Crucial role of iron in epigenetic rewriting during adipocyte differentiation mediated by JMJD1A and TET2 activity

Abstract Iron metabolism is closely associated with the pathogenesis of obesity. However, the mechanism of the iron-dependent regulation of adipocyte differentiation remains unclear. Here, we show that iron is essential for rewriting of epigenetic marks during adipocyte differentiation. Iron supply through lysosome-mediated ferritinophagy was found to be crucial during the early stage of adipocyte differentiation, and iron deficiency during this period suppressed subsequent terminal differentiation. This was associated with demethylation of both repressive histone marks and DNA in the genomic regions of adipocyte differentiation-associated genes, 　including Pparg, which encodes PPARγ, the master regulator of adipocyte differentiation. In addition, we identified several epigenetic demethylases to be responsible for iron-dependent adipocyte differentiation, with the histone demethylase jumonji domain-containing 1A and the DNA demethylase ten-eleven translocation 2 as the major enzymes. The interrelationship between repressive histone marks and DNA methylation was indicated by an integrated genome-wide association analysis, and was also supported by the findings that both histone and DNA demethylation were suppressed by either the inhibition of lysosomal ferritin flux or the knockdown of iron chaperone poly(rC)-binding protein 2. In summary, epigenetic regulations through iron-dependent control of epigenetic enzyme activities play an important role in the organized gene expression mechanisms of adipogenesis.


INTRODUCTION
Iron is closely associated with the pathophysiology of obesity ( 1 ). For example, a low-iron diet as well as administration of iron chelators significantly pre v ent diabetes in rodent models of obesity ( 2 ). It has also been reported that iron chelators inhibit adipogenesis (3)(4)(5). Howe v er, the mechanism underlying the iron-obesity interrelationship is complex, because whole-body iron homeostasis is affected by the systemic crosstalk of multiple organs. Ther efor e, elucidation of the cell-autonomous molecular actions of iron is r equir ed to understand the iron-dependent molecular mechanisms of obesity.
Iron is an essential metal that is r equir ed for a variety of biological processes, including fuel oxidation and electron transport in mitochondria, as well as for the activation of various 2-o x oglutarate-dependent dio xygenases (2OGDs) (6)(7)(8). The 2OGDs include the ten-ele v en translocation (TET) enzymes, alkB homolog (ALKBH) proteins, and jumonji C (JmjC)-domain containing demethylases, most of w hich catal yze the demethylation of DN A, RN A and histones, respecti v ely. Thus, iron regulates a wide range of epigenetic mechanisms and is expected to affect transcriptional and post-transcriptional control. Whereas iron is essential for biological processes, it can also be harmful because ferrous iron can induce the production of reacti v e o xygen species, causing o xidati v e damage. Ther efor e, intracellular iron homeostasis needs to be tightly controlled via its sequestration by the iron storage protein ferritin, and its transport in a nontoxic state by iron-binding chaperone proteins, namel y, pol y(rC)-binding proteins (PCBPs) ( 6 , 9 ). In response to increased iron demand, ferritin is degraded by a selecti v e type of autophagy called ferritinophagy ( 10 ). During the process of ferritinophagy, nuclear receptor coactivator 4 (NCOA4) acts as a selective cargo adap-tor to target ferritin to phagophor es, which ar e pr ecursors of autophagosomes that e v entually fuse with lysosomes ( 11 , 12 ).
Adipocyte dif ferentia tion is media ted by a highly organized gene expression program involving a series of transcription factors and epigenetic modifications (13)(14)(15). Particularly, per oxisome pr olifera tor-activa ted receptor ␥ (PPAR ␥ ) and C / CAAT enhancer-binding proteins (C / EBPs) are crucial transcription factors for adipocyte differentiation. For example, during the differentiation process of the 3T3-L1 adipocyte cell line, C / EBP ␤ is expressed within 4 h after the induction of adipocyte dif ferentia tion, and induces Pparg and Cebpa expression in concert with the epigenetic coordination of DNA methylation and histone modifications, such as the methylation of histones H3K4, H3K9 and H3K27 after a delay of about 12 h ( 14 , 16 , 17 ). The expression of a series of genes that define terminal differ entiation ar e then induced.
It has been reported that C / EBP ␤ activates the expression of the autophagy-related gene Atg4b ( 18 ), suggesting the possible role of autophagy in adipocyte differentiation. Indeed, multiple reports have shown that autophagy is not only acti v e but also essential in the early stage of adipocyte dif ferentia tion (18)(19)(20)(21). The suppression of autophagy was closely associated with the downregulation of PPAR ␥ and C / EBP ␣, and r epr essed adipocyte differentiation ( 18 , 19 , 22 ). Thus, autophagy is involved in the transcriptional regulation of adipocyte differentiation. In this stud y, we show tha t iron demand is increased in the early stage of adipocyte dif ferentia tion, and thus ferritinophagy is induced at this stage. We also show that the subcellular iron depletion through decreased iron supply or the insufficient subcellular iron transport reduces iron-dependent histone demethylation and DNA demethylation, and suppresses the terminal dif ferentia tion of adipocytes.

Real-time quantitative PCR (qPCR)
qPCR was performed as described previously ( 24 ) using primers in Supplementary Table S4. Data was analyzed by the standard curve method and normalized to the level of Cy c lophilin B ( Ppib ) as an internal control.

Immunoblotting
For preparation of w hole-cell l ysate, cells were washed with PBS and lysed in 1.5 × Laemmli SDS sample buffer containing 25 mM DTT and protease inhibitor cocktail (Roche, 05056489001). After boiling at 95 • C for 5 min, the sample was sonicated for 5 s × 20 times with 5 s intervals with Bioruptor II, Type 24 (Sonic Bio Co.). The supernatant was collected after centrifugation at 15 000 × g for 15 min at 4 • C and was used as the whole-cell lysate. The nuclear fractions were prepared as reported elsewhere with slight modifications ( 44 ). Cells on a 60 mm dish were allow ed to sw ell on ice in 0.4 ml of buffer A (10 mM KCl, 1.5 mM MgCl 2 , 1 mM EDTA, 1 mM EGTA, 1 mM DTT, 10 mM HEPES-KOH, pH 7.6, and protease inhibitor cocktail [Roche, 05056489001]), scraped in a 1.5 ml microcentrifuge tube, and centrifuged at 1000 × g at 4 • C for 7 min. The precipitate was collected as a nuclear pellet. The nuclear pellet was resuspended in 0.2 ml of buffer B (0.42 M NaCl, 1.5 mM MgCl 2 , 2.5% glycerol, 1 mM EDTA, 1 mM EGTA, 1 mM DTT, 20 mM HEPES-KOH, pH 7.6, and protease inhibitor cocktail [Roche, 05056489001]). After rotation at 4 • C for 1 h, the suspension was centrifuged at 10 5 × g for 30 min at 4 • C to collect the supernatant as the nuclear extract. Laemmli SDS sample buffer was added to these fractions at the final 1x concentration, and the samples were boiled at 95 • C for 5 min. The protein samples were used for immunoblotting and protein assay using Pierce 660nm Protein Assay Reagent (Thermo Fisher, 22660) with Ionic Detergent Compatibility Reagent (Thermo Fisher, 22663). The same amounts of proteins were separated on a SDS polyacrylamide gel or a 4-20% gradient SDS-polyacrylamide gel (Bio-Rad, 4561096) and transferred onto a polyvinylidene difluoride membrane (Millipore, ISEQ85R) or a nitrocellulose membrane (Bio-Rad, 1620097). Membranes were blocked with 5% skimmed milk or 5% bovine serum albumin in TBS-Tween 20 for 1 h at room temperature (RT) and incubated with primary antibodies at the indicated dilution (Supplementary Table  S5) overnight at 4 • C or for 1-2 h at RT, followed by an anti-rabbit or anti-mouse horseradish peroxidase (HRP)conjugated secondary antibody (anti-rabbit: Cell Signaling Technology, #7074, anti-mouse: Sigma-Aldrich, A4416) treatment for 1 h at RT. Alternati v ely, an anti-H3K9me2 antibody (from Dr. Kimura, Supplementary Table S5) was used after its direct conjugation to HRP using Peroxidase Labeling Kit (Dojindo, LK11). Band signals were detected with Immobilon Crescendo Western HRP substrate (Millipore, WBLUR0500) using LAS-4000 image analyzer (FUJIFILM). Band intensities were quantified with Multi Gauge software (FUJIFILM) or ImageJ / Fiji software. The band intensities on different membranes or gels were adjusted with the average intensity of the common group among them. The intensities of proteins of inter est wer e normalized to those of the indicated internal controls or the whole protein le v els. The whole protein le v els were measured using 4-20% Mini-PROTEAN TGX Stain-Free gels (Bio-Rad, 4561036), the GelDoc XR+ system (Bio-Rad), and the ImageLab software (Bio-Rad) in the setting of 5 min UV irradiation, and then the intensity values were used as an internal control as described in the guideline for autophagy evaluation ( 45 ).

Measurement of lysosome flux and ferritin degradation rate
Lysosome flux and ferritin degradation rate were measured by the immunoblotting-based method. For lysosome flux analysis, cells were incubated with 100 nM of bafilomycin A1 for 24 h before sample collection on Days 0, 1 and 2. For measurement of ferritin degradation rate, cells on Days 0, 1 and 2 wer e tr eated with 200 M of 2,2'-bipyridyl for 0, 3, 6 and 9 h. The involvement of lysosomal degradation was examined by treating cells sim ultaneousl y with 200 M of 2,2'-bipyridyl and 100 nM of bafilomycin A1 for 3 h on Days 0, 1 and 2. At the end of each drug treatment, the whole cell lysates wer e pr epar ed and subjected to immunoblotting.
To evaluate H3K9me2 levels in 3T3-L1 preadipocytes electroporated with pCMV-HA-Jmjd1a by GenePulser Xcell (Bio-Rad), cells wer e fix ed, permeabilized, and then stained with anti-H3K9me2 and anti-HA antibodies (Supplementary Table S5) in 1% normal goat serum in PBS for 2 h at RT, followed by double-staining with Alexa Fluor 488-labeled anti-mouse IgG (1:1000) and Alexa Fluor 647la beled anti-ra bbit IgG (1:1000; Thermo Fisher, A31574) in 1% normal goat serum in PBS. To evaluate 5-methyl cytosine (5mC) le v els in 3T3-L1 preadipocytes electroporated with pCMV-HA-Tet2, after fixation and permeabilization, cells wer e tr eated with 4 N HCl for 10 min at RT, and washed using 0.1 M borate buffer (pH 8.4) for 5 min twice. Cells were then stained with an anti-5mC antibody (Supplementary Table S5) overnight at 4 ºC in 1% normal goat serum and 0.3% Triton-X100 in PBS, followed by staining with Alexa Fluor 488-labeled anti-mouse IgG (1:1000) for 1 h at RT in 1% normal goat serum in PBS, and then by staining with an anti-HA antibody for 2 h at RT and Alexa Fluor 647-labeled anti-rabbit IgG (1:1000) for 1 h at RT, both in 1% normal goat serum in PBS. Staining with 4 ,6-diamidino-2-phenylindole (DAPI) (1 g / ml) was performed during the final antibod y incuba tion steps. Hoechst staining (10 g / ml) was performed after the final antibody incubation step. Fluorescence images were acquired using a FV1000 laser confocal microscope (Olympus) or fluorescence microscopeBZ-X810 (KEYENCE) and quantification analysis was performed using images acquired by FV1000.
Fluorescent intensities of indicated target proteins in nuclear regions were normalized to those of DAPI signal using FV10-ASW software (Olympus). For colocalization analysis between LC3 and NCOA4, and LC3 and Ferritin, fluorescence images were acquired with a 60 × objecti v e lens at a size of 1600 × 1600 pixels. Pearson's correlation coefficient (PCC) between pixel intensities in the cytoplasmic region of an individual cell from two fluorescence images was calculated using the Coloc 2 plugin of the Fiji (ImageJ) software.

Whole-genome bisulfite sequencing (WGBS)
Genomic DNA was purified from cells using DNeasy Blood & Tissue Kit (QIAGEN, 69504). PCR-free WGBS libraries wer e pr epar ed with the TdT-assisted adenylate connectormedia ted ssDNA liga tion-media ted Post-Bisulfite Adaptor-Tagging (tPBAT) protocol as described previously ( 51 ). One hundred nanograms of genomic DNA spiked with 1 ng of unmethylated lambda DNA was served for the library prepar ation. Each libr ary was index ed with differ ent sequences, and sequencing was performed on an Illumina HiSeq X Ten system (Macrogen Japan Corp.) with 2 × 150 paired-end chemistry. The same amounts of libraries prepared from three independent biological replicates were mixed in a tube, and one lane of sequencing was assigned per mix. Note that data of three replicates were merged to incr ease r ead depth for the downstr eam analysis. Sequenced r eads wer e mapped on the r efer ence comprised of mouse mm9 and lambda phage with Bmap as described previously ( 51 ). The alignments were summarized and exported to bed-GraphFiles with in-house de v eloped programs described previously ( 51 ). The summary of basic metrics of the DNA methylome data produced in the current study is provided in Supplementary Tables S6 and S7.

Genome-wide analysis of differentially methylated regions (DMRs) using metilene
The entire WGBS data were analyzed using the metilene software tool (version 0. [2][3][4][5][6][7][8] to detect DMRs ( qvalue < 0.05) in the whole genome by comparing the 2 conditions ( 52 ). Motif analysis for DMRs was carried out using the findMotifsGenome.pl of the HOMER software (version 4.11) ( 53 ). Aggregation plot for DNA methylation changes around DMRs was generated as described for Chroma tin immunoprecipita tion followed by high-throughput sequencing (ChIP-seq) but using the big-Wig files converted from bedGraph files by the bedGraph-ToBigWig program (downloaded from the UCSC Genome Browser w e bsite, version 2.8) as w ell as the bigwigCompare function of deepTools (version 3.5.0). 3D plots for methylation le v els of all CpG across thr ee conditions wer e generated using the Perl and R package 'rgl' (function plot3d).

WGBS data analysis within 1 kb upstream of the transcription start site (TSS)
To evaluate CpG methylation le v els within 1 kb upstream from TSSs, the following analysis was conducted using custom Perl scripts (available with test data at https://github.com/j-kohmaru-gunma/NGS Program/tree/ main/Suzuki et al WGBS ). First, CpGs located within 1 kb upstream from all TSSs wer e scr eened from the original data containing methylation le v els and cov erage / read depths for all CpGs (in a bedGraph format). After omitting CpGs having < 5 or > 1000 coverage / read depths, changes in methylation le v els on Day 8 were calculated by subtracting from those on Day 0. Finally, genes having three or more CpGs that showed more than a 50% reduction in methylation le v els from Day 0 wer e identified. Expr ession le v els of those genes were visualized using a heatmap as described above for RNA-seq, and pathway analysis was conducted by IPA.

WGBS data analysis within enhancer regions
Published peak-calling data for H3K27ac ChIP-seq (GSE21365) ( 54 ) and DNase I hypersensiti v e sites (DHSs) (GSE27826) ( 55 ) in 3T3-L1 cells were downloaded from Gene Expression Omnibus. For individual datasets, overlapping peaks across the time points were merged using the merge function of bedtools (version 2.28.0; H3K27ac: 71228 peaks; DHS: 46747 peaks), and the regions 500 bp upstream and downstream of each peak center were considered to be putati v e enhancers. DNA methylation le v els in these enhancers were analyzed by a ppl ying custom Perl scripts used for the TSS regions. The above merged peak files in a bedGraph format were applied to the script Enhancer Peak bed.pl ( https://github.com/j-kohmarugunma/NGS Program/tree/main/Suzuki et al WGBS ) to set the enhancer endpoint 500 bp downstream of each peak center. The resulting files were used for the same analysis as for the TSS regions described above. Motif analysis was performed as described above with the HOMER software (version 4.11), and all H3K27ac and DHS peaks were used as Custom Background Regions for ChIP-seq and DHSs, respecti v ely. Aggregation plots showing the distribution pa tterns for DNA methyla tion around the puta ti v e enhancers were generated using deepTools (version 3.5.0), as described above. In addition to analyzing all enhancers throughout the genome, further analysis was performed focusing on enhancers that are activated only after the induction of dif ferentia tion. The peaks of H3K27ac and DHS that were detected before dif ferentia tion induction (Day 0 and earlier) were omitted using the intersect function of bedtools, and the remaining regions were defined as dif ferentia tion-specific enhancers (H3K27ac: 25264 peaks; DHS: 36980 peaks).

ChIP
ChIP was performed essentially as described ( 17 , 25 ). Cells wer e fix ed with 0.5% formaldehyde at RT f or 10 min, f ollowed by quenching with glycine at the final concentration of 0.2 M. Cell pellets were resuspended in Hypotonic Buffer (10 mM HEPES-KOH, pH 7.5, 1.5 mM MgCl 2 , 10 mM KCl, 1 mM EDTA and 1 mM EGTA supplemented with 1 mM PMSF and Protease Inhibitor Cocktail (NACALAI TESQUE, 04080-11)) and then passed through a 22 G needle 10 times. After centrifugation, nuclear pellets were collected and resuspended in 2 ml of 1:4 mixture of SDS Lysis Buffer (50 mM Tris-HCl, pH 8.0, 10 mM EDTA, and 1% SDS) and ChIP Dilution Buffer (16.7 mM Tris-HCl, pH 8.0, 167 mM NaCl, 1.2 mM EDTA, 1.1% Triton X-100, 0.01% SDS) supplemented with the protease inhibitors above. Sonication was carried out to obtain 200-300 bp DNA fragments using a SONIFIER 250 (Branson) with the following setting: output 4, duty cycle 60%, 20 s × 15 times. Sonica ted lysa tes wer e clear ed by centrifugation, and protein concentrations wer e measur ed as described above. A total protein of 100 g for ChIP-qPCR and 500 g for ChIP-seq of lysate was diluted with SDS Lysis Buffer and ChIP Dilution Buffer with protease inhibitors to be 1:9 ratio. Antibodies (Supplementary Table S5) wer e pr ebound to 50 l of Dynabeads Protein G (Thermo Fisher, DB10004) and added to the lysates and rotated at 4 ºC for 2 h for ChIP-qPCR or overnight for ChIP-seq. The beads were successi v ely washed twice with the following buffers: Low Salt ChIP Wash Buffer (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, and 0.1% SDS), High Salt ChIP Wash Buffer (20 mM Tris-HCl, pH 8.0, 500 mM NaCl, 2 mM EDTA, 1% Triton X-100, and 0.1% SDS), LiCl ChIP Wash Buffer (10 mM Tris-HCl, pH 8.0, 250 mM LiCl, 1 mM EDTA, 1% IGEPAL CA-630 and 1% sodium deoxycholate), and TE Buffer (10 mM Tris-HCl, pH 8.0, and 1 mM EDTA). The beads were then resuspended in 200 l of buffer containing 1% SDS and 100 mM NaHCO 3 , treated with 200 g / ml proteinase K at 42 ºC for 2 h, and incubated at 65 ºC for more than 4 h for re v ersecrosslinking. DNA fragments were purified using QIAquick PCR Purification Kit (QIAGEN, 28106), and DNA concentrations were measured using Qubit dsDNA HS Assay Kit. ChIP-qPCR was carried out using the primers in Supplementary Table S8 by the standard curve method. To assess enrichment, %input was calculated relati v e to the input control.
Nucleic Acids Research, 2023, Vol. 51, No. 12 6125 ChIP using mouse epididymal white adipose tissue (WAT) All animal experiments were approved by the Gunma University Ethics Committee for Animal Experiments (protocol approval number: 22-051). Six 8-week-old male C57BL / 6N mice (Japan SLC), housed in a humidity-and temperature-controlled specific pathogen-free facility with a 12-h light / dark cycle, were fed ad libitum on a highfat, high-cholesterol diet (Research Diet, D12079B) for 2 weeks, with or without DFO treatment during the same period (3 mice each). DFO treatment was performed by intraperitoneal injection of 100 mg / kg of body weight per day (mg / kgBW / day), and the vehicle group received the same amount of PBS (10 ml / kgBW / day) intra peritoneall y. Mice were then sacrificed by cervical dislocation, and epididymal WAT was harvested for ChIP analysis. The resulting epididymal WAT was minced > 100 times with scissors, fixed in 0.5% formaldehyde at RT for 10 min, quenched with glycine (final concentration: 0.2 M), washed with ice-cold PBS, and homogenized on ice 50 times in hypotonic buffer using a W hea ton Dounce homogenizer with a loose pestle. After centrifuga tion a t 1000 × g for 7 min a t 4 • C , the pellet was collected and resuspended in 2 ml of SDS Lysis Buffer diluted 5 × with ChIP Dilution Buffer containing protease inhibitors and used for the sonication step described in the ChIP section.

ChIP-seq
ChIP-seq libraries were prepared using ThruPLEX DNA-Seq Kit (TAKARA BIO, R400674), and 75-cycle single-end sequencing was performed as described above for RNAseq. Reads were aligned to the mm9 mouse genome using Bowtie 2 (version 2.2.9) ( 56 ). Mapped reads having < 40 MAPQ were omitted (samtools view -bhS -F 0 × 4 -q 40 | samtools sort), and duplication removal was then carried out using the markdup -r function of samtools (version 1.11). Analysis of ChIP-seq signals around TSSs was performed on DEGs upregulated from Day 0 to Day 2 (DFO( −)) in RNA-seq as follows. First, ChIP-seq signals around TSSs ( ±5 kb) of these DEGs were calculated using featureCounts (version v1.5.0-p2) ( 57 ) followed by CPM normalization with R (version 4.0.3). Note that some genes have multiple TSSs, either overlapping or discrete. In genele v el anal ysis, m ultiple 'TSSs ±5 kb' r egions, if any, wer e merged for each gene, and read counting was conducted, followed by normalization with length. For analyzing individual TSSs, r eads wer e counted for each TSSs ±5 kb, e v en w hen overla pping with neighboring regions (using -O option of featureCounts). Then, using mean CPM scores from two replicates as an input, fuzzy c-means clustering was performed with the Mfuzz package (version 2.50.0) ( 58 ) on R (version 3.6.2), in which the number of clusters was determined as 7 based on the minimum distance Dmin between cluster centroids. For hea tmap genera tion, Z-scor es wer e calculated based on mean CPM values and visualized as described for RN A-seq. GO enrichment anal ysis on the clusters obtained was performed using clusterProfiler (version 3.18.1) ( 59 ) with org.Mm.eg.db 3.12.0 database on R (version 4.0.3) for selected GO terms. For principal component analysis (PCA), bigWig files were generated with the bam-Coverage function and subjected to the computeMatrix and plotPCA functions in deepTools (version 3.5.0). To visualize ChIP-seq data on IGV, bigWig files were generated as described for RNA-seq but with the following option in the bamCoverage function: -e 500. To visualize changes in ChIP-seq signals around DMRs, merged BAM files from two replicates were normalized with those of Day 0 samples and converted into bigWig files using the bamCompare function of deepTools (version 3.5.0) with the following options: -e 500 -operation subtract -normalizeUsing CPM -scaleFactorsMethod None. Heatmap and aggregation plot around DMRs ( ±3 kb from center of DMR) were generated using the computeMatrix (-skipZeros -binSize 10) and plotProfile functions in deepTools. Peak calling for H3K4me3 was carried out using the software MACS2 (version 2.7.1) ( 60 ) with the following options: -min-length 400 -q 0.000001 -g mm. Detection of overlapped peaks was done using the findOverlapsOfPeaks of chipPeakAnno (version 3.20.1) ( 61 ).

DNA methylation analysis
Genomic DNA was purified using DNeasy Blood & Tissue Kit at indicated days of differentiation. Bisulfite conversion of genomic DNA was performed with 500 ng of genomic DNA using EpiTect Plus DNA Bisulfite Kit (QI-AGEN, 59124) as the manufacture's manual. Target sequences were PCR-amplified using KOD -Multi & Epi-(TOYOBO, KME-101) with specific primer sets targeting the coding strands after the bisulfite conversion, which were designed with MethPrimer ( https://www.urogene.org/ methprimer/ ) ( 62 ). PCR products were subcloned into pCR-Blunt II TOPO vector (Thermo Fisher, 451245) or pCR4-Blunt TOPO vector (Thermo Fisher, 450031). Plasmids were isolated by boiling transformed E. coli for 1 min in buffer containing 0.7 mg / ml lysozyme, 10 mM Tris-HCl, pH 7.5, 63 mM EDTA, 2.5 M LiCl, and 4% Triton X-100 with immediate cooling on ice and purified by isopropanol precipitation. The inserts were PCR-amplified by M13 forward and reverse primers using KOD-Plus Neo (TOYOBO, KOD-401) and subjected to Sanger sequencing. All primers for DNA methylation analysis are listed up in Supplementary Table S9. Sequence data was analyzed using QUantification tool for Methylation Analysis (QUMA) ( http://quma.cdb.riken.jp/ ) ( 63 ).
The following equation was used to calculate Delta F % (DF%): DF% = ([665 nm / 620 nm of Enzyme (+) condition] / [665 nm / 620 nm of Enzyme ( −) condition] −1) × 100 In vitro H3K9me2 demethylation assay using purified cellderived JMJD1A To generate 3T3-L1 cells stably expr essing Flag-Twin-Str eptagged H3K9me2 demethylase, an artificially synthesized Flag-Twin-Strep sequence (IDT, gBlock) was incorporated into the shRNA-resistant retrovirus plasmid by Gibson assembly (Supplementary Table S2), and the plasmids were transfected into Platinum-E packaging cells to produce retroviruses as described above. Cells were then infected with the produced retrovirus, and infected cells were selected by adding 10 g / ml blasticidin S. Cells expressing Fla g-Twin-Strep-ta gged demethylase were collected in a precipita tion buf fer containing 50 mM HEPES-KOH, pH 7.9, 150 mM NaCl, 1.5 mM MgCl 2 and 1% Nonidet-P40 supplemented with EDTA-free protease inhibitor cocktail (Roche, 05056489001), and then sonicated using a Branson Sonifier SFX150 (Emerson) with the following setting: 4 cycles of 55% continuous amplitude for 10 s with a 50 sec interval. Fla g-Twin-Strep-ta gged demethylases were precipitated with prewashed StrepTactin beads (IBA Lifesciences, 2-5010-002) for 2 h at 4 ºC, and then washed with the precipita tion buf fer and PBS. A total of 1 g synthesized H3K9me2 peptide (Epigentek, R-1026) was added to the tube containing the Fla g-Twin-Strep-ta gged demethylases binding to StrepTactin beads in a reaction buffer (50 mM HEPES-KOH, pH 7.5, 1 mM ␣-KG and 2 mM ascorbic acid) in the presence or absence of 70 M ferrous ammonium sulfate, and incubated at 37 ºC for 1 h (preadipocytes) or 10 min (Day 2 samples). Samples were boiled at 95 ºC for 5 min in 1 × Laemmli SDS sample buffer containing 25 mM DTT and protease inhibitor cocktail (Roche, 05056489001), and were then applied for immunoblotting as above.

Measurement of ␣-KG
Nuclear ␣-KG concentration during differentiation was measured as described previously ( 24 ). For liquid chromato gra phy-tandem mass spectrometry (LC-MS / MS) analysis, metabolites were extracted from cells with the solvent of water: methanol: chloroform (1:2.5:1, v / v / v) containing 10 M 2-( N -morpholino) ethanesulfonic acid (2-MES) as an internal standard. The extracts were purified by using a CaptivaND Lipid filter plate (Agilent) according to the manufacturer's instruction. The filtrates were dried up by a vacuum evaporator, re-suspended with distilled water, and used for the detection of metabolites by LC-MS / MS. LC-MS / MS analysis of ␣-KG was performed by using a triple quadrupole mass spectrometer coupled with a liquid chromato gra ph (LCMS-8050 system, Shimadzu). The cell extract was separated on a MastroSP column (2.1 mm × 100 mm, 5 m, Shimadzu) by using a gradient of solvent A (10 mM ammonium aceta te in wa ter / acetonitrile 90 / 10) and solvent B (50 mM ammonium acetate in water / acetonitrile 80 / 20) with a flow rate of 0.5 ml / min. The initial solvent composition was 0% solvent B, and the following solvent gradient was applied: 0% solvent B for 4 min, increased linearly to 60% solvent B from 4 to 9 min, to 100% solvent B from 9 to 11 min, held at 100% solvent B for 2 min, then returned to 0% solvent B and held for 5 min. The column was maintained a t 40 • C . The separated analytes were ionized by electrospray ionization, and then measured by the mass spectrometer with the selected reaction monitoring (SRM) mode. The SRM transitions were The peak height of ␣-KG was divided by that of the internal standard and is presented as the relati v e ion intensity.

Statistical analysis
Statistical analysis was performed using R software (version 4.1.1). Sta tistical dif ference betw een two groups w ere examined by the two-tailed Student's t -test. For multiple comparison, groups were compared by one-way analysis of variance (ANOVA), followed by the post-hoc Tukey-Kramer test for comparison among groups or the post-hoc Dunnett test for comparison of groups with a specific control. Alternati v ely, the Kruskal-Wallis test was applied, followed by the posthoc Steel-Dwass test. Data ar e expr essed as mean ± standard error of the mean (s.e.m.) or as median with individual plots. P < 0.05 was considered statistically significant.

Requirement of iron in the early stage of adipocyte differentiation
Se v eral lines of evidence support the idea that both autophagy and iron are important for adipocyte differentiation ( 1 , 21 ). Howe v er, the role of autophagic degradation of the ir on-storage pr otein ferritin (ferritinophagy) in adipocyte dif ferentia tion r emained unclear. Ther efor e, we investigated the significance of ferritinophagy in adipocyte dif ferentia tion. First, we blocked the lysosomal degradation of ferritin by adding the lysosomal inhibitor bafilomycin A1 ( 45 , 64 ). Interestingly, the terminal dif ferentia tion of adipocytes was significantly inhibited by bafilomycin A1 in a dose-dependent manner, as determined by ORO staining on day 8 of dif ferentia tion (Day 8) (Figure 1 A). Of note, this inhibition occurred when bafilomycin A1 was added for the first 2 days in the entire process of differentiation for 8 days. Similar inhibition of adipogenesis was observed by the treatment of PIK-III, a specific inhibitor of vacuolar protein sorting 34, which plays a central role in the initiation of autophagy ( 11 ) (Supplementary Figure S1A). These data are consistent with previous reports that autophagy is crucial during the early stage of 3T3-L1 preadipocyte dif ferentia tion ( 18 , 19 ) and suggested the importance of lysosomal ferritin degradation in  the early stage of adipocyte dif ferentia tion. Therefore, we next analyzed the lysosomal flux of ferritin in the early stage of adipocyte dif ferentia tion, and found tha t the lysosomal flux of ferritin is gradually increased after the induction of dif ferentia tion (Figure 1 B , right; Supplementary Figure S1B). Accelerated ferritin degradation was also clearly observed when the translation of ferritin was inhibited by adding 2,2'-bipyridyl ( 65 ) (Figure 1 C, top; Supplementary Figure S1C). In addition, this degradation was inhibited by bafilomycin A1, indicating that ferritin degradation is mediated by lysosomal degradation (Figure 1 C, bottom; Supplementary Figure S1D). To verify that ferritin degradation during the early stage of adipocyte differentia tion is media ted by ferritinophagy, we analyzed the lysosomal flux of NCOA4, a specific cargo receptor of ferritinophagy, together with the flux of LC3, a marker of isolation membranes and autophagosomes ( 6 , 9 , 11 ). Importantl y, the l ysosomal flux of both NCOA4 and LC3 was increased during the first 2 days of dif ferentia tion ( We postulated that this is presumably due to a compensatory increase in iron uptake in cells with stable knockdown of NCOA4. This was supported by the finding that basal protein le v els of the transferrin receptor were higher in NCOA4-KD cells than in control cells (Figure 1 J). Taken together, these results indicate that iron is highly required during the early stage of adipocyte dif ferentia tion, which is accompanied by a corresponding increase in fer-ritinophagy and a concomitant increase in extracellular iron uptake. Ther efor e, we next tr eated cells with the iron chelator DFO to analyze the combined effects of both the depletion of stored iron and reduced iron uptake. When 3T3-L1 cells were treated with DFO during the entire course of dif ferentia tion, their terminal dif ferentia tion was markedly suppressed (Supplementary Figure S2A), as reported previously (3)(4)(5). Howe v er, no appar ent suppr ession was observed when DFO treatment was started on Day 2 (Supplementary Figure S2A). Further analysis demonstrated that DFO treatment during the first 2 days of dif ferentia tion was necessary and sufficient to suppress adipocyte differentiation ( Figure 2 A; Supplementary Figure S2B), and its effect was dose-dependent (Figure 2 B). This is in important agreement with the finding that ferritinophagy is crucial for the early stage of adipocyte dif ferentia tion. Indeed, ferritin levels were gradually decreased over the first 2 days of differentiation in response to DFO treatment (Figure 2 C; Supplementary Figure S2C). Thus, ferritin acts as an intracellular iron sink under the conditions of increased iron demand during the early stage of adipocyte dif ferentia tion. Therefore, in our subsequent experiments, we analyzed the effects of DFO treatment during the first 2 days of adipocyte differentiation.

DFO tr eatment r egulates the mRN A e xpr ession of r egulatory genes of adipocyte differentiation
We next determined the effects of iron depletion on gene expression profiles. The adipocyte dif ferentia tion of 3T3-L1 cells was induced either with or without the addition of DFO for the first 2 days, and mRNA was extracted from the cells on Days 0, 1, 2, 4 and 8. RNA-seq analysis detected 4907 variably expressed genes classified into four groups (Figure 2 D). Interestingly, the number of genes upregulated by DFO (cluster 1) was comparable to the number of genes downregulated by DFO (cluster 3). The top three pa thways activa ted by DFO on Day 1 included the pre-  viously reported iron-induced pathways ( 6 , 66 , 67 ), such as the hypoxia-inducible factor 1 ␣ (HIF-1 ␣) signaling, the glycolytic pathway, and the p53 signaling (Figure 2 E, left). In contrast, pathways suppressed by DFO included the PPAR signaling pathway, kinetochore metaphase signaling, and protein kinase A signaling (Figure 2 E, right). Importantly, the PPAR signaling pathway is crucial for adipocyte differentiation, as PPAR ␥ is the master regulator of adipocyte dif ferentia tion ( 13 ). Consistently, protein le v els of PPAR ␥ and mRNA expression of its encoding gene Pparg were se v er ely r epr essed by DFO (Figur e 2 F, G). This was associated with reduced mRNA expression of Cebpa and protein expression of C / EBP ␣, as PPAR ␥ and C / EBP ␣ regulate each other's transcription in a positi v e-feedback manner during adipocyte dif ferentia tion (Figure 2 F, G). In contrast, expression of the Cebpb transcript, which is induced at the very early phase of differentiation ( 13 ), did not show any significant reduction (Figure 2 F).

Iron-dependent demethylation of r epr essive histone marks during adipocyte differentiation
Sequential gene expression during adipocyte differentiation is regulated by both epigenetic regulation as well as transcriptional factors, including PPAR ␥ and C / EBP ␣. As iron is an essential cofactor for demethylases of histones and DNA, we next analyzed the epigenetic modifications altered by DFO in the early stage of adipocyte dif ferentia tion. First, whole-cell le v els of histone modifications were determined by immunocytochemistry. Repressi v e histone mar ks, such as histone H3 lysine 9 di-and trimethylation (H3K9me2 / me3), and H3K27me3 showed a decr easing tr end during differ entiation (from Day 0 to Day 2), which was inhibited by DFO treatment (Supplementary Figure S3A). In contrast, the le v els of H3K27 acetylation (H3K27ac) showed a reciprocal pattern (Supplementary Figure S3A). Next, changes in histone modifications were further analyzed in a genomic region-specific manner. ChIP-seq was performed using either anti-H3K4me3, anti-H3K9me2, anti-H3K9me3, or anti-H3K27me3 antibodies in duplicate samples. PCA of each ChIP-seq data showed a clear separation between the Day 0, Day 2 DFO( −) and Day 2 DFO(+) groups (Supplementary Figure S3B). The obtained ChIP-seq signals were calculated as normalized counts per million in the TSS ±5 kb region of each upwar dly e xpressed gene during the first 2 days of differentiation ( Figure 3 A; Supplementary Figure S3B). The patterns of changes in histone modifications were classified into 7 clusters (Figure 3 A) except for H3K4me3, which showed only limited changes, with its distribution overlapping well among the groups, namely Day 0, Day 2 DFO( −), and Day 2 DFO(+) (Supplementary Figure S3C). Gene ontolo gy (GO) enrichment anal ysis of terms associated with adipocyte biology across clusters showed that GO terms including 'fat cell differ entiation' wer e highly enriched in cluster 6 of H3K9me2 and H3K9me3, and some were also enriched in cluster 6 of H3K27me3, whereas 'ribosome biogenesis' was ubiquitously distributed (Figure 3 B). Of note, cluster 6 of each histone mark r epr esents a r egion that becomes demethylated during the first 2 days of differentiation, but not when treated with DFO (Figure 3 A). This in-dica tes tha t the demethyla tion of r epr essi v e histone mar ks in these regions is iron-dependent, and ther efor e, we postula ted tha t this is due to r equir ement of the iron-dependent activation of demethylases. The comparison of genes annotated in each cluster 6 of the r epr essi v e histone marks showed that only approximately 15% of the regions overlapped between any pair of histone marks (Figure 3 C), indica ting tha t most of the target genes of each histone modification are specific. The above findings were corroborated by indi vidual e xamples of adipocyte dif ferentia tion-associa ted genes, such as Pparg and Retn ( 68 ) in the clusters regulated by H3K9me2, Rarres2 ( 69 ) and Mrap ( 70 ) in the clusters regulated by H3K9me3, and Cebpa and G0s2 ( 71 ) in the clusters regulated by H3K27me3 (Figure 3 D; Supplementary Figure S3D, E).

Iron-dependent DNA demethylation during adipocyte differentiation
DNA methylation is another epigenetic mark implicated in the regulation of adipocyte dif ferentia tion ( 15 ). To determine whether the genomic distribution of DNA methylation changes in an iron-dependent manner, we performed WGBS (Supplementary Figure S3F, Supplementary Tables S6 and S7). Overall trends in the genome-wide distribution of DNA methylation were similar between before and after adipocyte dif ferentia tion, and between with and without DFO treatment (Supplementary Figure S3F). Howe v er, focusing on methylated CpGs, a decr easing tr end was observed with adipocyte dif ferentia tion, whereas no such decrease was observed upon DFO treatment (Supplementary Figure S3G). This suggested that DNA methylation is altered in a limited number of specific genomic regions. To identify these DMRs within whole-genome sequences during adipocyte dif ferentia tion, the metilene software tool ( 52 ) was used. When adipocytes were differentiated under normal conditions, i.e. DFO( −), 2521 downregulated DMRs were detected, whereas only 9 downregulated DMRs were detected when adipocytes were differentiated with DFO for the first 2 days, i.e. DFO(+) (Supplementary Figure S3H). The upregulated DMRs were less than 26 in conditions of both with and without DFO treatment (Supplementary Figure S3H). Thus, DNA was irondependently demethylated during adipocyte differentiation. Motif analysis of the downregulated DMRs demonstrated highly enriched binding motifs of transcriptional regulators of adipocyte dif ferentia tion, including C / EBPs, PPAR ␥ , nuclear factor I (NF1), and early B cell factor 2 (EBF2) ( 13 , 16 , 72 ) (Figure 3 E). To analyze the association between DNA methylation and neighboring gene expression, DNA methylation le v els wer e analyzed in a r egion 1 kb upstr eam from the TSS of each gene. A total of 330 demethylated regions were detected after dif ferentia ting adipocytes under the normal condition, i.e. DFO( −) (Figur e 3 F). P athway analysis of upstr eam r egulators showed high enrichment of PPAR ␥ and its heterodimeric partner, r etinoid X r eceptor, and its ligand rosiglitazone (Figure 3 G). When each region was annotated to flanking genes, and their expression profiles were classified into four groups (Figure 3 H), more than half of the 330 genes (171 genes) were classified into cluster 1, a group of genes upregulated during differenti-  Collecti v ely, our results demonstrated that the expression of se v eral genes during adipocyte dif ferentia tion is regulated by iron-dependent DNA demethylation.

JMJD1A mediates iron-dependent demethylation of H3K9me2 in the Pparg region
As inhibition of adipocyte dif ferentia tion by DFO trea tment was mediated by r epr essi v e histone marks, we next sought to identify the histone demethylases responsible for iron-dependent adipocyte dif ferentia tion. As the first screening, we knocked down a series of previously reported demethylases of H3K9 or H3K27, and screened them using Pparg expression as an indicator. Knockdown of each enzyme except for JMJD2A and JMJD3 resulted in a 15-65% reduction in Pparg expression on Day 2 ( Figure 4 A; Supplementary Figure S4A), which was partially consistent with previous findings ( 27 , 35 , 73-75 ). Rescue experiments were then performed by ov ere xpressing each enzyme with or without mutations in the iron-binding site (IBD). Ov ere xpression of JMJD1A, JMJD2B, JHDM1D, PHF2, and PHF8 increased Pparg expression in their corr esponding knockdown cells, wher eas their IBD mutants showed reduced effects (Figure 4 B; Supplementary Figure S4B). Thus, these enzymes regulate adipocyte differentiation in an iron-dependent manner. JMJD1A showed the most substantial effect among these enzymes and was further investigated. As JMJD1A is a demethylase targeting mono-or di-methylated H3K9, H3K9me2 le v els were determined by performing ChIP-qPCR analysis using a JMJD1A-KD cell line (  (Figure 4 F, bottom). This re v ersal in H3K9me2 le v els is thought to be due to the modulation of JMJD1A enzyme activity, considering that DFO treatment did not suppress Jmjd1a expression in the early stage of adipocyte differentiation (Figure 4 H).

Iron regulates the activity of histone demethylase JMJD1A
Next, the iron-dependent regulation of JMJD1A enzyme activity was analyzed. First, we performed in vitro H3K9 demethylation activity analysis using the HTRF system, and found that the activity of recombinant JMJD1A de-creases as the concentration of iron added to the reaction is reduced, and furthermore, H3K9 demethylation activity is not observed in the absence of iron (Figure 5 A). Thus, JMJD1A is dysfunctional under iron deficient conditions. In addition, the amount of unmethylated H3K9 (H3K9me0) produced by demethylation of H3K9me2 by JMJD1A is less than that produced from H3K9me1 in the low iron concentration range studied, when compared over the same reaction time (Figure 5 A). These data indica te tha t the two steps of demethylation of H3K9me2 (i.e. H3K9me2 to H3K9me1, and H3K9me1 to H3K9me0) by JMJD1A ( 76 ) are both iron-dependent. Second, to confirm iron-dependent JMJD1A demethylation activity in 3T3-L1 cells, imm unocytochemical anal ysis was performed under DFO( −) and DFO(+) conditions on Day 2. Immunocytochemistry of H3K9me2 showed that 3T3-L1 cells tr ansiently tr ansfected with HA-tagged JMJD1A showed demethylation of H3K9me2 under the DFO( −) condition, whereas such demethylation was not observed when treated with DFO [DFO(+) condition] (Figure 5 B). These results suggest that JMJD1A demethylation of H3K9me2 in 3T3-L1 cells is iron-dependent. Finally, to more directly demonstra te tha t intracellular JMJD1A demethylates H3K9me2 in an iron-dependent manner, we investigated iron-dependent JMJD1A activity using an in vitro assay system in which purified JMJD1A from 3T3-L1 cells reacts with an artificially synthesized H3K9me2 peptide. Fla g-Twin-Strep-ta gged JMJD1A was affinity purified from 3T3-L1 preadipocytes using StrepTactin beads, and its demethylation activity against the H3K9me2 pep-    showed higher H3K9me2 le v els in the Pparg region upon DFO treatment, suggesting the presence of a similar irondependent epigenomic regulation mechanism in the WAT of living mice.

TET2 regulates adipocyte differentiation in an irondependent manner
Ne xt, we inv estigated the iron-dependent effects of the Tet family of DNA demethylases. We found that knockdown of either TET2 or TET3 reduced Pparg expression ( Figure  6 A; Supplementary Figure S5A), and the latter was consistent with recent findings ( 77 , 78 ). Furthermore, in TET2-KD cells, the reduced Pparg expression was restored by the for ced expr ession of TET2, which is the shRNA-resistant form of TET2 with a wild-type IBD (WT-TET2), but not by ov ere xpression of the IBD mutant (Mut-TET2) containing both H1295Y and D1297A m utations, w hich pre v ent iron-dependent activity (Figure 6 B; Supplementary Figure S5B). To analyze the iron-dependent DNA demethylation by TET2, DNA methylation le v els in the Pparg2 promoter r egion wer e determined (Figur e 6 C, D), as the Pparg1 region is almost 100% methylated during the early stage of dif ferentia tion and hence cannot be increased any further (Supplementary Figure S5C). DNA methylation levels in the Pparg2 region gradually decreased by half during adipocyte dif ferentia tion, from 41% to 20% ( Figure  6 C, left). Howe v er, TET2-KD cells showed 41% or higher le v els of DNA methylation throughout the dif ferentia tion process (Figure 6 C, right). The forced expression of WT-TET2 showed reduced DNA methylation le v els on Day 2 in TET2-KD cells, but Mut-TET2 did not (Figure 6 D). These changes in DNA methylation le v els were inversely correlated with the le v els of Pparg mRNA (Figure 6 B) and PPAR ␥ protein (Figure 6 E) on Day 2 and lipid accumulation on Day 8 (Figure 6 F, G). This restoration of DNA demethylation by WT-TET2 during the early stage of adipocyte dif ferentia tion may be due to modulation of its activity, as DFO treatment did not suppress Tet2 expression during that period (Supplementary Figure S5D).
Next, to confirm that iron-dependent TET2 demethylation occurs in 3T3-L1 cells, immunocytochemistry of 5mC was performed. Cells transiently transfected with HA-tagged TET2 sho wed lo wer 5mC le v els under the DFO( −) condition, whereas such a change was not observed when cells wer e tr eated with DFO [DFO(+) condition] (Figur e 6 H). This data suggests that TET2 iron-dependently demethylates 5mC in 3T3-L1 cells.

Cooper ative r egulation of adipogenesis by histone methylation and DNA methylation
In summary, both JMJD1A and TET2 were found to be involved in the regulation of Pparg expression. This suggests that these enzymes may cooperati v ely regulate adipocyte dif ferentia tion. In other words, histone methylation and DNA methylation are both altered in an irondependent manner, suggesting that they cooperati v ely regulate adipocyte differ entiation. Ther efor e, we investigated the interrelationship between H3K9me2 and DNA methylation on a genome-wide basis by comparing WGBS with ChIPseq data. Notably, we found that iron-dependent demethylation of H3K9me2 was often observed in DMRs ( The two-tailed Student's t -test was performed for the cell lines expressing either sh-Jmjd2c, sh-Jmjd2d, or sh-Phf2. One-way ANOVA followed by the Dunnett test was perf ormed f or the other cell lines. * P < 0.05. ( B ) Each knockdown line was retrovirally transduced with the corresponding enzyme with or without the indicated mutation in the iron-binding site. Pparg mRNA le v els on Day 2 were measured by qPCR ( n = 3 biological replicates). The full-length mouse sequence of the corresponding enzyme gene was used for ov ere xpression, e xcept for the partial sequence encoding amino acids (a.a.) 71 to 940 of JHDM1D and the human version for UTX. Data are shown as the mean ± s.e.m. One-way ANOVA followed by the Tukey-Kramer test was perf ormed f or statistical analysis. * P < 0.05. ( C ) ChIP-qPCR analysis of H3K9me2 on Pparg and Actb genes in JMJD1A-KD cells (sh-Jmjd1a #1) and its control cell line (sh-Ctrl) on Day 2 (mean ± s.e.m. of three biological replicates). ChIP signals were presented as a percentage of input DNA. The two-tailed Student's t -test was performed for statistical analysis. * P < 0.05, *** P < 0.001, **** P < 0.0001, N.S., not significant.     Figure S6A, B). Additionally, enrichment motif anal ysis of DN A demethylated enhancers showed high enrichment of CEBP, PPARE, RXR, and NF1 binding motifs (Supplementary Figure S6C), and this result was also similar to the results of analysis of the whole genome ( Figure  3 E). Iron-dependent DNA demethylation in enhancer regions during adipocyte dif ferentia tion was limited on Day 2 and apparent on Day 8, and was markedly suppressed by DFO treatment (Supplementary Figure S6D). This DNA demethylation change is e v en more pronounced when focusing only on enhancer regions that are activated during adipocyte dif ferentia tion (Supplementary Figure S6D). These data support the existence of a mechanistic link between histone methylation and DNA methylation during iron-dependent adipocyte dif ferentia tion. Consistently, the inhibition of iron supply by bafilomycin A1 treatment during the first two days of dif ferentia tion was found to increase both H3K9me2 and methylated DNA le v el in the Pparg genomic region on Day 2 (Figure 6 J).

Requirement of iron chaperon PCBP2 for epigenetic changes in adipocytes
As both histones and DNA are localized in the nucleus, we specula ted tha t intr acellular iron tr ansport to the nucleus is important for the demethylation of both histones and DNA during adipocyte dif ferentia tion. Therefore, we analyzed the translocation of the cytoplasmic iron chaperone , PCBPs , to the nucleus (Figure 7 A, B). We first analyzed the nuclear le v els of PCBP2, which binds a wider range of proteins than PCBP1 ( 9 ), and found that it is transiently increased on Day 2 ( Figure 7 A; Supplementary Figure S7A). This increase in nuclear PCBP2 le v el on Day 2 was suppressed by the administration of either DFO or PIK-III (Figure 7 B, right). Similarly, nuclear PCBP1 le v els also showed an increase, albeit more transient, during the first 2 days of dif ferentia tion, which was also suppressed by either DFO or PIK-III (Figure 7 B, left). These data suggest that the nuclear translocation of PCPBs occurs in response to an increase in iron supply during the early stage of adipocyte dif ferentia tion. W hen PCBP2-KD cells were generated, both H3K9me2 and methylated DNA le v els in the Pparg genomic region on Day 2 of these cells were higher than in control cells (Figure 7 C). These findings collecti v ely indica te tha t iron deli v ery to the nucleus by PCBP2 is required for the regulation of histone and DNA demethylation during the early stage of adipocyte dif ferentia tion. We further established and analyzed single and double knockdown cell lines of PCBP1 and PCBP2 ( Supplementary Figure S7B), and found that Pparg expression was cooperati v ely suppressed in these cells (Figure 7 D). In addition, double knockdown of both PCBP1 and PCPB2 markedly inhibited terminal adipocyte dif ferentia tion (Figure 7 E). In summary, these results suggest that ir on chaper one PCBPs are transported to the nucleus during the early stage of adipocyte dif ferentia tion, and regula te epigenomic changes during adipocyte dif ferentia tion.

DISCUSSION
It has been reported that autophagy is essential for adipocyte dif ferentia tion, and is most acti v e during the early stage (18)(19)(20)(21). Howe v er, the importance of this time-specific induction of autophagy has rarely been recognized from the perspecti v e of iron supply through the autophagic degradation of ferritin, namely, ferritinophagy. In the present stud y, we demonstra ted tha t iron is essential for the early stage of adipocyte dif ferentia tion, and furthermore, tha t ferritinophagy is prominently activated during this period. These results suggest that iron demand is increased during early adipocyte dif ferentia tion, and tha t ferritin degradation is correspondingly accelerated to increase iron supply. Sequential gene expression during adipocyte differentiation is regulated by both transcription factors and epigenetic mechanisms. Regarding the serial regulation of transcription factors, it is well known that C / EBP ␤ induces the expression of Cebpa and Pparg , which in turn activate each other's expression and regulate subsequent gene expression in a coordinated manner. Our results demonstra ted tha t iron chela tion by DFO suppresses the mRNA le v els of Cebpa and Pparg but not Cebpb , indicating that iron is important for the expression of Cebpa and Pparg . It has been reported that autophagy promotes the transcription of Cebpa and Pparg by degrading Kruppel-like factor 2 (KLF2) and KLF3, which are negati v e regulators of adipocyte dif ferentia tion ( 18 ). In addition to this finding, we newly found that autophagy regulates Cebpa and Pparg transcription through iron-dependent epigenetic regulation. Regarding the epigenetic mechanisms of adipocyte dif ferentia tion, we noted that iron is essential for the activity of a series of epigenetic enzymes, including JmjC domain-containing histone demethylases and TET famil y DN A demeth ylases. As both histone demeth ylases and DNA demethylases affect gene transcription, we specula ted tha t iron is involved in genome-wide transcriptional r egulation. Ther efor e, we performed a genome-wide analysis of histone methylation and DNA methylation using next-generation sequencing, which demonstra ted tha t irondependent histone demethylation and DNA demethylation occur during adipocyte dif ferentia tion. A comprehensi v e screen of epigenetic enzymes that regulate Pparg expression in an iron-dependent manner identified JMJD1A, JMJD2B, JHDM1D, PHF2 and PHF8. These are all H3K9 demethylases, and the enzyme tha t regula tes H3K27 demethylation in an iron-dependent manner in the early stage of adipocyte dif ferentia tion was not identified in this study, and is hence the subject of future studies. For several histone demethylases, Pparg expression was reduced in their KD cells, but was not r estor ed by forced expression of the corresponding gene. These results can be interpreted in light of pr evious r eports that epigenetic enzymes regulate gene expression not only in an enzyme activity-dependent manner, but also in an enzyme activity-independent manner ( 25 , 79 , 80 ). Additionally, TET2 and TET3 were identified as DNA demethylases that regulate Pparg expression in an irondependent manner, which is consistent with recent reports demonstra ting tha t the la tter is a regula tor of adipocyte dif ferentia tion ( 77 , 78 ). These findings suggested that multiple enzymes cooperati v ely regulate transcription during adipocyte dif ferentia tion. Based on the integra ted analysis of WGBS and ChIP-seq data, we found that the demethylation of r epr essi v e histone mar ks, including H3K9me2, H3K9me3, and H3K27me3, were frequently observed around the downregulated DMRs in the CpG regions, indicating a mechanistic link between histone methylation and DNA methylation. We also found that the overall trend of the demethylation of r epr essi v e histone marks was evident as early as Day 2, whereas genome-wide patterns of DN A demethylation were onl y very slightl y observed on Day 2 and became apparent by Day 8 (Figure 6 I, Supplementary Figure S5E). Because of the difference in the onset times of histone demethylation and DNA demethylation, we considered the possibility that histone demethyla tion precedes DNA demethyla tion. In the enhancer regions, a similar trend of iron-dependent DNA demethylation, which is very limited on Day 2 and becomes apparent on Day 8, was more pronounced w hen anal yzed onl y in the regions that are activated during adipocyte differentiation (Supplementary Figure S6D). Considering that acti v e enhancer marks and r epr essive histone marks are often m utuall y e xclusi v e, these r esults ar e consistent with the hypothesis that the demethylation of r epr essi v e histone methylation modifications precedes the demethylation of DNA methylation modifications. It was previously reported that DNA methylation is a pr er equisite for H3K9 methylation, when forming a bivalent chromatin domain consisting of H3K9me3 and H3K4me3 to maintain the preadipocyte state of 3T3-L1 cells ( 17 ). Together with this previous data, our present results suggest that the demethyla tion of methyla ted H3K9 precedes DNA demethyla tion during the resolution of the bivalent domain of H3K9me3 and H3K4me3 during adipocyte dif ferentia tion. Howe v er, more detailed studies are needed to reach this conclusion because of technical limitations due to the difference in sensitivity between ChIP-seq and WGBS, and because bisulfite sequencing cannot dif ferentia te between 5mC and 5h ydroxymeth ylcytosine (5hmC).
The early stage of adipocyte dif ferentia tion is the time period when mitotic clonal expansion (MCE) of 3T3-L1 cells occurs. The number of cells counted during adipocyte differentiation was comparable between conditions with and without DFO addition on Day 1, but on Days 2 and 4, cell numbers decreased upon DFO treatment (Supplementary Figure S8A). Thus, DFO negati v ely affects MCE. In contrast, cell numbers of WT-JMJD1A-expressing JMJD1A-KD cells and Mut-JMJD1A-expressing JMJD1A-KD cells were comparable from Day 0 to Day 4 (Supplementary Figure S8B), and cell numbers of WT-TET2-expressing TET2-KD cells and Mut-TET2-expressing TET2-KD cells (Supplementary Figure S8C) showed the same trend. Al-though the detailed epigenomic mechanisms that regulate MCE in an iron-dependent manner are largely unknown, members of the KDM5 H3K4 demethylases (KDM5s) ma y pla y a role, because cell cycle-associated genes, such as Cdc20 , Plk1 , and Ccna2 , that show reduced mRNA expr ession upon DFO tr eatment (Supplementary Figur e S8D), are reported as direct targets of KDM5s ( 81 ). Based on the hypothesis that DFO suppresses the activity of KDM5s, which are demethylases of H3K4me3, it is assumed that H3K4me3 le v els are maintained and target gene expression is increased; howe v er, conv ersely, H3K4me3 le v els in the genomic regions of cell cycle-associated genes and their mRNA expr ession ar e suppr essed in cells treated with DFO [DFO(+)] compared with untreated cells [DFO( −)] (Supplementary Figure S8E). Considering the pr evious r eport that the KDM5s not only promotes but also r epr esses target gene expr ession in an enzyme acti v e site-dependent manner without necessarily mediating H3K4me3 ( 81 ), it is possible that other complex mechanisms besides the dir ect r egulation of histone demethylation are involved in regulating cell cycle gene expression by DFO.
A possible mechanism for the upregulation of Jmjd1a and Tet2 expression by DFO in the early stage of adipocyte differentiation shown in Figure 4 H and Supplementary Figure  S5D is the involvement of HIF-1 ␣. This is because Jmjd1a expression has previously been reported to be induced by HIF-1 ␣ in se v eral cell lines (e.g. endothelial and cancer cell lines) (82)(83)(84), and Tet2 is also regulated by HIF-1 ␣ in HepG2 hepatocellular carcinoma cells ( 85 ). Additionally, as pr esented in Figur e 2 E, the HIF-1 ␣ signaling pathway is activ ated b y DFO on Day 1. The fact that histone demethyla tion and DNA demethyla tion by JMJD1A and TET2, respecti v ely, ar e suppr essed by DFO tr eatment, wher eas the expression of Jmjd1a and Tet2 is increased by DFO treatment suggests that the activity of these enzymes is regulated by DFO.
Although the epigenomic mechanism by which specific iron-dependent enzymes regulate specific target genes is currently unknown, iron sensitivity may vary among enzymes or there may be context-dependent binding between enzymes and iron chaperones, such as PCBPs.
In addition to epigenetic enzymes, iron also binds in catalytic centers of other enzymes. Screening all such enzymes based on their effects on adipocyte dif ferentia tion is beyond the scope of this study. Howe v er, among these enzymes , mitochondrial aconitase , which r equir es iron-sulfur clusters ( 65 ), is an important iron-dependent enzyme in terms of its regulation of epigenetic enzyme activity. It is a constituent enzyme of the tricarboxylic acid (TCA) cycle and catalyzes the reversible isomerization of citrate to isocitrate. Because isocitrate is converted to ␣-KG in the TCA cycle, and ␣-KG is another cofactor essential for the activity of histone and DNA demethylases besides iron ( 16 , 86 ), the iron-dependent activation of mitochondrial aconitase may affect epigenetic demethylation. Indeed, intracellular and nuclear ␣-KG le v els wer e incr eased during the early stage of adipocyte dif ferentia tion, but were suppr essed by DFO tr eatment ( 24 ) (Supplementary Figur e  S9). This suggests that iron rewrites the epigenetic landscape during adipocyte dif ferentia tion by directly, and per-haps e v en indir ectly, r egulating the activity of epigenetic demethylases.
Taken altogether, our present study demonstrates that iron is essential during the early stage of adipocyte differentiation, and ferritinophagy is most acti v e during this stage, to meet the increased iron demand. Iron is subcellular ly tr ansported by PCBPs to the nucleus during the early stage of adipocyte dif ferentia tion, and media tes adipocyte dif ferentia tion. An inadequa te iron supply during the early stage of adipocyte dif ferentia tion disrupts the demethylation of r epr essi v e histone methylation marks and DNA methylation in the genomic regions of adipocyte dif ferentia tion-associa ted genes, resulting in suppression of the terminal dif ferentia tion of adipocytes. In particular, JMJD1A and TET2 are the major iron-dependent demethylases tha t regula te Ppar g expression during the early stage of adipocyte dif ferentia tion. Thus, iron is indispensable for epigenetic rewriting during the adipocyte dif ferentia tion process. Whether physiological changes in iron le v els epigenetically regulate gene expression in a dose-dependent manner during adipocyte dif ferentia tion in vivo remains unclear, and is a topic for future study.

DA T A A V AILABILITY
Next generation sequencing datasets reported in this study have been deposited to Gene Expression Omnibus with the accession numbers of GSE174136.

SUPPLEMENT ARY DA T A
Supplementary Data are available at NAR Online.