Global hyperactivation of enhancers stabilizes human and mouse naive pluripotency through inhibition of CDK8/19 Mediator kinases

Pluripotent stem cells (PSCs) transition between cell states in vitro, reflecting developmental changes in the early embryo. PSCs can be stabilized in the naive state by blocking extracellular differentiation stimuli, particularly FGF–MEK signalling. Here, we report that multiple features of the naive state in human and mouse PSCs can be recapitulated without affecting FGF–MEK signalling or global DNA methylation. Mechanistically, chemical inhibition of CDK8 and CDK19 (hereafter CDK8/19) kinases removes their ability to repress the Mediator complex at enhancers. CDK8/19 inhibition therefore increases Mediator-driven recruitment of RNA polymerase II (RNA Pol II) to promoters and enhancers. This efficiently stabilizes the naive transcriptional program and confers resistance to enhancer perturbation by BRD4 inhibition. Moreover, naive pluripotency during embryonic development coincides with a reduction in CDK8/19. We conclude that global hyperactivation of enhancers drives naive pluripotency, and this can be achieved in vitro by inhibiting CDK8/19 kinase activity. These principles may apply to other contexts of cellular plasticity. Lynch et al. demonstrate that inhibiting CDK8 and CDK19 kinases increases Mediator-driven recruitment of RNA Pol II to promoters and enhancers, therefore stabilizing the naive transcriptional program.

complexity-CDK8 can phosphorylate multiple Mediator subunits, the RNA Pol II C-terminal regulatory domain, chromatin regulators and transcription factors 4-6,11-17 . PSCs provide a prototypical model of cellular plasticity, the transcriptional program of which can be stabilized, extinguished or recaptured [18][19][20][21][22][23] . Although human PSCs offer great therapeutic promise, successful clinical applications remain limited, as human pluripotency is less characterized and less stable in vitro compared with in mice 21,22,24 . Chemical inhibition of MEK and GSK3 kinases with a two-inhibitor cocktail known as 2i shields mouse PSCs from extracellular differentiation in a state that is known as naive pluripotency 25 . Mouse PSCs cultured in 2i (referred to as 2i-naive cells) phenocopy the stable and homogenous state of undifferentiated naive pluripotency that exists transiently in embryonic day 4.5 (E4.5) preimplantation embryo epiblast 18,19,25 . By contrast, culture of PSCs without 2i shifts cell identity towards the postimplantation epiblast at ~E6.5, also known as primed pluripotency 18,19,23 . Enhancer destabilization by chemical blockade of BRD4, which is a key component of enhancers and SEs, triggers the loss of Mediator-driven gene expression in many cell types and induces differentiation in primed PSCs [26][27][28] . Notably, 2i-naive PSCs are highly resistant to enhancer destabilization 28 , indicating that there is an association between naive pluripotency and enhancer stability/resilience. MEK inhibition has been implicated upstream of potent and rapid reconfiguration of the transcriptome, proteome and DNA methylome, within embryonic or 2i-naive pluripotency [18][19][20]23 . However, the molecular mediators of 2i that are responsible for enhancer stabilization remain unclear.
Here we assessed the effect of inhibiting the activity of the Mediator CDK8/19 kinases, in order to elucidate the transcriptional basis of PSC identity and their plasticity. In summary, stimulating Mediator through its kinase module represses differentiation, favours self-renewal and upregulates preimplantation naive epiblast gene expression in mouse and in human.

Results
Inhibition of Mediator kinase stabilizes mouse naive pluripotency. GFP knock-in reporters at key stem cell marker genes such as Nanog represent well-established and precise indicators of the naive (GFP high ) and primed states (GFP low ) 18,22,29 . For example, in the 2i-naive state, Nanog promoter activity is enhanced, yielding a characteristically homogenous Nanog-GFP high cell expression pattern and uniform dome-shaped colonies (Fig. 1a-c and Extended Data Fig. 1a). By contrast, the Nanog promoter is metastable in primed-state PSCs, reversibly oscillating between high and low activity, presenting a heterogeneous Nanog-GFP expression pattern and flattened diffuse colonies, indicative of a general underlying switch in the transcriptional program 18,20,23,29,30 . The BRD4 inhibitor JQ1 destabilizes enhancers, resulting in colony flattening and a GFP low status (Fig. 1a), as reported previously [26][27][28] . In this experimental setting, we tested the effect of manipulating the transcriptional cyclin-dependent kinases (CDK7, CDK8/19 and CDK9) with a panel of small-molecule inhibitors. Several potent and structurally unrelated CDK8/19 inhibitors had a positive effect, inducing the formation of homogenous dome-shaped colonies, and upregulating both the Nanog-GFP reporter and endogenous Nanog expression, similar to PSCs in the 2i-naive state ( Fig. 1a-e, Extended Data Fig. 1a and Supplementary Table 1), whereas inhibition of CDK7 or CDK9 did not. We assessed the potency and selectivity of CDK8/19 inhibitors, commercially available or developed in-house, using multiple methods: (1) selectivity was suggested using a KinomeScan panel of 456 kinases; (2) LanthaScreen assays demonstrated inhibitory activity at nanomolar concentrations against pure recombinant CDK8-CCNC and CDK19-CCNC; (3) luciferase reporter cell assays (TOP-FLASH); and (4) potent inhibition of phosphorylation of STAT1 at Ser 727 in human PSCs, which is a well-documented CDK8 target site 11,14,16,31 (Fig. 1f On the basis of these data, we focused on the molecule that was generated at the Centro Nacional de Investigaciones Oncológicas (CNIO)-CDK8/19i-ETP-47799 (hereafter, CDK8/19i), which was the most effective at improving mouse PSCs (Fig.1a,b and Extended Data Fig. 1a; information about the structure and characterization of this inhibitor, as well as a comparison with other inhibitors used in this study, is provided in Supplementary Table 1 and Supplementary  Information). In addition to improvements in the Nanog-GFP profile and colony morphology described above, the effect of CDK8/19i on mouse PSCs resembled 2i in three other ways: (1) it was observed in serum-containing and serum-free-based media ( Fig. 1a and Extended Data Fig. 1a); (2) it was reversible after CDK8/19i withdrawal, with kinetics similar to that of 2i removal (Extended Data Fig. 1c); and (3) after removal of LIF or inhibition of LIF signalling using a JAK inhibitor, the presence of CDK8/19i delayed the downregulation of Nanog-GFP expression (Extended Data Fig. 1d,e). We conclude that inhibiting Mediator kinase CDK8/19 shifts mouse PSC morphology and Nanog expression towards their characteristic status in the naive state 18,23,29 .
As a genetic validation, depletion of CDK8, CDK19 and, most successfully, their regulatory partner cyclin C (CCNC; which is  Table 1 and the Supplementary Information b, FACS analysis of Nanog-GFP expression with three different CDK8/19i inhibitor molecules. Nanog-GFP low and Nanog-GFP high cell populations in the serum/LIF population (grey). The dotted line indicates the threshold at which >95% cells are Nanog-GFP High in 2i-naive culture conditions. Data are representative of three experiments. c, PSC colony morphology in the indicated treatments. Bright-field images and Nanog-GFP expression are shown. Data are representative of six experiments. d,e, Endogenous Nanog mRNA (d) or protein (e) expression levels in mouse PSCs adapted to the indicated conditions. Data are representative of three experiments. Data are mean ± s.d. Statistical analysis was performed using unpaired two-tailed t-tests; *P < 0.05. f, The levels of the CDK8-target STAT1 phosphorylation at Ser 727 (Ser 727P). HERVH human iPSCs treated with CDK8/19i concentrations for 3 h with or without the simultaneous induction of STAT1-Ser 727P by interferon-γ for 3 h. Data are representative of two experiments. g, Cell morphology and qPCR with reverse transcription (RT-qPCR) analysis of mouse PSCs after 7 d of shRNA-mediated knockdown of CDK8, CDK19 and CCNC (which encodes cyclin C). Data are the mean values from two experiments. h,i, Cell morphology and alkaline phosphatase staining (h) and FACS analysis of endogenous NANOG and OCT4 protein levels (i) in CDK8/19-dKO iPSCs stably expressing pMSCV-Empty (Empty) or pMSCV-CDK8-kinase dead (CDK8-KD). Data are representative of three independent clones. j,k, RT-qPCR analysis (j; data are mean ± s.d. from n = 3 independent clones) and western blot analysis of protein expression (k; data are representative of two experiments) in WT iPSCs or CDK8/19-dKO iPSCs stably expressing pMSCV-Empty or pMSCV-CDK8-KD, adapted to the indicated medium conditions. l, Alkaline phosphatase (AP) staining. Cells were fixed and stained 14 d after retroviral expression of pMSCV-Empty or pMSCV-CDK8-KD. Staining intensity was scored visually for each colony using ten fields of view. Data are mean ± s.d. from three experiments. m, Immunofluorescence in CDK8/19-dKO iPSCs expressing pMSCV-CDK8-KD-puro-IRES-GFP. Data are representative of four experiments. For c, g, h and m, scale bars, 100 μm. essential for full kinase activity 8 ) by short-hairpin RNA (shRNA) knockdown led to upregulation of Nanog expression and naive-like colony morphology ( Fig. 1g and Extended Data Fig. 1f,g). In another genetic approach, we generated CDK8/19 double-knockout (dKO) mouse PSCs (Extended Data Fig. 1h-k). CDK8/19-dKO PSCs could self-renew indefinitely, but CDK8/19-dKO was insufficient to confer naive morphological features or Nanog upregulation. Importantly, CDK8/19-dKO PSCs no longer responded to CDK8/19 inhibitors (Extended Data Fig. 1l,m). Together, this suggested that the beneficial effects observed may require the physical   Nanog-GFP presence of the inactive kinase. In agreement, we found that reconstituting CDK8/19-dKO PSCs with exogenous CDK8 rescued the ability of these cells to respond to CDK8/19i, observed by naive morphological features and Nanog, Klf4 and Oct4 upregulation (Extended Data Fig. 1l,m). Moreover, CDK8/19-dKO PSCs that were reconstituted with a CDK8 kinase-dead mutant (CDK8-KD; D173A) displayed homogenous naive colony morphology, high expression of naive-state markers ( Fig. 1h-m) and downregulation of Fgf5, which is a key marker of the primed state 18,19,23 (Fig. 1j), all without the need for any chemical inhibitor and despite maintaining active MEK-ERK signalling (Fig. 1k) Fig. 2b,c). Long-term CDK8/19i-adapted PSCs displayed typical developmental capacity after inhibitor withdrawal, specifically, retinoic-acid-induced differentiation, embryoid body cardiac centre formation, spheroid polarization and lumenogenesis 35 , generation of teratomas containing three germ layers, and robust chimaera contribution after morula aggregation and blastocyst microinjection assays (traced by constitutive GFP or RFP) evaluated at E4.5, E7.5, E14.5 and in fully developed adults that subsequently completed germline transmission ( Fig. 2d-i and Extended Data Fig. 2d-f). Notably, the continued presence of CDK8/19i impaired the early developmental events 35 of polarization and lumenogenesis in vitro (Fig. 2e), an observation that is discussed below. Thus, PSCs that are long-term adapted to CDK8/19i maintain upregulation of naive features, self-renewal and developmental capacity.

CDK8/19i induces and stabilizes the naive state in human PSCs.
We tested the effect of CDK8/19i on human stem cell identity.
STAT3 overexpression plus 2i induces the human naive state 36 , and we observed that CDK8/19i could replace 2i in this system (Fig. 2j). Even in the absence of STAT3 overexpression, other transgenes or chemicals, CDK8/19i treatment progressively converted human induced PSC (iPSC) colonies from flat and primed-like, to dome-shaped naive-like birefringent morphology. This was observed for a total of 7 human PSC lines treated with 0.4 µM or 1.1 µM CDK8/19i/LIF for 2-3 weeks ( Fig. 2k and Extended Data Fig. 2g), including human iPSCs carrying a specific HERVH-GFP reporter insertion that marks human naive cell identity 37 (Fig. 2k and Extended Data Fig. 2h). A 2i-based chemical cocktail (hereafter 2i p38iJNKi) induced naive colony morphology, as expected 33,37 , and combined with selection by cell sorting yielded cultures with homogeneous HERVH-GFP high (Extended Data Fig. 2h). Interestingly, treatment with CDK8/19 inhibitors (CDK8/19i or SnxA) also produced morphological conversion and increased GFP, similar to 2i p38iJNKi ( Fig. 2l and Extended Data Fig. 2h). The changes induced by CDK8/19 inhibition were gradual, required no selection after passage (sorting or manual picking), required no additional supplements except for rhLIF and were stable in the continuous presence of the inhibitor. By contrast, CDK7 inhibition failed to change colony morphology or GFP fluorescence, and produced cell death (Extended Data Fig. 2h). Culturing human PSCs in CDK8/19i, with or without p38iJNKi, increased their clonogenicity, alkaline phosphatase intensity and pluripotency markers [32][33][34]38 NANOG, OCT4, SSEA4, TRA1-81, TFCP2L1 and KLF17 (Figs. 2m and 3a, and Extended Data Figs. 2i and 3a-c). MYC, which is known to be reduced in naive cells 25,32 , was also reduced in cells maintained in CDK8/19i (Fig. 3a). Thus, similar to the observations in mouse PSCs described above, treatment of human PSCs with CDK8/19i establishes features that are characteristic of the naive state.

Developmental potential of CDK8/19i-adapted human PSCs.
Chemical induction of the human naive state can trigger genomic instability, severely impairing developmental potential 24,39 . We found that CDK8/19i-adapted human PSCs (five lines) had a normal karyotype over >16 passages (Extended Data Fig. 3d) and, after inhibitor withdrawal, maintained the capacity to contribute towards all three embryonic germ layers by embryoid-body differentiation in vitro and by teratoma assay in vivo (Fig. 3b-d  and Supplementary Table 1), comparable to control primed cells. Preimplantation interspecies chimerism tests for naive-specific properties, namely, capacity for clonal survival in a host embryo 40,41 . We tested CDK8/19i-adapted human iPSCs carrying a constitutive EpiSCs expressing pMSCV-Empty or pMSCV-CDK8-KD, then 7 d in EpiSC medium or standard embryonic stem cell (ESC) medium serum/LIF. Data are mean ± s.d. from n = 3 experiments. Scale bars, 100 μm. b,c, Clonogenicity of mouse PSCs. Nanog-GFP PSCs were FACS-sorted one cell per well, cultured for 7 d and then stained for alkaline phosphatase (b) or scored for Nanog-GFP intensity (c) to assess the pluripotent status of each colony, in standard medium serum/LIF, 2i-naive or CDK8/19i conditions. Data are representative of three experiments. d,e, The differentiation capacity of mouse PSCs that were previously adapted to serum/LIF, 2i or CDK8/19i. d, PSCs differentiated as indicated in two-dimensional culture. Analysis of PSC exit from pluripotency (Nanog downregulation) and differentiation (Nestin upregulation) using RT-qPCR. Data are the mean values of two experiments. e, Pluripotency exit assessed using immunofluorescence after PSC culture in three-dimensional Matrigel with or without CDK8/19i/LIF to observe early epiblast development (rosette formation and lumenogenesis) in PSC spheroids 35 . 'Disorganized' indicates differentiation failure. Data are representative of three experiments, n = 30 spheroids per condition. Statistical analysis was performed using unpaired two-tailed t-tests; ***P = 0.0097. Scale bars, 10 μm. f-i, In vivo assays of developmental capacity. Mouse CDK8/19i-treated PSCs, constitutively labelled with ROSA26-GFP or Tg.CAG-Katushka, were aggregated with, or microinjected into, host E2.5 morulae. Embryo chimerism was assessed visually. E4.5 blastocyst, n = 10 (f); E6.5 egg cylinder, n = 10 (g); E14.5, n = 2 (h); and perinatal E19.5, n = 4 (i). In i, three male (M) adult chimaeras (bottom left, the percentage of chimerism on the basis of coat colour is indicated) displayed germline transmission, generating three litters (bottom right, coat colour confirmed germline transmission per litter). Scale bars, 25 μm (f), 100 μm (g), 1 mm (h), 1 mm (i). H&E, haematoxylin and eosin; BF, bright field. j, Induction of naive colony morphology in human OSCAR ESCs. Tamoxifen-inducible constitutively active STAT3/LIF/2i (TL2i) 36 , or substituting CDK8/19i for 2i (TLCDK8/19i). Data are representative of three experiments. Scale bars, 100 μm. k, Induction of naive colony morphology in three human PSC lines; primed or cultured for 14 d with CDK8/19i. Scale bars, 100 μm. l, Cytometry analysis of HERVH-GFP intensity per cell in human PSCs; primed or cultured for 14 d with CDK8/19i. For k and l, data are representative of >5 experiments. m, Western blots of pluripotency markers in human PSCs; primed or cultured for 14 d with 2i-based or CDK8/19i-based medium, with or without p38iJNKi. SMC1 was used as a loading control.
Tomato-red marker for human-rabbit interspecies chimerism by microinjecting them into E2.5 rabbit morulae. Interestingly, the presence of human cells (Tomato + ) was detected 72 h later in up to 50% of the injected rabbit blastocysts (Fig. 3e). By contrast, human PSCs in the primed state were unable to integrate or survive in rabbit embryos (0 out of 24 rabbit embryos), similar to previous reports for primed state human PSCs within the embryos of mice, pigs and cattle 40,41 . In summary, long-term adaptation of human PSCs to CDK8/19i stabilizes naive pluripotency while preserving their developmental potential. We conclude that the role of CDK8/19 in pluripotency is conserved in mice and humans and, therefore, presumably across mammals.
CDK8/19i resets the transcriptome and proteome similar to 2i. Using RNA-seq, we compared global gene expression in mouse PSCs that were long-term adapted to CDK8/19i versus 2i. Overall, CDK8/19i altered gene expression with a magnitude similar to the magnitude in 2i conditions, and with a highly significant overlap in the identity and biological functions of genes that were up-or downregulated in both serum-containing and serum-free media (Fig. 4a, Extended Data Fig. 3e,f and Supplementary Table 2). Compared with control primed conditions, naive pluripotency markers were enhanced in CDK8/19i and 2i ( Fig. 4b and Extended Data Fig. 3g), whereas differentiation markers were globally downregulated in CDK8/19i and 2i conditions (Supplementary Table 2).
Endogenous retrovirus (ERV) expression is highly stage-specific during mammalian preimplantation and precisely defines naive and primed PSC identity [42][43][44][45][46] . The transcriptomic overlap between CDK8/19i or 2i treatments extended to ERVs; similar viral families were significantly up-or downregulated in mouse PSCs ( Fig. 4c and Supplementary   in close parallel, displaying highly similar pattern of expression in the CDK8/19i and 2i-naive states (Extended Data Fig. 3g,h). Another aspect of the plasticity of mouse PSCs is their ability to transition to a two-cell-like (2C) state, specifically marked by hyperactivation of the MERVL family of ERVs and by Zscan4c expression 46,47 . Stabilization of the naive state with 2i impairs the 2C-like fluctuation 46,47 . We also observed this in CDK8/19i-treated PSCs using multiple 2C markers, including MERVL and Zscan4c, and   : a summary of all tested lineage markers determined using RT-qPCR (n = 17) or immunofluorescence (n = 6) is provided in the Supplementary Information). EB, embryoid body. c,d, Human PSCs were adapted to primed or CDK8/19i conditions and then analysed using a teratoma differentiation assay. Data are representative of three experiments (human PSC lines: H1, D2#2 and HERVH). c, Immunofluorescence imaging shows markers for three embryonic germ layers in the H1 and D2#2 cell lines, as indicated. Scale bars, 50 μm. d, A summary of all of the tested lineage markers (6) determined on the basis of immunofluorescence in teratomas generated from the three human PSC lines described in c. The plus symbol (+) indicates detected; '0' indicates not detected. ND, not determined. A summary of all of the differentiation markers tested for all three cell lines in c and d is provided as source data. c-troponin, cardiac troponin. e, Interspecies chimaera assay in vivo to assess the developmental capacity of human PSCs that were adapted to primed or CDK8/19i conditions. Constitutively labelled human iPSCs (hiPSCs) (tdTomato (red); HERVH iPSC line) were introduced into host rabbit morulae of ~E2.5. Chimerism was assessed visually 72 h later in ~E5.5 rabbit blastocysts. The number of human cells introduced (5 or 10) and the number of embryos (n) in each of the three experiments (Exp 1-3) are indicated at the bottom. Quantification of the number of human cells observed in rabbit embryos is shown; data are from three independent experiments. A representative image (top) shows immunofluorescence of E5.5 rabbit blastocysts; the ICM is indicated by a dotted yellow line (determined by NANOG staining (inset)). Human PSCs that were adapted to CDK8/19i displayed a moderate contribution to human-rabbit chimaeras. Scale bars, 20 μm.
Although PSC plasticity has been examined in terms of RNA expression, its proteome remains relatively poorly defined. We analysed the proteome of mouse PSCs in serum/LIF versus 2i-naive or CDK8/19i-adapted conditions. Across five mouse PSC lines, CDK8/19i altered the expression levels of 465 proteins, of which 159 (34%) changed in the same direction in 2i conditions (Fig. 4l,m, Extended Data Fig. 5a,b and Supplementary Table 4). Importantly, among the overlapping changes in both 2i-naive and CDK8/19i conditions, we noted that key pluripotency regulators, such as KLF4, and metabolic pathways, such as oxidative phosphorylation, featured among the most upregulated; by contrast, LIN28A, MYC-target genes and differentiation markers were downregulated ( Table 4). Furthermore, proteomic changes in 2i and CDK8/19i conditions were significantly correlated with the transcriptomic changes observed (Extended Data Fig. 5d,e).
In summary, CDK8/19i upregulates pluripotency markers, reshapes the endogenous retroviral transcriptome and represses differentiation markers in a manner that is similar to the transcriptomic and proteomic resetting that was observed in previous studies of naive pluripotency, in vitro and in vivo, in mice and humans.

CDK8/19i does not reset global DNA methylation levels.
Many 2i-based chemical cocktails induce global DNA hypomethylation, both in mouse and human PSCs 21 . This has been attributed to MEK-dependent stabilization of UHRF1, which is a critical factor for recruiting DNMT1 to the DNA 61 . Importantly, the pattern of demethylation induced by 2i diverges significantly compared with the preimplantation naive epiblast state, and is associated with PSCs exhibiting genomic instability, chromosomal defects and loss of pluripotency 24,39,62,63 . Recent 2i-variant cocktails (with partial MEK inhibition) offer the advantage of largely preserving global DNA methylation [62][63][64] . Importantly, neither mouse nor human CDK8/19i-adapted PSCs showed evidence of global DNA hypomethylation (Fig. 5a,b). Moreover, 2i or MEK inhibition alone induced demethylation of LINE L1 repeat regions (Fig. 5c) and major satellite regions (Extended Data Fig. 5f) but had no effect on the methylation of IAP repeats (Extended Data Fig. 5g), all as previously reported 65 . By contrast, CDK8/19i did not reduce methylation at any of these mouse repeat elements ( Fig. 5c and Extended Data Fig. 5f,g) or UHRF1 protein levels (Supplementary Table 4). Thus, CDK8/19i induces naive features in the absence of global DNA hypomethylation, and this is probably due to its lack of MEK inhibition ( Fig. 1k and see below) or UHRF1 downregulation. By not recapitulating the partial demethylation of the naive epiblast, CDK8/19i has the advantage of preserving chromosomal stability and pluripotency after cell expansion (see above), which is particularly relevant for naive human PSCs. This is consistent with variant medium cocktails that are based on minimizing MEK inhibition both in mouse and human naive PSCs 62-64 .
X-chromosome reactivation status is another molecular signature that has been reported in human naive pluripotency during MEK inhibition 21,66,67 , which may be inferred by assessing XIST RNA expression in female cells. However, analysis using quantitative PCR (qPCR) revealed that there was very low XIST expression in our primed human PSCs (Extended Data Fig. 5h), suggesting that erosion of X-chromosome silencing may have already occurred in the parental cells under primed conditions, as observed previously 67 . Notably, some 2i-based cocktails reactivate XIST expression even in primed human PSCs displaying erosion of X-chromosome-silencing 66,67 , but this was not the case for our CDK8/19i-adapted cells (Extended Data Fig. 5h). In summary, CDK8/19i treatment does not recapitulate the reactivation of XIST in X-chromosome-silencing-eroded primed cells, indicating another distinction between CDK8/19i and most human medium cocktails that are based on MEK inhibition.

CDK8/19i induces phosphorylation changes similar to 2i.
We assessed the phosphoproteome of mouse PSCs after only 15 min exposure to CDK8/19i or 2i to explain their phenotypic similarity. Strikingly, out of 622 altered phosphorylation sites, 495 (79.6%) were similarly regulated by CDK8/19i and 2i (Fig. 5d,e). The co-regulated phosphorylation sites occurred on proteins that   Table 5). Note that CDK8/19i did not inhibit the kinase activity of purified recombinant GSK3 or MEK (Supplementary Table 1), and CDK8/19 inhibition did not reduce the relative levels of phosphorylated ERK (Figs. 1k and 5f,g and Extended Data Fig. 5j). However, 2i treatment reduced CDK8/19 kinase activity ( Fig. 5f and Extended Data Fig. 5k) and moderately downregulated CDK8 protein levels (Extended Data Fig. 5l). These data suggest that CDK8/19 inhibition occurs downstream of 2i, such that both treatments result in highly overlapping changes to phosphorylation sites.
CDK8/19i resets global RNA Pol II loading similar to 2i. 2i and CDK8/19i induce similar phosphoproteomic changes (converging on transcriptional machinery) and similar transcriptomic changes. Thus, to understand how CDK8/19 inhibition phenocopies the transcriptome of 2i-induced naive pluripotency, we investigated the global regulation of RNA Pol II abundance on chromatin using chromatin immunoprecipitation with sequencing (ChIPseq) in mouse PSCs in 2i or CDK8/19i, versus serum/LIF as control. Overall, total and Ser 5-phosphorylated RNA Pol II genomic distri bution was consistent with published resources 23,68,69 (Encode, https://www.encodeproject.org/). We observed that 2i globally increases RNA Pol II binding to promoters (Fig. 5h,i   Data Fig. 6a-c), which was confirmed by reanalysing independent data 23,69 . Notably, this global effect of 2i was phenocopied by CDK8/19i, regarding both total and Ser 5-phosphorylated RNA Pol II (Fig. 5h,i and Extended Data Fig. 6a-c). We measured the abundance of RNA Pol II at the promoter, gene body and transcription termination site (TTS) for each gene (Fig. 5j, Extended Data Fig. 6d and Supplementary Table 6). Consistent with previous analyses in mouse PSCs 23,68 , most genes (90%) possessed a ratio of promoter to gene body loading of >2.0 ( Fig. 5j and Supplementary  Genes + 2i Genes + CDK8/19i    or termination subregions of each gene indicated that 2i induces an increase in RNA Pol II binding selectively to the promoter region ( Fig. 5j and Extended Data Fig. 6d). Importantly, this was recapitulated by CDK8/19i, increasing RNA Pol II binding to promoters at a similar magnitude to that observed in 2i-induced naive pluripotency, following a gene-specific pattern (Figs. 5j and 6a-f and  Supplementary Table 6). Thus, 2i-and CDK8/19i-induced naive pluripotency is accompanied by widespread accumulation of RNA Pol II abundance at promoters. We also observed a correlation between changes in the abundance of RNA Pol II at promoters in 2i or in CDK8/19i conditions, as well as changes in mRNA expression for each gene (Extended Data Fig. 6e-i). In summary, gene-specific changes in RNA Pol II promoter loading may explain a significant proportion of the mRNA expression profile characteristic of 2i-or CDK8/19i-induced naive pluripotency.
CDK8/19i and 2i trigger activation of SEs. The primary role of Mediator is at enhancers, regulating the recruitment of RNA Pol II to promoters 4-6 . Using published ChIP-seq datasets 2,3 (Supplementary Table 7), we confirmed that CDK8/19 was enriched at promoter, typical enhancer (TE) and SE regions as previously defined in mouse PSCs 2 ( Fig. 6g and Extended Data Fig. 7a-c). There is a strong correlation between the abundance of CDK8/19, Mediator subunits and other factors that are critical for enhancer activity 1,6 (such as, p300, CBP, Pol II or BRD4; Extended Data Fig. 7d); the highest levels of CDK8/19 occurred within SE regions ( Fig. 6g and Supplementary  Table 7); and, finally, putative target genes proximal to genomic CDK8/19-binding loci were highly enriched in preimplantation functions characteristic of pluripotent cell identity (Extended Data Fig. 8a-c). We therefore hypothesized that, in mouse PSCs, CDK8/19 inhibition might act through Mediator to trigger changes in enhancer activity, explaining the observed increase in RNA Pol II loading at promoters and regulation of pluripotent states. As CDK8/19 protein was particularly enriched at SE regions, we examined the impact of CDK8/19i on SE function. Enhancers contain RNA Pol II, which transcribes enhancer-derived RNAs (eRNAs)a process that faithfully reflects enhancer activity 4,8,70 . We therefore measured the effect of CDK8/19i or 2i on the levels of RNA Pol II and eRNAs at SEs. Importantly, the abundance of RNA Pol II was selectively increased at CDK8/19-binding sites and, accordingly, RNA Pol II recruitment was also preferentially increased at SEs compared with TEs (Fig. 6h,i and Extended Data Fig. 8d). Consistent with this, mouse PSCs treated with 2i or CDK8/19i showed an increase in enhancer-derived eRNA levels, as well as RNA Pol II abundance, within enhancers specific for the naive state 71 (Fig. 7a and Extended Data Fig. 8e,f). Induction of naive-specific eRNAs and naive marker genes was an early event, occurring within 48 h after adding 2i or CDK8/19i, and it was rapidly reversible (Fig. 7a and Extended Data Fig. 8g). Finally, consistent with naive-specific enhancer activation, the expression levels of SE target genes were preferentially upregulated in both 2i and CDK8/19i conditions (Fig. 7b,c and Extended Data Fig. 8h). We conclude that, in PSCs, CDK8/19i and 2i hyperactivate existing SEs and upregulate SE-target genes in a manner that reinforces naive pluripotency.

CDK8/19 inhibition compensates for BRD4 inhibition.
Loss of Mediator function preferentially decreases the expression of enhancer target genes across multiple cell types 4-6,26-28 . In particular, BRD4 inhibition in primed state PSCs decreases the ability of Mediator to recruit RNA Pol II, and this results in the loss of Mediator-driven transcription, a collapse in pluripotency gene expression and differentiation 27,28 (Fig. 1a). Compared with primed PSCs, naive PSCs are highly resistant to the decreased Mediator activity and enhancer destabilization induced by BRD4 inhibition 28 . Interestingly, mouse PSCs lacking endogenous CDK8/19 and reconstituted with kinase-dead CDK8 were resistant to enhancer destabilization by BRD4 inhibition for ten passages (>3 weeks), maintaining naive morphology, and showed high expression of alkaline phosphatase, naive-specific pluripotency markers and naive-specific eRNAs, similar to 2i-naive PSCs (Fig. 7d,e and Extended Data Fig. 8i,j). Thus, PSCs expressing kinase-dead CDK8 phenocopy the robust resistance to enhancer destabilization that is characteristic of 2i-naive PSCs.

The roles of CDK8/19 during early embryonic development.
Given our observations that CDK8/19 inhibition stabilizes naive pluripotency, we investigated CDK8/19 function during early embryonic development. We focused on CDK8, which we found is highly expressed compared with CDK19 in both mouse and human PSCs (Extended Data Figs. 1k and 9a). Using a CDK8-specific antibody (Extended Data Fig. 1j), we detected CDK8 protein from the mouse zygote to morula (Extended Data Fig. 9b). Consistent  with this, CDK8-KO zygotes cannot progress beyond the 4-8-cell stage 72 , and we observed that CDK8/19i impaired the progression of zygotes to the two-cell stage (Extended Data Fig. 9c). CDK8 activity is therefore essential for the zygote to morula transition. We next investigated the role of CDK8 post-morula. CDK8 mRNA expression declines until the blastocyst stage, both in mouse and human preimplantation embryos (Extended Data Fig. 9d-f). CDK8 protein expression per cell was homogenous in the mouse inner cell mass (ICM) at E3.5 (Fig. 8a,b). Interestingly-at E4.5, when the ICM segregates into the naive epiblast and the primitive endoderm (PE)-CDK8 protein levels diverged, with lower levels in the epiblast compared to PE (Fig. 8a,b and Extended Data Fig. 10a).
This pattern was transient, and it reversed in postimplantation epiblast at E5.5 (Fig. 8a,b and Extended Data Fig. 10a). To further document that CDK8 levels are reduced in the naive epiblast, embryos were cultured from E3.5-E4.5 with MEK inhibitor (MEKi), which blocks PE formation and permits only the development of naive epiblast 73,74 . As expected, in MEKi, the ICM contained only naive epiblast cells and not PE, this simplified the quantification of CDK8 in the ICM and trophectoderm and we confirmed reduced CDK8 expression in the ICM (Fig. 8c,d). The CDK8-binding partner and essential activating subunit cyclin C also altered its nuclearcytoplasmic ratio during this developmental window. Specifically, E4.5 epiblast contained significantly less nuclear cyclin C compared  with E5.5 epiblast in vivo (Extended Data Fig. 10b,c), and a similar pattern was observed when comparing between 2i-naive and primed state PSCs in vitro (Extended Data Fig. 10d). In summary, the emergence of naive pluripotency during embryo development at E4.5 coincides with decreased CDK8 expression and decreased availability of its essential subunit cyclin C. This parallels the effect of MEKi on CDK8 expression and stabilization of naive epiblast identity in PSCs in vitro (Extended Data Fig. 5).
We wondered whether inhibition of CDK8/19 affects the emergence of naive pluripotency. Similar to MEKi, CDK8/19i treatment during E3.5-E4.5 did not interfere with embryo naive epiblast development (Fig. 8e-g) and enabled the derivation of PSC lines. In contrast to MEKi, CDK8/19i permitted PE formation (Fig. 8e-g). This suggests that the critical roles of MEK for PE segregation are independent of CDK8/19, and agrees with our observation that MEK activity is unaffected by CDK8/19i (Fig. 5f,g).
Finally, we examined the importance of CDK8/19 activity during preimplantation-to-postimplantation epiblast developmental progression. We focused on lumen formation within the epiblast, which marks the initiation of morphogenesis downstream of naive pluripotency exit 35 . We found that CDK8/19i treatment during E4.5-E5.5 impaired embryo epiblast lumenogenesis ( Fig. 8h; for spheroids, Fig. 2e). This indicates that CDK8/19 activity is required to support epiblast development, from the naive preimplantation to primed postimplantation embryonic stages, consistent with the significant increase in CDK8 expression that we observed at this time (Fig. 8a,b and Extended Data Fig. 9e).
These data suggest that CDK8/19 expression in early embryonic development mirrors its function-the transition from zygote to morula, and the formation of the postimplantation epiblast require CDK8/19 activity; the intervening naive ICM has low CDK8 expression and reduced nuclear cyclin C (a summary is provided in Extended Data Fig. 10e).

Discussion
Here we uncovered a role for the Mediator kinases CDK8/19 in defining the equilibrium between naive and primed pluripotent states, in both mouse and human pluripotent cells. Collectively, our data indicate the following model: 2i and CDK8/19i rapidly induce a highly overlapping set of phosphorylation changes focused on the transcriptional machinery, triggering enhancer hyperactivation, a global increase in RNA Pol II recruitment to promoters and resetting gene expression. This includes upregulation of eRNAs, as well as resetting the expression of endogenous retroviral and repeat elements, as part of this cell identity conversion. Further evidence supporting transcriptional stabilization of naive pluripotency includes repression of 2C fluctuation in PSC identity, similar to in 2i conditions. Thus, the ability of 2i and CDK8/19i to induce naive features seems to originate from their common effect on Mediator and RNA Pol II transcriptional activity. In support of this, SEs interact with more target promoters 75 , engage in more long-range interactions 75 and display increased H3K27ac 76 in the naive state versus primed. Our model agrees with the concept that transitions in cell identity are driven by early reconfiguration of the active enhancer network, which resets the transcriptional machinery to the new program 70,71,77 .
The evidence presented here suggests a signalling hierarchy; in particular, MEK inhibition results in CDK8/19 inhibition, whereas inhibition of CDK8/19 does not affect MEK activity. Accordingly, we observed that (1) the ability of MEK and GSK inhibition (2i) to induce naive features in PSCs in vitro is recapitulated by CDK8/19i; (2) 78,79 ) drive postimplantation epiblast differentiation, a process that we found is impaired by CDK8/19i, and a period during which CDK8 is upregulated. Thus, we propose that CDK8/19 inhibition is a common downstream feature of naive-inducing medium cocktails. Further studies will elucidate how MEK-ERK signalling regulates CDK8/19 Mediator activity in PSCs. Interestingly, Mediator hyperactivation through CDK8/19 inhibition triggers cancer cell death 14 , while we find a similar approach reinforces naive pluripotent identity. Cancer cells commonly develop novel oncogenic SEs, becoming addicted to a defined range of enhancer-driven transcription that seems to be sensitive to perturbation 9,80 . This provides an interesting parallel with MEK inhibition, which is also detrimental to many cancer cells, but beneficial to pluripotency.
Stabilization of the human naive pluripotent state in vitro is challenging and remains to be optimized 21,24 . Our understanding of stem cell identity indicates a continuum of molecular changes along a spectrum from naive to primed states, which also reflects the developmental path in early embryos 18,19,21,22 . Where does CDK8/19i position PSCs along this gradient? We found that CDK8/19 inhibition recapitulates the majority of the molecular characteristics that are associated with the primed-to-naive transition. However, other molecular features that are associated with the more-naive end of this spectrum are not recapitulated by CDK8/19 inhibition, particularly, global DNA hypomethylation, X-chromosome reactivation 66,67 and SSEA4 downregulation 24,39 . Achieving these last features of naive pluripotency seems to come at a price. Naive-inducing medium cocktails that are dependent on MEK inhibition can generate harmful side effects, specifically acute chromosomal instability and imprinting erasure 24,39,62,63 . Interestingly, those other cocktails which do not downregulate SSEA4 and produce modest DNA demethylation, are not associated with genomic instability 24,33,64 . Similarly, CDK8/19i installs many naive features in human cells while maintaining SSEA4, DNA global methylation and genomic stability. CDK8/19i-treated cells retain a normal karyotype after prolonged culture. We suggest that these important differences are due to CDK8/19i not impinging directly on MEK signalling.
In summary, CDK8/19i stimulates the recruitment of RNA Pol II by Mediator. This hyperactivates enhancers and stabilizes the transcriptional program of naive pluripotent cell identity. Thus, chemical inhibition of CDK8/19 may help to solve remaining challenges in unstable human naive PSC culture. Similarly, these principles of stabilizing cellular identity may apply to other contexts of cellular plasticity.

M et ho ds
Human PSC resources. HERVH iPSCs were shared by the laboratory of Z. Izsvak (Max Delbruck Centre for Molecular Medicine) 37 . WIBR3 ESCs were shared by the laboratory of J. Hanna (Weizmann Institute of Science). OSCAR ESCs carrying inducible STAT3 were shared by the laboratory of P. Savatier (SBRI, Stem Cell and Brian Research Institute) 36 . H1 and H9 human ESCs, and CB5, D2#2 and D2#4 human iPSCs were shared by the laboratory of N. Montserrat (IBEC, Institute for Bioengineering).
Resetting human PSCs from primed to naive state using 2i-based medium cocktail. The naive human pluripotent state was obtained using two methods. OSCAR PSCs were reset to the naive state with 2i (TL2i) or CDK8/19i (1.1 μM or 0.4 μM) plus rhLIF and STAT3 transgene induction, as described previously 36 . In a transgene-free approach, human PSCs were cultured in a 2i-based chemical cocktail 33 referred to here as 2i p38iJNKi. Cells were maintained on Matrigel (BD Biosciences, 356231) using mTeSR1 (Stem Cell Technologies). Medium was supplemented with 20 ng ml −1 of recombinant human LIF (Peprotech, as described previously 33 ), 1 μM PD0325901 (MEKi, Axon Medchem), 1.5 μM CHIR 99021 (GSK3i, Axon Medchem), 10 μM SP600125 (JNKi, TOCRIS) plus 2 μM BIRB796 (p38i, Axon Medchem). To obtain and maintain the naive state using the 2i p38iJNKi medium cocktail, cells were selected at each passage by cytometry sorting for the top-10% HERVH-GFP levels or by repeated manual picking to select colonies with dome-shaped morphology. Conversion of human PSCs from primed to naive required three passages/rounds of selection over 14-18 d.
Derivation of mouse ESCs. ESC line derivation was performed using standard methods. Eight-cell stage mouse embryos obtained from the oviducts of pregnant female mice were cultured in serum/LIF on mitomycin-C-inactivated MEF feeders plus 2i or CDK8/19i (added fresh every 2 d) until the emergence of colonies from hatched blastocysts. The feeders were not compatible with several days of CDK8/19i treatment; cells were therefore passed to fresh feeders every 2 d, then transferred to 0.1% gelatin only.
Derivation of mouse EpiSCs. PSCs in 2i/LIF cultured on gelatin were first induced to differentiate into epiblast-like cells over 48 h by seeding on fibronectin-coated plates (10 ng ml −1 ) and switching to medium containing 1% KSR, N2B27, FGF2 (12 ng ml −1 ) and activin A (20 ng ml −1 ) 50 . After 48 h, the cells were in a flat epiblast-like cell state, and the medium was switched to include 20% KSR, and expanded for five passages to stabilize the EpiSC primed state, which was confirmed by typical flat colony morphology and Fgf5 expression. EpiSC colonies were passaged as clumps.
Analysis of PSC self-renewal. Mouse or human PSC self-renewal and pluripotency was scored by colony morphology, cytometry (in mouse cells, Nanog-GFP heterogeneity and overall intensity, and costaining for ICAM1; in human cells, HERVH-GFP intensity, and assessing the expression of NANOG, OCT4, SSEA4 and TRA1-81), alkaline phosphatase staining (fixed cells; Promega, S3771), and using immunofluorescence and RT-qPCR (for pluripotency markers, indicated in each figure). The intensity of alkaline phosphatase staining was quantified by scoring colonies observed using bright-field microscopy in ten random fields of view per well.
Mouse PSC differentiation with LIF removal and retinoic acid. LIF was first removed for 24 h by culturing in LIF-free differentiation medium (as described for serum/ LIF medium, except LIF was omitted). Next, retinoic acid was added (10 μM) from 24 h to 72 h, followed by LIF-free differentiation medium alone from 72 h to 96 h. Differentiation was also assessed using the same protocol of LIF-withdrawal except without adding retinoic acid.

Mouse PSC differentiation by hanging-drop culture and as embryoid bodies.
PSCs were transferred to LIF-free differentiation medium (as described above) and suspended in hanging-drop culture at 1,000-5,000 cells per 20 μl for 48 h to form embryoid bodies (EBs), followed by transfer to suspension culture in low-adherence Petri dishes. Fresh medium was added every 3 d, and development of beating cells in cardiac centres was scored daily.
Morula aggregation and blastocyst microinjection in mouse chimaera assays. After ten passages in serum/LIF, 2i or CDK8/19i, mouse PSCs labelled constitutively with Rosa26-GFP or Tg.CAG-Katushka 81 underwent morula aggregation at E2.5 or blastocyst microinjection at E3.5 as described previously 25 . The extent of GFP + or Katushka-red + cell chimeric contribution was assessed on the basis of confocal fluorescence at E4.5 or embryos were introduced into CD1 pseudopregnant females for implantation, and collected at the following postimplantation time points: E6.5, E14.5 or E19.5. Chimaeras that developed to adulthood were assessed by coat colour contribution and capacity for germline transmission.

Cardiac-tissue-and endoderm-directed differentiation of EBs derived from hPSCs.
Human PSC colonies were dissociated and cultured in suspension for 3 d to form EBs in DMEM/F12, 15% FBS, 2 mM l-glutamine, non-essential amino acids and penicillin-streptomycin. To generate endoderm, EBs were transferred to 0.1% gelatin-coated plates for 2 weeks in differentiation medium (DMEM, 20% FBS, 2 mM l-glutamine, 0.1 mM 2-mercaptoethanol, non-essential amino acids and penicillin-streptomycin). To generate cardiac tissue, differentiation medium was supplemented with 100 μM ascorbic acid (Sigma-Aldrich). In all conditions, EBs spontaneously gave rise to neural cell clusters.
Teratoma assays. For mouse PSCs, 10 6 cells in 100 μl were injected subcutaneously in nude mice. For human PSCs, 2 × 10 6 cells in 30 μl were injected into the testis of male SCID beige mice.
Mouse embryo manipulation and analysis. Mouse embryo collection, culture for preimplantation embryo development in vitro and fixation for immunofluorescence analysis was performed as described previously 74,35 . Pre-to postimplantation embryo development in vitro, immunofluorescence analysis of CDK8 and cyclin C levels in preimplantation mouse embryos, and lumenogenesis by mouse PSC embryoid formation in Matrigel were performed as described previously 35 ; further details are available from the corresponding author on request.
Viral production and iPSC reprogramming. Viral production and iPSC reprogramming were performed as described previously 84 . In brief, retroviral and lentiviral supernatants were produced in HEK293T cells. Filtered supernatants were collected after 48 h, and added to recipient cells in four infections. Retroviral supernatants delivered exogenous CDK8 expression constructs and iPSC reprogramming vectors. Lentivirus supernatants delivered shRNA knockdown vectors and CRISPR-Cas9 vectors. A list of plasmids is provided in Supplementary Table 8.
Interspecies chimaera developmental potency. Primed human iPSCs were precultured with ROCK inhibitor for 24 h, prepared as a unicellular suspension and electroporated (Neon Transfection System; Invitrogen; 1 pulse at 1,400 V for 20 ms) with 10 μg of DNA constructs for constitutive tdTomato expression (PB-Hygro-PGK-CAG-tdTomato and PBase pCMV-Transposase). Cells were subsequently plated on Matrigel in mTeSR1 medium supplemented with ROCK inhibitor for 24 h, then antibiotic selection with 20 μg ml −1 hygromycin was applied for 12 d, before cytometric sorting for tdTomato constitutively labelled cells.
Human PSCs were dissociated into single-cell suspension with trypsin, and 5-10 cells were microinjected under the mucus coat and zona pellucida of morula eight-cell stage rabbit embryos, the day after collection. After microinjection, embryos were sequentially cultured in CDK8/19i medium for 4 h, followed by 20 h incubation with a 1:1 mixture of RDH:CDK8i medium and finally in RDH medium for extended in vitro culturing. After 24 h of in vitro culture, early blastocyst stage embryos (E3.5) were rinsed three times in embryo-holding medium (IMV Technologies) and treated with 5 mg ml −1 protease E (Sigma-Aldrich) for 3 min at 37 °C to digest the mucus coat and weaken the zona pellucida. Embryos were then rinsed three times in 199 HEPES medium (Sigma-Aldrich) and cultured in RDH medium for 3 d until the late-blastocyst stage (E5.5). Rabbit embryos were fixed in 2% paraformaldehyde for 20 min at room temperature, washed in PBS + 0.1% Tween-20 and permeabilized in PBS + 1% Tween-20 overnight at 4 °C. After 1 h blocking with 5% donkey serum, embryo immunofluorescence was performed as described previously 36 . A list of the antibodies used is provided in Supplementary Table 8.

Molecular methods.
Transcriptional CDK inhibitors. The structure and characterization of the CNIO CDK8/19 inhibitor (CDK8/19i-47799) as well as notes on all of the other transcriptional CDK inhibitors used in this study are provided in Supplementary Table 1.

Generation of CDK8/19-dKO iPSCs.
To target mouse CDK19, we designed sgRNA against CDK19 exon 1, targeting 76 bp downstream of the ATG translation start site to generate indels (a schematic of which is provided in Extended Data Fig. 1i). sgRNA sequences and plasmid details are provided in Supplementary  Table 8. Primary CDK8-flox/flox RERT-Cre MEFs of passage 1-4 were infected with lenti-CRISPR-Cas9 containing the CDK19 sgRNA (pLenti-CRISPRV2; Addgene, 52961) followed by selection with puromycin (1 μg ml −1 ). CDK19-KO was assessed using western blot. The MEFs were reprogrammed to iPSCs, single clones were picked and expanded, and CRISPR-induced indels were characterized by sequencing the CDK19 target region for frameshift mutations. CDK19-KO iPSC clones were compared versus iPSC clones that retained wild-type CDK19 expression, and no effect of CDK19-KO was observed in MEFs or in iPSCs. CDK8-KO was induced by 6 d of culture with 0.5 μM 4-hydroxy-tamoxifen to induce Cre-mediated deletion of CDK8 exon 2 (a schematic of which is provided in Extended Data Fig. 1j). CDK8-KO was confirmed using allele-specific PCR (to demonstrate deletion of exon 2; Extended Data Fig. 1h) and using western blot (to demonstrate complete loss of CDK8 protein; Extended Data Fig. 1j,k).
Stable exogenous expression of CDK8. Wild-type CDK8 (CDK8-WT) and catalytically inactivated kinase-dead CDK8 (CDK8-KD; D173A) were cloned into pMSCV-puro-IRES-GFP (Addgene, 21654) using the BglII and HpaI restriction enzymes, and confirmed by sequencing. Retroviral supernatants were generated in HEK293T cells with the packaging plasmid pCl-Eco (Addgene, 12371), followed by retroviral expression into CDK8/19-dKO iPSCs. Two rounds of FACS-selection by GFP expression were performed to enrich for expressing cells, and CDK8-WT or CDK8-KD protein expression was confirmed by western blot (a schematic of which in addition to western blot data are provided in Fig. 1k and Extended Data Fig. 1l).
FACS cytometry. FACS analysis of SSEA1 or ICAM1 was performed using FlowJo v.9.6.2 as described previously 84 . Live-cell analysis of the Nanog-GFP used 2i-adapted mouse PSCs to define the threshold (95% of cells) for the homogenous Nanog-GFP high population, against which other treatment groups were compared (Fig. 1a,b). Live-cell sorting for human PSCs carrying HERVH-GFP selected the top 10% GFP-expressing cells, as previously described 37 . The FACS gating strategy for live/dead cell discrimination is provided in Extended Data Fig. 9g.
Cell lysis, fractionation and western blot. Cell lysis, fractionation and western blots were performed as described previously 84 . A list of the antibodies used is provided in Supplementary Table 8. Nuclear/cytoplasmic fractionation was performed using the NE-PER kit (Thermo Fisher Scientific, 78833).
G-banding karyotype methodology. Subconfluent mouse and human PSC lines were arrested in metaphase by adding 0.02 μg ml −1 KaryoMax Colcemid (Gibco). Twenty metaphase spreads were analysed per condition.
Immunohistochemistry and immunofluorescence analysis of mouse embryos. Mouse tissues were fixed in formalin at 4 °C, embedded in paraffin block and sectioned at a thickness of 5 µm. Staining was performed using standard methods. A list of the antibodies is provided in Supplementary Table 8.
Mouse teratoma and embryoid body immunohistochemistry. Mouse teratoma and embryoid body immunohistochemistry analysis was performed as described previously 84 . A list of the antibodies used is provided in Supplementary Table 8.
Cell immunofluorescence. PSCs were grown on chamber slides using culture conditions indicated in each experiment. Confocal immunofluorescence staining and microscopy was performed as described previously 35,84 using a Leica SP5 microscope. A list of the antibodies is provided in Supplementary Table 8.
DNA methylation. Global DNA methylation status was quantified by mass spectrometry (MS). CpG methylation status at individual CpG sites of repeat DNA regions was assessed by DNA bisulphite conversion and pyrosequencing. A list of the primers used for PCR amplification and sequencing is provided in Supplementary Table 8.
Image analysis. All image analyses were performed using Fiji (http://fiji.sc).
Proteomics. Full proteome quantitative analysis of five mouse ESC lines. Five mouse ESC lines (ZS, TNGA, TON, BL6 and V6.4) were cultured in serum/LIF (as a control) or, additionally, with either 2i or CDK8/19i for >2 weeks. Cell pellets were collected by trypsinization, washed with cold 1× PBS and preserved immediately at −80 °C for further analysis. Protein sample preparation for MS, protein digestion, our scheme for isobaric labelling with iTRAQ8plex, detailed settings for high pH reverse-phase fractionation, detailed settings for the whole-proteome analysis using liquid chromatography coupled with tandem MS (LC-MS/MS) and bioinformatics analyses with the whole-proteome data were reported previously 84 .

Phosphoproteome analysis of mouse PSC lines after 15 min of inhibitor treatment.
Two mouse ES PSC lines (TON and ZS) were cultured in serum/LIF (as a control) or, additionally, with either 2i or CDK8/19i. Cells were treated with inhibitor for precisely 15 min, after which the cells were collected rapidly by scraping in ice-cold PBS, washed with ice-cold PBS, snap-frozen on dry ice and preserved at −80 °C for further analysis. Sample preparation for MS, protein digestion, isobaric labelling, phosphopeptide enrichment, micro high pH reverse-phase fractionation, settings used for phosphoproteome LC-MS/MS and bioinformatics analyses with phosphoproteomic data were performed as described previously 11,84 .
Transcriptomics. RNA isolation and RT-qPCR. Total RNA was extracted (on-column; RNeasy kit with DNA digestion; Qiagen, 74104, 79254) and retrotranscribed into cDNA (Superscript Reverse Transcriptase; Biorad, 170-889). RT-qPCR was performed using the GoTaq qPCR Master Mix (Promega A6002) in an ABI PRISM 7700 thermocycler (Applied Biosystem). Input normalization of all RT-qPCR data was performed using the 2 ÀΔΔCt I method, using housekeeping genes Actb or Gapdh as indicated in each figure. A list of the primers used is provided in Supplementary Table 8.
RNA-seq transcriptomic analyses. The complete set of reads has been deposited in GEO (GSE112208 and GSE127186). A complete list of meta-analyses expression comparisons between this study and multiple mouse and human published datasets, in vitro and in vivo, is provided in Supplementary Table 3.
For RNA-seq analysis in mice, samples of 1 μg of total RNA (RIN numbers: 9.8-10; Agilent 2100 Bioanalyzer) were used. PolyA+ fractions were processed using the TruSeq Stranded mRNA Sample Preparation Kit (Agilent). Adapter-ligated library was completed by PCR with Illumina PE primers (8 cycles) and sequenced for 40 bases in a single-read format (Genome Analyzer IIx, Illumina).
For RNA-seq analysis in human cells, samples of total RNA (RIN numbers: 9.0-10; Agilent 2100 Bioanalyzer) were used. For library construction, 10 ng of total RNA samples were processed using the SMART-Seq v4 Ultra Low Input RNA Kit (Clontech) according to the manufacturer's instructions. The resulting cDNA was processed using the NEBNext Ultra II DNA Library Prep Kit for Illumina (NEB, E7645). Adapter-ligated libraries were completed by PCR (8 cycles), and sequenced for 50 bases in a single-read format, (Illumina HiSeq2500).
Reads were aligned to the reference mouse genome (GRCm38/mm10) or the human genome (GRCh37/hg19) using TopHat-2.0.4 (using Bowtie v.0.12.7 and Samtools v.0.1.16, allowing for two mismatches and five multihits). Transcript assembly, estimation of abundance and differential expression were calculated using Cufflinks v.1.3.0. When comparing samples, total read numbers were normalized and visualized using SeqMiner v.1.3.3e or Integrated Genome Viewer from the Broad Institute (http://software.broadinstitute.org/software/igv/) Functional analyses of differential gene expression. Lists of differentially expressed genes are provided in Supplementary Table 2 for mouse PSCs adapted to control serum/LIF, +2i or +CDK8/19i; and Supplementary Table 3, for human PSCs adapted to control/primed, +2i or +CDK8/19i. Genes were ranked using the FDR q-value statistic to identify significant genes (q < 0.05), and then by fold change in expression. Venn diagrams and hypergeometric testing were performed to assess any significant overlaps. GSEA (GSEA_Pre-ranked) was performed with MsigDB Hallmarks, C5 GO terms, C2 Curated, KEGG, Reactome and NCI databases, using the standard settings, and with 1,000 permutations for Kolmogorov-Smirnoff correction for multiple testing. GSEA enrichment data were obtained and ranked according to FDR q value (significance threshold, q < 0.25). Heat maps of expression data were generated using GenePattern. Rank-rank hypergeometric overlap (RRHO) analysis was performed using the ranked list of log 2 -transformed fold changes in gene expression or RNA Pol II abundance using the standard settings 85 . The colour intensity of the RRHO heat map indicates the −log 10 -transformed P value after Benjamini-Yekutieli correction of the hypergeometric overlap (http://systems.crump.ucla.edu/rankrank/rankranksimple.php) 85 .
Analysis of repeat sequences and ERV expression was performed using Repbase datasets for rodent or human repeat elements and featureCounts. In Extended Data Fig. 3h, the total fragments per kb of transcript per million mapped reads for RNA expression of LINE L1 subtypes was calculated by grouping and summing by family, and was then arranged by evolutionary age 86 . A full list of three biological replicates for each viral subtype and the calculation for the summary of each viral LINE L1 family are provided in Supplementary Table 2.

Differential gene expression comparison of published mouse and human studies.
Gene expression changes have been comprehensively characterized in mouse, primate and human PSCs in response to overexpression of transcription factors after culture in various medium cocktails or in vivo during the development of the mouse or human embryos 87,88 ; a full list of datasets and references used here is provided in Supplementary Tables 2 and 3. We used the marker gene sets for each developmental stage to perform GSEA on the ranked list of genes that were up-or downregulated in the cellular studies of mice and humans. We also performed the analysis in reverse, comparing the gene sets of significantly differentially expressed mRNAs that were up-or downregulated in our cells versus the complete ranked list of differential gene expression in other studies. GSEA results are shown in Fig. 4e (mouse) and Fig. 4k (human). The readout is the NES. Data with P < 0.05 and q < 0.05 were considered to be significant and are indicated by asterisks in heat maps of GSEA NES scores.
ChIP-seq and genomic analyses. ChIP-qPCR was performed as described previously 84 ; a list of the primers and antibodies used is provided in Supplementary  Table 8. ChIP-seq was performed as described previously 2,3,68,84 . We performed six biological replicates for each condition (three conditions: serum/LIF, 2i and CDK8/19i) and for each antibody (three antibodies: anti-total RNA Pol II, anti-Ser 5P-RNA Pol II and control IgG). Three replicates were used for ChIP-qPCR validations, and the other three replicates were pooled for sequencing. Note that our RNA Pol II ChIP-seq data in this study for serum/LIF and 2i-naive cells very closely match previous ChIP-seq data involving the same comparison, that is, mouse PSCs in primed versus 2i-naive states 23,69 (compare Fig. 5h,i with Extended Data Fig. 6c).
Promoter and gene body regions were defined and RNA Pol II total and Ser 5P abundance along genes was calculated as described previously by Young and colleagues 68,84 (a schematic of which is provided in Fig. 5; Pol II abundance data are provided in Supplementary Table 6). RNA Pol II abundance was assessed by normalizing the total number of reads between treatments, and using featureCounts to calculate the background-subtracted log 2 -transformed RPKM of RNA Pol II abundance in the indicated regions. TSS and the transcription termination zone were identified using the Database of Transcriptional Start Sites (http://dbtss.hgc.jp). Metagenes were aligned to ±5 kb or ±2 kb around the TSS, and visualized using SeqMiner.
The promoter, gene body and transcription termination zone, as well as the ratios between these three regions for each gene (Fig. 5j, Extended Data Fig. 6d and Supplementary Table 6), were defined similar to that described previously 68,84 .  Total and Ser 5P RNA Pol II abundance were quantified at the promoter, gene  body and transcription termination zone for 31,167 RefSeq gene loci in which  the transcription start and stop sites are known (Supplementary Table 6), in four steps, similar to previous reports 68 : (1) the number of reads per nucleotide was computed using BEDTools genomecov; (2) to extend this number to the number of reads per gene promoter or gene body, BEDTools map was used; (3) to correct for region size, the RNA Pol II abundance was calculated as follows: ((number of reads in region/region size) × scaling factor) × 10 5 , where the scaling factor = (total number of reads in sample/genome length); (4) for the analysis of Pol II abundance according to inhibitor treatment, genes were first filtered for high-confidence Pol II detected at a threshold of >3,000 units at the promoter, and detected in all three conditions (serum/LIF, 2i or CDK8/19i), yielding 12,072 genes (Supplementary Table 6, for filtering and calculations). In Fig. 5h and Extended Data Fig. 6a,c, genes were arranged in rank by the abundance of RNA Pol II at the promoter region in the control serum/LIF condition.
CDK8/19 enrichment across the genome of wild-type mixed background V6.5 (C57BL/6-129) mouse PSCs was determined using a published dataset (GSE44286, GSM1082346) as previously described 2,3 ; peak calling was performed with MACS v.1.4.1 using the standard settings and compared to the input negative control. Note that the ChIP antibody for this ChIP-seq analysis (Santa Cruz, sc-1521) is reported to bind to both CDK8 and CDK19 (ref. 89 ). Peak annotation within local genomic features was performed using HOMER and the enhancer regions previously defined as constituent regions of TEs (n = 9,981) or SEs (n = 646) 2,3 , and of SE extended regions (n = 231) as defined previously 2,3 , where enhancers were defined by coenrichment for OCT4, SOX2, NANOG and MED1. Details about peak calls, CDK8/19 abundance at called peaks and loci annotations are provided in Supplementary Table 7. Naive-specific or primed-specific enhancer regions were defined by filtering the PREStige database of enhancers 77 , which identifies enhancers by enrichment of H3K4me1 monomethylation in multiple tissues and lineages. Using the PREStige dataset, we identified enhancer regions with H3K4me1 enrichment of >20 units, and that were specific to only preimplantation naive PSCs or postimplantation EpiSCs versus all other tissue-specific enhancer regions listed in the database (~120,000), by subtracting overlapping enhancers (1 bp overhang threshold) as outlined in Extended Data Fig. 8e. Source data are available online, including lists of naive ESC-specific enhancers (n = 1,424) and EpiSC-specific enhancers (n = 1,005). To identify the single nearest target gene to each PSC SE and analyse their biological functions, we performed an analysis using GREAT v.3.0.0 (ref. 90 ) with the standard settings, using the list of CDK8/19 peaks identified above (Supplementary Table 7). We used GREAT v.3.0.0 for GO analysis of target-gene functions, reporting the −log 10 -transformed binomial P value with conservative Bonferroni correction for multiple-hypothesis testing 90 . Correlation matrix of ChIP-seq data in Extended Data Fig. 7d was produced using Morpheus, which is available from the Broad Institute (https://software.broadinstitute.org/ morpheus/). For Fig. 7b,c, GSEA was run with a gene set of the single nearest genes to SEs (as identified in the GREAT analysis described above using the standard settings (GREAT v.3.0.0) 90 , using the SEs that were previously described in mouse PSCs 2,3 versus the ranked list of differential gene expression determined by RNA-seq for serum/LIF control compared with CDK8/19i-adapted mouse ES PSCs. Source data are available online, including lists of SE-target and expression-matched control genes.
For Extended Data Fig. 8h, GREAT analysis using the standard settings (GREAT v.3.0.0) 90 was used to identify the set of single nearest genes (n = 3,553 genes) to enhancer regions that were previously identified in mouse PSCs (n = 10,627) 2,3 . The log 2 -transformed fold change in RNA expression of these genes from this study was then ranked from high to low (serum/LIF versus 2i; serum/LIF versus CDK8/19i), and the extent of the overlap, calculated using a hypergeometric test of significance of these two ranked lists, is shown as a heat map in Extended Data Fig. 8h, performed using RRHO 85 with the standard settings (http://systems. crump.ucla.edu/rankrank/rankranksimple.php). The colour intensity of the RRHO heat map indicates the −log 10 -transformed P value after Benjamini-Yekutieli correction of the hypergeometric overlap.
Statistics and reproducibility. Unless otherwise specified, quantitative data are presented as mean ± s.d. and significance was assessed using two-tailed Student's t-tests; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. RRHO was performed as described previously 85 using the standard settings and after Benjamini-Yekutieli correction of the hypergeometric overlap. Except when annotated otherwise, each experiment shown was performed three times with similar outcomes. Statistical analyses are described in detail for each panel. No statistical methods were used to predetermine the sample size. In brief, for differential gene expression using RNA-seq analysis, a threshold of q < 0.05 or q < 0.01 was applied, as indicated in each case. For GSEA, the standard threshold for significance was applied (P < 0.05 and q < 0.25). Genes that were differentially expressed in the RNA-seq analysis were called using DESeq2 or Cufflinks v.1.3.0 (as described above). Immunofluorescence image analysis is described in detail above in the section about embryo analysis. Statistics were performed using MACS for peak calling of the ChIP-seq experiments. Statistical analyses of ChIP-qPCR, quantitative RT-qPCR and cell culture experiments was performed using Prism (v.7.03; GraphPad) or Microsoft Excel.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.