Unmasking of epigenetically silenced candidate tumor ... - Nature

Oncogene (2008) 27, 3556–3566

& 2008 Nature Publishing Group All rights reserved 0950-9232/08 $30.00 www.nature.com/onc

ORIGINAL ARTICLE

Unmasking of epigenetically silenced candidate tumor suppressor genes by removal of methyl-CpG-binding domain proteins L Lopez-Serra1, E Ballestar1, S Ropero1, F Setien1, L-M Billard2, MF Fraga1, P Lopez-Nieva1, M Alaminos3,4, D Guerrero5, R Dante2 and M Esteller1 1 Cancer Epigenetics Laboratory, Spanish National Cancer Research Centre (CNIO), Madrid, Spain; 2Unite´ INSERM 590, Laboratoire d’Oncologie Molećulaire, Centre León Be´rard, Lyon, France; 3Department of Histology, Granada University, Granada, Spain; 4Fundacioń Hospital Clıńico, Granada, Spain and 5Centro de Investigacioń Biome´dica, Servicio Navarro de Salud, Navarra, Spain

Methyl-cytosine-phosphate-guanine (CpG)-binding domain (MBD) proteins are bound to hypermethylated promoter CpG islands of tumor suppressor genes in human cancer cells, although a direct causal relationship at the genome-wide level between MBD presence and gene silencing remains to be demonstrated. To this end, we have inhibited the expression of MBD proteins in HeLa cells by short hairpin RNAs; and studied the functional consequences of MBD depletion using microarray-based expression analysis in conjunction with extensive bisulfite genomic sequencing and chromatin immunoprecipitation. The removal of MBDs results in a release of gene silencing associated with a loss of MBD occupancy in 50 -CpG islands without any change in the DNA methylation pattern. Our results unveil new targets for epigenetic inactivation mediated by MBDs in transformed cells, such as the cell adhesion protein c-parvin and the fibroblast growth factor 19, where we also demonstrate their bona fide tumor suppressor features. Our data support a fundamental role for MBD proteins in the direct maintenance of transcriptional repression of tumor suppressors and identify new candidate genes for epigenetic disruption in cancer cells. Oncogene (2008) 27, 3556–3566; doi:10.1038/sj.onc.1211022; published online 28 January 2008 Keywords: DNA methylation; epigenetics; methyl-CpGbinding domain proteins; tumor suppressor genes; RNA interference

Introduction During the past few years, considerable attention has been focused upon the possibility that proteins that Correspondence: Dr M Esteller, Cancer Epigenetics Laboratory, Molecular Pathology Program, Spanish National Cancer Research Centre (CNIO), C/Melchor Fernandez Almagro 3, Madrid 28029, Spain. E-mail: [email protected] Received 12 October 2007; revised 26 November 2007; accepted 4 December 2007; published online 28 January 2008

‘read’ DNA methylation patterns may play a central role in cellular transformation. This is mainly due to the recognition that DNA hypermethylation of the promoter cytosine-phosphate-guanine (CpG) islands of tumor suppressor genes, and resulting gene inactivation, represent a major event in tumor etiology and progression (Jones and Laird, 1999; Herman and Baylin, 2003; Esteller, 2007). The methyl-CpG-binding domain (MBD) family of proteins is the largest group of these factors that bind methylated DNA, and their exact role in the epigenetic silencing of tumor suppressor genes has functional and translational consequences that need to be clarified (Ballestar and Esteller, 2005; Fatemi and Wade, 2006). The MBD family of proteins is composed of five bona fide members, MeCP2, MBD1, MBD2, MBD3 and MBD4, which share an MBD that allows them to bind methylated DNA (Hendrich and Bird, 1998; Ballestar and Esteller, 2005; Fatemi and Wade, 2006). With the exception of MBD4, which is involved in DNA repair (Hendrich et al., 1999), and MBD3, whose MBD is unable to bind methylated DNA selectively (Saito and Ishikawa, 2002; Fraga et al., 2003), MBDs couple DNA methylation with transcriptional repression through association with histone deacetylases (HDACs; Jones et al., 1998; Nan et al., 1998; Wade et al., 1999) and histone methyltransferases (Fujita et al., 2003; Fuks et al., 2003). These properties have led them to being proposed as having a major role in the aberrant epigenetic silencing of tumor suppressor genes (Ballestar and Esteller, 2005; Fatemi and Wade, 2006). Genetic analysis of MBD proteins has shown that most single MBD-deficient mouse models do not exhibit dramatic phenotypes (Guy et al., 2001; Hendrich et al., 2001; Zhao et al., 2003). However, detailed analysis shows that subtle but important changes are associated with deficiency in individual MBD proteins. For instance, loss of MBD2 is associated with a significant change in the abundance of transcripts for certain cytokines that are crucial to the process of T-lymphocyte differentiation (Hutchins et al., 2002). Crossing Mbd2-null mice (Hendrich et al., 2001) with ApcMin also inhibits the development of intestinal adenomas (Sansom et al., 2003) and MeCP2 is required for

MBD depletion in transformed cells L Lopez-Serra et al

3557

prostate cancer cell growth (Bernard et al., 2006). In addition, disruption of the MeCP2 gene in Rett syndrome samples from mouse models (Jordan et al., 2007) or human patients (Ballestar et al., 2005) is associated with upregulation of a subset of genes as a result of the loss of interaction of MeCP2 with methylated CpG sites at the promoter region. Since DNA methylation patterns differ dramatically between cancer cells and their normal counterparts (Jones and Laird, 1999; Herman and Baylin, 2003; Esteller, 2007), it is likely that the distribution and, ultimately, the biological relevance of MBD proteins are also significantly different in normal and transformed cells. In normal cells, promoter CpG islands are mostly unmethylated, with the exception of those of imprinted genes, X-chromosome genes in females and a number of tissue-specific genes (Jones and Laird, 1999; Herman and Baylin, 2003; Esteller, 2007). In contrast, cancer cells are characterized by the generation of specific patterns of hypermethylation at the promoter CpG islands of tumor suppressor genes (Jones and Laird, 1999; Herman and Baylin, 2003; Esteller, 2007). DNAmethylated heterochromatic sequences are probably the primary binding site for MBD proteins in normal cells, as suggested by the fact that MeCP2 is enriched in pericentromeric heterochromatin in murine cells, in accordance with its content of major satellite DNA, the largest fraction of methylated DNA sequences in mice (Lewis et al., 1992). Another normal set of targets for MBDs are imprinted genes, where MBDs associate with the differentially methylated allele (Fournier et al., 2002). In cancer cells, the hypermethylated promoter CpG islands of tumor suppressor genes constitute new and aberrant targets for MBD proteins and accumulated evidence indicates that hypermethylation of tumor suppressor genes is accompanied by association of MBDs (Magdinier and Wolffe, 2001; Nguyen et al., 2001; Bakker et al., 2002; Koizume et al., 2002; Ballestar et al., 2003; Lopez-Serra et al., 2006). Given the association of MBDs with HDACs and methyltransferases and their DNA methylation-dependent effects on gene transcription, their essential contribution to the epigenetic silencing of tumor suppressor genes is generally accepted, even though no formal evidence has been presented to date. We have addressed this issue in transformed cells by studying the global gene expression patterns upon knocking down MBDs using short hairpin RNA molecules (interfering RNA, RNAi) targeting for individual (MeCP2, MBD1 and MBD2) and combined MBDs. Our results show that removal of MBDs cause a release in gene silencing without changing the underlying DNA methylation patterns of the respective DNA regulatory regions. Our comprehensive epigenomic screening also identifies new candidate genes undergoing transcriptional silencing in human cancer cells in association with CpG island promoter hypermethylation and MBD occupancy. Functional analysis of these genes, including colony formation assays and the use of xenografts in nude mice, indicates their putative tumorsuppressing properties. Therefore, these data underline

the critical role of MBDs in the maintenance of epigenetic gene silencing and the usefulness of MBDdepletion strategies to ‘catch’ new hypermethylated genes in human cancer. Results Removal of MBD proteins results in release of epigenetic gene silencing To demonstrate functionally the direct role of MBDs in gene silencing, we carried out a systematic depletion of MBD proteins in HeLa cells (human cervical cancer) in conjunction with comprehensive expression microarray analyses. We interfered with the expression of the three MBD proteins (MeCP2, MBD1 and MBD2) that have a functional MBD and for which association with histone modification enzymes and transcriptional repression properties has been demonstrated (Figure 1a; Jones et al., 1998; Nan et al., 1998; Ng et al., 1999, 2000). Seven different RNA interference experiments were performed: single MeCP2, MBD1 and MBD2, combined MeCP2/MBD1, MeCP2/MBD2 and MBD1/ MBD2, and the triple MeCP2/MBD1/MBD2 combination. First, we confirmed the robust depletion of the corresponding MBD proteins after transient transfection with RNAi oligonucleotides at both the RNA (Figures 1b and c) and protein levels (Figure 1d) for the single, double and triple combinations. Upon depletion of one single MBD, we did not observe any significant changes in the expression of the other members of the MBD family (Figures 1b and d). No significant changes in the global 5-methylcytosine DNA content determined by high-pressure liquid chromatography were observed upon MBD depletion (Supplementary Figure S1). From the HeLa-untreated and MBD-interfered cells, total RNA was extracted, reverse transcribed, hybridized to cDNA microarrays and the data were analysed, as described in ‘Materials and methods’. We observed a release of transcriptional silencing upon depletion of MBD proteins by RNAi (Figure 2a). Of the 6386 genes represented in the expression array, 967 (15%) experienced an expression change between untreated and triple MBD-depleted cells (Figure 2a). Most importantly, 99% (964 of 967) of these differences corresponded to increased transcription of each respective gene (Figure 2a). We observed the presence of a CpG island in their 50 -ends in 74.27% (716 of 964) of the described genes. Supplementary Table S1 lists the 964 genes upregulated upon triple-MBD interference in HeLa cells. For single MBD interference, MBD2 depletion was the protein most commonly involved in the observed release of gene silencing by far (Figure 2a). The presence of a putative described MeCP2-binding motif, (A/T)>4 (Klose et al., 2005), in the 50 -end regions of reactivated genes upon single MBD depletion was similar for MeCP2 (29%, 54 of 187 genes), MBD1 (25%, 66 of 264 genes) and MBD2 (24%, 219 of 913). Downregulated genes upon MBD removal were almost absent (Figure 2a). Oncogene


3558 MBD1 NA

CxxCxxC

BD

1

siM

siM

BD

m

P2

do m

0.006 MeCP2/GAPDH

siM

Ra

MBD

eC

MBD MBD3

2

siR

MBD MBD2

MBD1

MBD4

MBD2

Repair

MBD MeCP2

MeCP2 MBD

TRD GAPDH

0.003 0.002 0.001

0.025

0

2 siM 2 BD siM

l ro

siM BD 1

Co

80

0.01 0.005 0

nt ro l siM eC P2

100

0.01 0.005

nt

MBD2/ MeCP2/ MBD1

Co

MBD1/ MBD2

MBD2/GAPDH

MBD2/ MeCP2

0.02 0.015

2

MBD2

BD

WT

0.015

siM

20

0.025

0.02

siM BD 1

40

siM eC P2

60

0

BD M

MBD1

MBD1/ MeCP2

MBD1/ MBD2

MBD2/ MeCP2/ MBD1

100

Anti-MeCP2

2 BD M si

WT

co

nt

ro

l

si

P2 eC M

co

20

si

nt

ro

l

40

1

60

0

Relative expression of MeCP2

BD

nt ro l siM eC P2 siM BD 1

Co

80

MBD1/GAPDH

Relative expression of MBD2

0.004

0

100

Relative expression of MBD1

0.005

Anti-MBD2 Anti-MBD1 Anti-β-actin

80 Anti-β-actin

60 40 20 0 WT

MeCP2

MeCP2/ MBD1

MBD2/ MeCP2

MBD2/ MeCP2/ MBD1

Figure 1 Depletion of methyl-CpG-binding domain (MBD) proteins in HeLa cells. (a) Depiction of the MBD family of proteins. Other features shown are the well-defined transcriptional repression domain (TRD) of MeCP2, the thymine glycosylase domain (Repair) of MBD4 and the CxxCxxC of MBD1. (b) Conventional (inset gel) and real-time (graphs) reverse transcription (RT)–PCR analyses showing specific depletion of the targeted MBD in the RNAi experiments. (c) Quantitative RT–PCR showing MBD depletion in the corresponding RNAi combinations. (d) Western blot showing specific depletion of the targeted MBD in the RNAi experiments.

Genes undergoing silencing release upon MBD removal are characterized by promoter CpG island hypermethylation and MBD occupancy Among the hundreds of gene upregulated following MBD interference, imprinted genes and those on the X-chromosome (HeLa cells are derived from a female donor) are obvious candidates for MBD transcriptional repression associated with promoter CpG island hypermethylation (Ballestar and Esteller, 2005; Fatemi and Wade, 2006). A list of genes from this upregulated category upon MBD depletion is shown in Supplementary Table S2. We have instead focused on those genes lying in a different category: putative tumor suppressor genes that are epigenetically inactivated in human cancer (Jones and Laird, 1999; Herman and Baylin, 2003; Esteller, 2007). Thus, we have randomly selected 26 50 -CpG island-containing genes, from different cellular pathways disrupted in cancer cells (Hanahan and Weinberg, 2000), which are at least twofold Oncogene

overexpressed in the triple MBD-interfered cells for further DNA methylation and MBD-binding studies. These candidate genes are described in Supplementary Table S3. Extensive bisulfite genomic sequencing of the promoter CpG islands of these genes reveals dense DNA hypermethylation in 38% (10 of 26) of these candidate genes in HeLa cells (Figures 2b and 3a). The remaining genes with MBD depletion-mediated upregulation, but unmethylated promoter CpG islands, may represent either targets indirectly regulated by MBD proteins, the presence of DNA methylation sites outside the canonical proximal promoter CpG island or methylation-independent targets of MBD proteins, particularly those related to the CxxC motifs of MBD1 (Fujita et al., 1999). Genes previously described as hypermethylated and bound by MBDs in HeLa cells, such as CDH1 and PR (Lopez-Serra et al., 2006), were also found reactivated upon triple MBD depletion (data not shown), however, for other hypermethylated

MBD depletion in transformed cells L Lopez-Serra et al Number of upregulated genes Number of downregulated genes

3559 1000 900 800 700 600 500 400 300 200 100 0

1000 900 800 700 600 500 400 300 200 100 0

siMeCP2

siMBD1

siMBD2

siMeCP2

siMBD1

siMBD2

siMeCP2/ siMBD1

siMeCP2/ siMBD1

siMeCP2/ siMBD2

siMeCP2/ siMBD2

siMBD1/ siMBD2

siMBD1/ siMBD2

siMeCP2/ siMBD1/ siMBD2

siMeCP2/ siMBD1/ siMBD2

HeLa cells Triple MBD siRNA

3 downregulated genes

964 upregulated > 2 fold

Bisulfite sequencing of 26 randomly chosen sequences

16 Unmethylated 5’-CpG islands

10 Hypermethylated 5’-CpG islands Figure 2 Genes undergoing release of silencing upon methyl-CpG-binding domain (MBD) removal are characterized by promoter cytosine-phosphate-guanine (CpG) island hypermethylation and MBD occupancy. (a) Number of upregulated (top) and downregulated (down) genes in the seven different MBD RNAi combinations. (b) Schematic strategy used to unmask MBD-bound hypermethylated 50 -CpG islands in transformed cells.

genes in HeLa cells, such as CHFR and TIPM3, which do not present occupancy by MeCP2, MBD1 or MBD2 in these cells (Lopez-Serra et al., 2006), the triple MBD knockdown approach was not able to restore their expression.

To establish a functional link between MBDs and DNA methylation-associated transcriptional silencing in the candidate genes, it is essential to demonstrate physical occupancy by MBDs in these hypermethylated 50 -CpG islands. To accomplish this aim, we performed chromatin Oncogene

3560

Oncogene

LTBP3

COL11A2

HeLa

HeLa

siRNA HeLa

siRNA HeLa


PTPRN

PARVG

HeLa HeLa

siRNA HeLa

LTBP3

UB COL11A2 B

B

P2 D1 D2 B eC MB M M

B e NA M

CP

2 M

B

Demethylating Agent

1 2 B eCP BD BD M M NA M

UB PTPRN

B

Demethylating Agent

2 1 2 B eCP BD BD M M NA M

1 2 BD BD M

UB COLL11A2

B

PARVG

Demethylating Agent

UB LTBP3

UB

PTPRN

Demethylating Agent

N

N A B M eC P2 M B D 1 M B D 2

UB

UB B

AB




siRNA HeLa

UB PARVG

B

B

Figure 3 (a) Illustrative bisulfite genomic sequencing analyses of four candidate genes (latent transforming growth factor b-binding protein 3 (LTBP3), collagen type XI a-2 (COL11A2), protein tyrosine phosphatase-like N precursor (PTPRN) and g-parvin (PARVG)) in HeLa cells. Cytosine-phosphate-guanine (CpG) dinucleotides are represented as short vertical lines. The transcriptional start site is represented as a long black arrow and the location of bisulfite genomic sequencing PCR primers is indicated as gray arrows. Ten single clones are represented for each sample. Presence of a methylated or unmethylated cytosine is indicated by a black or white square, respectively. Dense hypermethylation of the four CpG islands is observed in HeLa cells. Triple methyl-CpG-binding domain (MBD) depletion by RNA interference (siRNA HeLa) does not induce CpG island hypomethylation events. (b) Chromatin immunoprecipitation (ChIP) analysis for MBDs in the hypermethylated 50 -CpG islands of the above-described four candidate genes in HeLa cells. Both unbound (U) and bound (B) fractions are shown. A negative no antibody (NAB) control is included. MBD occupancy in the hypermethylated CpG islands is observed. (c) ChIP analysis in HeLa cells treated with a DNA demethylating agent (5-aza-2-deoxycytidine) shows the release of MBDs from the described CpG islands.


3561

immunoprecipitation (ChIP) analyses of the 10 genes for which we had observed dense CpG island hypermethylation and of 5 unmethylated genes in HeLa cells. We consistently observed that MBDs were exclusively bound to the hypermethylated promoter CpG islands (Figure 3b), while unmethylated 50 -CpG islands were found to be devoid of MBDs (Supplementary Figure S2), reinforcing the notion of the in vivo preference of MBD proteins for DNA-methylated sequences. Importantly, treatment of HeLa cells with a DNA demethylating agent (50 -aza-2-deoxycytidine) induced the release of MBDs from the described CpG islands (Figure 3c). Both sets of data reinforce the notion of the in vivo preference of MBD proteins for DNA-methylated sequences. Finally, we considered whether the removal of the MBDs by the RNA interference approach affected the 50 -CpG island DNA methylation patterns of these candidate genes, in which we had demonstrated MBD occupancy by ChIP. We found the same DNA methylation pattern in these sequences in the untreated and MBD-interfered HeLa cells—a densely hypermethylated CpG island (Figure 3a).

MBD-bound hypermethylated genes contribute to cell transformation and human tumorigenesis The acceptance of the epigenetic silencing of tumor suppressor genes, such as p16INK4a, hMLH1 and BRCA1, by CpG island promoter hypermethylation as a major hallmark of human transformed cells (Jones and Laird, 1999; Herman and Baylin, 2003; Esteller, 2007) prompted us to investigate the contribution of the newly identified MBD target genes to tumorigenesis. First, we studied the 10 candidate genes showing CpG island methylation and MBD occupancy in HeLa cells to determine whether the presence of hypermethylation was a cancer-specific event. Extensive bisulfite genomic sequencing and methylation-specific PCR analyses for the corresponding 50 -CpG islands showed that for g-parvin (PARVG), fibroblast growth factor 19 (FGF19), protein tyrosine phosphatase-like N precursor (PTPRN), collagen type XI a-2 (COL11A2) and latent transforming growth factor b-binding protein (LTBP) 3, CpG island hypermethylation was commonly observed in human cancer cell lines (n ¼ 23) from different tumor types (cervix, breast colon, lung, leukemia and

COL11A2 MDA-MB-231 Normal Breast Normal Lymphocyte

LTBP3 SiHa

Normal Cervix Normal Lymphocyte

LTBP3

GAPDH

GAPDH

H

2O

HeLa 5’ -a za N L

C

COL11A2

5’ -a z C a

2O

SiHa H

-a za N L

5’

C

5’ -a

C

za

MDA-MB-231 HeLa

Figure 4 Methyl-CpG-binding domain (MBD)-bound genes with specific DNA hypermethylation and associated transcriptional silencing in transformed cells. (a) Illustrative bisulfite genomic sequencing analyses of two candidate genes (collagen type XI a-2 (COL11A2) and latent transforming growth factor b-binding protein 3 (LTBP3)) in normal and transformed cells. Six single clones are represented for each sample. Presence of a methylated or unmethylated cytosine is indicated by a black or white square, respectively. Cytosine-phosphate-guanine (CpG) island hypermethylation of COL11A2 and LTBP3 is restricted to cancer cells (such as HeLa, MDA-MB-231 and SiHA) and it is absent in normal tissues (normal cervix, breast and lymphocytes are shown). (b) Expression analyses for COL11A2 and LTBP3 using reverse transcription–PCR. The COL11A2-hypermethylated MDA-MB-231 and HeLa cells and the LTBP3-hypermethylated SiHa and HeLa cell show loss of expression of the respective transcripts in untreated control cells (‘C’ lane) and restoration of expression is observed upon treatment with the demethylating agent 5-aza-2-deoxycytidine (50 -aza). The water reaction and normal lymphocytes (NL) are shown as negative and positive controls, respectively. Oncogene


3562

lymphoma), but was absent from all normal tissues studied (n ¼ 13), indicating its cancer-specific profile (Figure 4a; Supplementary Table S4). In contrast, TUBA-2, QKI, GJB3, KRT14 and TSSC4 50 -CpG island methylation was observed in both cancer cell lines and normal tissues (Supplementary Figure S3). No particularly different MBD-binding profile was observed in the cancer-specific compared with the normal hypermethylated CpG islands (data not shown). For the MBD-bound cancer-specific hypermethylated genes, we further demonstrated that not only the presence of CpG island hypermethylation was associated with the lack of each respective mRNA transcript (Figure 3b), but also that treatment of the cancer cells with the DNA demethylating agent 5-aza-20 -deoxycytidine restored gene expression (Figure 4b), providing a further link between DNA methylation, MBD binding and transcriptional silencing. Furthermore, to determine whether these genes display putative tumor suppressor features, as do other classical hypermethylated genes (for example, p16INK4a, hMLH1 and BRCA1), we adopted a double approach, in vitro and in vivo, for two illustrative cases, PARVG and FGF19. First, we transfected PARVG or FGF19 in HeLa cells, which have DNA methylation/MBDassociated silencing of both genes, as demonstrated above, and assessed cell viability using the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay (see ‘Materials and methods’). Restoration of PARVG or FGF19 expression, demonstrated in Figure 5a, caused a marked reduction of the viability of HeLa cells (Figure 5b). We completed the demonstration of the growth-inhibitory features of PARVG and FGF19 in colony focus assays and nude mouse models. In the colony formation assay, we used G418 selection after transfecting the HeLa cell line with the PARVG or FGF19 gene or the empty vector (Figure 5c). PARVG or FGF19 reexpression revealed tumor suppressor activity whereby there were marked reductions of 82 and 90%, respectively, in colony formation density with respect to the empty vector (Figure 5c). We next tested the ability of PARVG- or FGF19-transfected HeLa cells to form tumors in nude mice compared with empty vector-transfected cells (Figure 5d). Cells transfected with the empty vector formed tumors rapidly, but cells infected with the PARVG or FGF19 expression vector had much lower tumorigenicity (Figure 5d). Finally, we observed that the presence of 50 -CpG island hypermethylation for PARVG and FGF19, in addition to the other three cancer-specific hypermethylated genes identified using our MBD-depletion microarray approach (PTPRN, COL11A2 and LTBP), was not an in vitro culture phenomenon of the transformed cells. The analyses of a comprehensive collection of primary human tumor samples (n ¼ 223) from the most common tissue types (breast, colon, lung, cervix, leukemia and lymphoma) demonstrated that promoter CpG island hypermethylation of these newly identified MBD targets is a common event in human cancer (Figure 6; Supplementary Table S5). Oncogene

Discussion The most widely studied epigenetic modification in humans is the cytosine methylation of DNA within the dinucleotide CpG. A total of 3–6% of all cytosines are methylated in normal human DNA (Jones and Laird, 1999; Herman and Baylin, 2003; Esteller, 2007). Potentially ‘methylable’ CpG dinucleotides are not randomly distributed in the human genome; instead, CpG-rich regions known as CpG islands, which span the 5-end region (promoter, untranslated region and exon 1) of many genes, are usually unmethylated in normal cells. This unmethylated status is linked to the ability of CpG island-containing genes to be transcribed in the presence of the necessary transcriptional activators. In cancer cells, the transcriptional silencing of tumor suppressor genes by CpG island promoter hypermethylation is key to the tumorigenic process, contributing to all of the typical hallmarks of a cancer cell (Hanahan and Weinberg, 2000) that result from tumor suppressor inactivation (Jones and Laird, 1999; Herman and Baylin, 2003; Esteller, 2007). Proteins of the MBD family are thought to be involved in promoter CpG island hypermethylation-associated silencing due to their ability to silence genes through the recruitment of HDAC and methyltransferase activities to methylated DNA (Jones et al., 1998; Nan et al., 1998; Wade et al., 1999; Fujita et al., 2003; Fuks et al., 2003). Despite the accumulated evidence demonstrating the presence of MBD proteins at CpG island-methylated promoters of tumor suppressor (Magdinier and Wolffe, 2001; Nguyen et al., 2001; Bakker et al., 2002; Koizume et al., 2002; Ballestar et al., 2003; Lopez-Serra et al., 2006), their direct involvement in the silenced gene status has not been functionally demonstrated. Here, we show the direct involvement of MBDs in the maintenance of the silenced status of a significant number of genes in transformed cells model. The removal of three MBD proteins (MeCP2, MBD1 and MBD2) singly or in combination results in the release of gene silencing mainly associated with DNA-methylated 50 -regulatory regions that were bound by the corresponding MBD(s) in untreated cells. Most importantly, the upregulation of these MBD-associated genes occurs without any change in the DNA methylation pattern of the underlying DNA sequence. MBD2 appears to be the MBD family member with the greatest effect on gene silencing. These results are in agreement with the preliminary findings of our group and others obtained using several models (Ballestar et al., 2003; Fraga et al., 2003; Sansom et al., 2003). In vitro, MBD2 is the MBD protein with the highest biochemical affinity for methylated DNA (Fraga et al., 2003). In vivo ChIP experiments with MBDs associated with genomic CpG island arrays (ChIP-on-CHIP) show that MBD2 binds to the largest number of CpG islands in breast cancer cells (Ballestar et al., 2003). Interestingly, crossing Mbd2-null mice with ApcMin colon adenoma-prone mice results in the inhibition of intestinal tumor development (Sansom et al., 2003), suggesting that the loss of MBD2 might ‘trigger’ a


PARVG

Actin

Actin

DN

Absorbance (595nm)

FGF19

pc

k oc M

pc

M

oc

k

DN

A3

A3

-P

AR

-F G F1

9

VG

3563 MTT assay 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0

empty vector PARVG

1

2

3

4

5

Empty vector PARVG+

FGF19+

Absorbance (595nm)

Days 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0

empty vector FGF19

1

2

3

4

5

Mean tumour volume (mm3)

Number of colonies

1000 400 800 600 400 200 0 Mock

PARVG

FGF19

PA R VG

M oc k

FG F1 9

M oc k

Days

1600 1400 1200 1000 800 600 400 200 0 Mock

PARVG

FGF19

Figure 5 Methyl-CpG-binding domain (MBD)-bound hypermethylated genes exhibit features of tumor suppressor genes. (a) g-Parvin (PARVG) and fibroblast growth factor 19 (FGF19) expression monitored by western blot in untransfected and transfected HeLa cells. (b) Effect of transfection of PARVG or FGF19 on the in vitro growth of HeLa cells using the 3-(4,5-dimethylthiazol-2-yl)-2,5diphenyltetrazolium bromide (MTT) proliferation assay over time. Both genes induce marked reduction. (c) Colony formation assay. Three independent experiments were developed. Example of the colony focus assay after a 2-week selection with G418. PARVG and FGF19 induce strong inhibition of colony formation. (d) Effect of PARVG and FGF19 transfection on the in vivo growth of HeLa xenotransplants in nude mice. Note the large tumor on the left flank, corresponding to empty vector cells and the presence of smaller tumors on the opposite flank, corresponding to PARVG- or FGF19-transfected cells. Tumor weight at the time of killing is shown.

comprehensive loss of silencing of hypermethylated tumor suppressor genes and thus induce inhibition of cell growth and transformation. Our results suggest that the removal of the MBD proteins associated with an expression microarray approach is also a useful strategy for identifying new gene targets undergoing DNA methylation-associated silencing in transformed cells. The use of candidate gene approaches (Magdinier and Wolffe, 2001; Nguyen et al., 2001; Bakker et al., 2002; Koizume et al., 2002; Lopez-Serra et al., 2006) and ChIP-on-CHIP (Ballestar et al., 2003) strategies had already established important genes whose transcriptional inactivation was mediated by MBD binding to the corresponding hypermethylated promoters, but we have gone one step further by

depleting the MBDs, the ‘interpreters’ of the DNA methylation signal itself. We have also shown that the CpG island hypermethylation of these newly identified genes is not a unique feature of HeLa cells, but is also common in human tumorigenesis, being found in cancer cell lines and human primary tumors from different organs and tissues. The list of epigenetically silenced genes revealed covers most of the disrupted pathways of transformed cells (Hanahan and Weinberg, 2000), such as signal transduction mediated by the families of tyrosine phosphatases and transforming growth factor-b and their regulators, exemplified by PTPRN (Lan et al., 1994) and LTBP3 (Yin et al., 1995), respectively, or cell matrix adhesion, exemplified by PARVG (Olski et al., 2001). In the case of the later gene and of FGF19 we Oncogene


3564 FGF19

PARVG

S1 S2 S3 S4 S6 S5 U M U M U M U M U M U M

S1 S2 S3 S4 S6 S5 U M U M U M U M U M U M

Cervix

Cervix

Breast

Breast

Colon

Colon

Lung

Lung

Leukemia

Leukemia

Lymphoma

Lymphoma NL U

IVD M

U

M

NL

H2O U

M

Control

U

M

IVD U

M

H2O U

M

Control

Figure 6 Methyl-CpG-binding domain (MBD)-bound hypermethylated genes contribute to human carcinogenesis. Detection of gparvin (PARVG) and fibroblast growth factor 19 (FGF19) 50 -cytosine-phosphate-guanine (CpG) island methylation in primary tumors from different organs and tissues (cervix, breast, colon, lung, leukemia and lymphoma) using methylation-specific PCR analysis. The presence of a PCR band under lanes M or U indicates methylated or unmethylated genes, respectively. Normal lymphocytes (NL) are used as positive controls for unmethylated DNA and in vitro methylated DNA (IVD) is used as positive control for methylated DNA.

provide further experimental evidence to demonstrate their features as tumor suppressor by showing how both induce inhibition of colony formation and block the growth of xenotransplanted tumors in nude mice. Thus, overall, we have demonstrated that MBD proteins exert a key role in the maintenance of the transcriptional silencing of those genes containing a 50 -hypermethylated CpG island, and that the use of epigenomic technologies combining removal of MBD transcripts and expression microarray approaches unveils new tumor suppressor genes undergoing epigenetic inactivation in transformed cells.

Materials and methods Human cancer cell lines and primary tumors HeLa cells were cultured in Dulbecco’s modified Eagle’s medium with 4.5 g l1 glucose and L-glutamine supplemented with 10% of fetal calf serum by Hyclone and 1% penicillin/ streptomycin in a humidified 37 1C, 5% CO2 incubator. All the remaining human cancer cell lines were grown as previously described (Lopez-Serra et al., 2006). HeLa cells were treated with 5-aza-2-deoxycytidine (1 mmol l1) for 72 h. All cell lines (n ¼ 23) were obtained from the American Type Culture Collection (Rockville, MD, USA). Tissue samples of human primary tumors (n ¼ 223) and corresponding normal tissues (n ¼ 13) were all obtained at the time of the clinically indicated procedures. RNA interference of MBD proteins MeCP2-, MBD1- and MBD2-specific small RNAi were designed and synthesized by Qiagen (Valencia, CA, USA). Two different RNAi duplexes, recognizing two different sequences, were used against each of the MBDs genes (Supplementary Table S6). As control, we used scramble RNAi (Qiagen). Transfections were carried out using Oncogene

oligofectamine (Invitrogen, Carlsbad, CA, USA) and cells were recollected 42 h after transfection. MeCP2, MBD1 and MBD2 content was analysed by western blotting, conventional reverse transcription–PCR (RT–PCR) and quantitative RT–PCR. Microarray analysis The profiles of gene expression were determined by a cDNA microarray (Ballestar et al., 2005) that contains 7237 sequencevalidated IMAGE clones, including 5253 clones representing known genes and the remaining 1984 clones representing expressed sequence tags (ESTs). Time-course experiments were performed using an extended version of the expression microarray in which cancer-related clones (n ¼ 2489) were printed twice. Total RNAs were converted to double-stranded cDNA using the superscript choice system (Life Technologies, Gaithersburg, MD, USA) using oligo-dT primer containing a T7 RNA polymerase promoter. Fluorescent first-strand cDNA is made in the presence of Cy5-dCTP (red) for the sample or Cy3-dCTP (green) for a universal RNA standard. Slides are simultaneously hybridized with labeled sample and standard. Slides are then scanned for Cy3 and Cy5 fluorescence using Scanarray 5000 XL (GSI Lumonics Kanata, Ontario, Canada) and quantified using the Quantarray software (GSI Lumonics) and/or GenePix Pro 4.0 software (Axon Instruments Inc., Union City, CA, USA). Data were preprocessed in the following way: (1) log-transformation to obtain symmetrical ratios, (2) replicate handling (removing inconsistent replicates and merging the remaining ones), (3) missing value management, (4) flat pattern filtering by standard derivation and (5) pattern standardization by subtracting the pattern average and dividing the values by the standard derivation. Genes with an average fold change of more than two were considered as differentially expressed between both groups of comparison. DNA methylation analysis We carried out bisulfite modification of genomic DNA as described previously (Lopez-Serra et al., 2006). We established the methylation status of gene promoters by PCR analysis of bisulfite-modified genomic DNA using two procedures. First,


3565 all genes studied were analysed by bisulfite genomic sequencing of their corresponding promoter CpG islands (Lopez-Serra et al., 2006). Both strands were sequenced. The second analysis used methylation-specific PCR analyses (Herman et al., 1996). Placental DNA treated in vitro with Sss I methyltransferase was used as positive control for all methylated genes. We designed all of the bisulfite genomic sequencing and methylation-specific PCR primers according to genomic sequences around presumed transcription start sites of investigated genes. Primer sequences are mentioned in Supplementary Table S6. PCR conditions for methylation analysis are available upon request. Chromatin immunoprecipitation assay Chromatin immunoprecipitation assays were performed as previously described (Ballestar et al., 2003; Lopez-Serra et al., 2006). Fixation was performed with 1% formaldehyde and sonication was optimized to obtain 300–1000 bp chromatin fragments. Antibodies against each of the MBDs studied were obtained from Abcam, Cambridge, UK. PCR amplification was carried in 25 ml with specific primers for each of the analysed promoters. For each promoter, the sensitivity of PCR amplification was evaluated on serial dilutions of total DNA collected after sonication (input fraction). After PCR amplifications of the input, as positive control, and the bound fraction for each antibody, samples were run in 2% agarose gels. Primer sequences are mentioned in Supplementary Table S6. RT–PCR expression analysis We reverse transcribed total RNA (2 mg) treated with DNase I (Ambion, Austin, TX, USA) using Oligo(dT) primer with SuperScript Reverse Transcriptase (Life Technologies). We used 100 ng cDNA for PCR amplification and amplified the candidate genes with multiple cycle numbers (20–35 cycles) to determine the appropriate conditions for obtaining semiquantitative differences in their expression levels. RT–PCR primers were designed between different exons to avoid residual genomic DNA amplification. Glyceraldehyde-3-phosphate dehydrogenase was amplified as internal control to test cDNA quality and loading accuracy. Primer sequences are mentioned in Supplementary Table S6. Cell transfection PARVG and FGF19 cDNAs were cloned into pCDNA3 expression vector (Invitrogen). Transfection of HeLa cell line was performed by cell electroporation (Gene Sensor II, Bio-Rad, Hercules, CA, USA) at 250 mV, 950 mF and maximal capacitance. After electroporation cells were cultured for 2 days in 20% fetal bovine serum medium and were then selected in complete medium supplemented with 1 mg ml1 G418. Expression of FGF19 and PARVG were tested by western blotting using antibodies raised against FGF19 (Upstate, Bedford, MA, USA, 1:500) and PARVG (courtesy of Dr Fa¨ssler laboratory, 1:2500), respectively.

Western blotting Total protein was separated on 10% SDS–polyacrylamide gel electrophoresis gel and blotted onto a polyvinylidene difluoride membrane of 45 mm pore size (Immobilon PSQ; Millipore, Bedford, MA, USA). The membrane was blocked in 5% milk phosphate-buffered saline with 0.1% Tween-20 (PBS-Tween) and immunoprobed with antibodies raised against PARVG and FGF19 The secondary antibodies used were rabbit antigoat conjugated to horseradish peroxidase (1:3000) and goat anti-rabbit horseradish peroxidase (1:3000) (both from Amersham Biosciences, Piscataway, NJ, USA), respectively. Colony formation and cell viability assays Colony formation assays were performed adding transfected cells to a medium containing 80% methylcellulose (StemCell Technologies, Vancouver, BC, Canada) and 20% conditioned medium from HeLa cell cultures, and 600 mg ml1 G418. The mixture was then placed in a six-well plate and incubated for 15 days. Colonies containing more than 20 cells were scored as positive. Cell viability was determined by the MTT assay. Aliquots of 1.5 104 cells were plated in 96-well microdilution plates. Following overnight cell adherence, experimental media containing the drugs or control media was added to appropriate wells. After 48 h, the media was replaced by drug-free fresh media (100 ml per well) containing 50 mg of MTT. After a 3 h incubation at 37 1C in 5% CO2 atmosphere, the MTT was removed and MTT formazan crystals were dissolved in dimethyl sulfoxide (100 ml per well). Absorbance at 570 nm was determined on an automatized microtiter plate reader. It was established that optical density was directly proportional to the cell number up to the density reached by the end of the assay. Mouse xenograft model Female athymic nude mice (6-week old) were used for tumor xenografts. Animals were randomly separated in three groups of seven specimens each (those injected with HeLa cells carrying the empty vector as a control, those injected with HeLa cells expressing FGF19 and those injected with HeLa cells expressing PARVG). Both flanks and a shoulder of each animal were injected subcutaneously with 106 (mock and FGF19 þ ) or 106 (PARVG þ ) cells in a total volume of 200 ml of PBS. Tumor development at the site of injection was evaluated daily. Mice were killed 28 days after injection. Acknowledgements We are grateful to Professor Reinhard Fa¨ssler for providing us with the PARVG antibody. The work was supported by the Health (FIS01-04) and Education and Science (I þ D þ I MCYT08-03, FU2004-02073/BMC and Consolider MEC0905) Departments of the Spanish Government, the European Grant Transfog LSHC-CT-2004-503438 and the Spanish Association Against Cancer (AECC). LL-S is a recipient of a BEFI predoctoral fellowship.

References Bakker J, Lin X, Nelson WG. (2002). Methyl-CpG binding domain protein 2 represses transcription from hypermethylated pi-class glutathione S-transferase gene promoters in hepatocellular carcinoma cells. J Biol Chem 277: 22573–22580. Ballestar E, Esteller M. (2005). Methyl-CpG-binding proteins in cancer: blaming the DNA methylation messenger. Biochem Cell Biol 83: 374–384. Ballestar E, Paz MF, Valle L, Wei S, Fraga MF, Espada J et al. (2003). Methyl-CpG binding proteins identify novel sites

of epigenetic inactivation in human cancer. EMBO J 22: 6335–6345. Ballestar E, Ropero S, Alaminos M, Armstrong J, Setien F, Agrelo R et al. (2005). The impact of MECP2 mutations in the expression patterns of Rett syndrome patients. Hum Genet 116: 91–104. Bernard D, Gil J, Dumont P, Rizzo S, Monte D, Quatannens B et al. (2006). The methyl-CpG-binding protein MECP2 is required for prostate cancer cell growth. Oncogene 25: 1358–1366. Oncogene


3566 Esteller M. (2007). Cancer epigenomics: DNA methylomes and histone-modification maps. Nat Rev Genet 8: 286–298. Fatemi M, Wade PA. (2006). MBD family proteins: reading the epigenetic code. J Cell Sci 119(Part 15): 3033–3037. Fournier C, Goto Y, Ballestar E, Delaval K, Hever A, Esteller M et al. (2002). Allele-specific histone lysine methylation marks regulatory regions at imprinted mouse genes. EMBO J 21: 6560–6570. Fraga MF, Ballestar E, Montoya G, Taysavang P, Wade PA, Esteller M. (2003). The affinity of different MBD proteins for a specific methylated locus depends on their intrinsic binding properties. Nucleic Acids Res 31: 1765–1774. Fujita N, Takebayashi S, Okumura K, Kudo S, Chiba T, Saya H et al. (1999). Methylation-mediated transcriptional silencing in euchromatin by methyl-CpG binding protein MBD1 isoforms. Mol Cell Biol 19: 6415–6426. Fujita N, Watanabe S, Ichimura T, Tsuruzoe S, Shinkai Y, Tachibana M et al. (2003). Methyl-CpG binding domain 1 (MBD1) interacts with the Suv39h1-HP1 heterochromatic complex for DNA methylation-based transcriptional repression. J Biol Chem 278: 24132–24138. Fuks F, Hurd PJ, Wolf D, Nan X, Bird AP, Kouzarides T. (2003). The methyl-CpG-binding protein MeCP2 links DNA methylation to histone methylation. J Biol Chem 278: 4035–4040. Guy J, Hendrich B, Holmes M, Martin JE, Bird A. (2001). A mouse Mecp2-null mutation causes neurological symptoms that mimic Rett syndrome. Nat Genet 27: 322–326. Hanahan D, Weinberg RA. (2000). The hallmarks of cancer. Cell 100: 57–70. Hendrich B, Bird A. (1998). Identification and characterization of a family of mammalian methyl-CpG binding proteins. Mol Cell Biol 18: 6538–6547. Hendrich B, Guy J, Ramsahoye B, Wilson VA, Bird A. (2001). Closely related proteins MBD2 and MBD3 play distinctive but interacting roles in mouse development. Genes Dev 15: 710–723. Hendrich B, Hardeland U, Ng HH, Jiricny J, Bird A. (1999). The thymine glycosylase MBD4 can bind to the product of deamination at methylated CpG sites. Nature 401: 301–304. Herman JG, Baylin SB. (2003). Gene silencing in cancer in association with promoter hypermethylation. N Eng J Med 349: 2042–2054. Herman JG, Graff JR, Myohanen S, Nelkin BD, Baylin SB. (1996). Methylation-specific PCR: a novel PCR assay for methylation status of CpG islands. Proc Natl Acad Sci USA 93: 9821–9826. Hutchins AS, Mullen AC, Lee HW, Sykes KJ, High FA, Hendrich BD et al. (2002). Gene silencing quantitatively controls the function of a developmental trans-activator. Mol Cell 10: 81–91. Jones PA, Laird PW. (1999). Cancer epigenetics comes of age. Nat Genet 21: 163–167. Jones PL, Veenstra GJ, Wade PA, Vermaak D, Kass SU, Landsberger N et al. (1998). Methylated DNA and MeCP2 recruit histone deacetylase to repress transcription. Nat Genet 19: 187–191. Jordan C, Li HH, Kwan HC, Francke U. (2007). Cerebellar gene expression profiles of mouse models for Rett syndrome reveal novel MeCP2 targets. BMC Med Genet 8: 36. Klose RJ, Sarraf SA, Schmiedeberg L, McDermott SM, Stancheva I, Bird AP. (2005). DNA binding selectivity of MeCP2 due to a

requirement for A/T sequences adjacent to methyl-CpG. Mol Cell 19: 667–678. Koizume S, Tachibana K, Sekiya T, Hirohashi S, Shiraishi M. (2002). Heterogeneity in the modification and involvement of chromatin components of the CpG island of the silenced human CDH1 gene in cancer cells. Nucleic Acids Res 30: 4770–4780. Lan MS, Lu J, Goto Y, Notkins AL. (1994). Molecular cloning and identification of a receptor-type protein tyrosine phosphatase, IA-2, from human insulinoma. DNA Cell Biol 13: 505–514. Lewis JD, Meehan RR, Henzel WJ, Maurer-Fogy I, Jeppesen P, Klein F et al. (1992). Purification, sequence, and cellular localization of a novel chromosomal protein that binds to methylated DNA. Cell 69: 905–914. Lopez-Serra L, Ballestar E, Fraga MF, Alaminos M, Setien F, Esteller M. (2006). A profile of methyl-CpG binding domain protein occupancy of hypermethylated promoter CpG islands of tumor suppressor genes in human cancer. Cancer Res 66: 8342–8346. Magdinier F, Wolffe AP. (2001). Selective association of the methylCpG binding protein MBD2 with the silent p14/p16 locus in human neoplasia. Proc Natl Acad Sci USA 98: 4990–4995. Nan X, Ng HH, Johnson C, Laherty CD, Turner BM, Eisenman RN et al. (1998). Transcriptional repression by the methyl-CpG-binding protein MeCP2 involves a histone deacetylase complex. Nature 393: 386–389. Ng HH, Jeppesen P, Bird A. (2000). Active repression of methylated genes by the chromosomal protein MBD1. Mol Cell Biol 20: 1394–1406. Ng HH, Zhang Y, Hendrich B, Johnson CA, Turner BM, ErdjumentBromage H et al. (1999). MBD2 is a transcriptional repressor belonging to the MeCP1 histone deacetylase complex. Nat Genet 23: 58–61. Nguyen CT, Gonzales FA, Jones PA. (2001). Altered chromatin structure associated with methylation-induced gene silencing in cancer cells: correlation of accessibility, methylation, MeCP2 binding and acetylation. Nucleic Acids Res 29: 4598–4606. Olski TM, Noegel AA, Korenbaum E. (2001). Parvin, a 42 kDa focal adhesion protein, related to the alpha-actinin superfamily. J Cell Sci 114: 525–538. Saito M, Ishikawa F. (2002). MBD3 and HDAC1, two components of the NuRD complex, are localized at Aurora-A-positive centrosomes in M phase. J Biol Chem 277: 35434–35439. Sansom OJ, Berger J, Bishop SM, Hendrich B, Bird A, Clarke AR. (2003). Deficiency of Mbd2 suppresses intestinal tumorigenesis. Nat Genet 34: 145–147. Wade PA, Gegonne A, Jones PL, Ballestar E, Aubry F, Wolffe AP. (1999). Mi-2 complex couples DNA methylation to chromatin remodelling and histone deacetylation. Nat Genet 23: 62–66. Yin W, Smiley E, Germiller J, Mecham RP, Florer JB, Wenstrup RJ et al. (1995). Isolation of a novel latent transforming growth factor-beta binding protein gene (LTBP-3). J Biol Chem 270: 10147–10160. Zhao X, Ueba T, Christie BR, Barkho B, McConnell MJ, Nakashima K et al. (2003). Mice lacking methyl-CpG binding protein 1 have deficits in adult neurogenesis and hippocampal function. Proc Natl Acad Sci USA 100: 6777–6782.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc).

Oncogene