Short sequences can efficiently recruit histone H3 PNAS PLUS lysine 27 trimethylation in the absence of enhancer activity and DNA methylation

Philip Jermanna,b, Leslie Hoernera, Lukas Burgera,c, and Dirk Schübelera,b,1

aFriedrich Miescher Institute for Biomedical Research, 4058 Basel, Switzerland; bFaculty of Science, University of Basel, 4053 Basel, Switzerland; and cSwiss Institute of Bioinformatics, 4056 Basel, Switzerland

Edited by Mark Groudine, Fred Hutchinson Cancer Research Center, Seattle, WA, and approved July 16, 2014 (received for review January 14, 2014) Trimethylation of histone H3 at lysine 27 (H3K27me3) is a chroma- recruitment (18, 19). In mammals, PRC2 and the H3K27me3 mark tin mark associated with Polycomb-mediated repression. De- localize mainly to transcriptionally inactive regions rich in CpG spite its critical role in development, it remains largely unclear how dinucleotides, referred to as CpG-islands (CGIs) (20–23). CGIs are this mark is targeted to defined loci in mammalian cells. Here, we unique to vertebrate genomes that show global DNA methylation use iterative genome editing to identify small DNA sequences ca- (24). These regulatory regions make up two-thirds of all promoters pable of autonomously recruiting Polycomb. We inserted 28 DNA and are under high selection for the presence of regulatory motifs elements at a defined chromosomal position in mouse embryonic more complex than CG, which are occupied by TFs (25). It thus stem cells and assessed their ability to promote H3K27me3 deposi- remains critical to understand if high CpG dinucleotide frequency tion. Combined with deletion analysis, we identified DNA elements is indeed sufficient to direct PRC2 recruitment. as short as 220 nucleotides that correctly recapitulate endogenous Here we aimed to identify DNA elements that can autono- H3K27me3 patterns. Functional Polycomb recruiter sequences are mously establish an H3K27me3 domain and dissect their function invariably CpG-rich but require protection against DNA methyla- through iterative testing of sequence variants. To control for po- tion. Alternatively, their activity can be blocked by placement of sition effect and copy number, we used recombination-mediated an active promoter-enhancer pair in cis. Taken together, these data cassette exchange (RMCE) to insert DNA elements into the same GENETICS support the model whereby PRC2 recruitment at specific targets in genetic . These repeated insertions identify DNA fragments mammals is positively regulated by local CpG density yet ob- as short as 220 nt capable of autonomously acquiring H3K27me3 structed by transcriptional activity or DNA methylation. in a heterologous genomic context. This ability is determined by the presence of unmethylated CpG dinucleotides, yet the cell- olycomb group (PcG) are required for proper fine- type–specific acquisition of H3K27me3 during differentiation is Ptuning of gene expression and act in part through mod- dependent on the presence of a cell-type–specific enhancer. ifications of histones (1–3). Initially identified in Drosophila melanogaster as regulators of homeobox gene expression and Results body segmentation, PcG proteins are also essential in mammals Small DNA Elements Autonomously Recruit PcG Proteins and Establish and implicated in many cellular processes, including mainte- an H3K27me3 Domain When Placed at an Ectopic Locus in ES Cells. nance of pluripotency and lineage-commitment (4, 5). PcG Candidate recruiter sequences for PRC2 were chosen based on proteins form two main complexes, PRC1 and PRC2 (6). PRC1 their enrichment for H3K27me3 and SUZ12 in genome-wide monoubiquitylates histone H2A lysine 119 (H2AK119), and ChIP datasets from murine ES cells (26, 27). Because the ma- PRC2 harbors histone methyltransferase activity toward histone jority of SUZ12 peaks lie within CGIs (20, 21), we asked if H3 lysine 27 (H3K27) (7, 8). Both chromatin marks are associ- candidate CGIs can act autonomously to promote H3K27me3 ated with gene repression and are essential for embryonic stem when inserted in a heterologous genomic environment. We (ES) cell differentiation (1, 9). Mammalian PRC2 consists of three core proteins essential for its catalytic activity: En- Significance hancer of zeste homolog 2 (EZH2), Suppressor of zeste homolog (SUZ12), and Embryonic ectoderm development (EED). Ge- Polycomb repressive complex 2 functions in gene repression nome-wide analysis of the H3K27me3 mark in different cell and acts by methylating histone H3 at lysine 27 (H3K27me3). types and during in vitro differentiation of ES cells revealed the Despite its relevance, it remains elusive how this complex mark to be dynamic during lineage-commitment, suggesting that is recruited to its target sites in the genome. Here, we used the complex is recruited to target sites in a cell-type–specific repeated genomic targeting in embryonic stem cells to identify manner (10, 11). Canonical PRC1 consists of RING1A/B, sev- DNA sequence determinants that autonomously confer eral chromodomain proteins (CBX), one of six polycomb ring H3K27me3 recruitment. We show that surprisingly small CG- finger proteins (PCGF), and three different polycomb-like rich DNA sequences are sufficient to recruit H3K27me3, but (PCL) proteins (12). It has been demonstrated that canonical only if they are devoid of DNA methylation and transcriptional PRC1 is recruited to sites marked by H3K27me3 via the CBX activity. This study provides new insights into the mechanisms proteins that specifically interact with the methylated lysine (9). recruiting H3K27me3 and the cross-talk between diverse PRC1-bound sites are often ubiquitylated on lysine 119 of histone chromatin modifications. H2A (H2AK119ub) (13), a chromatin modification that can impart gene repression through inhibition of the transcrip- Author contributions: P.J. and D.S. designed research; P.J. and L.H. performed research; P.J., tional machinery (14, 15). L.H., and L.B. analyzed data; and P.J. and D.S. wrote the paper. The mechanisms by which PcG proteins are recruited to spe- The authors declare no conflict of interest. cific genomic sites are still elusive. In D. melanogaster, transgenic This article is a PNAS Direct Submission. experiments showed that Polycomb can be recruited to polycomb Freely available online through the PNAS open access option. responsive elements (PREs), leading to repression of reporter 1To whom correspondence should be addressed. Email: [email protected]. – (16 18). Several transcription factor (TF) binding motifs This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. are enriched within fly PREs and might contribute to Polycomb 1073/pnas.1400672111/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1400672111 PNAS Early Edition | 1of7 Downloaded by guest on September 25, 2021 chose eight candidate CGIs, ranging in size from 550 nt to 1,000 recombination (28) (Fig. 1A). Importantly, the globin locus and nt, and included elements that are putative promoters as well as several hundred kilobases of surrounding region contain no elements located within exons and intergenic regions (Fig. 1 B–E preexisting H3K27 methylation in stem cells (Fig. S1A), making and Fig. S1 B–E). In addition, we inserted a ninth CGI over- this site suitable to test for the autonomous ability of DNA lapping with the promoter of the exosome complex component elements to recruit this chromatin mark. Moreover, the RMCE gene Exosc9. This element does not harbor H3K27me3 at its approach relies only on negative selection and the preexisting endogenous site and served as a negative control (Fig. 1F). marker gene is removed during the insertion, leaving behind only All nine elements were inserted separately into a previously the sequence of interest flanked by loxP sites (Fig. 1A). Fol- described target site in the β-globin locus using CRE-mediated lowing verification of correct insertion, we tested for the presence of H3K27me3 at all elements by ChIP. For each of the eight putative recruiter elements, we observed the acquisition of A B CGI11 H3K27me3 after insertion at the globin locus (Fig. 1 B–E). This recruitment is specific because no enrichment was observed at 2.5kb 20 Suz12 the element that served as a negative control (Fig. 1F), whereas 0 the endogenous positive control showed consistent H3K27me3 CGI 10 H3K27me3 levels in all generated stem cell lines (Fig. S2J). Importantly, the 0

coverage / 100bp Lhx6 H3K27me3 levels detected at the eight functional elements are CGIs comparable to those observed at their endogenous loci, consis-

CGI11 (780nt) tent with full sequence autonomy in recruitment of this chro- + - 5’ 3’ matin mark. This finding is particularly intriguing in light of the β-globin locus 1.2 H3K27me3 Suz12 Ring1B relatively small sizes of tested sequences that range from 542 to 1 901 bp, and demonstrates that elements of less than 1 kb can 0.8 0.6 establish a local H3K27me3 domain. To control for the possi- 0.4 bility that ectopic H3K27me3 enrichment merely reflects in- CGI 0.2 creased histone H3 abundance, we measured the levels of total 0 IP / Input rel. to Evx2 5' 3' 5' 3' H3 at both the endogenous and ectopic sites. All tested se- vx2 E Hprt ectopic endog. quences revealed no significant differences in nucleosome oc- cupancy between ectopic and endogenous sites. Thus, differential C CGI6 D CGI10 H3 occupancy cannot explain H3K27me3 enrichment signals 5kb – 20 Suz12 10kb 20 Suz12 (Fig. S2 A I). We next sought to test for presence of Polycomb components 0 0 10 H3K27me3 10 H3K27me3 at the ectopic sites capable of establishing H3K27me3 domains.

0 0 All insertion elements with detectable H3K27me3 recruitment coverage / 100bp coverage / 100bp Otp were also enriched for the PRC2 component SUZ12 (Fig. 1 B–E CGIs CGIs and Fig. S1 B–E). Furthermore, five of eight K27 recruiter ele- CGI6 (901nt) CGI10 (640nt) ments displayed SUZ12 levels comparable to that of their en- 5’ 3’ 5’ 3’ dogenous counterpart. We also tested for the presence of H3K27me3 Suz12 Ring1B H3K27me3 Suz12 Ring1B 1.2 1.2 RING1B at the inserted elements (Fig. 1 B–F and Fig. S1 B–E). 1 1 Indeed, the majority of elements displayed detectable levels of 0.8 0.8 0.6 0.6 RING1B, which positively correlate with the presence of SUZ12 0.4 0.4 (Fig. S1F). Although we detect Polycomb components at ectopic 0.2 0.2 H3K27me3 recruiter fragments, the strength of Polycomb en- 0 0 IP / Input rel. to Evx2 2 IP / Input rel. to Evx2 5' 3' 5' 3' richment is often less than what is seen at the endogenous x 5' 3' 5' 3' vx2 v Hprt E Hprt ectopic endog. E ectopic endog. equivalent. Whereas this finding may indicate that the tested DNA elements do not fully stabilize Polycomb complex binding to E CGI13 F CGI14 chromatin, this recruitment is nevertheless sufficient for the es-

20 5kb tablishment of an H3K27me3 domain. Another possible expla- 20 Suz12 Suz12 10kb nation for the lower levels of Polycomb components at the ectopic 0 0 10 site lies in how enrichment can be quantified between ectopic and 10 H3K27me3 H3K27me3 endogenous amplicons. To distinguish ectopic from endogenous 0 0 coverage / 100bp

coverage / 100bp B3gnt6 Exosc9 enrichment, one primer must be placed outside the recruiting CGIs CGIs sequence, thereby limiting quantification to the border of the el-

CGI13 (772nt) CGI14 (442nt) ement to be tested (Fig. 1A). This discrepancy is far less pro- 5’ 3’ 5’ 3’ nounced regarding H3K27me3, an observation in agreement with 1.2 H3K27me3 Suz12 Ring1B 1.2 H3K27me3 Suz12 Ring1B broad distribution patterns detected in H3K27me3 ChIP datasets. 1 1 to Evx2 0.8 0.8 Regardless, all tested CGIs are able to autonomously establish 0.6 0.6 an H3K27me3 domain and recruit Polycomb components. This 0.4 0.4 observation suggests that local sequence autonomy may be a gen- 0.2 0.2 eral feature of Polycomb recruitment in the mouse genome. We 0

IP / Input rel. 0 2 5' 3' 5' 3' IP / Input rel. to Evx2 5' 3' 5' 3' next asked if further deletions could identify even smaller sequence vx2 Evx Hprt ectopic endog. E Hprt ectopic endog. elements as functional recruiters for the H3K27me3 mark.

Fig. 1. Short DNA sequences can establish H3K27me3 domains. (A) Sche- Small Functional PRC2 Recruiter Elements Cover Less than Two matic representation of RMCE workflow. (B–F, Upper) Suz12 and H3K27me3 Nucleosomes. profiles around the candidate (B–E) and control (F) CGIs. (Lower) H3K27me3, To dissect length requirements and potentially SUZ12, and RING1B ChIP-qPCR data at the ectopically inserted and endog- identify regions essential for PRC2 recruitment, we tested sub- enous loci. All qPCR data are shown relative to Evx2, a positive control for fragments of the identified recruiter elements ranging from 218 H3K27me3. Hprt was used as an endogenous negative control. Error bars to 445 nt. The resulting eight fragments were separately inserted represent SD of at least two biological replicates. and SUZ12 and H3K27me3 enrichments were measured (Fig. 2

2of7 | www.pnas.org/cgi/doi/10.1073/pnas.1400672111 Jermann et al. Downloaded by guest on September 25, 2021 ADB C PNAS PLUS CGI10 (640nt) CGI11 (777nt) CGI3 (665nt) CGI20 (542nt) 10.1 (318nt) 11.1 (352nt) 3.1 (362nt) 20.1 (321nt) 10.2 (312nt) 11.2 (445nt) 3.2 (328nt) 20.2 (218nt) 1.2 1.2 1.2 1.2 1 1 1 1 0.8 0.8 0.8 0.8 0.6 0.6 0.6 0.6 0.4 0.4 0.4 0.4 0.2 0.2 0.2 0.2 0 IP / Input rel. to Evx2 IP / Input rel. to Evx2 IP / Input 0 rel. to Evx2 IP / Input 0 rel. to Evx2 IP / Input 0 5' 3' 5' 3' 5' 3' 5' 3' 5' 3' 5' 3' 5' 3' 5' 3'

Evx2 Hprt ectopic endog. Evx2 Hprt ectopic endog. Evx2 Hprt ectopic endog. Evx2 Hprt ectopic endog.

Fig. 2. Dissection of H3K27me3 recruiters identifies 220 nt recruiter sequence. (A–D) Recruiter elements were dissected into smaller parts and analyzed for H3K27me3 enrichments by ChIP-qPCR. All data are shown relative to Evx2. Error bars represent SD between at least two biological replicates.

and Fig. S3). Enrichments varied between elements, indicating Factor 1 (Utf1) gene-body because it harbors no detectable that these smaller elements do not show a uniform ability to H3K27me3 in stem cells but gains the mark during neuronal recruit a functional polycomb complex. For example, the 3′ half differentiation (Fig. 3A). Remarkably, when inserted in the of CGI3 shows no recruitment of H3K27me3, but the 5′ half β-globin locus, however, this CGI recruits the H3K27me3 mark shows strong recruitment, despite the fact that both are com- already in stem cells even though it is absent at the endogenous parable in size (Fig. 2C and Fig. S3C). Importantly these in- locus (Fig. 3B). Similarly, both SUZ12 and RING1B are sertion experiments show that elements as short as 218 nt can be enriched only at the ectopic locus, but not endogenously (Fig. sufficient to create a local H3K27me3 domain (Fig. 2D and Fig. 3B). Thus, outside of the native genomic locus, this sequence S3D). Thus, DNA regions spanning less than two nucleosomes recruits the Polycomb machinery in a cell-type–independent can be sufficient to recruit PRC2 components and establish manner. The Utf1 locus codes for an established pluripotency

H3K27 methylation patterns similar to their endogenous site. factor (29) and is exclusively expressed in ES cells (30). Upon GENETICS differentiation into neuronal progenitors, Utf1 expression is shut Cell-Type–Specific Reconstitution of H3K27me3 Patterns Requires off, accompanied by the gain of H3K27me3 (Fig. 3A). Expression Endogenous Regulatory Context. The above experiments argue of the Utf1 gene is regulated by a proximal downstream enhancer that CGIs that recruit H3K27me3 in stem cells function similarly positioned adjacent to the CGI (31). To determine whether the when placed in an ectopic site. Because Polycomb recruitment is transcriptional context is required to regulate cell-type–specific variable between cell types, we next asked whether sequences H3K27me3 recruitment, we inserted the entire Utf1 locus in- that do not harbor H3K27me3 in stem cells but gain it during cluding the promoter and enhancer. Addition of these cis-regu- differentiation recapitulate this behavior when placed at an ec- latory elements resulted in loss of H3K27me3 acquisition in stem topic site. As a test sequence, we chose a 1.1-kb CGI overlapping cells, mimicking the endogenous locus (Fig. 3C). Furthermore, with the Undifferentiated Embryonic Cell Transcription upon in vitro differentiation of ES cells into neural progenitors

ΔTFB A B uCGI C Pro uCGI Enh D Pro uCGI Enh chr7:147,128,498-147,132,268 5’ 3’ 5’ 3’ 5’ 3’ 1.6 1kb 1.2 H3K27me3 H3K27me3 1.2 1.4 20 H3K27me3

Evx2 ES ES Sox2 1 Suz12 1 1.2 Ring1B NP 0.8 0.8 1 0 0.8 250 0.6 0.6 0.6 0.4 0.4 ES RNA 0.4 0 0.2 0.2 0.2

25 0 IP / Input rel. to Evx2 0

IP / Input rel. to Evx2 0 IP / Input rel. to 5' 3' 5' 3' rt 5' 3' 5' 3' rt 5' 3' 5' 3' ES H3K27me3 p Hprt ectopic endog. H Hp 0 Evx2 Evx2 ectopic endog. Evx2 ectopic endog. 20 E F ΔTFB G coverage / 10bp NP Sox2 Pro CGI10 Enh Pro CGI10 Enh Pro CGI14 Enh 0 12 3 4 12 3 4 12 3 4 250 NP RNA 1.2 ES 1.2 1.2 H3K27me3 NP H3K27me3 WT H3K27me3 ES 0 1 1 ∆TFB 1 NP 25 NP H3K27me3 0.8 0.8 0.8 0.6 0.6 0.6 0 0.4 0.4 0.4 0.2 0.2 0.2 CGI 0 0 0 IP / Input rel. to Evx2 IP / Input rel. to Evx2 IP / Input rel. to Evx2 12345'3' 2 rt 12345'3' t 12345'3' Pro Enh vx2 vx E Hprt ectopic endog. E Hp ectopic endog. Evx2Hpr ectopic endog. CGI10 CGI10 CGI14

Fig. 3. Transcriptional activity inhibits H3K27me3 acquisition. (A) Sox2, RNA, and H3K27me3 profiles around the Utf1 locus in ES cells and NPs. (B) Ectopic insertion of the Utf1 CGI and corresponding H3K27me3, SUZ12, and RING1B levels in ES cells. (C) Insertion of the whole Utf1 locus, including its promoter and enhancer and corresponding H3K27me3 levels in ES cells and NPs. (D) H3K27me3 enrichments at the Utf1 locus with the mutated enhancer in ES cells. (E) Insertion of CGI10 flanked by the Utf1 promoter and enhancer and corresponding H3K27me3 enrichments in ES cells and NPs. (F) Comparison of H3K27me3 levels between CGI10 in the context of the wild-type or mutant Utf1 enhancer in ES cells. (G)AsinE but with the negative control CGI14 placed between the Utf1 promoter and enhancer. All qPCR data are shown relative to Evx2. Error bars represent SD of at least two biological replicates.

Jermann et al. PNAS Early Edition | 3of7 Downloaded by guest on September 25, 2021 (NPs), we observed acquisition of H3K27me3 as well as SUZ12 H3K27me3 recruitment. Such a model has previously been and RING1B at the ectopic locus similar to the endogenous site, proposed based on the strong presence of H3K27me3 at CGIs indicating that the dynamic pattern of this chromatin mark is now (20, 21) and the observation that two CpG-rich elements from reconstituted at the ectopic site in both ES cells and NPs (Fig. 3C the Escherichia coli genome become H3K27-methylated when and Fig. S4 A and B). Again, there was no significant difference in placed into stem cells as part of larger transgenes (22, 23). The total H3 occupancy between ES cells and NPs, excluding that the identification of small endogenous sequences that can autono- observed changes in H3K27me3 are caused by increased histone mously recruit H3K27me3 enabled us to rigorously test this abundance (Fig. S4C). A potential explanation for the loss of model. To determine if the density of CpGs and their individual H3K27me in stem cells is that the presence of promoter and position is sufficient for recruitment, we synthetically modified enhancer leads to transcriptional activation, which in turn coun- three of the identified sequence elements (Fig. 4). Specifically, teracts Polycomb recruitment. To test this theory, we sought to we kept all CpG dinucleotides and their relative positions but reduce the transcriptional activity of the promoter-enhancer pair replaced the rest of the sequence with random stretches of pro- with minimal changes in its sequence. The Utf1 enhancer contains karyotic DNA from the E. coli genome. The purpose of this ap- an OCT4/SOX2 binding site (31), both key transcription factors proach was to remove sequence cues (e.g., TF motifs) other than of the pluripotency network active in stem cells. Genome-wide the CpG dinucleotides. Upon insertion, only one of three tested profiles of SOX2 in ES cells and NPs confirmed that the factor is elements recruited SUZ12 and showed H3K27me3 levels remi- bound to the Utf1 enhancer in ES cells, but not in NPs (32). We niscent of the wild-type fragment (Fig. 4A and Fig. S5A), whereas thus inserted a construct with a 15-nt deletion within the enhancer, the other two displayed strongly reduced levels (Fig. 4 B and C effectively removing the OCT4/SOX2 binding site while leaving the and Fig. S5 B and C). Thus, CpG frequency and position cannot enhancer otherwise unchanged. Expression from this construct was solely account for the ability to recruit H3K27me3. Notably, of analyzed by replacing the Utf1 gene with a luciferase reporter as the three modified recruiter elements above, the construct ca- transcripts from the endogenous Utf1 locus and ectopic site are pable of recruitment also has the highest CpG density, suggesting otherwise indistinguishable (Fig. S4D). After insertion, we com- that very high CpG density could indeed be sufficient (Fig. 4). pared expression of both constructs revealing that deletion of the We have previously suggested that TF binding within CGIs OCT4/SOX2 binding site abolishes transcription, leading to a 100- contributes to their low DNA methylation (28), which could be to 200-fold decrease in luciferase activity (Fig. S4D). To determine relevant for the recruitment of the H3K27me3 mark because oc- if this affects H3K27me3 acquisition, we next inserted the enhancer cupancy of DNA methylation and H3K27me3 in ES cells has been mutation in combination with the Utf1 gene. As expected, SOX2 observed to be mutually exclusive at CGIs (33), leading to the binding was strongly reduced at the mutated enhancer compared hypothesis that DNA methylation could inhibit H3K27me3 ac- with the wild-type construct (Fig. S4E). More importantly, how- quisition. To explore whether increased DNA methylation could ever, and in contrast to the unaltered enhancer construct, this account for the absence of H3K27me3 at the two elements with mutant now showed premature acquisition of H3K27me3 in stem modified sequence backbone, we first determined their DNA cells (Fig. 3D). Thus, the presence of the Utf1 promoter-enhancer methylation state. Bisulphite sequencing revealed that elements pair is necessary to inhibit H3K27me3 acquisition of the over- depleted in H3K27me3 upon alteration of the sequence backbone lapping CGI in stem cells and loss of enhancer activity is sufficient were indeed fully DNA-methylated (Fig. 4 B and C and Fig. S6 A to cause premature recruitment of H3K27me3 to the Utf1 CGI. and B). Conversely, their wild-type counterparts showed little to no DNA methylation upon insertion (84% vs. 19% and 94% vs. 17%, Transcriptional Context Arbitrates Recruitment Abilities. To test respectively). The third element retaining H3K27me3 showed very whether the antagonism between recruitment and transcription low methylation, even with the prokaryotic backbone (14% vs. 0%) was specific to the Utf1 recruiter element, we replaced the Utf1 (Fig. 4A and Fig. S6C). Thus, loss of Polycomb recruitment co- gene-body CGI between the promoter-enhancer pair with the incides with increased susceptibility to DNA methylation. This recruiter CGI10 (Fig. 1D). Insertion of CGI10 in the context of finding is compatible with the observation of increased H3K27 the Utf1 promoter-enhancer combination indeed inhibited methylation upon global loss of DNA methylation (23, 24, 33, 34), H3K27me3 acquisition in ESCs (Fig. 3E). Furthermore, upon which we also observe in our genome-wide datasets (Fig. S7A). differentiation into NPs, the CGI reacquired the H3K27me3 With this in mind, we also analyzed the DNA methylation of the mark concomitant with loss of transcriptional activity (Fig. 3E smaller fragments described above (Fig. 2C). Again, the smaller and Fig. S4F). Using mutated constructs, we also observed that fragment that lost the ability to recruit H3K27me3 became fully the OCT4/SOX2 binding site in the enhancer is essential for DNA-methylated (Fig. S6D). Taken together, these results sup- preventing H3K27me3 deposition in ES cells (Fig. 3F). To fur- port a model of competition between DNA methylation and ther verify that H3K27me3 recruitment in NPs is indeed medi- H3K27me3 recruitment at the tested CGIs. ated by the CGI10 and not the Utf1 promoter or enhancer, we Trimethylation of H3 at lysine 4 (H3K4me3) has been dem- repeated the experiment with the CGI10 replaced by the non- onstrated to co-occupy PRC2-bound CpG islands in ES cells recruiting CGI14 (Figs. 1F and 3G). This construct no longer resulting in bivalent domains (35). When determining H3K4me3 acquired H3K27me3 upon differentiation, confirming that the levels at the three elements above, CGI3 and CGI10 both dis- observed gain of this histone mark at the CGI10 construct was played low levels of H3K4me3 at the endogenous site and at indeed due to the presence of the inserted CGI. wild-type inserts (Fig. S5 E and F). This signal was reduced upon Thus, cell-type–specific recruitment of the H3K27me3 mark sequence mutation in agreement with observations that can be counteracted by transcriptional activation mediated by H3K4me3 is absent at methylated CGIs (36). In the case of surrounding regulatory regions. CGI6, the wild-type insert showed high levels of H3K4me3, whereas the signal in the mutant was reduced even though it DNA Methylation and H3K27me3 Recruitment Exclude Each Other at remains DNA unmethylated and maintains wild-type levels of CpG Islands. All tested recruiter sequence elements are CpG- H3K27me3 (Fig. 4A and Fig. S5D). One potential explanation is dense because they originate from CGIs, suggesting that CpG that mutation of the sequence backbone removed TF binding dinucleotides might play an important role in the recruitment of sites that positively affect H3K4 methylation at this sequence. the H3K27me3 mark to these target sites. In support of this model, we did not observe H3K27me3 at the CpG-free luciferase DNA Methylation Counteracts H3K27me3 Recruitment. If DNA methy- reporter construct described above, even in the inactive state lation at the two modified recruiter elements inhibits deposition of (Fig. S4G), indicating that CpG dinucleotides are required for H3K27me3, removing DNA methylation should recover H3K27me3.

4of7 | www.pnas.org/cgi/doi/10.1073/pnas.1400672111 Jermann et al. Downloaded by guest on September 25, 2021 PNAS PLUS A CGI6 (901 nt) B CGI10 (640 nt) C CGI3 (571 nt) WT WT WT

CGI6 (783 nt, 13.2 CpGs/100 nt) CGI10 (640nt, 6.4 CpGs/100 nt) CGI3 (571 nt, 8.4 CpGs/100 nt) mut mut mut

WT WT WT 1.6 1.2 1.2 1.4 mut 100 mut 100 mut 100 1 1 1.2 . -5 75 0.8 2 10 75 0.8 75 1 -4 * 0.004 1.10 * 0.8 0.6 * 50 0.6 0.046 50 50 * 0.6 0.4 0.4 0.4 25 25 25 0.2 0.2 0.2 % DNA methylation % DNA methylation IP / Input rel. to Evx2 IP / Input rel. IP / Input rel. to Evx2 IP / Input rel. IP / Input rel. to Evx2 IP / Input rel. 0 0 0 % DNA methylation 0 0 0 5' 3' 5' 3' rt 5' 3' 5' 3' 5' 3' 5' 3' Hp Hprt Evx2 Hprt ectopic endog. Evx2 ectopic endog. Evx2 ectopic endog.

Fig. 4. CpG dinucleotides cannot solely account for the establishment of H3K27me3 domains. Replacement of all non-CpG nucleotides in CGI6 (A), CGI10 (B), and CGI3 (C) with prokaryotic DNA. (Upper) CpG dinucleotide positions within the sequences and the color indicates the level of DNA methylation (white through red indicates 0–100%). (Lower) Bar plots show H3K27me3 ChIP-qPCR levels at the tested sequences (Left) and total DNA methylation percentages (Right). All qPCR data are shown relative to Evx2. Error bars represent SD between at least two biological replicates. Numbers above asterisks indicate P values.

To test this theory, we treated the respective cells with the DNMT local H3K27me3 domains. Furthermore, we could recapitulate inhibitor 5-Aza-2′-deoxycytidine (37), which resulted in loss of DNA H3K27me3 dynamics upon differentiation into neural precursors, methylation to levels closely reflecting those observed at wild-type demonstrating that the cellular signal for recruiting the Polycomb CGIs (Fig. S7 B and C). Strikingly, reacquisition of H3K27me3 levels machinery is encoded within the target DNA sequence. Of par- comparable to the corresponding wild-type sequences occurred co- ticular note, by genetically uncoupling the contribution of CpGs incident with reduced DNA methylation (Fig. 5 A and B). We thus from the rest of the DNA sequence we show that CpG frequency conclude that DNA methylation can block the ability of CpG-rich and positioning is sufficient to recruit PRC2 and establish an GENETICS DNA stretches to recruit the H3K27me3 mark. H3K27me3 domain. Additionally, we demonstrate that DNA The above results predict that a CGI that is not K27-methyl- methylation inhibits H3K27me3 deposition at CpG-rich seque- ated should be either DNA-methylated or transcriptionally ac- nces and that this inhibition can be reversed upon chemical re- tive. This result is indeed what we observe at the level of the moval of DNA methylation. These observations readily explain genome (Fig. S7D) and at the CGI that was used as a negative the occupancy of H3K27me3 and DNA methylation observed control and which did not recruit PRC2 (CGI14) (Fig. 1D). This throughout the genome (23, 24, 33, 34). They extend previous particular sequence is not DNA-methylated after insertion but studies that already suggested a role for CpG dinucleotides in recruits efficiently RNA polymerase, similar to its endogenous PRC2 recruitment (22, 23) but are also more comprehensive and site, again in line with this model (Fig. S6 E and F). provide novel mechanistic details. Interestingly, a recent study suggests that in vitro PRC2 activity is not directly inhibited by Discussion DNA methylation (38). It is thus possible that the antagonistic Using iterative genome editing, we define minimal DNA se- effect of DNA methylation on H3K27me3 deposition that we quence elements capable of recruiting PRC2 as well as PRC1 observe in vivo is dependent on additional factors, possibly the and acquire H3K27me3 in stem cells. The genomic locus used in presence of methyl-CpG–binding proteins (39). our studies is otherwise devoid of this mark, allowing us to identify Our data reveal that the transcriptional context surrounding DNA elements as short as 220 nt that autonomously create recruiter sequences is critical in H3K27me3 establishment. By

A CGI10 B CGI3 5-AzadC 5-AzadC

WT mut mut + 5-AzadC WT mut mut + 5-AzadC -5 1.6 -4 1.4 . -4 2.8.10 1.10 1 10 . -5 * * 1.4 * 5 10 1.2 * 1.2 100 100 1 1 75 0.8 75 0.8 0.6 0.6 50 50 0.4 0.4 25 25 0.2 0.2 IP / Input rel. to Evx2 rel. to IP / Input IP / Input rel. to Evx2 rel. to IP / Input % DNA methylation 0 % DNA methylation 0 0 0 Evx2Hprt5'3'5'3' Evx2Hprt5'3'5'3' ectopic endog. ectopic endog.

Fig. 5. Removal of DNA methylation reestablishes acquisition of H3K27me3 at the modified fragments. ES cells harboring the modified versions of fragments CGI10 (A) and CGI3 (B) were treated with 5-Aza-2′-deoxycytidinefor4d.(Upper) CpG dinucleotide positions within the sequences and the color indicates the level of DNA methylation (white through red indicates 0–100%). (Lower) Bar plots show H3K27me3 ChIP-qPCR levels at the tested sequences (Left) and total DNA methylation per- centages (Right). All qPCR data are shown relative to Evx2. Error bars represent SD between at least two biological replicates. Numbers above asterisks indicate P values.

Jermann et al. PNAS Early Edition | 5of7 Downloaded by guest on September 25, 2021 coupling enhancer-dependent activation to otherwise transcrip- pied by default in the absence of active transcription. Presence of tionally inert constructs, we were able to inhibit premature ac- the H3K27me3 mark may serve to suppress sporadic expression of quisition of the H3K27me3 mark. This result suggests that active target genes, while concurrently allowing for future transcriptional transcription prevents PRC2 binding, a finding confirmed by mini- activation. The dispensability of PRC2 in ES cells lends support to mal deletions in motifs for activating transcription factors. Impor- this model, as well as the fact that only few—if any—genes are up- tantly, although the transcriptional state of a promoter is critical regulated in PRC2-null stem cells (1, 41, 59). in H3K27me3 deposition, our data argue that it is codependent Through analysis of several sequence variants in a controlled on CpG density. It is important to note that alternate transcrip- system we were able to delineate a set of guidelines for H3K27me3 tional contexts have been implicated in recruiting the polycomb recruitment in mouse embryonic stem cells. Critical parameters machinery to chromatin. For example, PRC2 has been suggested for this establishment include CpG density, transcriptional com- to interact with RNAs in a promiscuous manner, leading to an petency, and DNA methylation. The robustness of these criteria elaborate model whereby nascent RNA accumulation results in a is exemplified by our experiments recapitulating Polycomb re- chromatin state refractory to H3K27me3 deposition (40). Our cruitment on an element spanning less than two nucleosomes. observations implicate transcriptional activity as an antagonistic factor for PRC2 activity and are thus compatible with these find- Methods ings. Notably, our results argue that acquisition of H3K27me3 is Cell Lines and Cell Culture. TC-1 ES cells (28) and DnmtTKO cells (60) were not the cause of target gene repression, but rather occurs in the cultured on 0.2% gelatin-coated dishes. Growth medium consisted of DMEM absence of a transcriptional activator at CG-rich sequences. This (Invitrogen) supplemented with 15% (vol/vol) FCS (Invitrogen), 1× non- result is in line with the very recent finding that global inhibition essential amino acids (Invitrogen), 1 mM L-glutamine, LIF, and 0.001% of transcription increases H3K27 methylation in stem cells (41). β-mercaptoethanol. Differentiation was performed as previously described TFs themselves have been implicated in modulating local levels (61). Treatment with 5-Aza-2′-deoxycytidine (Sigma) was performed over μ of H3K27 methylation, and in line with the above model many of a course of 4 d at a concentration of 0.1 M. these factors serve as transcriptional repressors (e.g., REST and SNAI1) (42–44). Notably, ES cells lacking REST display only Recombination Mediated Cassette Exchange. For targeted insertion, DNA fragments were cloned into a plasmid containing a multiple cloning site subtle effects on H3K27me3 levels at target loci, whereas more flanked by two inverted L1 LoxP sites (L1-poly-1L). Promoter regions were pronounced implications are observed upon neuronal differenti- amplified from TC-1 ES cell genomic DNA (primers in Table S1) or synthesized ation (42). Additionally, loss of REST often accompanies tran- (modified CGIs shown in Table S2), and TF binding site deletions were in- scriptional up-regulation (42), an observation compatible with troduced by Quick-Change PCR. RMCE was performed as described previously our enhancer mutation experiments. A precedent for TF-medi- (28), with slight modifications. See SI Methods for detailed RMCE protocol. ated H3K27me3 deposition was originally described in D. mela- nogaster where, among others, the TF pleiohomeotic (Pho) seems ChIP. ChIP was carried out as previously described (36), using 70 μg of chro- to be involved in recruiting Polycomb (18, 45–47). The mam- matin and 5 μg of the corresponding antibody. A detailed ChIP protocol and malian homolog Yin-Yang1 (YY1), however, displays mutually list of used antibodies is available in SI Methods. Primers used for quantitative exclusive binding patterns with PRC2 in mouse stem cells (48). PCR (qPCR) are listed in Table S3. H3K27me3 ChIP in TKO-133 cells was sub- Notably, the fly genome lacks DNA methylation and CGIs and jected to high-throughput sequencing on an Illumina GAII sequencer using it is thus tempting to speculate that the adaptation to these ge- standard Illumina library preparation kits and protocols (GEO accession no. nomic elements in mammals played a role in shaping H3K27me3 GSE56110). For details regarding sequencing data analysis, see SI Methods. distribution. A critical question is thus how unmethylated CpGs Methylation Analysis. Genomic DNA (2 μg) was bisulfite converted with the are interpreted to become H3K27me3 domains in mammalian EpiTec Bisulfite Kit (Qiagen). Regions of interest were amplified by PCR and genomes. Although several CXXC domain-containing proteins can cloned by TOPOTA cloning (Invitrogen). Sequences were analyzed using BiQ recognize the unmethylated CpG dinucleotide (49), it is unclear if Analyzer (62). Bisulphite-specific PCR primers used are listed in Table S3. they account for the observed genome-wide H3K27me3 distribu- tion. Indeed, two recent studies have shown that the CXXC domain RNA Expression. RNA was isolated using TRIzol/chloroform extraction and KDM2B interacts with noncanonical PRC1 and binds to cDNA synthesis was performed using the PrimeScript RT reagent Kit (Takara). CGIs in stem cells (50, 51). Importantly, however, KDM2B cannot Corresponding cDNA levels were measured by quantitative PCR and nor- solely account for PRC1 recruitment as the protein also binds active malized to endogenous LaminB levels. CGIs (50, 51). Furthermore, loss of KDM2B only results in a mild reduction of PRC1 recruitment at relatively few targets (50, 51). Luciferase Activity Measurement. Lysate preparation and measurement of Another factor implicated in PRC2 recruitment is the lysine luciferase activity was done according to the Luciferase Assay Kit (Promega). methyltransferase cofactor JARID2 (27, 52–55). Although JARID2 Luminescence was normalized to total protein content in the lysate measured may be involved in this process, it actually binds CG-rich DNA with with a BCA Protein Assay Kit (Pierce). relatively low affinity (53, 56) and its loss results only in mild ACKNOWLEDGMENTS. We thank C. Wirbelauer for technical support, and H3K27me3 changes (53, 55), indicating that JARID2 may fine-tune M. Lorincz and members of the D.S. laboratory, in particular P. Ginno, for rather than directly recruit PRC2. critical reading of the manuscript. Research in the D.S. laboratory is In summary, our findings are in agreement with the concept supported by the Novartis Research Foundation, the European Union (NoE ‘‘EpiGeneSys’’ FP7- HEALTH-2010-257082, and the ‘‘Blueprint’’ consortium that PRC1 and PRC2 are recruited to transcriptionally inactive, FP7-282510), the European Research Council (EpiGePlas), the Swiss National unmethylated CpG-rich DNA sequences (22, 23, 57, 58). We Science Foundation Sinergia program, and the Swiss Initiative in Systems envision a model whereby unmethylated CGIs are PRC2 occu- Biology (RTD Cell Plasticity).

1. Boyer LA, et al. (2006) Polycomb complexes repress developmental regulators in 6. Levine SS, King IF, Kingston RE (2004) Division of labor in Polycomb group repression. murine embryonic stem cells. Nature 441(7091):349–353. Trends Biochem Sci 29(9):478–485. 2. Pereira CF, et al. (2010) ESCs require PRC2 to direct the successful reprogramming of 7. de Napoles M, et al. (2004) Polycomb group proteins Ring1A/B link ubiq- differentiated cells toward pluripotency. Cell Stem Cell 6(6):547–556. uitylation of histone H2A to heritable gene silencing and X inactivation. Dev Cell 3. Pengelly AR, Copur Ö, Jäckle H, Herzig A, Müller J (2013) A histone mutant reproduces the 7(5):663–676. phenotype caused by loss of histone-modifying factor Polycomb. Science 339(6120):698–699. 8. Kuzmichev A, Nishioka K, Erdjument-Bromage H, Tempst P, Reinberg D (2002) Histone 4. Lewis EB (1978) A gene complex controlling segmentation in Drosophila. Nature methyltransferase activity associated with a human multiprotein complex containing 276(5688):565–570. the Enhancer of Zeste protein. Genes Dev 16(22):2893–2905. 5. Di Croce L, Helin K (2013) Transcriptional regulation by Polycomb group proteins. Nat 9. Cao R, et al. (2002) Role of histone H3 lysine 27 methylation in Polycomb-group si- Struct Mol Biol 20(10):1147–1155. lencing. Science 298(5595):1039–1043.

6of7 | www.pnas.org/cgi/doi/10.1073/pnas.1400672111 Jermann et al. Downloaded by guest on September 25, 2021 10. Mohn F, et al. (2008) Lineage-specific polycomb targets and de novo DNA methyla- 37. Jüttermann R, Li E, Jaenisch R (1994) Toxicity of 5-aza-2′-deoxycytidine to mammalian PNAS PLUS tion define restriction and potential of neuronal progenitors. Mol Cell 30(6):755–766. cells is mediated primarily by covalent trapping of DNA methyltransferase rather than 11. Bracken AP, Dietrich N, Pasini D, Hansen KH, Helin K (2006) Genome-wide mapping of DNA demethylation. Proc Natl Acad Sci USA 91(25):11797–11801. Polycomb target genes unravels their roles in cell fate transitions. Genes Dev 20(9): 38. Cooper S, et al. (2014) Targeting Polycomb to pericentric heterochromatin in em- 1123–1136. bryonic stem cells reveals a role for H2AK119u1 in PRC2 recruitment. Cell Reports 7(5): 12. Levine SS, et al. (2002) The core of the Polycomb repressive complex is compositionally 1456–1470. and functionally conserved in flies and humans. Mol Cell Biol 22(17):6070–6078. 39. Baubec T, Ivánek R, Lienert F, Schübeler D (2013) Methylation-dependent and 13. Wang H, et al. (2004) Role of histone H2A ubiquitination in Polycomb silencing. Na- -independent genomic targeting principles of the MBD protein family. Cell 153(2): – ture 431(7010):873–878. 480 492. 14. Stock JK, et al. (2007) Ring1-mediated ubiquitination of H2A restrains poised RNA 40. Davidovich C, Zheng L, Goodrich KJ, Cech TR (2013) Promiscuous RNA binding by – polymerase II at bivalent genes in mouse ES cells. Nat Cell Biol 9(12):1428–1435. Polycomb repressive complex 2. Nat Struct Mol Biol 20(11):1250 1257. 15. Zhou W, et al. (2008) Histone H2A monoubiquitination represses transcription by 41. Riising EM, et al. (2014) Gene silencing triggers Polycomb repressive complex 2 re- cruitment to CpG islands genome wide. Mol Cell, 10.1016/j.molcel.2014.06.005. inhibiting RNA polymerase II transcriptional elongation. Mol Cell 29(1):69–80. 42. Arnold P, et al. (2013) Modeling of epigenome dynamics identifies transcription 16. Cavalli G, Paro R (1998) The Drosophila Fab-7 chromosomal element conveys epige- factors that mediate Polycomb targeting. Genome Res 23(1):60–73. netic inheritance during mitosis and meiosis. Cell 93(4):505–518. 43. Herranz N, et al. (2008) Polycomb complex 2 is required for E-cadherin repression by 17. Chan CS, Rastelli L, Pirrotta V (1994) A Polycomb response element in the Ubx gene the Snail1 transcription factor. Mol Cell Biol 28(15):4772–4781. that determines an epigenetically inherited state of repression. EMBO J 13(11): 44. Dietrich N, et al. (2012) REST-mediated recruitment of polycomb repressor complexes 2553–2564. in mammalian cells. PLoS Genet 8(3):e1002494. 18. Ringrose L, Rehmsmeier M, Dura JM, Paro R (2003) Genome-wide prediction of Pol- 45. Brown JL, Fritsch C, Mueller J, Kassis JA (2003) The Drosophila pho-like gene encodes ycomb/Trithorax response elements in Drosophila melanogaster. Dev Cell 5(5): a YY1-related DNA binding protein that is redundant with pleiohomeotic in homeotic – 759 771. gene silencing. Development 130(2):285–294. 19. Müller J, Kassis JA (2006) Polycomb response elements and targeting of Polycomb 46. Brown JL, Mucci D, Whiteley M, Dirksen ML, Kassis JA (1998) The Drosophila Polycomb – group proteins in Drosophila. Curr Opin Genet Dev 16(5):476 484. group gene pleiohomeotic encodes a DNA binding protein with homology to the 20. Ku M, et al. (2008) Genomewide analysis of PRC1 and PRC2 occupancy identifies two transcription factor YY1. Mol Cell 1(7):1057–1064. classes of bivalent domains. PLoS Genet 4(10):e1000242. 47. Klymenko T, et al. (2006) A Polycomb group protein complex with sequence-specific 21. Tanay A, O’Donnell AH, Damelin M, Bestor TH (2007) Hyperconserved CpG domains DNA-binding and selective methyl-lysine-binding activities. Genes Dev 20(9): underlie Polycomb-binding sites. Proc Natl Acad Sci USA 104(13):5521–5526. 1110–1122. 22. Mendenhall EM, et al. (2010) GC-rich sequence elements recruit PRC2 in mammalian 48. Vella P, Barozzi I, Cuomo A, Bonaldi T, Pasini D (2012) Yin Yang 1 extends the Myc- ES cells. PLoS Genet 6(12):e1001244. related transcription factors network in embryonic stem cells. Nucleic Acids Res 40(8): 23. Lynch MD, et al. (2012) An interspecies analysis reveals a key role for unmethylated 3403–3418. CpG dinucleotides in vertebrate Polycomb complex recruitment. EMBO J 31(2): 49. Long HK, Blackledge NP, Klose RJ (2013) ZF-CxxC domain-containing proteins, CpG 317–329. islands and the chromatin connection. Biochem Soc Trans 41(3):727–740.

24. Long HK, et al. (2013) Epigenetic conservation at gene regulatory elements revealed 50. Wu X, Johansen JV, Helin K (2013) Fbxl10/Kdm2b recruits polycomb repressive com- GENETICS by non-methylated DNA profiling in seven vertebrates. eLife 2:e00348. plex 1 to CpG islands and regulates H2A ubiquitylation. Mol Cell 49(6):1134–1146. 25. Cohen NM, Kenigsberg E, Tanay A (2011) Primate CpG islands are maintained by 51. Farcas AM, et al. (2012) KDM2B links the Polycomb Repressive Complex 1 (PRC1) to heterogeneous evolutionary regimes involving minimal selection. Cell 145(5): recognition of CpG islands. eLife 1:e00205. 773–786. 52. Landeira D, et al. (2010) Jarid2 is a PRC2 component in embryonic stem cells required 26. Tiwari VK, et al. (2012) A chromatin-modifying function of JNK during stem cell dif- for multi-lineage differentiation and recruitment of PRC1 and RNA Polymerase II to – ferentiation. Nat Genet 44(1):94–100. developmental regulators. Nat Cell Biol 12(6):618 624. 27. Pasini D, et al. (2010) JARID2 regulates binding of the Polycomb repressive complex 2 53. Li G, et al. (2010) Jarid2 and PRC2, partners in regulating gene expression. Genes Dev – to target genes in ES cells. Nature 464(7286):306–310. 24(4):368 380. 28. Lienert F, et al. (2011) Identification of genetic elements that autonomously de- 54. Peng JC, et al. (2009) Jarid2/Jumonji coordinates control of PRC2 enzymatic activity and target gene occupancy in pluripotent cells. Cell 139(7):1290–1302. termine DNA methylation states. Nat Genet 43(11):1091–1097. 55. Shen X, et al. (2009) Jumonji modulates polycomb activity and self-renewal versus 29. Kooistra SM, et al. (2010) Undifferentiated embryonic cell transcription factor 1 differentiation of stem cells. Cell 139(7):1303–1314. regulates ESC chromatin organization and gene expression. Stem Cells 28(10): 56. Kim TG, Kraus JC, Chen J, Lee Y (2003) JUMONJI, a critical factor for cardiac de- 1703–1714. velopment, functions as a transcriptional repressor. J Biol Chem 278(43):42247–42255. 30. Tippmann SC, et al. (2012) Chromatin measurements reveal contributions of synthesis 57. Klose RJ, Cooper S, Farcas AM, Blackledge NP, Brockdorff N (2013) Chromatin sam- and decay to steady-state mRNA levels. Mol Syst Biol 8:593. pling—An emerging perspective on targeting polycomb repressor proteins. PLoS 31. Chew JL, et al. (2005) Reciprocal transcriptional regulation of Pou5f1 and Sox2 via the Genet 9(8):e1003717. – Oct4/Sox2 complex in embryonic stem cells. Mol Cell Biol 25(14):6031 6046. 58. van Heeringen SJ, et al. (2014) Principles of nucleation of H3K27 methylation during 32. Lodato MA, et al. (2013) SOX2 co-occupies distal enhancer elements with distinct POU embryonic development. Genome Res 24(3):401–410. factors in ESCs and NPCs to specify cell state. PLoS Genet 9(2):e1003288. 59. Leeb M, et al. (2010) Polycomb complexes act redundantly to repress genomic repeats 33. Brinkman AB, et al. (2012) Sequential ChIP-bisulfite sequencing enables direct genome- and genes. Genes Dev 24(3):265–276. scale investigation of chromatin and DNA methylation cross-talk. Genome Res 22(6): 60. Tsumura A, et al. (2006) Maintenance of self-renewal ability of mouse embryonic 1128–1138. stem cells in the absence of DNA methyltransferases Dnmt1, Dnmt3a and Dnmt3b. 34. Reddington JP, et al. (2013) Redistribution of H3K27me3 upon DNA hypomethylation Genes Cells 11(7):805–814. results in de-repression of Polycomb target genes. Genome Biol 14(3):R25. 61. Bibel M, Richter J, Lacroix E, Barde YA (2007) Generation of a defined and uniform 35. Bernstein BE, et al. (2006) A bivalent chromatin structure marks key developmental population of CNS progenitors and neurons from mouse embryonic stem cells. Nat genes in embryonic stem cells. Cell 125(2):315–326. Protoc 2(5):1034–1043. 36. Weber M, et al. (2007) Distribution, silencing potential and evolutionary impact of 62. Bock C, et al. (2005) BiQ Analyzer: Visualization and quality control for DNA meth- promoter DNA methylation in the . Nat Genet 39(4):457–466. ylation data from bisulfite sequencing. Bioinformatics 21(21):4067–4068.

Jermann et al. PNAS Early Edition | 7of7 Downloaded by guest on September 25, 2021