1, and GAPDH. Densitometry analysis of Drp1 immunoblot (Right). GAPDH was used as the loading control. (K) Representative MitoTracker green staining of WAT, demonstrating the presence of aberrant mitochondrial morphology (arrowhead) in Dnmt1 AKO mice. Selected cells are contoured by a dashed line in MitoTracker panels. Magnified images of the boxed regions are shown as insets. (Scale bar, 5 μm.) (L) Representative transmission electron microscopy images (Left) of eWAT. Magnified images of the boxed regions are shown as insets. (Scale bar, 1 μm.) Quantification of mitochondrial number per image (Middle; n = 6 for WT and n = 9 for Dnmt1 AKO mice) and mitochondrial size (Right; n = 76 for WT and n = 56 for Dnmt1 AKO mice) at 3,200×. LD, lipid droplet; M, mitochondria. (M) WAT DNM1L (Left) and OPA1 (Right) expression in nonobese and obese humans. (N) Correlation of WAT Dnmt1 with Dnm1L (Left) and OPA1 (Right) from Fig. 5L. *P < 0.05; **P < 0.01. Data are mean ± SEM.

Mitochondrial biogenesis and dynamics are orchestrated to maintain mitochondrial abundance as well as fitness. In Dnmt1 AKO adipocytes, mitochondrial fission protein, especially dynamin-1–like protein Drp1 (encoded by the Dnm1l gene), were significantly down-regulated with no variation in the expression of genes involved in mitochondrial biogenesis, lipid oxidation, oxidative phosphorylation, and fusion (Fig. 4 I and J and SI Appendix, Fig. S6 D–H). The identification of Dnm1l as a potential target of epigenetic dysregulation by Dnmt1 depletion was of particular interest given its prominence as a key regulator of mitochondrial dynamics. Also, closer examination of expression patterns in the GTEx dataset revealed that the expression of human white adipose Dnmt1 was positively correlated with Dnm1l expression (SI Appendix, Fig. S6K), consistent with the findings in Dnmt1 AKO mice. However, human white adipose Dnmt1 displayed no correlation with PGC1-α and a weak correlation with OPA1 (SI Appendix, Fig. S6K). In line with previous reports of Dnm1l KO cells, MitoTracker staining and ultrastructural examination showed irregular mitochondrial network paralleled by an increase in large swollen mitochondria with perturbed cristae, while mitochondrial density was partially lost in Dnmt1 AKO adipocytes (Fig. 4 K and L). Given that defective mitochondrial fission is frequently associated with abnormal degree of mitophagy (24), we assessed mitophagic signaling in Dnmt1 AKO WAT. While Drp1 protein was significantly decreased by Dnmt1 depletion, the levels of major regulators of mitophagy, including Parkin and Pink1, and its putative outer mitochondrial membrane target, Mfn2, were not greatly changed (SI Appendix, Fig. S6 G and I). In parallel, the translocation of Parkin to mitochondria was not affected by Dnmt1 depletion (SI Appendix, Fig. S6J). Furthermore, Dnmt1 AKO adipocytes were more susceptible to HFD-induced defects in mitochondrial fission and its adverse consequences in lipid oxidation (SI Appendix, Fig. S7 A–D). Together, these data suggest that Dnmt1 ablation would disrupt fission-mediated mitochondrial quality control, interfering with the maintenance of healthy mitochondrial population in adipocytes. Similar to Dnmt1 AKO mice, white adipose DNM1L expression, but not OPA1 expression, was reduced in obese human subjects as compared to lean human subjects (Fig. 4M). Additionally, DNM1L expression positively correlated with DNMT1 expression in human adipose tissue, a finding that further proposes a potential role of DNMT1 in the transcriptional regulation of adipose DNM1L (Fig. 4N).

Mitochondrial Fission Mediates Metabolic Fitness of Adipocytes through Maintaining Mitochondrial Bioenergetics and Lipid Metabolism.

In human WAT, the expression of adipose mitochondrial fission genes, including DNM1L, FIS1, and MFF, is as high as their expression in oxidative tissues, such as heart and muscle, implying potential roles of mitochondrial fission in WAT homeostasis (SI Appendix, Fig. S8 A–D). Nonetheless, the significance of mitochondrial fission in this context remained undefined. To explore the connection between mitochondrial fission and WAT homeostasis, we manipulated mitochondrial fission in vivo through pharmacological intervention. Intraperitoneal injection of Mdivi-1, a potent Drp1 inhibitor, resulted in excessive lipid accumulation, concomitantly with compromised mitochondrial membrane potential in WAT (SI Appendix, Fig. S9 A–C). We examined more directly the effect of Mdivi-1 on mitochondrial homeostasis in adipocytes. In vitro, treatment of adipocytes with Mdivi-1 induced abnormal mitochondrial homeostasis characterized by an irregular mitochondrial network and reduced mitochondrial membrane potential (SI Appendix, Fig. S9 D and E). Although a number of studies utilize Mdivi-1 as a Drp1 inhibitor, recent studies also report that Mdivi-1 would impact other pathways in addition to Drp1 activity (25, 26). To further validate the direct role of Drp1 in mitochondrial homeostasis in adipocytes, we transfected Dnm1l small interfering RNA into adipocytes. Similarly, Dnm1l knockdown in adipocytes also promoted aberrant mitochondrial network and mitochondrial depolarization (SI Appendix, Fig. S9 F–H). These in vivo and in vitro findings suggest that mitochondrial fission would be required for proper mitochondrial function and consequent lipid metabolism in adipocytes.

Next, we tested the hypothesis that if decreased Dnm1l-meidated fission would account for defective mitochondrial homeostasis and consequent adipocyte hypertrophy in Dnmt1 AKO mice, restoration of Dnm1l expression should rescue these defects. Indeed, restoration of Dnm1l expression in Dnmt1 AKO adipocytes mitigated excessive lipid accumulation, increased the mitochondrial membrane potential, and alleviated the irregular mitochondrial network (SI Appendix, Fig. S10 A–D). These findings suggest that defects in Dnmt1–Dnm1l axis would contribute to the observed mitochondrial dysregulation and metabolic abnormalities of Dnmt1 AKO adipocytes.

Adipocyte Dnmt1 Governs DNA Methylation, which Facilitates Chromosomal Looping Required for Dnm1l Expression, by Preventing Aberrant CTCF Bindings.

Next, we determined the mechanism of Dnmt1-dependent regulation of Dnm1l expression. In adipocytes, Dnm1l promoter was linked to three active enhancers, one proximal enhancer (E2) and two distal enhancers (E1 and E3, located ∼495 kb downstream and ∼498 kb upstream of the Dnm1l transcription start site, respectively), thereby suggesting pleiotropic effects of multiple enhancers on Dnm1l transcription (Fig. 5A). WGBS and locus-specific bisulfite sequencing (BS) analysis revealed that DNA methylation in E1 to ∼E3 and promoter of Dnm1l remained unchanged in Dnmt1 AKO adipocytes (SI Appendix, Fig. S11 A–D). Because this lack of difference did not comply with the main mode of the direct repression of regulatory elements by DNA methylation, and Dnm1l promoter was linked to two distal enhancers that would require chromosomal looping for exerting regulatory effects on Dnm1l promoter, we next investigated the effect of Dnmt1 deficiency on the interactions between Dnm1l promoter and two distal enhancers. Particularly, chromosomal looping linking enhancers and their targets that reside within the same topologically associating domain (TAD) have been shown to be strongly associated with a cell type–specific gene expression (12). Utilizing publicly available adipocyte Hi-C data revealed that Dnm1l locus appeared to be simultaneously controlled by two adjacent regulatory TADs located on the centromeric side (C-TAD) and telomeric side (T-TAD) of the Dnm1l locus (Fig. 5A). Also, we found that E1 and Dnm1l were located within C-TAD, whereas E3 and Dnm1l were positioned within T-TAD (Fig. 5A). The chromosome conformation capture (3C) assay further revealed that physical contacts of E1 with the Dnm1l promoter were decreased in Dnmt1 AKO adipocytes, whereas the frequency of E3-Dnm1l loops was comparable between the two genotypes (Fig. 5B). These findings suggest that reduced Dnm1l expression in Dnmt1 AKO adipocytes would be attributable to perturbation of the interaction between E1 and Dnm1l promoter.

Fig. 5.

Dnmt1-dependent DNA methylation is engaged in the stabilization of chromosome structure required for linking distal enhancer to Dnm1l promoter. (A) 3D representation of Hi-C chromosome conformation capture data from mouse adipocytes (ADs) are presented in the Dnm1l locus. Three enhancers (E1 to ∼E3) and the TADs are indicated as green rectangles and black bars, respectively. (B) 3C-qPCR data between the enhancers and the Dnm1l promoter in ADs. (C) Venn diagram showing potential sites of CTCF gain in Dnmt1 AKO ADs. (D) The number of potential sites of CTCF gain with and without CpG(s). (E) Boxplot of Δ CpG methylation for potential sites of CTCF gain with DMRs (Upper), arranged by ascending WT CpG methylation decile (Lower). (F) Genomic location of two candidate CTCF sites with Hypo-DMR in C-TAD in ADs. (G and H) View of DNA methylation level at genomic regions containing two candidate CTCF sites (purple rectangle). The green arrowhead indicates that these CTCF sites contain putative CTCF motifs with a CpG at a methylation-sensitive position estimated using the JASPAR algorithm. (I) Bisulfite sequencing analysis (Left) and quantification (Right) of a CpG dinucleotide (red arrow) in the CTCF site shown in G. Each row indicates sequencing results of each clone. Open circles and closed circles denote unmethylated CpGs and methylated CpGs, respectively. (J) Bisulfite sequencing analysis of a CpG dinucleotide in the CTCF site shown in H. (K) CTCF ChIP-qPCR for the two candidate CTCF sites in ADs. (L) Expression profiles of genes located within C- or T-TAD. (M) Heatmap indicating Pearson’s correlations of DNMT1 and DNM1L with genes located within C- or T-TAD from mouse ADs (ARCHS⁴). (N) Correlation analysis of mRNA levels (log2-transformed) of DNMT1 and DNM1L and genes within C- or T-TAD from Figure 5M. *P < 0.05 and **P < 0.01. Data are mean ± SEM.

Decreased DNA methylation is associated with a reoccurrence of CTCF binding at the sites that are often occupied in other cell types (27). The aforementioned findings, together with our observation that adipocyte DNA methylation was closely associated with CTCF binding landscape (Fig. 1 I–K), prompted us to speculate a role of CTCF in mediating the disruption of intra-TAD interactions by Dnmt1 deficiency. Specifically, we considered that Dnmt1 depletion would induce aberrant CTCF binding in sites where CTCF binding would be antagonized by preexisting methylation in WT adipocytes. To address this, we reanalyzed the merged mouse CTCF ChIP-seq data used in Fig. 1 and identified 187,865 CTCF sites that were bound in at least one cell type of the 29 analyzed mouse cell types but remained unbound in adipocytes (Fig. 5C). Overall, the majority (72.7%) of the 187,865 CTCF sites had more than one CpG (Fig. 5D), suggesting a potential for interaction between CTCF and CpG methylation. Additionally, by integrating WGBS data from two genotypes with these identified CTCF sites, we found that hypomethylation predominantly affected the CTCF sites that were heavily methylated in WT adipocytes (Fig. 5E). Therefore, the decrease in DNA methylation mediated by Dnmt1 depletion within the potential sites of CTCF gain was not randomly distributed but was dependent on the initial level of DNA methylation. Next, we scanned for the potential sites of CTCF gain within C-TAD in adipocytes. Among 27 potential CTCF sites in C-TAD, only CTCF sites (chr16:15,810,434 to 15,810,552 and chr16:16,239,055 to 16,239,245) displayed significantly lower DNA methylation in Dnmt1 AKO adipocytes (Fig. 5 F–H). Locus-specific BS analysis on these CTCF sites further confirmed hypomethylation in Dnmt1 AKO adipocytes (Fig. 5 I and J). JASPAR analysis indicated that the CpG with differential methylation in these CTCF sites was potentially located at one of the two positions previously linked to methylation sensitivity of CTCF (Fig. 5 G and H and SI Appendix, Fig. S11E). Furthermore, quantitative ChIP-PCR revealed that Dnmt1 AKO adipocytes acquired CTCF binding at these CTCF sites (Fig. 5K), supporting the potential linkage between defective intra-TAD interactions with hypomethylation-induced CTCF gain in C-TAD. In contrast, among 21 potential sites of CTCF gain within the T-TAD, none of these sites showed substantial difference in DNA methylation between the two genotypes. Additionally, we examined whether genes located within C-TAD would also be affected by defective intra-TAD interactions due to Dnmt1 deficiency. Gene expression profiling revealed that the majority of genes (five out of seven genes) in C-TAD exhibited lower expression in Dnmt1 AKO adipocytes, whereas expression of genes within T-TAD were not altered (Fig. 5L). Consistent with these data, DNMT1 messenger RNA (mRNA) levels of the mouse adipocytes from the ARCHS4 gene database (28) were also more strongly correlated with genes within C-TAD than genes within T-TAD, the highest correlation being with MCM4, SPIDR, and DNM1L (Fig. 5 M and N). Collectively, these concordant changes in the expression of genes within C-TAD imply that Dnmt1 depletion would promote the disruption of chromosomal topology, subsequently diminishing the regulatory interactions required for Dnm1l expression.

Discussion

Here, we demonstrate that adipocyte DNA methylation synergistically cooperates with cis-regulatory network and the chromosome structure to establish distinct gene expression profiles that confer metabolic features to adipocytes (SI Appendix, Fig. S12).

Previous systemic evaluations of DNA methylation landscape have identified numerous cell type–specific DMRs that reside mainly within noncoding regulatory regions of the genome (19). Since assigning of noncoding regulatory elements to target genes is not immediately apparent from systemic mapping of DNA methylation, identification of target genes and subsequent biological pathways of these previously identified cell type–specific DMRs has been challenging. Likewise, the majority of the adipocyte-specific DNA methylation signature was enriched in noncoding regions. To determine the functional role of these adipocyte-specific DMRs on gene regulation, we integrated adipocyte PCHi-C data and revealed the involvement of adipocyte-specific DMRs in adipocyte-specific distal enhancer activity whose target genes were closely linked to adipocyte biology. Our comparative epigenomic analyses further demonstrated that the sites in the genome containing adipocyte-specific CTCF occupancy and architecture might be predominantly affected by DNA methylation. These findings on enhancer activity and CTCF sites provide proof of concept that DNA methylation could render transcriptome features of adipocytes through simultaneously modulating cis-regulatory element activity and chromosome architecture. Indeed, one of the important findings in this study is that the expression of many adipocyte DEGs would be modulated by the converging effects of DNA methylation on multiple distal enhancers and their interactions with target genes. For instance, we found that expression of crucial adipose genes, such as PPAR-γ and CD36, entails concerted action of enhancer hypomethylation and strong interaction power. In contrast, collaboration of the strong interaction power with enhancers with strong hypermethylation significantly suppressed expression of genes (e.g., APOA4 and APOC3) that modulate liver-specific functions. Given that hypermethylation can suppress gene expression either by evicting TF binding or recruiting methylcytosine-specific repressive factors (29), further studies will be required to elucidate the involvement of TF eviction or repressive factors in the coordinated action of enhancer hypermethylation with strong interaction power for target gene suppression in adipocytes.

WAT is a remarkably flexible organ that expands both by hyperplasia and hypertrophy. Despite that the amount and distribution of WAT associate independently with metabolic disorders such as insulin resistance and T2D, compelling evidence suggests that the size of adipocytes within WAT is also crucial (4, 30). Moreover, clinical studies have suggested that hypertrophic WAT expansion is linked to reduced adipocyte turnover with impaired adipogenesis regardless of BMI (6, 31). Here, we found that Dnmt1 AKO mice manifest an increment of hypertrophic adipocytes in parallel with reduced adipocyte hyperplasia that would be attributable to diminished physiologic expansion of APs in WAT. This impaired WAT expansion consequently promotes systemic hyperlipidemia, glucose intolerance, and defective energy expenditure in Dnmt1 AKO mice. Also, DNMT1 expression in human adipose tissue is inversely correlated with adiposity and markers of metabolic health. In contrast, white adipose DNMT1 expression was positively correlated with human AP–specific markers whose expression was decreased in adipose tissue from T2D patients. Ex vivo coculturing further demonstrated that hypertrophic adipocytes induced by Dnmt1 deficiency are capable of inhibiting AP proliferations, implying potential roles of factors derived from adipocytes on hyperplastic potential of APs in Dnmt1 AKO WAT. These data suggest that adipocyte Dnmt1 would be a central factor controlling the balanced WAT expansion and metabolic phenotype of WAT as well as whole-body energy metabolism.

In the present study, Dnmt1 depletion in white adipocytes promoted mitochondrial dysfunction in WAT, whereas brown adipocyte Dnmt1 depletion failed to induce such changes in BAT. These differential effects of Dnmt1 depletion on mitochondria homeostasis could have resulted from several factors. First, the distinct maintenance mechanism for mitochondria homeostasis between white adipocyte and brown adipocyte may be attributed. For instance, mitochondrial biogenesis is constitutively active in brown adipocytes that express relatively higher levels of PGC-α, a key regulator of mitochondrial biogenesis, than white adipocytes (32, 33). Given that mitochondrial homeostasis is modulated by coordinated action of mitochondrial biogenesis and mitochondrial dynamics, the relative contribution of mitochondrial fission to the maintenance of mitochondrial quantity and quality would be lower in brown adipocytes than in white adipocytes. Therefore, this feature of brown adipocytes may counteract the effect of defective mitochondrial fission on mitochondrial quality and quantity control in Dnmt1 BKO mice. Second, while previous studies demonstrate the significance of Drp1-mediated fission in active brown adipocytes (34, 35), we only analyzed BAT function in adult Dnmt1 BKO mice housed at room temperature (22 °C). Because BAT in adult mice is reported to be relatively inert without additional challenges, such as cold acclimation and HFD, our observations suggest that Dnmt1 may not be involved in basal mitochondrial activity in BAT. Whether Dnmt1 is involved in the regulation of mitochondrial fission in active BAT requires further investigation.

Drp1 is a cytosolic protein recruited to mitochondria during fission, self-assembling into spirals that constrict and separate mitochondria into two daughter mitochondria. While changes in mitochondrial fission are likely mediated by the coordinated interplay among complex signaling cascades modulating transcription, protein stability, and translocation of Drp1, a majority of Drp1 studies have focused on posttranslational modifications (e.g., phosphorylation, S-nitrosylation, and SUMOylation) that affect protein stability and the interaction with factors that facilitate translocation. However, relatively few studies to date have addressed the regulatory pathways that modulate the transcription of Dnm1l in physiological and pathophysiologic states (36, 37). Here, we showed that Dnmt1-dependent DNA methylation is crucial for sustaining physiological level of Dnm1l expression in adipocytes. Mechanistically, while Dnmt1 was dispensable for methylation in promoter and enhancers of Dnm1l, hypomethylation in Dnmt1 AKO adipocytes affected potential CTCF sites and promoted aberrant CTCF binding, which impeded the chromosomal looping required for proper expression of Dnm1l. The positive correlation of Dnmt1 with Dnm1l transcripts in human and mouse gene databases further supports the present findings in Dnmt1 AKO mice. Collectively, this study provides evidence of an epigenetic mechanism that plays a key role in regulatory circuits that operate in the homeostatic expression of Dnm1l.

In humans, dominant negative mutations in the DNMT1 gene are linked to autosomal dominant cerebellar ataxia-deafness and narcolepsy (ADCA-DN) and hereditary sensory neuropathy with dementia and hearing loss (HSN1E) (38). Notably, ADCA-DN and HSN1E share clinical and biomedical features of mitochondrial disorders, such as optic atrophy and compromised ATP production, suggesting a functional relationship between Dnmt1 and mitochondrial homeostasis (38). However, comparative WGBS findings reveal comparable DNA methylation in the promoters of crucial genes required for mitochondria homeostasis (e.g., PGC1-α, DNM1L, and FIS1) between HSN1E and unaffected siblings despite locus-specific hypomethylation in the HSN1E siblings (39). By contrast, hypomethylation in HSN1E siblings was significantly colocalized with the repressive histone marks (H3K27me3 and H3K9me3) and CTCF (39). Nevertheless, genomic functions of such an association in linking Dnmt1 mutation to mitochondrial dysfunction have not been clarified. Given that DNA methylation signatures of Dnmt1 AKO adipocytes appeared to be reminiscent to HSN1E, and many of the DMRs induced in genetic diseases are frequently seen in more than one cell type, our findings may provide a rationale to test the relative contribution of Dnmt1-CTCF-Drp1 axis in the pathology of affected cells (e.g., sensory nerve cells and brain cells) of ADCA-DN and HSN1E.

Recently, the reversible nature of epigenetic modifications has attracted a great deal of attention in the development of epigenetic therapies in obesity and metabolic disease. Nevertheless, the use of epigenetic drugs has been limited due to the appearance of pleiotropic effects derived from the complexity of epigenome. To overcome this limitation, it is necessary to delineate precisely underlying mechanisms by which epigenetic modification affects cellular function and refine target specificity to fully exploit the clinical potential of epigenetic drugs. Here, we demonstrate that Dnmt1-dependent DNA methylation would be required for the delicate control of key biological pathways to prevent excess adiposity and related metabolic dysfunction. In contrast, a previous study with adipocyte-specific Dnmt3a KO mice suggest potentially negative effects of Dnmt3a-mediated DNA methylation on WAT function and insulin sensitivity independent of body weight and adiposity (40). Such differential effects of DNMT isoforms on adiposity and energy metabolism would be attributable to differences in target loci in the genome and downstream biological pathways. Also, these findings suggest that the use of epigenetic drugs targeting all DNMT isoforms to treat obesity and metabolic disorders may induce unintended off-target effects, possible excess adiposity, and systemic hyperlipidemia. By contrast, our study suggests that selective modulation of Dnmt1 action in adipose tissue would be an efficient therapeutic intervention for obesity and metabolic disorders that have minimal side effects.

Materials and Methods

Experimental details on animals, all sequencing method and analysis, metabolic phenotyping, immunoblotting, flow cytometry analysis, fatty acid oxidation assay, seahorse analysis, mitochondrial staining, ChIP-qpcr analysis, and statistical analysis for this study are described in detail in SI Appendix, Supplementary Methods.

Data Availability

WGBS and RNA-sequencing data have been deposited in Gene Expression Omnibus (GSE158944).

Change History

August 9, 2021: Figure 3A has been updated; please see accompanying Correction for details.