A protein complex regulates RNA processing of intronicheterochromatin-containing genes in ArabidopsisCheng-Guo Duana,b,c,1,2, Xingang Wangc,1, Lingrui Zhanga,b,c, Xiansong Xionga,b, Zhengjing Zhanga,b, Kai Tangc, Li Pand,Chuan-Chih Hsud, Huawei Xua,b,e, W. Andy Taod, Heng Zhanga,b, and Jian-Kang Zhua,b,c,2

aShanghai Center for Plant Stress Biology, Chinese Academy of Sciences, Shanghai 200032, China; bCenter of Excellence in Molecular Plant Sciences, ChineseAcademy of Sciences, Shanghai 200032, China; cDepartment of Horticulture and Landscape Architecture, Purdue University, West Lafayette, IN 47907;dDepartment of Biochemistry, Purdue University, West Lafayette, IN 47907; and eCollege of Agriculture, Henan University of Science and Technology,Luoyang 471023, China

Contributed by Jian-Kang Zhu, July 17, 2017 (sent for review June 14, 2017; reviewed by Elizabeth S. Dennis and Qingshun Quinn Li)

In several eukaryotic organisms, heterochromatin (HC) in the intronsof genes can regulate RNA processing, including polyadenylation,but the mechanism underlying this regulation is poorly understood.By promoting distal polyadenylation, the bromo-adjacent homology(BAH) domain-containing and RNA recognition motif-containingprotein ASI1 and the H3K9me2-binding protein EDM2 are requiredfor the expression of functional full-length transcripts of intronic HC-containing genes in Arabidopsis. Here we report that ASI1 andEDM2 form a protein complex in vivo via a bridge protein, ASI1-Immunoprecipitated Protein 1 (AIPP1), which is another RNA recogni-tion motif-containing protein. The complex also may contain the Pol IICTD phosphatase CPL2, the plant homeodomain-containing proteinAIPP2, and another BAH domain protein, AIPP3. As is the case withdysfunction of ASI1 and EDM2, dysfunction of AIPP1 impedes the useof distal polyadenylation sites at tested intronic HC-containing genes,such as the histone demethylase gene IBM1, resulting in a lack offunctional full-length transcripts. A mutation in AIPP1 causes silencingof the 35S-SUC2 transgene and genome-wide CHG hypermethylationat gene body regions, consistent with the lack of full-length func-tional IBM1 transcripts in the mutant. Interestingly, compared withasi1, edm2, and aipp1 mutations, mutations in CPL2, AIPP2, andAIPP3 cause the opposite effects on the expression of intronic HC-containing genes and other genes, suggesting that CPL2, AIPP2,and AIPP3 may form a distinct subcomplex. These results advanceour understanding of the interplay between heterochromatic epi-genetic modifications and RNA processing in higher eukaryotes.

RNA processing | heterochromatin | DNA methylation | polyadenylation |transposable element

Asubstantial portion of eukaryotic genomes consists of trans-posable elements (TEs) and other repetitive elements (TREs).

Because of their potential deleterious effects on genome in-tegrity, TREs are generally repressed by epigenetic silencingmechanisms (1, 2). Therefore, epigenetic silencing marks, suchas DNA methylation and repressive histone modifications likeH3K9me2 are enriched, and heterochromatin (HC) is formedin these regions. Heterochromatic TREs (HC-TREs) can affectthe expression of their associated genes in various ways. In bothplant and animal genomes, HC exists not only in promoter re-gions, but also within transcribed regions, especially within introns,and most intronic HC is a result of TRE insertion (3–6). In thehuman genome, 60% of TEs are located within introns that com-pose only 24% of the genome. Approximately 10% of the genes inthe maize genome have intronic TEs >1 kb in length, and ∼0.7% ofthe annotated genes contain intronic TEs in the genome of themodel plant Arabidopsis (6). Thus, intronic TEs are widespread inhuman and plant genomes. Although the repressive effect of pro-moter HC-TREs on the expression of downstream genes has beenwell documented (7, 8), the effect of intronic HC-TREs on theexpression of their associated genes remains largely unknown (9).The polyadenylation of mRNAs, which involves 3′-end cleav-

age of pre-mRNAs and addition of poly (A) tails, is an important

step in gene expression in eukaryotes that is achieved through afinely tuned biochemical process. This process involves numerousprotein factors, including the CPSF (cleavage and polyadenylationspecificity factor) and CstF (cleavage stimulation factor) com-plexes, as well as a number of cis elements, such as poly(A) signalelements (10). Previous studies have shown that >50% of humangenes, >70% of Arabidopsis genes, and 50–82% of rice genes havemultiple polyadenylation sites that generate various mRNA iso-forms with different 3′ ends (10–12). This regulatory mechanismof gene expression is termed alternative polyadenylation.The RNA Pol II carboxyl-terminal domain (CTD) is critical

for coupling transcription with RNA processing (13). Studies inyeasts and animals have found that the Ssu72 RNA polymeraseII Ser-5 CTD phosphatase regulates alternative polyadenylation(14, 15). Ssu72 binds CTD heptapeptides and forms a complexwith Symplekin, a subunit of the CPSF complex (16).Little is known about the interplay between RNA processing

and heterochromatic histone modifications. Recent reports haveindicated that histone modifications and nucleosome occupancymight influence polyadenylation site selection (17, 18). In hu-mans, strong nucleosome depletion around polyadenylation sitesand nucleosome enrichment downstream of polyadenylationsites have been observed (17). Moreover, the heterochromatic

Significance

How heterochromatin affects RNA processing is unclear. Thechromatin regulators ASI1 and EDM2 function in regulating al-ternative polyadenylation at genes with intronic heterochro-matin. We found that ASI1 and EDM2 are associated in plantathrough interactions with a putative RNA-binding protein,AIPP1. Protein interaction assays suggest that the RNA Pol IIC-terminal domain phosphatase CPL2 and two other proteins(AIPP2 and AIPP3) are associated with the ASI1-AIPP1-EDM2complex. Like ASI1 and EDM2, AIPP1 also functions in promotingthe expression of heterochromatin-containing genes. However,the function of CPL2, AIPP2, and AIPP3 is antagonistic to that ofASI1, EDM2, and AIPP1. Our discovery of the ASI1-AIPP1-EDM2complex and associated proteins is important for understandinghow heterochromatin regulates RNA processing.

Author contributions: C.-G.D., X.W., and J.-K.Z. designed research; C.-G.D., X.W., L.Z., X.X.,Z.Z., L.P., and H.X. performed research; X.W., K.T., C.-C.H., W.A.T., H.Z., and J.-K.Z. analyzeddata; and C.-G.D., X.W., L.Z., and J.-K.Z. wrote the paper.

Reviewers: E.S.D., Commonwealth Scientific and Industrial Research Organisation; andQ.Q.L., Western University of Health Sciences.

The authors declare no conflict of interest.

Data deposition: The aipp1-1 and aipp1-2 WGBS and mRNA-seq data reported in thispaper have been deposited in the Gene Expression Omnibus (GEO) database, https://www.ncbi.nlm.nih.gov/geo (accession no. GSE98655).1C.-G.D. and X.W. contributed equally to this work.2To whom correspondence may be addressed. Email: cgduan@sibs.ac.cn or jkzhu@sibs.ac.cn.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1710683114/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1710683114 PNAS | Published online August 14, 2017 | E7377–E7384

PLANTBIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

May

28,

202

1

H3K9me2 and H3K27me1 modifications have been shown to beenriched in low-use polyadenylation sites, whereas the H3K27me2/3marks are enriched in high-use polyadenylation sites (18). Themechanism underlying this polyadenylation site selection remainsunclear, however.Our previous study indicated that the Arabidopsis chromatin

regulator and RNA-binding protein ASI1 (anti-silencing 1)/IBM2/SG1 functions in controlling polyadenylation site selectionfor genes with intronic HC-TREs (19–21). ASI1 is a BAH do-main- and RNA recognition motif (RRM)-containing proteinthat represses the use of intronic proximal polyadenylation sitenear intronic TREs, thereby promoting the generation of full-length transcripts (19–21). Another chromatin regulator, EDM2(enhanced downy mildew 2), functions in a similar manner asASI1 in regulating alternative polyadenylation for intronic HC-TRE–containing genes (22, 23). EDM2 has three PHD (planthomeo domain) domains that bind H3K9me2 marks, but notH3K27me3 marks (22, 23). ASI1 and EDM2, as well as theheterochromatic markers H3K9me2 and H3K27me1, but notH3K27me3, are enriched in the intronic heterochromatic regionsof the affected genes (19, 20, 22, 23). Although we identifiedASI1 and EDM2 from the same genetic screen, whether thesetwo proteins may function in the same protein complex in al-ternative polyadenylation regulation for genes with intronic HCis not known. In oil palm, the DNA methylation status of anintronic LINE-related transposon Karma also affects the RNAprocessing of the homeotic gene DEFICIENS (24), althoughwhether homologs of ASI1 and EDM2 may be involved in thisregulation is unclear. Loss of DNA methylation in somaclonalvariants at the Karma transposon causes reduced accumulationof functional full-length DEFICIENS transcripts and increasedaccumulation of short transcripts, resulting in the unproductivemantled fruits (24).In this study, we found that ASI1 is associated with EDM2 in

vivo even though they do not directly interact with each other. Theassociation between ASI1 and EDM2 is bridged by another RRMprotein, ASI1 immunoprecipitated protein 1 (AIPP1). AIPP1dysfunction causes increased use of proximal polyadenylation sitesin tested ASI1 and EDM2 target genes. Like asi1 and edm2 mu-tations, AIPP1 mutations lead to defects in the expression of theintronic HC-TRE–containing gene IBM1, which encodes anH3K9me2 demethylase. These defects result in transgene silencingand genome-wide CHG hypermethylation at gene body regions.Our protein interaction results suggest that the ASI1-AIPP1-EDM2 complex is associated with three other proteins—AIPP2,AIPP3 and AIPP4/CPL2—and that these three associated proteinsfunction antagonistically with ASI1, AIPP1, and EDM2 in regu-lating intronic HC-TRE–containing genes. Given that CPL2 is aplant RNA Pol II Ser-5 CTD phosphatase homologous to theyeast Ssu72 (25), and that Pol II Ser-5 CTD phosphatase is knownto associate with CPSF (16) and regulate alternative polyadenylationin yeasts and animals (14, 15), our results suggest that the ASI1-AIPP1-EDM2 complex may regulate alternative polyadenylation bymodulating the function of CPL2. Moreover, we have identified anonintronic HC-TRE gene that is also regulated by the ASI1-AIPP1-EDM2 complex, suggesting that the complex may function beyondthe regulation of alternative polyadenylation. Our findings uncovera protein complex critical for the regulation of genes with HC.

ResultsAIPP1 Interacts with ASI1 and EDM2 to Form a Protein Complex inVivo. Previous findings have shown that both ASI1 andEDM2 are required to facilitate the distal polyadenylation ofIBM1 and other genes with intronic HC-TREs (19, 20, 22, 23)and to prevent transcriptional silencing at the 35S-SUC2 re-porter gene (19, 22). The shared functions of ASI1 andEDM2 prompted us to explore whether the two proteins mayphysically interact and thus function together in plants. In a yeast

two-hybrid (Y2H) assay, direct interaction of ASI1 with EDM2was not observed (Fig. 1A), consistent with the results of a splitluciferase assay (Fig. S1A). However, immunoprecipitation cou-pled with mass spectrometry (IP-MS) revealed that EDM2 couldbe copurified with 3×Flag-tagged ASI1 in a previously createdasi1 mutant complementation line (Fig. 1B and Fig. S2) (19).Based on these data, we suspected that one or more unknowncomponents may mediate the physical interaction betweenASI1 and EDM2 in vivo.Further examination of the IP-MS data revealed other pro-

teins copurified with 3×Flag-ASI1 besides EDM2, includingAIPP1, AIPP2, AIPP3, and AIPP4/CPL2 (Fig. 1B and Fig. S2).Like ASI1 (19), AIPP1 contains an RRM domain (Fig. S3A),although the RRM domains of ASI1 and AIPP1 share only a verylimited sequence homology (Fig. S3B). To test whether the twoRRM proteins may interact, we performed Y2H assays and splitluciferase assays in tobacco leaves. Both assays showed thatASI1 directly interacts with AIPP1, and that AIPP1 also directlyinteracts with EDM2 (Fig. 1A and Fig. S1A). These results sug-gest that AIPP1 may bridge the interaction between ASI1and EDM2.To test this hypothesis, we conducted a yeast three-hybrid

(Y3H) assay. EDM2 and AIPP1 were coexpressed in the sameprey vector under the control of ADH1 and MET promoters,respectively (Fig. 2A). This coexpression allowed yeast cells togrow in a high-stringency selection medium when the bait BD-ASI1 was present (Fig. 2B). However, the yeast cells failed togrow when AIPP1 was replaced with an irrelevant protein, GFP,or when AIPP1 was simply removed (Fig. 2 A and B). Similarly,

Fig. 1. AIPP1 interacts with ASI1 and EDM2. (A) Y2H assay showing thatASI1 does not directly interact with EDM2, but both ASI1 and EDM2 directlyinteract with AIPP1. (B) IP-MS results showing that EDM2, AIPP1, AIPP2,AIPP3, and CPL2 copurified with 3×Flag-ASI1. Results from one of five in-dependent IP-MS experiments are shown. The “score” was calculated as thesum of the negative algorithms of the posterior error probability values ofthe connected peptide-spectrum matches. “Coverage” indicates the per-centage of amino acid residues covered by the identified peptides. “Pep-tide” refers to the total number of identified peptide sequences matchingthe protein.

E7378 | www.pnas.org/cgi/doi/10.1073/pnas.1710683114 Duan et al.

Dow

nloa

ded

by g

uest

on

May

28,

202

1

the coexpression of ASI1 and AIPP1, but not of ASI1 and GFP,in the bait vector also enabled the survival of yeast cells in theselection medium in the presence of prey AD-EDM2 (Fig. 2B).These results strongly support our hypothesis that AIPP1 canfunction as a bridge protein between ASI1 and EDM2 in vivo.To test whether these three proteins may exist in the same

protein complex in planta, we performed gel filtration assays withepitope-tagged ASI1, EDM2 (19, 22), and AIPP1 transgeniclines. For 4×Myc-tagged AIPP1 transgenic plants, AIPP1 geno-mic DNA, including a 2-kb promoter sequence, was transformedinto Col-0 plants. The gel filtration results showed that ASI1,EDM2, and AIPP1 can be eluted in the same fractions, in-dicating that they exist in a complex with an estimated size of440–670 kDa (Fig. 2C). Taken together, our results show thatASI1 and EDM2 form a protein complex in vivo via a bridgeprovided by AIPP1.

AIPP1 Is Required for the Accumulation of Full-Length Transcripts ofIntronic HC-Containing Genes. ASI1 and EDM2 function in con-trolling the polyadenylation site selection of intronic HC-containing genes by facilitating distal polyadenylation (19, 22).Therefore, we next determined whether AIPP1 also controls theproper expression of intronic HC-containing genes targeted byASI1 and EDM2. Because we failed to identify a T-DNA insertionmutant of the AIPP1 gene, we used CRISPR/Cas9-mediated ge-nome editing to generate AIPP1 mutants. We designed twoCas9 target sites in the second exon of the AIPP1 gene (Fig. S4A).By DNA sequencing, we identified two homozygous mutants fromthe CRISPR/Cas9-transformed plants. The first mutant has asingle C insertion between the nucleotides 179 and 180 down-stream of the ATG translation initiation, and the second mutanthas a 17-nt deletion from nucleotides 269 to 285 downstream ofthe start codon (Fig. S4A). Both mutations would lead to a frameshift in the AIPP1 amino acid sequence. Hereinafter, we refer tothese two mutants as aipp1-1 and aipp1-2, respectively (Fig. S4A).

We used reverse-transcription quantitative PCR (RT-qPCR) toquantify the accumulation of full-length and short transcripts atthree loci targeted by ASI1 and EDM2, including the histoneH3K9me2 demethylase gene IBM1. We used 5′-terminal–specificand 3′-terminal–specific primer pairs to quantify total and full-length transcripts in aipp1-1 and aipp1-2 mutants. asi1-2/ibm2-3(26) and edm2-4 (22) mutants, which are in the Col-0 background,served as parallel controls. The IBM1 gene produces two tran-script forms, a large, full-length form (IBM1-L) and a short form(IBM1-S) (19). Our RT-qPCR results show that the IBM1-Ltranscript was substantially reduced in aipp1-1 and aipp1-2 mutantplants compared with Col-0 wild-type (WT) plants, similar to thereduction observed in asi1-2 and edm2-4 mutants (Fig. 3A). Theamount of total IBM1 transcripts (IBM1-5′), including IBM1-Land IBM1-S together, was reduced to a lesser extent than theamount of IBM1-L transcript (Fig. 3A). Reduced expression offull-length transcripts was also observed in aipp1 as well as asi1and edm2 mutants for two other genes with intronic HC,AT3G05410 (Fig. 3B) and AT1G58602 (RPP7) (Fig. 3C). Theseresults show that AIPP1 functions with ASI1 and EDM2 in reg-ulating the expression of intronic HC-containing genes.We further investigated the effect of aipp1, asi1, and edm2

mutations on the intronic HC-containing genes by examining thedistribution of mRNA-seq reads. Fig. 3D shows the mRNA-seqreads mapped to the IBM1 locus, which contains a large intronicand methylated repetitive sequence. In the WT, abundant readswere mapped to sequences 3′ to the intronic HC; however, the 3′reads were greatly decreased in all three mutants (Fig. 3D). Re-ductions in 3′ reads also were observed at two other genes,AT3G05410 (Fig. 3E) and RPP7 (Fig. 3F). We quantified the useof individual exons using the DEXSeq method (27). The DEXSeqresults showed greatly decreased expression levels of exons 3′ tothe HC-TREs in all three mutants, with expression levels of exons5′ to the HC-TREs reduced to a lesser extent (Fig. S5 A and B) oreven increased (Fig. S5C) in the three mutants. Previous workhas shown that asi1 and edm2 mutations cause decreased 3′

Fig. 2. ASI1, EDM2, and AIPP1 may form a protein complex in vivo. (A) Diagram of the construction of the entry vectors harboring two expression units.(B) The ASI1–EDM2 interaction mediated by AIPP1 was examined by yeast growth on a stringent medium deficient in Met, Leu, Trp, His, and adenine (SD-M/L/T/H/A). GFP, which was used as an unrelated donor to replace AIPP1, could not mediate the interaction between EDM2 and ASI1. The plates were photo-graphed after incubation at 28 °C for 3 d. (C) Western blot analysis showing that the epitope-tagged ASI1, EDM2, and AIPP1 proteins are present in the sameeluted fractions in gel filtration assays. The arrows indicate the molecular mass corresponding to the indicated fractions.

Duan et al. PNAS | Published online August 14, 2017 | E7379

PLANTBIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

May

28,

202

1

mRNA-seq reads of intronic HC-TRE genes by reducing the useof distal polyadenylation sites (19, 20, 22, 23). The altered patternsof mapped mRNA-seq reads in aipp1 mutants are similar to thoseseen in asi1 and edm2 mutants, suggesting that, like ASI1 andEDM2, AIPP1 also affects the selection of polyadenylation sites.From the mRNA-seq data, we identified intronic TE-containinggenes with reduced levels of RNA-seq reads downstream of theintronic TE relative to reads upstream of the intronic TE in aipp1-1as well as asi1-2 and edm2-4 mutants, as described previously(19). We found 39, 44, and 47 intronic TE-containing genes with

reduced 3′ reads in the aipp1-1, asi1-2, and edm2-4 mutants, re-spectively (Dataset S1). Substantial portions of these genes areshared by the three mutants, as shown in Fig. S6.We generated asi1-2edm2-4, asi1-2aiipp1-1, and edm2-4aipp1-

1 double mutants and compared the accumulation of full-lengthtranscripts between the double mutants and single mutants. OurRT-qPCR results indicate a similar reduction of full-lengthtranscripts in the double mutants and their corresponding singlemutants (Fig. S7). These results support the idea that AIPP1,ASI1, and EDM2 function in the same genetic pathway, consistent

Fig. 3. AIPP1 is required for the expression of full-length transcripts of selected ASI1 and EDM2 target intronic HC-TRE genes. (A–C) Three ASI1 andEDM2 target genes that contain intronic HC, including the H3K9me2 demethylase gene IBM1 (A), AT3G05410 (B), and AT1G58602 (RPP7) (C) were selected forexpression analysis in aipp1 mutants. Col-0 served as the WT control, and asi1-2 and edm2-4 served as positive controls. Two representative forms of pre-mRNA transcripts with different polyadenylation sites are shown in the diagram. Black and green boxes represent exons and intronic HC elements, re-spectively. 3′- and 5′-specific primer pairs were used for detection of long (L) and 5′ (sum of distal and proximal transcripts) transcripts by RT-qPCR. The relativeexpression of different transcripts was first normalized to ACTIN2 and then to Col-0 plants. Error bars indicate SD; n = 3. (D–F) Snapshots of mRNA-seq andDNA methylation profiles from the IGV showing the read coverage (Upper) and DNA methylation levels (Lower) of the IBM1 (D), AT3G05410 (E), and RPP7(F) genes. The genomic structure of each gene is shown at the top. Green boxes represent TREs.

E7380 | www.pnas.org/cgi/doi/10.1073/pnas.1710683114 Duan et al.

Dow

nloa

ded

by g

uest

on

May

28,

202

1

with the three proteins functioning together in a complex. Ourresults suggest that the ASI1-AIPP1-EDM2 complex is requiredfor promoting the accumulation of full-length transcripts of intronicHC-containing genes.

AIPP1 Is an Antisilencing Factor That Affects Gene Body CHGMethylation. Because ASI1 and EDM2 were identified as anti-silencing factors (19, 22), we determined whether AIPP1 may havea role in preventing transgene silencing as well. To explore theeffect of AIPP1 dysfunction on transgene silencing, we introducedthe aipp1-1 mutation into the 35S-SUC2 transgene background bycrossing aipp1-1mutant and 35S-SUC2 plants. Homozygous aipp1-1mutant plants containing the 35S-SUC2 transgene, as well as asi1-1(19) and edm2-6 (22) plants (already in the 35S-SUC2 back-ground), were grown on sucrose- and glucose-containing half-strength Murashige and Skoog medium. All plants grew normallyand had long roots on the glucose-containing medium (Fig. S8A).Like the asi1-1 and edm2-6 plants, the 35S-SUC2/aipp1-1 plantsgrew normally and had long roots in the sucrose-containing me-dium, compared with the inhibited growth and short roots of theWT parental 35S-SUC2 plants (Fig. S8A), suggesting AIPP1 dys-function as the cause of 35S-SUC2 transgene silencing. Consistentwith the root growth phenotype, our RT-qPCR results show thatSUC2, NPTII, and HPTII transgenes were silenced in aipp1-1mutant plants compared with WT 35S-SUC2 plants (Fig. S8B).These results indicate that AIPP1 is an antisilencing factor.IBM1 suppresses gene body CHG methylation by demethy-

lating histone H3K9me2 (28-30). Because AIPP1 is required for

the appropriate expression of functional IBM1-L transcripts (Fig.3 A and D), we next asked whether AIPP1 dysfunction may affectgene body CHG methylation. To explore this question, we sub-jected two aipp1 mutant alleles to whole-genome bisulfite se-quencing (WGBS) with two biological replicates for each allele.Col-0 served as the WT control. Our WGBS analysis identifiedthousands of differentially methylated regions with increasedmethylation (hyper-DMRs) from the four methylomes of theaipp1 mutants. We used boxplot analysis to compare the hyper-DMRs of aipp1 with those of ibm1-4, which were identified frompublished DNA methylome data (19). The results show that atibm1-4 hyper-DMRs, DNA methylation was greater in aipp1mutants compared with Col-0 WT. Similarly, at aipp1 hyper-DMRs, DNA methylation levels were higher in the ibm1 mu-tant than in the Col-0 WT (Fig. 4A). These results suggest thatAIPP1 and IBM1 suppress cytosine methylation of similar ge-nomic regions. We next compared the genomic features ofhyper-DMRs between aipp1 and ibm1-4 mutants, and found that≈80% of aipp1 hyper-DMRs were mapped to gene regions,similar to the percentage in the ibm1-4 mutant (Fig. 4B).To further characterize the DNA methylation pattern in the

aipp1 mutants, we examined the distribution of DNA methyl-ation along the gene bodies, TEs, and their 2-kb upstream anddownstream flanking sequences. The results indicate that CHGmethylation, but not CG or CHH methylation, in the aipp1mutants was significantly increased in gene body regions com-pared with the WT (Fig. 4C and Fig. S9), with greater increasesin CHG methylation observed in longer genes (Fig. S9A). No

Fig. 4. AIPP1 controls gene body CHG methylation. (A) Boxplot diagram showing DNA methylation levels of each genotype at ibm1-4, aipp1-1, and aipp-2hyper-DMRs. r1, replicate 1; r2, replicate 2. (B) Composition of the hyper-DMRs in aipp1-1, aipp1-2, and ibm1-4 mutants. (C) Distribution of DNA methylationalong gene (Upper)/TE (Lower) bodies and their upstream/downstream 2-kb flanking sequences in different cytosine contexts.

Duan et al. PNAS | Published online August 14, 2017 | E7381

PLANTBIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

May

28,

202

1

change was evident in TEs or their flanking regions, however(Fig. 4C and Fig. S9B). This change in DNA methylation patternof the aipp1 mutants is very similar to that seen in the ibm1-4mutant. Considering the down-regulation of functional IBM1-Ltranscript in aipp1 mutant plants, these results suggest thatAIPP1 suppresses gene body CHG methylation by facilitatingexpression of the functional IBM1-L transcript.

Potential Roles of AIPP2, AIPP3, and AIPP4/CPL2. To better un-derstand the functions of the ASI1-AIPP1-EDM2 complex, weinvestigated three other proteins that copurify with ASI1: AIPP2,AIPP3, and APIPP4/CPL2 (Fig. 1B and Fig. S2). AIPP2 isencoded by AT3G02890 and bears a PHD domain, whereasAIPP3 is encoded by AT4G11560 and bears a BAH domain anda TFIIS domain known to be involved in transcriptional regu-lation. AIPP4/CPL2 is an RNA Pol II CTD phosphatase thatdephosphorylates the Ser-5 of RNA Pol II CTD (25). Using Y2Hand split luciferase assays, we found that AIPP2 can interactdirectly with ASI1, AIPP3, and CPL2 (Fig. 5A and Fig. S1B).These results suggest that ASI1, EDM2, AIPP1, AIPP2, AIPP3,and CPL2 may form a large complex in plant cells.To probe the genetic function of AIPP2, AIPP3, and CPL2,

we identified the T-DNA insertion lines salk_057771 andGK_058D11 for AIPP2 and AIPP3 genes, referred to as aipp2-1 andaipp3-1, respectively (Fig. S4B). For the CPL2 gene, a previously

described mutant allele, cpl2-2 (31), was used. RT-qPCR wasperformed to examine the expression of full-length transcripts atthree intronic HC-TRE genes: IBM1, AT3G05410, and RPP7.Surprisingly, in contrast to the reduction in asi1-2, edm2-4, aipp1-1, and aipp1-2 mutants, full-length transcripts for all three geneswere increased in aipp2-1, aipp3-1, and cpl2-2 mutants comparedwith the WT Col-0 (Fig. 5B).

The ASI1-AIPP1-EDM2 Complex Is Required for Expression of aNonintronic HC-TRE Gene. Our examination of mRNA-seq reads atHC-containing genes revealed that the transcript level ofAT4G16870, a TE gene of Copia-like retrotransposon origin, wasdramatically affected by asi1, edm2, and aipp1 mutations (Fig. 6).The AT4G16870 transcript was abolished in asi1-2, edm2-4, aipp1-1, and aipp1-2 mutants, but was significantly increased in aipp2,aipp3, and cpl2mutants (Fig. 6). The AT4G16870 locus was highlymethylated in Col-0 WT, as well as in aipp1 and ibm1 mutants(Fig. 6). Given that the HC at AT4G16870 does not reside in anintron, our results suggest that the ASI1-AIPP1-EDM2 complexhas functions beyond the regulation of intronic HC-TRE genes.

DiscussionTRE insertions within transcribed regions, especially introns, arecommonly observed in eukaryotic cells (3–6). The intronic HC-TREs do not appear to disturb gene transcription, given that

Fig. 5. AIPP2 can directly interact with ASI1, AIPP3, and CPL2 and regulateintronic HC-TRE genes. (A) Y2H assay results showing interaction ofAIPP2 with AIPP3, ASI1, and CPL2. (B) Effect of the different mutations onthe accumulation of full-length transcripts from the indicated intronic HC-TRE genes. 3′-specific primer pairs were used for detection of full-lengthtranscripts (L) of IBM1, AT3G05410, and RPP7 by RT-qPCR. Relative expres-sion of different transcripts was first normalized to ACTIN2 and then to Col-0 plants. Error bars indicate SD; n = 3. **P < 0.01; *P < 0.05, Student’s t test.

Fig. 6. Dysfunction of ASI1 and its associated proteins affects the expressionof a nonintroinc HC gene. RT-qPCR shows that the ASI1-AIPP1-EDM2 sub-complex is required for the expression of AT4G16870, a TE gene with heavyDNA methylation. A snapshot of DNA methylation in this locus is shown.Relative expression in different mutants was first normalized to ACTIN2 andthen to Col-0 plants. Error bars indicate SD; n = 3. **P < 0.01, Student’s t test.

E7382 | www.pnas.org/cgi/doi/10.1073/pnas.1710683114 Duan et al.

Dow

nloa

ded

by g

uest

on

May

28,

202

1

chromatin immunoprecipitation assays have shown no effect onRNA Pol II occupancy (19, 22); however, intronic HC-TREs cancause alternative polyadenylation of the transcripts (19, 20, 22,23). In previous studies, we identified two chromatin regulators,ASI1 and EMD2, that are required to promote distal poly-adenylation so that full-length transcripts can accumulate (19,22). It has been hypothesized that ASI1 and EDM2 may functiontogether and interact through an unknown factor (32). In thisstudy, we found that EDM2 can be copurified with ASI1, al-though the two proteins do not interact directly (Fig. 1A).AIPP1 interacts with both ASI1 and EDM2 and thus is the hy-pothesized unknown factor that serves as a bridge protein tomediate the in vivo association of ASI1 and EDM2 (Fig. 2).Consistent with AIPP1 being in the same protein complex withASI1 and EDM2, AIPP1 also functions to promote the accu-mulation of full-length transcripts of intronic HC-TRE–con-taining genes. In addition, we found that AIPP1, as well asASI1 and EDM2, are required for expression of the hetero-chromatinized TE gene AT4G16870.RRM is an RNA-binding domain common in many proteins

involved in RNA processing (33, 34). Both ASI1 (19) andAIPP1 contain an RRM domain, consistent with their function inregulating RNA processing. Because their RRM domains arequite divergent in sequence (Fig. S3), the two proteins mayrecognize different features of the pre-mRNAs during regulationof alternative polyadenylation of intronic HC-TRE–containinggenes. Along with the RRM domain, ASI1 also contains a BAHdomain that may interact with the chromatin (19–21). EDM2contains three copies of the PHD and has been shown to bind tochromatin marks, including the heterochromatic mark H3K9me2(22, 23). The ASI1-AIPP1-EDM2 complex is capable of inter-acting with both the chromatin and RNA, and thus is well suitedfor its function in connecting the epigenetic regulation of HC-TREs and RNA processing. EDM2 also has a putative N6-adenine methyltransferase domain (22). Methylated adenine isinvolved in 3′-end formation of mRNAs (35). It would be in-teresting to determine whether EDM2 may indeed methylateRNA, and how this RNA methylation may alter the use ofproximal vs. distal polyadenylation sites.Mutations in AIPP1, ASI1, and EDM2 have a significant impact

on the DNA methylome. This can be explained by their role inregulating IBM1 expression. As a histone H3K9me2 demethylase,IBM1 is required for preventing gene body CHG methylation,because the H3K9me2 mark accumulates in the gene body regionsof thousands of genes and the H3K9me2 mark attracts CMT3 tocause CHG methylation (27, 28, 36). IBM1 is an intronic HC-TRE–containing gene, and the accumulation of functional, full-length IBM1 transcript requires the ASI1-AIPP1-EDM2 complex(Fig. 3 A and D) (19, 22). IBM1 is also an antisilencing factorimportant for the prevention of promoter DNA hypermethylationand transcriptional silencing of the 35S-SUC2 transgene (19, 22).Because of their role in ensuring the accumulation of functionalfull-length IBM1 transcript, AIPP1 (Fig. 3 A and D), as well asASI1 and EDM2 (19, 22), are antisilencing factors necessary forexpression of the 35S-SUC2 transgene.The estimated size of the ASI1-AIPP1-EDM2 protein com-

plex is between 440 and 670 kDa. It is likely that this largecomplex contains other components as well. Our results suggestthat CPL2, AIPP2, and AIPP3 are associated with this complex.Consistent with this notion, dysfunction in CPL2, AIPP2, andAIPP3 also affects the accumulation of full-length transcripts ofintronic HC-TRE genes and modulates the expression of the TE

gene AT4G16870; however, the role of these proteins appears tobe opposite to that of ASI1, EDM2, and AIPP1. It is possiblethat AIPP2, AIPP3, and CPL2 may be in a subcomplex thatdiffers from the ASI1-AIPP1-EDM2 subcomplex. In the putativeAIPP3-AIPP2-CPL2 subcomplex, AIPP2 with a PHD domainand AIPP3 with a BAH domain may bind the HC, whereasCPL2, like the yeast and animal RNA Pol II Ser-5 CTD phos-phatase SSU72, may be more directly involved in regulating al-ternative polyadenylation (14–16). Further work is needed todetermine whether there are indeed two such subcomplexes and,if so, how they may function antagonistically in the regulation ofintronic and nonintronic HC-TRE genes.

Materials and MethodsPlant Materials and Growth Conditions. All plants were grown under a long-day photoperiod (16-h light/8-h dark). The aipp1-1 and aipp1-2mutants weregenerated using CRISPR/Cas9 (37). For determination of root length phe-notype, seeds were sown on half-strength Murashige and Skoog mediumcontaining 1% sucrose or 1% glucose. Plants were photographed at 10 dafter germination. 4×Myc-tagged AIPP1 transgenic expression driven by itsnative promoter was achieved by introducing the AIPP1 genomic DNA intoCol-0 plants, and T3 generation plants were used for gel filtration analysis.

Protein Interaction Analysis. Y2H and split luciferase assays were performed asdescribed previously (36). For yeast three-hybrid (Y3H) assays, the entryvectors were constructed to harbor a second cassette expressing AIPP1 orGFP. The primers used are listed in Table S1. The corresponding entry vectorswere recombined into pGBKT7-GW and pGADT7-GW vectors using the LRreaction (Invitrogen). Sets of constructs were cotransformed into the Y2HGold yeast strain (Clontech). The AD-T and BD-53 combination served as apositive control, and the AD-T and BD-Lam combination served as a negativecontrol. Yeast transformants were selected on synthetic minimal dropoutmedium deficient in Met, Trp, and Leu (SD-M/L/T). Protein interactions wereassessed on the stringent dropout medium deficient in Met, Leu, Trp, His,and adenine (SD-M/L/T/H/A).

Immunoprecipitation and MS. Immunoprecipitation coupled with LC-MS/MS(IP-MS) assays were performed as described previously (30). The in-florescence tissues from 3×Flag-tagged ASI1 plants (16) were collected fortotal protein extraction. Total proteins were then precipitated by agarose-conjugated anti-Flag antibody (Sigma-Aldrich) for 3 h at 4 °C. After fivewashes with protein extraction buffer and five washes with PBS buffer, theprecipitated proteins were subjected to MS analysis.

WGBS and DNA Methylation Analysis. For whole-genome bisulfite sequencinganalysis, genomic DNA was extracted from 2-wk-old seedlings using theDNeasy PlantMini Kit (Qiagen). Differentially methylated regions (DMRs) andDNA methylation patterns in different genotypes were identified as de-scribed previously (38).

mRNA-seq Analysis. FormRNA-seq assays, total RNAswere extracted from 2-wk-old seedlings and subjected to RNA deep sequencing. mRNA-seq libraries wereprepared using the TruSeq Stranded mRNA LT Kit (Illumina) following themanufacturer’s protocol. Paired-end sequencing of the library was performedwith an Illumina HiSeq 2500 system using the Illumina v4 reagents at theGenomics Core Facility of the Shanghai Center for Plant Stress Biology. Cleanreads were mapped to the Arabidopsis reference genome using TopHat. Themapped reads were visualized with the genome browser Integrative GenomicsViewer (IGV) in tdf file format. The tdf files were generated with igvtools usingmapping results obtained from TopHat. The list of genes with reduced readsdownstream of their intronic TEs in the mutants was generated according to apreviously described method (19) with a threshold of a 1.5-fold reduction.

ACKNOWLEDGMENTS. We thank Wei Jia and Aojun Chen for help withmRNA-seq library construction and sequencing. This work was supported bythe Chinese Academy of Sciences and National Institutes of Health Grant R01GM070795 (to J.-K. Z.).

1. Rebollo R, Romanish MT, Mager DL (2012) Transposable elements: An abundant and

natural source of regulatory sequences for host genes. Annu Rev Genet 46:21–42.2. Bennetzen JL, Wang H (2014) The contributions of transposable elements to

the structure, function, and evolution of plant genomes. Annu Rev Plant Biol 65:

505–530.

3. van de Lagemaat LN, Medstrand P, Mager DL (2006) Multiple effects govern en-

dogenous retrovirus survival patterns in human gene introns. Genome Biol 7:R86.4. Sela N, et al. (2007) Comparative analysis of transposed element insertion within

human and mouse genomes reveals Alu’s unique role in shaping the human tran-

scriptome. Genome Biol 8:R127.

Duan et al. PNAS | Published online August 14, 2017 | E7383

PLANTBIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

May

28,

202

1

5. Zemach A, McDaniel IE, Silva P, Zilberman D (2010) Genome-wide evolutionaryanalysis of eukaryotic DNA methylation. Science 328:916–919.

6. West PT, et al. (2014) Genomic distribution of H3K9me2 and DNA methylation in amaize genome. PLoS One 9:e105267.

7. Hollister JD, Gaut BS (2009) Epigenetic silencing of transposable elements: A trade-offbetween reduced transposition and deleterious effects on neighboring gene ex-pression. Genome Res 19:1419–1428.

8. Wang X, Weigel D, Smith LM (2013) Transposon variants and their effects on geneexpression in Arabidopsis. PLoS Genet 9:e1003255.

9. To TK, Saze H, Kakutani T (2015) DNA methylation within transcribed regions. PlantPhysiol 168:1219–1225.

10. Tian B, Manley JL (2017) Alternative polyadenylation of mRNA precursors. Nat RevMol Cell Biol 18:18–30.

11. Wu X, et al. (2011) Genome-wide landscape of polyadenylation in Arabidopsis pro-vides evidence for extensive alternative polyadenylation. Proc Natl Acad Sci USA 108:12533–12538.

12. Fu H, et al. (2016) Genome-wide dynamics of alternative polyadenylation in rice.Genome Res 26:1753–1760.

13. Meinhart A, Cramer P (2004) Recognition of RNA polymerase II carboxy-terminaldomain by 3′-RNA-processing factors. Nature 430:223–226.

14. Lamas-Maceiras M, Singh BN, Hampsey M, Freire-Picos MA (2016) Promoter-terminator gene loops affect alternative 3′-end processing in yeast. J Biol Chem291:8960–8968.

15. Chen F, et al. (2015) Context-dependent modulation of Pol II CTD phosphatase SSUP-72 regulates alternative polyadenylation in neuronal development. Genes Dev 29:2377–2390.

16. Xiang K, et al. (2010) Crystal structure of the human symplekin-Ssu72-CTD phos-phopeptide complex. Nature 467:729–733.

17. Spies N, Nielsen CB, Padgett RA, Burge CB (2009) Biased chromatin signatures aroundpolyadenylation sites and exons. Mol Cell 36:245–254.

18. Khaladkar M, Smyda M, Hannenhalli S (2011) Epigenomic and RNA structural corre-lates of polyadenylation. RNA Biol 8:529–537.

19. Wang X, et al. (2013) RNA-binding protein regulates plant DNA methylation bycontrolling mRNA processing at the intronic heterochromatin-containing gene IBM1.Proc Natl Acad Sci USA 110:15467–15472.

20. Saze H, et al. (2013) Mechanism for full-length RNA processing of Arabidopsis genescontaining intragenic heterochromatin. Nat Commun 4:2301.

21. Coustham V, et al. (2014) SHOOT GROWTH1 maintains Arabidopsis epigenomes byregulating IBM1. PLoS One 9:e84687.

22. Lei M, et al. (2014) Arabidopsis EDM2 promotes IBM1 distal polyadenylation andregulates genome DNA methylation patterns. Proc Natl Acad Sci USA 111:527–532.

23. Tsuchiya T, Eulgem T (2013) An alternative polyadenylation mechanism coopted to

the Arabidopsis RPP7 gene through intronic retrotransposon domestication. Proc Natl

Acad Sci USA 110:E3535–E3543.24. Ong-Abdullah M, et al. (2015) Loss of Karma transposon methylation underlies the

mantled somaclonal variant of oil palm. Nature 525:533–537.25. Koiwa H, et al. (2004) Arabidopsis C-terminal domain phosphatase-like 1 and 2 are

essential Ser-5-specific C-terminal domain phosphatases. Proc Natl Acad Sci USA 101:

14539–14544.26. Li Y, Rosso MG, Viehoever P, Weisshaar B (2007) GABI-Kat SimpleSearch: An Arabi-

dopsis thaliana T-DNA mutant database with detailed information for confirmed

insertions. Nucleic Acids Res 35:D874–D878.27. Anders S, Reyes A, Huber W (2012) Detecting differential usage of exons from RNA-

seq data. Genome Res 22:2008–2017.28. Saze H, Shiraishi A, Miura A, Kakutani T (2008) Control of genic DNA methylation by a

jmjC domain-containing protein in Arabidopsis thaliana. Science 319:462–465.29. Miura A, et al. (2009) An Arabidopsis jmjC domain protein protects transcribed genes

from DNA methylation at CHG sites. EMBO J 28:1078–1086.30. Rigal M, Kevei Z, Pélissier T, Mathieu O (2012) DNA methylation in an intron of the

IBM1 histone demethylase gene stabilizes chromatin modification patterns. EMBO J

31:2981–2993.31. Ueda A, et al. (2008) The Arabidopsis thaliana carboxyl-terminal domain phosphatase-

like 2 regulates plant growth, stress, and auxin responses. Plant Mol Biol 67:683–697.32. Ma L, Guo C, Li QQ (2014) Role of alternative polyadenylation in epigenetic silencing

and antisilencing. Proc Natl Acad Sci USA 111:9–10.33. Lorkovi�c ZJ, Barta A (2002) Genome analysis: RNA recognition motif (RRM) and K

homology (KH) domain RNA-binding proteins from the flowering plant Arabidopsis

thaliana. Nucleic Acids Res 30:623–635.34. Maris C, Dominguez C, Allain FH (2005) The RNA recognition motif, a plastic RNA-

binding platform to regulate post-transcriptional gene expression. FEBS J 272:

2118–2131.35. Bodi Z, et al. (2012) Adenosine methylation in Arabidopsis mRNA is associated with

the 3′ end and reduced levels cause developmental defects. Front Plant Sci 3:48.36. Du J, et al. (2012) Dual binding of chromomethylase domains to H3K9me2-containing

nucleosomes directs DNA methylation in plants. Cell 151:167–180.37. Feng Z, et al. (2014) Multigeneration analysis reveals the inheritance, specificity, and

patterns of CRISPR/Cas-induced gene modifications in Arabidopsis. Proc Natl Acad Sci

USA 111:4632–4637.38. Duan CG, et al. (2017) A pair of transposon-derived proteins function in a histone

acetyltransferase complex for active DNA demethylation. Cell Res 27:226–240.

E7384 | www.pnas.org/cgi/doi/10.1073/pnas.1710683114 Duan et al.

Dow

nloa

ded

by g

uest

on

May

28,

202

1