Peroxisomal β-oxidation regulates histone acetylationand DNA methylation in ArabidopsisLishuan Wanga,1, Chunlei Wangb,1, Xinye Liuc, Jinkui Chenga, Shaofang Lia, Jian-Kang Zhud,e,2, and Zhizhong Gonga,2

aState Key Laboratory of Plant Physiology and Biochemistry, College of Biological Sciences, China Agricultural University, 100193 Beijing, China; bCollege ofHorticulture, Gansu Agricultural University, 730070 Lanzhou, China; cMinistry of Education Key Laboratory of Molecular and Cellular Biology, College of LifeSciences, Hebei Normal University, 050024 Shijiazhuang, China; dShanghai Center for Plant Stress Biology, National Key Laboratory of Plant MolecularGenetics, Center of Excellence in Molecular Plant Sciences, Shanghai Institute for Biological Sciences, Chinese Academy of Sciences, 200032 Shanghai, China;and eDepartment of Horticulture and Landscape Architecture, Purdue University, West Lafayette, IN 47907

Contributed by Jian-Kang Zhu, April 4, 2019 (sent for review March 12, 2019; reviewed by Bao Liu and Jim Peacock)

Epigenetic markers, such as histone acetylation and DNAmethylation,determine chromatin organization. In eukaryotic cells, metabolitesfrom organelles or the cytosol affect epigenetic modifications. How-ever, the relationships between metabolites and epigenetic modifica-tions are not well understood in plants. We found that peroxisomalacyl-CoA oxidase 4 (ACX4), an enzyme in the fatty acid β-oxidationpathway, is required for suppressing the silencing of some endoge-nous loci, as well as Pro35S:NPTII in the ProRD29A:LUC/C24 transgenicline. The acx4 mutation reduces nuclear histone acetylation and in-creases DNA methylation at the NOS terminator of Pro35S:NPTII andat some endogenous genomic loci, which are also targeted by thedemethylation enzyme REPRESSOR OF SILENCING 1 (ROS1). Further-more, mutations in multifunctional protein 2 (MFP2) and 3-ketoacyl-CoA thiolase-2 (KAT2/PED1/PKT3), two enzymes in the last two stepsof the β-oxidation pathway, lead to similar patterns of DNA hyper-methylation as in acx4. Thus, metabolites from fatty acid β-oxidationin peroxisomes are closely linked to nuclear epigenetic modifications,which may affect diverse cellular processes in plants.

β-oxidation | acetyl-CoA | histone acetylation | DNA methylation |gene silencing

Histone acetylation is important for neutralizing the positivecharges of lysine residues and promoting chromatin relaxation;

it is also required for transcription, DNA replication, histonemethylation, and other histone modifications (1–4). Histones areacetylated by acetyltransferases, which transfer acetyl groupsfrom acetyl-CoA to histone lysine residues.Acetyl-CoA is a central metabolite that can be produced via

several metabolic pathways involved in pyruvate, citrate, acetate,and fatty acid β-oxidation metabolism (5). In mammals, acetyl-CoA in mitochondria is produced from different pathways, in-cluding the fatty acid β-oxidation (6). In cytosol and nucleus,adenosine triphosphate (ATP)-citrate lyase (ACLY) cleavescitrate exported from mitochondria to regenerate acetyl-CoAthat can be used for other biosynthetic processes, such as fattyacid synthesis and histone acetylation (6). In mouse, conditionalloss of carnitine palmitoyltransferase 1A (CPT1A), which is requiredfor the transfer of fatty acid into mitochondria for β-oxidation,impairs dermal lymphatic formation via histone acetylation in anACLY-dependent manner (7). A pyruvate dehydrogenase com-plex can be translocated from mitochondria to nuclei to generateacetyl-CoA and mediate histone acetylation in mammalian cellsin certain conditions (2, 8). In Arabidopsis, the mutations in cytosolicacetyl-CoA carboxylase (ACC1), which converts cytosolic acetyl-CoA to malonyl-CoA for elongating the plastid-produced fattyacids, lead to high accumulation of cytosolic acetyl-CoA, spe-cifically resulting in increased H3K27 acetylation (H3K27ac) (9).These results underscore the importance of acetyl-CoA in

histone acetylation in nuclei in both mammals and plants. However,in plant cells, plastids, mitochondria, peroxisomes, and cytosol canproduce acetyl-CoA (10). Whether impairment of metabolism inplant organelles will affect the nuclear epigenetic modifications isstill being unraveled.

DNA methylation is a conserved epigenetic marker importantin genome organization, gene expression, genomic imprinting,paramutation, and X chromosome inactivation in organisms (3,11–13). DNA methylation patterns are coordinately determinedby methylation and demethylation reactions in plants and ani-mals (13, 14). The active removal of 5mC in Arabidopsis is car-ried out by a subfamily of bifunctional DNA glycosylases/lyasesrepresented by REPRESSOR OF SILENCING 1 (ROS1) andDEMETER (DME) (15, 16). ROS1 family proteins bind DNA non-specifically (17) and need other factors to find target genomic regions(13). Among these, ROS4/INCREASED DNA METHYLATION 1(IDM1) is a plant homeodomain finger-containing histoneacetyltransferase that catalyzes the acetylation of histone H3lysine 18 (H3K18) and lysine 23 (H3K23) to create a favorablechromatin environment for the recruitment of ROS1 at someloci (1, 18). ROS4/IDM1, together with other factors, such as ROS5/IDM2, IDM3, methyl-CPG-binding domain 7 (MBD7), Harbinger

Significance

Small-molecule metabolites from cell organelles or cytosol exerttheir influence on decision making processes in eukaryotic cells.However, the mechanisms underlying the regulation of cellularprocesses by these metabolites are not well understood. Amongthese small metabolites, acetyl-CoA, a most critical one, is pro-duced in the cytosol and different cell organelles. In plants, acetyl-CoA produced from fatty acid β-oxidation in peroxisomes playscritical roles in various developmental stages. Here we found thatdefects in β-oxidation in peroxisomes affect both histone acety-lation and DNA methylation in the nucleus. Our work providesevidence for retrograde signaling from peroxisomes to regulatenuclear epigenetic modifications in higher eukaryotes.

Author contributions: L.W., C.W., J.-K.Z., and Z.G. designed research; L.W. and C.W. per-formed research; L.W., C.W., X.L., J.C., and S.L. analyzed data; and L.W., J.-K.Z., and Z.G.wrote the paper.

Reviewers: B.L., Northeast Normal University; and J.P., Commonwealth Scientific and In-dustrial Research Organisation.

The authors declare no conflict of interest.

Published under the PNAS license.

Data deposition: Sequence data referred to in this article have been deposited in theGenBank/EMBL database (accession nos. AT2G36490 for ROS1, AT3G14980 for ROS4/IDM1, AT5G66750 for DDM1, AT4G11130 for RDR2, AT3G51840 for ACX4, AT3G06860for MFP2, AT2G33150 for KAT2, AT1G64230 for UBC28, and AT3G18780 for ACTIN2).Primary datasets for the whole-genome bisulfite sequences of Col-0, ros1-4, acx4-4,acx4-1, mfp2-2, and kat2-3 mutant plants have been deposited in the Gene ExpressionOmnibus (GEO) database (accession no. GSE98214). Histone acetylation ChIP-seq data alsohave been deposited in the GEO database (accession no. GSE98214). Whole-genome bi-sulfite sequencing data of C24 WT, ros1-1, and ros4 plants were obtained from the GEOdatabase (accession no. SRP042060).1L.W. and C.W. contributed equally to this work.2To whom correspondence may be addressed. Email: jkzhu@sibs.ac.cn or gongzz@cau.edu.cn.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1904143116/-/DCSupplemental.

Published online May 7, 2019.

10576–10585 | PNAS | May 21, 2019 | vol. 116 | no. 21 www.pnas.org/cgi/doi/10.1073/pnas.1904143116

Dow

nloa

ded

by g

uest

on

Oct

ober

27,

202

0

transposon-derived protein 1 (HDP1), and HDP2, forms a complexto regulate active DNA demethylation (19–23). MET18 is a com-ponent in the cytosolic iron-sulfur cluster assembly pathway in-volved in the transfer of the Fe-S cluster to ROS1, which isnecessary for its function (24). The expression of ROS1 is positivelyregulated by promoter DNA methylation, which requires a proteincomplex composed of Su(var)3–9 homologs (SUVHs) and SUVH-interacting DNAJ (SDJ) proteins (25–28).To identify the components required to prevent transgene si-

lencing and normal DNA methylation patterns in Arabidopsis, weperformed a forward genetic screen for kanamycin (Kan)-sensitivemutants using the ProRD29A:LUC/Pro35S:NPTII transgenic C24line (18). Several ROS1 alleles, multiple components of the RdDMpathway, ROS4/IDM1, ROS5/IDM2, andMBD7, were identified inthis screening (1, 18, 22, 23). Here we identified an antisilencingfactor, acyl-CoA oxidase 4 (ACX4), in the fatty acid β-oxidationpathway. In acx4 mutants, overall levels of H3Ac and H4Ac arereduced, and DNA methylation is increased at some genomic loci,resulting in enhanced transcriptional silencing of reporter and someendogenous genes. The mfp2 and kat2 mutants have similar DNAhypermethylation phenotypes to acx4. Our results uncover a con-nection between fatty acid β-oxidation and epigenetic regulationin plants.

ResultsACX4 Is a Suppressor of Transcriptional Silencing. To decipher themechanism that blocks transcriptional gene silencing, we per-formed a forward genetic screen using an ethyl methanesulfo-nate (EMS)-mutagenized population of a C24 transgenic linecarrying ProRD29A:LUC and Pro35S:NPTII transgenes, both ofwhich are actively expressed (used as the WT) in Arabidopsis (18,22, 23). From this population, a recessive kanamycin (Kan)-sensitive mutant, acx4 (hereinafter acx4-4), was isolated. Likeros4, ros5-1, andmbd7mutants (18, 22, 23), acx4-4 plants exhibitedsilenced NTPII expression (Fig. 1 A and C), but the expression ofLUC was unaltered (Fig. 1 B and D), while in ros1-1, bothPro35S:NPTII and ProRD29A:LUC were silenced (Fig. 1 A–D).ACX4 was identified through map-based cloning. We crossed

the acx4-4 mutant in the C24 accession with the WT Columbia-0 (Col-0) background and used the 2105 F2 plants for mapping.The acx4-4 mutation was localized on the bottom of chromo-some 3 and then narrowed down between bacterial artificialchromosome clones T18N14 and T25B25 (Fig. 1E). By sequencingcandidate genes, a G-to-A mutation was found at position 808 ofAT3G51840 (from the first putative ATG), which would changethe splicing acceptor site from GT to AT at the end of the thirdintron (Fig. 1F). By comparing the cDNAs of AT3G51840 am-plified by RT-PCR from WT and acx4-4 lines, we found that thecDNA of AT3G51840 from acx4-4 was 16 bp shorter than thatfrom WT. Furthermore, the acceptor splice site had moved tothe fourth exon, which would lead to premature termination oftranslation and a truncated protein (SI Appendix, Fig. S1 A and B).To further examine whether the mutation in AT3G51840 causesthe silencing of the Pro35S:NPTII transgene, WT AT3G51840,including the fragment 2543 bp upstream of the first putative ATGcodon and 312 bp downstream of the putative stop codon TAA,was cloned and transformed into the acx4-4 mutant. Like WT,three randomly selected independent transgenic lines exhibitedKan resistance, and accumulated the NPTII protein (Fig. 1 G andH). Two additional alleles, acx4-1 and acx4-3 (Col-0 accession)(29), were obtained, representing the T-DNA insertion linesSALK_000879 and SALK_065013, with the T-DNA inserted inthe third exon and 11th intron of ACX4, respectively (Fig. 1F). Weintroduced the acx4-1 and acx4-3 mutations into the WT trans-genic plants by crossing. As expected, both mutations resulted inhypersensitivity to Kan relative to WT with greatly reduced NPTIItranscripts, as determined by quantitative RT-PCR (qPCR) (SIAppendix, Fig. S2 A and B), and the expression of ProRD29A:LUC

was not obviously affected in the acx4-1 and acx4-3 mutants, asindicated by luciferase and LUC transcript assays (SI Appendix,Fig. S2 A and B).We examined the subcellular localization of ACX4 using

transient assays. ACX4 was fused to the C terminus of the redfluorescent protein mCherry, CD3-990 (a mitochondrial marker)was fused to the C terminus of yellow fluorescent protein (YFP)(30), CAT2 (a peroxisomal marker) was fused to the C terminusof green fluorescent protein (GFP) (31), and nuclear localizationsignal (NLS) was fused to the N terminus of GFP. Mcherry-ACX4 was transiently coexpressed with each of the fusedmarker proteins in tobacco (Nicotiana benthamiana) epidermalcells driven by the cauliflower mosaic virus 35S promoter.ACX4 did not colocalize with CD3-990 and NLS (SI Appendix,Fig. S3 A and B) but did colocalize with GFP-CAT2 (SI Appendix,Fig. S3C), consistent with immunoelectron microscopy data froma previous study (32). Beta-glucuronidase (GUS) staining oftransgenic plants expressing ProACX4:GUS revealed ACX4 ex-pression throughout the seedling (SI Appendix, Fig. S4).

The acx4Mutation Causes DNA Hypermethylation at the NOS Terminatorof Pro35S:NPTII and Some Endogenous Genomic Loci. To test whethersilencing of the Pro35S:NPTII transgene is associated with DNAmethylation, we treated the WT, acx4-4, ros1-1, and ros4 lines with

Fig. 1. Characterization of the acx4 mutant. (A) The acx4 mutation silencedPro35S:NPTII, as indicated by Kan sensitivity. The acx4-4, ros1-1, and ros4mutants were grown on MS medium or on MS medium supplementedwith 50 mg/L Kan. (B) The acx4 mutation does not affect the expression ofProRD29A:LUC. The acx4-4, ros1-1, and ros4 mutants grown on MS mediumwere treated with 30 μM abscisic acid for 3 h, after which a luciferase assaywas performed using a cold charge-coupled device camera. (C) Relative ex-pression levels of Pro35S:NPTII in the acx4 mutant and the WT(C24 accession), ros1-1, and ros4 seedlings by qPCR. UBC28 served as an in-ternal control. (D) Relative expression levels of ProRD29A:LUC in the acx4mutant and WT, ros1-1, and ros4 seedlings by qPCR. UBC28 served as an in-ternal control. (E) Diagram of the region identified as the acx4-4 mutation bymap-based cloning. (F) Model structure of the AT3G51840 (ACX4) gene. AG808-to-A808mutation occurs in the splice donor site of the third intron of theAT3G51840 gene in the acx4-4 mutant. Two T-DNA insertion mutants, acx4-1and acx4-3, are shown. The exon and intron are marked by a black box andblack line, respectively. (G) The acx4 mutant was complemented by ACX4, astested on MS medium containing 50 mg/L Kan in three complement lines(Com9, 10, 15). (H) NPTII protein levels in WT, ros1-1, acx4-4, and three com-plemented lines, as indicated by immunoblotting with NPTII antibodies. ACTINserved as a loading control.

Wang et al. PNAS | May 21, 2019 | vol. 116 | no. 21 | 10577

PLANTBIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

27,

202

0

the DNA methylation inhibitor 5-Aza-2′-deoxycytidine (5′-Aza).The DNA methylation inhibitor suppressed the Kan sensitivityof the acx4-4, ros1-1, and ros4 mutants (SI Appendix, Fig. S5A)and restored the expression of NPTII (SI Appendix, Fig. S5B).Furthermore, the acx4-4 ddm1 double mutant, but not theacx4-4 rdr2 double mutant, released the silencing of NPTII inacx4-4 (SI Appendix, Fig. S5 C and D). Both ddm1 and rdr2 havethe same genetic background as acx4-4 (18). Silencing of NPTII inacx4-4 was suppressed by the ddm1, but not the rdr2 mutation (SIAppendix, Fig. S5 C and D), suggesting that NPTII silencing isdependent on the DDM1-mediated methylation pathway, but notthe RdDM pathway. Published studies indicate that the mutationsin the components of RdDM pathway cause silencing of NPTIIdue to the repression of ROS1 that is positively regulated by DNAmethylation in its promoter region (18, 33).Previous studies have indicated that the ros1-1 mutation in-

creases DNA methylation in both the RD29A promoter and the3′-NOPALINE SYNTHASE (NOS) terminator region, but thatthe ros4-4, ros5-1, and mbd7 mutations increase DNA methyl-ation mainly in the NOS terminator (22, 23). Bisulfite sequencinganalyses indicated a slight increase in DNA methylation in the35S promoter in these mutants (Fig. 2A). Like the ros1-1 andros4 mutants, acx4-4 showed a clear increase in CG, CHG, andCHH DNA methylation (Fig. 2B) in the NOS terminator. Theseresults suggest that the acx4 mutation leads to DNA hyper-methylation in theNOS terminator and silencing of Pro35S:NPTII.To investigate the effect of the acx4-4 mutation on genomic

DNA methylation patterns, whole-genome DNA methylationprofiles of acx4-4 (two biological replicates) and the WT seed-lings (22, 23) were compared using next-generation sequencingafter bisulfite conversion. Analysis of differentially methylatedregions (DMRs) revealed 864 DMRs with increased DNAmethylation (hyper-DMRs) in acx4-4, including 405 DMRs withhyper-CG methylation, 55 with hyper-CHG methylation, and404 with hyper-CHH methylation. Heat map analysis of hyper-DMRs indicated that the levels of DNA methylation at acx4-4–specific loci were usually preferentially increased in ros1-1 andros4 (Fig. 2C), We picked several genes to confirm their DNAmethylation patterns. Individual locus bisulfite sequencing revealedincreased DNA methylation levels of At4G18250, At4G29380,AT5G52400, AT1G19160, and AT3G51440 in acx4-4 and ros1-1compared with WT (SI Appendix, Fig. S6 A and B). For AT4G29380,only the levels of CHG and CHH DNA methylation were in-creased in acx4-4 and ros1-1, while the level of CG methylation didnot change—probably because it was already very high. Analysis ofthe acx4-4 ros1-1 double mutant revealed no clear additive effectson DNA methylation for AT1G19160 and AT3G51440 (SI Appendix,Fig. S6B). Whole-genome bisulfite sequencing data confirmed that,like the ros4 mutant, acx4-4 exhibited increased DNA methylationonly in the NOS terminator region and not in the RD29A pro-moter (SI Appendix, Fig. S6C). The DNA methylation of someendogenous transposons (here we took two transposons ATREP10Dand HELITRON2 as examples) was increased (SI Appendix, Fig.S6D). These results suggest that ACX4 and ROS1 work in the samepathway to regulate DNA methylation of some loci in Arabidopsis.We next used qRT-PCR to test the expression levels of

13 genes (plus 2 kb of sequence on either side) overlappingwith hypermethylated DMRs in acx4-4, ros1-1, and acx4-4 ros1-1.Transcript levels of these genes were substantially reduced in thesemutants, and the acx4-4 ros1-1 double mutant had no additiveeffects over the single mutants (Fig. 2D). These data suggest that,like ROS1, ACX4 is critical for preventing the transcriptional si-lencing of some endogenous genes through DNA methylation.

Mutations in the Last Two Enzymes of the β-Oxidation Pathway CausePro35S:NPTII Silencing. The first step in the fatty acid β-oxidationpathway in peroxisomes is catalyzed by ACX enzymes and producestrans-2-enoyl-CoAs (34–36). Six ACX genes have been identified in

the Arabidopsis genome with different expression patterns anddistinct substrate specificities (37). The second and third enzymaticreactions are catalyzed by multifunctional proteins (MFPs), whichexhibit enoyl-CoA hydratase and β-hydroxyacyl-CoA dehydrogenaseactivities (35, 38). The last step in the pathway is catalyzed byL-3-ketoacyl-CoA thiolase (KAT), which converts L-3-ketoacyl-CoAs to acetyl-CoAs (35, 39).We next examined whether mutations in the downstream en-

zymes of the β-oxidation pathway also lead to silencing of thePro35S:NPTII transgene. Of the four catalytic activities of MFP2,two—2-trans enoyl-CoA hydratase for long-chain (C18:0) sub-strates and L-3-hydroxyacyl-CoA dehydrogenase for C6:0, C12:0,and C18:0 substrates—are required for β-oxidation (40). MPF2has one homolog, abnormal inflorescence meristem 1 (AIM1),whose mutation results in an abnormal inflorescence meristemphenotype in mature plants (41). The enzyme 3-ketoacyl-CoAthiolase-2 (KAT2/PED1/PKT3) catalyzes the last step of theβ-oxidation pathway to produce one molecule of acetyl-CoA ineach repeat cycle; this gene has higher expression than the tworedundant homologs KAT1 and KAT5 (42, 43). kat2 kat5 doublemutants recapitulate the aim1 phenotype in inflorescence de-velopment and fertility (44).We used the egg-specific CRISPR/Cas9 system (45) to gen-

erate the mfp2-cas9-1 and mfp2-cas9-2 mutants in the Pro-RD29A:LUC/C24 transgenic line (WT) (SI Appendix, Fig. S7 A

Fig. 2. The acx4 mutation causes hypermethylation of DNA in the NOSterminator. (A) Effects of the acx4-4, ros1-1, and ros4 mutations on DNAmethylation in the Pro35S:NPTII promoter by bisulfite sequencing. (B) Effectsof the acx4-4, ros1-1, and ros4 mutations on DNA methylation at the NOSregion by bisulfite sequencing. (C) Heat map showing the methylation levelsof the WT, acx4-4, ros1-1, and ros4 plants in hyper-DMR regions of acx4-4 indifferent sequence contexts. The lighter color indicates a low methylationlevel, and the darker color indicates a high methylation level. (D) Expressionprofiles of the hypermethylated genes or genes near DMRs in 7-d-old acx4-4,ros1-1, and acx4-4 ros1-1 seedlings by qPCR. Error bars represent ± SE. n = 3.

10578 | www.pnas.org/cgi/doi/10.1073/pnas.1904143116 Wang et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

27,

202

0

and D). Meanwhile, we obtained two T-DNA insertion alleles,Salk_098016 (mfp2-2) (40) and GK_787F01 (mfp2-10), and oneT-DNA insertion allele, Salk_024922 (kat2-3), in which MFP2and KAT2 expression were eliminated (SI Appendix, Fig. S7 A–C). We introduced these mutations (in Col-0 accession) intoProRD29A:LUC/C24 transgenic line by crossing. Consistent withNPTII expression levels (Fig. 3 A and C), the mfp2-cas9-1 andmfp2-cas9-2 mutants were more Kan-sensitive than the WT andless sensitive than acx4-4, but did not exhibit altered LUC ex-pression (Fig. 3 B and D). Similarly, the mfp2-2, mfp2-10, andkat2-3 mutations caused more Kan sensitivity than the WT, butthe ProRD29A:LUC transgene was not affected (SI Appendix,Fig. S8A). Bisulfite sequencing analysis showed increased DNAmethylation at the NOS region of the Pro35S:NPTII gene in mfp2-cas9-1 andmfp2-cas9-2mutants compared with theWT, although theDNA methylation at these regions was a little lower comparedwith that in acx4-4 (Fig. 3E). The lesser effect of mfp2 mutationon DNA methylation compared with acx4 suggests a redundancyof the MFP2 homologs in the β-oxidation pathway. Transcrip-tionally silent information (TSI) is regulated by the DDM1pathway, but not by the RdDM pathway (46, 47). qPCR dem-onstrated that the transcript levels of TSI were reduced in mfp2-2, mfp2-10, and kat2-3 mutants compared with Col-0 plants; asimilar reduction was observed in acx4-4 compared with the WT(SI Appendix, Fig. S8B). We found increased CG DNA methyl-ation at the AT4G28910 promoter in two independent mfp2-cas9mutant alleles, and in the acx4-4 mutant (Fig. 3F). qPCR analysis

showed that AT4G28910 was expressed at lower levels in acx4-4,mfp2-cas9-1, and mfp2-cas9-2 mutants compared with the WT(Fig. 3G). Taken together, our results suggest that the β-oxidationpathway in peroxisomes regulates gene silencing and DNAmethylation in Arabidopsis.

MFP2 and KAT2 Target Some Common Loci as ACX4 for DNAMethylation. To determine whether MFP2 and KAT2 also func-tion in modulating DNA methylation at the whole genome level,we performed bisulfite sequencing analyses and identified 1,153DMRs in acx4-1, 1,135 DMRs inmfp2-2, and 686 DMRs in kat2-3compared with the WT (all in the Col-0 background), respectively.Among these, 1,069 were hypermethylated and 84 were hypo-methylated in acx4-1, 1,045 were hypermethylated and 90 werehypomethylated in mfp2-2, and 623 were hypermethylated and63 were hypomethylated in kat2-3 (SI Appendix, Fig. S9). Morethan 60% of these hyper-DMRs were located in genic regions inacx4-1, mfp2-2, and kat2-3, whereas only 40% were located ingenic regions in ros1-4 (Col-0) (Fig. 4A and SI Appendix, Fig. S9).Furthermore, 60.5% of 1,069 hyper-DMRs in acx4-1, 60.7% of1,045 hyper-DMRs in mfp2-2, and 76.1% of 623 hyper-DMRs inkat2-3 overlapped with those in ros1-4 (Fig. 4B). Approximately50% of the 1,069 hyper-DMRs inmfp2-2 and 60% of the 623 hyper-DMRs in kat2-3 were hypermethylated in acx4-1 (Fig. 4B).Profiling the methylation levels of mfp2-2, kat2-3, and ros1-4 in

the loci that were hypermethylated in acx4-1 revealed increasedDNA methylation levels at these loci compared with Col-0 (Fig.4C). Boxplots indicate that at those hyper-DMRs that appearspecific for acx4-1, the average DNA methylation levels in the CGcontext but not the CHG or CHH contexts were higher in themfp2-2, kat2-3, and ros1-4 mutants than in Col-0 plants (Fig. 4D),suggesting that fatty acid β-oxidation preferentially affects genomicregions in the CG context. These data indicate that mutationsin ACX4, MFP2, and KAT2 affect the DNA methylation levels ofsome common genomic regions (SI Appendix, Fig. S10 A–E).

The acx4 Mutant Has Reduced Histone Acetylation. ACX4 catalyzesthe conversion of fatty acyl-CoAs to trans-2-enoyl CoA. This isthought to be the predominant way by which the rate of acetyl-CoAflux is controlled through β-oxidation in peroxisomes. We askedwhether the level of acetyl-CoA is reduced in acx4 mutants. High-performance liquid chromatography tandem mass spectrometry(HPLC-MS/MS) was used to measure the accumulation of acetyl-CoA and its related metabolites butyryl-CoA and hexanoyl-CoA,the main substrates for ACX4. Both butyryl-CoA and hexanoyl-CoA accumulated to much higher levels, but the level of acetyl-CoA was significantly decreased in acx4-4 compared with the WT(Fig. 5A).We next conducted immunoblotting assays to assess whether

defective β-oxidation in the acx4-4 mutant would change thelevel of histone acetylation in vivo. Compared with WT, levels ofall three acetylated (Ac) core histones (H2B, H3, and H4) weredecreased in the acx4-4 mutant (Fig. 5 B and C). Immunos-taining nuclei using H3Ac or H4Ac antibodies showed that levelsof these acetylated histones were significantly reduced in acx4-4compared with the WT or complemented acx4-4 (Fig. 5 D andE). Furthermore, chromatin immunoprecipitation (ChIP)-PCRassays with antibodies for the acetylated histones H2B, H3, andH4 revealed that levels of H3Ac and H4Ac, but not of H2BAc,were substantially reduced at both the 35S promoter and NOSregions (SI Appendix, Fig. S11). These results indicate that his-tone acetylation is reduced in the acx4-4 mutant and suggest thatperoxisome-derived acetyl-CoA is important for histone acetylation.

Decreased Histone H3K18Ac Is Associated with DNA Hypermethylationin acx4 Mutant. Recent studies have demonstrated that the DNAdemethylation complex recognizes acetylated H3K18 and H3K23at some target loci for DNA demethylation, suggesting that histone

Fig. 3. MFP2 prevents transcriptional gene silencing of Pro35S:NPTII. (A)Themfp2 CRISPR/Cas9 mutants are more sensitive to Kan than theWT, butmoreresistant to Kan than acx4-4. Seeds were planted on MS medium supplementedwith 150 mg/L Kan and cultured for 2 wk before being photographed. (B) Themfp2 CRISPR/Cas9 mutants do not affect ProRD29A:LUC. (C) The mfp2 CRISPR/Cas9 mutants and acx4-4 reduced Pro35S:NPTII expression levels compared withthe WT, as determined by qPCR analyses. UBC28 was used as an internalcontrol. (D) qPCR analysis of the expression levels of ProRD29A:LUC in theindicated genotypes.UBC28 served as an internal control. (E) Compared with theWT, increased methylation levels were observed in the NOS region of mfp2CRISPR/Cas9 and acx4-4 mutants, according to bisulfite sequencing analysis. (F)Bisulfite sequencing data showing methylation levels of the AT4G28910 pro-moter in the WT and indicated mutant plants. (G) qPCR analysis of expressionlevels of the AT4G28910 gene in the WT and different mutant plants. UBC28served as an internal control. Data are means ± SE (n = 3) with three replicate inone experiment. Three independent experiments were done with similar results.*P < 0.05.

Wang et al. PNAS | May 21, 2019 | vol. 116 | no. 21 | 10579

PLANTBIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

27,

202

0

acetylation is required for active DNA demethylation (1, 20–23).To better evaluate whether ACX4-dependent generation of acetyl-CoA selectively affects acetylation of histone H3 at lysine residues,we analyzed the levels of histone H3 acetylation at K9, K14, K18,K23, and K27 in acx4-4 and WT by immunoblotting. The levels ofH3K9Ac, H3K18Ac, and H3K23Ac were moderately reduced, butthose of H3K14Ac and H3K27Ac were unchanged (Fig. 6 A and Band SI Appendix, Fig. S12). To determine which histone acetylationmarks are associated with DNA hypermethylation in acx4 mutant,we conducted ChIP followed by sequencing (ChIP-seq) in WT andacx4-4 using antibodies against H3K9Ac, H3K14Ac, H3K18Ac,and H3K23Ac. For H3K14Ac, H3K18Ac, and H3K23Ac ChIP-seqexperiments, two biological replicates were performed. Pearson’scorrelation between two replicates are shown in SI Appendix, Fig. S13. Inour analysis, 15,421 H3K9Ac, 8,460 H3K14Ac, 9,040 H3K18Ac,and 14,884 H3K23Ac peaks were identified in the WT. Consistentwith the immunoblotting results, we found that ChIP-seq reads onH3K9Ac, H3K18Ac, and H3K23Ac peaks, but not those onH3K14Ac peaks, were decreased in acx4-4 relative to WT(Fig. 6C).

We next plotted ChIP-seq reads on acx4-4 hyper-DMRs in WTand acx4-4 and found that the enrichment levels of H3K18Acaround hyper-DMRs, but not other histone acetylation makers,were clearly decreased in acx4-4 relative to WT (Fig. 6D). Inagreement with this finding, boxplots show much lowerH3K18Ac levels of nearby hyper-DMRs in acx4-4 comparedwith WT (Fig. 6E). Several hyper-DMRs shared by ros1-1 andacx4-4 were selected for further investigation by ChIP-qPCR(qPCR) assays, which revealed lower H3K18Ac levels in the acx4-4mutant compared with WT (Fig. 7 A and B). As negative controls,no changes were found at ACTIN2, AT1G10950, and AT1G01260.These results suggest that reduced histone H3K18 acetylation inacx4-4 is correlated with DNA hypermethylation.

Overexpression of ATP-Citrate Lyase A and B Together Can Rescue theKanamycin Sensitivity of acx4-4. Acetyl-CoA is impermeable tomembranes. The foregoing data suggest that mutations of genesin the β-oxidation pathway reduce the acetyl-CoA level in per-oxisomes, which may in turn decrease the acetyl-CoA level incytosol and affect histone acetylation and DNA methylation oraffect other metabolisms, such as the reduced production of

Fig. 4. MFP2 and KAT2 target some common loci asACX4 for DNA methylation. (A) Comparison of thegenomic regions of hyper-DMRs in acx4-1, mfp2-2,kat2-3, and ros1-4 mutants. All mutants are from theCol-0 accession. TE, transposable element. (B) Num-bers of overlapping hyper-DMRs in acx4-1, mfp2-2,kat2-3, and ros1-4 mutants. (C) Heat map comparingthe methylation levels in acx4-1 hyper-DMRs withthe same regions in mfp2-2, kat2-3, and ros1-4 mu-tants. Columns represent the indicated genotypes;rows represent differentially methylated loci. Hyper-DMRs in acx4-1 relative to the WT were used tocompare DNA methylation levels with the same re-gions in mfp2-2, kat2-3, and ros1-4. Light yellow in-dicates a low methylation level, and black indicates ahigh methylation level. (D) Boxplots representingmethylation levels of Col-0, acx4-1, mfp2-2, kat2-3,and ros1-4 mutants in those regions that are hyper-DMRs in acx4-1 in different sequence contexts. Two-tailed Student’s t test, *P < 0.05.

Fig. 5. The acx4mutation reduces histone acetylation.(A) Relative levels of acetyl-CoA, butyryl-CoA, andhexanoyl-CoA determined by HPLC-MS/MS in 10-d-oldseedlings. SD ± SE (n = 3) of three independentbiological repeats is shown. Two-tailed Student’s ttest, *P < 0.05, **P < 0.01, ***P < 0.001. (B) Overalllevels of H2BAc, H3Ac, and H4Ac as determined byimmunoblotting. Total proteins extracted from theWT and acx4-4 seedlings were used for immunoblot-ting with anti-H2BAc, -H3Ac, and -H4Ac antibodies.H3 was detected as a loading control. (C) Quantifiedmean data normalized to H3 showing decreased lev-els of acetylated H2B, H3, and H4 in acx4-4 comparedwith the WT. Error bars represent ± SE from threeindependent experiments (n = 3). (D and E) Levels ofacetylated H3 (D) and H4 (E) in WT, acx4-4, and thecomplemented acx4 line 9, as indicated by immuno-fluorescence under confocal microscopy. Acetylated-H3/H4 colocalized with the nuclear stain DAPI (blue).Representative images (Left) and quantified meandata (Right) are shown. Values are means ± SE, n = 40.**P < 0.01.

10580 | www.pnas.org/cgi/doi/10.1073/pnas.1904143116 Wang et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

27,

202

0

some phytohormones, as some products of β-oxidation may go todifferent metabolite pathways to produce MeJA, SA, or auxin(48, 49). Under different hormone treatments, such as NAA, SA,or MeJA, the Kan sensitivity of acx4-4 mutant was not changed(SI Appendix, Fig. S14A), indicating that Pro35S:NPTII silencingis not regulated by these phytohormones.We further added some possible precursors for acetyl-CoA

biosynthesis, such as succinate, acetate, oxoglutarate, and citricacid, to the growth medium to see whether these chemicals couldrescue the Kan-sensitive phenotype of acx4-4, but found that theycould not. We then searched for a strategy to increase cytosolicacetyl-CoA in vivo. In the cytosol, ATP-citrate lyase (ACL) is thesole key enzyme for generating acetyl-CoA (50). Unlike in animals,in which ACL has a monomeric structure, plant ACL consists oftwo distinct subunits, ACLA (45 kDa) and ACLB (65 kDa), and theheterooctomer holoenzyme is expected to have an A(4)B(4) con-figuration (50). We speculated that if ACL is overexpressed inacx4-4 mutant, more acetyl-CoA will be produced, which maycompromise the Kan-sensitive phenotype of acx4-4. We expressedACLA1-GFP (AT1G10670) and ACLB2-MYC (AT5G49460)under the control of 35S promoter in acx4-4. We obtained trans-genic lines and tested their Kan sensitivity, but did not find anydifference between acx4-4 and acx4-4 expressing ACLA1-GFPor ACLB-MYC (SI Appendix, Fig. S14 B and C), suggestingthat overexpressing one subunit of ACL does not affect thesilenced Pro35S:NPTII.We then crossed different transgenic lines of ACLA1-GFP

with those of ACLB2-MYC and obtained F1 seeds. Proteinimmunostaining using GFP and MYC antibodies and qPCR in-dicated that the ACLA1-GFP/ACLB2-MYC F1 #12 and #6 linesshowed high levels of both ACLA1-GFP and ACLB2-MYC pro-tein and high transcript levels of the two genes (Fig. 8 A and B).The acetyl-CoA content in ACLA1-GFP/ACLB2-MYC F1 #12was three times that in the WT (Fig. 8C). Both lines #12 and #6had the Kan-sensitive phenotype of acx4-4 rescued and expressedhigher levels of NPTII than acx4-4 (Fig. 8 D and E). Bisulfite

sequencing revealed that the DNA methylation of endogenoustarget gene AT4G18250 was recovered to WT level in both lines#6 and #12 in F2 Kan-resistant seedlings (Fig. 8F). These resultsindicate that increasing acetyl-CoA content rescued the acx4-4 Kansensitivity.

DiscussionAcetyl-CoA is a central metabolite produced from fatty acids,glucose, and amino acids. It is a donor of single acetyl groups forhistone acetylation and is crucial for cell growth and pro-liferation (39, 51, 52). Because it is impermeable to membranes,acetyl-CoA can be produced in different organelles, such asperoxisomes, mitochondria, and plastids (5). In these organelles,acetyl-CoA must be condensed to other molecules before beingexported to the cytosol. However, whether organellar acetyl-CoA can affect nuclear DNA methylation in pants is unknown.In this study, we provide genetic and molecular evidence showingthat a deficiency of short-acid β-oxidation can reduce histoneacetylation and increase DNA methylation at a subset of geno-mic loci in Arabidopsis, suggesting a close connection betweenorganelle metabolites and nuclear epigenetic modifications.DNA methylation and histone modifications are primary

epigenetic markers for the regulation of chromatin stability andtranscription (12, 53). In our genetic screen, we identified theROS4/IDM1, ROS5/IDM2, and MBD7 genes for their roles inDNA demethylation in the ProRD29A:LUC/C24 transgenic line(18, 22, 23). The current favored model suggests that MBD7 bindsto genomic regions with high CG methylation density and physi-cally associates with ROS4/IDM1 to create acetylated H3K18 andH3K23 marks, which in turn facilitate recruitment of the DNAdemethylation enzyme ROS1 (1, 18). These studies suggest thathistone acetylation is required for DNAdemethylation enzymes to targetgenomic loci to prevent DNA hypermethylation and gene silencing.Using the same ProRD29A:LUC screening system, we identi-

fied ACX4 and found that compared with WT, acx4 mu-tants have greatly reduced histone acetylation levels. In acx4

Fig. 6. The hyper-DMRs in acx4-4 are associatedwith reduced H3K18Ac. (A) The acx4-4 mutant reduceshistone acetylation. Histone proteins were extractedand used for immunoblotting with different anti-bodies. Histone H3 served as a loading control. (B)Quantified mean data normalized to H3 showingthe relative levels of different H3 acetylation modi-fications in acx4-4 compared with the WT. Error barsrepresent ± SE from three independent experiments.n = 3. (C) Comparison of ChIP-seq reads with differenthistone acetylation antibodies between acx4-4 andtheWT. −2 kb and +2 kb represent 2 kb upstream and2 kb downstream of the peak middle point site, re-spectively. (D) Correlation analyses of DMRs with dif-ferent histone acetylation modifications betweenacx4-4 and the WT. Only the enrichment levels ofH3K18Ac are found to be decreased in acx4-4 relativeto the WT. −2 kb and +2 kb represent 2 kb upstreamand 2 kb downstream of middle point site of thehyper-DMR region, respectively. (E) Boxplots showingthat the H3K18Ac levels of nearby hyper-DMRs aresignificantly lower in acx4-4 than inWT. 50 bp, 100 bp,150 bp, and 200 bp represent the distances upstreamand downstream of the middle point site of thehyper-DMR region. Two-tailed Student’s t test, **P <0.01.

Wang et al. PNAS | May 21, 2019 | vol. 116 | no. 21 | 10581

PLANTBIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

27,

202

0

hyper-DMRs, the levels of H3K18Ac were reduced relative tothe WT (Fig. 6 D and E), suggesting a close connection betweenDNA hypermethylation and histone acetylation at these loci. Acomparison of DNA hypermethylated loci showed that in acx4,hyper-DMRs partially overlap with those identified in the ros1mutant (Fig. 4B), supporting the notion that ACX4 functions inROS1-mediated DNA demethylation. The presence of otherhyper-DMRs in the acx4 mutant that did not increase in the ros1mutant (Figs. 2C and 4C) suggests that in these loci, DNAdemethylation is controlled by ROS1 paralogs, or histone acetylationdirectly affects DNA methylation, as seen in had6 mutants (54).The expression levels of hypermethylated genes, or genes locatednear hyper-DMRs, were reduced in the acx4-4 and ros1-1 mu-tants, and the acx4-4 ros1-1 double mutant had no additive effecton gene expression relative to the single mutants (Fig. 2D),suggesting the importance of ACX4 in preventing transcriptionalgene silencing by DNA demethylation.We found that acx4-1,mfp2-2, and kat2-3 mutants have similar

hyper-DMRs, suggesting that β-oxidation in peroxisomes is crucialfor modulating DNA methylation in nuclei. In plants and microbes,β-oxidation occurs in peroxisomes, while in mammalian cells, itoccurs mainly in mitochondria. In mammalian cells, the addition ofbutyrate, a short-chain fatty acid that can be oxidized in mitochondria,induces ACL-mediated histone acetylation (55). Both cytosolic and

mitochondrial acetyl-CoA synthetases (ACSS2 and ACSS1) are re-quired for acetate-induced histone acetylation under hypoxia incancer cells (56). Acetyl-CoA produced by the β-oxidation pathwayin peroxisomes may have a similar role in modulating nuclear histoneacetylation in plants, which in turn would affect DNA methylation.We found that the levels of all three acetylated core histones

(H2B, H3, and H4) were down-regulated in acx4 compared withthe WT (Fig. 5 B and C), demonstrating that ACX4 is importantfor histone acetylation. However, along with affecting DNAmethylation, histone acetylation usually has a generally positiveeffect on gene expression. The globally reduced histone acetylationin acx4 mutants would also affect the expression of genes notrelated to DNA methylation, which merits future exploration. Sinceacetyl-CoA can be produced in various metabolic pathways in plantcells (5) and other organisms (39), further studies are needed todemonstrate whether these other pathways are also involved inhistone acetylation and DNA demethylation.In plants, the β-oxidation of storage fatty acids is required to

establish seedlings during germination. This is coupled with theglyoxylate cycle to convert acetyl-CoA into succinate. Succinatecan be used for amino acid biosynthesis or can be converted intosucrose by gluconeogenesis. Sucrose can then be catabolized intopyruvate by glucose oxidation and converted to acetyl-CoA bythe pyruvate dehydrogenase complex in mitochondria. Acetyl-CoA

Fig. 7. Comparison of hyper-DMRs and reduced H3K18Ac at some loci in acx4, ros1-1, and WT. (A) DMRs and H3K18Ac levels in different loci among acx4-4andWT. The DNAmethylation levels in ros1-1were included. (B) ChIP-qPCR confirmation of H3K18Ac levels in different DMRs between acx4-4 and WT. Valuesare means ± SE of three replicates in one experiment. Three independent experiments were done with similar results. Two-tailed Student’s t test, *P < 0.05,**P < 0.01, ***P < 0.001, n.s., no significant difference.

10582 | www.pnas.org/cgi/doi/10.1073/pnas.1904143116 Wang et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

27,

202

0

can enter the tricarboxylic acid cycle in mitochondria to producecitrate, which can be exported to the cytosol and converted toacetyl-CoA by ACL (10). However, in later developmental stages,neither β-oxidation nor the glyoxylate cycle is considered essential,because photosynthesis can provide the carbohydrates and precur-sors required for protein and nucleic acid biosynthesis. In our study,we used Murashige and Skoog (MS) medium supplemented with2% sucrose (sufficient for heterotrophic seedling growth) to isolatethe acx4 mutant. Contrary to previous studies indicating that ex-ogenous application of sucrose can largely rescue the growth defectsof various mutants in the β-oxidation pathways (29, 32, 57), ourstudy indicates that Kan-sensitive acx4 phenotypes could not berescued by sucrose or by any other possible precursors for acetyl-CoA biosynthesis, such as succinate, acetate, oxoglutarate, and citricacid. Peroxisomal β-oxidation not only functions in fatty acid ca-tabolism, but also is required for the metabolism of hormones andamino acids (58). Similarly, exogenous application of hormones,including jasmonate, salicylate, and NAA, could not rescue thephenotype of Kan sensitivity in acx4 (SI Appendix, Fig. S14A). Theseresults suggest that exogenous addition of these components cannotbypass the β-oxidation defect in the acx4 mutant to rescue the si-lencing phenotype. We speculate that the metabolites provided byβ-oxidation in cells might be used more efficiently than exogenoussources for biosynthesis of the acetyl-CoA donor for histone acet-ylation, or that active DNA demethylation is a slow process that

cannot be reflected so quickly after addition of sucrose or othercompounds in the medium. Another possibility is that active DNAdemethylation is preferentially active during early development,such that once high levels of DNA methylation are establishedduring seed development, they cannot be decreased by later addi-tion of the precursors for acetyl-CoA to the seedlings. We foundthat when acx4-4 expressed both ACLA1-GFP and ACLB2-MYC,more acetyl-CoA was produced, and the Kan-sensitive phenotypeand DNA hypermethylation were rescued, suggesting that acetyl-CoA plays a critical role in mediating DNA demethylation andantisilencing.The status of cellular metabolism is closely connected to epi-

genetic modifications (59, 60). Nutrition is well known to affectdevelopment and diseases through metabolism that modulatesepigenetic modifications in mammals (61). Further elucidationof the connections between peroxisomal metabolism and chro-matin regulation will enhance our understanding of epigeneticregulation and development.

MethodsPlant Growth Conditions and Mutant Screening. Seeds sterilized with 0.5%NaClOwere sown ontoMSmedium plates containing 2% (wt/vol) sucrose and0.8% (wt/vol) agar. After 3–4 d at 4 °C, the plates were transferred to agrowth chamber at 22 °C under long-day conditions (23-h light/1-h dark).Seven-day-old seedlings were transferred to soil and cultivated in a green-house at 20 °C under long-day conditions (16-h light/8-h dark).

The WT C24, ros1-1, ros4, rdr2, ddm1, acx4-4,mfp2-cas9-1, andmfp2-cas9-2 mutants (C24 background) mentioned in this study all carried the RD29A-LUC and Pro35S:NPTII transgenes (16, 18). The T-DNA lines ros1-4, SALK_000879(acx4-1), SALK_065013 (acx4-3), SALK_098016 (mfp2-2), GK_787F01 (mfp2-10),and SALK_024922 (kat2-3) were obtained from the Arabidopsis Stock Center.

The acx4-4 mutant was identified from an EMS-mutagenized WT pop-ulation as described previously (18). In brief, since WT seedlings can grow onMS medium containing 50 mg/L Kan, mutants that did not grow well on thismedium were selected and transferred to MS medium for recovery. TheirKan-sensitive phenotype was confirmed in the next generation. Putative mu-tants were crossed with a Columbia accession (gl1), and Kan-sensitive plantsfrom the F2 seedlings were assayed by PCR (using primers Pro35S:NPTII-F andPro35S:NPTII-R). Mutants (approximately 2,000) containing the T-DNA inser-tion were selected for mapping.

For the complementation experiment, the genomic sequence of ACX4from −2,543 to +3,546 bp was cloned into the SacI and SmaI (XmaI) sites ofpCAMBIA1300; this sequence contains the 2,543-bp promoter, the codingregion, and the 190-bp 3′ region. Using Agrobacterium tumefaciens GV3101,this construct was transformed into the acx4-4 mutant. Transgenic plantswere selected and analyzed for sensitivity to Kan.

Individual Loci DNA Methylation Analysis. Genomic DNA was extracted usingthe Qiagen DNeasy Plant Mini Kit. DNA methylation at individual loci wasanalyzed using bisulfite sequencing.

For bisulfite sequencing, 500 ng of genomic DNA was treated with the EZMethylation-Gold Kit (Zemo Research), following the manufacturer’s pro-tocol. Approximately 75 ng of bisulfite-treated DNA was used for PCR withthe specific primers listed in SI Appendix, Table S1. PCR products were clonedinto the pMD18-T vector (Takara Bio), and at least 15 independent clones fromeach sample were sequenced for each region.

Real-Time qPCR. Total RNA was extracted from 7-d-old seedlings using TRIzolreagent (Invitrogen), and contaminating DNAwas removed using RNase-freeDNase I (Takara Bio). Approximately 4 μg of mRNA was used for first-strandcDNA synthesis using Moloney Murine Leukemia Virus Reverse Transcriptase(Promega) in a 20-μL reaction volume, and 5 μL of a 1:20 dilution of the cDNAreaction mixture was used as a template in a 20-μL PCR with SYBR GreenMaster Mix (Takara Bio) on a Step One Plus machine (Applied Biosystems) withthree technical replicates. UBC28 served as an internal control. RNA transcriptlevels were determined by qPCR in a 20-μL reaction mixture, with TUB8 as aninternal control. The primers used for PCR are listed in SI Appendix, Table S1.

Histochemical GUS Staining. The ACX4 promoter fragment (from −2,620 to−1 bp) and the GUS-coding fragment were cloned into the pCAMBIA1391vector. The recombinant plasmid was introduced into A. tumefaciens strainGV3101 and then transformed into C24 WT plants. Histochemical GUS stainingwas conducted on the T2 transgenic plants as described previously (62).

Fig. 8. Overexpression of ACLA1 and ACLB2 together can rescue the Kan-sensitive phenotype of acx4-4. (A) Protein levels detected with differentantibodies in different F1 seedlings of acx4-4 expressing both ACLA1-GFPand ACLB2-MYC. The F1 seedlings were obtained by crossing the differentlines of acx4-4 expressing ACLA1-GFP with those expressing ACLB2-MYC.ACTIN served as a loading control. (B) Relative expression levels of ACLA1and ACLB2 in two F1 lines as determined by qPCR. (C) Relative levels ofAcetyl-CoA in F1 seedlings. Values are means ± SE (n = 3) of three in-dependent biological repeats is shown. Two-tailed Student’s t test, **P <0.01. (D) Comparison of Kan-resistant phenotypes among WT, acx4-4, andtwo F1 lines on MS containing 25 mg/L Kan. (E) Relative transcriptional levelsof NPTII in WT, acx4-4, and two F1 lines. (F) Bisulfite sequencing analyses ofDNA methylation for endogenous gene AT4G18250 in WT, acx4-4, and twoacx4-4 lines expressing both ACLA1 and ACLB2 that were isolated fromF2 seedlings with Kan resistance.

Wang et al. PNAS | May 21, 2019 | vol. 116 | no. 21 | 10583

PLANTBIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

27,

202

0

Histone Extraction and Immunoblotting. Ten-day-old seedlings were ground toa fine powder in liquid nitrogen and suspended in nuclear isolation buffer(250 mM sucrose, 60 mM KCl, 15 mM NaCl, 5 mM MgCl2, 1 mM CaCl2, 15 mMPipes pH 6.8, and 0.8% Triton X-100). The mixture was centrifuged, and thepellet was resuspended in 0.4 M H2SO4 and incubated for at least 1 h on ice.The preparation was centrifuged again, followed by the addition of a 20-foldvolume of acetone to precipitate the histone protein. The preparation waskept at −20 °C overnight, after which the proteins were dissolved in 4 M ureaand then used for protein blot analysis with the following antibodies: anti-H2BAc (ab1759; Abcam), anti-H3 (17–10046; EMD Millipore), anti-H3Ac (7–615;EMD Millipore), anti–Pan-H3Ac (ab47915; Abcam), anti-H3K9ac (07–352; EMDMillipore), anti-H3K14ac (07–353; EMD Millipore), anti-H3K18ac (ab1191;Abcam), anti-H3K23ac (07-355; EMD Millipore), and anti-H4Ac (17–630; EMDMillipore). Anti-H3 immunoblot was used as a loading control.

Immunofluorescence Assay. An immunofluorescence assay was performedusing 10-d-old seedlings. Nuclei were fixed in 4% paraformaldehyde andblocked with 3% BSA in PBS. The primary antibody H3Ac (7–615; EMD Mil-lipore) was diluted at 1:100, and H4Ac (17–630; EMD Millipore) was dilutedat 1:200, incubated overnight at 4 °C, and finally incubated with rabbit AlexaFluor 594 (A11012m; Invitrogen)-conjugated secondary antibodies for 2 h at37 °C. Chromatin was counterstained with DAPI in Prolong Gold (Invitrogen).Images were acquired with a Leica TCS SP8 STED 3X confocal microscope at100× magnification.

ChIP Assays. ChIP assayswere performed as described previously (63) using 10-d-old seedlings and the following antibodies: anti-H2BAc (ab1759; Abcam), anti-H3Ac (17–615; EMD Millipore), anti-H4Ac (17–630; EMD Millipore), anti-H3K9ac(07–352; EMD Millipore), anti-H3K14ac (07–353; EMD Millipore), anti-H3K18ac(ab1191; Abcam), anti-H3K27ac (07-360; EMD Millipore), and anti-H3K23ac (07-355; EMD Millipore). ChIP products were eluted into 50 μL of Tris-EDTA bufferand diluted to a ratio of 1:5, and then a 2-μL aliquot of this dilution was used foreach qPCR reaction. The primers used for qPCR are listed in SI Appendix, Table S1.

ChIP-Seq and Data Analysis. The ChIP DNA concentration was determinedusing a fluorescence-based quantification method (Qubit 3.0 system; LifeTechnologies). High-throughput sequencing libraries were prepared usingthe NEBNext ChIP-Seq Library Prep Master Mix Set for Illumina (E6240; NewEngland BioLabs) according to the manufacturer’s instructions. Sequencingwas performed on an Illumina HiSEq 2500 instrument with single-end 50-bpreads. The reads were aligned to the TAIR10 Arabidopsis reference genomeusing Bowtie (64) with default parameters. Reads that perfectly and uniquelymapped to the genome were retained for further analysis. For each ChIP-seq,mapped reads were pooled from both replicates. Wiggle (WIG) format wasgenerated using MACS program (65) and visualized using IGV (66). Acetylation-enriched peaks were identified using SICER (67) with a window size settingof 200 bp, gap size of 200 bp, false discovery rate (FDR) ≤0.05, and foldchange ≥1.5. The genome-wide occupancies, boxplots, and Pearson corre-lations were performed with R packages.

Localization of the mCherry-ACX4 Fusion Protein. The red fluorescent proteinmCherry was fused to the N terminus of ACX4 under the control of the 35Spromoter, and this construct was cloned into a modified pCAMBIA1300 vector.The vector carrying YFP-CD3-990 was generated in a previous study (30). NLSwas fused to the N terminus of GFP under the control of the 35S promoter and

cloned into the pCAMBIA1300 vector to generate NLS-GFP. The peroxisomalmarker GFP-CAT2 was obtained from a previous study (31). The transient expres-sion assay in tobacco and image acquisition were also described previously (68).

Whole-Genome Bisulfite Sequencing and Data Analysis. For plants in theC24 background, genomic DNAwas isolated from 7-d-old seedlings using theQiagen DNeasy Plant Mini Kit and sent to BGI (Shenzhen, China) for bisulfitetreatment, library preparation, and sequencing using the Illumina HiSeq2000 sequencing system. Raw reads obtained from sequencing were trimmedusing SolexaQA software (69). Clean reads were mapped to the Arabidopsisreference sequence (TAIR 10) using Bismark, allowing two mismatches. DMRidentification was performed as described previously (23). Bins applied inthis study were 100 bp from the reference genome. DMR identification usedthe standard of an absolute difference in methylation levels in the CG, CHG,and CHH contexts of at least 0.4, 0.2, and 0.1, respectively, and the Benjamini–Hochberg-corrected FDR was <0.01 (Fisher’s exact test).

For plants in the Col-0 background, genomic DNAwas extracted from 14-d-old seedlings using the DNeasy Plant Mini Kit and sent to BGI for bisulfitetreatment, library preparation, and sequencing using the Illumina HiSeq4000 system. For library preparation, 5 μg of genomic DNA was sonicatedinto 100- to 300-bp fragments, which were end-repaired and ligated withadenine at their 3′ ends. After ligation with Illumina DNA adaptors, unme-thylated cytosine residues were converted to uracils using the Qiagen Epi-Tect Bisulfite Kit.

For data analysis, clean reads were mapped to the TAIR 10 genome usingBismark, allowing up to two mismatches. DMR identification was performed asdescribed previously (70). In brief, the DNA methylation level in every 200-bpwindow at 50-bp intervals was compared between WT and mutant seedlingsusing Fisher’s exact test with P ≤ 0.05. FDRs were estimated using the Benjamini–Hochberg correction of calculated Fisher’s P values. Windows with seven ormore differentially methylated cytosines (defined as C with P < 0.01 in Fisher’sexact test) and a >1.5-fold change in DNA methylation levels were retained andcombined if the gap size was no more than 100 bp to generate DMRs. Finally,DMR length was adjusted to start at the first mC and end at the last mC.

Metabolite Measurement. Ten-day-old Arabidopsis seedlings grown on platescontaining MS medium were used to measure acetyl-CoA, butyryl-CoA, andhexanoyl-CoA levels. Sample preparation and procedures for measuringmetabolites were described previously (71). The experiments were per-formed with three biological replicates.

Data Availability. Sequence data referred to in this article can be found in theGenBank/EMBL databases under the following accession numbers: AT2G36490 forROS1, AT3G14980 for ROS4/IDM1, AT5G66750 for DDM1, AT4G11130 forRDR2, AT3G51840 for ACX4, AT3G06860 for MFP2, AT2G33150 for KAT2,AT1G64230 for UBC28, and AT3G18780 for ACTIN2. Primary datasets for thewhole-genome bisulfite sequences of Col-0, ros1-4, acx4-4, acx4-1, mfp2-2,and kat2-3 mutant plants have been deposited in the Gene ExpressionOmnibus (GEO) database (accession no. GSE98214), as have histone acetylationChIP-seq data (accession no. GSE98214). Whole-genome bisulfite sequencingdata of C24 WT, ros1-1, and ros4 plants were obtained from the GEO database(accession no. SRP042060) (23).

ACKNOWLEDGMENTS. This study was supported by the Natural ScienceFoundation of China (Grant 31330041).

1. Qian W, et al. (2012) A histone acetyltransferase regulates active DNA demethylationin Arabidopsis. Science 336:1445–1448.

2. Sutendra G, et al. (2014) A nuclear pyruvate dehydrogenase complex is important forthe generation of acetyl-CoA and histone acetylation. Cell 158:84–97.

3. Tariq M, Paszkowski J (2004) DNA and histone methylation in plants. Trends Genet 20:244–251, and erratum (2005) 21:36.

4. Vogelauer M, Rubbi L, Lucas I, Brewer BJ, Grunstein M (2002) Histone acetylationregulates the time of replication origin firing. Mol Cell 10:1223–1233.

5. Shen Y, Wei W, Zhou DX (2015) Histone acetylation enzymes coordinate metabolismand gene expression. Trends Plant Sci 20:614–621.

6. Sivanand S, Viney I, Wellen KE (2018) Spatiotemporal control of acetyl-CoA metab-olism in chromatin regulation. Trends Biochem Sci 43:61–74.

7. Wong BW, et al. (2017) The role of fatty acid β-oxidation in lymphangiogenesis.Nature 542:49–54.

8. Nagaraj R, et al. (2017) Nuclear localization of mitochondrial TCA cycle enzymes as acritical step in mammalian zygotic genome activation. Cell 168:210–223.e11.

9. Chen C, et al. (2017) Cytosolic acetyl-CoA promotes histone acetylation predominantlyat H3K27 in Arabidopsis. Nat Plants 3:814–824.

10. Fatland BL, et al. (2002) Molecular characterization of a heteromeric ATP-citrate lyasethat generates cytosolic acetyl-coenzyme A in Arabidopsis. Plant Physiol 130:740–756.

11. He XJ, Chen T, Zhu JK (2011) Regulation and function of DNA methylation in plantsand animals. Cell Res 21:442–465.

12. Law JA, Jacobsen SE (2010) Establishing, maintaining and modifying DNA methyl-ation patterns in plants and animals. Nat Rev Genet 11:204–220.

13. Zhang H, Lang Z, Zhu JK (2018) Dynamics and function of DNA methylation in plants.Nat Rev Mol Cell Biol 19:489–506.

14. Wu SC, Zhang Y (2010) Active DNA demethylation: Many roads lead to Rome. Nat RevMol Cell Biol 11:607–620.

15. Gehring M, et al. (2006) DEMETER DNA glycosylase establishes MEDEA polycombgene self-imprinting by allele-specific demethylation. Cell 124:495–506.

16. Gong Z, et al. (2002) ROS1, a repressor of transcriptional gene silencing in Arabi-dopsis, encodes a DNA glycosylase/lyase. Cell 111:803–814.

17. Parrilla-Doblas JT, Ponferrada-Marín MI, Roldán-Arjona T, Ariza RR (2013) Early stepsof active DNA demethylation initiated by ROS1 glycosylase require three putativehelix-invading residues. Nucleic Acids Res 41:8654–8664.

18. Li X, et al. (2012) Antisilencing role of the RNA-directed DNA methylation pathwayand a histone acetyltransferase in Arabidopsis. Proc Natl Acad Sci USA 109:11425–11430.

19. Duan CG, et al. (2017) A pair of transposon-derived proteins function in a histoneacetyltransferase complex for active DNA demethylation. Cell Res 27:226–240.

10584 | www.pnas.org/cgi/doi/10.1073/pnas.1904143116 Wang et al.

Dow

nloa

ded

by g

uest

on

Oct

ober

27,

202

0

20. Lang Z, et al. (2015) The methyl-CpG-binding protein MBD7 facilitates active DNAdemethylation to limit DNA hyper-methylation and transcriptional gene silencing.Mol Cell 57:971–983.

21. Qian W, et al. (2014) Regulation of active DNA demethylation by an α-crystallin do-main protein in Arabidopsis. Mol Cell 55:361–371.

22. Wang C, et al. (2015) Methyl-CpG-binding domain protein MBD7 is required for activeDNA demethylation in Arabidopsis. Plant Physiol 167:905–914.

23. Zhao Y, et al. (2014) REPRESSOR OF SILENCING5 encodes a member of the small heatshock protein family and is required for DNA demethylation in Arabidopsis. Plant Cell26:2660–2675.

24. Duan CG, et al. (2015) MET18 connects the cytosolic iron-sulfur cluster assemblypathway to active DNA demethylation in Arabidopsis. PLoS Genet 11:e1005559.

25. Gong Z (2019) A SUVH-DNAJ/SDJ protein complex activates the expression ofpromoter-methylated genes in Arabidopsis. J Integr Plant Biol 61:90–92.

26. Harris CJ, et al. (2018) A DNA methylation reader complex that enhances genetranscription. Science 362:1182–1186.

27. Xiao X, et al. (2019) A group of SUVH methyl-DNA binding proteins regulate ex-pression of the DNA demethylase ROS1 in Arabidopsis. J Integr Plant Biol 61:110–119.

28. Zhao QQ, Lin RN, Li L, Chen S, He XJ (2019) A methylated-DNA-binding complex re-quired for plant development mediates transcriptional activation of promotermethylated genes. J Integr Plant Biol 61:120–139.

29. Adham AR, Zolman BK, Millius A, Bartel B (2005) Mutations in Arabidopsis acyl-CoAoxidase genes reveal distinct and overlapping roles in beta-oxidation. Plant J 41:859–874.

30. Nelson BK, Cai X, Nebenfuhr A (2007) A multicolored set of in vivo organelle markersfor co-localization studies in Arabidopsis and other plants. Plant J 51:1126–1136.

31. Li J, et al. (2015) A chaperone function of NO CATALASE ACTIVITY1 is required tomaintain catalase activity and for multiple stress responses in Arabidopsis. Plant Cell27:908–925.

32. Hayashi H, et al. (1999) A novel acyl-CoA oxidase that can oxidize short-chain acyl-CoAin plant peroxisomes. J Biol Chem 274:12715–12721.

33. Lei M, et al. (2015) Regulatory link between DNA methylation and active demethy-lation in Arabidopsis. Proc Natl Acad Sci USA 112:3553–3557.

34. Kindl H (1993) Fatty acid degradation in plant peroxisomes: Function and biosynthesisof the enzymes involved. Biochimie 75:225–230.

35. Pan R, Liu J, Hu J (December 21, 2018) Peroxisomes in plant reproduction and seed-related development. J Integr Plant Biol10.1111/jipb.12765.

36. Shockey JM, Fulda MS, Browse JA (2002) Arabidopsis contains nine long-chain acyl-coenzyme a synthetase genes that participate in fatty acid and glycerolipid metab-olism. Plant Physiol 129:1710–1722.

37. Graham IA, Eastmond PJ (2002) Pathways of straight and branched chain fatty acidcatabolism in higher plants. Prog Lipid Res 41:156–181.

38. Preisig-Müller R, Gühnemann-Schäfer K, Kindl H (1994) Domains of the tetrafunc-tional protein acting in glyoxysomal fatty acid beta-oxidation: Demonstration ofepimerase and isomerase activities on a peptide lacking hydratase activity. J BiolChem 269:20475–20481.

39. Pietrocola F, Galluzzi L, Bravo-San Pedro JM, Madeo F, Kroemer G (2015) Acetyl co-enzyme A: A central metabolite and second messenger. Cell Metab 21:805–821.

40. Rylott EL, et al. (2006) The Arabidopsis thaliana multifunctional protein gene (MFP2)of peroxisomal beta-oxidation is essential for seedling establishment. Plant J 45:930–941.

41. Richmond TA, Bleecker AB (1999) A defect in beta-oxidation causes abnormal in-florescence development in Arabidopsis. Plant Cell 11:1911–1924.

42. Castillo MC, León J (2008) Expression of the beta-oxidation gene 3-ketoacyl-CoAthiolase 2 (KAT2) is required for the timely onset of natural and dark-induced leafsenescence in Arabidopsis. J Exp Bot 59:2171–2179.

43. Germain V, et al. (2001) Requirement for 3-ketoacyl-CoA thiolase-2 in peroxisomedevelopment, fatty acid beta-oxidation and breakdown of triacylglycerol in lipidbodies of Arabidopsis seedlings. Plant J 28:1–12.

44. Wiszniewski AAG, Bussell JD, Long RL, Smith SM (2014) Knockout of the two evolu-tionarily conserved peroxisomal 3-ketoacyl-CoA thiolases in Arabidopsis recapitulatesthe abnormal inflorescence meristem 1 phenotype. J Exp Bot 65:6723–6733.

45. Wang ZP, et al. (2015) Egg cell-specific promoter-controlled CRISPR/Cas9 efficientlygenerates homozygous mutants for multiple target genes in Arabidopsis in a singlegeneration. Genome Biol 16:144.

46. Zhang J, et al. (2016) The second subunit of DNA polymerase delta is required forgenomic stability and epigenetic regulation. Plant Physiol 171:1192–1208.

47. Steimer A, et al. (2000) Endogenous targets of transcriptional gene silencing inArabidopsis. Plant Cell 12:1165–1178.

48. Wasternack C, Song S (2017) Jasmonates: Biosynthesis, metabolism, and signaling byproteins activating and repressing transcription. J Exp Bot 68:1303–1321.

49. Bussell JD, Reichelt M, Wiszniewski AAG, Gershenzon J, Smith SM (2014) PeroxisomalATP-binding cassette transporter COMATOSE and the multifunctional protein ab-normal INFLORESCENCE MERISTEM are required for the production of benzoylatedmetabolites in Arabidopsis seeds. Plant Physiol 164:48–54.

50. Fatland BL, Nikolau BJ, Wurtele ES (2005) Reverse genetic characterization of cytosolicacetyl-CoA generation by ATP-citrate lyase in Arabidopsis. Plant Cell 17:182–203.

51. Cai L, Sutter BM, Li B, Tu BP (2011) Acetyl-CoA induces cell growth and proliferationby promoting the acetylation of histones at growth genes. Mol Cell 42:426–437.

52. Öst A, Pospisilik JA (2015) Epigenetic modulation of metabolic decisions. Curr OpinCell Biol 33:88–94.

53. Jacob Y, et al. (2010) Regulation of heterochromatic DNA replication by histoneH3 lysine 27 methyltransferases. Nature 466:987–991.

54. Earley KW, et al. (2010) Mechanisms of HDA6-mediated rRNA gene silencing: Suppression ofintergenic Pol II transcription and differential effects on maintenance versus siRNA-directedcytosine methylation. Genes Dev 24:1119–1132.

55. Donohoe DR, et al. (2012) The Warburg effect dictates the mechanism of butyrate-mediated histone acetylation and cell proliferation. Mol Cell 48:612–626.

56. Gao X, et al. (2016) Acetate functions as an epigenetic metabolite to promote lipidsynthesis under hypoxia. Nat Commun 7:11960.

57. Eastmond PJ, Graham IA (2000) The multifunctional protein AtMFP2 is co-ordinatelyexpressed with other genes of fatty acid beta-oxidation during seed germination inArabidopsis thaliana (L.) Heynh. Biochem Soc Trans 28:95–99.

58. Reumann S, Ma C, Lemke S, Babujee L (2004) AraPerox: A database of putativeArabidopsis proteins from plant peroxisomes. Plant Physiol 136:2587–2608.

59. Keating ST, El-Osta A (2015) Epigenetics and metabolism. Circ Res 116:715–736.60. Meng J, et al. (2018) METHIONINE ADENOSYLTRANSFERASE4 mediates DNA and

histone methylation. Plant Physiol 177:652–670.61. Carrer A, Wellen KE (2015) Metabolism and epigenetics: A link cancer cells exploit.

Curr Opin Biotechnol 34:23–29.62. Xia R, et al. (2006) ROR1/RPA2A, a putative replication protein A2, functions in epi-

genetic gene silencing and in regulation of meristem development in Arabidopsis.Plant Cell 18:85–103.

63. Saleh A, Alvarez-Venegas R, Avramova Z (2008) An efficient chromatin immunopre-cipitation (ChIP) protocol for studying histone modifications in Arabidopsis plants.Nat Protoc 3:1018–1025.

64. Langmead B, Trapnell C, Pop M, Salzberg SL (2009) Ultrafast and memory-efficientalignment of short DNA sequences to the human genome. Genome Biol 10:R25.

65. Zhang Y, et al. (2008) Model-based analysis of ChIP-seq (MACS). Genome Biol 9:R137.66. Robinson JT, et al. (2011) Integrative genomics viewer. Nat Biotechnol 29:24–26.67. Zang C, et al. (2009) A clustering approach for identification of enriched domains

from histone modification ChIP-seq data. Bioinformatics 25:1952–1958.68. Liu Q, et al. (2010) DNA replication factor C1 mediates genomic stability and tran-

scriptional gene silencing in Arabidopsis. Plant Cell 22:2336–2352.69. Cox MP, Peterson DA, Biggs PJ (2010) SolexaQA: At-a-glance quality assessment of

Illumina second-generation sequencing data. BMC Bioinformatics 11:485.70. Zhang H, et al. (2013) DTF1 is a core component of RNA-directed DNA methylation

and may assist in the recruitment of Pol IV. Proc Natl Acad Sci USA 110:8290–8295.71. Purves RW, Ambrose SJ, Clark SM, Stout JM, Page JE (2015) Separation of isomeric

short-chain acyl-CoAs in plant matrices using ultra-performance liquid chromatographycoupled with tandem mass spectrometry. J Chromatogr B Analyt Technol Biomed LifeSci 980:1–7.

Wang et al. PNAS | May 21, 2019 | vol. 116 | no. 21 | 10585

PLANTBIOLO

GY

Dow

nloa

ded

by g

uest

on

Oct

ober

27,

202

0