Main

Obesity develops when energy intake chronically exceeds total energy expenditure. All anti-obesity medications currently approved by the FDA act to repress energy intake, either by suppressing appetite or by inhibiting intestinal fat absorption. However, because of their side effects including depression, oily bowel movements and steatorrhoea, there is an urgent need for alternative approaches. BAT is specialized to dissipate energy through uncoupling protein 1 (UCP1). Recent studies with 18fluoro-labelled 2-deoxy-glucose positron emission tomography (18FDG-PET) scanning demonstrated that adult humans have active BAT deposits3,4,5,6 and that the amount of BAT inversely correlates with adiposity and body mass index4,5, indicating that it plays an important role in energy homeostasis in adult humans. Hence, a better understanding of the molecular control of BAT development may lead to an alternative approach to alter energy balance by increasing energy expenditure.

It has been reported that brown adipocytes in the interscapular and peri-renal BAT arise from Engrailed-1+ and Myf5+ dermotomal precursors1,7,8. The PRDM16–C/EBP-β complex in the myogenic precursors activates the brown adipogenic gene program through inducing peroxisome proliferator-activated receptor (PPAR)-γ expression1,2,9; however, the mechanism by which the PRDM16–C/EBP-β complex functions as a fate switch to control brown adipocyte versus myocyte lineage remains unexplored.

Previously we determined the essential domains of PRDM16 for converting myoblasts into brown adipocytes by generating two deletion mutants of PRDM16: a mutant lacking the PR-domain (ΔPR), a domain that shares high homology with methyltransferase SET domains10,11, and a mutant lacking the zinc-finger domain-1 (ΔZF-1) (Fig. 1a, top panel). Wild type (WT) and the ΔPR mutant, but not the ΔZF-1 mutant, were able to convert myoblasts into brown adipocytes, suggesting that the ZF-1 domain is required2. Consistent with the results, the PRDM16 complex purified from brown adipocytes expressing WT and ΔPR, but not ΔZF-1, had significant methyltransferase activities on H3 (Fig. 1a, bottom panel). Because this effect was independent of its SET domain, we searched for methyltransferases that were associated with differentiation-competent PRDM16 proteins (that is, WT and ΔPR), but not with differentiation-incompetent PRDM16 (ΔZF-1). By using high-resolution liquid chromatography coupled with tandem mass spectrometry (LC–MS/MS), we found EHMT1 as the only methyltransferase that was co-purified preferentially with the differentiation-competent PRDM16 complexes2. EHMT1 has enzymatic activity on H3K9 mono- or di-Me12. Notably, haploinsufficiency of the EHMT1 gene, because of 9q34.3 microdeletions or point mutations in humans13, is associated with clinical phenotypes including mental retardation. Importantly, 40–50% of patients with EHMT1 mutations develop obesity14,15; however, the underlying mechanism remains completely unknown. Given the essential role of the PRDM16 complex for BAT development, we considered that EHMT1 is a key enzymatic component that controls the lineage specification and thermogenic function of BAT.

Figure 1: Identification of EHMT1 in the PRDM16 transcriptional complex.
figure 1

a, Top: schematic illustration of PRDM16. Bottom: PRDM16 complex purified from brown adipocytes were subjected to in vitro histone methylation assay. b, Immunoprecipitation of EHMT1 protein followed by western blotting to detect PRDM16. Input is shown in lower panels. c, In vitro binding assay of 35S-labelled EHMT1 or CtBP1 and purified PRDM16 fragments. d, Histone methylation assay of PRDM16 complex from brown adipocytes expressing indicated constructs (n = 3 or 4). e, Western blotting for indicated proteins in adipose tissues. f, Transcriptional activities of PRDM16 using a PPAR-γ-responsive luciferase reporter (n = 3). Error bars, s.e.m. **P < 0.01, ***P < 0.001.

PowerPoint slide

Full size image

To test this hypothesis, we first confirmed the PRDM16–EHMT1 interaction by immunoprecipitation followed by western blotting in brown adipocytes (Fig. 1b and Supplementary Fig. 1). The purified ZF-1 (224–454) and ZF-2 (881–1038) domains of glutathione S-transferase (GST)–PRDM16 protein bound to the in-vitro-translated EHMT1 protein, whereas the 680–1038 region of PRDM16 bound to CtBP1 as previously reported16 (Fig. 1c and Supplementary Fig. 2). These results indicate that EHMT1 directly interacts with PRDM16. EHMT1 is the main methyltransferase of the PRDM16 complex in brown adipocytes, because the histone methyltransferase activity of the PRDM16 complex was largely lost when EHMT1 was depleted using two short hairpin RNAs (shRNAs) targeted to EHMT1 (Fig. 1d and Supplementary Fig. 3). Furthermore, expression of EHMT1 protein was highly enriched in BAT and in cultured brown adipocytes, correlating well with PRDM16 (Fig. 1e and Supplementary Fig. 4). In contrast, amounts of EHMT2 protein were higher in white adipose tissue (WAT) than in BAT. To test if EHMT1 modulates the PRDM16 transcriptional activity, we performed luciferase assays using a luciferase reporter gene containing PPAR-γ binding sites1. As shown in Fig. 1f, co-expression of EHMT1 and PRDM16 synergistically increased the reporter gene activity, whereas this induction was completely lost when the ΔZF-1 mutant was expressed. These data indicate that EHMT1 forms a transcriptional complex with PRDM16 and regulates its activity through direct interaction.

Next, we investigated the genetic requirement for EHMT1 in BAT development in vivo. Because a whole-body knockout of the Ehmt1 gene causes embryonic lethality before the emergence of brown adipocytes12, the Ehmt1 gene was deleted in brown adipocyte precursors by crossing Ehmt1flox/flox mice17 with Myf5-Cre mice1. As shown in Fig. 2a–c, the interscapular BAT of Ehmt1myf5 knockout mice was substantially smaller than in WT mice at postnatal stage (P)1. Haematoxylin and eosin staining showed that brown adipocytes in Ehmt1myf5 knockout mice were significantly smaller and contained fewer lipids than in WT mice, whereas other tissues near the BAT including skin seemed normal (Fig. 2b and Supplementary Fig. 5). Similar results were observed in embryos at embryonic day (E)18.5 (Supplementary Fig. 6). Subsequently, we analysed the global gene expression of BAT from the WT and Ehmt1myf5 knockout embryos by RNA-sequencing. The following gene ontology analysis found that the gene expression pattern in the Ehmt1myf5 knockout BAT showed a skeletal-muscle phenotype: that is, a broad activation of the skeletal muscle-selective genes, and a broad reduction of the BAT-selective genes. Strikingly, 78.7% of the differentially expressed genes (118 out of 150 genes) between WT and knockout mice were stratified into categories of skeletal muscle development, BAT development and BAT function (glucose/fatty-acid metabolism). Specifically, 77.5% of the ectopically activated genes in the knockout BAT were related to skeletal muscle development, including myogenin and myosin heavy chains. On the other hand, 80.0% of the reduced genes in the knockout BAT were involved in BAT development and fatty-acid/glucose metabolism, including Ucp1, Pgc1a, Cebpb, Cpt1a and Elovl3 (Fig. 2d and Supplementary Fig. 7). These results indicate that EHMT1 is absolutely required for the cell-fate specification between BAT versus muscle.

Figure 2: EHMT1 is required for BAT versus muscle lineage specification.
figure 2

a, Morphology of BAT from WT and Ehmt1myf5 knockout embryos at P1. Scale bar, 2.5 mm. KO, knockout. b, Haematoxylin and eosin staining of WT and Ehmt1myf5 knockout (KO) BAT. Scale bar, 600 μm. Bottom: high-magnification images. Scale bar, 30 μm. c, BAT weight from WT (n = 14) and knockout embryos (n = 8). d, Gene ontology analyses of RNA-sequencing data. The log2-fold changes in the expression of skeletal muscle (group M) and BAT (group B) genes are shown. e, Immunocytochemistry for MHC in C2C12 cells expressing indicated constructs under pro-myogenic culture conditions. Scale bar, 200 μm. f, Myogenin mRNA expression in e (n = 3). g, Chromatin immunoprecipitation assays using indicated antibodies (n = 3). Error bars, s.e.m. *P < 0.05, **P < 0.01, ***P < 0.001.

PowerPoint slide

Full size image

To investigate the mechanisms by which EHMT1 determines BAT lineage, retroviruses expressing a scrambled control RNA (scr) or shRNAs targeting EHMT1 (shEHMT1-1 and -2) were transduced into C2C12 myoblasts together with PRDM16 (Supplementary Fig. 8a). As shown in Fig. 2e (upper panels), PRDM16 expression powerfully blocked myogenic differentiation in a dose-dependent fashion, as shown by immunohistochemistry using a pan-skeletal myosin heavy chain (MHC) antibody. In contrast, EHMT1 depletion significantly impaired the PRDM16-mediated repression on myogenesis (Fig. 2e, lower panels). Gene expression analysis showed that the repression on muscle-selective genes such as myogenin was near completely abolished when EHMT1 was depleted (Fig. 2f and Supplementary Fig. 8b). This repressive effect was mediated through the methyltransferase activity of EHMT1, because ectopic expression of the EHMT1 mutant (N1198L;H1199E) that lacks methyltransferase activity18 significantly blunted the PRDM16-mediated repression on myogenesis (Supplementary Fig. 9a). Additionally, two chemical inhibitors of EHMT1/2, BIX-01294 and UNC0638, blocked the repressive effects of PRDM16 (Supplementary Fig. 9b, c). BIX-01294 treatment in brown adipocytes also significantly reduced the expression of BAT-selective genes (Supplementary Fig. 9d). Consistent with these data, chromatin immunoprecipitation assays found that EHMT1 depletion robustly reduced amounts of H3K9me2 and me3 at the proximal region of the myogenin gene promoter on which EHMT1 was recruited (Fig. 2g). On the contrary, the amounts of H3K9/14ac were significantly increased by EHMT1 depletion without any effect on total H3 amounts. Similar changes were observed at the promoter regions of other muscle-selective genes including Acts, Ryr1 and Mylpf1, where EHMT1 was recruited (Supplementary Fig. 10). Conversely, under pro-adipogenic culture conditions, knockdown of EHMT1 largely blocked the PRDM16-induced brown adipogenesis in C2C12 cells (Supplementary Fig. 11). Together, these results indicate that EHMT1 determines BAT versus muscle cell lineage through PRDM16 by controlling H3K9 methylation status of the muscle-selective gene promoters.

To investigate the role of EHMT1 in BAT thermogenesis, EHMT1 was depleted in immortalized brown adipocytes by retrovirus-mediated shRNA knockdown (Supplementary Fig. 12a, b). Total and uncoupled (oligomycin-insensitive) oxygen consumption rate in the EHMT1-depleted brown adipocytes was significantly reduced both at the basal and cyclic AMP (cAMP)-stimulated states (Fig. 3a). Conversely, EHMT1 overexpression significantly increased messenger RNA (mRNA) amounts of BAT-selective thermogenic genes, including Ucp1, Pgc1a and Dio2 (Fig. 3b), and oxygen consumption rate (Supplementary Fig. 12c). To test further if this EHMT1 action requires PRDM16, EHMT1 was ectopically introduced into mouse embryonic fibroblasts that did not express endogenous PRDM16. As shown in Fig. 3c, mouse embryonic fibroblasts expressing PRDM16 and C/EBP-β uniformly differentiated into lipid-containing adipocytes as previously reported2. Although EHMT1 alone did not stimulate brown adipogenesis, the combination of EHMT1 with PRDM16 and C/EBP-β synergistically increased mRNA amounts of the BAT-selective genes, including Ucp1, Cidea, Cox7a and Cox8b (Fig. 3d). These data indicate that EHMT1 positively regulates the BAT-selective thermogenic gene program through PRDM16.

Figure 3: EHMT1 controls BAT thermogenesis through stabilizing PRDM16 protein.
figure 3

a, Cellular respiration in brown adipocytes expressing indicated constructs (n = 6). b, BAT-selective gene expression in brown adipocytes expressing indicated constructs (n = 3). c, Oil-Red-O staining of mouse embryonic fibroblasts expressing indicated constructs under pro-adipogenic culture conditions. d, BAT-selective gene expression in c (n = 3). e, Effects of EHMT1 mutants on PRDM16 transcriptional activities (n = 3). f, Amounts of PRDM16 protein in COS7 cells expressing indicated constructs. g, Regression analysis of the PRDM16 protein stability (n = 3). h, Changes in rectal temperature during a cold challenge (n = 4 or 5). Error bars, s.e.m. *P < 0.05, **P < 0.01, ***P < 0.001.

PowerPoint slide

Full size image

To determine which domains of EHMT1 are required for the induction of PRDM16 transcriptional activity, we tested a series of EHMT1 mutants for their ability to interact and co-localize with PRDM16 (Supplementary Figs 13 and 14). A deletion mutant of EHMT1 (772–1009), which failed to interact with PRDM16, was unable to increase PRDM16 reporter gene activity (Fig. 3e). EHMT1 with mutations in the SET domain (N1198L; H1199E) had no methyltransferase activity, but was still able to bind to and activate PRDM16. Even a mutant that lacked the SET-domain altogether (1–1009) interacted with and activated PRDM16. Thus an interaction between EHMT1 and PRDM16 seems to be required to activate PRDM16 transcriptional activity. Notably, expression of EHMT1 robustly increased the amount of PRDM16 protein, independently of its mRNA expression (Fig. 3f and Supplementary Fig. 15). This effect was due to changes in the rate of protein degradation, because cycloheximide chase experiments showed that expression of EHMT1 extended the half-life of PRDM16 protein from 8.5 to 16.5 h. The N1198L;H1199E mutant also extended the half-life of PRDM16 protein as potently as the WT form (Fig. 3g). PRDM16 protein accumulation was induced only by the EHMT1 mutants that bind to PRDM16 (Supplementary Fig. 16). EHMT1 regulates endogenous PRDM16 protein amounts in vivo (Extended Data 1). EHMT1 protein stability was not affected by PRDM16 (Supplementary Fig. 17). These results collectively suggest that EHMT1 has dual functions: that is, repressive effects on the muscle-selective gene program through its methyltransferase activity, and activation of the BAT-selective gene program through stabilization of PRDM16 protein through direct association.

Next, we examined the requirement for EHMT1 in adaptive thermogenesis in vivo. To exclude potential defects in the skeletal muscle of Ehmt1myf5 knockout mice, we generated adipose tissue-specific Ehmt1 knockout mice (Ehmt1adipo knockout) using Adiponectin-Cre mice19. Of note, 62.3% of the differentially expressed muscle/BAT-selective genes in the Ehmt1myf5 mice were similarly dysregulated in the Ehmt1adipo knockout mice (Extended Data 2). Although Adiponectin-Cre is expressed both in BAT and WAT, expression of EHMT1 is highly enriched in BAT compared with WAT (Fig. 1e). Furthermore, lipolysis capacity in the WAT of Ehmt1adipo knockout mice was indistinguishable from WT mice (Supplementary Fig. 18). EHMT1 is required for beige/brite cell development (Extended Data 3). Hence, the Ehmt1adipo knockout mice allow us to examine the role of EHMT1 in BAT/beige fat-mediated thermogenesis in vivo. As shown in Fig. 3h, rectal temperature of Ehmt1adipo knockout mice strikingly dropped within 1 h after a cold challenge to 4 °C, whereas that of control mice remained constant. Expression of BAT-selective genes in skeletal muscle20 was not altered in Ehmt1adipo knockout mice (Supplementary Fig. 19). We subsequently measured oxygen consumption rate at thermoneutrality (29–30 °C)21 in response to an activation of the β3-adrenoceptor pathway. As shown in Fig. 4a, the oxygen consumption rate of WT mice was significantly increased after administering CL316,243 whereas this induction was completely lost in knockout mice. The impaired thermogenesis in knockout mice was accompanied by higher serum amounts of free fatty acids (FFAs) (Fig. 4b). This is consistent with previous findings that BAT serves as a major sink of FFAs for heat generation22, and that reduced fatty-acid oxidation in BAT leads to an increase in amounts of serum FFA23. Indeed, fatty-acid oxidation capacity in the knockout BAT was significantly lower than in WT mice at the basal state and after administering CL316,243 (Fig. 4c). Additionally, fatty-acid uptake in the knockout BAT was reduced (Supplementary Fig. 20). These results indicate that EHMT1 is absolutely required for BAT-mediated adaptive thermogenesis and fatty-acid metabolism in vivo.

Figure 4: EHMT1 deficiency in BAT causes obesity and insulin resistance.
figure 4

a, Oxygen consumption rate of WT and Ehmt1adipo knockout mice treated with CL316,243 (0.5 mg kg−1) at thermoneutrality (n = 6). b, Amounts of serum FFA in mice treated with saline or CL316,243. c, Fatty-acid oxidation in BAT (n = 6–10). d, Body mass change under a high-fat diet at thermoneutrality (n = 16). e, Adipose tissue mass after 4-week high-fat diet (n = 16). f, Haematoxylin and eosin staining of adipose tissues. Scale bar, 100 μm. g, Glucose tolerance test in 9-week high-fat diet-fed mice (n = 9). h, Insulin tolerance test in 10-week high-fat diet-fed mice (n = 9). i, Amounts of serum insulin at the fasted and glucose-stimulated states (n = 9). j, Haematoxylin and eosin staining of liver in d. Scale bar, 50 μm. k, Amounts of liver triglyceride in j (n = 9). l, Hepatic insulin signalling as assessed by phosphorylated (S473) and total Akt amounts. Error bars, s.e.m. *P < 0.05, **P < 0.01, ***P < 0.001.

PowerPoint slide

Full size image

Lastly, we tested whether EHMT1 deficiency in BAT affects the propensity for weight gain in response to an obesogenic diet at thermoneutrality (29–30 °C), because an obesity phenotype in Ucp1 knockout mice was observed only at thermoneutrality24. As shown in Fig. 4d and Supplementary Fig. 21, Ehmt1adipo knockout mice gained significantly more body mass than WT mice without any change in food intake (Supplementary Table 1). Knockout mice had higher amounts of epididymal WAT and interscapular BAT that contained substantially larger lipid droplets than WT mice (Fig. 4e, f). A glucose tolerance test found that knockout mice showed significantly higher blood glucose concentrations than WT mice (Fig. 4g). Similarly, knockout mice showed impaired responses to insulin during an insulin tolerance test (Fig. 4h) and higher amounts of serum insulin at the fasted and glucose-stimulated states (Fig. 4i). Knockout mice showed an insulin-resistance phenotype even at ambient temperature, whereas no statistically significant difference was observed in body mass (Supplementary Fig. 22). Notably, the liver from knockout mice contained higher amounts of lipids and triglyceride (Fig. 4j, k) and showed impaired insulin signalling as assessed by phosphorylation of Akt in response to insulin (Fig. 4l and Supplementary Fig. 23). Together, these results indicate that EHMT1 deficiency in BAT leads to obesity, systemic insulin resistance and hepatic steatosis under a high-fat diet.

In conclusion, we have identified EHMT1 as an essential BAT-enriched methyltransferase that controls brown adipose cell fate, adaptive thermogenesis and glucose homeostasis in vivo. Although presence of BAT in adult humans is now widely appreciated, no mutation that causes defects in human BAT development and thermogenesis had been described except polymorphisms in UCP1 and β3-adrenoceptor genes25. Delineating the causal link between EHMT1 mutations and BAT thermogenesis will provide a new perspective in understanding the molecular control of energy homeostasis through the epigenetic pathways, which may lead to effective therapeutic interventions for obesity and metabolic diseases.

Methods Summary

Animals

All animal experiments were performed according to procedures approved by University of California, San Francisco’s Institutional Animal Care and Use Committee. Ehmt1flox17 and Adiponectin-Cre mice19 were provided by A. Tarakhovsky and E. D. Rosen. For metabolic studies, male mice in Bl6 background were fed with a high-fat diet for 4 weeks at thermoneutrality and ambient temperature.

Bioinformatics

RNA-sequencing libraries were constructed at the University of California, San Francisco Genomic Core Facility. Gene ontology enrichment analyses were performed on the differentially expressed genes (P < 0.05, the delta-method-based hypothesis test) using RefSeq as the background data set. The accession number for the data is E-MTAB-1704.

Online Methods

Animals

All animal experiments were performed according to procedures approved by University of California, San Francisco’s Institutional Animal Care and Use Committee for animal care and handling. Ehmt1flox mice and AdiponectinCre/+ mice were provided by A. Tarakhovsky and E. D. Rosen. Myf5Cre/+ mice were obtained from the Jackson Laboratory26. To analyse embryonic BAT development, Ehmt1flox/flox or Myf5-Cre+/−;Ehmt1flox/flox embryos at E18.5 or newborn mice at P1 were collected and fixed in 4% paraformaldehyde for histological analyses. The presumptive BAT depots in the interscapular region were micro-dissected for histological and RNA expression analyses. For cold exposure experiments, male Ehmt1adipo knockout mice (Adipo-Cre+/−;Ehmt1flox/flox) and body-weight-matched control mice (Ehmt1flox/flox) at 10 weeks of age were single-caged and exposed to 4 °C for 5 h. Rectal temperatures were monitored every hour using a TH-5 thermometer (Physitemp).

Metabolic studies

Whole-body energy expenditure of Ehmt1adipo knockout mice or control mice matched for body mass at 14 weeks of age was measured at thermoneutrality (29–30 °C) using a Comprehensive Lab Animal Monitoring System (Columbus Instruments). The mice were injected intraperitoneally with a β3-adrenergic receptor-specific agonist CL316,243 at a dose of 0.5 mg kg−1. For diet-induced obesity studies, male mice at 6–7 weeks of age were fed a high-fat diet (D12492, Research Diet) for 4 weeks at thermoneutrality. At the end of the experiments, serum samples were collected. Amounts of serum insulin (Millipore), triglyceride (Thermo) and FFA (Wako) were measured using commercially available kits. For glucose tolerance test experiments, male mice were fed a high-fat diet for 9 weeks. After an overnight fast, the mice were injected intraperitoneally with glucose (1 g kg−1). For insulin tolerance test experiments, male mice under a high-fat diet for 10 weeks were used. After an overnight fast, the mice were injected intraperitoneally with insulin (0.75 U kg−1). Blood samples were collected at indicated time points and amounts of glucose were measured using blood glucose test strips (Abbott). To measure liver triglyceride contents, the liver tissue (25 mg) was homogenized in 1.25 ml of Folch solution (chloroform/methanol, 2:1, v/v). Subsequently, equal amounts (0.4 ml) of chloroform and water were added to the lysate. After centrifugation at 735g for 3 min, the chloroform phase was collected and dried. The pellet was dissolved in isopropanol. Amounts of triglycerides were determined by an Infinity Triglycerides kit (Thermo).

Fatty-acid oxidation assay

WT and Ehmt1adipo knockout mice at 11 weeks old were intraperitoneally injected with saline or CL316,243. Five hours after the injection, the interscapular BAT depots were isolated. Fatty-acid oxidation assay was performed according to the protocol described by Mao et al.27 Briefly, the adipose tissues were minced to small pieces and incubated with DMEM supplemented with 1 mM pyruvate, 1% FFA-free BSA and 0.5 mM oleate. [14C]oleic acid at 1 μCi μl−1 was added for 2 h at 37 °C. After adding 70% perchloric acid into each well, CO2 was captured by Whatman paper soaked in 3 M NaOH solution for 1 h. 14C radioactivity was measured by liquid scintillation counter and normalized to tissue mass. To assess fatty-acid uptake, BAT (approximately 100 mg) was isolated from WT and Ehmt1adipo knockout mice and incubated in DMEM containing oleic acid (250 μM, Nu-Chek Prep) supplemented with [14C]oleic acid (0.25 μCi μl−1) and 10% FBS for 15 min. 14C radioactivity in the BAT explants was measured by liquid scintillation counter and normalized to the total protein content.

In vivo insulin stimulation assay

Mice were anaesthetized with Tribromoethanol (Avertin). Insulin (5 U) was injected into the inferior venae cavae. Livers were removed 2 min after the injection and lysed in lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 10% (w/v) glycerol, 100 mM NaF, 10 mM EGTA, 1 mM Na3VO4, 1% (w/v) Triton X-100, 5 μM ZnCl2, 2 mM), with protease inhibitor cocktail (cOmplete, Roche). The lysates were isolated and separated by SDS–polyacrylamide gel electrophoresis (SDS–PAGE). Akt (Pan) antibody (Cell Signaling) and Phospho-Akt (Ser473) Antibody (Cell Signaling) were used for western blotting.

Fat lipolysis assay

Epididymal fat pads were collected and digested in a digestion buffer (121 mM NaCl, 4.9 mM KCl, 1.2 mM MgSO4, 0.33 mM CaCl2, 12 mM HEPES) containing collagenese D (1.5 U ml−1), dipase II (2.4 U ml−1), 3 mM glucose and fatty-acid-free 1% BSA (Akron Biotech). After digestion for 1 h at 37 °C with gentle shaking, adipocytes were filtrated through nylon mesh and centrifuged at 186g for 5 min. Floating adipocytes were collected and incubated in DMEM containing 10% FBS in the presence or absence of isoproterenol (1 μM) for 1.5 h at 37 °C. Glycerol release into the media was determined using a free glycerol reagent (Sigma). Amounts of glycerol were normalized to the total protein content of the primary adipocytes by using a Pierce BCA Protein Assay reagent (Thermo Scientific).

Cell culture

Immortalized brown fat cells were isolated from the interscapular BAT of WT mice at P1–P3. Mouse embryonic fibroblasts have been described previously2. HEK293 cells and C2C12 cells were obtained from the American Type Culture Collection. Adipocyte differentiation in C2C12 cells was induced by treating confluent cells with DMEM containing 10% FBS, 0.5 mM isobutylmethylxanthine, 125 μM indomethacin, 2 μg ml−1 dexamethasone, 850 nM insulin, 1 nM T3 and 0.5 μM rosiglitazone. Two days after induction, cells were switched to the maintenance medium containing 10% FBS, 850 nM insulin, 1 nM T3 and 0.5 μM rosiglitazone. For cAMP treatment, cells were incubated with 10 μM forskolin for 4 h. Myocyte differentiation in C2C12 myoblasts was induced by treating cells in DMEM containing 2% horse serum. For beige cell differentiation in culture, the stromal vascular (SV) fraction was isolated from Ehmt1flox/flox mice and plated in collagen-coated plates (BD Biosciences). Cells were differentiated in the absence or presence of rosiglitazone at 0.5 μM according to the previous paper28.

DNA constructs and viruses production

Deletion mutants of Flag-tagged PRDM16 and GST-fused PRDM16 fragments (1–223, 224–454, 455–680, 680–880, 881–1038 and 1039–1176) were described previously16. EHMT1 expression constructs were gifts from Y. Shinkai18 and E. Hara29. EHMT1 was cloned to pMSCV-puro vector for retroviral expression. The sequences used for retroviral shRNA expression vectors targeting EHMT1 were 5′-CGC TAT GAT GAT GAT GAA TAA-3′ (shEHMT1-1) and 5′-GAG GAT AGT AGG ACT TCT AAA-3′ (shEHMT1-2). The corresponding double-stranded DNA sequences were ligated into pSUPER-Retro (GFP-Neo) (Oligoengine) for retroviral expression. For retrovirus production, Phoenix packaging cells were transfected at 70% confluence by calcium phosphate method with 10 μg retroviral vectors. After 48 h, the viral supernatant was collected and filtered. Cells were incubated overnight with the viral supernatant and supplemented with 6 μg ml−1 polybrene. Subsequently, puromycin (PRDM16 and EHMT1), hygromycin (C/EBP-β) or G418 (shRNAs) were used for selection.

Gene expression analysis

Total RNA was isolated from tissues using Trizol (Invitrogen) or RiboZol reagents (AMRESCO) following the manufacturers’ protocols. Quality of RNA from all the samples was checked by spectrophotometer. Reverse transcription reactions were performed using an IScript complementary DNA (cDNA) synthesis kit (Bio-Rad). The sequences of primers used in this study can be found in Supplementary Table 2. Quantitative reverse transcriptase PCR (qRT–PCR) was performed with SYBR green fluorescent dye using an ABI ViiA7 PCR machine. Relative mRNA expression was determined by the ΔΔ-Ct method using TATA-binding protein as an endogenous control to normalize samples.

RNA-sequencing and gene ontology analysis

Total RNA was isolated from the presumptive interscapular BAT depots of WT and Ehmt1myf5 knockout mice at P1 or from the interscapular BAT depots of WT and Ehmt1adipo knockout mice at 12 weeks old. RNA-sequencing libraries were constructed from 50 ng of total RNA from the Ehmt1adipo knockout and Ehmt1myf5 knockout BAT using an Ovation RNA-sequencing system version 2 (NuGEN). mRNA was reverse transcribed to cDNAs using a combination of random hexameric and a poly-T chimaeric primer. The cDNA libraries were subsequently amplified by single primer isothermal amplification30 using an Ultralow DR library kit (NuGEN) according to the manufacturer’s instructions. The qualities of the libraries were determined by Bioanalyzer (Agilent Technologies). Subsequently, high-throughput sequencing was performed using a HiSeq 2500 instrument (Illumina) at the University of California, San Francisco Genomics Core Facility. RNA-sequencing reads for each library were mapped independently using TopHat version 2.0.8 against the University of California, Santa Cruz (UCSC) mouse genome build mm9 indexes, downloaded from the TopHat website (http://tophat.cbcb.umd.edu/igenomes.shtml). The mapped reads were converted to fragments per kilobase of exon per million fragments mapped by running Cuffdiff 2 (ref. 31) on the alignments from TopHat and the UCSC coding genes to estimate amounts of gene and isoform expression. Based on the list of genes that showed significant difference (P < 0.05, the delta-method-based hypothesis test) from the RNA-sequencing data, enrichment of the Gene Ontology biological process terms (GO FAT category) was analysed using the Gene Set Enrichment Analysis (GSEA) program, according to the method described by the previous paper32. RNA-sequencing reads have been deposited in ArrayExpress (http://www.ebi.ac.uk) under accession number E-MTAB-1704.

Immunocytostaining

Differentiated C2C12 myotubes or COS7 cells expressing green fluorescent protein (GFP)–PRDM16 and EHMT1 constructs were fixed with 4% paraformaldehyde for 10 min at room temperature (24 °C), rinsed with PBS and then exposed to 0.2% Triton X-100 in PBS for 5 min. The cells were subsequently incubated with anti-MF20 mouse antibody (DSHB, 1:50) for MHC and with Flag antibody (M2, 1:200) for EHMT1. After washing with PBS, Alexa 594-labelled anti-mouse IgG (1:800) was added as a secondary antibody.

Protein interaction analyses

Immortalized brown fat cells stably expressing Flag-tagged WT, PR-domain deletion mutant and ZF-1 deletion mutant of PRDM16 or an empty vector were grown to confluence2. Nuclear extracts were isolated from these cells and incubated with Flag M2 agarose beads, washed in a binding buffer (180 mM KCl) and subsequently eluted either by 3× or by 1× Flag peptides (0.2 μg ml−1). The eluted proteins were subjected to histone methyltransferase assay or to reverse-phase LC–MS/MS for peptide sequencing using a high-resolution hybrid mass spectrometer (LTQ-Orbitrap, Thermo Scientific) with TOP10 method. Data obtained was annotated using the IPI mouse database33. Proteins were considered significantly identified with at least two unique valid peptides, and the false discovery rate was estimated to be 0% using the target-decoy approach34. To confirm the interaction between PRDM16 and EHMT1 in brown adipocytes, the immunopurified complex was purified using anti-EHMT1 (R&D Systems) or Flag antibody (M2) and subjected to 4–12% SDS–PAGE. Rabbit polyclonal PRDM16 antibody35 or EHMT1 antibody (R&D Systems) was used for western blotting. COS7 cells expressing haemagglutinin (HA)-tagged PRDM16 or deletion fragments of Flag-tagged EHMT129 were collected 48 h after transfection. Total cell lysates were incubated overnight at 4 °C with Flag M2 agarose beads, washed and eluted by boiling. The immunoprecipitants were analysed by western blot analysis using HA antibody (Roche). For in vitro binding assays, various fragments of the GST–fusion PRDM16 fragments were purified as previously described16. 35S-labelled proteins (EHMT1, EHMT2, CtBP1, C/EBP-β) were prepared with a TNT reticulocyte lysate kit (Promega). Equal amounts of GST fusion proteins (2 μg) were incubated overnight at 4 °C with in vitro translated proteins in a binding buffer containing 20 mM HEPES pH 7.7, 300 mM KCl, 2.5 mM MgCl2, 0.05% NP40, 1 mM DTT and 10% glycerol. The sepharose beads were then washed five times with the binding buffer. Bound proteins were separated by SDS–PAGE and analysed by autoradiography.

Histone methylation assay

The PRDM16 transcriptional complex was immunopurified from nuclear extracts of brown adipocytes using Flag M2 agarose or IgG (negative control). The immunoprecipitants were incubated with 2 µg of core histone (Millipore) with [3H]S-adenosyl-methionine at 30 °C for 1 h. Subsequently, the reaction was stopped by addition of sample buffer. Core histone was resolved by 4–12% SDS–PAGE and detected by autoradiography or by scintillation counter.

Chromatin immunoprecipitation assay

After cross-linking with 1% formaldehyde at room temperature (24 °C) for 10 min, total cell lysates from brown adipocytes were sonicated to shear the chromatin, and immunoprecipitated overnight at 4 °C using antibodies for H3 di-methyl and tri-methyl K9 (Abcam), acetyl-H3K9/K14 (Millipore), pan-H3 (Cell Signaling), EHMT1 (R&D Systems) or IgG (Santa Cruz). After extensive washing, the immunoprecipitants were eluted with 2% SDS in 0.1 M NaH2CO3. Cross-linking was reversed by heating at 65 °C overnight. Input DNA and immunoprecipitated DNA were purified by a PCR purification kit (Qiagen) and analysed by qRT–PCR using SYBR green fluorescent dye (Bio-Rad). Enrichment of each protein was calculated as a ratio to input DNA. Primer sequences used in the chromatin immunoprecipitation assays are provided in Supplementary Table 2.

Protein stability assay

COS7 cells expressing HA-tagged PRDM16 and EHMT1 or vector control were incubated with a medium containing 60 μg ml−1 cycloheximide for up to 24 h. Total cell lysates were isolated and separated by SDS–PAGE. Horseradish peroxidase-conjugated HA antibody (Sigma) and β-actin (Sigma) were used for western blotting. Image J software was used to quantify the intensity of signals. Half-life of the protein was estimated by regression analysis.

Reporter gene assay

A luciferase reporter gene controlled by PPAR-γ binding sites (3× DR1-Luciferase) was transiently transfected with PPAR-γ/RXR-α, PRDM16 and EHMT1 expression plasmids in COS7 cells using Lipofectamine 2000 (Invitrogen). Forty-eight hours after the transfection, cells were collected and reporter gene assays used the Dual Luciferase Kit (Promega). Transfection efficiency was normalized by measuring expression of Renilla luciferase.

Cellular respiration assay

Immortalized brown adipocytes were transduced with retroviral shEHMT1 (shEHMT1-1) or scramble control and induced to differentiate. Brown adipocytes expressing EHMT1 or vector control were also differentiated under a pro-adipogenic condition. At day 6 of differentiation, oxygen consumption was measured as previously described36. Oligomycin was used to determine uncoupled respiration. In addition, antimycin A was added at the end of experiments to determine non-mitochondrial cellular respiration. For cAMP-induced respiration assays, fully differentiated brown adipocytes were incubated with 0.5 mM dibutyryl cyclic AMP for 12 h before measuring oxygen consumption.

Statistical analyses

Statistical analysis used JMP version 9.0 (SAS Institute). Two-way repeated-measures analysis of variance was applied to determine the statistical difference in glucose tolerance test, insulin tolerance test, body mass gain and rectal temperatures between genotypes. Effect size and power analysis were done by the pwr.t.test function of the R statistics package. Other statistical comparisons were assessed by an unpaired Student’s t-test. P < 0.05 was considered significant throughout the study.