Mouse embryonic stem cells cultured with MEK (mitogen-activated protein kinase kinase) and GSK3 (glycogen synthase kinase 3) inhibitors (2i) more closely resemble the inner cell mass of preimplantation blastocysts than those cultured with SL [serum/leukemia inhibitory factor (LIF)]. The transcriptional mechanisms governing this pluripotent ground state are unresolved. Release of promoter-proximal paused RNA polymerase II (Pol2) is a multistep process necessary for pluripotency and cell cycle gene transcription in SL. We show that β-catenin, stabilized by GSK3 inhibition in medium with 2i, supplies transcriptional coregulators at pluripotency loci. This selectively strengthens pluripotency loci and renders them addicted to transcription initiation for productive gene body elongation in detriment to Pol2 pause release. By contrast, cell cycle genes are not bound by β-catenin, and proliferation/self-renewal remains tightly controlled by Pol2 pause release under 2i conditions. Our findings explain how pluripotency is reinforced in the ground state and also provide a general model for transcriptional resilience/adaptation upon network perturbation in other contexts.
Pluripotency can be sustained in vitro through culture in specific conditions. Mouse embryonic stem cells (ESCs) in conventional serum/leukemia inhibitory factor (LIF) (SL) medium are considered to exhibit naïve, preimplantation-like pluripotency because they contribute to chimeras with relative high efficiency upon blastocyst complementation. Yet, only a proportion of ESCs in SL are truly naïve at a given time, and the entire population is highly metastable, periodically switching between naïve and early post-implantation–like (formative or partially primed) pluripotent states (1, 2). Culture in serum-free medium with mitogen-activated protein kinase kinase (MEK) and glycogen synthase kinase 3 (GSK3) inhibitors, PD0325901 (PD) and CHIR99021 (CHIR), produces a pluripotent ground state that more closely resembles the preimplantation inner cell mass (3–5). Addition of LIF to the 2i (2iL) facilitates pluripotency maintenance in the ground state but is not strictly necessary. ESCs in 2iL have lower expression of lineage-associated genes and more homogeneous expression of pluripotency genes than in SL (6, 7). They also display genome-wide DNA hypomethylation, reduced histone 3 lysine-27 trimethylation (H3K27me3) at promoters, and tolerate better the suppression of epigenetic/epitranscriptomic factors than ESCs in SL (6, 8–10). Overall, this suggests a rewiring of regulatory networks that confers additional robustness in 2iL, but the underlying mechanisms are unclear.
Gene transcription in eukaryotes has a highly regulated progression involving initiation, Pol2 pausing in the vicinity of the promoter, release of paused Pol2, gene body elongation, and termination (11). Recruitment of the Pol2 transcription initiation apparatus and Pol2 pause release are rate-limiting steps. Initiation is orchestrated by sequence-specific transcription factors (e.g., the pluripotency transcription factors), which, through chromatin remodeling, allow the recruitment of the basal transcription machinery including general transcription factors and Pol2. For many mammalian genes, Pol2 then pauses 20 to 60 nucleotides after the transcription start site (TSS), requiring pause release for subsequent productive gene body elongation. Pol2 pause release is mediated by CDK9, the catalytic subunit of the positive transcription elongation factor b (P-TEFb) complex. CDK9 resides in a catalytically inactive complex that is activated by different mechanisms; the bromodomain and extraterminal (BET) family member BRD4 plays a critical role in this process (12, 13). Pol2 pausing and the subsequent pause release represent a mechanism for ensuring potent but quick binary-switchable gene expression but, being a multistep process, could render gene transcription vulnerable to perturbation. Notably, both transcription initiation and Pol2 pause release are required for sustaining high expression levels of genes involved in pluripotency maintenance and proliferation/self-renewal of mouse ESCs in SL (12, 14–16), but it was unclear whether transcriptional regulation in ground-state culture conditions has the same essential requirements.
Here, we show that β-catenin potentiates the recruitment of coregulators—including BRD4, CDK9, mediator, cohesin, and p300—to strengthen pluripotency loci in ESCs in 2iL. This enhances transcription initiation at those loci, compensatorily lowering the dependence on Pol2 pause release for productive gene body elongation. By contrast, cell cycle–related genes are not bound by β-catenin and remain addicted to Pol2 pause release in 2iL, making self-renewal highly sensitive to BRD4/CDK9 suppression in both culture conditions. Our findings explain how pluripotency gene transcription is selectively reinforced in the ground state to protect against exogenous perturbation.
Pluripotency maintenance in the ground state requires a residual BRD4 level
To investigate distinctive transcriptional features of mouse ESCs cultured in SL or 2iL, we performed a short hairpin RNA (shRNA) screen for a panel of transcriptional regulators (fig. S1A). This panel included regulators of Pol2 pause release, histone methyltransferases/demethylases, histone acetyltransferases/deacetylases, histone acetylation readers, and splicing regulators, many of which are known to be necessary for ESC maintenance in SL (table S1). The effect of the knockdown was determined by measuring the expression of the core pluripotency markers Oct4 (Pou5f1), Nanog, and Klf2 by reverse transcription quantitative polymerase chain reaction (RT-qPCR). Suppressing most regulators had a stronger effect in reducing pluripotency genes in SL than 2iL (Fig. 1A). In particular, we noticed that knocking down two relevant mediators of Pol2 pause release, BRD4 and CDK9, was better tolerated in 2iL.
We first focused on BRD4 because we and others have reported that it is a master regulator of ESC pluripotency/self-renewal (in SL) and early embryonic development (12, 14, 15). Basal BRD4 expression tested by Western blotting was comparable in SL and 2iL (fig. S1B). We repeated the Brd4 knockdown in both conditions and confirmed that it was effective in reducing mRNA and protein expression (Fig. 1B and fig. S1C). In contrast to SL, ESC colonies in 2iL transduced with Brd4 shRNA remained domed and compact, as well as alkaline phosphatase (AP) positive, even after several passages as single cells (fig. S1, D and E). Likewise, pluripotency genes, measured by RT-qPCR, exhibited little change or up-regulation in 2iL compared to SL (Fig. 1B and fig. S1F), but we observed reduced proliferation in both conditions (albeit more obvious in SL) (Fig. 1C and fig. S1D). This was associated with a significant increase in the number of cells in the G0-G1 phase of the cell cycle (Fig. 1D). Analysis of chromatin immunoprecipitation–sequencing (ChIP-seq) for BRD4 showed a similar widespread binding pattern in SL and 2iL (fig. S1, G and H). We then validated the differential effects of Brd4 knockdown in SL and 2iL using two additional ESC lines and two more batches of ESC-qualified serum from different vendors (fig. S2, A to E). These results demonstrated that BRD4 is less required for preserving pluripotency in 2iL than SL but remains necessary for self-renewal (i.e., robust proliferative expansion in vitro) under both conditions.
To further verify the differential sensitivity of pluripotency characteristics to BRD4 suppression in ESCs cultured in SL and 2iL, we used JQ1, a well-known BET inhibitor that binds to the two BRD4 bromodomains to prevent their interaction with acetylated histones (17). At lower doses (100 and 200 nM) for 60 hours, JQ1 notably affected colony morphology, AP activity, and pluripotency gene expression in SL but had little effect in 2iL (Fig. 1, E and F). ESCs in 2iL remained competent for teratoma and chimera formation with 100 nM JQ1 (fig. S3, A and B). However, at higher doses (500 nM and above), pluripotency characteristics were also notoriously affected in 2iL (Fig. 1, E and G), especially upon passage as single cells (fig. S3C). Likewise, JQ1 reduced proliferation in 2iL, although at lower doses, this only became prominent after passaging as single cells (Fig. 1H). This was paralleled by an increase in the percentage of cells in G0-G1 and in apoptosis (Fig. 1I and fig. S3D). RNA sequencing (RNA-seq) confirmed the differential effect of 100 nM JQ1 on pluripotency in SL and 2iL and also showed the down-regulation of cell cycle genes (Fig. 1J and table S2). We confirmed that low doses of JQ1 impair pluripotency in SL but not 2iL using two additional ESC lines and two batches of ESC-qualified serum (fig. S3, E to G).
The experiments with JQ1 suggested that a certain level of BRD4 is necessary for maintaining pluripotency in 2iL. To exclude the possibility that higher doses impair pluripotency characteristics in 2iL through off-target effects, we used an inducible Cre/LoxP system for knocking out Brd4 (fig. S4, A to D). Despite extensive testing, we only obtained heterozygous Brd4fl/− clones in 2iL, which proliferated less and differentiated when changed to SL culture conditions (fig. S4, E and F). We also noticed that, in contrast to wild-type clones, low doses of JQ1 could effectively reduce pluripotency gene expression in heterozygous Brd4 knockout ESCs in 2iL (fig. S4G). We concluded that pluripotency maintenance is more resistant to BRD4 suppression in ESCs in 2iL than in SL, but reducing BRD4 beyond a threshold also affects pluripotency in 2iL.
Suppressing Pol2 pause release at pluripotency loci is better tolerated in the ground state
Pol2 pausing is mediated by pausing factors including DRB sensitivity–inducing factor (DSIF) and negative elongation factor (NELF), whereas pause release is triggered through phosphorylation of Pol2 on serine-2 (Ser2P) by CDK9. A major role of BRD4 is to induce Pol2 pause release by activating CDK9 (13), a target that was also identified as less necessary for 2iL in our shRNA screen (see above Fig. 1A). Consistently, analysis of CDK9 ChIP-seq in SL showed notable overlap with BRD4 ChIP-seq in SL or 2iL (Fig. 2A). Likewise, a sizeable proportion of genes down-regulated by JQ1 in SL or 2iL were cobound by BRD4 and CDK9 (Fig. 2B), including many pluripotency (in SL) and cell cycle genes (in SL and 2iL) (Fig. 2, B and C, and fig. S5A). To confirm the differential CDK9 dependence in SL and 2iL, we repeated the knockdown experiments and also used a specific CDK9 inhibitor [LDC000067; (18)]. As with Brd4 knockdown, Cdk9 knockdown severely affected colony morphology, AP activity, and pluripotency gene expression in SL but had no obvious effect in 2iL (Fig. 2, D and E), and this persisted for several passages (fig. S5, B and C). Proliferation and the cell cycle were significantly affected by Cdk9 knockdown in 2iL too (Fig. 2, F and G), although to a lesser extent than in SL (Fig. 2D). These effects were validated using an additional ESC line (fig. S5, D and E). Similarly, 10 μM LDC000067 impaired colony morphology, AP activity, and pluripotency gene expression in SL but not in 2iL (even after multiple passages as single cells), but a higher dose had severe consequences in both conditions (Fig. 2, H and I, and fig. S5, F and G). Likewise, LDC000067 reduced cell growth in SL and 2iL, impaired the cell cycle, and enhanced apoptosis significantly (Fig. 2, H and J to L). The consistent phenotypes of suppressing Brd4 and Cdk9 implied that reducing Pol2 pause release at pluripotency genes is better tolerated in 2iL than SL, suggesting a major change in transcriptional control in the two culture conditions.
β-Catenin induces resistance to Pol2 pause release suppression in the ground state
BET inhibitors including JQ1 are a promising therapeutic avenue for cancer, but recent reports have described resistance to BET inhibitors through activation of Wnt/β-catenin signaling (19, 20). In this pathway, Wnt ligands trigger stabilization and nuclear translocation of β-catenin, which then binds to and transactivates T cell factor/lymphoid enhancer factor (TCF/LEF) transcription factors to switch on gene expression (21). We envisaged that β-catenin could also confer resistance to BRD4 suppression in ESCs cultured in 2iL, as, similarly to Wnt ligands, CHIR stabilizes β-catenin through GSK3 inhibition (3). Moreover, β-catenin has been proposed to promote ground-state pluripotency by alleviating the repressor function of TCF3, which associates with pluripotency transcription factors at target loci (21–23). Yet, the specific mechanisms are not well understood. Accordingly, Tcf3 or Gsk3 depletion allow expansion of ESCs in serum-free medium with PD alone (3, 21, 23), whereas β-catenin is strictly required for expansion in 2iL medium without LIF (21, 22).
We first studied whether PD alone, CHIR alone, or the 2i added to ESCs in SL could rescue the negative effects of 100 nM JQ1 on colony morphology, AP activity, and pluripotency gene expression, which we tested using two ESC lines. PD alone had some rescue effect on Nanog expression but not on the other genes tested. CHIR was more effective in restoring pluripotency characteristics, but only the combined effect of PD and CHIR achieved a complete rescue (Fig. 3, A and B, and fig. S6A). As for cell proliferation, we observed that the moderate rescue effect of adding 2i to ESCs in SL treated with JQ1 was mostly mediated by PD (Fig. 3C).
To systematically dissect the role of specific components of the Wnt/β-catenin pathway, we treated wild-type ESCs in SL with WNT3A or used several knockout ESC lines lacking either Gsk3 (24), Ctnnb1 (encoding β-catenin) (22), or Tcf3 (23). In addition, we used wild-type ESCs overexpressing a mutant form of β-catenin (S33Y β-catenin) resistant to GSK3-mediated degradation (25). The authentication of Gsk3 and Ctnnb1 knockout ESC lines, and ESCs overexpressing S33Y β-catenin, was performed with a β-catenin/TCF reporter (fig. S6B), whereas Tcf3 knockout cells were validated by PCR amplification and sequencing (fig. S6C). WNT3A treatment significantly reversed the effects of 100 nM JQ1 on pluripotency characteristics in SL, and Gsk3 knockout achieved a stronger rescue (Fig. 3, D and E, and fig. S6, D and E). The latter was also confirmed by RNA-seq (Fig. 3F and fig. S6F). The stronger effect of Gsk3 knockout compared to WNT3A and CHIR alone is possibly related to the extent and length of GSK3 suppression. Likewise, Tcf3 knockout and S33Y β-catenin overexpression induced significant resistance to JQ1 in SL (Fig. 3, G to J). Moreover, Ctnnb1 knockout became sensitive to 200 nM JQ1 in 2iL, but the effect on pluripotency gene expression was not as strong as for wild-type ESCs in SL (Fig. 3, K and L). We also validated the resistance to the CDK9 inhibitor LDC000067 in Gsk3 or Tcf3 knockout ESCs (fig. S6, G and H). Therefore, GSK3 inhibition is the main mediator of the resistance of pluripotency genes to suppression of pause release in 2iL, of which β-catenin stabilization is a major component.
β-Catenin increases transcription initiation at pluripotency loci in the ground state
To understand how β-catenin mediates resistance to suppression of Pol2 pause release at pluripotency loci, we compared β-catenin bound sites (table S3) in ChIP-seq (from a study using SL + CHIR) (26) with BRD4 bound sites in 2iL. There was a good genome-wide overlap (Fig. 4A), mostly at distal enhancers but also at promoters (fig. S7A), although the binding of BRD4 was more widespread. Moreover, we noticed that most β-catenin/BRD4 cobound genes were not down-regulated by JQ1 in 2iL. We then named β-catenin/BRD4 cobound genes that are down-regulated by JQ1 in SL but not 2iL as group 1 genes (Fig. 4B). By contrast, group 2 genes were defined as genes down-regulated by JQ1 in 2iL that are bound by BRD4 but not β-catenin. Group 2 included many cell cycle genes, whereas group 1 included many pluripotency regulators (Fig. 4B and table S4). ChIP-qPCR confirmed enhanced β-catenin binding at selected group 1 pluripotency loci in 2iL compared to SL, whereas at group 2 cell cycle–related loci did not change (fig. S7B). ChIP-seq analysis also showed that TCF3 binds to a notable proportion of group 1 genes, whereas most of the group 2 genes were negative (fig. S7C and table S4). These findings suggested that β-catenin promotes resistance to Pol2 pause release suppression through cobinding with BRD4/CDK9 at target loci, including pluripotency loci.
Next, we sought to elucidate the molecular mechanism underlying the above observations. To rule out the possibility that β-catenin compensates for the negative effect of JQ1 on pluripotency genes by enhancing mRNA stability in 2iL (27), we measured a panel of pluripotency mRNAs after actinomycin D treatment, which blocks transcription. Their stability was similar or lower in 2iL compared to SL (fig. S7D). This hinted to β-catenin maximizing transcriptional flux at target genes in 2iL as a way to counteract a reduction in Pol2 pause release. So, we turned our attention to potential differences in transcriptional dynamics between ESCs cultured in SL and 2iL. In this regard, a recent Pol2 ChIP-seq study (6) showed a global increase of promoter-proximal signal in 2iL that was not matched in the gene body, concluding that Pol2 pausing is more prevalent in 2iL than in SL. This was attributed to low expression of c-MYC in ESCs in 2iL, as c-MYC induces Pol2 pause release via CDK9 (28). Our reanalysis of this dataset showed a strong increase of Pol2 signal at the proximal promoter in 2iL for both group 1 and group 2 genes (fig. S7, E to G), but this could represent either more Pol2 pausing or more transcriptional initiation. As opposed to Pol2 pausing, more transcriptional initiation implies more gene body elongation if the degree of pausing remains constant and, hence, often associates with increased gene expression. Consistent with the former possibility, the Pol2 signal along the gene body only increased moderately at both groups of genes in 2iL, especially at group 1 (fig. S7, E to G). To define the extent of pausing at these loci more accurately, we used the Pol2 traveling ratio (TR), which compares the ratio in the signal of the proximal promoter and the gene body (12, 28, 29). The TR of both group 1 and group 2 genes was higher in 2iL (fig. S7H), supporting the idea that there is indeed more Pol2 pausing in both groups of genes in 2iL. Yet, it is difficult to reconcile the resistance of group 1 genes to BRD4/CDK9 suppression in 2iL with an increased Pol2 pausing that, in principle, would reduce gene expression. In this regard, we noted several reports describing that, for being a static snapshot, Pol2 ChIP-seq signals cannot effectively distinguish pausing from transcription initiation, nor can they be a measure for effective transcription elongation (29, 30).
We then used genome-wide nuclear run-on sequencing (GRO-seq) (31), which labels nascent RNAs with the synthetic nucleoside 5-bromouridine 5’-triphosphate (BrUTP) to accurately map the distribution of transcriptionally engaged Pol2 throughout the genome. Group 1 genes in 2iL displayed enhanced GRO-seq signal not only in the proximal promoter but also in the gene body compared to SL, and the TR did not change significantly (Fig. 4, C to G), together indicating more transcription initiation followed by productive elongation. By contrast, group 2 genes in 2iL showed little difference in the GRO-seq signals in the proximal promoter or the gene body compared to SL, and the TR also remained fairly unchanged (Fig. 4, C to G). The relative pausing level of group 2 genes was higher than that of group 1 genes (Fig. 4F), in agreement with previous reports showing that Pol2 pausing is prevalent at cell cycle–related genes in both SL and 2iL (29, 32). We also performed ChIP-qPCR for both Pol2 phosphorylated in serine-5 (Ser5P) and Pol2 Ser2P. The former Pol2 modification is mediated by CDK7, a catalytic subunit of the transcription factor H (TFIIH) complex (11), and is considered representative of transcriptional initiation, whereas the latter marks elongating Pol2. Both Pol2 modifications showed a notably increased signal at the Nanog and Esrrb promoter and gene body in 2iL compared with SL (Fig. 4, H and I), but not at two selected cell cycle–related loci. These results further support the conclusions of the GRO-seq experiment. The GRO-seq of Gsk3 knockout ESCs in SL also showed a pattern consistent with higher transcription initiation and increased gene body elongation at group 1 genes but not group 2 genes compared to the wild-type control (fig. S8, A to D).
We next asked whether the increased transcription initiation at group 1 genes in 2iL may directly contribute to their resistance to Pol2 pause release suppression. Supporting this possibility, higher Pol2 occupancy at the promoter-proximal pause site driven by more transcription initiation can result in greater productive elongation if the function of BRD4/CDK9 is not yet saturated (13). It has also been proposed that increased transcription initiation can nudge paused Pol2 out of the proximal promoter to resume gene body elongation at BRD4 bound genes insensitive to JQ1 (30, 33). To study the relative dependence of pluripotency genes on transcription initiation of ESCs in SL and 2iL, we used the CDK7 inhibitor THZ1 (34). Notably, THZ1 down-regulated pluripotency genes more significantly in 2iL than in SL (Fig. 4J). Moreover, a low dose of THZ1 synergized with 100 nM JQ1 to reduce the expression of pluripotency genes in 2iL (Fig. 4K), supporting the idea that increased initiation compensates for the reduction in pause release mediated by BRD4 inhibition. To study whether the link between Wnt/β-catenin signaling, transcription initiation, and JQ1 resistance applies to other contexts, particularly cancer cells, we tested a widely used leukemia cell line, THP1. CHIR induced resistance of THP1 cells to JQ1 (fig. S8E) and also rendered them more sensitive to THZ1 (fig. S8F). Likewise, the combination of THZ1 and JQ1 was more effective than either of the two (fig. S8G). In summary, the recruitment of β-catenin to BRD4 bound sites in 2iL changes the mode of transcriptional regulation at target loci including pluripotency loci, which then rely more on transcription initiation for gene body elongation in detriment to Pol2 pause release.
β-Catenin supplies coregulators to maximize transcriptional flux at pluripotency loci
We searched β-catenin protein interaction networks looking for partners whose recruitment or reinforcement at group 1 genes in 2iL could explain the above phenomena. In addition to chromatin remodeling complexes (35), we observed two modules corresponding to transcription initiation and elongation (Fig. 5, A and B). Among other β-catenin interacting proteins in these modules, we noticed Pol2, TATA-binding protein–associated factors (TAF5/6/7), cohesin components (SMC1A and SMC3), and, interestingly, BRD4 and CDK9 as well. Pol2 and TAFs are critical for transcription initiation (11), whereas cohesin regulates transcription by forming ring-like structures that allow enhancer-promoter looping (36). We also noticed previous reports describing the interaction of β-catenin with mediator (37, 38) and p300 (39) in other cell contexts. Mediator was immediately interesting because it is a well-known partner of BRD4 that controls transcription initiation through both cross-talk with TFIIH and enhancer-promoter looping (40). Immunoprecipitation of β-catenin followed by Western blotting confirmed the interaction with mediator (MED1 and MED12), cohesin (SMC1A), and BRD4 in ESCs in 2iL (Fig. 5C). Likewise, ChIP-seq analysis showed genome-wide colocalization of β-catenin, MED1, SMC1A, and BRD4 in 2iL at many pluripotency genes belonging to group 1 (Fig. 5D and fig. S9, A and B).
To see whether β-catenin is actually promoting the recruitment of these coregulators at target loci, we compared ChIP-seq datasets for MED1, SMC1A, and BRD4 in 2iL and SL. We observed higher levels of the three coregulators at β-catenin binding sites in 2iL (Fig. 5, E to G). BRD4 also showed increased signal outside β-catenin binding sites but less remarkably, consistent with the idea that the 2i cause a global increase in BRD4 (41). In agreement, ChIP-qPCR for the same three coregulators at β-catenin binding sites of pluripotency loci showed reduced levels in Ctnnb1 knockout ESCs in 2iL compared to the wild-type control (Fig. 5, H to J). However, binding of these coregulators at group 2 cell cycle–related genes remained unchanged. A truncated β-catenin form (lacking amino acids 727 to 781) without the C-terminal domain responsible for transcriptional activation of TCF/LEF factors (22) was still competent for inducing resistance to JQ1 in 2iL (Fig. 5K and fig. S9, C and D). This truncated β-catenin also retained the ability to interact with coregulators in ESCs in 2iL (Fig. 5L). Overall, our findings support a model in which β-catenin strengthens transcriptional flux at pluripotency loci by acting as a scaffold for recruiting coregulators rather than forming a canonical β-catenin–dependent activation complex.
Permissive chromatin features help maximize transcriptional flux at pluripotency loci in the ground state
We also investigated chromatin features that could further contribute to maximizing transcriptional flux at group 1 genes in 2iL compared to SL. We focused on histone acetylation and DNA hypomethylation because these epigenetic marks associate with chromatin opening, transcription activation, and reduced Pol2 pausing (42, 43). H3K27 acetylation (H3K27ac) around β-catenin binding sites was higher in 2iL than in SL (Figs. 5D and 6A), consistent with the recruitment of histone acetyltransferases (e.g., p300) by β-catenin (39). Similarly, we observed an increase in H3K27ac in 2iL when comparing the −2-kb to +2-kb region around the TSS of group 1 genes. By contrast, H3K27ac did not increase at group 2 genes in 2iL compared to SL, and group 2 genes in 2iL had lower H3K27ac than group 1 genes (Fig. 6B). Consistent with the changes in H3K27ac, we noticed a clear increase in open chromatin with an assay for transposase-accessible chromatin sequencing (ATAC-seq) at β-catenin binding sites in 2iL compared to SL and more moderately also at group 1 genes, whereas, at group 2 genes, it was slightly reduced in 2iL (Figs. 5D and 6, C and D). Notably, DNA methylation at β-catenin binding sites was lower in 2iL than in SL (Fig. 6E). Yet, this effect extended to the entire locus of not only group 1 but also group 2 genes (Fig. 6F), indicating that it is not directly mediated by β-catenin. The latter is in agreement with the existence of global DNA hypomethylation in 2iL, which is mostly driven passively through the suppression of UHRF1 protein stability induced by PD (44). In this regard, the limited number of sites actively demethylated by the ten-eleven translocation (TET) enzymes in the conversion of ESCs from SL to 2iL (44) included few β-catenin binding sites (fig. S9E). Accordingly, Tet1/2 double and Tet1/2/3 triple knockout ESCs (45) did not show increased sensitivity of pluripotency genes to JQ1 in 2iL compared to the control (fig. S9, F to H). We concluded that permissive chromatin features, some of which are induced by β-catenin, likely contribute to strengthening pluripotency gene transcription in 2iL by facilitating the assembly of multiprotein complexes (see schematic in Fig. 7).
In addition to the recruitment of coregulators and the chromatin changes, other mechanisms may participate in inducing transcriptional resilience at pluripotency loci in 2iL. For example, alternative RNA splicing is a cotranscriptional event that can influence the speed with which Pol2 moves along the gene body (46), and it has also been shown that specific splicing regulators participate in Pol2 pause release (47). Likewise, Gsk3 knockout in ESCs in SL reduces the amount of alternative splicing due to impaired GSK3-mediated phosphorylation of splicing factors (48). We did not observe any notable difference in the number of alternatively spliced genes regulated by GSK3 between group 1 and group 2 genes (fig. S9I). Yet, we noticed that β-catenin interacts with multiple splicing regulators including SRSF3 and TRA2B (fig. S9J) (35), both of which also appeared in our screen as differentially required in SL and 2iL (see above Fig. 1A). We validated that Srsf3 and Tra2b knockdown is better tolerated in 2iL compared to SL (fig. S9K). This observation suggests that β-catenin helps stabilize splicing regulators at pluripotency genes to render ESCs more resistant to a splicing reduction in 2iL. Although a potential role in modifying the speed of gene body elongation would need to be investigated, these results support the model depicting β-catenin as a scaffold that strengthens transcription at pluripotency loci in 2iL.
Mouse ESC pluripotency can be viewed as a continuum of hierarchical interconvertible states on the road to a somatic phenotype. The more naïve or closer to inner cell mass characteristics, the more pluripotency is consolidated, but the underlying mechanisms are poorly understood. We have shown here that β-catenin stabilized by CHIR selectively reinforces the pluripotency gene network in 2iL by potentiating the recruitment of BRD4, CDK9, mediator, cohesin, p300, and other transcriptional coregulators to pluripotency loci. This selectively heightens transcription initiation at pluripotency loci, enhancing gene body elongation in 2iL and making it more—albeit not completely—independent of Pol2 pause release by BRD4/CDK9 than in SL. The enhanced transcriptional elongation in 2iL likely explains why expression of multiple pluripotency genes is higher than in SL and potentially also why there is less oscillation in gene expression (an underlying cause of metastability) (7). The removal of TCF3 from pluripotency loci causes a similar transcriptional consequence to β-catenin stabilization, conceivably by allowing closer interactions between coregulators and the pluripotency transcription factors or by removing detrimental epigenetic activities [e.g., histone deacetylases (49)]. PD also contributes to inducing resistance to suppression of Pol2 pause release in 2iL possibly by inducing Nanog mRNA and stabilizing NANOG protein (50). The former effect might be caused by preventing extracellular signal–regulated kinase–mediated phosphorylation and dissociation of coregulators including MED24 from Pol2-containing complexes at the Nanog locus (51). As opposed to pluripotency genes, proliferation genes are not bound by β-catenin and, thus, remain very sensitive to suppression of Pol2 pause release in 2iL.
In recent years, it has become evident that phase-separated biomolecular condensates compartmentalize biochemical reactions within cells, including transcription (52). This is caused by multivalent interactions between proteins, many of which have intrinsically disordered regions (IDRs) that confer the physicochemical properties of the condensate. In this regard, it was recently proposed that, thanks to its two disorganized domains at the N-terminal (amino acids 1 to 141) and C-terminal (amino acids 727 to 781) ends, β-catenin is attracted to stable chromatin phase-separated condensates formed by mediator and BRD4 to execute its signaling role in ESCs in 2iL (53). Our findings suggest that β-catenin might be a priming event for the stabilization of these condensates in ESCs in 2iL by enhancing the cooperative and multivalent interactions between coregulators at pluripotency loci (Fig. 7). This is consistent with our observation that the C-terminal domain of β-catenin containing one of its IDRs is not necessary for the resiliance of ESCs in 2iL to JQ1 and the fact that β-catenin IDRs are much shorter than those of BRD4 and MED1 (53, 54). The physicochemical forces created within these condensates and the interaction with β-catenin could cause a remnant of BRD4, CDK9, and other coregulators to tend to localize to pluripotency loci despite genome-wide depletion induced by shRNAs or chemical inhibitors.
Finley et al. recently reported that BRD4 is dispensable for pluripotency and self-renewal in the ground state (41). A reduced rather than abolished requirement for BRD4 in the early embryo is perhaps easier to understand from a developmental point of view, as it is supported by the observation that Brd4 null mouse embryos cannot maintain the inner cell mass (15, 55). Finley et al. also proposed that a strengthened network of pluripotency transcription factors and the recruitment of TET enzymes partially contribute to the resistance to BRD4 suppression in 2iL. The former mechanism fits well with our observations, as transcription factors can recruit coregulators and enhance transcription initiation (11). Yet, we did not observe any evidence for TET involvement, which may be related to variations among ESC lines or in the culture conditions. Despite the differences, both studies are relevant and highlight the striking similarities in transcriptional adaptation upon network perturbation between ESCs in the ground state and cancer cells. Further mechanistic knowledge will mutually contribute to understand ground-state pluripotency and cancer cell resistance to drugs. For example, ESCs in 2iL may prove to be a useful model to identify either more effective anticancer drugs or synergistic combinations. In this regard, our findings with ESCs in 2iL suggest that treatment of BRD4-addicted cancers with a combination of JQ1 and inhibitors of transcription initiation might be a more robust and applicable anticancer therapy for a general patient base than JQ1 alone.
In the future, it will be important to study whether the molecular interface regulating the interaction between β-catenin and transcriptional coregulators can be used to develop specialized anticancer drugs. It will also be interesting to test whether the principles presented here can yield optimized methods for sustaining ground-state pluripotency in vitro in a broad spectrum of mammals.
MATERIALS AND METHODS
Cell lines and culture conditions
Human embryonic kidney–293T (HEK293T) cells were purchased from the American Type Culture Collection and maintained in Dulbecco’s modified Eagle’s medium (DMEM)/high glucose (Corning, 10-017-CVR) containing 10% fetal bovine serum (FBS; Biowest). THP1 cells were purchased from the Cell Bank of Type Culture Collection of the Chinese Academy of Sciences (Shanghai, China) and maintained in RPMI 1640 medium (Thermo Fisher Scientific, C11875500CP) supplemented with 10% FBS (Biowest), GlutaMAX (Gibco, 35050079), penicillin/streptomycin (Hyclone, SV30010), and β-mercaptoethanol (Gibco, 2198503). ESCs in SL medium were cultured in DMEM/high glucose containing 15% FBS (Biological Industries; unless otherwise specified), GlutaMAX, penicillin/streptomycin, nonessential amino acids (Gibco, 11140050), sodium pyruvate (Corning, 25-000-CI), β-mercaptoethanol, and LIF (1000 U/ml) on mitomycin-C–treated mouse embryonic fibroblasts (as feeders); they were split onto 0.2% gelatin-precoated plates before each experiment. ESCs in 2iL medium were cultured in a 1:1 mix of DMEM/F12 (Hyclone, SH30023.01) and Neurobasal medium (Gibco, 21103049) with N2 (Gibco, 17502048) and B27 (Gibco, 17504044) supplements, GlutaMAX, penicillin/streptomycin, nonessential amino acids, sodium pyruvate, β-mercaptoethanol, LIF (1000 U/ml), 3 μM CHIR99021 (StemRD, CHIR-50), and 1 μM PD0325901 (StemRD, PD-50) on 0.2% gelatin-precoated plates. SL and 2iL media were changed daily. ESCs cultured in SL medium were cryopreserved in CELLBANKER 2 (Amsbio, 11891). After cell thawing, the same vial was used for culture in SL or 2iL. For the latter, ESCs cultured in SL were adapted to 2iL for three passages before each experiment. ESCs in SL or 2iL were passaged as single cells using 0.05% trypsin (Gibco, 25300054) every 3 days. The other two types of serum for SL medium were purchased from Fisher Scientific and Biowest; both were tested for ESC maintenance beforehand in the Esteban laboratory. Other inhibitors including JQ1 (BPS Bioscience, 27402), LDC000067 (Selleck Chemicals LLC, S7461), THZ1 (MedChemExpress, HY-80013), and actinomycin D (Sigma-Aldrich, A1410) were dissolved in dimethyl sulfoxide and added into the medium at the indicated concentrations. JQ1 and LDC000067 were added for 60 hours unless otherwise specified. E14gt2a (E14) ESCs were provided by I. Samokhvalov (Guangzhou Institutes of Biomedicine and Health, Chinese Academy of Sciences, China); they were used for all experiments unless otherwise specified. 129 and OG2 ESCs were provided by J. Liu (Guangzhou Institutes of Biomedicine and Health, Chinese Academy of Sciences, China). Tcf3 knockout ESCs (23) were provided by B. Merrill (University of Illinois at Chicago, USA). Gsk3 knockout ESCs, S33Y β-catenin–overexpressing ESCs, Ctnnb1 knockout ESCs, Ctnnb1 knockout ESCs rescued by either a wild-type or a C-terminal truncated form of β-catenin, Tet1/2 double knockout ESCs, and Tet1/2/3 triple knockout ESCs were previously reported (21, 24, 25, 45) .
shRNA transduction, RNA isolation, RT-qPCR, and RNA-seq
For shRNA experiments, ESCs cultured in SL or 2iL medium were infected with lentiviruses generated from HEK293T cells. Samples were extracted 96 hours after infection unless otherwise specified. shRNA inserts were cloned into pLKO.1 lentiviral vectors. All shRNA target sequences and RT-qPCR primers are listed in table S5. RNA samples were isolated using TRIzol reagent (Thermo Fisher Scientific, 15596026). RT-qPCR was performed using the SYBR Premix ExTaq Kit (Takara, RR420A) with an ABI 7500 real-time PCR machine. Data were analyzed in triplicate and normalized on the basis of Actb values. RNA-seq was performed by RiboBio Co. Ltd., China.
Animal experiments were compliant with all relevant ethical regulations regarding animal research and were conducted under the approval of the Animal Care and Use Committee of the Guangzhou Institutes of Biomedicine and Health under license number 2016012. For teratomas, ESCs were trypsinized, and 2 × 106 cells were injected into the flanks of immunocompromised nude mice. Mice were euthanized when the tumor diameter reached 1.5 cm, and the teratomas were processed for histological analysis. Chimeras were produced by injecting ESCs into blastocysts followed by implantation into a pseudopregnant C57BL/6J mice.
Proliferation, cell cycle, apoptosis, and AP activity assays
For proliferation assays, 60,000 ESCs were seeded, unless otherwise specified, per well of a six-well plate (three wells per time point). ESCs were counted at the indicated time points with a Bright-Line hemacytometer (Marienfeld). Cell cycle experiments were performed with propidium iodide staining (Beyotime, C1052) followed by flow cytometry analysis. Apoptosis experiments were performed with the Annexin V–FITC (fluorescein isothiocyanate) Apoptosis Detection Kit (Vazyme Biotech, A211) followed by flow cytometry analysis. Flow cytometry data were analyzed with FlowJo (v10.4) software. AP activity was detected with the BCIP-NBT Alkaline Phosphatase Color Development Kit (Roche, 11681451001).
Brd4 conditional knockout
Plasmid construction. Dual single guide RNAs (sgRNAs) were designed to target upstream and downstream intron of exon 5, respectively. Two sets of sgRNAs were designed, and the more efficient set was used for the experiments. sgRNAs were cloned into pX330-U6-Chimeric_BB-CBh-hSpCas9 (Addgene, 42230). PKD-EF1α-CreER with a puromycin resistance gene was obtained by subcloning pCAG-CreERT2 (Addgene, 14797) into a PKD-EF1α lentiviral backbone plasmid. The left and right homologous arms of the mouse genome and a fragment containing LoxP-exon5-FRT-PGK-Neo-FRT-LoxP were cloned into pMD-19 T donor plasmid (Takara, 6013).
Generation of Brd4fl/fl clones. ESCs cultured in 2iL medium were transduced with the donor and PX330-CAS9-sgRNA plasmids using Lipofectamine 3000 (Invitrogen, L3000015). G418 (Merk, 108321-42-2) was added 24 hours after transduction for selection. After selection, the remaining cells were seeded into a 96-well plate for genotyping. To obtain Brd4fl/fl clones, the remaining cells were again transfected with pCAG-FlpeGFP plasmid (Addgene, 13788) and the green fluorescent protein (GFP)–positive cells were sorted out to remove the selective marker Neo that was already integrated. Cells were then transduced with the donor and PX330-CAS9-sgRNA plasmids for a second round. After selection with G418, all the remaining cells were seeded again into a 96-well plate for genotyping. For Brd4fl/fl clones, the left-LoxP-exon5–containing fragment and right-LoxP–containing fragment were amplified for Sanger sequencing to make sure that the sequence and position of the LoxPs and exon 5 were correctly modified.
Generation of Brd4 fl/− and Brd4−/− clones. Brd4fl/fl clones were transduced with the PKD-EF1α-CreER plasmid and selected with puromycin (InvivoGen, ant-pr-1) for 2 days. The expression level of CreER was tested by RT-qPCR. Cells were seeded into a 96-well plate, and 4-hydroxytamoxifen was added to induce deletion of the floxed alleles. Genotyping was performed to obtain Brd4 fl/− and Brd4−/− clones. Brd4 fl/− clones were transduced with pCAG-CreGFP plasmid (Addgene, 13776), and GFP-positive cells were sorted 72 hours later. The sorted GFP-positive cells were seeded into a 96-well plate for genotyping to get Brd4−/− clones. All primers are listed in table S5.
Ten million cells were cross-linked in freshly prepared formaldehyde solution (1% final concentration for 10 min at room temperature) and then quenched with 125 mM glycine (for 5 min at room temperature). Fixed cells were washed with cold phosphate-buffered saline (PBS), harvested, flash-frozen in liquid nitrogen, and stored at −80°C for further use. For β-catenin, Pol2 Ser5P, and Pol2 Ser2P ChIP-qPCR, immunoprecipitation was performed as reported by Ward et al. (56). For MED1, SMC1A, and BRD4 ChIP-qPCR, immunoprecipitation was performed as reported by Finley et al. (41). After elution of antibody-bound complexes from the beads, cross-linking was reversed by overnight incubation at 65°C. Samples were diluted in TE (Tris-EDTA) buffer and then treated with ribonuclease A (Sigma-Aldrich, R6513) for 1 hour at 37°C, followed by incubation with proteinase K (Thermo Fisher Scientific, 25530049) for 2 hours at 55°C. DNA was purified using the QIAquick PCR Purification Kit (Qiagen, 28106). Antibodies used for ChIP-qPCR were immunoglobulin G (Abcam, ab172730), anti–β-catenin (Abcam, ab32572), anti–Pol2 Ser5P (Abcam, ab5131), anti–Pol2 Ser2P (Abcam, ab5095), anti-MED1 (Bethyl, A300-793), anti-SMC1A (Bethyl, A300-055), and anti-BRD4 (Bethyl, A301-985A). Primers for ChIP-qPCR are listed in table S5.
GRO-seq was performed as previously described (57). Briefly, nuclei from 107 ESCs were extracted and run-on-transcribed with BrUTP (Sigma-Aldrich, B7166) and other nucleoside 5’-triphosphates at 30°C for 5 min. Nascent RNA was enriched by agarose-coated anti-BrUTP (Santa Cruz Biotechnology, sc-32323). Poly(A) tail was added to the nascent RNA by poly(A) polymerase (New England Biolabs, M0276S) to synthesize complementary DNA with oligo(dT) primers. GRO-seq libraries were amplified by PCR for 10 cycles and separated with 10% tris-borate EDTA polyacrylamide gels. Bands ranging from 160 to 300 base pairs (bp) were cut and purified by isopropanol precipitation. Sequencing of GRO-seq libraries was performed by Berry Genomics Co. Ltd., China.
Co-immunoprecipitation and Western blotting
Cells (107) were lysed in 250 μl of TNE lysis buffer [50 mM tris-HCl (pH 7.5), 250 mM NaCl, 0.5% NP-40, and 1 mM EDTA] containing protease inhibitor cocktail (Roche, 04693132001) on ice for 15 min. Lysates were homogenized by a 0.4-mm syringe needle and centrifuged at 13,000g for 15 min at 4°C. Supernatants were diluted with TNEG buffer [50 mM tris-HCl (pH 7.5), 50 mM NaCl, 0.5% NP-40, 20% glycerol, and 1 mM EDTA] and then incubated with the relevant antibodies overnight at 4°C. The next day, 30 μl of prewashed Protein A/G–conjugated beads (Thermo Fisher Scientific, 10001D and 10003D) was added and incubated for 3 hours at 4°C. Beads were then washed three times with wash buffer 1 [20 mM tris-HCl (pH 7.4), 125 mM NaCl, and 0.1% NP-40] and two times with wash buffer 2 (1 × PBS with 0.02% NP-40) for 5 min under rotation at 4°C (for each wash). Last, the proteins were eluted with 60 μl of SDS loading buffer and boiled for Western blotting. The following primary antibodies were used for immunoprecipitation or Western blotting: anti-BRD4 (Bethyl, A301-985A), anti-tubulin (Sigma-Aldrich, T5201), anti–β-catenin (Abcam, ab32572), anti-MED1 (Bethyl, A300-793), anti-MED12 (Bethyl, A300-774A), and anti-SMC1A (Bethyl, A300-055).
ATAC-seq library construction
Cells (50,000) were washed once with cold PBS and resuspended in 50 μl of lysis buffer [10 mM tris-HCl (pH 7.4) 10 mM NaCl, 3 mM MgCl2, and 0.1% (v/v) IGEPAL CA-630]. The suspension was then centrifuged at 500g for 10 min at 4°C, followed by addition of 50 μl of transposition reaction mix of the TruePrep DNA Library Prep Kit (Vazyme Biotech, TD502). Samples were then incubated at 37°C for 30 min. Transposition reactions were cleaned up using the MinElute PCR Purification Kit (Qiagen, 28004). ATAC-seq libraries were subjected to five cycles for preamplification and amplified by PCR for an appropriate number of cycles. The amplified libraries were purified using the QIAquick PCR Purification Kit (Qiagen, 28104). Library concentration was measured using the VAHTSTM Library Quantification Kit (Vazyme Biotech, NQ101). Libraries were sequenced by Berry Genomics Co. Ltd., China.
For RNA-seq gene expression quantification, data were first aligned with STAR (v2.5.2) and quantified according to GENCODE vM15 in an RSEM-based pipeline (58). Differentially expressed genes were determined by DESeq2 (v1.18.1) and were defined as absolute fold change of >2 and q value of <0.1. Functional annotation was further performed by ClusterProfiler (v3.6.0) (59). For ChIP-seq and ATAC-seq, data were first aligned to the mm10 mouse genome assembly using Bowtie2 (v2.2.5) with the settings “–very-sensitive.” Low-quality mapped reads were removed using Samtools with the settings “-q 30.” Duplicated reads were collapsed using Picard (v1.9.0). For ChIP-seq, binding sites were called using MACS2 (v2.1.0) with the settings “–keep-dup 1 -q 0.01.” Peaks were annotated to all genes within 2 kb and the single closest gene within 20 kb, and duplicate genes were removed. Peaks were considered as overlapping if they intersect with each other. For GRO-seq, adaptors were first trimmed with fastp (v0.20.0), and only read 1 was kept for further analysis. PCR duplicates were collapsed using FASTX-Toolkit. A 20-bp polyA sequence and an 8-bp random sequence were trimmed from 3′ end. Clean data were then aligned to the mm10 mouse genome assembly using Bowtie2. For quantification of Pol2 ChIP-seq and GRO-seq signals, the proximal promoter was considered as the −100- to +300-bp region around the annotated TSS, and the gene body was considered as the +300-bp to +2-kb region downstream of the annotated TSS. Reads were first normalized as reads per million mapped reads or reads per kilobase per million mapped reads using deepTools (v 3.3.1) and further assigned to the corresponding regions, while the top 1% of the values were trimmed. For whole-genome bisulfite sequencing analysis, data were aligned to the mm10 mouse genome assembly using BSMAP with the settings “-v 0.1 -g 1 -p 8 -R -u” and further assigned to corresponding regions. Occupancy plots were generated by deepTools. Cumulative plots, violin plots, and boxplots were generated by ggplot2 (v2.2.1); the black central line of boxplots is the median, the boxes indicate the upper and lower quartiles, and the whiskers indicate the 1.5 interquartile range.
Wnt reporter activity assay
ESCs were seeded 12 hours before transfection on gelatin-precoated 24-well plates at a density of 30,000 cells per well. TOPflash or FOPflash report plasmids (Millipore, 17-285) and Renilla luciferase plasmids were transduced using Lipofectamine 3000. Twenty-four hours after transfection, luciferase activity was measured using a Dual-Luciferase Reporter Assay System (Promega, E1910).
Cell viability assay for THP1 cells
CCK-8 Cell Counting Kit (Vazyme Biotech, A311-02) was used to evaluate the cell viability of THP1 cells. THP1 cells (8,000 per well) were seeded in a 96-well plate. For measurements, 10 μl of CCK-8 solution was added to each well, and the plates were incubated for 1 to 4 hours at 37°C before the absorbance was measured at 450 nm using an Epoch 2 microplate spectrophotometer from BioTek.
Statistics and reproducibility
Data of bar charts are represented as mean ± SEM. The P value was calculated using the unpaired two-tailed Student’s t test or two-way analysis of variance (ANOVA). The number of replicates for each experiment is indicated in the figure legends. For violin plots and boxplots, the P value was calculated using Wilcoxon rank sum test.
Acknowledgments: We thank all members of the Esteban laboratory for their comments. We also thank M. Oren (Weizmann Institute of Science, Israel) for technical advice and J. T. Lis (Cornell University, USA) and X. Fu (University of California, USA) for helpful comments on this manuscript. We also thank the technical support from the Guangzhou Branch of the Supercomputing Center of Chinese Academy of Science and the Experimental Animal Center of Guangzhou Institutes of Biomedicine and Health, Chinese Academy of Sciences. Funding: This work was supported by the National Key Research and Development Program of China (2016YFA0100102, 2016YFA0100701, 2016YFA0100300, and 2018YFA0106903), the Strategic Priority Research Program of the Chinese Academy of Sciences (XDA16030502), the National Natural Science Foundation of China (31671537, 31571524, 31501192, 31430049, 31850410463, 31970619, 31950410553, and 31900617), the Guangdong Province Science and Technology Program (2014A030312001, 2015A030308007, 2016B030229007, 2016A050503037, and 2017B050506007), the Guangzhou Science and Technology Program (201807010066), the Innovative Team Program of Guangzhou Regenerative Medicine and Health Guangdong Laboratory (2018GZR110103001), and the Science and Technology Planning Project of Guangdong Province, China (2017B030314056). J.H.H. was funded by Pascal and Ilana Mantoux, Helen and Martin Kimmel Institute for Stem Cell Research, Flight Attendant Medical Research Council (FAMRI), European Research Council (ERC-CoG), and an Israel-China Israel Science Foundation (ISF) grant. C.W. was supported by a Zhujiang Overseas Young Talents Postdoctoral Fellowship. S.K. was supported by a Chinese Academy of Sciences President’s International Fellowship. M.M.A., D.P.I., and M.T. were supported by the Chinese Academy of Sciences–Third World Academy of Sciences (TWAS) President’s PhD Fellowship. A.S. was supported by the Deutsche Forschungsgemeinschaft (REBIRTH and SFB738). Author contributions: M.A.E., M.Z., and Y. Lai conceived the idea and designed the experiments. M.Z. conducted most of the experiments and Y. Lai performed most of the bioinformatics study. M.A.E., M.Z., and Y. Lai analyzed the data. V.K. and L.C. contributed critically to the experiments. P.G., X.G., Jianguo Zhou, Y.X., Z.Y., L.L., A.J., W.L., M.M.A., G.M., N.L., X.F., Y. Lv., M.J., M.T., S.K., H.L., X.X., H.Z., Y.H., L.W., S.C., I.A.B., Z.L., D.W., T.Z., C.W., M.H., D.P.I., Y. Li, Jiajian Zhou, J.Y., Y.F., K.A., U.D.V., F.G., A.P.H., and G.V. contributed to the experiments and/or the analyses. X.B., G.W., A.S., H.W., H.S., B.Q., A.P.H., B.W.D., C.H., M.P.C., Y.Q., G.-L.X., R.C., and G.V. provided relevant advice, essential materials, and/or infrastructural support. M.A.E. supervised the study and provided most of the financial support. J.H.H. contributed to the supervision and also provided financial support. M.A.E. wrote the manuscript with help from M.Z. and Y. Lai. M.A.E. approved the final version of the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors. RNA-seq, GRO-seq, and ATAC-seq data have been deposited in the Gene Expression Omnibus database under the accession number GSE123692. Published datasets used in this study are listed in table S6.
- Copyright © 2020 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. Distributed under a Creative Commons Attribution NonCommercial License 4.0 (CC BY-NC).