Gene silencing in cancer by histone H3 lysine 27 ... - Nature

© 2008 Nature Publishing Group http://www.nature.com/naturegenetics

ARTICLES

Gene silencing in cancer by histone H3 lysine 27 trimethylation independent of promoter DNA methylation Yutaka Kondo1,2,9, Lanlan Shen1,9, Alfred S Cheng3,8, Saira Ahmed1, Yanis Boumber1, Chantale Charo1, Tadanori Yamochi4, Takeshi Urano5, Koichi Furukawa5, Bernard Kwabi-Addo6, David L Gold7, Yoshitaka Sekido2, Tim Hui-Ming Huang3 & Jean-Pierre J Issa1 Epigenetic silencing in cancer cells is mediated by at least two distinct histone modifications, polycomb-based histone H3 lysine 27 trimethylation (H3K27triM) and H3K9 dimethylation. The relationship between DNA hypermethylation and these histone modifications is not completely understood. Using chromatin immunoprecipitation microarrays (ChIP-chip) in prostate cancer cells compared to normal prostate, we found that up to 5% of promoters (16% CpG islands and 84% non-CpG islands) were enriched with H3K27triM. These genes were silenced specifically in prostate cancer, and those CpG islands affected showed low levels of DNA methylation. Downregulation of the EZH2 histone methyltransferase restored expression of the H3K27triM target genes alone or in synergy with histone deacetylase inhibition, without affecting promoter DNA methylation, and with no effect on the expression of genes silenced by DNA hypermethylation. These data establish EZH2-mediated H3K27triM as a mechanism of tumor-suppressor gene silencing in cancer that is potentially independent of promoter DNA methylation.

Epigenetic silencing in mammalian cells is mediated by at least two distinct histone modifications, polycomb-based histone H3 lysine 27 trimethylation (H3K27triM) and H3K9 dimethylation1. Cancer cells usurp silencing mechanisms to extinguish functional pathways by DNA hypermethylation2. The relationship between DNA hypermethylation in cancer and the developmental silencing mechanisms marked by histone modifications can now be studied on a wholegenome scale. Trimethylation at H3K27 (H3K27triM) is a distinct histone modification involved in regulation of homeotic (Hox) gene expression and in early steps of X-chromosome inactivation in women3,4. This process is mechanistically distinct from H3K9 silencing in that it uniquely involves polycomb group (PcG) proteins. A polycomb group (PcG) protein, enhancer of zeste 2 (EZH2), which is a member of polycomb repressor complex 2 (PRC2), has a histone methyltransferase activity with substrate specificity for H3K27 (ref. 3). H3K27triM serves as a signal for specific binding of the chromodomain of another polycomb repressor complex, PRC1, which includes BMI-1, RING1, HPC and HPH5. Binding of PRC1 blocks the recruitment of

transcriptional activation factors, such as SWI/SNF, and the presence of PRC1 prevents initiation of transcription by RNA polymerase II6,7. Deregulation of PcG proteins has been observed in several types of cancer. EZH2 is upregulated in prostate and breast cancers with poor prognosis8,9. BMI-1 is overexpressed frequently in human medulloblastoma cell lines and primary tumors10. These studies imply that deregulation of PcG proteins is closely related to tumorigenesis. However, it remains unclear whether PcG deregulation in cancer results in aberrant H3K27triM, and links between this modification and DNA methylation in cancer have not been well established. By applying chromatin immunoprecipitation (ChIP) coupled with CpG promoter microarrays (ChIP-chip), we found that up to 5% of the genes on the CpG microarray were silenced in cancer cells by H3K27triM independent of DNA methylation. Two of these genes, GAS2 and PIK3CG, were shown to induce growth suppression and/or promotion of apoptosis when reexpressed. This minimal overlap between genes silenced by H3K27triM and the DNA hypermethylated gene subsets was also true for breast cancer. Our data show a newly identified epigenetic aspect of gene deregulation in cancer cells.

1Department of Leukemia, University of Texas M. D. Anderson Cancer Center,1515 Holcombe Boulevard, Houston, Texas 77030, USA. 2Division of Molecular Oncology, Aichi Cancer Center Research Institute, 1-1 Kanokoden, Chikusa-ku, Nagoya 464-8681, Japan. 3Human Cancer Genetics Program, Department of Molecular Virology, Immunology, and Medical Genetics, Comprehensive Cancer Center, The Ohio State University, 420 West 12th Avenue, Columbus, Ohio 43210, USA. 4Department of Lymphoma/Myeloma, University of Texas M.D. Anderson Cancer Center, 1515 Holcombe Boulevard, Houston, Texas 77030, USA. 5Department of Biochemistry II, Nagoya University Graduate School of Medicine, 65 Tsurumai, Showa-ku, Nagoya 466-8550, Japan. 6Department of Pathology, Baylor College of Medicine and Michael E. DeBakey, Department of Veterans Affairs Medical Center, 1 Baylor Plaza, Houston, Texas 77030, USA. 7Department of Biostatistics and Applied Mathematics, University of Texas M.D. Anderson Cancer Center, 1515 Holcombe Boulevard, Houston, Texas 77030, USA. 8Present address: Institute of Digestive Disease, Faculty of Medicine, Chinese University of Hong Kong, Li Ka Shing Medical Sciences Building, Prince of Wales Hospital, Shatin, N.T. Hong Kong SAR. 9These authors contributed equally to this work. Correspondence should be addressed to J.-P.J.I. ([email protected]).

Received 5 June 2007; accepted 19 March 2008; published online 18 May 2008; doi:10.1038/ng.159

NATURE GENETICS VOLUME 40

[

NUMBER 6

[

JUNE 2008

741

DNA methylation mRNA expression

0% +

10 8 6 4 2 0

88% –

15 12

9 6 3 0

9 6 3 0

9 6 3 0

2% +

10 8

10 8

10 8

6

6

6

4

4

2 0

2 0

4 2 0

4 2 0

95% –

92% –

15 12 9 6

19% –

10 8 6

97% –

27% –

K4-2M K9-Ac K9-2M K27-1M K27-3M Ab(–)

12

K4-2M K9-Ac K9-2M K27-1M K27-3M Ab(–)

15

K9-2M K27-1M K27-3M Ab(–)

15 12

96% –

–

39% –

K4-2M K9-A c K9-2M K27-1M K27-2M K27-3M Ab(–)

K4-2M K9-A c K9-2M K27-1M K27-2M K27-3M Ab(–) 13% –

3 0

95% –

K4-2M K9-Ac K9-2M K27-1M K27-3M Ab(–)

K4-2M K9-Ac K9-2M K27-1M K27-2M K27-3M Ab(–) 35% –

ESR1 10 8 6 4 2 0

K4-2M K9-A c K9-2M K27-1M K27-2M K27-3M Ab(–)

K4-2M K9-Ac K9-2M K27-1M K27-2M K27-3M Ab(–) 94% –

K4-2M K9-Ac

K9-2M K27-1M K27-3M Ab(–)

K4-2M K9-Ac

0% +

RAR2 10 8 6 4 2 0

6% +

10

10

8 6 4 2 0

8 6

97% –

96% –

4 2 0

K4-2M K9-Ac K9-2M K27-1M K27-3M Ab(–)

0% +

Homozygous deletion

RAR1 10 8 6 4 2 0

K4-2M K9-Ac K9-2M K27-1M K27-3M Ab(–)

SW48

9 6 3 0

15 12 9 6 3 0

K4-2M K9-Ac K9-2M K27-1M K27-3M Ab(–)


25 20 15 10 5 0

K4-2M K9-Ac K9-2M K27-1M K27-2M K27-3M Ab(–)

15 12

K4-2M K9-Ac K9-2M K27-1M K27-3M Ab(–)

6 3 0

94% –

PGR 10 8 6 4 2 0

K4-2M K9-Ac K9-2M K27-1M K27-3M Ab(–)

15 12 9

K4-2M K9-Ac K9-2M K27-1M K27-3M Ab(–)

MCF7

0% +

0% +

RASSF1 10 8 6 4 2 0



10 8 6 4 2 0



PC 3

20 15 10 5 0

CDKN2A

MGMT 25 20 15 10 5 0

Global assessment of H3K27 modifications by CpG array The above data suggest that DNA methylation in cancer is indeed associated with a specific histone code (H3K4 hypomethylation and H3K9 hypoacetylation and hypermethylation), but they also point to H3K27triM as an alternate code of tumor-suppressor gene silencing in cancer. However, focusing solely on a study of genes known to have

K4-2M K9-Ac K9-2M K27-1M K27-3M Ab(–)

CDKN1A 25

(RAR1 and RAR2). This pattern of H3K27triM was associated with relatively low levels of promoter DNA methylation, as measured by bisulfite pyrosequencing of several adjacent CpG sites (Fig. 1). Thus, in PC3, CDKN2A and RASSF1 had 490% promoter DNA methylation, whereas the four promoters modified at H3K27triM averaged 28% (P o 0.002). The degree of DNA methylation of adjacent CpG sites was very similar in these cell lines. H3K27triM was also observed in breast (MCF7) and colon (SW48) cancers at the RAR1 and RAR2 promoters. However, in the colon cancer cell line SW48, we observed high DNA methylation at both the RAR1 and RAR2 promoters, suggesting that there is tissue and cell line specificity to the process of epigenetic gene silencing. This might reflect the fact that, in some tissues, both DNA methylation and H3K27 can target the same genes. Indeed, SW48 is affected by the CpG island methylator phenotype12,13, in which multiple genes are silenced by aberrant DNA methylation simultaneously, whereas PC3 is not, which may explain these differences.

K4-2M K9-Ac K9-2M K27-1M K27-3M Ab(–)

RESULTS Two distinct histone codes at silenced loci We first used chromatin immunoprecipitation to analyze histone code modifications in genes silenced in cancer. ChIP on cells derived from three different types of malignancies (prostate cancer, PC3; breast cancer, MCF7; colon cancer, SW48) confirmed the previously described11 predominance of H3K9 alterations in genes silenced in association with DNA methylation. Expressed genes (for example, CDKN1A in all three cell lines and MGMT in PC3 and MCF7) are characterized by high H3K4 methylation and H3K9 acetylation and low methylation at H3K9. Genes showing dense promoter CpG island methylation (for example, CDKN2A and RASSF1; Fig. 1) showed a histone code switch to high H3K9 dimethylation (H3K9diM). This was consistent across the three cell lines. Patterns at H3K27, however, were notably different. Active genes show low H3K27 monomethylation (H3K27monoM) and practically undetectable H3K27triM. Genes silenced in association with promoter DNA methylation showed a slightly elevated enrichment for H3K27monoM but no H3K27triM (for example, CDKN2A and RASSF1). Three genes were distinct (Fig. 1). In PC3, RARB, a tumor-suppressor gene on chromosome 3p21, PGR (progesterone receptor) and ESR1 (estrogen receptor a), which are silenced in many cancers, showed marked elevation in H3K27triM, and relatively low H3K27monoM (Fig. 1). This was true for both promoters of RARB

K4-2M K9-Ac K9-2M K27-1M K27-3M Ab(–)


ARTICLES

97% –

Figure 1 Two distinct histone codes at loci silenced in cancer. ChIP and qPCR were used to study histone H3 modifications of multiple genes in prostate (PC3), breast (MCF7) and colon (SW48) cells. Modifications studied are indicated below each column, including K4 dimethylation, K9 acetylation, K9 dimethylation and K27 mono-, di- and trimethylation. The y axis represents the ratio of the H3-methylation or acetylation immunoprecipitations to a core histone H3 immunoprecipitation. Error bars, s.e.m. of the averaged values. Promoter DNA methylation measured by bisulfite pyrosequencing is indicated below each panel, as is mRNA expression measured by qPCR. Three patterns are observed: (i) high K4 dimethylation and K9 acetylation at CDKN1A and MGMT, associated with active gene expression, (ii) high K9 dimethylation and K27 monomethylation, associated with gene silencing and high promoter DNA methylation (CDKN2A and RASSF1), (iii) high K27 trimethylation, associated with gene silencing and variable and/or low promoter DNA methylation (PGR, RARB and ESR1).

742

VOLUME 40

[

NUMBER 6

[

JUNE 2008 NATURE GENETICS

[

NUMBER 6

[

JUNE 2008

2 AS A SF 1 N AB 1 N KX 2 O -3 TO P3 W N SA T1 M D TR 11 IM 3 R 6 AR 2 KC

KN

R

D

D


C

KN

1A

promoter DNA methylation would considerably underestimate the transforming gene when activated18, but its association with b-catenin frequency of H3K27triM-based silencing in cancer. To search for is also consistent with different effects when silenced. NKX2-3 is a H3K27triM targets in an unbiased way, we first carried out ChIP homeodomain transcription factor involved in stem cell physiology microarrays11 on the PC3 prostate cancer line using a CpG island and a homolog of NKX3-1, a putative tumor-suppressor gene that microarray containing 12,220 loci, which is associated with MeCP2 maps to an area of loss of heterozygosity in prostate cancer and that is protein binding (12K CpG-array14; Fig. 2). Of the Cot-1 negative spots epigenetically silenced in that disease without promoter DNA methy(presumed unique genes), high H3K27monoM was present in lation19. Our data are consistent with the fact that WNT1 and NKX2-3 193 genes (2.5%), whereas high H3K27triM was detected at 327 have previously been reported to be targets of SUZ12, a member of genes (4.2%). PRC2, and to be associated with H3K27 methylation by ChIP-chip15. Analysis of 148 loci modified at H3K27monoM showed that most were repetitive elements (short and long interspersed nuclear elements H3K27triM silencing machinery as a cancer-specific event (SINEs and LINEs, respectively) or long terminal repeats (LTRs)), and To determine whether this enrichment at H3K27triM on the three the four genes that were present among the top 40 modified loci were known genes and the nine newly recovered genes was a cancer-specific not silenced in PC3 (Supplementary Table 1 online and data not silencing event, we next studied normal peripheral blood cells and shown). Thus, H3K27monoM shows an inconsistent association with cultured normal prostate epithelial cells. These genes showed no gene silencing in cancer. Alternately, the H3K27monoM antibody used enrichment for H3K27triM in noncultured normal blood cells and here could have poor specificity or cross reactivity with other no or generally much less enrichment in cultured normal prostate modifications, which would confound this interpretation. Analysis epithelial cells (REPE1 and PrEC, Fig. 3a). By RT-PCR and qPCR, we of the top 40 loci modified at H3K27triM showed 9 unique genes found that H3K27triM target genes had undetectable expression in where the arrayed CpG island was in or close to the gene promoter (Table 1 and Supplea b 2520 mentary Table 1). H3-K4 15 We confirmed that each of the nine recov-diM 10 Chromatin IP with antibodies to monomethyl histone H3-K27 5 or trimethyl histone H3-K27 ered genes was enriched at H3K27triM by 0 5 ChIP and qPCR (Fig. 2). In search of a 4 H3K9 3 mechanism for H3K27triM-mediated silen2 Ac 12,220 loci 1 cing, we noted that EZH2, a specific H3K27 0 Applied to CpG 12K arrays Cot-I (+); 4,465 5 Cot-I (–); 7,755 histone methyltransferase and a member of 4 H3K9 3 2 polycomb repressive complex 2 (PRC2), is -diM 1 0 frequently overexpressed in cancers3,8,15,16. 5 4 H3K27triM serves as a signal for specific H3K27 3 -monoM 2 binding of the PRC1, which includes BMI1 0 1, RING1, HPC and HPH12. We used ChIP 10 8 to examine the recruitment of EZH2 and H3K27 6 4 -triM BMI-1 on the target genes we identified 2 H3-K27 monoM (Cy5 red) H3-K27 triM (Cy5 red) 0 (Fig. 2b). Consistent with increased vs. vs. 10 8 Ab (–) (Cy3 green) Ab (–) (Cy3 green) H3K27triM, EZH2 was enriched in all the 6 EZH2 4 genes we identified. BMI-1 was also enriched 2 0 but only in a subset of the genes (KCNAB1, 10 Cot-I (–) 8 K27monoM >2.0 K27triM >2.0 TRIM36 and RARB2). These data suggest that 6 BMI-1 4 the PRC2 complex, including EZH2, is 2 0 148 45 282 indeed associated with H3K27triM100 (1.9%) (0.6%) (3.6%) 80 DNA 60 dependent silencing in cancer cells, whereas methylation 40 20 subsequent recruitment of BMI-1, which is (%) 0 part of the PRC1 complex, is involved in only some the affected genes. Notably, the genes enriched at H3K27triM had low H3K9 acetylation (H3K9Ac), which is a character- Figure 2 Global assessment of H3K27 modifications in cancer. (a) DNA obtained by ChIP analysis in istic mark for expressed genes, but had vari- PC3 using antibodies to mono- or trimethylated H3K27 was used in a ChIP-chip experiment to able H3K4 dimethylation (H3K4diM), a hybridize a CpG island microarray. The fluorescent dyes used were Cy5 and Cy3 coupled to monomethyl modification previously associated with H3-K27 (K27monoM) or trimethyl H3-K27 (K27triM)-immunoprecipitated DNA and control (noantibody) DNA, respectively. The human CpG 12K array contains 12,798 spots, of which 578 are active genes. Recently, a bivalent chromatin negative controls (for example, Arabidopsis sequences). Cot-1 hybridization identified approximately structure with this modification and 4,465 spots of repetitive sequences. Among the 7,755 cot-1–negative loci, 193 (2.5%) loci showed H3K27triM was found in embryonic stem high signal (twofold more than no-antibody background) for monomethyl H3-K27, and 327 (4.2%) cells1,17, and our data suggest that it can showed a high signal of trimethyl H3-K27. (b) Histone modifications and recruitment of EZH2 and occasionally be found in cancer cells as well. BMI-1 on the identified genes by ChIP-chip. We carried out ChIP and qPCR in PC3 cells. The y axis is Some of the genes recovered are of poten- as in Figure 1. CDKN1A was used as an active gene control, and CDKN2A and RASSF1 were used as DNA methylation–dependent silencing controls. All the genes identified by ChIP-chip showed high tial interest in cancer. Of note, one of these H3K27triM with enrichment of EZH2. They also showed low H3K9Ac, low H3K27monoM and variable was RARB, which was shown earlier to be H3K4diM. BMI-1 is also enriched at several loci. Error bars, s.e.m. of the averaged values. DNA a target of H3K27triM, thus confirming methylation was analyzed by bisulfite pyrosequencing of 2–6 adjacent CpG sites in the promoter the validity of the microarrays. WNT1 is a region. DNA methylation was generally low in identified H3K27triM targets. C


ARTICLES

743


ARTICLES PC3, whereas 9 of 12 were expressed in normal prostate, and all were expressed in normal testis (Fig. 3b). To determine the relevance of these findings to uncultured human cancers, we next examined the expression of H3K27triM target genes by qPCR in matched normal and cancerous tissues from individuals with prostate cancer (Fig. 3c). Several of these individuals had higher expression of EZH2 in cancers than in normal tissues (as previously reported8,15,16); further, they showed marked decreases of expression of several H3K27triM target genes in cancers. Expression of PGR, RAR2, KCNAB1, RGMA and SAMD11 were significantly lower in cancers than in normal tissues (P o 0.05). Expression of NKX2-3, OTOP3 and WNT1 was not detectable in normal or cancerous tissues, and we were unable to develop a qPCR assay specific for RAR1. Only TRIM36 and ERa showed no evidence of silencing in prostate cancer. These data show that, for many genes, H3K27triM is used by cancer cells to silence genes that should be active in normal cells, an event that presumably provides a selective advantage by modifying the function of key tumor-suppressor pathways.

in the 88K promoter array, indicating that both arrays could detect target genes with B80% sensitivity. ChIP-chip (88K array) analysis in PrEC showed that 737 genes were also targets of H3K27triM in PrEC. However, a large number of genes, including RARB (chromosome 3p) and PIK3CG (chromosome 7q) were observed to be enriched specifically in PC3 (Supplementary Fig. 1 online), suggesting that H3K27triM was a cancer-specific event. ChIP-chip (88K array) on the breast cancer cell line MCF7 identified 852 genes (4.8%) enriched at H3K27triM, which was a similar number to that observed in PC3. ChIP-chip (12K CpG-array) on the colon cancer SW48 cell line, however, showed a notably different pattern (Supplementary Table 3 online). Only 21 genes (0.3%) showed high H3K27triM, and 76 genes (1%) showed high H3K27monoM, supporting the idea that disruption of the epigenetic silencing machinery in cancer is tissue and cell line specific. In SW48, we also found RARB among H3K27triM targets, confirming the validity of the microarrays.

H3K27triM is not closely associated with DNA methylation An important question with regards to epigenetic inactivation of Global assessment of H3K27 modifications by promoter array tumor-suppressor genes in cancer concerns the extent to which We recovered the genes discussed above on the basis of a CpG island mechanisms distinct from DNA methylation participate in the promicroarray, potentially introducing biases to the results. We therefore cess. A similarly key issue is the relation between PcG proteins and applied the ChIP-chip method to a promoter array containing 88K DNA methylation. Several groups have reported that genes marked by probes, which cover the proximal promoter region of 18,300 genes20. PcG in embryonic stem cells are more susceptible to DNA methylation We found that 887 genes (4.8%) showed high signal intensities in cancer21–23. However, it remains unclear whether this is an indirect indicating bound H3K27triM ChIP products. We annotated the 200 effect (for example, related to low baseline occupancy by transcription genes most significantly bound and found that 31 (16%) had CpG factors), or a direct consequence of PcG proteins recruiting DNA islands and 169 genes (84%) did not, suggesting that the silencing methyltransferases, as suggested by a previous report24. To address this machinery associated with H3K27triM tends to target non-CpG island issue, we next analyzed DNA methylation at several H3K27triMpromoters in this cell line (Supplementary Table 2 online). Among 62 modified CpG islands using bisulfite pyrosequencing of 2–6 adjacent genes that showed enrichment for H3K27triM on the 12K CpG array CpG sites (Fig. 2b and Table 1). Consistent with our earlier observa(Supplementary Table 1), we found 38 (61%) that were also enriched tions on candidate genes, methylation was generally low in these, ranging from 6.3% to 39.3%, with only three out of eight genes having methylation above Table 1 Summary of histone H3K27 methylation status in PC3 15%. These levels of DNA methylation are considerably lower than what has been genMicroarray signal intensity ChIP-PCR (IP/Input) erally observed for tumor suppressors silenced by promoter DNA methylation in cancer (for DNA example, CDKN2A, RASSF1). There was no methylation mRNA Gene Mono-Me Tri-Me Mono-Me Tri-Me K4-Me K9-Me (%) expression consistent relationship between H3K27triM and DNA methylation, which was in marked Identified by ChIP on chip contrast to our previous finding that genes 0.8 6.7 0.9 4.1a 0.5 1.5 39.3 – RAR2 enriched for H3K9 dimethylation identified by 20.0 2.1 31.7 – WNT1 0.6 5.1 0.5 5.3a ChIP-chip showed a strong and linear correla5.8 1.0 12.2 – KCNAB1 1.0 4.1 0.5 1.8a tion with DNA methylation11. To more pre9.1 0.9 7.9 – TRIM36 0.5 3.9 0.2 3.2a cisely quantitate DNA methylation, we carried 0.5 1.2 11.1 – SAMD11 0.3 3.4 0.3 2.8a out bisulfite sequencing at five H3K27triM RGMA 0.4 3.0 0.9 1.5 12.4 – target CpG islands (Supplementary Fig. 2 1.7 1.3 14.6 – NKX2-3 0.5 2.7 0.4 3.0a online) and confirmed the low levels of 0.3 1.1 31.1 – OTOP3 0.4 2.6 0.4 2.5a methylation observed by bisulfite pyrosequenCR2 0.5 2.6 0.2 0.9 6.3 – cing. We also could not find any specific CpG Candidates studied by ChIP-PCR site preferentially methylated in the promoter a 0.8 2.1 35.1 – PGR 0.8 5.4 region of each gene. These observations a 0.9 5 0.7 2.0 13.2 – RAR1 further support the concept of independence 0.9 1.4 27 – ESR1 0.9 9a between H3K27-based silencing and H3K4/ a 0.5 2.3 93.8 – CDKN2A 3.1 0.2 K9- (and DNA methylation-) based silencing. a 0.5 1.8 94.1 – RASSF1 4.4 0.2 We next examined this issue in a global way 13.4 0.7 0 + CDKN1A 1.5 0.4a using methylated CpG island microarrays 17.4 1.4 0 + MGMT 2.4 0.2a (MCAM), a recently described sensitive and The microarray header refers to spot intensity by microarray analysis. The ChIP-PCR header refers to PCR confirmation specific technique25,26. Analysis of the same data. DNA methylation of the promoter CpG island was determined by bisulfite/pyrosequencing. mRNA expression was determined by RT-PCR. aAssays were done by real-time PCR. cells, PC3 and MCF7, for which we had

744

VOLUME 40

[

NUMBER 6

[


ARTICLES

a

10 8 Normal 6 4 PB 2 0


PrEC

K27-1M

K27-3M

RT-PCR PC3 Normal Positive prostate Ctrl PGR

c

EZH2 0.020

*

0.015 0.010 0.005

10 8 6 4 2 0

RAR2 0.000 KCNAB

Normal cancer

PGR 0.008 0.007 0.006 0.005 0.004 0.003 0.002 0.001 0.000

**

Normal cancer KCNAB

RAR2 0.020

ESR1

10 8 6 RPWE1 4 2 0

PC3

b

**

0.005

**

0.004

0.015

0.003 RAR1

0.010

0.002

0.005 SAMD11

10 8 6 4 2 0 CDKN1A CDKN2A MGMT RAR1 RAR2 SAMD11 WNT1 OTOP3 TRIM36 PGR

TRIM36 RGMA

0.000 0.100 0.075

0.001 Normal cancer RGMA

**

0.000

0.03

Normal cancer SAMD11

**

0.02

0.050 CR2 0.025

0.01

Figure 3 H3K27triM and its related gene silencing as a cancer-specific 0.000 0.00 WNT1 event in prostate cancers. (a) Comparison of H3K27monoM and Normal cancer Normal cancer H3K27triM between cancer (PC3), normal peripheral blood cells (normal TRIM36 ESR1 0.006 0.03 OTOP3 PB) and cultured normal prostate epithelial cell lines (PrEC and RPWE1). 0.005 The y axis is as in Figure 1. H3K27triM is highly enriched specifically in 0.004 0.02 NKX2-3 PC3 but lower or absent in normal PB and normal prostate epithelial cell 0.003 lines. Gray and black bars indicate H3K27monoM and H3K27triM, 0.002 0.01 GAPDH respectively. CDKN1A, CDKN2A and MGMT in normal PB, PrEC and 0.001 0.000 0.00 RWPE1 were used as active gene controls. Error bars, s.e.m. of the Normal cancer Normal cancer averaged values. (b) Expression of the H3K27triM affected genes is examined by RT-PCR in normal prostate, and the prostate cancer cell line (PC3) used for ChIP. The gene name is indicated to the left of each row. Normal testis or placenta was used as a positive control for each gene expression. High H3K27triM was associated with gene silencing in PC3 (summarized in Table 1). The results shown here by conventional RT-PCR are consistent with qPCR results. (c) Expression of H3K27triM target genes in matched normal and cancerous tissues from individuals with prostate cancer examined by qPCR. The average EZH2 expression in cancers is significantly higher than that in normal tissues (*P o 0.05). By contrast, in most of the H3K27triM target genes, average gene expression in cancerous tissues is significantly lower than that in normal tissues (**P o 0.05).

H3K27triM data focusing on genes detectable on both ChIP-chip and DNA methylation arrays, showed that most of the genes enriched at H3K27triM had no detectable DNA hypermethylation, and most genes showing DNA hypermethylation had no enrichment for H3K27triM (Figs. 4a,b). This was also true when we limited the analysis to the genes differing between PrEC and PC3 (Fig. 4a). Although largely independent, there was, however, some overlap between DNA methylation and H3K27triM. Some of this may represent low but detectable levels of DNA methylation (as shown for the genes studied in detail earlier), but there likely are some genes targeted by both silencing mechanisms, although these seem to be relatively rare. H3K27-based silencing is independent of DNA methylation To confirm that H3K27triM-dependent silencing does not require or rapidly lead to DNA methylation in cancer cells, we selected one of the H3K27triM target genes, RAR2 promoter, and inserted it into a reporter vector (Supplementary Fig. 3a online)27. After transfection of the RAR2 reporter constructs into PC3 and SW480 cells, in which the endogenous RAR2 gene is silenced, we found that they promptly lost promoter activity (data not shown). We then established single stable clones of RAR2 reporter constructs. In contrast to the SV40 promoter, the RAR2 promoter was inactivated and stably silenced over 6 months (Supplementary Fig. 3b and data not shown). Analysis by bisulfite sequencing showed no or minimal DNA methylation in the RAR2 promoter region of these constructs (Supplementary Fig. 3c). Consistent with the endogenous RAR2 promoter status, we detected increased H3K27triM on the exogenously transfected


[

NUMBER 6

[

JUNE 2008

promoter. This promoter silencing was reversed by the histone deacetylase (HDAC) inhibitor, trichostatin A (TSA; Supplementary Fig. 3d), supporting the idea that RAR2 is downregulated through H3K27triM-PRC2–mediated silencing machinery, which requires HDAC activity4 (see below). We next used treatment with the DNA methyltransferase inhibitor 5-aza-2¢-deoxycytidine (DAC), the HDAC inhibitor, TSA and short hairpin RNA (shRNA) specific for EZH2 to dissect the contribution of each event to gene silencing in PC3 (Fig. 5a). DAC efficiently reactivated p16 and RASSF1, both of which are characterized by H3K9diM and H3K27monoM, whereas neither TSA nor EZH2 inhibition had an impact on reactivation of these genes. These data confirm the previously noted dependence of H3K9diM on DNA methylation11. In marked contrast, EZH2 inhibition alone activated the expression of six of seven of the promoters showing H3K27triMassociated silencing. TSA was also able to increase the expression of all of these promoters, and the combination of TSA and EZH2 activated the expression of all seven promoters silenced by H3K27triM but none of the genes silenced by H3K9diM. Similar results were obtained with three different shRNA constructs targeting EZH2, and no effect was noted when a scrambled shRNA control construct was used (data not shown). These epigenetic treatments could only partially impair the epigenetic silencing machinery, as expression of reactivated genes was only B1% of that in normal prostate tissues (y axis; Fig. 3c). Even when DNA methylation inhibitors are used, it is rare for gene expression to be restored to normal levels in a short time—that would likely require multiple treatments over a prolonged period of time. However, clearly distinct responses to each targeted intervention

745

ARTICLES

a MCF7

PC3

141

90

490

H3K27 triM

PC3 relative to PrEC Both 4

Both 38

Both 34

DNA methylation

738

76 H3K27 triM

H3K27 triM

446

DNA methylation

b Signal ratio (Cy5/Cy3)

Chromosome 3p 40

Signal ratio (Cy5/Cy3)

40

30

PrEC PC3 DNA Me (PC3/PrEC)

20 10 0

30

Chromosome 7q PrEC PC3 DNA Me (PC3/PrEC)

20 10 0

point to different epigenetic silencing mechanisms regulating different genes silenced in cancer. As it has been suggested that EZH2 inhibition might lead to loss of DNA methylation24, we also analyzed promoter methylation of identified H3K27triM target genes by bisulfite pyrosequencing after inhibition of EZH2 or after DAC (Fig. 5b). In contrast to DAC treatment, EZH2 inhibition did not result in any DNA demethylation of H3K27triM target genes. DNA methylation in SAMD11, KCNAB1, NKX2-3, TRIM36 and WNT1 averaged 24.5% at baseline, 28.5% after EZH2 inhibition (P ¼ 0.7 compared to baseline) and 9.1% after DAC treatment (P ¼ 0.004 compared to baseline). Thus, we are unable to confirm an effect of EZH2 on DNA methylation. Given that gene activation was found after EZH2 inhibition at most H3K27triM targets with no effects on DNA methylation, these data further support the concept that this process is independent of DNA methylation.

microarray analyses (P ¼ 0.00038 and 0.00045, respectively). These two genes have no CpG islands and thus cannot be silenced permanently by DNA methylation. Consistent with increased H3K27triM, these two genes are silenced in PC3 (Fig. 6a,b). GAS2 was also silenced in MCF7 with increased H3K27triM. PIK3CG was expressed in MCF7 in concordance with active histone modifications (Fig. 6b). This gene silencing was reactivated by downregulation of EZH2 or TSA treatment (Fig. 6b and data not shown), as we found in the other H3K27triM target genes. We transfected GAS2 in PC3 and MCF7 cells (Fig. 6c). Overexpression of GAS2 alone could not induce apoptosis as was previous reported28 but could increase drug-induced apoptosis. Treatment with etoposide (100mM) for 24 h resulted in apoptosis in 13% of mock transfected cells, which increased to 32% in GAS2-transfected MCF7 cells, which have wild-type p53. However, GAS2 overexpression in PC3, which lacks p53 protein, did not

a

CDKN2A

0.012

0.15

0.009 0.10

0.006

0.05

0.003

The H3K27triM targets GAS2 and PIK3CG are tumor-suppressors To confirm that targets of H3K27triM, which were detected by our arrays, include tumor-suppressor genes inactivated in cancer independently of DNA methylation, we selected two genes for analysis—GAS2 (growth arrest specific 2) and PIK3CG (phosphoinositide 3-kinase catalytic subunit g)—with high signal intensities on the 88K promoter

0.000

C D

T

E

D E

T E

0.00

PGR

–4

(×10 ) 0.10

E D E

T E

0.2

0.04

0.1 C D

T E

D E

0.0

T E

TRIM36

–3

(×10 ) 0.4

C D T

E

D E

T E

NKX2-3

–4

(×10 ) 0.10

0.02

0.0

0.00

C D

T E

D E

T E

E

D T E E

WNT1

0.3 0.2

0.04

0.1

T

–4

0.06

0.2

(×10 ) 0.6 0.5 0.4 0.3 0.2 0.1 0.0 C D

T E

KCNAB1

(×10 ) 0.4

0.08

0.3

–4 RAR2 (×10 ) 0.6 0.5 0.4 0.3 0.2 0.1 0.0 C D T E D E –4

0.3

0.06

b

T

SAMD11

(×10 ) 0.4

0.08

0.00

746

C D –3

0.02

Figure 5 Reactivation of silenced gene expression by inhibition of DNA methylation, histone deacetylation or EZH2 expression. (a) Gene expression for the genes indicated was measured by qRT-PCR in the PC3 cell line after treatment with the DNA methylation inhibitor 5-aza-2-deoxycytidine (DAC, D), the HDAC inhibitor trichostatin A (TSA, T), an siRNA construct specific for the H3K27 methyltransferase EZH2 (E) or a combination of siRNA and 5-aza-2-deoxycytidine (DE) or siRNA and TSA (TE). The two genes silenced in association with promoter DNA methylation (CDKN2A and RASSF1) are reactivated only by DNA methylation inhibition. All other genes, silenced in association with H3K27 trimethylation, are reactivated by EZH2 knockdown (except PGR), TSA or a combination of the two. (b) Promoter methylation was analyzed by bisulfite/pyrosequencing. Black, white and gray bars indicate DNA methylation status in PC3 with mock viral infection, with EZH2 knockdown and with DAC treatment, respectively. The assays were done at least in triplicate. Error bars, s.e.m. from independent experiments in triplicate.

RASSF1

–3

(×10 ) 0.20

0.015

DNA methylation (%)


DNA methylation

Figure 4 Global assessment of DNA methylation and H3K27triM in normal and cancer cell lines. (a) Comparison of global analysis of DNA methylation targets using MCAM and H3K27triM targets using ChIP microarrays in PC3, MCF7 and PrEC. Venn diagrams show that most genes enriched at H3K27triM are not targets of DNA hypermethylation and vice versa. When compared to PrEC, PC3 has many specific targets of DNA methylation and H3K27triM that also rarely overlap. The Venn diagrams shown underestimate H3K27triM targets because they focus on CpG island containing genes, and most H3K27triM targets are not CpG island associated. (b) The y axis and x axis indicate signal ratio of Cy5/Cy3 and relative probe location on chromosome 3p (top) and chromosome 7q (bottom). ChIP products (enriched for H3K27triM) and MCA products (enriched for DNA methylation) from PC3 and PrEC were labeled with Cy5 (red) and Cy3 (green), respectively. Each dot represents the value of each probe (red, PrEC; blue, PC3; green, relative value (PC3/PrEC) of DNA methylation). There was some overlap between DNA methylation and H3K27triM, although this is generally rare (as shown in a).

0.1 C

D T

E

D T E E

0.0

40

C D

T

E

D T E E

Mock ctrl EZH2 KD DAC

30 20 10 SAMD11

KCNAB1

VOLUME 40

[

NKX2-3

NUMBER 6

TRIM36

[

WNT1


ARTICLES

2.0

10

1.6

8

Actin

GAS2

Ab(–)

K27-3M

PC3

0.00

(×10–4)

(×10–4)

EZH2 0.01

0.8 0.4 0.0

0.2 0.0

(×10–5)

Ab(–)

K27-3M

K27-1M

PIK3CG

MCF7 0.030 0.015 0.000

0.02

GAS2

K9-2M

0

Ab(–)

0

0.4

0.6

0.2

Mock EZH2 KD

GAS2 Mock Etoposide

0.0

– + –

0

2%

Effects of EZH2 knockdown in cancer cell lines EZH2 overexpression is a poor prognosis indicator in prostate and breast cancer3, and our results suggest that a possible mechanism of this effect is gene silencing through H3K27triM. To explore this mechanistic link between EZH2 expression, gene silencing and transformation, we analyzed the effects of EZH2 shRNA on gene expression and cellular biology. The EZH2 shRNA constructs efficiently downregulated EZH2 expression as measured by real-time PCR or protein blots. Consistent with the downregulation of EZH2, total H3K27triM was decreased (Fig. 7). In parallel, using ChIP, we found decreased levels of BMI-1 as well as H3K27triM in KCNAB1, TRIM36, RAR2 and PGR after EZH2 downregulation (Fig. 7), consistent with a direct involvement of EZH2 in establishing the silencing code. As discussed earlier, shRNA efficiently activated the expression of multiple genes

[

NUMBER 6

[

JUNE 2008

0

13%

29%

PC 3 0

increase susceptibility to apoptosis, suggesting that GAS2 only enhances p53-dependent apoptosis (Fig. 6d)28. We obtained similar results with another p53-dependent apoptotic stimulus, UVC irradiation (180 J/m2, data not shown). Growth curves and colony formation assays showed that PIK3CG overexpression in PC3 resulted in strong growth inhibition (Fig. 6e). By contrast, MCF7, in which PIK3CG is well expressed, showed no growth inhibition by overexpression of PIK3CG (data not shown). We examined the expression status of these genes in individuals with prostate cancer and found that 6 (30%) and 5 (25%) out of 20 individuals showed a more than twofold decrease in expression of GAS2 and PIK3CG in prostate cancer tissues compared to surrounding normal prostate tissues, respectively. These data indicate that H3K27triM target genes are clearly involved in the neoplastic phenotype.

+ – +

0

SubG1

Mock EZH2 KD

– + +

MCF7

(×10–2)

1.2 0.0

d

0

2%

0

15%

11%

e Mock

6 3

Cell number (×10 )

Figure 6 H3K27triM and its related gene silencing are cancer-relevant events in prostate and breast SubG1 cancers. (a) GAS2 and PIK3CG, which showed high signal intensities on microarray analyses (P ¼ 0.00038 and 0.00045, respectively) are modified with high H3K27triM. In MCF7, GAS2 is a target of H3K27triM. The y axis is as in Figure 1. Error bars, s.e.m. of the averaged values. (b) Expression of the GAS2 and PIK3CG genes is examined by RT-PCR (conventional PCR in the upper and qPCR in the lower columns) in normal prostate, PC3 and MCF7. Gene expression is reactivated in EZH2 downregulated cells (EZH2 KD) by siRNA. (c) Protein blotting shows that transfection of GAS2 or PIK3CG expression vectors effectively induces the expression of each gene. (d) Cells transfected with GAS2 were treated with 100 mM of etoposide for 24 h. FACS analyses show that GAS2 expression increases susceptibility to apoptosis (13% to 32% in sub-G1 fraction) in MCF7 with wild-type p53 but not in p53-defective cell line, PC3 (15% to 11% in sub-G1 fraction). (e) Ectopic PIK3CG expression induces strong growth inhibition in PC3. After 48 h of PIK3CG transfection, cells were plated (1 104/each wells in 24-well plate) at the same density and counted daily by trypan blue exclusion (left column). Filled circle and filled triangle indicate the growth of mock control and PIK3CG transfected cells, respectively. Error bars, s.d. from independent experiments in triplicate. Colony formation is also inhibited in PIK3CG-transfected PC3 (right column).


Mock GAS2 IB GAS2

IB

GAPDH

K4-2M K9-Ac

2 K27-3M

4

0.4 K27-1M

0.8

K9-2M

6

K9-Ac

1.2

MCF7 PIK3 GAS2 CG

Mock

PIK3CG

0.4

PIK3CG

PC3

Actin

K27-1M

0

c MCF7

PIK3CG

K9-Ac

0.6

0

Normal prostate PC3 GAS2

K4-2M

0.6 Ab(–)

1.2

K27-3M

1.2

K27-1M

1.8

K9-Ac

2.4

1.8

K9-2M

2.4

K4-2M

b

MCF7 3.0

K4-2M


GAS2

PC3 3.0

K9-2M

a

4 PIK3CG 2

0

0

1

2 (Days)

3

4

silenced in cancer. Using expression microarrays (Affymetrix), we found that 257 genes were activated twofold or more in these cells, including genes involved in growth, differentiation and adhesion, such as SPARC and ITGB8, which are related to cancer metastasis (Supplementary Table 4 online). The growth of the cells with EZH2 downregulation was markedly inhibited, and this was associated with an increased number of cells undergoing apoptosis, as well as a delay in G2/M progression (Fig. 7). The morphology of the cells acquired typical characteristics of senescence, and this was confirmed by a marked increase in staining for the senescence-associated marker b-galactosidase (Fig. 7). Similar results were seen in multiple independent clones of shRNA-transfected cells, whereas two clones that did not show EZH2 downregulation had no detectable change in growth, cellular morphology or senescence characteristics (data not shown). We observed a similar effect in H1299 (lung cancer) and RKO (colon cancer) cells (Fig. 7). This inhibition of cell growth and senescence upon deletion of EZH2 has also been reported previously8,9,16, suggesting that EZH2-mediated H3K27triM is indeed a key to overcome senescence and promote tumorigenesis. DISCUSSION The data presented here establish H3K27triM as an epigenetic mark pathogenically involved in neoplasia through the silencing of tumor-suppressor genes such as RARB, GAS2 or PIK3CG. This process is mechanistically distinct from DNA methylation–associated silencing, which is characterized by H3K4 and H3K9 modifications. One technical drawback to our findings is the difficulty in analyzing chromatin modifications in uncultured cancer cells. The findings that genes silenced in cell lines are also downregulated in primary

747

ARTICLES

a

b

0.015

c

10

d

256 Ctrl

PC3

0.012 0.009

K27-3M

EZH2 KD

Ctrl 6

Counts

Ctrl

Cell number (×103)

0.003

EZH2

RKO

8

0.006 0.000

SA-β-gal H1299

PC3

4

1,023 256 EZH2 KD EZH2 KD

1,023

KCNAB

20

3

cancers are reassuring in this regard, but it will be important in future studies to confirm the data as technology evolves or in model systems. Recently, links between PcG-mediated methylation on H3K27and de novo DNA methylation in cancers were described using ChIP analyses coupled with bioinformatic database mining21,22. However, our data suggest that a direct interaction between these two pathways is unlikely. Four lines of evidence support the independence of H3K27triM and DNA methylation (and associated H3K9diM). First, classical gene targets of DNA methylation such as CDKN2A and RASSF1 show no modification of H3K27triM. Second, H3K27triM targets generally show no or low DNA methylation in their promoters (with exceptions discussed below). Third, inhibition of EZH2, the main methyltransferase responsible for H3K27triM, reduces H3K27triM, suppresses clonogenicity, and reactivates hundreds of genes including genes silenced by H3K27triM, but has no effect on silencing by DNA methylation, which contradicts a previous report24. In addition, inhibition of histone deacetylation, which is a key component of PcG complexes, is enough to reactivate genes silenced by H3K27triM and is synergistic with EZH2 inhibition in this respect, but has no effect at all on genes silenced by DNA methylation. Fourth, transfection of one of the H3K27triM targets (RAR2) into the cells led to rapid silencing that was stable over 6 months, accompanied by H3K27triM but no or minimal DNA methylation. The conflict between our study and previous studies might be explained by tissue- and cancer-specific differences related to activation of specific silencing pathways. Indeed, the colon cancer cell line SW48 had little H3K27triM overall and also showed high DNA methylation at H3K27triM target genes. It is possible that, in the setting of activation of the DNA methyltransferase machinery, as postulated for colon cancers with a hypermethylator phenotype29 (such as the SW48 cells studied here), the two mechanisms of gene silencing will also interact, as they do on the inactive-X

748

4

0

0

K27 K27 1M 3M

EZ H 2 BM I1

4

3 2

8

1

WNT1

1 0

K27 K27 1M 3M

PGR

4

3

2

2

2

1

1

1

0

0 H 2 BM I1

3

K27 K27 1M 3M

K27 K27 1M 3M

CDKN1A

4

3

K27 K27 1M 3M

RAR2

4

12

2

Figure 7 Effects of EZH2 knockdown (EZH2 KD) in cancer cell lines. (a) Real-time PCR (upper) and protein blotting (lower) show that shRNAi effectively downregulated EZH2 expression by 490% in PC3 cell line. EZH2 KD also reduced total H3K27triM as measured by protein blotting, with no effects on total H3 content. (b) EZH2 KD resulted in strong growth inhibition in PC3. Cells were plated at the same density and counted daily by trypan blue exclusion. Filled circle and filled triangle indicate the growth of mock control and EZH2 KD cells, respectively. Error bars, s.d. from independent experiences in triplicate. (c) FACS analysis showing a high degree of sub-G1 cells (apoptotic) and G2/M cells in the EZH2 KD PC3 cell line compared to mock control. (d) Photomicrographs showing that EZH2 KD in PC3, H1299 (lung cancer) and RKO (colon cancer) cells are greatly enlarged and stain positively for SA-b-gal, indicating induction of senescence. Cell size is relative to the 50-mm bar. (e) ChIP shows that EZH2 KD decreased H3K27triM and BMI-1 at specific gene promoter loci. ChIP experiments were done and ratios established by real-time PCR at least twice. Gray and black bars indicate ChIP samples from mock control and EZH2 KD, respectively. Error bars, s.e.m. of the averaged values.

TRIM36

16

EZ H 2 BM I1

(Day)

4

0

K27 K27 1M 3M

H 2 BM I1

5

EZ

4

I1

3

BM

2

EZ H 2 BM I1

1

2

0

H

e

0

EZ

Actin

EZ


2 H3

Ctrl

EZH2 KD

chromosome30. Future studies will need to address mechanistically why DNA methylation in cancer affects some silenced PcG targets but not others and why this shows remarkable variability between different cancers. Our data add a novel layer of complexity to epigenetic dysregulation in cancer and also establish H3K27triM-mediated silencing as a promising therapeutic target. METHODS Cell lines and culture conditions. The prostate cancer cell line PC3, the lung cancer cell line H1299 and the breast cancer cell line MCF7 were grown in RPMI; the colon cancer cell line SW48 was grown in L-15 media (Invitrogen), and the colon cancer cell lines SW480 and RKO were grown in DMEM (Invitrogen) plus 10% FBS. All six cell lines were obtained from the American Type Culture Collection. Human prostate epithelial cell line PrEC was obtained from Cambrex Bioscience and grown in the PrEC-specific culture medium (Cambrex Bioscience). Human prostate epithelial cell line RWPE1 was obtained from ATCC and maintained in the keratinocyte-serum free medium (K-SFM) containing 5 ng/ml human recombinant epithelial growth factor (hrEGF), 0.05 mg/ml bovine pituitary extract (Invitrogen).The cells were grown in plastic tissue culture plates in a humidified atmosphere containing 5% CO2 at 37 1C. They were grown to a density of 1.0 106 to 3.0 106 cells per dish before being harvested for crosslinking experiments. Trichostatin A (TSA) and 5-aza-2¢-deoxycytidine (DAC) treatment of cells. Cells were split 12–24 h before treatment. Cells were then treated with either 1 mM of DAC (Sigma) or PBS (control) daily for 3 days. TSA (300 nM, ICN Biomedicals) or an identical volume of ethanol (control) treatment was done for 20 h. ChIP. ChIP assays were done based on a modification of previously published methods3. Cells (8 106) were treated with 1% formaldehyde for 8 min to crosslink histones to DNA. After washing by cold PBS, the cell pellets were resuspended in lysis buffer (150 mM NaCl, 25 mM Tris (pH 7.5), 5 mM EDTA, 1% Triton X-100, 0.1% SDS, 0.5% sodium deoxycholate) and sonicated 8 s seven times. We then divided the lysate into three fractions. The first lysate was

VOLUME 40

[

NUMBER 6

[



ARTICLES incubated with 10 ml each of antibody to K4 dimethylated histone H3, antibody to K9 acetylated histone H3, antibody to K9 dimethylated histone H3, antibody to K27 monomethylated histone H3, antibody to K27 dimethylated histone H3, and antibody to K27 trimethylated histone H3 (Upstate Biotechnology; we also used antibody to K9 dimethylated histone H3 established by ourselves and obtained consistent results), antibody to histone H3 (Abcam), antibody to EZH2 or antibody to Bmi (Santa Cruz Biotechnology) at 4 1C overnight. The specificity of the histone modification antibodies used for ChIP was confirmed by ChIP-PCR for control genes using GAPDH and p21 promoters as active gene controls and Alu sequences as silenced gene controls31,32. Once identified, WNT1 and NKX2.3 promoters in SW480 cell line were used for positive controls for antibody to K27 trimethylated histone H315. We also carried out protein blotting to confirm that these antibodies could recognize the modified histones or proteins. The second lysate was incubated with TE buffer (10 mM Tris, 1 mM EDTA (pH 8.0), 10 ml) at 4 1C overnight as a negative control. The third lysate (2% of total) was used for input control. To collect the immunoprecipitated complexes, we added protein G-Sepharose beads (GE Healthcare) and left them to incubate for 1 h at 4 1C. After washing, the beads were treated with RNase (50 mg/ml) for 30 min at 37 1C and then proteinase K overnight. The crosslinks were then reversed by heating the sample at 65 1C for 6 h. DNA was extracted by the phenol/chloroform method, ethanol-precipitated and resuspended in water. To obtain DNA enriched with monomethylated or trimethylated H3-K27 or no-antibody control for microarray, we carried out ChIP from 20 plates of 15 cm dishes (3 106/dish), each with monomethylated or trimethylated H3-K27 or without antibody. Promoter microarray analyses. The protocol for CGI microarray hybridization has been previously described33. The combined monomethylated H3K27 or trimethylated H3K27-specific samples and the combined no-antibody samples were labeled and used to probe the microarray. Incorporation of amino-allyl dUTP (aa-dUTP, Sigma) into 5 mg each of monomethylated or trimethylated H3-K27-DNA and control no-antibody DNA was conducted by using the Bioprime DNA-labeling system protocol (Life Technologies). Cy5 and Cy3 fluorescent dyes were coupled to aa-dUTP–labeled monomethylated or trimethylated H3-K27–precipitated DNA and no-antibody DNA, respectively, and cohybridized to the microarray panel. Microarray protocols, including the hybridization and posthybridization washing procedures, are available online (see URLs section below). Labeled chromatin was hybridized with human CpGisland microarrays obtained from the Ontario University Health Network (see URLs section below). Hybridized slides were scanned with a GenePix 4000A scanner (Axon Instruments), and the acquired images were analyzed with the software GENE PIX PRO 6.0. The ratio of the two fluorescent dyes was log2 transformed and normalized using intensity-dependent normalization. We then used the average value from the two independent replicates for further analysis. Loci with features of poor intensity and those that had obvious blemishes were manually flagged and removed from the putative positive list. We also analyzed ChIP products as probes on the 88K promoter array (Agilent Technologies). ChIP products were labeled with cy-5 (red) and input with cy-3 (green) using a random primed Klenow polymerase reaction (Invitrogen) at 37 1C for 3 h. Labeled samples were then hybridized to the 88K human promoter array (Agilent Technologies) in the presence of human Cot-1 DNA for 40 h at 65 1C. After washing the array according to the manufacturer’s protocol, arrays were scanned on an Agilent scanner and analyzed using Feature Extraction software (Agilent Technologies). Results were analyzed by the neighborhood error model in ChIP Analytics software (version 1.1, Agilent Technologies)34. Confirmation ChIP-PCR. ChIP product from cells (PC3, MCF7, SW48, RWPE1 and PrEC) was used for confirmation ChIP-PCR with the oligonucleotide primers shown in Supplementary Table 5 online. The PCR products were visualized by 6% PAGE and quantitated by densitometry. For quantitation, TaqMan Q-PCRs were done in an ABI Prism 7000 (Applied Biosystems) in triplicate for the target genes. RT-PCR analyses. Total RNA was isolated using TRIzol (Invitrogen), and 2 mg were reverse transcribed with Super-scriptase (Invitrogen). Conventional RT-PCRs were carried out with the oligonucleotide primers shown in Supplementary Table 5. The PCR products were visualized by 3% agarose


[

NUMBER 6

[

JUNE 2008

gel electrophoresis and quantitated by densitometry (Bio-Rad). If the density of a target gene was less than 0.1% than that of GAPDH, we defined it as an expression negative in a sample. TaqMan Q-PCRs were then carried out in duplicate for the target gene EZH2, PGR, SAMD11, KCNAB1, WNT1, TRIM36, NKX2-3, GAS2 and PIK3CG using probe sets Hs00544830_m1, Hs00172183_m1, Hs00153462_m1, Hs00600966_g1, Hs00399414_m1, Hs00180529_m1, Hs00218867_m1, Hs00414553_g1, Hs 00169477_ml and Hs 00176916_ml, respectively (Applied Biosystems). Bisulfite pyrosequencing methylation analysis. We carried out bisulfite treatment as reported previously35. Briefly, 2 mg of genomic DNA was denatured with 2 M NaOH for 10 min, followed by incubation with 3 M sodium bisulfite (pH 5.0) for 16 h at 50 1C. After treatment, DNA was purified by using a Wizard Miniprep Column (Promega), precipitated with ethanol and resuspended in 30 ml of diluted water. We carried out a highly quantitative method to assess DNA methylation levels using the Pyrosequencing technology36 (Pyrosequencing AB). This method is not restricted to restriction enzyme sites, avoids sequencing multiple clones and allows accurate quantitation of multiple CpG methylation sites in the same reaction. Primer sequences and conditions used for methylation determination are described in Supplementary Table 5. For selected genes, we also cloned the PCR products and sequenced multiple clones for higher-resolution analysis. RNA interference. We designed three different retrovirus vectors (RNAi-Ready pSIREN-RetroQ Vector, Clontech Laboratories), each encoding a shRNA directed against EZH2 in PC3. Target sequences are listed in Supplementary Table 5. The target vector was co-transfected with pVSV-G (Clontech Laboratories) into GP2-293 cells (Clontech Laboratories) and obtained viruscontaining medium according to the manufacturer’s protocol. Viruses were incubated with PC3 (multiplicity of infection ¼ 0.4) in the presence of polybrene (8 mg/ml). After selection with puromycin (5 mg/ml), stable clones were established in 3–4 weeks. All three constructs could reduce EZH2 expression more than 90% and induced similar morphology changes. As a control, we used shRNAs for luciferase (Luc), synthesized by BD Biosciences, or shRNA vector without hairpin oligonucleotides (mock). SEAP (secreted alkaline phosphatase) reporter assay. We inserted 601 bp and 1,151 bp (–171 to +430 and –721 to +430 relative to the transcription start site, respectively) of RAR2 promoter fragments into KpnI- and NheI-digested pSEAP2-Basic Vector (Takara Bio) to give pSEAP-601 and pSEAP-1151. Cells were seeded in 10-cm plates at 80% confluence in complete medium 18 h before transfection. We used Lipofectamine 2000 (Invitrogen) to transfect the pSEAP-601 or pSEAP-1151 plasmids (20 mg) and pcDNA3 (2 mg) that contained the neomycin-resistant gene (Invitrogen) according to the manufacturer’s instructions. pSEAP2-Control vector containing SV40 promoter and pSEAP2-Basic vector without promoter were transfected as a positive and base line controls for this assay, respectively. After 48 h of transfection, G418 (Invitrogen) was added for selection. Finally, we picked up resistant clones to G418 and assayed SEAP activity according to manufacturer’s protocol (Toyobo). Units of SEAP activity were expressed as means of triplicate experiences. Plasmid constructs for H3K27triM target genes. Full-length cDNA of GAS2 and PIK3CG were obtained (Ultimate ORF human clones and MGC full-length clones, respectively; Invitrogen). After transferring the target sequences into expression vector pcDNA-DEST40 (Invitrogen) using Gateway systems, we confirmed the sequences in the expression vector. Cells were seeded in 10-cm plates at 80% confluence in complete medium 18 h before transfection. After 48 h of transfection, we assayed the transfected cells for growth curve, colony formation assay and FACS analyses. Empty pcDNA-DEST40 was used as a mock control. Colony formation. Cells (1 105) were plated in 100-mm culture dishes for 24 h before transfection with the expression vector pcDNA-DEST40-PIK3CG or empty vector (10 mg each) using Lipofectamine 2000 (Invitrogen). After transfection, cells were selected for 14 d in medium containing 0.6 mg/ml G418. The resultant colonies were then stained with methylene blue. SA-b-gal analysis. PC3, H1299 and RKO cells infected with retrovirus encoding EZH2-shRNA or control-shRNA were examined for SA-b-gal activity as previously described37.

749

ARTICLES


Antibodies and protein analyses. Immunoblotting was done with primary antibodies against EZH2 (Santa Cruz Biotechnology), K27 trimethylated histone H3 (Upstate Biotechnology), histone H3 (Abcam), b-Actin (Sigma), GAS2 (Abnova) and PIK3CG (R&D Systems). RNA microarray analysis. Total RNA was extracted from cells infected with either EZH2 shRNA or control Luc shRNA as described above. Targets for microarray hybridization were generated from the RNA according to the manufacturer’s instructions (Affymetrix). The human U133A gene chip (Affymetrix), which contains 33,000 transcripts, was used for gene expression profiling. Hybridization, washing, scanning and analysis of gene chips were done according to the manufacturer’s instructions. Expression was analyzed by the statistical algorithm in the Microarray Analysis Suite (MAS) 5.0 software (Affymetrix) using the default parameters. The data from the Luc shRNA treatment were used as a baseline expression for comparison with the EZH2 shRNA–treated sample. MCAM. Methylated CpG island amplification (MCA) was carried out using DNA from the cell lines. A detailed protocol for MCA was described previously26. Briefly, 2 mg of genomic DNA was digested with 100 U of methylation-sensitive restriction endonuclease SmaI (New England Biolabs) for 8 h at 20 1C two times, which cuts unmethylated DNA and leaves blunt ends (CCC/GGG). Subsequently, the DNA was digested with 20 U of methylationinsensitive restriction endonuclease XmaI for 9 h at 37 1C, which creates sticky ends (C/CCGGG). In total, 500 ng of digested DNA was ligated. After filling in the overhanging ends of the ligated DNA fragments at 72 1C, DNA was amplified in a condition of 95 1C for 3 min followed by 25 cycles of 1 min at 95 1C and 3 min at 77 1C using 100 pmol of RMCA24 primer. MCA products were labeled with cy-5 (red) for PC3 and cy-3 (green) for reference DNA (PrEC) using a random primed Klenow polymerase reaction (Invitrogen) at 37 1C for 3 h. Labeled samples were then hybridized to a customized array26 based on the 88K human promoter array (Agilent Technologies) in the presence of human Cot-1 DNA for 40 h at 65 1C. After washing the array according to the manufacturer’s protocol, we scanned arrays on an Agilent scanner and analyzed them using Feature Extraction software (Agilent Technologies). URLs. Ontario University Health Network, http://www.microarrays.ca. Accession codes. ArrayExpress: E-MEXP-1581 (gene expression array data) and E-MEXP-1585 (ChIP experiment in PC3). Note: Supplementary information is available on the Nature Genetics website. ACKNOWLEDGMENTS This work was supported in part by US National Institutes of Health by grants P50CA100632, R33CA89837 and RO1CA098006 to J.-P.J.I. and grant P50CA058204 to B.K.-A. J.-P.J.I. is an American Cancer Society Clinical Research Professor supported by a generous gift from the F.M. Kirby Foundation. Y.K. was supported by an Odyssey Fellowship at the M.D. Anderson Cancer Center and the Cell Science Research Foundation. We especially thank Y. Guo and X. Chen for their technical assistance. AUTHOR CONTRIBUTIONS Y.K. and J.-P.J.I. designed the research; Y.K., L.S., A.S.C., S.A., Y.B. and C.C. performed the research; T.Y., T.U., K.F., Y.S. and T.H.-M.H. designed the assays; D.L.G. analyzed microarray data; B.K.-A. collected and analyzed clinical samples; and Y.K. and J.-P.J.I. wrote the paper. Published online at http://www.nature.com/naturegenetics/ Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/ 1. Jenuwein, T. & Allis, C.D. Translating the histone code. Science 293, 1074–1080 (2001). 2. Herman, J.G. & Baylin, S.B. Gene silencing in cancer in association with promoter hypermethylation. N. Engl. J. Med. 349, 2042–2054 (2003).

750

3. Cao, R. & Zhang, Y. The functions of E(Z)/EZH2-mediated methylation of lysine 27 in histone H3. Curr. Opin. Genet. Dev. 14, 155–164 (2004). 4. Lachner, M. & Jenuwein, T. The many faces of histone lysine methylation. Curr. Opin. Cell Biol. 14, 286–298 (2002). 5. Levine, S.S. et al. The core of the polycomb repressive complex is compositionally and functionally conserved in flies and humans. Mol. Cell. Biol. 22, 6070–6078 (2002). 6. Dellino, G.I. et al. Polycomb silencing blocks transcription initiation. Mol. Cell 13, 887–893 (2004). 7. Shao, Z. et al. Stabilization of chromatin structure by PRC1, a Polycomb complex. Cell 98, 37–46 (1999). 8. Varambally, S. et al. The polycomb group protein EZH2 is involved in progression of prostate cancer. Nature 419, 624–629 (2002). 9. Kleer, C.G. et al. EZH2 is a marker of aggressive breast cancer and promotes neoplastic transformation of breast epithelial cells. Proc. Natl. Acad. Sci. USA 100, 11606–11611 (2003). 10. Leung, C. et al. Bmi1 is essential for cerebellar development and is overexpressed in human medulloblastomas. Nature 428, 337–341 (2004). 11. Kondo, Y., Shen, L., Yan, P.S., Huang, T.H. & Issa, J.P. Chromatin immunoprecipitation microarrays for identification of genes silenced by histone H3 lysine 9 methylation. Proc. Natl. Acad. Sci. USA 101, 7398–7403 (2004). 12. Toyota, M. et al. CpG island methylator phenotype in colorectal cancer. Proc. Natl. Acad. Sci. USA 96, 8681–8686 (1999). 13. Toyota, M. et al. Identification of differentially methylated sequences in colorectal cancer by methylated CpG island amplification. Cancer Res. 59, 2307–2312 (1999). 14. Cross, S.H., Charlton, J.A., Nan, X. & Bird, A.P. Purification of CpG islands using a methylated DNA binding column. Nat. Genet. 6, 236–244 (1994). 15. Kirmizis, A. et al. Silencing of human polycomb target genes is associated with methylation of histone H3 Lys 27. Genes Dev. 18, 1592–1605 (2004). 16. Bracken, A.P. et al. EZH2 is downstream of the pRB-E2F pathway, essential for proliferation and amplified in cancer. EMBO J. 22, 5323–5335 (2003). 17. Bernstein, B.E. et al. A bivalent chromatin structure marks key developmental genes in embryonic stem cells. Cell 125, 315–326 (2006). 18. Brennan, K.R. & Brown, A.M. Wnt proteins in mammary development and cancer. J. Mammary Gland Biol. Neoplasia 9, 119–131 (2004). 19. Asatiani, E. et al. Deletion, methylation, and expression of the NKX3.1 suppressor gene in primary human prostate cancer. Cancer Res. 65, 1164–1173 (2005). 20. Boyer, L.A. et al. Core transcriptional regulatory circuitry in human embryonic stem cells. Cell 122, 947–956 (2005). 21. Schlesinger, Y. et al. Polycomb-mediated methylation on Lys27 of histone H3 pre-marks genes for de novo methylation in cancer. Nat. Genet. 39, 232–236 (2007). 22. Widschwendter, M. et al. Epigenetic stem cell signature in cancer. Nat. Genet. 39, 157–158 (2007). 23. Ohm, J.E. et al. A stem cell-like chromatin pattern may predispose tumor suppressor genes to DNA hypermethylation and heritable silencing. Nat. Genet. 39, 237–242 (2007). 24. Vire, E. et al. The Polycomb group protein EZH2 directly controls DNA methylation. Nature 439, 871–874 (2006). 25. Estecio, M.R. et al. High-throughput methylation profiling by MCA coupled to CpG island microarray. Genome Res. 17, 1529–1536 (2007). 26. Shen, L. et al. Genome-wide profiling of DNA methylation reveals a class of normally methylated CpG island promoters. PLoS Genet. 3, 2023–2036 (2007). 27. Dey, A., Minucci, S. & Ozato, K. Ligand-dependent occupancy of the retinoic acid receptor beta 2 promoter in vivo. Mol. Cell. Biol. 14, 8191–8201 (1994). 28. Benetti, R. et al. The death substrate Gas2 binds m-calpain and increases susceptibility to p53-dependent apoptosis. EMBO J. 20, 2702–2714 (2001). 29. Issa, J.P. CpG island methylator phenotype in cancer. Nat. Rev. Cancer 4, 988–993 (2004). 30. Plath, K. et al. Role of histone H3 lysine 27 methylation in X inactivation. Science 300, 131–135 (2003). 31. Kondo, Y. & Issa, J.-P. Enrichment for histone H3 lysine 9 methylation at Alu repeats in human cells. J. Biol. Chem. 278, 27658–27662 (2003). 32. Martens, J.H. et al. The profile of repeat-associated histone lysine methylation states in the mouse epigenome. EMBO J. 24, 800–812 (2005). 33. Weinmann, A.S., Yan, P.S., Oberley, M.J., Huang, T.H. & Farnham, P.J. Isolating human transcription factor targets by coupling chromatin immunoprecipitation and CpG island microarray analysis. Genes Dev. 16, 235–244 (2002). 34. Pokholok, D.K. et al. Genome-wide map of nucleosome acetylation and methylation in yeast. Cell 122, 517–527 (2005). 35. Clark, S.J., Harrison, J., Paul, C.L. & Frommer, M. High sensitivity mapping of methylated cytosines. Nucleic Acids Res. 22, 2990–2997 (1994). 36. Colella, S., Shen, L., Baggerly, K.A., Issa, J.-P. & Krahe, R. Sensitive and quantitative universal Pyrosequencing methylation analysis of CpG sites. Biotechniques 35, 146–150 (2003). 37. Campisi, J. Analysis of tumor suppressor gene-induced senescence. Methods Mol. Biol. 223, 155–172 (2003).

VOLUME 40

[

NUMBER 6

[
