Polycomb recruitment attenuates retinoic acid–induced transcription of the bivalent NR2F1 gene

Polycomb proteins play key roles in mediating epigenetic modifications that occur during cell differentiation. The Polycomb repressive complex 2 (PRC2) mediates the tri-methylation of histone H3 lysine 27 (H3K27me3). In this study, we identify a distinguishing feature of two classes of PRC2 target genes, represented by the Nr2F1 (Coup-TF1) and the Hoxa5 gene, respectively. Both genes are transcriptionally activated by all-trans retinoic acid (RA) and display increased levels of the permissive H3K9/K14ac and tri-methylated histone H3 lysine 4 epigenetic marks in response to RA. However, while in response to RA the PRC2 and H3K27me3 marks are greatly decreased at the Hoxa5 promoter, these marks are initially increased at the Nr2F1 promoter. Functional depletion of the essential PRC2 protein Suz12 by short hairpin RNA (shRNA) technology enhanced the RA-associated transcription of Nr2F1, Nr2F2, Meis1, Sox9 and BMP2, but had no effect on the Hoxa5, Hoxa1, Cyp26a1, Cyp26b1 and RARβ2 transcript levels in wild-type embryonic stem cells. We propose that PRC2 recruitment attenuates the RA-associated transcriptional activation of a subset of genes. Such a mechanism would permit the fine-tuning of transcriptional networks during differentiation.


INTRODUCTION
The ability to self-renew and differentiate into specific cell lineages in response to external stimuli is a unique property of pluripotent stem cells. This ability makes embryonic stem (ES) cells an excellent model for in vivo differentiation (1). All-trans retinoic acid (RA), a metabolite of vitamin A, induces epigenetic and transcriptional changes underpinning the differentiation of various stem cells, including ES cells (2,3). Several key regulators of stem cell differentiation exhibit a bivalent chromatin structure possessing both repressive and permissive histone modification, trimethylated histone H3 lysine 27 (H3K27me3) and trimethylated histone H3 lysine 4 (H3K4me3), respectively (4). Differentiation of stem cells, e.g. during neurogenesis, involves epigenetic changes, which resolves bivalent regions into either active H3K4me3-rich, or repressive H3K27me3rich, domains (5,6).
The Polycomb Repressive Complex 2 (PRC2) is a multiprotein complex that confers transcriptional repression via the placement of the repressive H3K27me3 histone mark. Indeed, Polycomb repressive complexes (PRC1/2) silence many genes in ES cells (7,8). The PRC2 protein Ezh2 is a H3K27-specific histone methyltransferase that, via epigenetic modification of histones, controls aspects of cell fate choice during differentiation (9). Ezh2 deposits the H3K27me3 repressive mark recognized by PRC1 factors, which leads to Ring1 mono-ubiquitination of histone H2A lysine 119 (8,10,11). Ezh2, Suz12 and Eed proteins form the core of the PRC2 complex, and the methyltransferase activity of PRC2 requires both Ezh2 and Suz12 (12,13). While the role of epigenetic modifications at promoter proximal regions has been extensively studied (4), the mechanisms by which epigenetic changes at distal enhancer sites influence transcription and how these relate to the PRC function at proximal promoters are only now emerging (14,15).
Vitamin A (retinol) and its natural and synthetic analogs, retinoids, exert profound effects on many biological processes [for review see (2,3)]. The retinol metabolite all-trans RA mediates most biological effects of retinol (16), and has been implicated in numerous in vivo differentiation pathways (17). The actions of RA are primarily mediated by two classes of nuclear retinoid receptors: retinoic acid receptors (RARs) and retinoid X receptors (RXRs) (18). These nuclear receptors are members of the steroid hormone or nuclear receptor superfamily that also includes estrogen, androgen, thyroid hormone, peroxisome proliferator activated receptors and vitamin D receptors. These receptors act as ligand-modulated transcription factors that activate transcription of specific target genes (19,20). We have previously shown that RA treatment of ES and F9 cells leads to the removal of the PRC2 complex from several RA target genes, including Hoxa1, Cyp26a1 and RAR 2 (21)(22)(23)(24), and that the removal of PRC2 is a key step in the transcriptional induction of these direct/primary RA target genes (21,25). It is unclear whether PRC2 displacement is a common feature associated with RA-induced transcription.
Nr2F1 and Nr2F2 are also referred to as the chickenovalbumin upstream promoter-transcription factors (Coup-TF1/2). Nr2F1 and Nr2F2 belong to a diverse group of nuclear receptors, which are termed orphan nuclear receptors because physiological ligands have not yet been identified (26). In mammals, only two genes, Nr2F1 (Coup-TF1, EAR-3) and Nr2F2 (Coup-TF2, ARP-1), have been identified, but homologs have been cloned from numerous species (27,28). The high degree of evolutionary conservation of the Coup-TF proteins strongly suggests that they are primordial members of the nuclear receptor family and that they have important biological functions (29)(30)(31)(32). We and others have shown that RA-induced endodermal differentiation is associated with increased Nr2F1 mRNA levels (33)(34)(35). We have further demonstrated that ectopic expression of Nr2F1 enhances the RA-induced differentiation of ES cells into extra-embryonic endoderm (35), which may suggest that induction of Nr2F1 is a key event in the generation of endodermal tissue. Consistent with this expression of Laminin, which is required to separate the primitive endoderm from the epiblast, is regulated by Nr2F1 (36). Although Nr2F2 has been shown to bind RA (37), it is not clear if Nr2F2 is activated by RA under physiological conditions. The expression patterns of Nr2F1 and Nr2F2 are partially overlapping in the early mouse embryo (E7.5), but later in development, Nr2F1 is expressed mainly in the nervous system, whereas Nr2F2 is predominantly expressed in the mesenchyme of internal organs such as the pancreas (28). The roles of Nr2F1 and Nr2F2 in RA-induced endodermal differentiation and the potential association with RA suggest that these transcription factors are key players in mediating the cellular response to RA (38), yet how the expression of the Nr2F1 and Nr2F2 is regulated by RA remains largely unknown.
In this study, we first evaluate the effects of RA on the epigenetic states of Nr2F1, Nr2F2, Hoxa5 and Hoxa1 in F9 embryonal carcinoma stem cells. We demonstrate differential Suz12 dynamics between two types of PRC2 target genes, represented by the Nr2F1 (Coup-TF1) and Hoxa5 genes. We further evaluate both transcriptionally permissive and PRC2-associated repressive epigenetic marks. We extend these findings to ES cells, and delineate the functional role of Suz12 in the RA-induced transcription of the genes Nr2F1, Nr2F2, Meis1, Sox9, BMP2, Hoxa5, Hoxa1, Cyp26a1, Cyp26b1 and RAR 2 . Our findings identify PRC2 dynamics as a distinguishing feature between two classes of RA-inducible stem cell genes, both of which include several key regulators of differentiation. We suggest that in addition to maintaining transcriptional repression, the PRC2 complex attenuates the transcriptional activation of specific genes during stem cell differentiation.

Generation of Suz12 knockdown cell lines
Generation of viral particles and transduction of ES cells was previously described (42,43). In brief, knockdown vectors pLKO shSuz12 (Cat. # TRCN0000038728, Sigma Aldrich, MO) or pLKO shLuc (control), together with packaging vectors pCMVÁ8.9 and pVSV-G (Cat. #631530, Clontech, CA), were transfected into HEK293T cells using Lipofectamine 2000 (Cat. # 11668019, Invitrogen, CA). On overnight recovery, the cells were replenished with fresh media and allowed to produce virus for an additional 48 h before the supernatant was harvested, filtered through 0.45 mm filters and supplemented with polybrene. F9 and ES cells were transduced with viral supernatant in a 1:1 ratio with 2Â growth medium. About 16 h later, the cells were replenished with media supplemented with puromycin (0.5 mg/ml) for 10 days of propagation in the selection media. In agreement with a previous publication (44), the knockdown was validated by western blotting.

Semi-quantitative and Quantitative RT-PCR
Total RNA (3 mg) was used to synthesize cDNA with random primers. The cDNA synthesis was performed at 42 C for 1 h in a final volume of 20 ml using qScript (Quanta, MD). PCR were performed using 2.5 Â 10 À2 U Taq DNA polymerase (Invitrogen, CA). Each cycle included 94 C for 30 s (denaturation), 58-64 C for 45 s (annealing) and 72 C for 1 min (extension). The number of cycles required for PCR amplification in the linear range was determined experimentally for each gene. For semiquantitative PCR, amplified PCR products were resolved on 2% agarose gels and visualized by staining with ethidium bromide. Primer pairs were either obtained from published articles or designed using the PrimerSelect program (DNAstar). The primer pairs were evaluated using in silico PCR analysis (http://genome. ucsc.edu/) to avoid pseudogenes. All primers were designed to anneal to different exons to avoid any contribution of genomic DNA to the signal. Sequences of genespecific primers are specified in the Supplementary Table S1. All gene expression amplicons were validated by sequencing.

Data processing and Statistical analysis of quantitative PCR
The transcript levels in the biological triplicates (n = 3) were normalized to 36B4 transcript levels and statistical significance was determined by t-test (P < 0.05). The chromatin immunoprecipitation (ChIP) signals in the biological triplicates (n = 3) were normalized to percent input and statistical significance was determined by t-test (P < 0.05). The standard error of the mean was determined for each of the data sets (plotted as error bars in the graphs), and P-values <0.05 between compared samples were assigned statistical significance.

Library screening and promoter sequence analysis
To isolate the mouse Nr2F1 promoter, a genomic library (129SVJ Mouse Genomic Library in the Lambda FIX II vector, Stratagene, CA) was screened. A PCR product containing $1 kb of mouse Nr2F1 promoter sequence was used as the probe for library screening. Positive plaques were further screened three times, and phage DNA was isolated using the Wizard kit (Promega, WI). The positive clones were verified by sequencing a portion of each clone. The inserts in the positive plaques were cloned into the pGL3 firefly luciferase reporter plasmid (Clontech, CA), and assayed for RA responsiveness. This identified a 1 kb region as sufficient for RA-induced transcription. The fragment corresponded to a 1 kb region of sequence located $2.4 kb upstream of the P3 RefSeq transcriptional start site (TSS; genome coordinates, NCBI bld 37, MM9, chr13:78339594-78340675).

Transient transfections and luciferase assays
F9 cells were transfected with luciferase reporter constructs containing different lengths of 5 0 flanking sequences of the Nr2F1 gene using the Lipofectamine method (Invitrogen, CA) (45). The reporter plasmid pRL-TK (Renilla luciferase-thymidine kinase) (Promega, WI) was used as control for transfection efficiency. Cells were then cultured with or without RA for another 24 or 48 h. Firefly and Renilla luciferase activities were sequentially measured using the Dual-luciferase Reporter Assay system (Promega, WI) with a luminometer.

ChIP assays
ChIP assays were performed as previously described (22,46,47). In brief, a one-step ChIP protocol that uses formaldehyde cross-linking was employed for histone ChIP assays. For Suz12, Ring1B and polII-CTD ChIP assays, we used a two-step ChIP protocol. Cells were resuspended in variable amounts of lysis buffer thereby normalizing for differences in cell numbers between plates. ChIPs of sonicated chromatin from 5.0 Â  Table S2. Each ChIP assay was performed at least three times starting with independently propagated cells each time (n ! 3).

Bioinformatics analyses
The gene location, exon-intron boundaries and TSSs were specified in accordance with the NCBI Reference Sequences (50). Processed, mapped, next-generation RNA-seq, ChIP-seq and DNA methylation/hydroxymethylation data sets for undifferentiated and RAtreated mouse ES cells were downloaded from the NCBI Geo database (http://www.ncbi.nlm.nih.gov/geo/) and viewed in the integrated genome viewer (IGV 2.0) (51). Where appropriate the GALAXY implementation of the UCSC liftover tool was used to convert genome coordinates between mouse genome builds v36 (mm8) and NCBI37 (mm9) (52).

RESULTS
Identification and cloning of the murine Nr2F1 (Coup-TF1) 5 0 flanking region We determined the time-dependent increases in Nr2F1 and Hoxa5 transcript levels in response to RA treatment of F9 WT embryonal carcinoma stem cells ( Figure 1A). We found that Nr2F1 transcript levels exhibit a moderate/slow increase, with a half-maximal induction at 24 h, whereas Hoxa5 transcript levels increase dramatically/rapidly in response to RA, with a half-maximal induction at 12 h. Consequently, the induction of Nr2F1 is delayed relative to that of Hoxa5. We have previously shown that increased transcriptional activity of the Hoxa-d clusters in response to RA is associated with decreased PRC2 occupancy of the Hox promoters (23). Therefore, we used ChIP to evaluate the levels of the PRC2 core component protein Suz12 at the Nr2F1 and Hoxa5 promoter regions in response to RA treatment of F9 WT cells ( Figure 1B). We found that the Suz12 levels at the Nr2F1 promoter region increased during the first 24 h, and then declined to below the initial levels. In contrast, the Suz12 levels decreased rapidly at the Hoxa5 promoter region ( Figure 1B).
To evaluate the involvement of the RA receptors RARa, RARb 2 and RARg, respectively, we next assessed the Nr2F1 and Hoxa5 transcript levels in F9 RAR knockout cell lines devoid of individual RAR isotypes (21,(39)(40)(41)53). We found that 24 h of RA treatment significantly increased Nr2F1 transcript levels in F9 WT ($4-fold). In the RAR 2 À/À and RAR À/À cell lines, Nr2F1 transcript levels were induced by RA to only 43 and 57% of WT levels, respectively ( Figure 1C, left). In contrast, the RAR À/À cell line displayed Nr2F1 transcript levels similar to those of WT (97%). For comparison, we measured the increase in Hoxa5 transcript levels in response to RA ( Figure 1C, right). The Hoxa5 gene is directly activated by RA in vivo and in cultured ES cells (54). In F9 WT cells Hoxa5 transcript levels were potently induced by 24 h of RA treatment ($8-fold). Consistent with Hoxa5 expression being dependent on RARb 2 (54), we found that Hoxa5 transcript levels were increased to only 20% and 38% of WT levels following RA addition to (B) ChIP analysis shows that the PRC2 core protein component Suz12 initially increases at the Nr2F1 promoter, but decreases at the Hoxa5 promoter in response to RA (RefSeq promoter regions). (C) Nr2F1 and Hoxa5 transcript levels in F9 WT and in RARa, RARb 2 and RARg knockout cells, untreated or treated with RA for 8 or 24 h. Statistical significance (P < 0.05) is indicated for the effect of RA (24 h RA relative to vehicle-treated cells), and for the effect of RAR knockout in 24 h RA treatment conditions (relative to WT). All experiments were performed three or more times starting with fresh cells (n ! 3). the RAR 2 À/À and RAR À/À cell lines, respectively ( Figure 1B). In contrast, the RAR À/À cell line displayed Hoxa5 transcript levels that were similar to WT (110%). Therefore, we conclude that both RARg and RARb 2 (or downstream targets thereof) are required for the full RAassociated increases in Nr2F1 and Hoxa5 transcript levels to be achieved.
Deletion analysis of the Nr2F1 upstream region and analysis of DNA methylation status To identify DNA elements required for Nr2F1 transcriptional activation, we isolated $14 kb of the mouse Nr2F1 locus from a phage library. The phage DNA that contained portions of 5 0 Nr2F1 sequence (55) was digested and fragments were assayed in the pGL3 reporter to identify the DNA region(s) responsible for the RA induction. We identified a 1 kb region upstream of the RefSeq TSS (P3, Figure 2A) as sufficient to drive RA-dependent transcription (data not shown). Importantly, this fragment contains at least two previously identified TSS (P1 and P2, Figure 2A) for the murine Nr2F1 gene (50,55).
To delineate the region responsible for the RA induction of Nr2F1, we generated further deletions of the RA responsive 1 kb construct. Together with a series of shorter deletion constructs, all of which contain the most upstream promoter identified for Nr2F1 (P1), the 1 kb construct was assayed for RA responsiveness in F9 WT cells ( Figure 2B). Using the Nr2F1 promoter numbering system relative to the P1 promoter described by Salas and colleagues (55), RA increased the reporter activity of the À441 bp promoter/luciferase construct by 3.4-fold ( Figure 2A). A 5 0 -truncation to À382 bp reduced the RA induction to <2-fold. Constructs with further deletions of the Nr2F1 gene displayed no statistically significant RA induction ( Figure 2B). These results suggest the presence in Nr2F1 of an RA-inducible enhancer element (IEE) positioned at 383-441 bp upstream of the P1 promoter (2.4 kb upstream of the P3 RefSeq promoter).
We next wanted to evaluate the functionality of the P1 promoter in ES cells. Using publically available RNA-seq data, we performed bioinformatic analysis of transcription in RA-treated ES cells ( Figure 2C). This analysis revealed transcriptional activity upstream of the Nr2F1 RefSeq promoter (P3), thereby confirming upstream promoter activity also in ES cells. Interestingly, Nr2F2 displayed similar transcriptional activity upstream of the RefSeq promoter, whereas both Hoxa5 and Cyp26a1 displayed only background levels of transcription upstream of their respective RefSeq promoters ( Figure 2C).
DNA methylation, which occurs at CpG dinucleotides, is an epigenetic modification that can potently repress transcription. Regulation by DNA methylation is particularly pronounced in genes with CpG-rich promoter regions. We therefore evaluated the CpG content proximal to the RefSeq TSS of Nr2F1, Nr2F2, Hoxa5 and Cyp26a1. The 1000 bp regions upstream of the RefSeq TSS of Nr2F1 and Nr2F2 displayed a higher density of CpG dinucleotides ($100 CpGs) as compared with the Hoxa5 and Cyp26a1 promoters, which possess a modestly lower CpG density (57 and 52, respectively) as compared with a random 1000 bp sequence ($62 CpGs) ( Figure 2C). As promoter proximal CpG sites are commonly regulated by DNA methylation, we evaluated the DNA methylation status of candidate regions in the Nr2F1 gene. We found that the Nr2F1 IEE (P1/P2) and the RefSeq promoter (P3) displayed only minimal DNA methylation both before and after RA treatment of F9 WT cells (Supplementary Figure S2). Thus, changes in DNA methylation are not involved in the changes in Nr2F1 transcript levels after RA addition.
RA modifies histone marks and PRC association with the Nr2F1 and Hoxa5 genes We next examined the effects of RA on the epigenetic signatures at the Nr2F1 gene in F9 embryonal carcinoma stem cells. As discussed above, three distinct Nr2F1 TSSs have been detected in differentiating ES cells (Figures 2C  and 3A). We measured the levels of both permissive and repressive histone marks at the IEE (P1/P2) and the RefSeq promoter (P3). To compare Nr2F1 with a previously identified direct target of RA, we analyzed the epigenetic marks at the Hoxa5 proximal promoter and retinoic acid response element (RARE) in parallel (54). We found that in F9 WT cells there was an RA-dependent, >2-fold increase in the levels of Suz12 both at the IEE and at the RefSeq promoter (P3) of the Nr2F1 gene, whereas Suz12 levels were decreased at the Hoxa5 gene ( Figure 3B, left). We observed parallel changes in the PRC2-associated H3K27me3 mark and in the levels of Ring1B, a core component of the PRC1 repressive complex, at both the Nr2F1 and the Hoxa5 genes ( Figure 3B, left). Notably, for the Nr2F1 gene, the increase in response to RA was observed at both the RefSeq promoter (P3) and the IEE (P1/P2). Similarly, RA-associated increases in the PRC repressive epigenetic marks occurred in the three F9 RAR knockout lines, RAR À/À , RAR 2 À/À and RAR À/À at the Nr2F1 gene ( Figure 3B, right). To ascertain that RA was not perturbing the PRC2, we evaluated the association of Suz12 with the PRC2 catalytic subunit EZH2 (Supplementary Figure  S3). We found that RA had no effect on the overall interaction between Suz12 and EZH2 core components. Transcriptional activation is often associated with a decrease in epigenetic repressive marks, as we observed for the Hoxa5 gene ( Figure 3B, left), and for Cyp26a1, Hoxa1 and RAR 2 genes (22). Despite the increased levels of PRC repressive marks at the Nr2F1 RefSeq promoter (P3) and IEE (P1/P2) after RA addition, we observed an increase in Nr2F1 transcript levels in response to RA ( Figure 1A and C).
We next examined the H3K9/K14ac and H3K4me3 histone modifications associated with transcriptional activation and the levels of Ser5-phosphorylated (transcriptionally poised) RNA polymerase II (polII-CTD). The Nr2F1 RefSeq promoter (P3) displayed increased levels of the H3K9/K14ac and H3K4me3 marks (1.9-fold and 2.1-fold, respectively) on RA treatment of F9 WT cells ( Figure 3C, left). Likewise, a 2-fold increase in the polII-CTD level was detected at the Nr2F1 RefSeq promoter (P3) on a 24 h RA treatment of F9 WT cells ( Figure 3C, The numbers refer to the 5 0 terminal nucleotide included in each construct with respect to the upstream transcription start site (P1-TSS). The nucleotide sequence of the À441 to 382 region, which is required for a potent RA induction (>2-fold), is shown at the bottom of the figure. Note that in the bar diagram only constructs with the P1-TSS located proximal to the luciferase coding region are shown. Luciferase (firefly) reporter constructs containing the indicated regions of the Nr2F1 sequence were transfected into F9 WT cells, and assayed for RA responsiveness. Relative luciferase activities were expressed as a ratio over the untreated F9 WT cells transfected for each promoter construct, and statistical significance was determined. Data are compiled from at least three independent experiments. **P < 0.05. (C) Transcriptional activity at the Nr2F1, Nr2F2, Hoxa5 and Cyp26a1 genes assessed in publically available RNA sequencing data set generated from mouse ES (CCE) cells treated with RA for 5 days (GSM566812) (56). The number of CpG dinucleotides within the 1000 bp sequence upstream of the RefSeq promoter is indicated in the left side of the figure. The Nr2F1, Nr2F2, Hoxa5 and Cyp26a1 reference transcripts are indicated as blue boxes below the RNA-seq intensity data (white arrow heads). Notice the high transcriptional activity extending >1 kb upstream of the Nr2F1 and Nr2F2 reference sequence TSSs (P3). This contrasts with the low transcriptional activity upstream of the Hoxa5 and Cyp26a1 reference sequence TSSs. The genomic location relative to the RefSeq TSS is indicated at the bottom.  left). In contrast, we observed only modest changes (<1.5fold) in the levels of polII-CTD and in the levels of the permissive marks H3K9/K14ac and H3K4me3 at the Nr2F1 IEE in response to RA treatment ( Figure 3C, left). Consequently, the RA-associated increase in transcriptionally permissive marks was more pronounced at the RefSeq promoter (P3) than at the IEE (P1/P2) of the Nr2F1 gene. At the Hoxa5 gene, we observed an RAassociated increase in the H3K9/K14ac mark at the promoter and RARE (1.6-and 1.8-fold, respectively), and a large increase in polII-CTD association (>5-fold) with the promoter and the RARE (Figure 3C, left). Also RA-associated increases in the H3K4me3 mark at the promoter and RARE (1.2-and 1.4-fold, respectively) were detected at the Hoxa5 promoter and RARE ( Figure 3C, left). We observed an RA-associated increase in the levels of H3K9/14ac in RARa knockout cells, whereas the H3K9/14ac levels did not increase in RARb 2 and RARg knockout cells ( Figure 3C, right). The RA-associated increases in H3K4me3 and polII-CTD at the Nr2F1 gene in the RAR knockout cells were similar to those observed in F9 WT cells ( Figure 3C). The histone density at the Nr2F1 gene was not affected by RA, whereas RA led to nucleosomal depletion at the Hoxa5 gene ( Figure 3D, bottom). We conclude that the increase in Nr2F1 transcriptional activity following a 24 h RA treatment of F9 WT cells occurs even when epigenetic repressive factors such as Suz12, Ring1B and the H3K27me3 mark are present in Nr2F1 regulatory regions.
Suz12 knockdown reveals that RA differentially regulates the transcription of Nr2F1 and Hoxa5 genes The RA-induced changes in PRC association with the Nr2F1 promoter were dramatically different from those of the Hoxa5 promoter ( Figures 1B and 3B), yet both genes display increased transcriptional activity in response to RA (as evident by increased promoter association with permissive histone marks and by increased transcript levels). To define further the role of the Suz12 protein, a key protein component of the PRC2 polycomb repressive complex, we generated knockdown stem cell lines depleted of Suz12 transcripts. We performed the shRNA-induced Suz12 depletion in the F9 WT cell line and in each of the RAR knockout cell lines. Western blot analyses confirmed that the levels of Suz12 protein (and the levels of K27me3 modified histone 3) in each of the Suz12 knockdown lines were reduced by >90% relative to each of the shLuc (control) transfected parent cell lines (Figure 4). We then evaluated the effects of Suz12 depletion on Nr2F1, Nr2F2, Hoxa5 and Hoxa1 transcript levels in each of these cell lines after 24 h of RA treatment. The Suz12 depletion increased the RA-responsiveness of Nr2F1 and Nr2F2 in F9 WT, whereas we detected no significant effects on the RAresponsiveness of Hoxa5 and Hoxa1 (Figure 4). In the RAR knockout lines, we observed similar effects of Suz12 depletion on these four genes (Figure 4).
We next examined the effects of Suz12 depletion on additional RA responsive genes in WT ES cells. We found that the knockdown of Suz12 by shRNA technology dramatically increased the Nr2F1, Nr2F2, Meis1, Sox9 and BMP2 transcript levels in response to RA ( Figure 5A). In contrast, Hoxa1, Hoxa5, Cyp26a1, Cyp26b1 and RAR 2 transcript levels were not significantly affected by the Suz12 depletion in both control and RA-treated cells ( Figure 5A). The levels of Suz12 transcripts in ES cells stably transfected with shSuz12 constructs were significantly reduced compared with the levels in shLuc (control) transfected ES cells ( Figure 5B). As expected, the transcript levels of the 36B4 reference gene did not change in response to RA ( Figure 5B). An RAdependent increase in Nr2F1 protein was observed on Suz12 depletion, whereas the levels in control cells were below detection ( Figure 5C). The western blot analysis also confirmed that the total levels of Suz12 protein do not change in response RA. Furthermore, the Suz12 knockdown reduced protein levels by >90% relative to control ES cells ( Figure 5C). The depletion of Suz12 also reduced the levels of EZH2 protein ( Figure 5C).
Overall, our data suggest differential PRC2 dynamics in RA-induced transcription of the Nr2F1, Nr2F2, Meis1, Sox9 and BMP2 genes versus Hoxa5, Hoxa1, Cyp26a1, Cyp26b1 and RAR 2 genes. Specifically, Suz12 exhibits an inhibitory effect on the RA-associated increases in Nr2F1, Nr2F2, Meis1, Sox9 and BMP2 transcript levels, but not on the RA-associated increase in Hoxa5, Hoxa1, Cyp26a1, Cyp26b1 and RAR 2 transcript levels ( Figure 5A). Importantly, whereas the Nr2F1 promoter shows increased PRC2 association on RA treatment ( Figures 1B and 3B), the Hoxa5, Hoxa1, Cyp26a1 and RAR 2 promoters all respond to RA addition by dissociation of the PRC2 component ( Figures 1B and 3B and previous publications), thus potentially explaining the more rapid transcriptional induction by RA of this latter group of genes.
In Silico ChIP-seq analysis of RARc, RXRa and RNA polII association in F9 WT cells A recently published ChIP-seq analysis of RARg, RXRa and RNA polII in F9 WT cells (57) allowed us to expand our analysis of the epigenetic signature of RA responsive genes. We analyzed the RARg, RXRa and RNA polII chromatin association with the PRC2-attenuated RA target genes Nr2F1, Nr2F2 and Sox9, and with the nonattenuated RA target genes Hoxa5 and Cyp26a1 at various time points (Supplementary Figure S4). The ChIP-seq analysis confirmed recruitment of RNA polII to the Nr2F1, Nr2F2, Sox9, Hoxa5 and Cyp26a1 proximal promoter regions in response to RA treatment (Supplementary Figure S4). Importantly, with respect to Hoxa5 and Cyp26a1, RNA polII was not only recruited to the proximal promoter regions in response to RA but was also detected distributed throughout the length of each gene, consistent with highly active transcription. In contrast, recruitment of RNA polII appeared to be lower at both the promoters (as observed in Figure 3C) and throughout the Nr2F1, Nr2F2 and Sox9 genes in the presence of RA (Supplementary Figure S4).
The co-localization of RARg and RXRa at specific genomic locations is strong evidence for an RARE. Indeed, the ChIP-seq analysis confirmed the previously identified RAREs in Hoxa5 and Cyp26a1 (indicated in Supplementary Figure S4). The Cyp26a1 distal RARE showed association with both RARg and RXRa, whereas the proximal RARE was predominantly associated with RXRa. This may indicate that the proximal RARE is primarily bound by RARa or RARb, whereas the distal RARE is primarily bound by RARg. Neither RARg nor RXRa was associated with the Nr2F1, Nr2F2 and Sox9 genes (Supplementary Figure S4). This suggests that Nr2F1, Nr2F2 and Sox9 are secondary targets in the RA signaling cascade. Alternatively, Nr2F1, Nr2F2 and Sox9 could be targeted by different RAR and RXR isotypes (e.g. RARa or RARb dimerized with RXRb or RXRg).

DISCUSSION
In the current study, we have identified a novel class of PRC target genes, represented by Nr2F1, Nr2F2, Meis1, Sox9 and BMP2, that is activated during RA-induced stem cell differentiation. The RA-associated transcriptional activation of Nr2F1 occurs along with both increases in PRC and H3K27me3 repressive marks and permissive histone marks such as H3K4me3. The differential effects of Suz12 depletion on Nr2F1/Nr2F2 and Hoxa5/Hoxa1 (Figures 5 and 6) point to PRC2 dynamics as a distinguishing feature between the canonical direct RA target genes (rapidly induced) and a class of attenuated RA target genes (slowly induced). PRC2 was identified as a repressor of Hox gene transcription in Drosophila (58,59). It it was thus somewhat surprising that functional depletion of PRC2 in murine ES cells did not enhance transcription activation of Hoxa1 (61) or Hoxa5 (Figure 4). The functional characterization of PRCs is further complicated by the variable PRC complex composition and the numerous PRC target genes (62,63). Several groups have evaluated individual polycomb components (10,64), and Pasini et al. established a role for the Suz12 polycomb protein in ES cell differentiation (8). They further identified distinct groups of PRC-regulated genes: (i) genes expressed in pluripotent ES cells and silenced in differentiated cells, (ii) genes transcriptionally activated on differentiation and (iii) genes that were initially activated and then silenced in differentiated cells. This classification was corroborated by a recent report in which Mendoza-Parra et al. distinguished between (i) genes silenced by RA, (iia) genes rapidly induced by RA and (iib) genes exhibiting delayed induction by RA (57). This genome-wide ChIPseq study involved algorithm-based grouping of RA responsive genes, thereby providing a blinded validation of the grouping. The emerging consensus is that genes expressed after differentiation show strong polycomb protein association in the pluripotent stem cell state, whereas genes expressed in stem cells show strong polycomb association in the differentiated state (8,57). Our findings, however, reveal that this broad consensus does not fit genes that display a delayed induction in response to RA (e.g. Nr2F1 and Nr2F2). We show here that such genes initially display an increase in PRC2 levels concurrent with transcriptional activation. We demonstrate that PRC2 attenuates transcription of Nr2F1 and describe the associated epigenetic changes (Figures 4-6). This provides new mechanistic insights into genes exhibiting delayed induction by RA i.e. Nr2F1 (iib) and genes rapidly induced by RA i.e. Hoxa5 (iia) during stem cell differentiation.
To understand better the regulatory regions and chromatin environment of the Nr2F1 gene, we compared the local epigenetic contexts of the Nr2F1 and Nr2F2 genes with those of Cyp26a1 and Hoxa5. The Nr2F1, Nr2F2, Cyp26a1 and Hoxa5 genes are embedded within bivalent chromatin regions associated with CpG islands (Figure 3 and Supplementary Figure S5), as is typically seen for genes, with the potential for transcriptional activation or repression depending on the ES cell differentiation state (4,65). The Nr2F1 and Nr2F2 promoter regions are enriched for CpG sites relative to the Cyp26a1 and Hoxa5 promoter regions ( Figure 2C). Mammalian promoters show a bimodal distribution based on CpG content (66,67), and blinded computational analyses have confirmed that key epigenetic histone marks differ among promoters of high versus low CpG content (68). RA resulted in enrichment of PRC proteins at the Nr2F1 promoter (high CpG content), but resulted in a decrease in PRC proteins located at the Hoxa5 promoter (low CpG content). This difference could be related to the presence or absence of a TATA-box, which specifies a distinct TSS only in promoters with low CpG content (69). The observation that Nr2F1 and Nr2F2 use loosely defined TSSs, whereas Cyp26a1 and Hoxa5 each uses a clearly defined TSS ( Figure 2C), further supports this grouping of genes into different functional classes.
Differences in the epigenetic environment of the Nr2F1/ Nr2F2 versus the Cyp26a1/HoxA5 genes in untreated conditions could potentially explain the differential response to RA, i.e. slow versus rapid induction ( Figure 1A). In the absence of RA (untreated ES cells), Nr2F1, Nr2F2, Cyp26a1 and HoxA5 are transcriptionally silent ( Figure  5). We therefore compared the epigenetic signatures in untreated ES cells using publicly available ChIP-seq data (Supplementary Figure S4). The key role of PRC2 ( Figures 1B and 3B) prompted us to further evaluate PRC2 associated epigenetic marks (H3K27me3, EZH2, Suz12 and Ring1B) and H3K4 methylation of Nr2F1, Nr2F2, Cyp26a1 and HoxA5 genes. Our bioinformatics analysis suggests that in the absence of RA, the epigenetic distribution of H3K27me3 and PRC1/2 is lower at the Nr2F1 and Nr2F2 genes in comparison with the Cyp26a1 and HoxA5 genes in ES cells. The RefSeq promoter regions of Nr2F1 and Nr2F2 show a localized decrease of H3K27me3 on either side of the TSS, which extends to a localized decrease in H3K4 methylation. In contrast, in the absence of RA, the Hoxa5 and Cyp26a1 proximal promoter regions show high levels of H3K27me3 and H3K4me3 and decreased H3K4 methylation proximal to the TSS, which may be the result of nuclesome depletion around the TSS. Interestingly, RNA-seq data ( Figure 2C) suggest that whereas the TSSs of Hoxa5 and Cyp26a1 are clearly defined, Nr2F1 and Nr2F2 each may use a number of alternative sites for transcriptional initiation. Consistent with these data, the RA responsive region of Nr2F1 (IEE, Figure 2) is enriched for the H3K4me1 modification (70), progressively transitioning on RA addition to H3K4me2/3, which co-localizes with H3K27me3 and Jarid2a upstream of the TSS (Supplementary Figure S4). These features are similar to the features recently described for PRC-associated permissive enhancers (14), which permit cell type-specific transcriptional activation of PRC-repressed stem cell genes depending on the promoter context. The attenuated induction of Nr2F1 may involve recruitment of Ring1B and/or Jarid2, which has been reported to introduce pausing of polII at loci primed for future transcriptional activation (64,71). Indeed, we found that Ring1B was recruited to the IEE and RefSeq promoter of Nr2F1 in response to RA ( Figure 3B). The RAR 2 and RAR knockout cell lines displayed reduced induction of Nr2F1 by RA ( Figure 1C). However, polII was recruited even more efficiently in RAR À/À cells than in WT cells ( Figure 3C, right). One explanation could be that the absence of RARb 2 reduces polII recruitment, whereas the absence of RARg affects the initiation of transcriptional elongation. The RAassociated increase in H3K4me3 levels observed at the Nr2F1 gene in both the RARb 2 and RARg knockout cell lines suggests that RA induces a partially permissive chromatin structure ( Figure 3C). In contrast, we observed no increase in H3K9/14ac levels at the Nr2F1 gene in the RARb 2 and RARg knockout cell lines, which suggests that histone acetylation is required for full transcriptional activation of Nr2F1.
The role of Nr2F1 in RA-associated endodermal differentiation of ES cells (35) and the potential of RA to function as a ligand for Nr2F2 (37), and as a regulator of mesenchymal differentiation (38) point to the Coup-TFs as key players in RA-induced differentiation. Nr2F1 basal transcription is mediated by three ETS response Figure 6. Summary model of Nr2F1 and Hoxa5 epigenetic signatures in response to RA. Nr2F1 and Hoxa5 display different epigenetic signatures on RA treatment of stem cells. Nr2F1 is characterized by increased levels of PRC (Suz12 and Ring1B) and the associated H3K27me3 histone mark in response to RA. In contrast, the Hoxa5 epigenetic signature is characterized by dissociation of PRC (Suz12 and Ring1B) and reduction of the H3K27me3 histone mark. Note that the epigenetic signatures of Hoxa5 are similar in presence and absence of Suz12 after RA addition. In contrast, the epigenetic signature of Nr2F1 on RA treatment differs in the presence and absence of Suz12, thus potentially explaining the increased transcriptional activity on Suz12 knockdown. For both Nr2F1 and Hoxa5, the transcriptional induction is marked by increased levels of H3K4me3, H3K9ac and H3K14ac permissive histone marks. The different marks are depicted as shapes whose sizes reflect the relative abundance in the specified condition. Histone H3 is depicted as a gray circle. PRC1 and PRC2 are represented by Ring1B and Suz12, respectively. elements (55), yet the DNA elements within the Nr2F1 promoter that mediate the RA responsiveness of Nr2F1 are unknown (27,34). The Nr2F1 promoter region contains no consensus RARE (DR2 or DR5), and thus far no functional element has been identified that can explain the RA responsiveness of Nr2F1 (Supplementary Figure S4). This is the first study identifying the Nr2F1 promoter region that is responsible for the RA induction. The complex dynamics of PRC association in response to RA suggest that several transcription factors cooperate to regulate the transcription of the Nr2F1 gene. Thus, it will be important to identify the cis-regulatory DNA elements responsible for the RA-associated PRC recruitment and to characterize the transcription factors that recognize these DNA elements.
Here we identify PRC2 dynamics as a distinguishing feature between two classes of PRC2 target genes represented by the Nr2F1 (Coup-TF1) and the Hoxa5 gene, respectively. We conclude that PRC1/2, in addition to specifying transcriptional repression, can function to attenuate transcriptional activation by RA of specific genes during stem cell differentiation. Attenuation of the maximal transcriptional activation may allow for more exquisite precise regulation of commitment to a specific differentiation pathway.