Nucleoside Drugs Induce Cellular Differentiation by Caspase-Dependent Degradation of Stem Cell Factors

Background Stem cell characteristics are an important feature of human cancer cells and play a major role in the therapy resistance of tumours. Strategies to target cancer stem cells are thus of major importance for cancer therapy. Differentiation therapy by nucleoside drugs represents an attractive approach for the elimination of cancer stem cells. However, even if it is generally assumed that the activity of these drugs is mediated by their ability to modulate epigenetic pathways, their precise mode of action remains to be established. We therefore analysed the potential of three nucleoside analogues to induce differentiation of the embryonic cancer stem cell line NTERA 2 D1 and compared their effect to the natural ligand retinoic acid. Methodology/Principal Findings All nucleoside analogues analyzed, but not retinoic acid, triggered proteolytic degradation of the Polycomb group protein EZH2. Two of them, 3-Deazaneplanocin A (DZNep) and 2′-deoxy-5-azacytidine (decitabine), also induced a decrease in global DNA methylation. Nevertheless, only decitabine and 1β-arabinofuranosylcytosine (cytarabine) effectively triggered neuronal differentiation of NT2 cells. We show that drug-induced differentiation, in contrast to retinoic acid induction, is caused by caspase activation, which mediates depletion of the stem cell factors NANOG and OCT4. Consistent with this observation, protein degradation and differentiation could be counteracted by co-treatment with caspase inhibitors or by depletion of CASPASE-3 and CASPASE-7 through dsRNA interference. In agreement with this, OCT4 was found to be a direct in-vitro-target of CASPASE-7. Conclusions/Significance We show that drug-induced differentiation is not a consequence of pharmacologic epigenetic modulation, but is induced by the degradation of stem-cell-specific proteins by caspases. Our results thus uncover a novel pathway that induces differentiation of embryonic cancer stem cells and is triggered by the established anticancer drugs cytarabine and decitabine. These findings suggest new approaches for directly targeting the stem cell fraction of human tumours.


Introduction
Cancer is widely considered a developmental disease, caused by the destabilisation of the balance of cellular proliferation and differentiation [1,2]. Tumours thus originate from cells with stem cell characteristics that have acquired aberrant gene expression patterns, mostly due to mutations and epimutations. These aberrant gene expression patterns lead to a block in differentiation and trigger uncontrolled proliferation. Loss of differentiation is thus an important characteristic of tumour cells and represents a defining feature of human cancers. As a consequence, differentiation therapy has been developed into an important approach for the treatment of cancers, particularly leukaemias [3,4].
Stem cell features are controlled by a small group of transcription factors, like NANOG and OCT4, that maintain the expression of stem cell genes and repress the induction of differentiation-specific genes by recruiting repressive complexes of the Polycomb Group (PcG) [5,6]. Differentiation-associated genes in mouse ES cells are repressed by PcG proteins and are often also marked by DNA methylation [7]. These repressive marks are successively lost during differentiation. A prominent example is the loss of PcG repression at the HOXA gene cluster during neuronal differentiation [8,9]. It has also been shown that promoters of lineage-specific genes become methylated during differentiation, suggesting context-dependent interactions between DNA methylation and Polycomb repression [10]. A key component of PcG repression is the histone methyltransferase EZH2 (enhancer of zeste homolog 2), the enzymatic core component of the Polycomb repressive complex 2 (PRC2). This protein creates specific trimethylation patterns of lysine 27 of histone H3 (H3K27me3), which leads to concomitant transcriptional silencing [11,12].
The characterization of drugs that modulate epigenetic processes and induce differentiation in human cancer cells represents an important aspect in the development of epigenetic cancer therapies. Retinoic acid (RA), which induces differentiation in many stem cell populations, was among the first substances used for differentiation therapy [3]. The seminal finding that the differentiation-inducing cytosine analogue 29-deoxy-5-azacytidine (decitabine, DAC) acts as an effective inhibitor of DNA methyltransferases provided an important link between cellular differentiation and epigenetic regulation [13]. Another example is 3-Deazaneplanocin A (DZNep), which has been shown to cause proteolytic degradation of PRC2 components, to influence histone modification patterns and to induce moderate differentiation effects in acute myeloid leukaemia cells [14][15][16][17]. DZNep was originally synthesised as an inhibitor of S-adenosylhomocysteine hydrolase (SAHH), a key enzyme in S-adenosylmethionine (SAM) dependent methylation processes [18]. Treatment of MCF-7 breast cancer cells with DZNep led to the derepression of a defined set of Polycomb targets, which again suggested that the compound might also induce cellular differentiation [14]. While these findings proposed a close connection between epigenetic modulation and drug-induced differentiation, cytarabine (1 b-arabinofuranosylcytosine, araC), a cytosine analogue closely related to decitabine, effectively induces differentiation, without inhibiting DNA methylation [19]. Both decitabine and cytarabine have been shown to be effective in the treatment of myeloid leukaemias, a group of diseases that is characterised by a differentiation block of precursor cells [20]. However, it is still not clear how these substances induce cellular differentiation and whether the inhibition of epigenetic modifiers plays a significant role in these mechanisms.
Over the past few years, the human embryonic teratocarcinoma cell line NTERA2 D1 (NT2) has been established as an intriguing human cancer stem cell model and represents a valuable tool for the analysis of the mechanisms regulating cellular differentiation. We have used NT2 cells, which can be induced to differentiate with natural ligands, like retinoic acid [21], to characterise the differentiation-inducing mechanisms triggered by RA, araC, DAC and DZNep. The three nucleoside analogues caused degradation of EZH2, but drug-induced differentiation could be observed only for araC and DAC. The latter drugs became integrated into DNA and induced DNA damage, which triggered the caspasedependent degradation of NANOG and OCT4. Our results suggest that drug-induced differentiation is not a consequence of pharmacologic inhibition of DNA methylation and/or histone methylation but caused by the direct degradation of stem cell factors. Our study uncovers a novel mechanism mediating cellular differentiation by nucleoside drugs and suggests that cytarabine and decitabine might be useful for targeting cancer stem cells.

DAC and araC induce differentiation of NT2 EC cells
Embryonal carcinoma (EC) cells derived from teratocarcinomas show many features of pluripotent cells and are often considered to be the malignant counterpart of human embryonic stem cells [22]. The EC cell line NTERA D1 (NT2) is a subclone of the line TERA2, originally derived from a lung metastasis associated with a testicular germ cell tumour that was established from a nude mouse xenocraft [23]. NT2 cells have not only been shown to differentiate along the neuronal linage, but also have mesodermal and ectodermal lineage potential. NT2 cells thus represent a valuable stem cell and cancer stem cell model [22,24]. Morphological neuronal differentiation of NT2 cells with RA usually requires several weeks of treatment [21]. Nevertheless, gene expression patterns change rapidly, and strong derepression of the anterior part of the PcG-repressed HOXA cluster becomes detectable after a few hours and is accompanied by significant epigenetic changes [8,9].
We assumed that activation of differentiation in NT2 cells would require substantial epigenetic modulation, especially changes in PcG presence, histone modification patterns and also DNA methylation.
We thus chose DZNep, a drug that has been shown to cause degradation of the PcG protein EZH2 [14], and the established DNA methylation inhibitor DAC as potential drugs to induce differentiation. In addition, we used araC, an inhibitor of mitosis and potent inducer of apoptosis [25], as a structurally related drug with no direct epigenetic activity, and retinoic acid as a natural differentiation-inducing agent. All three nucleoside drugs showed similar, concentration-dependent cytotoxicity on NT2 cells and on other human cancer cell lines ( Figure S1). For further treatments of NT2 cells we thus used drug concentrations equitoxic to the DZNep concentration established previously [5 mM, ref. 14], i.e. 20 mM for araC and 1 mM for DAC. Interestingly, we observed a neuron-like morphology in NT2 cells following incubation with araC and DAC already after 3 days, whereas treatment with DZNep and RA only led to weak or no morphological differentiation effects ( Figure 1A). RT-PCR analysis of two neuronal markers known to be prominently induced in differentiating NT2 cells [26][27][28], NEFL (neurofilament light polypeptide) and SNAP25 (synaptosomalassociated protein 25), demonstrated robust induction of both genes following treatment with araC or DAC ( Figure 1B). DZNep and RA caused only minor activation of these genes, which is consistent with the absence of neuronal morphology after 3 days of treatment with both substances. These results establish araC and DAC, but not DZNep, as strong inducers of neuronal differentiation in NT2 cells.

Analysis of drug-dependent changes in PcG repression
In order to identify the mechanisms that lead to drug-induced differentiation, we first analysed the effects on PcG repression. PRC2 degradation has been demonstrated for DZNep-treated MCF7 breast cancer cells [14] and the release from PcG repression is a molecular hallmark of differentiation [8,9]. All three nucleoside drugs caused degradation of EZH2 after three days of treatment at equitoxic concentrations whereas RA had only a weak effect (Figure 2A). EZH2 degradation by these compounds could also be observed in other human cancer cell lines, although to varying degrees ( Figure S2A). Drug treatment did not significantly affect the level of EZH2 mRNA expression ( Figure 2B), which confirmed that drug-dependent EZH2 degradation is regulated at the protein level [14].
It has been suggested that DZNep-mediated EZH2 degradation in MCF-7 cells might be related to the drug's ability to inhibit Sadenosylhomocyteine hydrolase (SAHH), a key enzyme in Sadenosylmethionine dependent methylation processes [14]. In order to determine the requirement of SAHH activity for EZH2 stability, we used siRNAs to efficiently deplete SAHH in NT2 cells. Subsequent Western analysis did not indicate any changes in the EZH2 protein level and in H3K27 trimethylation ( Figure 2C). Similar results were also obtained in MCF-7 cells ( Figure S2B). These findings indicate that the pharmacologic activity of DZNep is not limited to the inhibition of SAHH and that the molecular mechanisms triggering DZNep-induced EZH2 degradation remain to be established.
The 59 genes of the HOXA cluster are among the first genes to be activated by RA and show rapid loss of PcG complexes and H3K27me3 upon in vitro differentiation [8,9]. We therefore determined the expression levels of HOXA genes upon drug treatment. As expected, the natural ligand RA triggered strong expression in the 59-region of the cluster ( Figure 3A). DAC and araC also caused significant gene activation in a concentrationdependent manner, whereas DZNep-induced changes were relatively minor ( Figure 3A and Figure S3A). These findings, together with the effects on EZH2 stability, suggested that RA, araC and DAC induce cellular differentiation independent of EZH2 degradation. This was confirmed by siRNA-mediated knockdown of EZH2, which only resulted in a very weak activation of the HOXA cluster ( Figure 3A and S3B), indicating that EZH2 degradation is not sufficient for cellular differentiation.
Reactivation of HOXA genes by RA is accompanied by epigenetic changes in their histone modification patterns [8,9]. Cytarabine and DAC have not been reported to affect global histone modifications. In contrast, DZNep has been shown to cause genome-wide loss of H3K27me3 [14] and additional repressive and active histone modifications [16]. We therefore analysed the effect of drug treatment on the histone methylation pattern of the HOXA1 promoter by chromatin immunoprecipitation (ChIP). In control cells this region is marked by H3K27me3 and lacks the activating mark H3K4me3 (trimethylation of lysine 4 of histone H3), consistent with its silent state ( Figure 3B). When HOXA1 was induced by RA, the promoter was marked mainly by H3K4me3. DZNep did not cause any significant changes, whereas the HOXA1-activating drugs araC and DAC caused a substantial increase in H3K4me3, in parallel with a reduction in H3K27me3 levels ( Figure 3B). This indicates that drug-induced activation of HOXA1 is accompanied by histone modification changes, reflecting the loss of PcG repression at this target gene.

Analysis of drug-dependent DNA methylation changes
To further elucidate the pathways mediating drug-induced differentiation we also investigated DNA methylation. Previous studies have linked altered HOX gene expression patterns to changes in the DNA methylation status [29][30][31]. In addition, it has been suggested that DZNep might cause concomitant DNA methylation changes [14]. Lastly, DAC is an established inhibitor of DNA methylation and inhibition of DNA methylation has also been associated with cellular differentiation [13]. We therefore compared DZNep, araC, DAC and RA in their ability to alter HOXA methylation in NT2 cells. Methylation analysis by COBRA indicated dense DNA methylation of several HOXA CpG islands, some of which were demethylated upon incubation with DZNep or DAC, but not with araC or RA ( Figure 4A). Similar results were also obtained using capillary electrophoresis to measure global cytosine methylation. Untreated NT2 cells showed a methylation level of 4.3%, which was not significantly altered after treatment with araC or RA ( Figure 4B). However, treatment with DAC or DZNep significantly reduced global DNA methylation levels to 2.4% and 3.2%, respectively ( Figure 4B). These results provide important confirmation for the DNA methylation inhibiting activity of DZNep [14,16]. However, DZNep did not cause degradation of DNA methyltransferase 1 (DNMT1) ( Figure 4C), which represents the main pathway for DAC-induced DNA demethylation [32]. Interestingly, depletion of SAHH by RNA interference (RNAi) also led to significant demethylation of HOXA CpG islands ( Figure 4D and Figure S3C), which suggests that the demethylating activity of DZNep is a consequence of the inhibition of S-adenosylhomocysteine hydrolase. Our results thus confirm the demethylating activity of DZNep and suggest that DZNep-induced DNA demethylation is not related to DNMT1 degradation. Nevertheless, the lack of detectable araC-induced methylation changes suggests that inhibition of DNA methylation is not required for HOXA derepression and cellular differentiation.

Drug-induced differentiation of NT2 cells requires degradation of stem cell factors
In our previous experiments we found differentiation by RA, araC and DAC to be accompanied by a loss of PcG-repression at the HOXA-cluster. However, RA did not trigger EZH2-depletion and EZH2-degradation by DZNep did not lead to HOXAactivation. This suggests that the drug-induced loss of PcG repression is caused by an upstream event. A recent study described the CASPASE-3-dependent degradation of the stem cell-specific transcription factor NANOG as important for RAinduced differentiation of human ES cells [33]. NANOG and the core transcription factor OCT4 maintain the expression of stem cell genes, and repress the induction of differentiation-specific factors [5]. A reduction of NANOG and OCT4 protein levels is thus a prerequisite for the induction of differentiation.
Both araC and DAC, but not DZNep [14], are incorporated into DNA and have been shown to induce DNA damage, apoptosis and proteolytic cascades [25]. Consistently, we observed a robust accumulation of the DNA-damage marker cH2AX in cells treated with araC and DAC ( Figure 5A). As DNA-damage is generally a trigger for caspase activation [34,35], we next measured the enzymatic CASPASE-3/7 activity of drug-treated cells. Indeed, treatment with araC and DAC led to a substantial increase in caspase activity, while RA treatment had only a moderate effect and DZNep did not lead to clear caspase activation ( Figure 5B). Of note, CASPASE-3/7 activation was paralleled by an effective depletion of NANOG and OCT4 ( Figure 5C and S4A). DZNep had no effect on cH2AX, NANOG and OCT4, whereas RA-treatment did not lead to an increase of cH2AX, but also reduced both stem cell factors ( Figure 5A, 5C and S4A). The lack of OCT4 reduction in DZNep-treated NT2 cells indicates that the mild morphological changes observed ( Figure 1A) are not due to differentiation processes. Downregulation of NANOG and to lesser extent OCT4 by araC and DAC was observed also at the mRNA level, while RA led to a very pronounced transcriptional repression of both genes ( Figure 5D).
To further distinguish between protein degradation and regulation on the RNA level, we established a transfected NT2 cell line expressing EGFP under the control of the OCT4 regulatory region [36]. Untreated cells strongly expressed EGFP, whereas treatment with retinoic acid for three days led to a significant reduction of the EGFP-signal ( Figure 5E), indicating that OCT4 expression is silenced on the mRNA level. Treatment with araC for the same period of time resulted in cellular differentiation, but the cells retained significant EGFP expression ( Figure 5E), which is consistent with OCT4 downregulation at the protein level.
In order to test if direct depletion of OCT4 and NANOG induces differentiation of NT2 cells, we reduced the protein levels of OCT4, NANOG, EZH2, DNMT1 and SAHH by dsRNA interference and monitored neuronal morphology and expression mRNA expression levels in NT2 cells upon 3 days of drug treatment do not change significantly, as determined by quantitative RT-PCR. qRT-PCR measurements with at least three biological replicates were internally normalised to the corresponding lamin-b and b-actin expression levels. Deoxycytidine (dC) was included as a mock treatment control. (C) DZNep-mediated EZH2 depletion is not caused by SAHH inhibition. Western blot showing the effect of SAHH depletion (kd SAHH) on EZH2 and H3K27me3 levels in NT2 cells. Cells were treated for 72 hours with siRNAs. Knockdown of EZH2 (kd EZH2) was used as positive control, b-actin and histone H3 (H3) were used as loading controls. A mix of scrambled siRNAs was used as negative control (kd cont.). Reduction of SAHH-levels has no effect on EZH2 stability or H3K27me3 levels. doi:10.1371/journal.pone.0010726.g002 of SNAP25. Depletion of SAHH, DNMT1, EZH2, NANOG did induce minor changes in SNAP25 expression, but only the knockdown of OCT4 clearly induced neuronal differentiation, and, in parallel, significantly increased expression of SNAP25 ( Figure S5A and S5B). Combined knockdown of OCT4 and NANOG or OCT4, NANOG and EZH2, did not enhance the effect. NANOG expression is already strongly reduced in cells depleted for OCT4 ( Figure S5B), as OCT4 is a major activator of the NANOG gene [5]. This also explains the lower levels of NANOG transcription upon drug treatment ( Figure 5D). Nevertheless, depletion of stem cell factors, in particular OCT4, by RNAi shows similar effects on NT2 cells as treatment with araC and DAC and is sufficient to induce differentiation.
To confirm the involvement of proteolytic pathways in druginduced NANOG and OCT4 degradation, we analysed the expression levels of several key regulatory factors of these pathways in a time course experiment by RT-PCR. As expected, we observed a significant induction of the CASPASE-7 gene by araC and DAC, but also RA ( Figure 6). Cytarabine and DAC also triggered moderate upregulation of CASPASE-3 and the c-Jun Nterminal kinase II (JNKII). This was paralleled by the expression of the differentiation marker SNAP25 ( Figure 6). We observed no significant changes in gene expression for the apoptotic factor p53, for c-Jun N-terminal kinase I (JNKI), CASPASE-9 and EZH2 ( Figure 6). Thus, CASPASE-7, and, to a lesser extent, CASPASE-3, seem to be critical factors in araC-and DAC-induced degradation processes.

Drug-induced differentiation is caused by CASPASE-7dependent degradation of OCT4
In order to further prove the role of caspases in drug-induced differentiation, we combined nucleoside drug treatment with caspase inhibitors (Inhibitor I and II) and a JNKII inhibitor (JNK Inhibitor I). These substances caused a significant reduction of the ability of araC or DAC to activate HOXA1 and SNAP25, whereas RA-mediated induction of these genes and CASPASE-7 expression was not significantly affected ( Figure 7A). In addition, all three inhibitors partially rescued OCT4-and EZH2-degradation  Figure  S3B). Deoxycytidine (dC) was included as a mock treatment control. (B) Chromatin immunoprecipitation (IP) analysis of the HOXA1 promoter region in RA-, drug-treated and control NT2 cells. IP was repeated at least three times with chromatin from biological replicates using antisera specific for H3K27me3 and H3K4me3. Immunoprecipitated DNA was analyzed by real time PCR and a primer specific for the HOXA1 promoter region. Enrichments are shown as percentage of the total input. doi:10.1371/journal.pone.0010726.g003 induced by araC ( Figure S4B). No such effect could be observed for the reduction of OCT4 upon RA-treatment ( Figure S4C), which is consistent with the notion that OCT4 is downregulated on the RNA level in RA-treated NT2 cells. DZNep has been shown to trigger proteasomal degradation of PRC2 components [14]. We thus also employed the proteasome inhibitor Lactacystin [37]. We observed no an effect on araC-or DAC-induced expression of HOXA1 or SNAP25 ( Figure S6A) and Lactacystin did not rescue araC-induced OCT4 or EZH2 degradation ( Figure  S6B and S6C). In contrast, DZNep-induced EZH2 degradation was affected by Lactacystin ( Figure S6C), which is in line with the established role of DZNep in inducing proteasome degradation [14].
The rescue effects exerted by caspase and JNK inhibitors on araC-induced differentiation could also be observed morphologically ( Figure 7B). Combinatorial treatments with araC and caspase inhibitors led to clearly less differentiated cells compared to araC-treatment alone. Similar effects were observed when CASPASE-3 and CASPASE-7 were depleted by corresponding siRNAs in araC-treated cells ( Figure S7A). NT2 cells induced to differentiate with araC showed significantly less differentiated cells when CASPASE-7 or both caspases were knocked down after 6 hours of araC-treatment and subsequent incubation for 48 hours ( Figure S7A). In addition, combined knockdown of CASPASE-3 and -7 in araC-induced NT2 cells led to significantly lower activation of HOXA1 and SNAP25 ( Figure S7B) and also partially restored OCT4 protein levels in araC-treated NT2 cells ( Figure S7C).
In a final set of experiments we also tested if active recombinant human CASPASE-7 or CASPASE-3 is able to degrade OCT4 or NANOG in nuclear extracts of NT2 cells in vitro. As shown in Figure 7C, treatment with increasing amounts of CASPASE-7 led to a significant reduction of OCT4 signals, whereas similar treatment with CASPASE-3 had no effect on OCT4 stability ( Figure 7C). In contrast, NANOG is degraded under similar conditions by both caspases, with CASPASE-3 showing a significantly stronger effect ( Figure 7C). These experiments confirm that OCT4 and NANOG are direct substrates for CASPASE-7 and support published data showing that NANOG is targeted by CASPASE-3 [30].
Taken together, we identified CASPASE-7 and, to a lesser extent, CASPASE-3 as important mediators of araC-and DAC- induced OCT4 and NANOG degradation and drug-induced differentiation.

Discussion
Differentiation-inducing compounds represent important tools for molecular biology and promising drugs for cancer therapy [4,19,38]. In this study, we compared the differentiation-inducing mechanism of three compounds -DZNep, cytarabine and decitabine -in a cancer stem cell model and identified a novel pathway for drug-induced differentiation. Importantly, our results show that drug-induced differentiation is not a consequence of pharmacologic inhibition of DNA methylation and/or histone methylation, but is triggered directly by the CASPASE-3/7dependent degradation of stem cell specific proteins. In particular, we identified CASPASE-7 as a central enzyme specifically inducing OCT4 depletion. Cytarabine and decitabine have previously been reported to induce DNA damage and to trigger caspase-and JNK-dependent apoptotic pathways [25,[39][40][41][42]. Consistent with this notion, we can reverse the differentiation-associated effects caused by araC and DAC by using inhibitors of these pathways or by reducing levels of CASPASE-3 and CASPASE-7 by dsRNA interference.
The potential of cytarabine to induce differentiation has been known for a long time. The most prominent example is the use of the drug for the differentiation of bone marrow progenitor cells in leukaemias [4,43]. Decitabine has first been described to induce differentiation more than 25 years ago in a landmark study linking DNA demethylation to cellular differentiation [13]. Both cytarabine and decitabine are incorporated into DNA and lead to double strand breaks (as monitored by the increase of hyperphosphorylated H2AX). Double strand breaks lead to the release of mitochondrial cytochrome c into the cytosol, which triggers apoptosome formation and mediates the activation of caspases-9, -3 and -7 [34,35]. A recent publication describes that araC-induced differentiation of acute myeloid leukaemia cells positive for oncogenic RAS depends on p53 [44]. NT2 cells express p53 [45] and, even if we observed no effect of araC-or DAC-treatment on p53 expression, we can not rule out a role for p53 in druginduced differentiation of NT2 cells. p53 is generally thought to  . Diagrams show fold differences compared to araC-, DAC-and RA-treatment alone. Cotreatment with caspase inhibitors leads to reduced gene expression after araC and DAC treatment. All qRT-PCR measurements were repeated at least three times and internally normalised to the corresponding lamin-b and b-actin expression levels. P-values (Student's t-test) for the expression differences between only drug-treated (2) and co-treated cells for highly significant cases (p#0.01) are shown. Asterisks indicate expression values trigger DNA repair and to halt the cell cycle upon DNA damage [34]. Nevertheless, p53 function can differ from cell type to cell type and accumulation of p53 can also activate pro-apoptotic genes, which finally leads to caspase activation [34].
We had initially speculated that the drug-induced degradation of EZH2 alone could lead to derepression of differentiationspecific genes in embryonic cancer stem cells. However, DZNep efficiently degraded EZH2 (without causing DNA damage in NT2 cells) and did not induce robust differentiation in our assays. The exact mechanism of DZNep-dependent degradation of EZH2 remains to be elucidated. It was assumed that DZNep-mediated inhibition of S-adenosylhomocysteine hydrolase (SAHH) leads to the proteasomal degradation of EZH2 [14]. However, we found that efficient knockdown of SAHH had no effect on EZH2 stability or H3K27 trimethylation. This indicates that the pharmacologic activity of DZNep is not limited to the inhibition of SAHH and that the molecular mechanisms triggering DZNep-induced EZH2 degradation remain to be established. Also, direct reduction of EZH2 by RNA interference did not trigger differentiation, induced only very weak HOXA expression in NT2 cells and also did not enhance the effects of OCT4 and NANOG depletion ( Figure S5). It is possible that repressive PRC1 complexes remain attached to their target regions even in the absence of EZH2. Alternatively, reduced EZH2-repression could be compensated by its functionally redundant homologue EZH1 [46,47]. Of note, our experiments confirm and extend earlier observations for a DNA demethylating activity of DZNep [14]. This activity could indeed be related to the DZNep-mediated inhibition of SAHH [18], as suggested by our finding that depletion of SAHH by RNA interference led to significant demethylation of HOXA-CpGislands.
Degradation of OCT4 and NANOG (detectable already after 24 hours) clearly preceded EZH2 degradation, HOXA activation and upregulation of differentiation markers. Thus, degradation of PRC2 components is most likely a consequence of the depletion of stem cell markers, but nevertheless necessary for the subsequent differentiation-specific epigenetic changes and the progression of differentiation. As we show for retinoic acid, natural differentiation is slow, even if OCT4 and NANOG are rapidly downregulated. This downregulation, however, appeared mainly to be caused by transcriptional repression. EGFP expression under the control of the OCT4 regulatory region became clearly reduced after the addition of RA, whereas expression persisted in araC-treated cells, where transcriptional repression of the OCT4 gene is much slower than proteolytic degradation of the gene product. In addition, RAinduced OCT4 downregulation could not be rescued by caspase inhibitors. In line with these findings, OCT4-degradation was also not observed in RA-treated mouse embryonic stem cells [33]. We found that RA-induced CASPASE-7-expression was not paralleled by a similar increase in enzymatic activity, indicating that the enzyme remained mostly inactive, which is also consistent with the absence of EZH2 degradation in RA-treated cells at this time point. Significantly reduced EZH2-levels could only be observed after 2 weeks of RA-treatment ( Figure S3D). Thus, PcGdependent repression is only slowly removed in naturally differentiating NT2 cells, leading to a slow activation of neuronal markers like NEFL and SNAP25. In summary, it appears plausible that efficient and rapid differentiation, as induced by araC and DAC, requires the concomitant proteolytic degradation of stem cell factors and PcG proteins. Notably, the NANOG gene is activated by OCT4 [5] and reduced OCT4 will therefore also lead to reduced NANOG levels, as observed in our experiments.
In contrast to NANOG, OCT4 has so far not been described as a caspase substrate and is not cleaved by CASPASE-3 in vitro [33].
Our results indicate that CASPASE-7, but not CASPASE-3, can degrade OCT4 in vitro. This is in line with accumulating recent evidence showing that these two enzymes are functionally distinct [35,48]. The CASVM web server for support-vector-machinesbased prediction of caspase cleavage sites [49] predicts 5 cleavage sites in OCT4 ( Figure S8). When overexposing the blots shown in Figure 7C, we observed a band of about 30 kD, which showed increased intensity in the lanes with the CASPASE-7-treated extracts ( Figure S8C). As we observed this band only with an OCT4-antibody directed against an N-terminal peptide of the protein, but not with a C-terminal antibody (data not shown), it most likely represents a caspase-dependent N-terminal degradation product of OCT4. A proteolytic cut at the predicted P1-site 272 within the homeodomain of OCT4 would result in a Nterminal fragment of 30 kD ( Figure S8B). This potential CASPASE-7 cleavage site needs to be confirmed in future experiments using recombinant OCT4 proteins bearing specific point mutations at this recognition sequence.
Stem cell characteristics are increasingly recognised as an important feature of human cancer cells and it is now widely assumed that a stem cell population plays a major role in the therapy resistance of tumours [50][51][52]. Strategies that target the stem cell properties of cancer cells will thus be critically important for translating this concept into therapeutical applications [3]. Our results uncover a novel pathway that triggers differentiation of embryonic cancer stem cells by the established anticancer drugs cytarabine and decitabine. These findings not only shed light on the mechanistic background of the biological effects caused by these drugs, but also show that cytarabine and decitabine might be useful in directly targeting the stem cell fraction of human tumours.

Materials and Methods
Cell culture and drug treatment that are significantly different from controls. (B) Microscopic images (106 magnification) of NT2 control cells (cont.) and NT2 cells treated for 72 h with 1, 3 and 10 mM araC (upper panels) and NT2 cells co-treated for 72 h with a mixture of all three inhibitors (lower panels) are shown. The araCinduced neuron-like morphology can be reverted to some degree by co-treatment with caspase and JNK inhibitors. (C) Western blots showing the levels of OCT4 and NANOG in nuclear extracts from NT2 cells without treatment (cont.), in nuclear extracts incubated at 37uC alone in the presence of 1 (+) or 2 (++) units of active recombinant human CASPASE-7 or CASPASE-3. b-actin was used as loading control. Antibodies specific for N-terminal peptides of OCT4 and NANOG were employed. OCT4 signals are efficiently reduced only by CASPASE-7, whereas NANOG is degraded by both caspases. doi:10.1371/journal.pone.0010726.g007 1 mM DAC, 20 mM araC, 20 mM dC and 5 mM DZNep (if not otherwise indicated). Substances, from aqueous stock solutions (5 or 50 mM), were added once to freshly seeded cells (2610 5 in a 6well culture plate in 2 ml of medium). Treatment with JNK Inhibitor 1 (Alexis Biochemicals, 1 mM stock solution in water) at 1 mM, caspase Inhibitor I and II (Calbiochem, 5mM stock solution in DMSO) at 5 mM and with the proteasome inhibitor Lactacystin (Sigma L6785, 5 mg/ml stock solution in water) at 2 mM was done 2 hours before araC, DAC and RA were added. Inhibitors were then added every day. Cells were harvested by trypsin treatment and stored at 280uC.

Viability assay
Cell viability was assayed using the CellTiter-Blue Cell Viability Assay (Promega). NT2, MCF7, HeLa or A549 cells were seeded into 96-well assay plates at a density of 1000 cells per well in 100 ml of medium. Test compounds araC, DAC, DZNep and dC were added in increasing concentrations from 40 nM up to 100 mM and cells were grown for 72 hours. Assays were performed in 6 biological replicates. After 3 days 20 ml of CellTiter-Blue reagent was added to each well, mixed and incubated for 1-4 hours. Absorbance at 570 and 600 nm was measured for each well and resazurin reduction was calculated according to the instructions of the kit. Viability was plotted as percentage compared to the untreated cells.

RT PCR and real time analysis
Total RNA from drug-treated and untreated NT2 tissue culture cells was prepared using the Trizol-reagent (Invitrogen) according the manufacturer's specifications. Real-time RT-PCR was performed using the QuantiTect Reverse Transcription Kit (Qiagen). One mg of total RNA was processed in 20 ml reactions according to the manufacturer's specifications. One ml of the cDNA was used for a 10 ml PCR reaction using the Absolute QPCR SYBR Green Mix (Thermo Scientific) and a Roche LightCycler 480. PCR conditions: 1 cycle: 95uC615 minutes; 42 cycles: 95uC615 seconds, 60uC640 seconds, read; melting curve 65uC-95uC, read every 1uC. Cycle threshold numbers for each amplification were measured with LightCycler 480 software and relative quantifications of expression status of the genes under observation comparing treated to non-treated cells were calculated and normalised using lamin-b and b-actin as internal standards. qRT-PCR measurements were repeated at least three times on biological replicates. P-values were calculated using Student's ttest. For primer sequences see Table S1.

COBRA assay
For methylation analysis of CpG-rich regions in the HOXA cluster genomic DNA was treated with sodium bisulfite and analysed by COBRA [53]. PCR cycling conditions were 95uC for 30 seconds, 55-60uC for 30 seconds, and 72uC for 40 seconds for 35 cycles. The PCR products were gel purified and digested with BstUI or TaqIa (New England Biolabs, Beverly, MA). Digested PCR products were separated on 3% agarose gels. For primer sequences see Table S2.

Capillary electrophoresis
Genomic DNA was prepared from cells using the DNeasy Kit (Qiagen). Global methylation levels were determined by capillary electrophoresis, as described previously [54]. Briefly, genomic DNA was enzymatically hydrolysed to single nucleotides and the nucleotides were derivatised with the fluorescent marker BODIPY (Molecular Probes). Derivatised nucleotides were separated by capillary electrophoresis and analysed in a Beckman PACE MDQ Molecular Characterization System. P-values were calculated using Student's t-test.

DNMT1 trapping assay
NT2 D1 cells were grown and treated with the indicated concentrations of DZNep, araC and DAC as described above. After 3 days of treatment, cells were harvested, washed with PBS and lysed for 30 minutes on ice in trapping buffer (PBS pH 7.4, 0,1% SDS, 0,5% NP40, 0,5% Sodiumdeoxycholate, 1 mM DTT and Roche complete protease inhibitor cocktail). Insoluble debris was pelleted and supernatants were separated by 8% SDS-PAGE and transferred to polyvinylidene difluoride membranes (Millipore). DNMT1 was detected using an antibody from Santa Cruz Biotechnologies (C-17, sc-10222).

Protein depletion via siRNAs
For depletion of SAHH, DNMT1, EZH2, NANOG and OCT4 ON-TARGETplus SMARTpool siRNAs (Dharmacon) were used. NT2 cells were seeded into 6-well plates at a density of 2610 5 in 2 ml of medium. siRNAs were transfected with the Dharma-FECT1 transfection reagent (Dharmacon) according to the manufacturer's instructions. The final siRNA concentration was 50 nM. For combined knock downs the reactions were scaled up accordingly. After 48 h or 72 hours cells were retransfected (48 h), analysed by microscopy or harvested by trypsin treatment. Cell pellets were then analysed by Western blot and RT-PCR. Scrambled siRNAs for negative control experiments were also obtained from Dharmacon. For depletion of araC-induced CASPASE-3 and/or CASPASE-7 cells were treated for 6 hours with 10 mM araC and were then transfected with scrambled siRNAs, CASPASE-3 specific siRNAs, CASPASE-7 specific siRNAs and a mix of both as described above. After transfection the cells were grown for further 48 hours (without changing the medium), analysed by microscopy, then harvested by trypsin treatment and analysed by Western blot and RT-PCR.

ChIP assay
Crosslinked chromatin was prepared and immunoprecipitated as described previously [9]. ChIP-grade antibodies specific for H3K27me3 (07-449) and H3K4me3 (07-473) were purchased from Upstate. Immunoprecipitates were finally dissolved in 30 ml of TE buffer (10 mM Tris-HCL pH 8, 1 mM EDTA). One ml was analysed by real time PCR using a primer pair specific for the HOXA1 promoter region (IPHOXA1prom_up ACTGGAAAGT-TGTAATCCTATG; IPHOXA1prom_lo AGAAAGTTGGCA-CAGTCACG) in 10 ml PCR reactions, using the Absolute QPCR SYBR Green Mix (Themo Scientific) and a Roche LightCycler 480. PCR conditions: 1 cycle: 95uC615 minutes; 42 cycles: 95uC615 seconds/60uC640 seconds, read; melting curve 65uC-95uC, read every 1uC. Cycle threshold numbers for each amplification were measured with LightCycler 480 software and enrichments were calculated as percentage of the input.

Caspase assay
Activity of CASPASE-3 and -7 in DZNep-, araC-, DAC-and RA-treated NT2 cells was assayed with the Caspase-Glo 3/7 Assay System (Promega) according to the instructions of the manufacturer. P-values were calculated using Student's t-test.

Preparation of nuclear extracts and caspase treatment
Nuclei from NT2 cells were prepared according to Wu [55]. The nuclear pellet was resuspended in lysis buffer ( Aliquots corresponding to roughly 15 mg of total protein were treated for 20 minutes with 1 or 2 units of active recombinant human CASPASE-7 or CASPASE-3 (BioVision Research Products) in a total volume of 20 ml at 37uC. SDS-PAGE loading buffer (86) was then added and the samples were separated on a 12% SDS-PAGE gel and transferred to nitrocellulose. Proteins were immunodetected using specific antibodies against NANOG (abcam ab21624), OCT4 (abcam ab18976, N-terminal peptide), OCT4 (abcam ab19857, C-terminal peptide) and b-actin (abcam ab8226).

Histone preparation
Pellets of NT2-cells were washed once with cold PBS and then lysed on ice in cold lysis buffer (10 mM Tris pH 6.5, 5.2 g/l sodium bisulfite, 1% Triton-X100, 10 mM MgCl 2 , 486 g/l sucrose, 1,2 g/l sodium butyrate). Lysates were centrifuged for 5 minutes at 4uC, max. speed in a tabletop centrifuge. Pellets were washed 36 with cold lysis buffer and once with wash buffer (10 mM Tris-HCL pH 7.4, 13 mM EDTA). Final pellets were resuspended in 100 ml water, 1 ml H 2 SO 4 was added and the samples were incubate on ice for 1 hour. Then, precipitates were pelleted and the supernatants were transferred to new tubes. Isolated histones were precipitated by adding 10 volumes of acetone at 220uC over night. Precipitates were collected and pellets were dissolved in water. After Bradford assay the histone solution was set to 1 mg/ml, stabilised by adding PAGE loading buffer to 16 and stored at 220uC.

Generation of cell lines expressing OCT4-specific EGFP
Roughly 1610 6 NT2 cells were transfected with the plasmid phOCT4-EGFP, containing a 4 kb fragment of the human OCT4 regulative region (23917 to +55) upstream of EGFP [35], using the Effectene Transfection Kit (Qiagen) according to the manufacturer's instructions. Cells were plated into 10 cm culture dishes and G148 (Sigma) selection was applied 24 hours after transfection at 350 mg/ml. After 2 weeks of selection surviving colonies were picked and expanded. One clone, showing approximately 90% of EGFP positive cells, was chosen for treatment with araC and RA. ). The transcription of DNMT1, SAHH, EZH2, NANOG and OCT4 is efficiently reduced after the respective siRNA mediated knock down. Expression of the differentiation marker SNAP25 is weakly induced by NANOG knock down, but significantly upregulated upon depletion of OCT4, which is in line with the morphological observations. A combined knock down of NANOG and OCT4, or NANOG, OCT4 and EZH2 does not enhance the effect. qRT-PCR measurements are from two independent experiments and were internally normalised to the corresponding lamin-b and b-actin expression levels. P-values (Student's t-test) for the expression differences between untreated cells (cont.) and transfected cells for highly significant cases (p#0.01) are indicated. Asterisks indicate expression values that are significantly different from controls. Found at: doi:10.1371/journal.pone.0010726.s005 (5.49 MB TIF) Figure S6 The effects of araC and DAC can not be rescued by the proteasome inhibitor Lactacystin. (A) Quantitative RT-PCR expression analysis of the differentiation markers SNAP25 and HOXA1 in NT2 cells after 3 days of treatment with araC, DAC or RA alone or in combination with the proteasome inhibitor Lactacystin (LAC) or a combination of two caspase inhibitors (Casp-I/II). Diagrams show fold differences compared to araC-, DAC-and RA-treatment alone. Co-treatment with caspase inhibitors leads to reduced gene expression after araC-and DAC-treatment, whereas Lactacystin had no significant effect. All qRT-PCR measurements were repeated at least three times and internally normalised to the corresponding lamin-b and b-actin expression levels. P-values (Student's t-test) for the expression differences between only drug-treated and co-treated cells for highly significant cases (p#0.01) are indicated. Asterisks indicate expression values that are significantly different from controls. (B) Western blots showing that drug-induced degradation of OCT4 in NT2 cells after 36 hours of treatment with araC is reduced (the signal is restored) by the mix of caspase inhibitors (Casp), but not by treatment with Lactacystin (LAC, middle). b-actin served as loading control. (C) Western blots showing that drug-induced degradation of EZH2 in NT2 cells after 72 hours of treatment with araC is reduced (the signal is restored) by the mix of caspase inhibitors (Casp), but not by treatment with Lactacystin (LAC, middle). In contrast, DZNep-induced degradation of EZH2 is also inhibited by Lactacystin (blot on the right). b-actin served as loading control. Found at: doi:10.1371/journal.pone.0010726.s006 (4.70 MB TIF) Figure S7 The effect of differentiation-inducing drugs in NT2 cells can be rescued by depletion of CASPASE-3 and -7. (A) Microscopic images (106 magnification) of NT2 control cells (cont.) and NT2 cells treated for 54 h with 10 mM araC (araC). The remaining images show cells that were treated for 6 hours with 10 mM araC, then were transfected with a mix of scrambled siRNAs (kd cont.), CASPASE-3 specific siRNAs (kd CASP-3), CASPASE-7 specific siRNAs (kd CASP-7) and a mix of both (kd CASP-3/7). After transfection the cells were grown for further 48 hours (with araC present). The araC-induced neuron-like morphology can be reverted to some degree by the double knock down. (B) Quantitative RT-PCR expression analysis of the differentiation markers SNAP25, HOXA1 and of CASPASE-3 and CASPASE-7 after the treatment described in (A). Diagrams show fold differences compared to araC-treatment alone. CASPASE-3 and CASPASE-7 transcription is efficiently reduced after the respective siRNA mediated knock down. Combined knock down of both caspases leads to a significantly reduced araCinduced expression of SNAP25, HOXA1. qRT-PCR measurements are from two independent experiments and were internally normalised to the corresponding lamin-b and b-actin expression levels. P-values (Student's t-test) for the expression differences between only araC-treated (cont.) and transfected cells for highly significant cases (p#0.01) are indicated. Asterisks indicate expression values that are significantly different from controls. The calculated molecular weight in kD for potentially originating N-terminal fragments after caspase digest are indicated below the arrows. The potential site used by CASPASE-7 is marked with an asterisk (C) Western blots showing the levels of OCT4 in nuclear extract from NT2 cells without treatment (cont.), in nuclear extract incubated at 37uC or treated with 1 (+) or 2 (++) units of active recombinant human Caspase-7 at 37uC. An antibody specific for an N-terminal peptide of OCT4 was used. b-actin was used as loading control. The blot is identical to the one shown in Fig. 7C but was exposed longer. A digestion product of ca. 30 kD becomes visible, that is significantly increased upon CASPASE-7 digestion (arrow with asterisk). This would correlate with a cut at the predicted site within the homeodomain of OCT4 (asterisk in B). Found at: doi:10.1371/journal.pone.0010726.s008 (5.86 MB TIF)