1. Cell Biology
Download icon

Dnmt3a is an epigenetic mediator of adipose insulin resistance

  1. Dongjoo You
  2. Emma Nilsson
  3. Danielle E Tenen
  4. Anna Lyubetskaya
  5. James C Lo
  6. Rencong Jiang
  7. Jasmine Deng
  8. Brian A Dawes
  9. Allan Vaag
  10. Charlotte Ling
  11. Evan D Rosen  Is a corresponding author
  12. Sona Kang  Is a corresponding author
  1. University of California, Berkeley, United States
  2. Lund University Diabetes Centre, Scania University Hospital, Sweden
  3. Beth Israel Deaconess Medical Center, United States
  4. Broad Institute of Harvard and MIT, United States
  5. Weill Cornell Medical College, United States
  6. Rigshospitalet, University of Copenhagen, Denmark
  7. AstraZeneca, Innovative Medicines, Sweden
Research Article
  • Cited 13
  • Views 2,368
  • Annotations
Cite this article as: eLife 2017;6:e30766 doi: 10.7554/eLife.30766

Abstract

Insulin resistance results from an intricate interaction between genetic make-up and environment, and thus may be orchestrated by epigenetic mechanisms like DNA methylation. Here, we demonstrate that DNA methyltransferase 3a (Dnmt3a) is both necessary and sufficient to mediate insulin resistance in cultured mouse and human adipocytes. Furthermore, adipose-specific Dnmt3a knock-out mice are protected from diet-induced insulin resistance and glucose intolerance without accompanying changes in adiposity. Unbiased gene profiling studies revealed Fgf21 as a key negatively regulated Dnmt3a target gene in adipocytes with concordant changes in DNA methylation at the Fgf21 promoter region. Consistent with this, Fgf21 can rescue Dnmt3a-mediated insulin resistance, and DNA methylation at the FGF21 locus was elevated in human subjects with diabetes and correlated negatively with expression of FGF21 in human adipose tissue. Taken together, our data demonstrate that adipose Dnmt3a is a novel epigenetic mediator of insulin resistance in vitro and in vivo.

https://doi.org/10.7554/eLife.30766.001

Introduction

Insulin resistance (IR) is generally defined as a reduced ability of insulin to provoke metabolic responses, such as glucose uptake. Several mechanistic drivers of IR have been proposed, including defects in the insulin signal transduction pathway, dysfunctional mitochondrial respiration, and ER stress (Boucher et al., 2014; Gregor and Hotamisligil, 2011; Houstis et al., 2006; Ozcan et al., 2004; Samuel and Shulman, 2012). The data supporting the involvement of these non-mutually exclusive pathways are convincing, but additional mechanisms are also likely to participate in the development of IR. Nuclear pathways in particular (i.e. transcriptional and epigenomic mechanisms) are likely to play an important role, given that IR often develops over a time course that is inconsistent with rapid signal transduction events (Kang et al., 2015; Tan et al., 2015). Furthermore, there is a large body of data from humans and rodents indicating that the propensity for IR can be passed transgenerationally; such ‘fetal programming’ studies strongly imply an epigenetic basis for IR (Carone et al., 2010; Heijmans et al., 2008; Li et al., 2010; Tobi et al., 2014). Finally, thiazolidinedione drugs, which activate the transcription factor PPARγ, provide one of the only pharmacological therapies for IR (Soccio et al., 2014; Step et al., 2014; Sugii et al., 2009). Taken together, these data strongly suggest that epigenomic and transcriptional mechanisms must be in play in the etiopathogenesis of IR (Kang et al., 2016).

DNA methylation is a reversible epigenetic mark involving the covalent transfer of a methyl group to the C-5 position of a cytosine residue, usually in the context of a CpG doublet, by DNA methyltransferases (Dnmts) (Robertson, 2005). Long thought to be a static epigenetic mark, emerging evidence suggests that methylated DNA undergoes dynamic and reversible remodeling in somatic cells during developmental, physiological, and pathogenic pathophysiological processes (Barres et al., 2013; 2009; Benner et al., 2015; Kirchner et al., 2016; Multhaup et al., 2015; Siersbæk et al., 2017). As an example, a bout of intense exercise can induce DNA hypomethylation at the PPARGC1a locus in human skeletal muscle within hours, followed by increased gene expression; these effects were reversed after resting (Barrès et al., 2012). Mounting evidence points to a role for DNA methylation in the pathogenesis of metabolic disorders. For example, the obesity-prone agouti mouse has reduced DNA methylation at the regulatory region for the Agouti gene, leading to enhanced expression and subsequent inhibition of the satiety-inducing MC4 receptor (Xie et al., 2013). Multiple studies have shown that changes in DNA methylation at key metabolic genes, such as Cox7a1 (Rönn et al., 2008), IGF-2 (Heijmans et al., 2008), and Pomc (Ehrlich et al., 2010), associate with various metabolic insults including aging, obesity, anorexia, and prenatal exposure to famine. Moreover, recent genome-wide profiling studies have identified distinct global DNA methylation patterns that associate with obesity and diabetes in humans (Dayeh et al., 2014; Feinberg et al., 2010; Kirchner et al., 2016; Multhaup et al., 2015; Nilsson et al., 2015; Volkmar et al., 2012; Volkov et al., 2017). Altered DNA methylation is linked to the transgenerational passage of metabolic disorders (Ng et al., 2010; Wei et al., 2014). Despite numerous studies showing a link between DNA methylation and metabolic dysregulation, the cause-and-effect relationship between the two remains largely unknown.

In mammals, five Dnmt family members have been identified: Dnmt1, Dnmt2 (also known as Trdmt1), Dnmt3a, Dnmt3b and Dnmt3l, yet only three (Dnmt1, 3a, and 3b) possess DNA methyltransferase activity (Robertson, 2005). Dnmt1 (the maintenance DNA methyltransferase) has a preference to methylate hemimethylated DNA whereas Dnmt3a and 3b prefer unmethylated DNA as substrate, and thus act as de novo DNA methyltransferases (Robertson, 2005). Dnmt3l is homologous to the other Dnmt3s but lacks catalytic activity (Robertson, 2005) and Dnmt2 has sequence homology to all Dnmts but methylates cytoplasmic tRNA, and not DNA (Robertson, 2005). Administration of a small molecule Dnmt inhibitor has been shown to improve insulin sensitivity in the setting of obesity, in part by demethylating the Adipoq promoter, and a role for Dnmt1 in this process was suggested by genetic knockdown studies in cultured adipocytes (Kim et al., 2015).

Here, we identify Dnmt3a as a key epigenetic determinant of obesity-associated IR in adipose tissue. Dnmt3a plays a causal role in the development of cell autonomous IR in mouse and human adipocytes in a manner that is dependent on its catalytic activity. Consistent with results from adipocytes in vitro, we find that adipose-specific Dnmt3a deficiency in vivo confers protection from diet-induced IR and glucose tolerance without accompanying changes in body weight or adiposity. In addition, unbiased gene expression profiling suggests that Dnmt3a acts as primarily as a gene repressor, and reveals a set of adipocyte-specific target genes, including Fgf21. Further, we demonstrate that Dnmt3a mediates IR, at least in part, by methylating specific cis-regulatory elements in the Fgf21 gene and thus suppressing its expression. We also provide evidence that similar mechanisms might be at play in human adipose tissue. These studies identify Dnmt3a as a novel epigenetic mediator of IR in adipocytes.

Results

Adipose expression of Dnmts inversely correlates with insulin sensitivity

To identify a potential role for Dnmts in the pathogenesis of insulin resistance, we compared their expression levels in insulin-sensitive metabolic tissues (i.e. liver, muscle, and adipose depots) in various metabolic settings. In a diet-induced obesity (DIO) mouse model, Dnmt1 and Dnmt3a mRNA expression was elevated in adipose tissue, especially in epididymal and mesenteric WAT (Figure 1A, Figure 1—figure supplement 1A). Dnmt3a expression was significantly induced by HFD in skeletal muscle, but to a lesser degree than in adipose tissue. No effect of HFD was observed in liver (Figure 1—figure supplement 1B–C). In genetically obese ob/ob mice, mRNA expression of all three functional Dnmts was elevated in adipose tissue (Figure 1B). Importantly, expression of these genes was significantly reduced after treatment with the insulin sensitizer rosiglitazone (Rosi) (Figure 1B). Increased adipose expression of Dnmt3a was confirmed by immunoblotting in DIO mice (Figure 1C,D). Relative expression of Dnmt3a was higher in eWAT compared to skeletal muscle and liver compared in normal chow-fed mice (Figure 1—figure supplement 1D–E). Dnmt2 and Dnmt3l mRNA displayed minimal expression in adipose tissue (not shown). Together, this led us to hypothesize that Dnmts may have a direct functional role in the pathogenesis of adipose IR.

Figure 1 with 1 supplement see all
Increased adipose expression of Dnmts in obesity.

(A) Dnmt mRNA expression in adipose from chow- and high fat-fed (HFD) C57Bl/6 mice (A); n = 6 mice, p<0.05, Student’s t-test, mean ±s.e.m.) and (B) in adipose tissue from ob/+and ob/ob mice in the presence and absence of ​rosiglitazone treatment (right; n = 7 for ob/+, n = 8 for ob/+plus Rosi, n = 7 for ob/ob with and without Rosi, p<0.05, Student’s t-test, mean ±s.e.m.). (C) Western blot showing Dnmt3a protein levels in adipose tissue from chow vs. HFD mice. (D) Quantification of western blot in (C).

https://doi.org/10.7554/eLife.30766.002

Knock-down of Dnmt3a confers protection from drug-mediated IR

Given the inverse correlation of adipose Dnmt expression with insulin sensitivity, we sought to test the cell autonomous role of Dnmts in the development of IR. To that end, individual Dnmts were knocked-down in mature 3T3-L1 adipocytes using two distinct short hairpin RNAs (shRNAs) (Figure 2—figure supplement 1A–F). To promote IR, cells were treated with dexamethasone (Dex) and TNF-α (TNF) (Houstis et al., 2006; Kang et al., 2015) for an additional four days. Knock-down of individual Dnmts did not have a major impact on the basal level of insulin sensitivity when assessed by insulin-stimulated glucose uptake (Figure 2—figure supplement 2A–E). However, knock-down of Dnmt3a, but not Dnmt1 or Dnmt3b, protected adipocytes from Dex- and TNF-mediated IR as assessed by insulin-stimulated glucose uptake and signal transduction (Figure 2A,B, Figure 2—figure supplement 2A–E). Given the elevated expression of adipose Dnmts in obesity, we were prompted to test whether overexpression of Dnmts per se is sufficient to drive IR in adipocytes. Individual Dnmts were expressed in mature 3T3-L1 adipocytes by lentiviral delivery. Interestingly, overexpression of Dnmt3a and 3b, but not Dnmt1, potently inhibited insulin-stimulated glucose uptake (Figure 2C). Additionally, we tested whether the causal role of Dnmt3a in the development of insulin resistance is conserved in human adipocytes. Overexpression of Dnmt3a powerfully inhibited insulin-stimulated glucose uptake in in vitro differentiated adipocytes from human adipose derived stem cells (Figure 2D). Together, our gain- and loss-of-function studies in vitro show that Dnmt3a is both necessary and sufficient to mediate IR in adipocytes in a cell autonomous manner.

Figure 2 with 2 supplements see all
Dnmt3a is sufficient and necessary to mediate cell autonomous IR in vitro.

(A) Two independent hairpins against ​Dnmt3a (versus scrambled control shRNA; shScr) were delivered to mature 3T3-L1 adipocytes (Day 8) via lentiviral transduction. Cells were then treated with ​Dex and ​TNF for 4 days and assessed for insulin-stimulated ​glucose uptake (3H-2-DG assay, n = 6, p<0.05). Shown is the % of insulin-stimulated glucose uptake rescued by Dnmt3a knockdown (n = 6, p<0.05). (B) Cells in (A) with #2 shDnmt3a hairpin and control cells were withdrawn from serum for 6 hr, stimulated with 20 nM insulin for 5 min and subjected to immunoblotting with antibodies against total and phospho-AKT and –IRS-1 with and without insulin. (C) Basal and insulin-stimulated glucose uptake in mature 3T3-L1 and (D) in in vitro-differentiated primary human adipocytes transduced with lentivirus expressing Dnmts or the empty plasmid vector, pCDH (n = 6, p<0.05, Student’s t-test, mean ±s.e.m.).

https://doi.org/10.7554/eLife.30766.006

DNA methyltransferase activity is critical for Dnmt3a to mediate IR

Studies have shown that Dnmt3a can act as a gene repressor using both catalytic-dependent and independent mechanisms (Bachman et al., 2001; Fuks et al., 2001). Thus, we sought to determine whether DNA methyltransferase activity is required to mediate IR. We initially pursued a pharmacological approach, using two synthetic Dnmt inhibitors, 5-azacytidine (5-Aza) (Stresemann and Lyko, 2008), a nucleoside analogue serving as a pseudo-substrate for Dnmts, and RG-108 (Brueckner et al., 2005), a non-nucleoside-based drug. Remarkably, both Dnmt inhibitors blocked Dex- and TNF-mediated IR (Figure 3A,B, Figure 3—figure supplement 1A,B). In addition to these synthetic inhibitors, the naturally occurring Dnmt inhibitors (Rajavelu et al., 2011) epigallocatechin gallate (ECCG) and theaflavin-3,3′-digallate (TF-3) also showed insulin-sensitizing activity in the setting of Dex- and TNF-mediated IR (Figure 3—figure supplement 1C–F). These data suggest that DNA methyltransferase activity is necessary for Dnmt3 to cause IR. To further probe this hypothesis, we overexpressed a catalytically inactive allele of Dnmt3a (C706S) (Hsieh, 1999). Unlike the wild-type allele, Dnmt3a C706S was unable to suppress insulin-stimulated glucose uptake (Figure 3C). Together, the pharmacological and genetic evidence indicates that DNA methyltransferase activity is essential for Dnmt3a to mediate IR.

Figure 3 with 1 supplement see all
DNA methyltransferase activity is critical for Dnmt3a to mediate IR.

Insulin-stimulated glucose uptake assay was conducted using 3T3-L1 adipocytes treated with Dnmt inhibitors (A, 5-Aza; B, RG-108) and Dex or TNF for 4 days. Shown is the % of insulin-stimulated glucose uptake rescued by Dnmt inhibitors (n = 6, p<0.05). (C) Basal and insulin-stimulated glucose uptake in mature 3T3-L1 adipocytes transduced with wild-type Dnmt3a vs. a catalytically inactive allele (C706S) (n = 6, p<0.05, Student’s t-test, mean ±s.e.m.).

https://doi.org/10.7554/eLife.30766.012

Adipose-specific Dnmt3a deficiency confers protection from diet-induced insulin resistance and glucose intolerance

We next sought to determine whether loss of Dnmt3a in adipocytes could confer protection from diet-induced adipose tissue and systemic IR in vivo. To address this, we generated adipose-specific Dnmt3a knock-out mice (Dnmt3aAdiKO) by crossing Dnmt3a floxed mice (Dnmt3af/f) (Nguyen et al., 2007) to adiponectin-Cre mice (Eguchi et al., 2011) (Figure 4—figure supplement 1A,B). On chow diet, male WT and Dnmt3aAdiKO animals did not show any obvious change in their body weight, insulin sensitivity, glucose tolerance, or insulin levels (Figure 4—figure supplement 2A–D). On HFD (45% fat), Dnmt3aAdiKO mice still showed no significant differences in body weight or composition relative to littermate controls (Figure 4A,B). Moreover, the effect of long-term fasting (16 hr) on body weight or body composition did not differ between genotypes (Figure 4C), suggesting that gross energy homeostasis is not altered in the knock-out animals. Despite no obvious change in energy balance, high fat fed male Dnmt3aAdiKO mice exhibited improved glucose tolerance (Figure 4D,E) and insulin sensitivity (Figure 4F,G) with reduced serum insulin in the fed state (Figure 4H). Consistent with their metabolic improvement, male Dnmt3aAdiKO mice display increased insulin signal transduction in adipose tissue, as demonstrated by increased insulin-stimulated phosphorylation of IRS and AKT (Figure 4I,J). Similarly, female Dnmt3aAdiKO mice display improved insulin sensitivity and glucose tolerance on HFD without a change in body weight (Figure 4—figure supplement 3A–C). We next assessed whether insulin sensitization could be due to either enhanced browning or reduced inflammation of WAT. With regard to the former, we found no consistent change in the expression of major thermogenic genes such as Ucp1 and Ppargc1a in either iWAT and BAT (Figure 4—figure supplement 4A,B). Additionally, gene expression analysis showed a nonsignificant trend toward reduced expression of proinflammatory genes such as Tnfa and Ccl2 (which encodes MCP-1) in the eWAT of Dnmt3aAdiKO mice (Figure 4—figure supplement 5A). We also saw no obvious change in the density of crown-like structures between WT and KO tissues (Figure 4—figure supplement 5B). Together, these results demonstrate that adipose-specific deletion of Dnmt3a confers protection from diet-induced IR in a manner that is independent of adiposity, inflammation, or browning of white adipose tissue.

Figure 4 with 5 supplements see all
Adipose-specific Dnmt3a KO mice display improved whole body insulin sensitivity and glucose tolerance on HFD.

(A) Body weight of Dnmt3aAdiKO and control mice on HFD (n = 7, 9, p<0.05, Student’s t-test, mean ±s.e.m.). (B) Body composition after 10 weeks of HFD (n = 7, 9). (C) Effect of long-term fasting (16 hr) on body weight or body composition by EchoMRI. Shown is % loss after fasting (n = 14, 14, p<0.05, Student’s t-test, mean ±s.e.m.). (D) Glucose tolerance test after 9 weeks on HFD (n = 7, 9, p<0.05, Student’s t-test, mean ±s.e.m.). (E) The area under curve (AUC) in (D). (F) Insulin tolerance test after 11 weeks on HFD (n = 7, 9, p<0.05, Student’s t-test, mean ±s.e.m.). (G) AUC in (F). (H) Fed and fasted insulin levels measured by ELISA (n = 7, 9 p<0.05, Student’s t-test, mean ±s.e.m.) (I) Immunoblot of total and phospho-AKT and IRS-1 in eWAT after IP insulin. (J) Quantification of western blot in (I).

https://doi.org/10.7554/eLife.30766.016

Unbiased gene profiling studies reveal that Fgf21 is a key metabolic target gene of Dnmt3a in adipocytes

Dnmt3a affects developmental processes by regulating non-overlapping sets of cell type-specific target genes (Challen et al., 2011; Nguyen et al., 2007; Nishikawa et al., 2015). To gain mechanistic insight into Dnmt3a-mediated IR at the molecular level, we performed RNA-seq in 3T3-L1 adipocytes, in both gain- and loss-of-function contexts. We have successfully used this approach to enhance our ability to identify bona fide targets (Kang et al., 2012). We found that 97 genes were repressed after shRNA-mediated Dnmt3a knock-down, while only eight genes were up-regulated in the presence of excess Dnmt3a; the single gene showing concordant regulation was Dnmt3a itself (Figure 5A). Conversely, 290 genes were induced by Dnmt3a knock-down, and 36 genes showed diminished expression after Dnmt3a overexpression; seven genes were concordantly regulated by both manipulations (Figure 5B and C and Supplementary file 1). These data are consistent with the expectation that Dnmt3a works primarily as a repressor of gene expression. Among the concordantly-regulated negative target genes of Dnmt3a, we were immediately struck by the presence of Fgf21. FGF21 is known to facilitate glucose uptake in adipocytes (Ge et al., 2011; Lee et al., 2014; Minard et al., 2016), and is a positive target gene of PPARγ (Muise et al., 2008). Furthermore, mice lacking FGF21 do not support Rosi-mediated insulin sensitization in vivo (Dutchak et al., 2012). We confirmed by Q-PCR that Fgf21 expression is indeed reduced in cultured adipocytes overexpressing Dnmt3a, is increased in cells expressing shDnmt3a, and most importantly, is up-regulated in the adipose tissue of Dnmt3aAdiKO mice (Figure 5D). Unlike adipose tissue levels, serum levels of Fgf21 were reduced in HFD-fed Dnmt3aAdiKO mice compared to controls (Figure 5—figure supplement 5–). To determine whether restoration of FGF21 levels could ameliorate Dnmt3a-mediated IR, we treated 3T3-L1 adipocytes that overexpress Dnmt3a with recombinant FGF21, which restored insulin sensitivity in a dose-dependent manner (Figure 5E).

Figure 5 with 1 supplement see all
Fgf21 is a key target gene of Dnmt3a.

Venn diagram showing the number of positive (A) and negative (B) Dnmt3a target genes through comparative analysis of RNA-seq profiles from Dnmt3a knockdown and overexpressing adipocytes. (C) Heat map showing differentially regulated genes by Dnmt3a overexpression and knock-down. Group 1: Genes that are down-regulated by Dnmt3a overexpression; Group 2: Genes that are down-regulated by Dnmt3a overexpression and up-regulated by Dnmt3a knock-down; Group 3: Genes that are up-regulated by Dnmt3a knock-down. (D) Fgf21 mRNA expression in Dnmt3a overexpressor (left) and knock-down (middle) L1 adipocytes and in fractionated adipocytes (right) from Dnmt3aAdiKO and Dnmt3af/f mice on HFD (n = 5). (E) Mature 3T3-L1 adipocytes expressing GFP or Dnmt3a treated with indicated amount of recombinant Fgf21 for 48 hr (or vehicle), and tested for insulin sensitivity by 3H-2-DG assay. Shown is basal and insulin-stimulated glucose uptake. (n = 6 mice, p<0.05, Student’s t-test, mean ±s.e.m.).

https://doi.org/10.7554/eLife.30766.028

Dnmt3a-mediated changes in Fgf21 gene expression involve changes in DNA methylation

In general, the level of DNA methylation at gene promoters is inversely correlated with mRNA expression (Jones, 2012). Given the functional significance of FGF21 in Dnmt3a-mediated IR, we further examined whether Dnmt3a directly alters methylation of the 5 kb upstream promoter regions of the Fgf21 gene (Figure 6A). Using methylated DNA precipitation PCR (MeDIP-qPCR), we detected genomic DNA hypermethylation at Fgf21 promoter regions in Dnmt3a overexpressor 3T3-L1 cells compared to controls (Figure 6B). Conversely, DNA hypomethylation of these same regions was seen in Dnmt3a knock-down cells (Figure 6B). In addition to the in vitro models, we conducted MeDIP-qPCR analysis using genomic DNA from fractionated adipocytes of Dnmt3aAdiKO and control animals, finding reduced DNA methylation at the Fgf21 promoter in KO cells (Figure 6C). Consistent with DNA methylation pattern at the Fgf21 promoter region, co-transfection of Dnmt3a potently inhibited promoter activity of Fgf21 (Figure 6D). Together, these data suggest that Dnmt3a-mediated suppression of Fgf21 involves changes in DNA methylation profile at key promoter regions.

Dnmt3a-mediated methylation of the Fgf21 gene promoter.

(A) Schematic showing the tested regions for MeDIP-qPCR at the Fgf21 locus. Target regions were chosen using MethFinder (http://www.urogene.org/methprimer) on regions proximal (5 kb) and distal (50 kb) to the Fgf21 transcription start site. (B) MeDIP-qPCR was performed using genomic DNA extracted from 3T3-L1 adipocytes transduced with GFP vs. Dnmt3a or shScr vs. shDnmt3a. Shown is relative fold enrichment of 5mC of Dnmt3a overexpressor and KD cells over control cells. (C) MeDIP-qPCR was performed using genomic DNA from Dnmt3a KO and Flox mice on HFD. Shown is relative fold enrichment of 5mC of KO over WT samples. (D) Fgf21-promoter reporter (1497/+5) assay performed by co-transfection of the indicated amounts of Dnmt3a expression vector. (n = 3 mice, p<0.05, Student’s t-test, mean ±s.e.m.).

https://doi.org/10.7554/eLife.30766.032

DNA methylation of FGF21 is increased in human adipose tissue of subjects with type 2 diabetes

To translate some of our findings to humans, we investigated DNA methylation and gene expression of FGF21 in human adipose tissue collected from diabetic vs. non-diabetic human subjects. In line with our mouse data, we found hypermethylation of four CpG sites annotated to FGF21 in adipose tissue of subjects with type 2 diabetes compared with non-diabetic controls (Figure 7A and Supplementary file 2). Further, we found a negative correlation between DNA methylation and mRNA expression of FGF21 in human adipose tissue (Figure 7B–E). Together, these data support a role for increased DNA methylation of FGF21 in the regulation of adipose FGF21 expression in association with metabolic disease.

FGF21 DNA methylation in adipose tissue from subjects with T2D compared with controls.

(A) Four CpG sites annotated to FGF21 with significant difference in adipose tissue DNA methylation between subjects with type 2 diabetes and controls. (p<0.01, paired Wilcoxon statistics, mean ±s.d.). (B–D) Correlations between DNA methylation and mRNA expression for CpG sites significantly associated with T2D in human adipose tissue. (r = Pearson correlation coefficient.).

https://doi.org/10.7554/eLife.30766.034

Discussion

While genetics plays an important role in obesity and T2D, genetic differences cannot fully explain many features of these conditions, such as discordance in monozygotic twins, and the close relationship with lifestyle factors (McCarthy, 2010; Morris et al., 2012; Voight et al., 2010; Zeggini et al., 2008). Of note, the vast majority of GWAS-proven allelic variants associated with T2D correspond much better to altered insulin secretion and islet function rather than to IR per se (Voight et al., 2010). Hence, forms of non-genetic variation, such as epigenetic alterations, must also be considered. This notion has been borne out by a recent epigenome-wide association study (EWAS) linking alterations in DNA methylation to whole-body insulin sensitivity (Zhao et al., 2012). Importantly, even genetic variants may cause disease in a DNA methylation-dependent manner, as has recently been shown for T2D and obesity (Dayeh et al., 2013; Elliott et al., 2017; Nilsson et al., 2014; Olsson et al., 2014; Volkov et al., 2016). We have also shown that adipose tissue from patients with T2D exhibit numerous methylation and expression differences relative to unrelated healthy controls, with many of the same changes also seen in twin pairs discordant for T2D (Nilsson et al., 2014). In this study, we identified increased expression of an enzymatic effector of DNA methylation, Dnmt3a, as a key epigenetic driver of IR in vitro and in vivo.

DNA methylation, like other forms of epigenomic modification, has been an attractive therapeutic target because of their plasticity and because they offer an opportunity to reprogram cells into a more healthy state. For example, administration of pan-inhibitors of histone deacetylase exerts beneficial metabolic effects in both mice and humans, such as increased energy expenditure, insulin sensitivity and secretion (Christensen et al., 2011; Daneshpajooh et al., 2017; Sharma and Taliyan, 2016; Ye, 2013). DNA methyltransferase inhibitors have also been shown to have an insulin-sensitizing effect in vivo and in vitro (Kim et al., 2015). In line with pharmacological studies, mutant mouse models carrying loss-of function histone modifiers (e.g. Mll2, Ehmt1, Jmhd2a Lsd1) were shown to have profound impact on whole body metabolism (Lee et al., 2006; Ohno et al., 2013; Sambeat et al., 2016; Tateishi et al., 2009; Zeng et al., 2016). It should be noted that metabolic consequences in mouse models carrying mutations of histone modifiers appear to be mainly attributable to changes in body weight. Unlike these examples, adipose-specific Dnmt3a KO mice are relatively unique in that they display improved insulin sensitivity without a change in adiposity or body weight. Similarly, the effect on insulin action is not dependent on increased beige or brown adipocytes, or on reducing intra-adipose inflammation.

Our in vitro study indicates that overexpression of Dnmt3a is sufficient to induce cell autonomous IR, however, it does not seem to be the case in vivo. Kamei et al showed that transgenic expression of Dnmt3a per se was not sufficient to render whole body IR and glucose intolerance (Kamei et al., 2010) on chow, HFD, or high methyl diet with chow diet. Authors observed only a mild increase in the gene expression of proinflammatory cytokines (i.e. Mcp1 and Tnfa) when animals were put on HFD. It should be noted that the major caveat of this study is the use of aP2 promoter to drive the transgene expression (Jeffery et al., 2014; Kang et al., 2014; Lee et al., 2013); Potentially, non-adipose expression and earlier expression during development might have a compounding factor in this study. Despite these notions, it will be highly interesting to address how the combination of high methyl donor diet and HFD would affect the metabolic profile of these transgenic animals compared to controls.

Our unbiased gene expression profiling studies led us to identify Fgf21 as a key target gene in adipocytes, and the data from diabetic humans suggests that this could be a conserved mechanism of IR in both mice and humans. Fgf21 is better known as a hepatokine, and the liver is the main source of circulating Fgf21 (Markan et al., 2014) Adipose Fgf21 probably does not circulate, but rather acts in an autocrine/paracrine manner (Dutchak et al., 2012; Markan et al., 2014). In fact, levels of Fgf21 in serum were moderately reduced in Dnmt3a KO animals compared to controls, in contrast to significantly elevated adipose tissue levels. Studies have shown a positive correlation between insulin and circulating Fgf21 levels (Chavez et al., 2009; Chen et al., 2008; Zhang et al., 2008), suggesting that insulin may be a key determinant of Fgf21 in serum. Thus, it is possible that the reduced insulin levels in Dnmt3a KO mice on HFD may have led to reduction of Fgf21 production in liver. This notion is in line with the finding that Fgf21 serum levels are often elevated in obesity-associated IR (Fisher and Maratos-Flier, 2016; Morrice et al., 2017), possibly reflecting resistance to Fgf21 (Fisher et al., 2010).

We propose that the local activity of Fgf21 within the fat pad is important for the improved insulin sensitivity in the knock-out mice, although formal proof of this idea would require generating adipose tissue-specific Dnmt3a and Fgf21 double knockout mice. In line with this view, there are previous studies indicating that adipose Fgf21 is insulin-sensitizing. For example, three independent studies demonstrated that mutant mice devoid of Fgf21 signaling in adipose tissue secondary to ablation of FGFR1 or bKlotho, were refractory to the insulin-sensitizing effect of Fgf21 administration (Adams et al., 2013; BonDurant et al., 2017; Ding et al., 2012). Some of these studies (Adams et al., 2013; Ding et al., 2012) suggested that adiponectin is an essential mediator of Fgf21 metabolic action; the other study (BonDurant et al., 2017) suggested that adiponectin is dispensable. In our studies, albeit it did not reach statistical significance, there was a trend toward increased mRNA expression of Adipoq in both L1 and primary Dnmt3a KO adipocytes (not shown).

Control of Fgf21 expression appears to be highly tissue-specific. For example, PPARγ is an important driver of Fgf21 in adipose tissue, while PPARα serves that role in liver (Dutchak et al., 2012). Other studies have also demonstrated that Fgf21 is positively regulated by PPARγ in fat (Muise et al., 2008; Wang et al., 2008). In addition to focused studies on Fgf21 promoter regions, global DNA methylation profiling will be necessary to identify differentially methylated regions in the setting of Dnmt3a gain- and loss-of function.

A recent study demonstrated that Dnmt1 may contribute to obesity-associated inflammation and insulin resistance by promoting hypermethylation at the Adipoq locus (Kim et al., 2015). Concordant with this study, we also see elevated Dnmt1 levels in the adipose tissue of obese animals. However, knockdown of Dnmt1 did not rescue insulin sensitivity in the setting of Dex or TNF, and overexpression of Dnmt1 did not reduce insulin-stimulated glucose uptake in cultured adipocytes. Based on those criteria, we chose to focus on Dnmt3a, and showed that adipose-specific deletion of this gene promoted insulin sensitivity even in weight-matched obese animals. Of note, we did not identify Adipoq as a regulated target after either deletion or overexpression of Dnmt3a.

Despite some effort, we have not yet identified the mechanism by which Dnmt3a expression is increased in obesity. We found that Dnmt3a is not induced by lipids (e.g. palmitate), pro-inflammatory signals (e.g. IL-6, LPS), high doses of insulin, or high glucose in cultured adipocytes (not shown). In our study, in vivo administration of thiazolidinediones normalized the expression of Dnmt3a in ob/ob mice but not in lean control animals. The mechanism by which PPARγ represses gene expression is not entirely clear. Trans-repression through binding of inflammatory transcription factors like NF-κB and AP-1 has been proposed as a general mechanism for this type of effect (Glass and Saijo, 2010) although we note that there are PPARγ binding sites near the Dnmt3a TSS (not shown). It is thus unclear if direct or indirect (i.e. trans-repression) mechanisms account for the actions of rosiglitazone on Dnmt3a expression in obesity.

In summary, we have identified a novel role for Dnmt3a as an epigenetic mediator of adipose IR in vitro and in vivo. Future studies will be required to fully elucidate the full range of its target genomic regions and metabolic effects.

Materials and methods

Cell culture

3T3-L1 preadipocytes were obtained from ATCC and maintained and differentiated as described (Kang et al., 2012). These cells were authenticated by the ability to differentiate and confirmed to be mycoplasma negative. To generate lentivirus particles, lentiviral constructs were co-transfected with pM2DG- and psPAX-expressing plasmids into 293T cells. After 48 hr, virus-containing supernatant was collected, filtered through 0.45 μm filters, and added to mature 3T3-L1 adipocytes for 24 hr along with 8 μg/ml Polybrene. Transduction efficiency was determined by comparison with cells transduced in parallel with a GFP-expressing lentivirus. For the ex vivo system, subcutaneous adipose tissue from wild-type C57Bl/6 mice was fractionated with digestion buffer (10 mg/ml collagenase D, 2.4 units of dispase II, 10 mM ​CaCl2 in PBS). Cells from the stromal-vascular fraction (SVF) were plated in culture and differentiated as described previously (Qureshi et al., 2014). ​To differentiate hASCs, Rosi (5 µM) was added in addition to IBMX (0.5 mM), insulin (5 µg/ml), and dexamethasone (1 µM), from day 0–2.

3H-2-DG assay

Mature 3T3-L1 adipocytes were incubated in serum-free DMEM for 4–6 hr. Cells were then washed three times with KRH buffer (12 mM HEPES, pH 7.4, 121 mM NaCl, 5 mM KCl, 0.33 mM CaCl2, and 1.2 mM MgSO4) and incubated for 20 min in KRH buffer in the absence or presence of 50 nM insulin. Cells were treated with 2-deoxy-d-[2,6-3H]-glucose (0.33 μCi/ml) for another 10 min. Glucose uptake was stopped quickly by three rapid washes with KRH buffer containing 200 mM glucose and 10 μM cytochalasin B on ice. Cells were solubilized in 0.1% SDS for 30 min, and radioactivity was measured by liquid scintillation counting.

Reagents

2-Deoxy-d-[2,6-3H]-glucose was purchased from PerkinElmer NEN radiochemicals. Insulin, dexamethasone, isobutylmethylxanthine (IBMX), cytochalasin B, glucose, 2-deoxyglucose, 5-Aza, ECCG, and TF-3 were purchased from Sigma-Aldrich. Recombinant TNF-α and Fgf21 were purchased from Millipore. RG-108 was purchased from Stemgent.

Antibodies

Antibodies were purchased from Cell Signaling (Dnmt3a, 3598S; Akt, 9272; pAkt [S473], 3787; IRS-1, 2382), and from Thermo Fisher (β-actin, MA5-14739, pIRS1[pY612], 44–816G).

Animals

For ​rosiglitazone studies, female mice ob/ob and ob/+were treated starting at 8 weeks of age with ​rosiglitazone by gavage (10 mg/kg body weight) in 0.5% carboxymethylcellulose versus 0.5% carboxymethylcellulose vehicle control daily for 6 weeks. For chow and high-fat feeding studies, male C57Bl/6J mice were put on diet beginning at 8 weeks of age and continued for three months (n = 8 per dietary condition). Samples were collected from the perigonadal fat pad.

For histology, adipose tissues were fixed with neutral-buffered formalin and embedded in paraffin and sections were stained with H&E.

For in vivo insulin signaling assay, after overnight fast, insulin (10 U/kg, 5 min) or saline IP were given to WT and KO mice on HFD. After 10 min various tissues were harvested and stored at –80°C until use. Tissue samples were homogenized in cell signaling lysis buffer containing protease inhibitors (Roche) and phosphatase inhibitors (Sigma-Aldrich) and subjected to western blotting. All animal work was approved by the BIDMC IACUC and/or the UC Berkeley ACUC.

RNA extraction and quantitative PCR

Total RNA was extracted from cells or tissues using TRIzol reagent according to the manufacturer’s instructions. cDNA was reverse-transcribed from 1 μg of RNA using the RETROscript first strand synthesis kit (Ambion). Quantitative PCR (qPCR) was performed with SYBR Green qPCR Master Mix (Applied Biosystems) using a 7900HT Fast Real-Time PCR System (Applied Biosystems) and CFX96 Touch (Bio Rad). Primer sequences are listed in Supplementary file 3. The relative amount of mRNA normalized to ​cyclophilin B was calculated using the delta–delta method (Livak and Schmittgen, 2001).

Plasmids

Hairpins against Dnmt1, Dnmt3a, and Dnmt3b were purchased from Sigma. Lentiviral overexpression vectors for Dnmt1, 3a, and 3b were subcloned into pCDH using various multicloning sites (XbaI/NotI for ​Dnmt1, EcoRI/NotI for Dnmt1, Dnmt3a, Dnmt3a-CM) and NotI sites and hairpins targeting ​Dnmts were subcloned at AgeI/EcoRI or purchased from Open Biosystems. Hairpin sequences are shown in Supplementary file 3. Fgf21-luc (1497/+5) was generously provided by Dr. Steven Kliewer (UT Southwestern).

MeDIP-qPCR

Genomic DNA was sheared by sonication to an average of 200–800 bp size. Two microgram of denatured DNA was incubated with 2 μg of anti-5-methylcytidine antibody (Epigentek) in IP buffer (10 mM Na-Phosphate pH 7.0, 0.14 M NaCl, 0.05% Triton X-100) for 2 hr at 4°C. Antibody-bound DNA was collected with 20 μl of Dynabeads anti-mouse IgG (Invitrogen Dynal, Oslo, Norway) for 1 hr at 4°C on a rotating wheel and successively washed five times with washing buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl pH 8.1, 150 mM NaCl), and twice with TE (10 mM Tris·Cl, 1 mM EDTA pH 8.0). DNA was recovered in digestion buffer 125 μl buffer (50 mM Tris pH 8.0, 10 mM EDTA, 0.5% SDS, 35 μg proteinase K) and incubated for 3 hr at 65°C. Recovered DNA was used for qPCR analysis. Primers for MeDIP-qPCR studies are listed in Supplementary file 3. All data are normalized to input.

RNA-seq studies

mRNA was purified from 100 ng of total RNA using the Ribo-Zero Magnetic Gold Kit (catalog MRZG126, Illumina). Libraries were prepared using the TruSeq RNA Library Preparation Kit v2 (catalog RS-122–2001, Illumina) according to the manufacturer’s protocol starting with the EPF step. Sequencing was performed on the Illumina HiSeq2500. RNA-Seq data are aligned using TopHat2 (Kim et al., 2013). Reads are assigned to transcripts using featureCounts and an mm9 genome modified to minimize overlapping transcripts (Liao et al., 2014). Differential expression analysis of the data is performed using EdgeR (Robinson et al., 2010).

DNA methylation and mRNA expression in human adipose tissue

DNA methylation was analyzed in adipose tissue from 28 subjects with type 2 diabetes and 28 controls as well as in adipose tissue of 14 monozygotic twin pairs (n = 28) using Infinium HumanMethylation450 BeadChips (Illumina) according to our previous study where we also present the characteristics of these subjects (Nilsson et al., 2014). Their characteristics are presented in Supplementary file 4. mRNA was analyzed in adipose tissue from the 12 monozygotic twin pairs (n = 24) using GeneChip Human Gene 1.0 ST arrays (Affymetrix, Santa Clara, CA) according to the manufacturer’s recommendations. In the present study, we studied mRNA expression and DNA methylation of CpG sites annotated to FGF21. DNA and RNA were extracted from human adipose tissue as presented (Nilsson et al., 2014).

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
    Large-scale association analysis provides insights into the genetic architecture and pathophysiology of type 2 diabetes
    1. AP Morris
    2. BF Voight
    3. TM Teslovich
    4. T Ferreira
    5. AV Segrè
    6. V Steinthorsdottir
    7. RJ Strawbridge
    8. H Khan
    9. H Grallert
    10. A Mahajan
    11. I Prokopenko
    12. HM Kang
    13. C Dina
    14. T Esko
    15. RM Fraser
    16. S Kanoni
    17. A Kumar
    18. V Lagou
    19. C Langenberg
    20. J Luan
    21. CM Lindgren
    22. M Müller-Nurasyid
    23. S Pechlivanis
    24. NW Rayner
    25. LJ Scott
    26. S Wiltshire
    27. L Yengo
    28. L Kinnunen
    29. EJ Rossin
    30. S Raychaudhuri
    31. AD Johnson
    32. AS Dimas
    33. RJ Loos
    34. S Vedantam
    35. H Chen
    36. JC Florez
    37. C Fox
    38. CT Liu
    39. D Rybin
    40. DJ Couper
    41. WH Kao
    42. M Li
    43. MC Cornelis
    44. P Kraft
    45. Q Sun
    46. RM van Dam
    47. HM Stringham
    48. PS Chines
    49. K Fischer
    50. P Fontanillas
    51. OL Holmen
    52. SE Hunt
    53. AU Jackson
    54. A Kong
    55. R Lawrence
    56. J Meyer
    57. JR Perry
    58. CG Platou
    59. S Potter
    60. E Rehnberg
    61. N Robertson
    62. S Sivapalaratnam
    63. A Stančáková
    64. K Stirrups
    65. G Thorleifsson
    66. E Tikkanen
    67. AR Wood
    68. P Almgren
    69. M Atalay
    70. R Benediktsson
    71. LL Bonnycastle
    72. N Burtt
    73. J Carey
    74. G Charpentier
    75. AT Crenshaw
    76. AS Doney
    77. M Dorkhan
    78. S Edkins
    79. V Emilsson
    80. E Eury
    81. T Forsen
    82. K Gertow
    83. B Gigante
    84. GB Grant
    85. CJ Groves
    86. C Guiducci
    87. C Herder
    88. AB Hreidarsson
    89. J Hui
    90. A James
    91. A Jonsson
    92. W Rathmann
    93. N Klopp
    94. J Kravic
    95. K Krjutškov
    96. C Langford
    97. K Leander
    98. E Lindholm
    99. S Lobbens
    100. S Männistö
    101. G Mirza
    102. TW Mühleisen
    103. B Musk
    104. M Parkin
    105. L Rallidis
    106. J Saramies
    107. B Sennblad
    108. S Shah
    109. G Sigurðsson
    110. A Silveira
    111. G Steinbach
    112. B Thorand
    113. J Trakalo
    114. F Veglia
    115. R Wennauer
    116. W Winckler
    117. D Zabaneh
    118. H Campbell
    119. C van Duijn
    120. AG Uitterlinden
    121. A Hofman
    122. E Sijbrands
    123. GR Abecasis
    124. KR Owen
    125. E Zeggini
    126. MD Trip
    127. NG Forouhi
    128. AC Syvänen
    129. JG Eriksson
    130. L Peltonen
    131. MM Nöthen
    132. B Balkau
    133. CN Palmer
    134. V Lyssenko
    135. T Tuomi
    136. B Isomaa
    137. DJ Hunter
    138. L Qi
    139. AR Shuldiner
    140. M Roden
    141. I Barroso
    142. T Wilsgaard
    143. J Beilby
    144. K Hovingh
    145. JF Price
    146. JF Wilson
    147. R Rauramaa
    148. TA Lakka
    149. L Lind
    150. G Dedoussis
    151. I Njølstad
    152. NL Pedersen
    153. KT Khaw
    154. NJ Wareham
    155. SM Keinanen-Kiukaanniemi
    156. TE Saaristo
    157. E Korpi-Hyövälti
    158. J Saltevo
    159. M Laakso
    160. J Kuusisto
    161. A Metspalu
    162. FS Collins
    163. KL Mohlke
    164. RN Bergman
    165. J Tuomilehto
    166. BO Boehm
    167. C Gieger
    168. K Hveem
    169. S Cauchi
    170. P Froguel
    171. D Baldassarre
    172. E Tremoli
    173. SE Humphries
    174. D Saleheen
    175. J Danesh
    176. E Ingelsson
    177. S Ripatti
    178. V Salomaa
    179. R Erbel
    180. KH Jöckel
    181. S Moebus
    182. A Peters
    183. T Illig
    184. U de Faire
    185. A Hamsten
    186. AD Morris
    187. PJ Donnelly
    188. TM Frayling
    189. AT Hattersley
    190. E Boerwinkle
    191. O Melander
    192. S Kathiresan
    193. PM Nilsson
    194. P Deloukas
    195. U Thorsteinsdottir
    196. LC Groop
    197. K Stefansson
    198. F Hu
    199. JS Pankow
    200. J Dupuis
    201. JB Meigs
    202. D Altshuler
    203. M Boehnke
    204. MI McCarthy
    205. Wellcome Trust Case Control Consortium
    206. Meta-Analyses of Glucose and Insulin-related traits Consortium (MAGIC) Investigators
    207. Genetic Investigation of ANthropometric Traits (GIANT) Consortium
    208. Asian Genetic Epidemiology Network–Type 2 Diabetes (AGEN-T2D) Consortium
    209. South Asian Type 2 Diabetes (SAT2D) Consortium
    210. DIAbetes Genetics Replication And Meta-analysis (DIAGRAM) Consortium
    (2012)
    Nature Genetics 44:981–990.
    https://doi.org/10.1038/ng.2383
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
  66. 66
  67. 67
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72
  73. 73
  74. 74
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
    Twelve type 2 diabetes susceptibility loci identified through large-scale association analysis
    1. BF Voight
    2. LJ Scott
    3. V Steinthorsdottir
    4. AP Morris
    5. C Dina
    6. RP Welch
    7. E Zeggini
    8. C Huth
    9. YS Aulchenko
    10. G Thorleifsson
    11. LJ McCulloch
    12. T Ferreira
    13. H Grallert
    14. N Amin
    15. G Wu
    16. CJ Willer
    17. S Raychaudhuri
    18. SA McCarroll
    19. C Langenberg
    20. OM Hofmann
    21. J Dupuis
    22. L Qi
    23. AV Segrè
    24. M van Hoek
    25. P Navarro
    26. K Ardlie
    27. B Balkau
    28. R Benediktsson
    29. AJ Bennett
    30. R Blagieva
    31. E Boerwinkle
    32. LL Bonnycastle
    33. K Bengtsson Boström
    34. B Bravenboer
    35. S Bumpstead
    36. NP Burtt
    37. G Charpentier
    38. PS Chines
    39. M Cornelis
    40. DJ Couper
    41. G Crawford
    42. AS Doney
    43. KS Elliott
    44. AL Elliott
    45. MR Erdos
    46. CS Fox
    47. CS Franklin
    48. M Ganser
    49. C Gieger
    50. N Grarup
    51. T Green
    52. S Griffin
    53. CJ Groves
    54. C Guiducci
    55. S Hadjadj
    56. N Hassanali
    57. C Herder
    58. B Isomaa
    59. AU Jackson
    60. PR Johnson
    61. T Jørgensen
    62. WH Kao
    63. N Klopp
    64. A Kong
    65. P Kraft
    66. J Kuusisto
    67. T Lauritzen
    68. M Li
    69. A Lieverse
    70. CM Lindgren
    71. V Lyssenko
    72. M Marre
    73. T Meitinger
    74. K Midthjell
    75. MA Morken
    76. N Narisu
    77. P Nilsson
    78. KR Owen
    79. F Payne
    80. JR Perry
    81. AK Petersen
    82. C Platou
    83. C Proença
    84. I Prokopenko
    85. W Rathmann
    86. NW Rayner
    87. NR Robertson
    88. G Rocheleau
    89. M Roden
    90. MJ Sampson
    91. R Saxena
    92. BM Shields
    93. P Shrader
    94. G Sigurdsson
    95. T Sparsø
    96. K Strassburger
    97. HM Stringham
    98. Q Sun
    99. AJ Swift
    100. B Thorand
    101. J Tichet
    102. T Tuomi
    103. RM van Dam
    104. TW van Haeften
    105. T van Herpt
    106. JV van Vliet-Ostaptchouk
    107. GB Walters
    108. MN Weedon
    109. C Wijmenga
    110. J Witteman
    111. RN Bergman
    112. S Cauchi
    113. FS Collins
    114. AL Gloyn
    115. U Gyllensten
    116. T Hansen
    117. WA Hide
    118. GA Hitman
    119. A Hofman
    120. DJ Hunter
    121. K Hveem
    122. M Laakso
    123. KL Mohlke
    124. AD Morris
    125. CN Palmer
    126. PP Pramstaller
    127. I Rudan
    128. E Sijbrands
    129. LD Stein
    130. J Tuomilehto
    131. A Uitterlinden
    132. M Walker
    133. NJ Wareham
    134. RM Watanabe
    135. GR Abecasis
    136. BO Boehm
    137. H Campbell
    138. MJ Daly
    139. AT Hattersley
    140. FB Hu
    141. JB Meigs
    142. JS Pankow
    143. O Pedersen
    144. HE Wichmann
    145. I Barroso
    146. JC Florez
    147. TM Frayling
    148. L Groop
    149. R Sladek
    150. U Thorsteinsdottir
    151. JF Wilson
    152. T Illig
    153. P Froguel
    154. CM van Duijn
    155. K Stefansson
    156. D Altshuler
    157. M Boehnke
    158. MI McCarthy
    159. MAGIC investigators
    160. GIANT Consortium
    (2010)
    Nature Genetics 42:579–589.
    https://doi.org/10.1038/ng.609
  81. 81
  82. 82
  83. 83
  84. 84
  85. 85
  86. 86
  87. 87
  88. 88
  89. 89
  90. 90
  91. 91

Decision letter

  1. Clifford J Rosen
    Reviewing Editor; Maine Medical Center Research Institute, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Dnmt3a is an epigenetic mediator of adipose insulin resistance" for consideration by eLife. Your article has been favorably evaluated by Mark McCarthy (Senior Editor) and three reviewers, one of whom, Clifford J Rosen (Reviewer #1), is a member of our Board of Reviewing Editors. The following individual involved in review of your submission has agreed to reveal their identity: Rob Koza (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission. Overall there is significant enthusiasm for this manuscript, if the suggestions noted below by the reviewing editor are considered within the context of the other comments and the time frame for resubmission.

Essential revisions:

There were a couple of points that need to be addressed in the response and incorporated within the manuscript. These include three major comments:

1) A clearer rationale for studying epididymal fat as the major tissue source for the studies and elaboration about the possible beiging that could occur in the conditional deletion as a mechanism for the enhanced insulin sensitivity.

2) More elaboration about the regulation of the enzyme itself and the impact of activated PPARG on expression of Dnmt3.

3) Considerations for more in vivo proof that FGF-21 is the target repressor of Dnmt3.

Reviewer #1:

Rosen and colleagues in this paper found enhanced expression of DNA methyltransferases (Dnmts) in insulin resistance. Pharmacological and genetic inhibition studies demonstrated that Dnmt3a was both necessary and sufficient to mediate insulin resistance in cultured mouse and human adipocytes. In addition, adipose-specific Dnmt3a knock-out mice were protected from diet-induced insulin resistance and glucose intolerance without accompanying changes in adiposity. Gene expression profiling studies revealed Fgf21 as a key negatively regulated Dnmt3a target gene in adipocytes with concordant changes in DNA methylation at the Fgf21 promoter region. Fgf21 rescued Dnmt3a-mediated insulin resistance in vitro, and DNA methylation at the FGF21 locus was elevated in human subjects with diabetes and correlated negatively with expression of FGF21 in human adipose tissue. Thus they concluded that adipose Dnmt3a is a novel epigenetic mediator of insulin resistance in vitro and in vivo.

This manuscript breaks new ground in describing an epigenetic mechanism for insulin resistance that is mediated through adipose specific FGF-21. The experiments are straightforward, the results consistent and the discussion is not over-reaching. The in vivo data are strong, and the in vitro rescue and silencing studies stand up to closer scrutiny. The human studies provide a translational perspective although somewhat limited by the confounders in T2d. The dose dependent suppression of Fgf21 luc with Dnmt3a in figure 6D is particularly striking. Thus overall there are considerable strengths that are both basic and translational. The concerns are relatively limited and are noted below:

1) Rosiglitazone suppressed Dnmt3a – have the authors looked at whether activated PPARG directly mediates the suppression of Dnmt3a expression, or is it one of the many downstream targets of PPARG? This is particularly relevant since the enzyme's regulation is not defined, although it increases with a HFD, albeit without a known mechanism.

2) Can the authors discuss the discrepancy in vivo between circulating FGF-21 (likely hepatic- low) and adipose FGF-21 levels (high) in the Dnmt3a KO mice?

3) The authors mention that Fgf21 null mice do not support Rosi-mediated insulin sensitivity – does crossing the Fgf21-/- with the Dnmt3 adipose specific KO restore insulin resistance in those mice fed a HFD? This certainly would provide further in vivo proof of concept that FGF-21 is the target repressor of Dnmt3a.

4) The human studies are interesting but somewhat limited by the small number of subjects and the relatively weak, albeit significant negative regressions between methylation at the Fgf21 locus and mRNA expression at CpG sites. Of concern is the lack of a description of the diabetic subjects and their controls. For example, were the T2D patients treated with oral agents and/or insulin and did concurrent treatment affect methylation expression status?

5) One of the novel aspects of these studies is the similar weight gain and fat mass change in the Drnmt3a adipose specific KO, compared to the control. Thus insulin sensitivity is maintained even in the face of greater adiposity, a key finding that needs more emphasis in the Results and Discussion.

6) The authors fail to cite the paper by Kamei et al. in Obesity 2010, which demonstrated that an aP2 transgenic Dnmt3 mouse (PMID: 19680236) exhibit similar weight gain to wt mice; but importantly those mice have increased expression of Tnfa and MCP-1 in fat from their adipose depots. Similarly in that paper Dnmt3 expression was higher in WT mice fed a HFD. There should be a citation and also discussion about that paper (limitations based on the aP2 promoter, etc.).

Reviewer #2:

The manuscript entitled: 'Dnmt3a is an epigenetic mediator of adipose insulin resistance' submitted presents substantial and compelling evidence that adipose tissue Dnmt3a plays a regulatory role in insulin resistance. The PIs present a clear and logical progression of studies with knockdown and transgenic overexpression of Dnmt3a in cultured adipocytes; and, adipose tissue specific inactivation of Dnmt3a in mice. In addition, RNA-seq analyses of 3T3-L1 adipocytes with shRNA knockdown or transgenic overexpression of Dnmt3a identified Fgf21 as a Dnmt3a target gene. Analyses of the Fgf21 promoter using MeDIP-qPCR confirmed DNMT3A-mediated methylation as a mechanism for Fgf21 repression. Overall, these studies are comprehensive and very well presented; and, begin to elucidate epigenetic mechanisms that can modulate adipose tissue 'health' with respect to insulin resistance and glucose uptake.

1) What is the rationale for using the epididymal adipose tissue depot to measure HFD-mediated induction of Dnmt3a? Are there differences in HFD regulation of Dnmt3a between visceral and subcutaneous adipose tissue? Fgf21?

2) Although not as dramatic as in adipose tissue, there is a relatively robust and significant increase of Dnmt3a in skeletal muscle in mice fed a HFD (Figure 1—figure supplement 1). What is the relative expression of skeletal muscle Dnmt3a with respect to adipose tissue and liver? What are the potential implications for HFD-mediated up-regulation of skeletal muscle Dnmt3a with respect to Fgf21 expression?

Reviewer #3:

The manuscript "Dnmt3a is an epigenetic mediator of adipose insulin resistance" by You et al. presents new data linking the DNA methyltransferase Dnmt3a to insulin resistance in adipocytes. The authors start by demonstrating increased expression of Dnmt3a in models of insulin resistance. They then do pharmacologic and genetic studies in vitro to support a role for Dnmt3a in adipocyte insulin sensitivity. They then go on to generate an adipocyte-specific Dnmt3a knockout mouse and show that it is protected from diet-induced insulin resistance. Finally, they perform mechanistic studies showing that Fgf21 is negatively regulated by Dnmt3a. This is a well-written paper, with novel data, and well-designed experiments. As such, this would be of broad interest to the readership of this journal.

If the authors are able to address the point below, the manuscript would be further strengthened:

1) In their studies here, the authors only look at Dnmt3a in gonadal white fat (eWAT). Since Dnmt3a is apparently broadly expressed, it would be interesting to know whether it is also increased in subcutaneous/beige fat and in brown fat in models of insulin resistance. Moreover, it would be relevant to know whether beige/brown fat are altered in the adipocyte-specific Dnmt3a knockout model. If for example, the mutant mouse has activated brown or beige fat, this could be an explanation for the improved glucose and insulin tolerance.

https://doi.org/10.7554/eLife.30766.041

Author response

Essential revisions:

There were a couple of points that need to be addressed in the response and incorporated within the manuscript. These include three major comments:

1) A clearer rationale for studying epididymal fat as the major tissue source for the studies and elaboration about the possible beiging that could occur in the conditional deletion as a mechanism for the enhanced insulin sensitivity.

We now show that visceral fat depots (i.e. epididymal and mesenteric WAT) express higher levels of Dnmt3a than liver and muscle, and we also show that the levels in these depots are more responsive to the effects of HFD than inguinal WAT and interscapular BAT. Additionally, we have looked for evidence of subcutaneous browning but we see none, either by histological examination or by analysis of thermogenic gene expression.

2) More elaboration about the regulation of the enzyme itself and the impact of activated PPARG on expression of Dnmt3.

While we agree that this is an interesting topic, it is very difficult to tease this apart, as explained below in the response to reviewer 1. We have done several things to address this issue that have produced data but not a tremendous amount of insight. To summarize, we have tried to model the effects of obesity in cultured adipocytes using FFAs, cytokines, insulin, and glucose, none of which raise Dnmt3a levels. Treatment of cultured adipocytes with TZD does not reduce basal levels of Dnmt3a, which is consistent with the in vivo results showing that Rosi reduces Dnmt3a levels in obese but not lean mice. There are PPARγ binding sites in the murine genome that lie near the Dnmt3a TSS; whether these are functional is not easy to test given the constraints already mentioned. Complicating this further is the fact that inhibiting PPARγ in mature adipocytes leads to a general process of de-differentiation, confounding efforts to parse a specific effect on Dnmt3a expression. Very little is known about how PPARγ (or any nuclear receptor) mediates negative gene expression; we submit that detailed analysis of this is beyond the scope of this study.

3) Considerations for more in vivo proof that FGF-21 is the target repressor of Dnmt3.

The definitive experiment to demonstrate this in vivo would be a five allele cross of adiponectinCre, Dnmt3a Flox, and Fgf21 null mice. This experiment can’t be completed within the time frame laid out by the eLife review process, and completing it would require at least a year and half of breeding. We have therefore chosen not to pursue this as part of the current manuscript.

Reviewer #1:

[…] 1) Rosiglitazone suppressed Dnmt3a – have the authors looked at whether activated PPARG directly mediates the suppression of Dnmt3a expression, or is it one of the many downstream targets of PPARG? This is particularly relevant since the enzyme's regulation is not defined, although it increases with a HFD, albeit without a known mechanism.

We appreciate (and share) the reviewer’s curiosity about this important issue, but we would argue that is a very complex subject to address. First, we know very little about negative regulation by PPARγ in general. Like many nuclear receptors that are primarily transcriptional activators, PPARγ has negative targets, including leptin and adipsin. How this occurs is unknown, although trans-repression has been proposed as a mechanism for TZD antagonism of inflammatory genes. We note that there are bona fide PPARγ sites near the Dnmt3a locus (see Author response image 1), suggesting at least the possibility of direct regulation. Proving that would require CRISPRing those sites in mature adipocytes; not an easy feat. One can’t simply perform loss of function studies of PPARγ in mature adipocytes, as the cells immediately begin to dedifferentiate. This has an effect on Dnmt3a gene expression, but not the effect we are looking to study. Given these constraints, we did the following: first, we assessed whether lipids, glucose, insulin, or cytokine administration could model the effect of obesity on Dnmt3a expression in vitro. Unfortunately, they do not. Next, we tested whether rosiglitazone could repress Dnmt3a in 3T3-L1 adipocytes, and if so, whether cycloheximide would have an effect on the process. Those data are shown below. The upshot is that TZD does not repress basal Dnmt3a in adipocytes, consistent with the in vivo data shown in Figure 1B, in that TZD rescues the increased Dnmt3a of obesity but does not cause Dnmt3a levels to go lower than those seen in lean animals. We have added discussion of these points to the manuscript. We have not included the data shown in Author response image 1 because we feel they don’t bring additional enlightenment to the reader, but we can do so if the reviewer and editor wish.

Author response image 1
(A) Histograms of PPARγ-ChIP and histone modification fragments near the mouse Dnmt3a locus (Mikkelsen et al., 2010).

Depicted with arrow are PPARγ peaks that co-localize with H3K4me1 and/or H3K27ac enhancer peaks. (B) The effect of Rosi (1uM) on the expression of Dnmt3a in the presence/absence of cycloheximide (10ug/ml) in mature 3T3-L1 adipocytes.

2) Can the authors discuss the discrepancy in vivo between circulating FGF-21 (likely hepatic- low) and adipose FGF-21 levels (high) in the Dnmt3a KO mice?

We have added additional thoughts on this topic to the Discussion.

3) The authors mention that Fgf21 null mice do not support Rosi-mediated insulin sensitivity – does crossing the Fgf21-/- with the Dnmt3 adipose specific KO restore insulin resistance in those mice fed a HFD? This certainly would provide further in vivo proof of concept that FGF-21 is the target repressor of Dnmt3a.

We absolutely agree that this is the definitive in vivo experiment. However, it is a five allele cross (two Dnmt3aflox alleles, two FGF21 KO alleles, and one adiponectin Cre), and we fear that the time spent on this (approximately 18 months) would unnecessarily delay this report and would dilute the impact of our findings.

4) The human studies are interesting but somewhat limited by the small number of subjects and the relatively weak, albeit significant negative regressions between methylation at the Fgf21 locus and mRNA expression at CpG sites. Of concern is the lack of a description of the diabetic subjects and their controls. For example, were the T2D patients treated with oral agents and/or insulin and did concurrent treatment affect methylation expression status?

Based on this valid comment, we added a table (see Supplementary file 4) describing the characteristics of the diabetics and the controls included in this study. We also added information about the treatments of the diabetics in the legend of the table. We did not find any effects of treatment (insulin and metformin) on methylation of three of the studied CpG sites or expression of Fgf21 (data not shown). However, a weak effect of treatment on methylation of cg21591807 was found, but this effect did not remain significant after adjustment for multiple testing (data not shown).

5) One of the novel aspects of these studies is the similar weight gain and fat mass change in the Drnmt3a adipose specific KO, compared to the control. Thus insulin sensitivity is maintained even in the face of greater adiposity, a key finding that needs more emphasis in the Results and Discussion.

We appreciate reviewer’s suggestion, and we have now emphasized the importance and uniqueness of this finding in the Discussion.

6) The authors fail to cite the paper by Kamei et al. in Obesity 2010, which demonstrated that an aP2 transgenic Dnmt3 mouse (PMID: 19680236) exhibit similar weight gain to wt mice; but importantly those mice have increased expression of Tnfa and MCP-1 in fat from their adipose depots. Similarly in that paper Dnmt3 expression was higher in WT mice fed a HFD. There should be a citation and also discussion about that paper (limitations based on the aP2 promoter, etc.).

We thank the reviewer for bringing up this reference, which we had overlooked. We now cite it. As noted, the authors of this paper generated adipose-selective Dnmt3a-overexpressor mice, which did not show dramatic changes in whole body metabolism on chow, HFD and nor with high methyl donor diet on chow. Of note, those authors did minimal physiological characterization of these mice, restricted mainly to a single time point of body weight and serum lipids, glucose and insulin. An additional caveat, also noted by the reviewer, is the use of the aP2 (Fabp4) promoter to drive transgene expression. As described in recent papers (Jeffery et al., 2014; Kang, Kong and Rosen, 2014; Lee et al., 2013), there are significant concerns about specificity as the aP2 gene is also expressed in non-adipose tissues including macrophages. Another potential compounding factor is the ‘timing’ of aP2 expression, which is speculated to be early during adipose development. Some studies have shown that Dnmt3a affects adipogenesis (Guo, Chen, Yang, Zhu and Wu, 2016), and so, a potential developmental effect of transgene expression might have confounded interpretation of the metabolic effect in this model. Per the reviewer’s suggestion, we investigated inflammation in Dnmt3a KO mice. We observe a non-significant trend toward reduced TNF and MCP-1 expression in KO mice (Figure 4—figure supplement 5A). We also saw no obvious change in macrophage infiltration by histological examination in Dnmt3a KO mice after HFD (Figure 4—figure supplement 5B).

Reviewer #2:

[…] 1) What is the rationale for using the epididymal adipose tissue depot to measure HFD-mediated induction of Dnmt3a? Are there differences in HFD regulation of Dnmt3a between visceral and subcutaneous adipose tissue? Fgf21?

Epididymal adipose tissue is a large and accessible depot, and its mass correlates well with insulin resistance. We have now measured Dnmt3a expression in BAT, subcutaneous WAT, and mesenteric WAT from chow vs HFD mice (Figure 1—figure supplement A). Dnmt3a expression in all depots were significantly elevated but most prominently in visceral WAT (i.e. epididymal and mesenteric WAT).

2) Although not as dramatic as in adipose tissue, there is a relatively robust and significant increase of Dnmt3a in skeletal muscle in mice fed a HFD (Figure 1—figure supplement 1). What is the relative expression of skeletal muscle Dnmt3a with respect to adipose tissue and liver? What are the potential implications for HFD-mediated up-regulation of skeletal muscle Dnmt3a with respect to Fgf21 expression?

Based on the reviewer’s suggestion, we measured the relative expression of Dnmt3a between those three tissues. Protein expression of Dnmt3a is highest in epididymal fat compared to liver and muscle (Figure 4—figure supplement 1D, E). It seems possible that Dnmt3a could suppress Fgf21 levels in skeletal muscle as it does in adipose tissue. Overall, however, serum FGF21 levels rise in response to a HFD (Figure 4—figure supplement 3).

Reviewer #3:

[…] If the authors are able to address the point below, the manuscript would be further strengthened:

1) In their studies here, the authors only look at Dnmt3a in gonadal white fat (eWAT). Since Dnmt3a is apparently broadly expressed, it would be interesting to know whether it is also increased in subcutaneous/beige fat and in brown fat in models of insulin resistance. Moreover, it would be relevant to know whether beige/brown fat are altered in the adipocyte-specific Dnmt3a knockout model. If for example, the mutant mouse has activated brown or beige fat, this could be an explanation for the improved glucose and insulin tolerance.

This is a fair point. We now add data showing how Dnmt3a expression changes in additional depots after HFD (Figure 1—figure supplement 1A); the upshot is that there is a greater effect in visceral depots (epiWAT and mesenteric WAT) than in iWAT or BAT.

In addition, Dnmt3a KO mice do not display changes in body weight or adiposity at baseline or in response to fasting (Figures 4A-C), suggesting that increased thermogenesis/energy expenditure is not a likely mechanism for improved insulin sensitivity in the KO mice. However, to rule out a subtle but significant change in beige/BAT activity, we assessed the expression of major thermogenic genes in subcutaneous WAT and BAT. We confirm there is no major impact on thermogenic genes in beige and brown fat in KO animals; if anything, we note a slight reduction of Ucp1 expression in subcutaneous WAT (Figure 4—figure supplement 4). We also did not observe any obvious histological changes that would suggest increased browning (Figure 4—figure supplement 5B).

https://doi.org/10.7554/eLife.30766.042

Article and author information

Author details

  1. Dongjoo You

    Nutritional Sciences and Toxicology Department, University of California, Berkeley, Berkeley, United States
    Contribution
    Formal analysis, Validation, Investigation, Carried out most of biological experiments and assisted with preparing the manuscript
    Competing interests
    No competing interests declared
  2. Emma Nilsson

    Epigenetics and Diabetes Unit, Department of Clinical Sciences, Lund University Diabetes Centre, Scania University Hospital, Malmö, Sweden
    Contribution
    Data curation, Formal analysis, Investigation, Collected human adipose tissue and analyzed DNA methylation and gene expression with human sample
    Competing interests
    No competing interests declared
  3. Danielle E Tenen

    Division of Endocrinology, Diabetes and Metabolism, Beth Israel Deaconess Medical Center, Boston, United States
    Contribution
    Investigation, Methodology, Generated RNA-Seq library
    Competing interests
    No competing interests declared
  4. Anna Lyubetskaya

    Broad Institute of Harvard and MIT, Cambridge, United States
    Contribution
    Formal analysis, Generated and analyzed RNA-Seq data
    Competing interests
    No competing interests declared
  5. James C Lo

    Weill Cornell Medical College, New York, United States
    Contribution
    Resources, Helped with metabolic characterization of knock-out mouse models
    Competing interests
    No competing interests declared
  6. Rencong Jiang

    Nutritional Sciences and Toxicology Department, University of California, Berkeley, Berkeley, United States
    Contribution
    Investigation, Carried out insulin signaling assay
    Competing interests
    No competing interests declared
  7. Jasmine Deng

    Nutritional Sciences and Toxicology Department, University of California, Berkeley, Berkeley, United States
    Contribution
    Investigation, Carried out gene expression analysis of several biological experiments
    Competing interests
    No competing interests declared
  8. Brian A Dawes

    Division of Endocrinology, Diabetes and Metabolism, Beth Israel Deaconess Medical Center, Boston, United States
    Contribution
    Formal analysis, Helped with RNA-Seq data analysis, Generated figures, Submitted GEO data
    Competing interests
    No competing interests declared
  9. Allan Vaag

    1. Diabetes and Metabolism, Department of Endocrinology, Rigshospitalet, University of Copenhagen, Copenhagen, Denmark
    2. Early Clinical Development, AstraZeneca, Innovative Medicines, Göteborg, Sweden
    Contribution
    Investigation, Collected human adipose tissue and analyzed DNA methylation and gene expression with human sample
    Competing interests
    No competing interests declared
  10. Charlotte Ling

    Epigenetics and Diabetes Unit, Department of Clinical Sciences, Lund University Diabetes Centre, Scania University Hospital, Malmö, Sweden
    Contribution
    Resources, Investigation, Collected human adipose tissue and analyzed DNA methylation and gene expression data, Supervised overall human data set
    Competing interests
    No competing interests declared
  11. Evan D Rosen

    1. Division of Endocrinology, Diabetes and Metabolism, Beth Israel Deaconess Medical Center, Boston, United States
    2. Broad Institute of Harvard and MIT, Cambridge, United States
    Contribution
    Supervision, Funding acquisition, Writing—original draft, Writing—review and editing, Supervised experiments and edited the manuscript and revision for the revision
    For correspondence
    erosen@bidmc.harvard.edu
    Competing interests
    No competing interests declared
  12. Sona Kang

    Nutritional Sciences and Toxicology Department, University of California, Berkeley, Berkeley, United States
    Contribution
    Supervision, Investigation, Writing—original draft, Writing—review and editing, Designed most of biological experiments, Supervised experiments, Wrote draft and revision for the revision
    For correspondence
    kangs@berkeley.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-9831-677X

Funding

National Institutes of Health (102173)

  • Evan D Rosen

National Institutes of Health (102170)

  • Evan D Rosen

National Institutes of Health (085171)

  • Evan D Rosen

American Heart Association (15SDG25240017)

  • Sona Kang

National Institutes of Health (116008)

  • Sona Kang

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

Work was funded by NIH R01 102173, 102170, and 085171 to EDR and AHA Award #15SDG25240017 to SK. We thank Jeff Leung for technical assistance and the Rosen and the Kang Lab members for helpful conversations. We are also grateful to Drs. Hei Sook Sul, Jen-Chywan Wally Wang, and Andreas Stahl for constructive comments.

Ethics

Animal experimentation: This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All animal work was approved by the BIDMC IACUC (056-2017) and/or the UC Berkeley ACUC (AUP-2015-08-7887).

Reviewing Editor

  1. Clifford J Rosen, Maine Medical Center Research Institute, United States

Publication history

  1. Received: July 26, 2017
  2. Accepted: October 29, 2017
  3. Accepted Manuscript published: November 1, 2017 (version 1)
  4. Version of Record published: December 14, 2017 (version 2)

Copyright

© 2017, You et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 2,368
    Page views
  • 580
    Downloads
  • 13
    Citations

Article citation count generated by polling the highest count across the following sources: Scopus, Crossref, PubMed Central.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

  1. Further reading

Further reading

    1. Cell Biology
    2. Chromosomes and Gene Expression
    Alexander JR Booth et al.
    Research Article Updated
    1. Cell Biology
    Naama Flint Brodsly et al.
    Research Article