High Histone Acetylation and Decreased Polycomb Repressive Complex 2 Member Levels Regulate Gene Specific Transcriptional Changes During Early Embryonic Stem Cell Differentiation Induced by Retinoic Acid
Elliot R. Lee,
Cancer Biology Program, University of Wisconsin-Madison, Madison, Wisconsin, USA
Histone modifications play a crucial role during embryonic stem (ES) cell differentiation. During differentiation, binding of polycomb repressive complex 2 (PRC2), which mediates trimethylation of lysine 27 on histone H3 (K27me3), is lost on developmental genes that are transcriptionally induced. We observed a global decrease in K27me3 in as little as 3 days after differentiation of mouse ES cells induced by retinoic acid (RA) treatment. The global levels of the histone K27 methyltransferase EZH2 also decreased with RA treatment. A loss of EZH2 binding and K27me3 was observed locally on PRC2 target genes induced after 3 days of RA, including Nestin. In contrast, direct RA-responsive genes that are rapidly induced, such as Hoxa1, showed a loss of EZH2 binding and K27me3 after only a few hours of RA treatment. Following differentiation induced by leukemia inhibitor factor (LIF) withdrawal without RA, Hoxa1 was not transcriptionally activated. Small interfering RNA-mediated knockdown of EZH2 resulted in loss of K27me3 during LIF withdrawal, but the Hoxa1 gene remained transcriptionally silent after loss of this repressive mark. Induction of histone hyperacetylation overrode the repressive K27me3 modification and resulted in Hoxa1 gene expression. Together, these data show that there are multiple temporal phases of derepression of PRC2 target genes during ES cell differentiation and that other epigenetic marks (specifically, increased acetylation of histones H3 and H4), in addition to derepression, are important for gene-specific transcriptional activation. This report demonstrates the temporal interplay of various epigenetic changes in regulating gene expression during early ES cell differentiation.
Disclosure of potential conflicts of interest is found at the end of this article.
Embryonic stem (ES) cells are pluripotent and capable of differentiation into cells of the three primary germ layers: endoderm, mesoderm, and ectoderm [1, 2]. ES cells can be manipulated under proper growth conditions to form desired cell types [3, –5]. Mouse ES cells are maintained in an undifferentiated state by treatment with the cytokine leukemia inhibitor factor (LIF). Withdrawal of LIF leads to multilineage differentiation . This process can be altered with addition of retinoic acid (RA) . RA is a ligand for the RA receptor (RAR) family of nuclear receptors. RAR members heterodimerize with retinoid X receptor family members and bind to DNA at consensus RA response elements (RAREs), thereby altering transcriptional levels of target genes and influencing differentiation predominantly toward a neuroectoderm lineage [8, –10]. The duration and timing of RA exposure can alter lineage selection.
Early ES cell differentiation involves large-scale changes in gene expression and morphology [2, 11, –13]. Recent evidence has shown that covalent histone modifications play a vital role in the process of differentiation [14, , , , , , –21]. Histone modifications, which include acetylation, methylation, and phosphorylation, can be either repressive or permissive for transcriptional activation, depending on location and context, and are part of the histone code hypothesis [22, 23]. Trimethylation of lysine 27 on histone H3 (K27me3) is a repressive modification that is enriched across the genome of undifferentiated ES cells. This mark is lost from genes during differentiation [14, 18, 24, 25]. Genes enriched for K27me3 in ES cells include those involved in early embryonic development, organogenesis, and cell fate decisions. Some of these are located in regions with both K27me3 and trimethylation at lysine 4 of histone H3 (a permissive modification), forming a structure that has been termed bivalent chromatin domain and that is thought to keep these repressed genes in a state poised for transcriptional activation during differentiation [18, 20, 26]. For the genes so marked, it is proposed that simple removal of the repressive K27me3 is sufficient for transcriptional activation . Recent work suggests that derepression (loss of K27me3) is not sufficient to lead to increased transcription at all genes . Acetylation of histone H3 is generally permissive for transcriptional activation and has also been shown to mark regions in ES and embryonal carcinoma (EC) cells for activation during differentiation .
The K27me3 modification is carried out by polycomb repressive complex 2 (PRC2), which includes SUZ12, EED, and the histone methyltransferase EZH2 [27, , –30]. Binding of PRC2 to target genes is lost during differentiation, following the same pattern as K27me3 . Expression levels of Ezh2 have previously been shown to decrease during embryoid body formation . In addition, animal studies have shown early embryonic lethality of Ezh2 and Suz12 knockout mice [32, 33]. Recent evidence suggests that Suz12 is necessary for differentiation to occur . Although histone methylation was originally thought to be a long-term stable mark, the dynamic nature of methylation is evidenced by the identification of histone demethylases to remove these marks, except for the K27 demethylase, which has not yet been identified [34, , –37].
Although recent studies have shown high levels of K27me3 and PRC2 binding in undifferentiated ES cells, the exact timing and mechanism of derepression of PRC2 target genes and loss of K27me3 during differentiation remain unclear. Previous studies have used long-term differentiation methods (5–10 days). In our study, we analyzed the very earliest stages of ES cell differentiation induced by RA to identify temporal patterns of derepression of PRC2 target genes. We found striking differences in global and local changes in K27me3 and EZH2 binding during this early time frame. There was a global decrease in K27me3 on histone H3 and loss of both EZH2 protein and mRNA after 3 days of RA treatment. This was also observed locally on genes such as Nestin (Nes) that require 2–3 days of differentiation for expression. In contrast, there was a dynamic, rapid loss of K27me3 and EZH2 binding after only a few hours of RA treatment at direct response genes, such as Hoxa1. These direct response genes all contain specific RARE DNA binding sites. These differences show that there are at least two phases of derepression of polycomb complex target genes during differentiation. The genes that are derepressed at the later time (3 days) are associated with the global loss in EZH2 levels and a resulting decrease in K27me3. Hoxa1 is strongly activated by RA treatment but does not become transcriptionally activated by LIF withdrawal alone. Analysis of Hoxa1 showed that decreased levels of EZH2 protein and loss of K27me3 via RNA interference for Ezh2 was not sufficient to induce Hoxa1 transcription in LIF-withdrawn ES cells, suggesting that derepression alone is not sufficient for gene activation of Hoxa1. Multiple other histone modifications, including K4 trimethylation on histone H3 and acetylation on histones H3 and H4, combine to play a role in Hoxa1 gene transcription. Even in the presence of repressive K27me3 at Hoxa1, transcriptional activation was induced by treatment with the histone deacetylase inhibitor trichostatin A (TSA). This resulted in histone H3 and H4 hyperacetylation and Hoxa1 gene transcription, despite the presence of the repressive K27me3 mark. These data establish temporal stages of derepression of polycomb complex target genes during early differentiation and emphasize the complexity of the histone code in regulating gene transcription.
Materials and Methods
CCE mouse ES cells were used for all studies and were obtained from Dr. John Gearhart (Johns Hopkins University) with permission from Dr. Gordon Keller (Mount Sinai School of Medicine). Cells were cultured as previously described . All-trans-RA in ethanol was added at a final concentration of 1 μM. TSA was added at a final concentration of 100 nM at the time of LIF withdrawal.
Extraction of Acid-Soluble Proteins (Histones)
Extraction of acid-soluble proteins was performed as previously described  in a protocol adapted from Upstate Biotechnology (Billerica, MA, http://www.millipore.com) and He et al. .
Immunoblotting was performed as previously described . Primary antibodies against histone H3 K27me3 (1:5,000; 2-hour room temperature [RT]) and EZH2 (1:5,000; 2-hour RT) were from Upstate Biotechnology. Antibody against histone H3 (1:5,000; 2-hour RT) was from Cell Signaling Technology (Beverly, MA, http://www.cellsignal.com). Antibodies against EED (1:2,000; 2-hour RT) and Actin (#C-11; 1:1,000; 2-hour RT) were from Santa Cruz Biotechnology Inc. (Santa Cruz, CA, http://www.scbt.com). Antibody against SUZ12 (1:2,000; 2-hour RT) was from Abcam (Cambridge, U.K., http://www.abcam.com). Image analysis was performed on a Macintosh computer (Apple, Cupertino, CA, http://www.apple.com) using the public domain NIH Image program (available at http://rsb.info.nih.gov/nih-image).
Cells were grown on coverslips and prepared using standard methods . Primary antibody was diluted at 1:500. Secondary antibody (Alexa 488 fluorescein isothiocyanate chicken anti-rabbit IgG) was diluted 1:1,000. Coverslips were mounted (Fluoromount G; SouthernBiotech, Birmingham, AL, http://www.southernbiotech.com) on a glass slide.
RNA was prepared using the Qiagen RNeasy mini-kit with a Qiashredder (Qiagen, Hilden, Germany, http://www1.qiagen.com), following the manufacturer's instructions, including a DNase digestion step.
cDNA was synthesized using 1.25 μg of RNA with avian myeloblastosis virus-reverse transcriptase enzyme and random hexamers (Promega, Madison, WI, http://www.promega.com). A Bio-Rad iCycler system (Bio-Rad, Hercules, CA, http://www.bio-rad.com) was used for quantitative analysis. Real-time polymerase chain reaction (PCR) was performed using a SYBR Green supermix kit (Bio-Rad). The level of each gene transcript was normalized to β-actin expression levels. All primers showed 80%–100% efficiency. Expression was calculated as 2(Ct[gene of interest control treatment − gene of interest experimental treatment)]/2(Ct[β-actin control treatment − β-actin experimental treatment)] . Primers used are included in supplemental online Table 1.
Small Interfering RNAs and Transfection
RNA interference was performed as previously described, with slight modifications . Cells were plated in +LIF medium 24 hours prior to transfection. The first transfection was for 24 hours in −LIF medium. The second transfection was for an additional 24 hours. Cells were harvested at 72 hours of culture time (total of 48 hours in −LIF medium). Mock treatments contained either BLOCK-iT fluorescent oligo (Invitrogen, Carlsbad, CA, http://www.invitrogen.com) or no small interfering RNA, with no differences between these two mock treatments.
Chromatin immunoprecipitation (ChIP) experiments were performed in a protocol adapted from Upstate and previously described . Chromatin from 2 × 106 cells was used for each immunoprecipitation. Approximately 5 μg of primary antibody was added to precleared chromatin, and samples were rotated at 4°C overnight. Antibodies used were against H3 K27me3 (Upstate), EZH2 (Upstate), acetylated H3 (Upstate), acetylated H4 (Upstate), trimethylation of histone H3 on lysine 4 (K4me3; Abcam) with normal rabbit IgG (Santa Cruz Biotechnology) used as control. DNA was purified using a PCR purification kit (Qiagen) following the manufacturer's instructions and used in quantitative PCR analysis.
Global Levels of K27me3 on Histone H3 Decreased During Early RA-Induced Differentiation of ES Cells
We analyzed total levels of K27me3 on histone H3 during the first 3 days of RA-induced differentiation of mouse ES cells to determine whether there was a global change in this repressive histone modification. Immunoblotting for K27me3 was performed on acid-extracted histones using an antibody against K27me3 with native histone H3 used as a loading control (Fig. 1A). Histones were extracted from undifferentiated ES cells (+LIF), or cells were withdrawn from LIF and treated with RA (1 μM) (−LIF/+RA) for 14, 48, or 72 hours. Quantitative analysis from multiple experiments showed a global decrease of three- to fourfold in K27me3 by 3 days of RA-induced differentiation of ES cells (Fig. 1B). Immunofluorescent analysis for K27me3 showed a diffuse nuclear staining pattern, consistent with previous evidence of large K27me3 domains present in ES cells . This genome-wide distribution was lost with 72 hours of −LIF/+RA, but low levels of K27me3 were still observed (supplemental online Fig. 1).
Expression of Members of PRC2 Decreased During Early RA-Induced Differentiation of ES Cells
To understand the mechanism responsible for the global loss of K27me3 during the first 3 days of RA-induced differentiation, we analyzed expression levels of components of PRC2. Whole cell extracts of +LIF and −LIF/+RA cultured ES cells were used in immunoblotting analysis with anti-EZH2, anti-SUZ12, and anti-EED antibodies. An antibody to Actin was used as a loading control (Fig. 2A). We observed decreased protein expression of the PRC2 components EZH2 and SUZ12 that closely followed the global loss of K27me3. EED did not show a change in protein level during this time frame. Immunofluorescent staining with anti-EZH2 showed high levels of nuclear protein in +LIF ES cells (Fig. 2B). After 72 hours of −LIF/+RA treatment, there was a marked decrease in the amount of EZH2 protein. However, as with K27me3, low levels of EZH2 protein expression were maintained at 3 days of RA treatment of ES cells (supplemental online Fig. 1). To further understand this process, we analyzed mRNA expression of Ezh2 by quantitative RT-PCR. This analysis showed a gradual decrease in Ezh2 transcript level, mirroring that of the protein level. These data indicate that decreased gene expression of Ezh2 may be responsible for loss of PRC2, which subsequently leads to a loss of global K27me3 by 72 hours following RA treatment of ES cells.
The Kinetics of Changes in Local Levels of EZH2 Binding and K27me3 During RA-Induced Differentiation Define Two Subclasses of Target Genes
The global loss of K27me3 occurs at 72 hours following RA treatment, yet several RA-responsive genes show increased transcription within only a few hours, whereas other genes respond more slowly. We chose to analyze the local changes in EZH2 binding and K27me3 at specific genomic regions using ChIP on two different sets of genes, each with a different kinetics of transcriptional activation by RA. Nes is known to be upregulated during RA-induced differentiation of ES cells , yet it does not have a known RARE. We determined by quantitative RT-PCR (qRT-PCR) analysis that Nes responded slowly to RA, with transcript levels showing large increases only after 48 and 72 hours of −LIF/+RA treatment (Fig. 3A, diamonds). We used cDNA microarray analysis as a screen (data not shown) to identify other time-dependent RA-inducible genes in our ES cell system. Two genes, transforming growth factor β2 (Tgfβ2) and dikkopf-1 (Dkk-1) were selected for further study based on their transcriptional induction kinetics being similar to those of Nes in our microarray analysis and on the fact that they are PRC2 target genes in human ES cells . The qRT-PCR data showed that these two genes, like Nes, respond slowly with transcript levels, showing large increases only after 48–72 hours of −LIF/+RA treatment (Fig. 3A, diamonds). We used ChIP to study the presence of EZH2 binding and K27me3 at the promoter regions of Nes, Dkk-1, and Tgfb2 (Fig. 3A, squares [K27] and circles [EZH2]) in the undifferentiated state (+LIF) or following 14 or 72 hours of −LIF/RA treatment. Normal rabbit IgG was used as a control for nonspecific binding (Fig. 3A, triangles). As a negative control for ChIP, we also examined a genomic region of Nes that showed very low levels of EZH2 binding and K27me3 (data not shown). We observed a relatively high level of EZH2 binding and K27me3 and a temporal pattern of decreased EZH2 binding and K27me3 by 72 hours of −LIF/+RA treatment. However, EZH2 binding and K27me3 were maintained at the early, 14-hour −LIF/+RA treatment time point. All three of these genes showed a similar temporal pattern of change.
Hoxa1 is known to be upregulated during RA-induced differentiation of ES cells [40, 41]. Hoxa1 is a direct RA response gene with a direct repeat DR5 RARE downstream of the coding region at the 3′ end of the gene [44, , , –48]. Hoxa1 showed a rapid increase in accumulation of mRNA with high levels present within 14 hours of −LIF/+RA treatment (Fig. 3B, diamonds). To determine whether the rapid increase in Hoxa1 mRNA also occurred at other Hox sites, we analyzed the Hoxb1 gene. Hoxb1 can be regulated by Hoxa1 and contains multiple RARE sites . The qRT-PCR analysis showed that Hoxb1 also rapidly responded with high levels of mRNA accumulation within 14 hours of −LIF/+RA treatment (supplemental online Fig. 2). We identified two other known rapid response targets of RA treatment, which included Rarβ, a member of the family of nuclear receptor proteins that bind RA, and Cyp26a1, a cytochrome P450 enzyme involved in RA metabolism . The rapid response of these genes was confirmed by qRT-PCR analysis with rapid upregulation of transcript levels by 14–24 hours of −LIF/+RA treatment, similar to Hoxa1 and Hoxb1 (Fig. 3B, diamonds). ChIP analysis of EZH2 binding and K27me3 at the Hoxa1 3′ DR5 and the RAREs of Rarβ and Cyp26a1 showed dramatic differences from the Nes promoter, with marked loss of EZH2 binding and K27me3 within 14 hours of −LIF/+RA treatment (Fig. 3B, squares [K27] and circles [EZH2]) and a further decrease by 72 hours of −LIF/+RA (Fig. 3B). Normal rabbit IgG was used as a control for nonspecific binding (Fig. 3B, triangles). Hoxb1 3′ DR5 showed a similar temporal loss of EZH2 binding and K27me3 (supplemental online Fig. 2). A more detailed, early time course analysis of this loss of EZH2 binding and K27me3 (supplemental online Fig. 3) showed that at the Hoxa1 3′ DR5, there was a detectable loss of EZH2 binding and K27me3 as early as 6 hours of −LIF/+RA treatment. These data establish at least two sets of PRC2 target genes defined by temporal alterations in transcriptional derepression. One set appears to be regulated by the loss of EZH2 protein and mRNA by 72 hours after RA, and the other set, each member of which contains an RARE, occurs early even in the presence of unchanged global levels of EZH2 protein and RNA by an unknown mechanism.
K27me3 Increased at the Regulatory Regions of Two Genes Known to Be Transcriptionally Repressed During Differentiation
We further studied regulatory regions of two genes expressed in undifferentiated, pluripotent ES cells, Oct4 and Sox2, to determine whether K27me3 was present at any time [39, 51, 52]. Oct4 and Sox2 expression levels decreased by 72 hours after RA-induced differentiation, as seen by qRT-PCR in Figure 4 (diamonds). Surprisingly, there was an increase in K27me3 at regulatory regions of Oct4 and Sox2 at 72 hours of −LIF/+RA treatment despite the global decrease in K27me3. The levels of K27me3 were still relatively low (compare the maximum of 0.3%–0.6% [Fig. 4, squares] with 1%–4% [Fig. 3A, 3B]). Low levels of EZH2 binding were present at Oct4 and Sox2 in the undifferentiated state and remained steady during differentiation (Fig. 4, circles), despite the global loss of EZH2 protein expression. These data point to a complex multiphasic loss and gain of K27me3 during early RA-induced differentiation of ES cells.
Loss of K27me3 Alone at the Hoxa1 Gene Did Not Induce Gene Expression During Early ES Cell Differentiation
Previous studies, along with data presented here, point to the importance of derepression of PRC2 target genes during ES cell differentiation [14, 24, 25]. Current models propose that loss of the repressive histone modification regulates gene transcription during differentiation, by allowing permissive activating marks such as histone H3 K4 trimethylation to dominate. We took advantage of the unique nature of Hoxa1 gene expression to study whether loss of a repressive mark is sufficient for transcription . Unlike Hoxb1, which we previously showed to be induced during withdrawal of LIF from ES cells, Hoxa1 was not induced by LIF withdrawal alone, as seen by qRT-PCR analysis in Figure 5A. ChIP analysis showed that with LIF withdrawal alone, there was a high level of K27me3, even at 72 hours, and a gradual but decreased amount of EZH2 binding at the Hoxa1 3′ DR5 (Fig. 5B) by 72 hours. Global EZH2 protein levels were maintained at 72 hours of LIF withdrawal (data not shown). We proceeded to knock down EZH2 protein levels with RNA interference technology to determine whether the presence of the K27me3 repressive mark alone dictated repression of Hoxa1 gene expression, as observed in the LIF-withdrawn cells compared with the RA-induced cells (Fig. 5C, 5D) . Successful knockdown was observed by immunoblotting analysis that shows decreased total EZH2 protein levels (Fig. 5C). ChIP analysis in these knockdown cells showed a loss of K27me3 and EZH2 binding at the Hoxa1 3′ DR5 after 48 hours of LIF withdrawal alone (compare Fig. 5B with Fig. 5C). Despite the loss of this repressive mark at the Hoxa1 3′ DR5, there was not an induction of Hoxa1 gene transcription (Fig. 5D). However, Dkk-1 was induced threefold upon knockdown of EZH2 protein levels (supplemental online Fig. 4), suggesting that for some genes, derepression is a critical mechanism regulating gene transcription. This indicates that for Hoxa1, loss the K27me3 repressive mark was not sufficient for transcriptional activation during this early period of differentiation.
Increased Histone Acetylation and Loss of K27me3 Regulate Hoxa1 Gene Expression During RA-Induced Differentiation of ES Cells
To determine what role other histone modifications may play in regulating Hoxa1 expression during ES cell differentiation, we analyzed the local presence of other histone tail modifications using ChIP. ChIP analysis at the Hoxa1 3′ DR5 showed the presence of the transcriptionally permissive mark, K4me3, in the pluripotent and differentiated state (Fig. 6A) . There was a high level of H3 acetylation (acH3) present in both the undifferentiated and differentiated state. There was also a rapid fourfold increase in histone H4 acetylation (acH4) observed during RA-induced differentiation, visible by 14 hours of −LIF/+RA treatment at the Hoxa1 3′ DR5 (Fig. 6A). However, the increase in acH4 was not observed by ChIP analysis of the Hoxa1 3′ DR5 in cells differentiated by LIF withdrawal alone, which do not express Hoxa1 (Fig. 6B). These data suggest that histone H4 acetylation plays an important role in the proper induction of Hoxa1 transcription during differentiation.
We next used a chemical inhibitor of histone deacetylases, TSA, to determine whether increased histone acetylation at Hoxa1 could induce gene transcription even in the presence of a repressive histone modification (K27me3). Treatment of cells with a short pulse of only 14 hours of TSA along with withdrawal of LIF showed increased acH3 and acH4 at the Hoxa1 3′ DR5, as observed by ChIP analysis (Fig. 6C). However, the TSA treatment did not alter the high levels of K27me3 present at the Hoxa1 3′ DR5. Analysis by qRT-PCR showed a rapid increase in Hoxa1 gene expression observed by 14 hours of −LIF/+TSA (Fig. 6D). These results demonstrate a highly complex regulation of gene transcription at the early RA response gene Hoxa1 via loss of K27me3 and the presence of other permissive histone modifications, including high levels of K4me3, acH3, and particularly acH4.
Recent evidence shows the importance of histone tail modifications during ES cell differentiation [14, , –17]. Specifically, K27me3 is present on key developmental genes that are repressed in the pluripotent ES cell [14, 24, 25]. During differentiation, there is a loss of EZH2 binding and K27me3, leading to derepression of targets. However, previous studies have used long-term differentiation methods over 5–10 days. In this study, we demonstrated that (a) the rapid, dynamic nature of histone tail modifications during early ES cell differentiation is likely regulated by unknown gene-specific targeting mechanisms; (b) loss of K27me3 and EZH2 binding occurs during at least two different time periods and likely via two different mechanisms at subsets of PRC2 target genes during the first 3 days of RA-induced differentiation; (c) derepression by loss of K27me3 is not sufficient to activate gene transcription at all genes; and (d) increased histone acetylation can override the K27me3 repressive mark to induce gene transcription, at least at some genes. Taken together, these observations suggest that targeting mechanisms must regulate the specific genes or gene sets that undergo altered histone tail modifications in a time-dependent manner during early ES cell differentiation.
We observed a global loss of K27me3 in ES cells after 3 days of differentiation by LIF withdrawal with RA treatment. This time period was earlier than has previously been reported. In addition to the loss of K27me3, there was a decrease in protein expression of PRC2 members EZH2 and SUZ12 after 3 days of RA-induced differentiation, which likely accounts for the global decrease in K27me3. The decrease in SUZ12 may be transient, as recent work has demonstrated that Suz12 is necessary for differentiation of ES cells . The decrease in SUZ12 protein levels may be unique to RA-induced differentiation, as others have not seen this decrease during embryoid body differentiation . Gene expression of Ezh2 also decreased during this time period, producing a cascade of effects including loss of EZH2 protein and K27me3. Global K27me3 was still maintained, but at low levels after 3 days of RA.
Our examination of patterns of K27me3 revealed two classes of K27me3 target genes during early differentiation. Genes such as Nes, Dkk-1, and Tgfβ2 showed a late response to RA treatment, with gene transcription induction occurring only after 2–3 days. Local EZH2 binding and K27me3 were lost in target regions in these genes at approximately 2–3 days of RA treatment. This was the time period of global loss of EZH2 binding and K27me3. In contrast, genes such as Hoxa1, Hoxb1, Cyp26a1, and Rarβ showed rapid induction of gene transcription, with transcript accumulation observed within the first 14 hours of RA treatment. Local analysis of EZH2 binding and K27me3 at these targets differed from global levels, with rapid loss occurring as early as 6 hours of RA treatment. All of these genes are direct RA response genes with RAREs. The loss of EZH2 binding and K27me3 on the Hoxa1 gene appeared to occur first at the RARE, which is important in regulating gene transcription . This suggests that the RARE may represent the initial nidus of EZH2 ejection and loss of K27me3. Mechanisms responsible for this early loss of EZH2 binding and K27me3 are still unclear. There is likely a histone demethylase that removes methyl groups at EZH2 targets. However, unlike LSD1 and the JMJ proteins responsible for demethylation at K4 or K9 and K36, the identity of a K27 demethylase remains unknown . In addition, it is not clear how EZH2 is ejected from target regions at early response target genes, such as Hoxa1. The EZH2 and RARα proteins are reported to bind a common partner, PRAME; however, we did not detect an interaction between EZH2 and RARα by coimmunoprecipitation (data not shown) . Another possible mechanism is the activation of upstream signaling enzymes, such as AKT, which is reported to negatively regulate EZH2 binding and K27me3 via phosphorylation of EZH2 . We observed a rapid increase in AKT phosphorylation after RA treatment, as have others in EC cells , but were unable to demonstrate increased phosphorylation of EZH2 (data not shown). It is also possible that the binding of RARα protein and associated cofactors alters local chromatin structure in such as way as to reduce EZH2 binding efficiency without a direct interaction.
We observed another class of K27me3 target genes, exemplified by the pluripotent genes Oct4 and Sox2. These genes are expressed in the pluripotent ES cells and showed significantly decreased gene expression levels and increased levels of K27me3 at promoter regions by 72 hours of RA treatment (Fig. 4), despite the global loss of EZH2 and K27me3. This increase in K27me3 has also been recently reported for Oct4 in human EC cells during longer RA-induced differentiation [19, 20]. Although increased, the absolute levels of K27me3 on these genes in our study remained relatively low. The major long-term repressive histone modification for Oct4 has been proposed to be K9 methylation of histone H3 [19, –21, 57]. Other epigenetic mechanisms, such as DNA methylation, also regulate Oct4 expression . DNA methyltransferases are recruited by the PRC2 complex linking these two repressive mechanisms . Increased K27me3 on these two pluripotent genes during differentiation also raises the question of how PRC2 is targeted. Although there are well-known polycomb response elements in Drosophila, such a targeting system has not been found in mammalian systems . Previous studies have shown that alterations in EED isoforms during differentiation alter PRC2-binding targets [61, 62]. Also, recent evidence indicates that Oct4, Sox2, and Nanog may recruit PRC2s to target regions in the pluripotent state . Previous studies have identified genes that are expressed in the pluripotent state despite PRC2 occupancy and K27me3 . Taken together, these results point to a very complex system regulating polycomb group proteins and K27me3 during ES cell differentiation.
Increased transcription from PRC2 target genes during differentiation is more complex than simple derepression via loss of K27me3. Bivalent chromatin domains that contain small regions of the permissive K4me3 modification within large regions of repressive K27me3 are present in the pluripotent ES cell . These bivalent domains tend to resolve during differentiation to produce either a transcriptionally active gene or a repressed gene. Hoxa1 is a gene that contains such a bivalent domain. Hox genes are classic polycomb target genes that occur within four cluster regions and have large regions of K27me3 [64, –66]. Genes at the 3′ end of these clusters are activated first, with subsequent activation of Hox genes necessary for proper body segmentation. RA is a potent inducer of Hoxa1, which contains an RARE in its 3′ region (3′ RAIDR5) . We observed that Hoxa1 was not induced during differentiation with LIF withdrawal alone, unlike Hoxb1, which was activated by both LIF withdrawal alone and RA treatment. This unique nature of Hoxa1 allowed us to test the importance of the loss of K27me3 in Hoxa1 gene transcriptional activation. We found that loss of K27me3 early in differentiation was not sufficient for Hoxa1 gene transcription. This raised the question of what other histone modifications were necessary for activation of Hoxa1 gene transcription. We found high levels of the permissive K4me3 mark on the Hoxa1 gene in both the pluripotent and differentiated states. Despite the loss of K27me3 with RNA interference of Ezh2 expression and high levels of K4me3, transcription of Hoxa1 was still not induced by LIF withdrawal alone. Previous studies have reported that histone acetylation is important during ES cell differentiation . H3 acetylation occurs at discrete sites in the pluripotent state and spreads out during differentiation . This is hypothesized to mark the gene for transcriptional competence during differentiation . In our laboratory, we have observed that acH4 increased during early differentiation induced by LIF withdrawal in both a global and gene-specific manner . In the current study, we observed that Hoxa1 contained high levels of acH3 in the pluripotent state, in agreement with it being poised for activation. However, a four- to fivefold increase in acH4 was seen at Hoxa1 with RA treatment, which did not occur during LIF withdrawal alone. Pharmacologic induction of acH3 and acH4 with TSA treatment resulted in Hoxa1 transcription, despite the continued presence of repressive K27me3. This suggests that a poised Hoxa1 gene contains high levels of K27me3, K4me3, and acH3 in the pluripotent state and that RA treatment causes a rapid loss of K27me3, while maintaining high K4me3 and acH3. In addition, the level of acH4 increases to aid in converting the poised potential of Hoxa1 into activated transcription. Alterations in chromatin structure are likely influenced by RAR and cofactor binding [68, 69], but we have yet to detect a direct interaction between PRC2 complex members and members of the RAR family. Increasing the permissive histone modifications, such as histone acetylation, appears to be able to overcome repressive modifications, such as K27me3. These data point to a complex multivalent epigenetic model of differentiation that includes the dynamic interplay of a multitude of histone modifications, as predicted by the histone code and diagrammed in our model (Fig. 7). These results suggest that altering early epigenetic events, such as the composition of histone tail modifications, may alter gene expression patterns and ultimately lineage selection during differentiation.
Disclosure of Potential Conflicts of Interest
The authors indicate no potential conflicts of interest.
We thank Ruben Alexanian and Rebecca McDermid for technical support and Yi Mu for assistance with primer design. This work was supported in part by NIH Grant R01-DK064243 (to M.K.F.) and a Rath Foundation Graduate Fellowship (to E.R.L.).