HeLa (ATCC, CCL-2) and MEF (ATCC, SCRC-1045) cells were purchased from ATCC. Cells were cultured with modified Eagle’s medium (Thermo Fisher Scientific, 12100046) containing 10% fetal bovine serum (FBS) (Biowest, S1810) in 5% CO₂ incubator. SiRNAs were transfected using Lipofectamine 3000 as described in the manufacturer’s protocol. The sequences of siRNAs are shown in Table S1 .

The University of California Santa Cruz UCSC Xena database (https://xena.ucsc.edu) is an online tool for visualization of gene expression profiles and clinical data (21). The gene expression profiles of cervical cancer (CESC), acute myeloid leukemia (LAML), bladder cancer (BLCA), breast cancer (BRCA), head and neck cancer (HNSC), liver cancer (LIHC), lung adenocarcinoma (LUAD), pancreatic cancer (PAAD), prostate cancer (PRAD), and stomach cancer (STAD) were collected in the data hubs of UCSC Xena. The set of Illumina HiSeq (version 2017-10-13) in each type of tumor was downloaded. The gene expression profiles of mice, chromatin immunoprecipitation (ChIP)-seq of mice, RNA immunoprecipitation (RIP)–ChIP, fRIP-seq, and par-CLIP-seq data were downloaded from the Gene Expression Omnibus (GEO) datasets (https://www.ncbi.nlm.nih.gov/geo/). The ChIP-seq data of seven human cell lines were downloaded from the Encyclopedia of DNA Elements (ENCODE) database (https://www.encodeproject.org/). The detailed information of every datasets is shown in Table S2 .

First, cells from 10-cm culture dish were washed with 5 ml of cold phosphate-buffered saline (PBS) twice, and then 1 ml of polysome lysis buffer (10 mM of KCl, 5 mM of MgCl₂, 10 mM of HEPES, pH 7.0, 0.5% NP-40, 1 mM of DTT, 100 U/ml of RRI, 20 μl/ml of PIC, and 2 mM of vanadyl ribonucleotide complex solution) was added. Subsequently, cells were scraped and transferred to a fresh microtube on ice and centrifuged at 14,000 ×g for 15 min at 4°C. The supernatant was collected; 2 μg of primary antibody (EZH2, 21800-1-AP; H3K27me3, ab192985) was added and incubated on an end-to-end rotator overnight at 4°C; 2 μg of IgG (abcam, ab171870) was used as negative control. Dynabeads protein G measuring 40 μl (Thermo Fisher Scientific, 88848) was added and incubated for another 4 h. The complex was washed four times with 500 μl of polysome lysis buffer. Subsequently, the complex was re-suspended using 100 μl of polysome lysis buffer containing 0.1% sodium dodecyl sulfate (SDS) and 30 μg of Proteinase K and incubated at 50°C for 30 min. The supernatant was collected, 100 μl of phenol–chloroform–isoamyl alcohol was added, and then the water phase was collected. Subsequently, 5 μl of linear polyacrylamide, 10 μl of 3 M sodium acetate, and 250 μl of ethanol were added and then precipitated at −20°C overnight. Finally, the RNA was collected and analyzed using RT-PCR.

First, we identified PRC2-bound transcripts through analyzing the RIP–ChIP or par-CLIP-seq data of EZH2 and SUZ12 proteins (GSE16226 and GSE49435). The genes with H3K27me3, EZH2, and SUZ12 occupation on their promoter region (−5 kb, +1 kb) were considered to be regulated by PRC2 complex. Therefore, we secondly analyzed the PRC2-regulated genes through analyzing the ChIP-seq data of H3K27me3, SUZ12, and EZH2 in seven human cell lines of ENCODE database or in mouse cells of GEO. Finally, the Bayesian network was performed on the expression profiles of PRC2-bound transcripts and PRC2-regulated genes mentioned above. Briefly, the RIP–ChIP and ChIP-seq data were download and cleaned using trimmomatic [trimmomatic PE -threads -phred33 -basein <input> -baseout <paired output 1> <unpaired output 1> <paired output 2> <unpaired output 2>] (22). Then the RIP-seq data were aligned using hisat2 [hisat2 –phred33 -x <hisat2-idx> -1 <m1> -2 <m2> -S <output. sam>] and mapped using StringTie [stringtie <aligned_reads.bam> -p 25 –G <indexes> -B –o <output. gtf>] (23, 24). Subsequently, the ChIP-seq data were aligned using bowtie2 [bowtie2-build <reference_in> <bt2_base>] and mapped using macs2 [macs2 callpeak -t <text.bam> -c <input.bam> -g hs -B -f BAM -n test -q 0.01] (25, 26). Then mRNA expression profiles of CESC, LAML, BLCA, BRCA, HNSC, LIHC, LUAD, PAAD, PRAD, and STAD tissues in The Cancer Genome Atlas (TCGA) database were normalized together using z-score. Furthermore, the Bayesian gene regulatory network analysis was performed using pcalg package (version 2.6-7) of R software (27). The regulatory between control genes and PRC2-regulated genes was also studied. Two unbiased ways were used to select control genes. The ways to select control genes based on their expression are as follows: 1) all genes were ranked according to their average mRNA levels in all samples of the above 10 cancers. 2) Forty-five genes with their mRNA expression nearest to the 45 PRC2-bound transcripts were defined as control genes. Another method was selecting genes randomly: 1) all genes were ranked according to their name in the Excel; 2) using the rand function to obtain a set of random numbers and corresponding them to the gene name; 3) all genes were ranked according to the random numbers, and the top 45 genes were considered as control genes. Please see the Supplementary Methods for the R scripts of Bayesian gene regulatory network.

Library of Integrated Network-Based Cellular Signatures (LINCS) project studies how cells respond to various genetic and environmental stressors and helps us understand the cell pathways (28, 29). An online tool “query” of LINCS (https://clue.io/query) was performed to evaluate the gene signature patterns. The 48 PRC2-regulated genes were defined as PRC2-regulated signature and put into the upregulated gene sets of “query” tool. The similarity score would be calculated by “query” tool online. Finally, the genes were ranked according to the similarity score. The control genes were selected based on their mRNA expression nearest to PRC2-bound RNAs, as described in the Bayesian gene regulatory network analysis.

The gene set enrichment analysis (GSEA) was performed using GSEA v4.1.0 software (30). The patients were divided into two groups according the median of mRNA expression of each gene. Gene sets were analyzed using “c5.all.v7.4.symbols.gmt” downloaded from MSigDB (http://www.gsea-msigdb.org/gsea/msigdb/index.jsp). One thousand permutations of each gene set were used. The pathways with p < 0.05 and false discovery rate (FDR) <0.25 were considered markedly enriched. All the histone modification-related pathways were displayed using heatmap.

The gene ontology (GO) analysis was performed using the clusterProfiler package (version 3.18.1) of the R software (version 3.6.3) (31). The pvalueCutoff was set as 0.05, and the items with p < 0.05 were considered to have significant enrichment. The sub-ontologies of biological processes (BP) and molecular function (MF) were studied.

Total RNA was isolated using RNA iso Plus (Takara, D9108B). Subsequently, reverse transcription was performed using ReverTra Ace qPCR RT Kits (TOYOBO, FSQ-101), and qRT-PCR was carried out using a SYBR^® Green Real time PCR Master Mix (TOYOBO, QPK-201). All samples were repeated in triplicate. Data were normalized to ACTB, and 2^−ΔΔCt was used to calculate the relative fold change. The primers of this study are shown in Table S3 .

Cultured cells in the six-well were extracted on ice using 200 μl of cell extraction buffer (50 mM of Tris-HCl, pH 8.0, 4 M of urea, and 1% Triton X-100) containing protease inhibitor mixture (Roche Diagnostics, 04693132001). Cell lysates were analyzed by Western blotting using EZH2 (CST, 5246, 1:1000), TMEM117 (ProteinTech, 21314-1-AP, 1:500), H3K27me3 (H3K27me3, ab192985, 1:1,000), and GAPDH (ProteinTech, 60004-1-Ig, 1:5,000) antibodies.

The weighted gene co-expression network analysis (WGCNA) of the gene expression profiles of CESC was performed using the WGCNA package (version 1.67) of the R software (version 3.6.3) (32). The appropriate power value in the co-expression module construction was 0.8. Functional enrichment analysis of every co-expression module was performed by GO analysis using clusterProfiler package (version 3.15.1) (31). Only 40 PRC2-bound transcripts with their expression information were shown in CESC. The module–gene relationships were estimated using the correlation between modules and the expression of 40 PRC2-bound transcripts, or control genes. The control genes were selected randomly or based on their expression, as described in the Bayesian gene regulatory network analysis. The gene with absolute value of coefficient greater than 0.2 and p < 0.05 was considered to be related to the module. Finally, the proportion of related genes in each group of control genes was analyzed and compared with that in the PRC2-bound transcripts.

The nuclear and cytoplasmic protein extraction kit (Beyotime Biotechnology, P0027) was used to separate the RNA in nucleus and cytoplasm fraction. Briefly, cells from 10-cm culture dish were washed using 5 ml of PBS and then harvested. Subsequently, 800 μl of buffer A containing phenylmethylsulfonyl fluoride (PMSF) was added and vortexed vigorously for 5 s, and then the cells were incubated on ice for 15 min. Next, 40 μl of buffer B was added and vortexed vigorously for 5 s, and then the cells were incubated on ice for 1 min. Cells were centrifuged at 14,000×g for 5 min at 4°C, and then the supernatant was collected. The precipitation was washed three times using 500 μl of PBS each. The RNA in both supernatant and precipitation fraction was isolated using RNA iso Plus and then quantified using qRT-PCR assay.

The fluorescence in situ hybridization (FISH) assay was performed using TMEM117-specific probes (Biosearch Technologies). First, cells in six wells were fixed with 1 ml of 3% paraformaldehyde for 10 min at room temperature and then permeated for 7 min on ice using 1 ml of PBS containing 0.5% v/v Triton X-100. Cells were washed twice using 500 μl of 70% ethanol for 5 min each and then successively dehydrated in 500 μl of 80%, 95%, and 100% ethanol for 3 min each. Subsequently, 1 μl of RNA probe was added to incubate the cells overnight at 37°C. Then the cells were washed three times and each time for 5 min at 42°C using 500 μl 2× SSC. Subsequently, the cells were blocked with 1 ml of 3% bovine serum albumin (BSA) for 15 min and then incubated with EZH2 antibody (CST, 5246, 1:250) for 1 h at room temperature. The samples were washed three times using 1 ml of PBS and incubated with secondary rabbit antibody (CST, 4412, 1:250) for 1 h. Finally, the samples were stained with DAPI and imaged with an Olympus FV1000 confocal microscope.

First, HeLa cells from 10-cm culture dish were washed with 1 ml of cold PBS and then fixed using 5 ml of 4% formaldehyde for 10 min on ice. Cells were harvested on ice and re-suspended with 600 μl of ChIP lysis buffer (50 mM of Tris-HCl, 5 mM of EDTA, 0.1% deoxycholated, 1% Triton X-100, 150 mM of NaCl, and proteinase inhibitor). Subsequently, cells were sonicated (10/20 s of pulse/pause, 18 cycles, and 30% power) and then centrifuged at 14,000 ×g for 15 min at 4°C, and the supernatant was collected. For the ChIP assay, cell lysates, 1 μg of ChIP grade H3K27me3 primary antibodies (abcam, ab192985), and 25 μl of Dynabeads protein G (Thermo Fisher Scientific, 88848) were mixed and incubated for 2 h at 4°C; 1 μg of IgG antibody (abcam, ab171870) was used as negative control. The complex was washed three times with 500 μl of PBS. To release immune complex, 40 μl of 10% Chelex-100 was added, and the suspension was collected. Finally, boiling the samples for 10 min and the DNA fractions were analyzed using qRT-PCR.

The MEME suite (https://meme-suite.org/meme/) is an online motif-based sequence analysis tool, helping biologists discover novel motifs. The fRIP-seq of EZH2 (GSE67963) and par-CLIP-seq of EZH2 (GSE49435) were used to study PRC2 recognition motif of humans and mice, respectively. The association between putative motifs and EZH2 or SUZ12-bound transcripts was analyzed using Poisson distribution (MATLAB), as described by Wang et al. (15).

HeLa cells from 10-cm culture dish were washed with 1 ml of cold PBS and then harvested using 0.5 ml of cell lysis buffer (10 mM of HEPES, pH 7.0, 200 mM of NaCl, 1% Triton X-100, 10 mM of MgCl₂, 1 mM of DTT, and protease inhibitor). Subsequently, cells were sonicated (20/30 s of pulse/pause, 4 cycles, and 30% power). Cells were centrifuged at 14,000 ×g for 15 min at 4°C, and then the supernatant was collected. Yeast RNA (~0.5 μg/mg protein) was added and incubated on a rotation shaker for 20 min at 4°C. Cell lysates were centrifuged at 14,000 ×g for 15 min at 4°C, and then the supernatant was collected. The plasmids of TMEM117 with or without the RNA recognition motif mutation were construction using a pcDNA3.1 plasmid. Subsequently, the plasmids were linearized and transcribed in vitro for 1 h using a transcription kit (Thermo Fisher Scientific, AM1312). After transcription, the samples were boiled for 10 min, and poly-A tail was added to the 3′ end of RNA in vitro for 1 h using a tailing kit (Thermo Fisher Scientific, AM1350). Subsequently, the RNA was purified using phenol/chloroform/isoamyl alcohol (25:24:1), and then the total amount of RNA in each group was detected. Following the pull-down assay, 100 μl of cell lysis, 200 μl of Dynabeads Oligo (dT)25 (Thermo Fisher Scientific, 61005), and 10 μg of RNA in each group were mixed and incubated together for 2 h at 4°C. Subsequently, the samples were washed three times using 500 μl of washing buffer (10 mM of Tris-HCl pH 7.5, 0.15 M of LiCl, 1 mM of EDTA) and then harvested and analyzed using Western blotting. The plasmid containing the antisense of TMEM117 was used as a negative control.

The statistical analysis was performed using IBM SPSS software (version 24.0), and the data plotting was performed with Prism Graph Pad 6.0 software (33). A value of p < 0.05 was considered statistically significant. Unpaired, two-tailed Student’s t-tests were used to compare the differences between two groups. One-way ANOVA was used to compare the differences between multiple groups.

To investigate whether protein-coding RNAs epigenetically regulate gene expression in cancer, we first studied whether protein-coding RNAs were physically associated with PRC2 complex. It was pointed out that only EZH2 and SUZ12 proteins could bind RNA with high affinity in vitro. Therefore, we only studied the RNAs bound to EZH2 and SUZ12 proteins but not bound to other subunits of PRC2 complex. The fragments of RNAs that bound to EZH2 and SUZ12 proteins were analyzed in the fRIP-seq data of EZH2 and SUZ12 (GSE67963) using the annotatePeak function of ChIPseeker package of R language (34). The results revealed that of the total fragments identified as bound to EZH2 protein, 4.49% were located in the introns and 41.59% were located in exons ( Figure S1A ). Of the fragments of ncRNAs identified as bound to EZH2 protein, 11.47% were located in introns and 31.8% were located in exons ( Figure S1B ). However, of the fragments of protein-coding RNAs identified as bound to EZH2 protein, only 3.79% were located in introns and 42.55% were located in exons ( Figure S1C ). Of the total RNAs with their exons bound to EZH2 or SUZ12 protein, more than 90% were protein-coding RNAs ( Figure 1A , Tables S4 , S5 ). The RIP–ChIP data of EZH2 and SUZ12 in HeLa cells (GSE16226) were also analyzed ( Table S6 ). The results showed that 45 protein-coding RNAs were bound to both EZH2 and SUZ12 proteins ( Figure 1B ). Subsequently, the RIP assay was exploited to confirm that some of them indeed interacted with EZH2 protein and histone H3 with tri-methylation of lysine 27. The result showed that the RNA of MTF2, MYO1C, PHIP, PBRM1, and TBX2 indeed bound to EZH2 protein and histone H3 with tri-methylation of lysine 27 ( Figure 1C ). Genes with their promoter region enriched with EZH2, SUZ12, and H3K27me3 were considered to be regulated by PRC2-mediated histone H3K27me3 modification. Therefore, we identified PRC2-regulated genes in the ChIP-seq data of EZH2, SUZ12, and H3K27me3 in seven cell lines of ENCODE database ( Table S7 ). The result showed that the promoters of 48 genes were occupied by EZH2, SUZ12s and H3K27me3 in all seven cell lines and defined as PRC2-regulated genes ( Figure 1D ). Bayesian network can effectively identify the causal relationship between biological factors (35). Therefore, we studied whether these 45 PRC2-bound transcripts regulated the expression of 48 PRC2-regulated genes in cancer using Bayesian gene regulatory networks ( Figures 1E, F , Table S8 ). Forty-five genes with their mRNA levels nearest to PRC2-bound transcripts were selected as control genes, and the same analysis was repeated ( Figure S2A , Table S9 ). In addition, another 45 genes selected randomly by rand function of Excel were also considered as control genes and repeated the same analysis ( Figure S2B , Table S9 ). In comparison with control genes, PRC2-bound transcripts indeed regulated the expression of PRC2-regulated genes ( Figure 1F ). The LINCS project studies how cells respond to various genetic and environmental stressors. To further confirm our results, 48 PRC2-regulated genes were defined as PRC2-regulated signature, and then the similarity scores between PRC2-bound transcripts and PRC2-regulated signature were evaluated using query tool of LINCS database. Forty-five genes with their mRNA expression level nearest to PRC2-bound transcripts were selected as control genes. In the LINCS project, only 11 PRC2-bound transcripts and 10 control genes were studied. The absolute value of similarity scores of PRC2-bound transcripts were higher than that of control genes, indicating that PRC2-bound transcripts were related to PRC2-regulated signature ( Figures 1G, H ). In conclusion, PRC2-bound transcripts were associated with PRC2-regulated genes.

To further study the functions of PRC2-bound RNAs, we performed the GSEA in the cervical cancer patients of TCGA database (CESC). The histone modification-related pathways were displayed using heatmap. The results revealed more than half of PRC2-bound transcripts markedly enriched on the histone modification-related pathways, indicating these transcripts participated in the regulation of histone modification ( Figure 2A , Table S10 ). To further confirm our conclusion, the GO analysis was also performed to analyze the items enrichment of PRC2-bound transcripts. The results revealed that histone modification-related items were enrichment in both BP and MF categories ( Figures 2B, C , Table S11 ). MYO1C, one of the PRC2-bound transcripts, was selected to prove that it regulated the modification level of H3K27me3. We knocked down the expression of MYO1C in HeLa cells using siRNAs and then analyzed the level of H3K27me3 using Western blotting. We found that the level of H3K27me3 markedly increased after 48 h of transfection ( Figures 2D, E , Figure S3 ).

In addition, the WGCNA was also performed in CESC to confirm our idea. First, the gene co-expression modules were calculated ( Figure S4 ). Subsequently, we analyzed the relationship between co-expression modules and PRC2-bound transcripts, or control genes ( Figure 3A ). The genes with absolute value of correlation coefficient greater than 0.2 and p < 0.05 were considered to be related to modules. We next calculated the proportion of related genes in the PRC2-bound transcripts or in the control genes of each module. Among the PRC2-bound transcripts, the two highest modules were red and turquoise modules ( Figure 3B ). GO analysis revealed that genes in red module were associated with RNA processing and those in turquoise module were associated with histone modification and RNA transcription ( Figure S5 , Table S12 ), indicating that PRC2-bound transcripts were related to histone modification and RNA transcription. Next, we focused on the relationship between turquoise module and PRC2-bound transcripts. We randomly selected 10 sets of genes as control genes and analyzed their relationship with turquoise module ( Figure 3C ). The proportion of related genes in PRC2-bound transcripts was the highest ( Figure 3D ), confirming that PRC2-bound transcripts were associated with histone modification.

We analyzed the RNAs bound on both EZH2 and SUZ12 proteins in the fRIP-seq data (GSE67963). The result showed that both SUZ12 and EZH2 protein bound to the exon of TMEM117 (transmembrane protein 117), a typical protein-coding gene ( Figure 4A ). The RIP assay was also exploited to confirm the interaction between RNA of TMEM117 and EZH2 protein in both K562 and HeLa cells. Interestingly, this interaction in HeLa was approximately twice than that of K562 cells ( Figure 4B ). In addition, the Kaplan–Meier analysis was applied to study the relationship between the expression level of TMEM117 and the prognosis of cancer patients. The result revealed that the expression of TMEM117 was associated with the survival of CESC patients rather than LAML patients of TCGA database ( Figures 4C, D , Figure S6 ). Therefore, TMEM117 was selected for further study of the HeLa cells. First, we studied whether TMEM117 regulated the expression of PRC2-regulated genes. We knocked down TMEM117 in HeLa cells using siRNAs ( Figure 4E , Figure S7 ). Subsequently, we detected the expression of TSLP, a gene downstream of TMEM117 calculated by the Bayesian gene regulatory network. The result revealed that TSLP significantly increased after 48 h of siTMEM117s transfection ( Figure 4F ). We also detected whether TMEM117 altered the expression of TSLP in the HeLa cells with EZH2 knockout ( Figure 4G , Figure S8 ). Nevertheless, knockdown of TMEM117 did not affect the expression TSLP in the HeLa cells with EZH2 knockout ( Figure 4H ), suggesting that TMEM117-regulated TSLP expression was EZH2 dependent.

Therefore, we asked whether TMEM117 suppressed TSLP expression through EZH2-mediated H3K27me3 modification. Because the protein of EZH2 located in nucleus and the modification of histone happened in the nucleus, we first analyzed whether the RNA of TMEM117 occurred in the nucleus using cell fractions analysis and FISH assay. The result revealed that part of the RNA of TMEM117 is localized in the nucleus ( Figures 5A, B ). The RIP assay was also performed to demonstrate that the RNA of TMEM117 interacted with histone 3 with tri-methylation of lysine 27 ( Figure 5C ). We next studied whether TMEM117 affected the modification level of H3K27me3. We knocked down TMEM117 in HeLa cells using siRNAs and then detected the modification level of H3K27me3. The result revealed that TMEM117 knockdown did not affect the modification level of H3K27me3 ( Figure 5D ). However, the ChIP assay proved that TMEM117 silencing dramatically reduced the occupation of H3K27me3 on the promoter region of TSLP ( Figure 5E ). In conclusion, TMEM117 regulated TSLP expression through EZH2-mediated H3K27me3 modification.

Previous studies indicated that RNAs with some characteristic sequences help them bind to proteins more easily. Therefore, we analyzed the RNA recognition motif of PRC2 complex in the fRIP-seq data using the MEME algorithm. The result revealed there were two RNA recognition motifs of PRC2 complex ( Figure 6A ). We next quantified the association between PRC2 complex and these two motifs. The result revealed that both of them were associated with EZH2 and SUZ12 proteins ( Figure 6B ). A G-rich motif was identified, consistent with previous reports (15, 16). Subsequently, the YCAGCYTCC motif was pinpointed and focused on. To study the function of YCAGCYTCC motif, we performed the RNA-pulldown assay to study the interaction between EZH2 protein and RNA of TMEM117 with or without the YCAGCYTCC motif mutation. The results revealed that RNA of TMEM117 with YCAGCYTCC motif mutation dramatically decreased its interaction with EZH2 protein ( Figure 6C , Figure S9 ), indicating that this motif contributed to their interaction. We next compared the characteristics between PRC2-bound transcripts and protein-coding RNAs, such as length, the number of exons, and the guanine–cytosine (GC) content of transcripts. The result showed there was no difference among them ( Figures 6D–F ).

PRC2 complex is highly conserved between plants and animals. In humans and mice, PRC2 complex binds to RNA in a similar manner in vitro (17). Therefore, we investigated whether the protein-coding RNAs bound to PRC2 complex and regulated histone methylation in mice using Bayesian gene regulatory network analysis ( Figure 7A ). The PRC2-bound RNAs were identified from par-CLIP-seq data of EZH2 protein (GSE49435). Subsequently, genes with their promoter region enriched with EZH2, SUZ12, and H3K27me3 were considered to be PRC2-regulated genes and identified in the eight datasets of ChIP-seq data ( Figure 7B , Table S13 ). The Bayesian gene regulatory network analysis was performed to study their causal relationship. The results revealed that PRC2-bound RNAs regulated the expression of PRC2-regulated genes ( Figure 7C , Table S14 ). Subsequently, the RNA recognition motifs of EZH2 protein in mice were analyzed in the par-CLIP-seq data (GSE49435) using the MEME algorithm. Three motifs were identified, and one of them was considered as G-rich motif ( Figure 7D ), consistent with a previous report that EZH2 protein preferentially bound to G-rich RNA was conserved (16). Analyzing four repeats of par-CLIP-seq data of EZH2 (GSE49435) revealed that EZH2 protein directly bound to the RNA of ZDHHC20 ( Figure 7E ). We also performed the RIP assay to confirm that the RNA of ZDHHC20 indeed bound to EZH2 protein and H3 with tri-methylation on lysine 27 in MEF cells ( Figure 7F ). Therefore, ZDHHC20 was selected for further study. We asked whether ZDHHC20 regulated the expression of PRC2-regulated genes. Therefore, we knocked down the expression of ZDHHC20 in MEF cells using siZDHHC20 and then detected the expression of PCNX, a gene downstream of ZDHHC20 calculated by Bayesian gene regulatory network. The result revealed that knockdown of ZDHHC20 in MEF cells significantly decreased the expression of PCNX ( Figure 7G ). We next analyzed the RNA location of ZDHHC20 using the cell fraction analysis. The result revealed that part of the RNA of ZDHHC20 localized in the nucleus, indicating that it had the opportunity to bind to EZH2 protein and mediate H3K27me3 modification ( Figure 7H ). To further prove this, we knocked down ZDHHC20 in MEF cells using siRNA and studied whether it altered the occupation of H3K27me3 modification using ChIP assay. The result revealed that ZDHHC20 knockdown significantly increased the occupation of H3K27me3 on the promoter of PCNX ( Figure 7I ), indicating that ZDHHC20 might regulate PCNX expression through altering the level of H3K27me3 modification.