An Essential Role for DNA Methyltransferase DNMT3B in Cancer Cell Survival*

Abnormal methylation and associated silencing of tumor suppressor genes is a common feature of many types of cancers. The observation of persistent methylation in human cancer cells lacking the maintenance methyltransferase DNMT1 suggests the involvement of other DNA methyltransferases in gene silencing in cancer. To test this hypothesis, we have evaluated methylation and gene expression in cancer cells specifically depleted of DNMT3A or DNMT3B,de novo methyltransferases that are expressed in adult tissues. Here we have shown that depletion of DNMT3B, but not DNMT3A, induced apoptosis of human cancer cells but not normal cells. DNMT3B depletion reactivated methylation-silenced gene expression but did not induce global or juxtacentromeric satellite demethylation as did specific depletion of DNMT1. Furthermore, the effect of DNMT3B depletion was rescued by exogenous expression of either of the splice variants DNMT3B2 or DNMT3B3 but not DNMT1. These results indicate that DNMT3B has significant site selectivity that is distinct from DNMT1, regulates aberrant gene silencing, and is essential for cancer cell survival.


From the Department of Molecular Biology, MethylGene Inc., Montreal H4S 2A1, Canada
Abnormal methylation and associated silencing of tumor suppressor genes is a common feature of many types of cancers. The observation of persistent methylation in human cancer cells lacking the maintenance methyltransferase DNMT1 suggests the involvement of other DNA methyltransferases in gene silencing in cancer. To test this hypothesis, we have evaluated methylation and gene expression in cancer cells specifically depleted of DNMT3A or DNMT3B, de novo methyltransferases that are expressed in adult tissues. Here we have shown that depletion of DNMT3B, but not DNMT3A, induced apoptosis of human cancer cells but not normal cells. DNMT3B depletion reactivated methylation-silenced gene expression but did not induce global or juxtacentromeric satellite demethylation as did specific depletion of DNMT1. Furthermore, the effect of DNMT3B depletion was rescued by exogenous expression of either of the splice variants DNMT3B2 or DNMT3B3 but not DNMT1. These results indicate that DNMT3B has significant site selectivity that is distinct from DNMT1, regulates aberrant gene silencing, and is essential for cancer cell survival.
DNA methylation is a highly plastic (1) and critical component of mammalian development (2,3). The maintenance DNA methyltransferase enzyme, Dnmt1, 1 and the de novo methyltransferases, Dnmt3a and Dnmt3b, are indispensable for development because mice homozygous for the targeted disruption of any of these genes are not viable (2,3). DNA methylation is also strongly implicated in tumorigenesis (4 -6). Dnmt1ϩ/Ϫ mice develop fewer precancerous intestinal lesions than Dnmt1 wild type animals when bred with adenomatous polyposis coli (APC) multiple intestinal neoplasia (Min) animals predisposed to this neoplasia (7). Elevated levels of DNMT1 are also required to silence p16ink4a in bladder cancer cells (8) and to maintain the phenotype of fibroblasts transformed with the Fos oncogene (9). Overexpression of DNMT3A, DNMT3B, and various DNMT3B splice variants has also been reported in tumor cells (10 -12); however, the extent to which they are involved in cancer remained to be investigated. In this report, we have investigated the role of DNMT3B in the aberrant methylation and inactivation of genes in human tumor cells as well as its role in the maintenance of the transformed phenotype.
Western Blot Analysis-Western blots were done following standard protocols. Preparation and use of polyclonal antibodies to DNMT1 were described previously (8). Antibodies raised in rabbits by the same method against purified glutathione S-transferase fused to residues 10 -118 of DNMT3A and 4 -101 of DNMT3B were used at a dilution of 1:2000. Horseradish peroxidase-linked antibodies (Sigma) and ECL (Amersham Biosciences) were used for chemiluminescent detection.
Northern Blot Analysis-Northern blot analysis using 10 -15 g of total RNA/sample was performed according to standard protocols. Detection of histone H4 employed an end-labeled antisense oligonucleotide probe as previously described (13). Detection of DNMT1, DNMT3A, or DNMT3B mRNA was done as previously described (11). Northern blots were scanned and quantified using an alpha imager (Alpha Innovotech).
Cell Cycle and TUNEL Analyses-Cells were fixed with paraformaldehyde and processed for TUNEL (TdT-mediated dUTP nick-end labeling) analysis using a fluorescein apoptosis detection system (Promega) or cell-death detection ELISA Plus kit (Roche Molecular Biochemicals) performed according to the manufacturers' protocols.
Cytoplasmic histone DNA assay was performed using the cell-death detection ELISA Plus kit (Roche) according to the manufacturer's protocol.
Southern Blot Analysis-Standard protocols were used. Briefly, 5 g of DNA were digested with either HpaII or MspI (Invitrogen) using the manufacturer's conditions. Samples were loaded on a 1% agarose gel and transferred to a Zeta-Probe nylon membrane (Bio-Rad). Blots were hybridized to a classic satellite 2 oligonucleotide as previously described (14,15).
RT-PCR Analysis-RT-PCR for RASSF1A was performed as described (16). As a sample loading control, 100 ng of the same reversetranscribed cDNA was used to amplify ␣-actin using the forward primer 5Ј-ACGAAACTACCTTCAACTCCATC-3Ј and the reverse primer 5Ј-TG-GTCTCAAGTCAGTGTACAGGT-3Ј. Specificity of RASSF1A amplification was verified by Southern blot using as a probe the 32 P-labeled internal oligonucleotide probe 5Ј-GCAACCTCTTCATGAGCTTG-3Ј.
Bisulfite Genomic Sequencing-Genomic DNA from lipofectin control and antisense-treated cells were bisulfite treated (Intergen Company) and used to amplify the RASSF1A promoter region as described in Ref. 16. PCR products were cloned into TA cloning vectors (Invitrogen) and sequenced.
* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
cDNA Expression Arrays-Atlas human cancer cDNA expression arrays (CLONTECH Laboratories, Inc.) were performed according to the manufacturer's protocols.
Radiolabeled [ 3 H]dCTP Extension Assay-DNA samples were digested with HpaII and AciI or with MspI (normalization control) and incubated with 1ϫ Taq buffer, 1.0 mM MgCl 2 , 0.25 unit of AmpliTaq (PerkinElmer), and [ 3 H]dCTP at 56°C for 1 h as described in Ref. 17. Samples were precipitated in trichloroacetic acid, applied to nitrocellulose filters, and processed for scintillation counting.

RESULTS AND DISCUSSION
To test the roles of the DNMT isoforms in cancer cells, we identified potent isotype-specific antisense inhibitors. We screened numerous antisense phosphorothioate oligodeoxynucleotides (AS p-ODN, 20 bases in length) complementary to 5Ј-and 3Ј-untranslated regions as well as coding sequences of the respective mRNA for the ability to specifically reduce the mRNA of the target isotype after 24 h of treatment. The location of the most potent of these inhibitors within their target mRNA is shown in Fig. 1A. Potency and dose dependence of the inhibition was assessed by treating human cancer cells growing in culture with increasing concentrations (0 -75 nM) of inhibitors DNMT1-AS1, DNMT3A-AS1, DNMT3B-AS1 (antisense oligonucleotides), or mismatch (MM) (Fig. 1B). All inhibitors had IC 50 values for mRNA inhibition of less than 50 nM. To determine the specificity of the inhibition, Northern blots with total RNA isolated from treated cells were hybridized with probes for the given target DNMT isoform as well as with probes for non-target DNMT mRNA and the non-target mRNA glyceraldehyde-3-phosphate dehydrogenase. Non-target mRNA were not affected by any of the inhibitors tested, and, as expected, mismatch control oligonucleotides had no effect on any of the mRNA examined (Fig. 1B, left panel). Several human tumor cell lines of various tissue origins were similarly tested with essentially identical results (data not shown). The DNA methyltransferases (DNMT1, DNMT3A, and DNMT3B) are S-phase-specific genes. Therefore, to control for possible indirect effects on DNMT3B mRNA levels, Northern blots with mRNA from DNMT3B-depleted A549 cells were also hybridized with probes for another S-phase-specific gene, histone H4. DNMT3B depletion had no effect on histone H4 expression or on DNMT1 mRNA levels (Fig. 1B, middle panel). The ability to deplete DNMT3B from two normal cells, human mammary epithelial cells (HMEC) and normal foreskin fibroblasts (MRHF), was also shown (Fig. 1B, right panel).
To demonstrate reduction in target protein levels, we raised antibodies specific for DNMT1 (8), DNMT3A, and DNMT3B. All antibodies were shown to specifically recognize target human proteins (expressed in baculovirus or mammalian systems) with high affinity; however, endogenous levels of DNMT3A protein were below the level of detection in all cell lines tested (data not shown). As expected from the mRNA inhibition, Western blot analysis with anti-DNMT1 and anti-DNMT3B antibodies on nuclear extracts prepared from cancer cells treated with increasing doses of DNMT1, DNMT3A, or DNMT3B inhibitors for 48 h showed selective dose-dependent inhibition of only the target DNMT isoform (Fig. 1C).
DNMT1 has been shown to be in a complex with proliferating cell nuclear antigen (PCNA), an auxiliary factor for DNA replication. This interaction can be disrupted by the cell cycle  (8) and newly raised anti-DNMT3B antibodies or ␣-actin-specific antibody to control for specificity of inhibition and protein loading. regulator p21 WAF1/CIP1 (18), suggesting that DNMT1 may play an active role in replication in addition to its role in regulation of gene expression. In fact, Dnmt1 mutant fibroblasts proliferate at only one-third the rate of control cells (19). To determine the effect of DNMT3A or DNMT3B depletion on the proliferation of human cancer cells, we treated lung cancer cells (A549) or breast cancer cells (MDA-MB-231) with increasing doses (0 -75 nM) of the inhibitors DNMT3A-AS1 or DNMT3B-AS1 over several days. A potent dose-dependent antiproliferative effect upon DNMT3B depletion was observed in both cancer cell lines (Fig. 2A). In contrast, DNMT3A depletion had no significant effect on the proliferation of either cell type ( Fig.  2A). The antiproliferative effect observed was independent of cellular p53 status because MDA-MB-231 cells harboring mutant p53 (20) and the p53 wild type A549 cell line (21) were both sensitive to DNMT3B depletion. In addition, the antiproliferative effect of DNMT3B depletion was also seen in T24 human bladder cancer cells that lack p53 protein (22) (data not shown) as well as the p53 wild type HCT116 human colon cancer cell line (23) (Fig. 2B, right panel).
Additional diversity within the DNMT3A and DNMT3B gene family exists because several splice variants for both have been identified in mouse and human tissues (10,11,24). These splice variants have the potential to encode proteins with altered activities. DNMT3B3, for example, lacks exons 10, 21, and 22 and yields a protein with altered spacing between conserved methyltransferase motifs VI and IX in the catalytic domain (11). DNMT3B2 lacks only exon 10, which lies outside the conserved catalytic domain. DNMT3B4 and DNMT3B5 give rise to truncated proteins without the conserved motifs IX and X (11), which could conceivably encode a dominant negative function. The conservation of these splice variants from mouse to human suggests that they encode important cellular functions. The antisense inhibitors described here target 3Ј-UTR sequences of DNMT3A and DNMT3B common to all the splice variants. Thus, the inhibitors tested lead to a depletion of all DNMT3A and DNMT3B mRNA and proteins (Fig. 1, B and C). To test whether the effects of DNMT3B depletion could be ascribed to ablation of a single DNMT3B splice variant, we evaluated the ability of individual, exogenously expressed DNMT3B splice variants to reverse or rescue the phenotype produced by depletion of all DNMT3B forms. It is important to note that the exogenously expressed DNMT isoforms do not contain the target sequence of the antisense inhibitors, which target sequences in the 3Ј-untranslated regions and thus do not compete for inhibitor and are not themselves inhibited. Several cell lines were tested for the ability to express sufficient levels of exogenous DNMT3B2, DNMT3B3, and DNMT1 after transient transfections. HCT116 cells were used because they were the only cell type found to express exogenous DNMT3B protein 48 h after transient transfection in a significant proportion of the population. Expression of the HA-tagged DNMT3B2, DNMT3B3, and DNMT1 proteins was confirmed 48 h after transfection of HCT116 cells by Western blot with both anti-HA and anti-DNMT3B antibodies (Fig. 2B, left panel). HCT116 cells were treated for 72 h with a combination of either DNMT3B-AS1 mRNA expression levels were assayed using RT-PCR analysis with primers specific for RASSF1A. As a loading control, PCR of the same reverse-transcribed cDNA with primers specific for ␣-actin was performed.
(75 nM) and the empty vector pCGN, or DNMT3B-AS1 (75 nM) and pCGN-DNMT3B2, pCGN-DNMT3B3, or pCGN-DNMT1 plasmids. After the initial treatment period, cells were left untreated for an additional 24 h at which point cell colonies were visualized by methylene blue staining. The effect of DNMT3B depletion was rescued by exogenous expression of either DNMT3B2 or DNMT3B3 but not by DNMT1 or vector sequences alone (Fig. 2B, right panel). These results demonstrate the selectivity of the inhibition for the DNMT3B enzymes and suggest that DNMT3B2 and DNMT3B3 perform similar roles in cancer cells clearly distinct from those of DNMT1.
In an attempt to identify the gene expression changes induced by DNMT3B depletion (48 h treatment), we performed cDNA microarray assays (Atlas human cancer arrays). DNMT3B depletion significantly altered the expression of a limited set of genes on the array (ϳ10 of 588 genes). The genes that showed the greatest fold changes are shown in Fig. 2C, left panel. The observed changes in gene expression are consistent with the inhibition of proliferation. Induction of caspases 9 and 10 is suggestive of induction of an apoptotic pathway. Western blot analysis confirmed the induction of p21 WAF1/CIP1 and suppression of PCNA at the protein level (data not shown). However, because these genes are not known to be regulated by methylation, these changes may either be secondary to the reactivation of methylation-silenced genes or may be regulated by the cellular levels of DNMT3B protein. In addition to the broad analysis of gene expression by cDNA arrays, we also began a systematic analysis of specific genes previously shown to be silenced by methylation at high frequencies in lung cancers. Promoter methylation and silencing of death-associated protein kinase is associated with non-small cell lung cancer (25). However, methylation-specific PCR analysis revealed that DAP kinase is not methylated (within the region of the MSP-PCR primers) in either A549 or MDA-MB-231 cells (data not shown). The RAS effector homologue RASSF1A is a candidate tumor suppressor gene that has very recently been shown to be inactivated by methylation in a high percentage of lung, breast, and renal tumors (16,26,27). Most importantly, RASSF1A has been shown to be silenced by methylation in both A549 and MDA-MB-231 cells (16). We examined whether DNMT3B depletion could lead to reactivation of RASSF1A. RT-PCR analysis demonstrated that RASSF1A expression was induced in a dose-dependent manner by DNMT3B depletion in both MDA-MB-231 (data not shown) and A549 cells (Fig. 2C, middle  panel), demonstrating the requirement of DNMT3B for active suppression of this gene in these cancer cells. Interestingly, higher doses of DNMT3B-AS1 were required to reactivate RASSF1A in A549 cells when compared with MDA-MB-231 cells. This is consistent with the observation that higher doses of 5-aza-dC are also required for its reactivation in A549 cells relative to MDA-MB-231 cells (16).
DNMT3A and DNMT3B have recently been shown to be transcriptional repressors that function in part by recruiting the chromatin remodeling histone deacetylase enzymes (HDAC) (28,29). Therefore, DNMT3B-mediated gene suppression may involve both methylation-dependent and methylation-independent HDAC-dependent mechanisms. To investigate the potential involvement of DNMT3B-associated HDAC activity in the reactivation of RASSF1A induced by DNMT3 depletion, we treated cells with increasing doses of the HDAC inhibitor trichostatin A (TSA) alone or with a combination of TSA and DNMT3B-AS1. Treatment of MCF-7 cells with TSA alone did not reactivate silenced RASSF1A but, as expected, DNMT3B depletion alone did (Fig. 2C, right panel). The combination of TSA and DNMT3B depletion resulted in higher levels of RASSF1A expression (Fig. 2C, right panel), suggesting that although the dominant mechanism of RASSF1A silencing in these cells is methylation-dependent, inhibition of HDAC activity cooperates with DNMT3B inhibition. Bisulfite genomic sequencing confirmed that the RASSF1A promoter was highly methylated in A549 cells and revealed demethylation of CpG sites within the RASSF1A promoter in DNMT3B-depleted A549 cells (Fig. 4A). The demethylation observed was limited to fewer CpGs compared than that generally seen by either 5-aza-dC-induced demethylation of CpG islands or by specific inhibition of the maintenance methyltransferase DNMT1 by antisense inhibitors, where demethylation of most CpG sites within hypermethylated CpG islands is seen (8). This suggests that DNMT3B differs from DNMT1 in terms of its site selectivity and regulates selected CpG sites within the RASSF1A promoter, whereas DNMT1 is required to maintain methylation of all CpG sites. However, because DNMT3B depletion resulted in significant antiproliferative activity, it remains possible that the most affected (demethylated) cells are in fact underrepresented due to selection.
Because expression of RASSF1A transcripts is associated with both loss of anchorage-independent growth and apoptosis of cancer cells (16,30), we investigated the possibility that cell death by apoptosis was a contributing factor to the observed antiproliferative effect. Flow cytometry studies on DNMT3Bdepleted A549 cells demonstrated altered cell cycle distributions with an accumulation of cells in G 1, a decreased S-phase population, and, most strikingly, a large increase in a sub-G 1 population, suggestive of apoptotic cells. Vehicle controls or DNMT3A-depleted cells showed no such alterations (Fig. 3A,  left panel). A549 cells treated for 24 h with DNMT3B inhibitors showed minimal effects, suggesting that prolonged DNMT3B depletion is required to elicit the apoptotic response. TUNEL assays (Fig 3A, right panel) and cytoplasmic histone-DNA assays (data not shown) confirmed that DNMT3B depletion induced apoptotic cell death of cancer cells. The induction of apoptosis was dependent on both the dose and the potency of the three separate DNMT3B inhibitors tested (data not shown). To determine whether normal cells exhibited sensitivity to DNMT3B depletion similar to that of the cancer cells, we treated HMEC and MRHF cells with DNMT3B inhibitors. DNMT3B depletion did not result in cell cycle alterations or induce apoptosis in these normal cells (Fig. 3B, left panel). Apoptosis was again seen by the cytoplasmic histone apoptosis assay in DNMT3B-depleted A549 cells (pos. control) but not in similarly treated normal cells (Fig. 3B, right panel). The cancer-specific apoptosis observed is consistent with a mechanism involving re-expression of genes aberrantly silenced in cancer cells. However, whether activation of the apoptotic program is the result of reactivation of a limited set of genes or is the net effect of reactivation of many silenced genes acting in concert will be of interest for future studies, as will be investigation of the tissue specificity of DNMT3B-suppressed genes.
Global demethylation of mouse fibroblasts can induce p53dependent apoptosis (19). Although the apoptosis described here is p53-independent, we determined the effect of DNMT3B depletion on global methylation levels and juxtacentromeric satellite methylation. Global methylation was evaluated by [ 3 H]dCTP single nucleotide extension assays on DNA from A549 cells treated with DNMT3B inhibitors for up to 4 days. DNMT3B depletion did not produce significant global demethylation at any of these time points (Fig. 4B, left panel). Juxtacentromeric classic satellite DNA is normally heavily methylated, but DNA from Dnmt3bϪ/Ϫ mice or humans with inactivating mutations in DNMT3B (chromosome instability and immunodeficiency syndrome (ICF) patients) show almost no methylation in this region (31,32). DNMT1-depleted cells showed strong demethylation of the classic satellite 2 region (Fig. 4B, right panel) similar to that produced by the global demethylator 5-aza-dC. In contrast, demethylation was not seen in DNMT3B-depleted cells (Fig. 4B, right panel). These results demonstrate that the apoptosis observed is not a consequence of genome-wide hypomethylation and that DNMT3B is required to establish, but not to maintain, methylation of classic satellites in cancer cells.
Because limited inhibition of DNMT3B in cancer cells produced primarily antiproliferative effects, whereas DNMT3B depletion with higher doses of DNMT3B-AS1 resulted in apoptosis, it is interesting to speculate that genes regulating the proliferation of cancer cells are more sensitive to DNMT3B depletion and those controlling survival pathways require greater levels of DNMT3B depletion.
At the time of submission of this paper, Rhee et al. (33) reported the findings on two DNMT3BϪ/Ϫ HCT116 colon cancer clones that they selected following two rounds of homologous recombination. They also demonstrated that DNMT3B, in cooperation with DNMT1, is required for optimal neoplastic growth. Consistent with our findings, the lack of DNMT3B did not induce demethylation of juxtacentromeric satellite sequences. In contrast, however, the two DNMT3BϪ/Ϫ clones did not show altered growth properties or apoptosis. This discrepancy most likely results from the differences in methodologies employed to deplete cellular DNMT3B levels. Gene disruption by homologous recombination requires two rounds of selection to yield DNMT3BϪ/Ϫ clones. This process would clearly select against cells that underwent apoptosis or had any other growth disadvantage, for example due to reactivation of tumor suppressor genes by demethylation. The method described in this study specifically depletes DNMT3B in an entire treated population of cells and therefore is not biased by selection. Moreover, because the gene-targeting strategy targets only 2 of the more than 20 exons of the DNMT3B gene, while the antisense inhibitors were entirely depleted of DNMT3B transcripts, it remains possible that other domains or alternatively spliced DNMT3B mRNA retain DNMT3B functions.
In summary, our results provide evidence that DNA methyltransferase isoforms display significant methylation site specificity and demonstrate that the de novo methyltransferase DNMT3B is required for the active suppression of genes and for the survival of cancer cells. These results also suggest that DNMT3B inhibitors may have potential as novel cancer therapeutics.