Epigenetic Regulation of Adipogenesis by PHF2 Histone Demethylase

A bacterial artificial chromosome (BAC) DNA containing mouse Phf2 (BAC clone RP23-114C14) was obtained from the BAC-PAC Resources Center. LoxP was inserted between exons 6 and 7 of Phf2 using the Escherichia coli-based BAC modification system (12). Modification cassettes were generated by PCR amplification of the PL452 vector with the following primers:5′-TATATAAGGAGCACTTGGGACCAGTGACATACATGTGTCTAATGTCCTGAGAATTCATTCCGATCATATTCAATAACCC-3′ and 5′-ACTCTGGACACTAGGTGACCCAGTGGCCTCTCCTAATAGTTAATGAGCTCGAACTAGTGGATCCCCTCGA-3′. LoxP-inserted fragments of mouse Phf2 were subcloned into the pBSIIKS+ vector using the E. coli-based BAC recombination system. Modification cassettes and retrieval cassettes were generated by PCR amplification of the pBSIIKS+ vector with primers corresponding to each promoter. LacZ-PGK-Neo was digested from the pNTR-lacZ-PGK-neo-lox vector and inserted into the pMC1DTpA vector that contained DT-A. Oligonucleotides of loxP and frt were inserted into this vector (cassette vector). LoxP-inserted fragments of mouse Phf2 were inserted into the cassette vector to form the final Phf2 knockout construct. The Phf2 knockout construct was linearized by SacII and was introduced into M1 mouse embryonic cells (RIKEN) by electroporation and screened by genomic Southern blotting. Chimeric mice were generated by aggregation of embryonic stem cells with eight cell embryos of BDF1 mice. Phf2^Z/+ mice were generated by crossing Phf2^Nf/+ mice with CMV-Cre mice (13). Phf2^fl/+ mice were generated by crossing Phf2^Nf/+ mice with Flpe deleter strain ACTB-Flpe mice (Jackson Laboratory). Mice with Phf2^Nf/+, Phf2^Z/+, and Phf2^fl/+ were maintained by backcrossing to C57BL/6J mice under a specific pathogen-free environment. All animals were maintained according to the protocol approved by the Animal Care and Use Committee of The University of Tokyo.

Cre-ERT2 transgenic mice were provided by Dr. Daniel Metzger (14). Cre-ERT2 transgenic mice were crossed with Phf2^fl/+ mice to generate Phf2^fl/fl;Cre-ERT2 mice. Genotyping was performed by PCR using corresponding primers. Sequences of primers were as follows: P1, 5′-CACCCTCTGTGTCCTCCTGT-3′; P2, 5′-CAGTTCTCTTAGCTCCCCCTTT-3′; P3, 5′-GACAGGAAGCCAAGGAGATG-3′; P4, 5′-GACAGCCTGGTCAGGTGAAT-3′; and P5, 5′-TGTAATACACCTGGGGCTCA-3′. Cre-ERT2 transgenic mice were identified by amplification of the Cre allele using primers 5′-TTACGGCGCTAAGGATGACT-3′ and 5′-TTGCCCCTGTTTCACTATCC-3′.

Mice were killed by cervical dislocation. Epididymal, gonadal, lumbar subcutaneous, and mesenteric white adipose tissues (WATs) were carefully separated from testis, uterus, skin, and intestines, respectively. Wet weights of these WATs were measured.

Food was removed 2 h before glucose (2 g/kg) or insulin (1 units/kg) was administered by intraperitoneal injection. Blood samples were collected from the tail vein at various times after the glucose load, as indicated. Plasma glucose was immediately determined on an Accu-Chek Aviva (Roche).

3T3-L1 mouse fibroblasts were maintained and differentiated into adipocytes as previously described (15). Mouse primary embryonic fibroblasts were isolated on embryonic day 13.5 and cultured in Dulbecco’s modified Eagle’s medium with 10% FBS.

FLAG-tagged, full-length cDNAs of human CEBPA or PPARG were inserted into pcDNA3 vectors (Invitrogen).

Protein levels were determined by blotting with anti-FLAG antibody (Sigma-Aldrich), anti-CEBPA antibody (Santa Cruz Biotechnology), and anti-PHF2 antibody (5). ECL Plus (Amersham) was used for detection of the immunoreactive bands.

Total RNA was isolated with TRIzol reagent (Invitrogen) according to the manufacturer’s protocol. First-strand cDNA was synthesized from total RNA using PrimeScript RT Master Mix (Takara) and subjected to real-time PCR using KAPA SYBR Fast qPCR Kits (Kapa Biosystems) with Thermal Cycler Dice (Takara) according to the manufacturers’ instructions. The following primers were used: mouse Phf2, 5′-TACTGCCTGATCTGTGTGAAGGA-3′ and 5′-CTGGCCGGATGAGATAGAAGA-3′; mouse Pparg, 5′-TTAACTGCCGGATCCACAAAA-3′ and 5′-TGAGACATCCCCACAGCAAG-3′; mouse Cebpa, 5′-TGCCAAACTGAGACTCTTCACTAAC-3′ and 5′-CCCAAACATCCCTAAACCAAAA-3′; mouse Adipoq, 5′-CAAGGCCGTTCTCTTCACCTAC-3′ and 5′-TCCCCATCCCCATACACCT-3′; mouse Fabp4, 5′-CAGCGTAAATGGGGATTTGG-3′ and 5′-GCTCTTCACCTTCCTGTCGTCT-3′; mouse Plin2, 5′-GAGCTGGAGATGGAAGCAAAA-3′ and 5′-GTGATAAGCCCGAGAGCAGAG-3′; mouse Cd36, 5′-AAGAACAGCAGCAAAATCAAGG-3′ and 5′-AGACAGTGAAGGCTCAAAGATGG-3′; mouse Lpl, 5′-TGGGACTGAGGATGGCAAG-3′ and 5′-GGCAGGGTGAAGGGAATGT-3′; and mouse Arbp, 5′-GCTCCAAGCAGATGCAGCA-3′ and 5′-CCGGATGTGAGGCAGCAG-3′.

Platinum-E cells were transfected with either pSuper-retro-puro-shLacZ or pSuper-retro-puro-shPHF2 (5′-GCCTGACTCAGTTTCCCAACT-3′) using Lipofectamine (Invitrogen). Forty-eight hours after transfection, the medium containing retroviruses was harvested, filtered, and transferred to 3T3-L1 cells. Infected cells were selected with 1 μg/mL puromycin.

Chromatin immunoprecipitation (ChIP) assays were carried out essentially as previously described (16). For immunoprecipitation, antibody against CEBPA (sc-61; Santa Cruz Biotechnology) or PHF2 (5) was used. The precipitated DNA fragments were amplified by real-time PCR. The following primers were used: mouse Cebpa-CEBPRE, 5′-TTGCGCCACGATCTCTCTC-3′ and 5′-CTTAGAGCCCGCCTTCTCCT-3′; mouse Pparg-CEBPRE, 5′-CCACTGGTGTGTATTTTACTGCAA-3′ and 5′-GTTCTGTGAGGGGCGTGAA-3′; mouse Fabp4-CEBPRE, 5′-CATTGCCAGGGAGAACCAA-3′ and 5′-CCATGTGACTGTAGGAGTGACCAA-3′; and mouse negative control region, 5′-CAGACATGTCAAATCAAGAAGACAG-3′ and 5′-ACTTTGGAGGAAGAGGCAGAAA-3′.

These procedures were conducted according to the Percellome method (17). Briefly, mRNA expression values were normalized to the cell numbers in each sample by adding external spike mRNAs to them in proportion to the genomic DNA concentration and using the spike RNA quantity data as a dose-response standard curve for each sample. Total RNAs were purified from epididymal adipose tissue using an RNeasy Mini kit (Qiagen). First-strand cDNAs were synthesized using SuperScript II reverse transcriptase (Invitrogen). After second-strand synthesis, the double-stranded cDNAs were purified using a GeneChip (Affymetrix) Sample Cleanup Module and labeled by in vitro transcription using a BioArray HighYield RNA Transcript Labeling Kit (Enzo Life Sciences). The labeled cRNA was then purified using a GeneChip Sample Cleanup Module. Purified cRNA were hybridized with GeneChip Mouse Genome 430 2.0 Array. Washing and staining were performed in a GeneChip Fluidics Station using the appropriate antibody amplification, washing, and staining protocols. The phycoerythrin-stained arrays were scanned as digital image files, which were analyzed with GeneChip Operating Software. The expression data were converted to copy numbers of mRNA per cell by the Percellome method and analyzed using Percellome software.

Paraffin sections of epididymal WAT were processed for hematoxylin-eosin staining and observed by light microscopy. Adipocyte areas were measured using OsteoMeasure.

The number of adipocytes in epididymal WAT was estimated as previously described (18). Briefly, adipose tissue was minced and fixed at 37°C for 96 h in 1.88% osmium tetroxide. After washing twice with PBS, it was replaced with 8 mol/L urea at room temperature for 48 h. Urea-free particles were washed through a 180-μm net (Millipore), trapped on a 10-μm net (Millipore), and suspended using 0.01% Triton X-100 in PBS. Particles were counted with a counting chamber.

Cells were stained with oil red O, and the quantification was performed as previously described (19). Briefly, cells were fixed with 10% formalin and stained in a working solution of oil red O. Isopropyl alcohol was added to the stained culture dishes, and the extracted dye was monitored spectrophotometrically at 510 nm.

Data were analyzed by two-tailed Student t test, Pearson χ² test, log-rank test, or two-way repeated-measures ANOVA. For all graphs, data are presented as mean ± SEM. Statistical significance was accepted at P < 0.05.

Figure 1A shows the strategy to generate the mutant Phf2 mutant allele. Successful insertion of LacZ-PGK-Neo^R cassettes and loxP at sites flanking exon 6 of the Phf2 allele was confirmed by Southern blotting (Fig. 1B). Cre-mediated recombination excised exon 6 and PGK-Neo^R from the mutant allele, resulting in the LacZ allele. P1, P2, and P3 primers were used to differentiate the wild-type and LacZ alleles (Fig. 1C). Flpe-mediated recombination excised LacZ and PGK-Neo^R from the mutant allele, resulting in a floxed allele. P4 and P5 primers were used to differentiate the wild-type and floxed alleles (Fig. 1D). Cre-mediated recombination excised exon 6 from the floxed allele, resulting in a null allele. Exon 6 corresponds to part of the JmjC domain of Phf2, and its deletion is expected to result in an alteration of the reading frame of the Phf2 transcript. As expected, the PHF2 protein was not detected in mouse embryonic fibroblasts of Phf2^Z/Z (Fig. 1E).

Phf2^Z/Z mice were born according to expected Mendelian ratios (Fig. 2A). However, <30% of Phf2^Z/Z pups were alive when they reached 2 weeks of age (Fig. 2B). In fact, daily observation revealed that ∼70% of Phf2^Z/Z mice died within 3 days of birth (P = 3.0 × 10⁻⁵ compared with wild type) (Fig. 2C). However, after 3 days of age, the remaining pups retained viability and lived for no less than 6 months (data not shown). Phf2^Z/Z mice had progressively reduced growth compared with littermate controls until 10 days after birth, although there was little difference in body weight among newborn mice of different genotypes (Fig. 2D). Two weeks after birth, Phf2^Z/Z mice grew in a manner similar to wild-type or heterozygous littermates, but with significant differences in body weight (Fig. 2E). The extent of these weight reductions was not significantly different between males and females (two-way repeated-measures ANOVA). These data suggest that PHF2 is dispensable for the survival of embryos but is necessary for survival and growth in the neonatal period.

Next, we surveyed the weight and appearance of each tissue present in male Phf2^Z/Z mice at 5 weeks of age. As mentioned previously, Phf2^Z/Z mice had decreased body weights at that time (Fig. 3A). Phf2^Z/Z mice exhibited significantly shortened body lengths (Fig. 3A). Various tissues of Phf2^Z/Z, including brown adipose tissue (BAT), appeared normal and weighed the same as in control littermates. However, WAT and brain showed a different trend. The brain weights were significantly increased compared with wild-type littermates, the potential significance of which will be discussed later (Fig. 3A). The weight of epididymal WAT was only 50% of that found in control littermates, even when normalized by body weight. Weight reduction was also observed in subcutaneous WAT to an extent similar to that in epididymal WAT (P = 0.06) but to a lesser extent in mesenteric WAT (Fig. 3B). Additionally, female Phf2^Z/Z mice exhibited a similar reduced weight of gonadal WAT as observed in male mice (Fig. 3C). However, these weight reductions in adipose tissues seemed to be limited to a young age because there was no significant difference of WAT weights between Phf2^Z/Z and wild-type littermates when the mice reached 8 weeks of age (Supplementary Fig. 1). To elucidate whether the decreased weight of WAT was caused by a reduction of lipid droplets per cell or adipocyte number, we measured adipocyte size and number in epididymal WAT of Phf2^Z/Z and wild-type littermates. Evaluation of the mean adipocyte area in the section of WAT revealed that Phf2^Z/Z mice have smaller adipocytes than do control littermates (Fig. 3D). Furthermore, counting the number of adipocytes in WAT revealed that Phf2^Z/Z mice have fewer adipocytes in adipose tissue (Fig. 3E). These data suggest that the decreased weight of WAT in Phf2^Z/Z mice was caused by a reduction in both the size and the number of adipocytes. Because impaired adipogenesis often results in decreased size (20) and a decreased number of adipocytes, we used microarray analysis to compare the gene expression profiles in WAT between Phf2^Z/Z mice and control littermates. As a result, the expression of various genes associated with adipogenesis, such as Pparg, Cebpa, Fabp4, Adipoq, LPL, Plin2, and Cd36, tended to be decreased in Phf2^Z/Z mice (Supplementary Fig. 2). The expression levels of more than half of these genes were significantly reduced when confirmed by real-time quantitative PCR (qPCR) (Fig. 3F). These data suggest that decreased WAT weight in Phf2^Z/Z mice resulted, at least in part, from impaired adipogenesis in these mice. Finally, we measured insulin sensitivity in Phf2^Z/Z mice because adipose tissue is well known to be involved in glucose metabolism. There was no significant change in the glucose or insulin tolerance tests (Supplementary Fig. 3), suggesting that reduced WAT weight did not affect insulin sensitivity in mice at 5 weeks of age.

To elucidate the role of PHF2 in adipogenesis, we generated Phf2^fl/fl;Cre-ERT2 mice by crossing Phf2 floxed mice with Cre-ERT2 mice. Stromal vascular cells (SVCs) were then obtained from Phf2^fl/fl;Cre-ERT2 mice or control Cre-ERT2 mice and treated with 4-hydroxytamoxifen (4-OHT) to induce Cre-mediated excision of Phf2. As expected, treatment with 4-OHT efficiently reduced mRNA expression of Phf2 in Phf2^fl/fl;Cre-ERT2 but not in control Cre-ERT2 mice (Fig. 4C). When SVCs were treated with a differentiation cocktail to induce their differentiation into adipocytes, differentiation was impaired in 4-OHT-treated SVCs from Phf2^fl/fl;Cre-ERT2 mice as shown by oil red O staining (Fig. 4A and B). This was also confirmed by decreased expression of adipogenic marker genes in 4-OHT–treated SVCs from Phf2^fl/fl;Cre-ERT2 mice (Fig. 4C). Next, we generated 3T3-L1 cell lines in which PHF2 was stably knocked down by retrovirus carrying Phf2-targeted short hairpin RNA. Infection of retrovirus carrying shPhf2 successfully decreased the expression of Phf2 (Fig. 4F). In accordance with the results from SVCs, short hairpin RNA-mediated knockdown of Phf2 resulted in impaired adipogenesis in the 3T3-L1 cell line as assessed by oil red O staining (Fig. 4D and E) and expression of adipogenic marker genes (Fig. 4F).

The results indicated that PHF2 promoted adipogenesis. Moreover, we show that PHF2 works as a coactivator for several transcription factors through H3K9me2 demethylation (5). Therefore, we hypothesized that PHF2 coactivates transcription factors that promote adipogenesis. Among such transcription factors, PPARG and CEBPA are the master regulators for adipogenesis (21). First, we assessed whether PHF2 could physically interact with PPARG or CEBPA. When transfected into HEK293 cells, FLAG-CEBPA (but not FLAG-PPARG) was coimmunoprecipitated with endogenous PHF2 (Fig. 5A). Conversely, FLAG-PHF2 was coimmunoprecipitated with endogenous CEBPA in 3T3-L1 adipocytes (Fig. 5B). In accordance with these results, ChIP analysis revealed that PHF2 was recruited to known CEBPRE (CEBP responsive elements) in the promoter regions of Cebpa, Pparg, and Fabp4 in differentiated 3T3-L1 adipocytes (Fig. 5C). These results indicate that PHF2 might work as a coactivator for CEBPA. PHF2 seemed to be necessary for recruitment of CEBPA to chromatin because the recruitment of CEBPA to CEBPRE was impaired in 3T3-L1 in which Phf2 had been knocked down (Fig. 5D). Moreover, modification of H3K9me2 in these CEBPRE was significantly increased in Phf2 knocked-down 3T3-L1 adipocytes (Fig. 5E). These data indicate that PHF2 can play a role as a coactivator, positively regulating adipogenic gene expression with CEBPA through H3K9me2 demethylation near CEBPA-binding regions.

To investigate the role of PHF2 in vivo, we generated Phf2 knockout mice. Phf2 knockout mice showed partial neonatal death, growth retardation, and reduced body weight. Reduced body weight seems to be mainly related to growth retardation and reduced lean mass in the knockout mice rather than reduced WAT mass because the body weight of Phf2 knockout mice is still significantly lower than that of wild type at 8 weeks of age (Fig. 2E) when the weight of WAT is not different between Phf2 knockout and wild type (Supplementary Fig. 1). The reason for this phenotype is not clear. PHF2 has been reported to be highly expressed in the neural tube and dorsal root ganglia (22), and Phf2 represents a candidate gene for hereditary sensory neuropathy type I (HSN1) (23). In fact, the brain weights of Phf2 knockout mice were larger than wild-type littermates. Taken together, we speculate that Phf2 knockout mice may exhibit partial neonatal death attributable to defects in the central nervous system. Conditional deletion mutants using brain-specific Cre mice would clarify the precise roles of PHF2 in brain development.

On the other hand, subsequent study revealed that PHF2 plays an important role in adipogenesis. Phf2 knockout mice produced lipoatrophy in which adipocytes were decreased in size and number. It seemed to be limited to a young age; for example, in Klf5 knockout mice, atrophic changes in adipose tissue were abolished until 4 weeks of age (20). Although the reduced WAT phenotype in Phf2 knockout mice disappeared after the mice reached 8 weeks of age and Phf2 was systemically knocked out in these mice, this phenotype should be the result of a cell-autonomous mechanism because conditional knockout of Phf2 in primary SVCs obtained from Phf2^fl/fl;Cre-ERT mice resulted in impaired adipogenesis. Moreover, PHF2 interacts with CEBPA, one of the master adipogenic regulators, and is recruited to CEBPRE. In addition, the lipoatrophic changes in Phf2 knockout mice were stronger in epididymal and subcutaneous WAT than in mesenteric WAT, suggesting that PHF2 may be a possible candidate factor that determines regional variations in adipogenesis reported previously (24). Although adipose tissue is well known to be involved in glucose homeostasis, there was no significant change in glucose metabolism in Phf2 knockout mice (Supplementary Fig. 3). There are two possible reasons for this result. First, 5 weeks of age is too young to develop insulin resistance, and second, only slight lipoatrophic change was observed in mesenteric WAT, the mass of which is strongly correlated with insulin resistance. It is meaningful to evaluate glucose metabolism of Phf2 knockout adult mice under the treatment of a high-fat diet to clarify whether PHF2 plays a role in diabetes.

It is well known that factors that increase cAMP, such as isobutylmethylxanthine, strongly accelerate adipogenesis (25). Elevation of cAMP is known to lead to suppression of Wnt10b (26) and Sp1 (27), induction of C/EBPβ (28), and production of PPARγ ligands (29). Moreover, cAMP signaling is mediated by two major pathways, PKA and Epac (exchange proteins directly activated by cAMP), that synergistically induce adipogenesis (30). However, it is not fully understood how PKA stimulates adipogenesis. PHF2 seems to be one possible candidate because it is reportedly activated through phosphorylation by PKA (5), and the current study shows that PHF2 stimulates adipogenesis. It seems that Phf2 was not transcriptionally regulated during adipogenesis (Supplementary Fig. 4). However, PHF2 might be activated by PKA-mediated phosphorylation by a cAMP inducer in the adipogenic differentiation cocktail.

On the other hand, the weight reduction of BAT was not observed in Phf2 knockout mice, although PKA and CEBPA were also reported to be necessary for the development of BAT (31). Recently, it was reported that brown adipocytes were differentiated from Myf5-positive precursor cells, which are more closely related to myoblasts rather than to the white preadipocytes (32). Thus, PHF2 may play a role in white preadipocytes but not in brown preadipocytes or myoblasts because the histone demethylases seem to be involved in differentiation in specific cell types.

PHF2 reportedly interacts with ARID5B, which is necessary for the coactivator function of PHF2 (5). Because phenotypes of systemic Arid5b knockout animals have already been reported (33–35), it is worth comparing the phenotypes between Arid5b knockout and Phf2 knockout mice. Of note, most Arid5b knockout mice die within 24 h of birth, but some pups survive (33). The surviving Arid5b knockout animals show dramatically reduced body weight in neonates and adults. The WAT of Arid5b knockout mice weighed less than controls because of a reduction in the amount of lipid per cell. Because the phenotype of Arid5b knockout mice is close to that of Phf2 knockout mice, it is likely that PHF2 and ARID5B work together in multiple organs, including adipose tissue.

Transcriptional control of the adipocyte lineage has been studied extensively (21,36). However, little is known about the role of histone demethylases during adipogenesis, even though several studies have clarified the roles of histone deacetylase HDAC1 (37,38) and HDAC3 (39) and histone demethylase LSD1 (40) in adipogenesis. In the current study, we found that histone demethylase PHF2 plays an important role in adipogenesis in vivo. Notably, several studies reported the role of histone demethylation in cell differentiation; for example, LSD1 controls pituitary terminal cell-type differentiation (6), JMJD3, an H3K27me3 demethylase, potentiates epidermal differentiation (41); and JMJD1A, an H3K9 demethylase, potentiates smooth muscle cell differentiation (42). In general, histone demethylases may play roles in cell differentiation in specific cell types. We suggest that PHF2 may be a significant histone demethylase in adipocytogenesis.

Adipose tissue plays an essential role in energy homeostasis. In mammals, WAT stores excess energy as triglycerides from fatty acids imported from circulating lipoproteins. Moreover, recent studies established adipose tissue as an active endocrine organ that secretes various humoral factors called adipokines that work in various physiological pathways, such as feeding, insulin resistance, inflammation, and atherogenesis (43). Thus, clarifications of precise molecular mechanisms that control adipose tissue development should improve our understanding of the pathogenesis and pathophysiology of the metabolic syndrome, diabetes, and other metabolism-related diseases. In this study, PHF2 appears to be a novel molecule that controls adipogenesis in vivo. Modulation of enzymatic activities has been a good target of small molecules. In fact, some drugs that can modulate the activities of histone-modifying enzymes have been developed for clinical use (44–46). Thus, pharmacological modulation of the histone demethylase activity of PHF2 may be a new target in the treatment of human lipodystrophies or adipocyte hyperplasia in diet-induced obesity.

This work was supported by Grants-in-Aid 22790848 from the Ministry of Education, Culture, Sports, Science and Technology (to Y.O.) and a grant for specially promoted research from the Japan Society for the Promotion of Science (to S.K.).

The funder played no role in the conduct of the study, collection of data, management of the study, analysis of data, interpretation of data, or preparation of the manuscript.

No potential conflicts of interest relevant to this article were reported.

Y.O. wrote the manuscript and analyzed data. F.O., T.M., I.T., S.K., and Y.I. designed the experiments. F.O. and Y.I. reviewed and edited the manuscript. K.I. and J.K. performed the microarray analyses. Y.I. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

The authors thank Yoko Yamamoto, Kazuki Inoue, and Erina Inoue, Laboratory of Epigenetic Skeletal Diseases, Institute of Molecular and Cellular Biosciences, The University of Tokyo, Japan, for technical support in the generation of knockout mice.