Common modes of ERK induction resolve into context-specific signalling via emergent networks and cell-type-specific transcriptional repression

Cell fate decisions during embryonic development are driven by a limited number of extracellular signals that operate at the precise time and place to activate distinct lineage-specific programmes. Given there is a finite number of signals but thousands of different cell types, it becomes necessary to repurpose these signals to produce cell-type-specific outcomes. Fibroblast Growth Factor (FGF) signalling via Extracellular signal-Regulated Kinase (ERK) is a prime example of this problem, exploiting the same set of kinases to produce diverse responses in multiple tissues during embryonic development (Corson et al., 2003; Dorey and Amaya, 2010; Ornitz and Itoh, 2015).

Early mammalian development is concerned with the segregation of the embryonic epiblast from the extra-embryonic lineages required for post-implantation development, the trophectoderm (TE) and the primitive endoderm (PrE); and FGF/ERK is required to make both. Initially, the outside cells are specified into TE and distinguished from the inner cell mass (ICM) (Lu et al., 2008; Nichols et al., 1998, 2009). FGF/ERK signalling then drives the specification of PrE as opposed to epiblast from the ICM at embryonic day (E)3.5 (Chazaud et al., 2006; Krawchuk et al., 2013; Yamanaka et al., 2010). Epiblast cells secreting FGF4 activate ERK in a neighbouring cell, inducing PrE identity (Azami et al., 2019; Chazaud et al., 2006; Frankenberg et al., 2011; Kang et al., 2017; Molotkov et al., 2017). As a result, culture of pre-segregation mouse embryos with either FGF4 or an FGF/ERK signalling inhibitor produces an ICM composed entirely of PrE or epiblast cells, respectively (Nichols et al., 2009; Schrode et al., 2014; Yamanaka et al., 2010).

Based on the capacity of ERK inhibition to block PrE differentiation, mouse embryonic stem cells (mESCs) can be efficiently derived from peri-implantation epiblast in the presence of an FGF/ERK inhibitor (Ying et al., 2008). This supports their expansion as cells with the competence to differentiate into all lineages, a property termed pluripotency. As the gene expression in these cells resembles the earliest stage of embryonic development, they are called naïve (Nichols and Smith, 2009). ERK induces exit from naïve pluripotency, suppressing the expression of pluripotency factors at the same time as activating differentiation by acting directly on enhancer activity (Hamilton et al., 2019). However, the activity of ERK rapidly changes at implantation. Post-implantation epiblast cells can also be expanded as pluripotent cultures but, in these cases, FGF/ERK is required for their expansion and to block somatic differentiation (Brons et al., 2007; Tesar et al., 2007). Their gene expression profiles approximate later stages of embryonic development than naïve cells, and therefore are termed primed (Endoh and Niwa, 2022; Morgani et al., 2017; Nichols and Smith, 2009).

FGF/ERK is not the only pathway to promote opposing outcomes in pre- versus post-implantation epiblast: at the naïve stage WNT/β-catenin activation (or GSK3 inhibition) blocks differentiation to promote pluripotency, whereas in primed epiblast it induces differentiation (Sumi et al., 2013; Ying et al., 2008). In vivo, the epiblast progresses out of the naïve state through two morphogenetic changes that have been exploited to identify the temporal hierarchy of WNT and FGF signalling: first, epiblast cells become polarized to produce rosette-like structures at E5.0, then a central lumen arises to form a cup-shaped epiblast at E5.5 (Bedzhov and Zernicka-Goetz, 2014; Shahbazi et al., 2017). Naïve mESCs and ex vivo cultured embryos can be captured in a rosette-like state [termed rosette stem cells (RSCs), analogous to the E5.0 epiblast] in the presence of FGF and WNT inhibition, and then advanced further into epiblast development upon FGF/ERK release to generate epiblast-like cells (EpiLCs) (Endoh and Niwa, 2022; Hayashi et al., 2011; Neagu et al., 2020). The existence of these defined culture models that mimic the progression through naïve into primed epiblast development provides an ideal platform to investigate the molecular cues that regulate FGF/ERK signalling context.

In this paper, we developed mouse and cell culture models for cell-type-specific ERK activation and used them to map signalling responses in different pluripotent states. We found that inductive events are conserved across cell types, whereas repression is cell-type specific. At a chromatin level similar trends are apparent, but induction in naïve pluripotency involves chromatin remodelling, whereas in primed pluripotency it exploits existing accessible regions.

To investigate changes in ERK response in pre- and post-implantation epiblast, we designed a strategy for ERK induction in vivo and in vitro. We took advantage of a genetic model (Hamilton and Brickman, 2014; Hamilton et al., 2019) which employs a constitutively active c-RAF (Raf1) to activate the kinase cascade upstream of ERK (Fig. 1A). The engineered c-RAF kinase fragment is fused to a tamoxifen (4OHT)-specific ligand binding domain of the oestrogen receptor (ER^T2) that facilitates folding and stabilises the RAF fragment in the cytoplasm (Samuels et al., 1993), allowing for precise temporal activation with 4OHT (Fig. 1A). To enable in vivo lineage-specific Cre-mediated ERK activation, we introduced a lox-stop-lox cassette upstream of the inducible c-RAF. The inducible c-RAF was tagged with HA and followed by a T2A self-cleaving peptide and a H2B-mCherry reporter (Fig. 1B) to visualise transgene expression by either H2B-mCherry fluorescence or HA immunodetection. We targeted mESCs and verified ERK induction and mCherry expression (Fig. S1A-E) before injecting into morula-stage embryos to produce chimeric mice, which were then crossed to obtain a stable homozygous line, CRAFR26 (Conditional RAF in Rosa26 locus).

To validate the system, we crossed homozygous CRAFR26 females with heterozygous Sox2Cre males (Hayashi et al., 2002), expressing Cre in the epiblast lineage. We observed mCherry signal restricted specifically to the epiblast of E6.5 embryos (Fig. 1C) and induced the c-RAF kinase with 4OHT for 2 or 4 h. The 4OHT-treated embryos produced higher pERK expression in the epiblast compared with non-treated embryos (vehicle) (Fig. 1D; Fig. S1F-H) or non-floxed embryos (Sox2Cre^+/+) (Fig. S1F-H). CRAFR26 mESCs were also transiently transfected with Cre, generating a cell line that constitutively expressed cRAF-HA-T2A-H2B-mCherry from the Rosa26 locus (Fig. 1E). The cells show homogeneous induction of pERK after 2 or 4 h of 4OHT treatment, rapidly reversed by a MEK inhibitor (MEKi) (Fig. 1F). Cells were cultured in the presence of FGF receptor inhibitor (FGFRi) for 48 h before and during the induction, to block exogenous signalling. SOX2 expression during ERK induction confirmed that cells maintained ESC identity (Fig. 1F).

To map the transcriptional response to ERK in in vitro pluripotent cell types around implantation, we differentiated targeted ESCs into RSCs, which are analogous to the E5.0 epiblast (Neagu et al., 2020), and EpiLCs, related to the E5.5 epiblast (Hayashi et al., 2011) (Fig. 2A). We analysed the transcriptome of these cells by bulk RNA-sequencing (RNA-seq) before and after ERK stimulation and reversion with MEKi (Fig. 2A). Principal component analysis (PCA) projection of the top 500 most variable genes shows that developmental time in the progression from ESCs to RSCs and eventually EpiLCs is the main source of variation in the dataset (Fig. 2B). Naïve ESC populations express naïve pluripotency markers, whereas EpiLC populations express primed pluripotency markers (Fig. 2C). RSCs, as an intermediate population between ESCs and EpiLCs (Fig. 2B), express both naïve and primed markers (Fig. 2C). We confirmed that genes induced by FGF4 stimulation in ESCs (Fernkorn and Schröter, 2024 preprint) were also being upregulated in our samples (Fig. S2A). Genes annotated as MAPK (GO:0000165) were upregulated in all cell types (Fig. 2D), to similar levels (Fig. 2E).

As expected, ERK repressed pluripotency genes in naïve ESC (Hamilton et al., 2019), but in EpiLCs it induced them (Fig. 2F), consistent with the role of FGF in supporting primed pluripotency (Greber et al., 2010; Tesar et al., 2007). This included genes normally associated with naïve pluripotency, such as Klf2, Esrrb, Nanog and Prdm14 (Fig. 2F; Fig. S2B). RSCs, although an intermediate population, are cultured in ERK inhibition, and their transcriptional ERK response was similar to ESCs. We confirmed these results with NANOG immunostaining before and after 8 h of ERK induction in each cell type, showing that ERK repressed NANOG also at the protein level in ESC and RSC but induced it in EpiLCs (Fig. S2C,D).

To determine the extent to which ERK regulates common sets of genes in different contexts, we quantified the number of common genes upregulated [defined as log2 fold change (LFC)>1.5 and adjusted P-value (Padj)<0.05] or downregulated (LFC<1.5 and Padj<0.05) by ERK (Table S1). At all the evaluated time points of ERK induction, we observed a higher shared number of genes in the activated responses than in the repressed ones (Fig. 3A,B; Table S1). Reverted genes (upregulated with MEKi) are also cell-type specific, similar to downregulated genes at 4 h (Fig. 3C).

We observed the same effect when we clustered the differentially expressed genes according to the dynamics of activation and repression (Fig. S3A-E; Table S2). The trend is similar in all three cell types, with 70% of clusters activated by ERK in ESCs exhibiting a positive fold change in EpiLCs, and the mean above zero (Fig. S3F). This contrasts with ERK-repressed genes, where only 50% of genes in clusters repressed in ESCs show negative fold changes in EpiLCs and the mean of the fold change is zero (Fig. S3G). For both repression and activation, the response in RSCs and naïve ESCs is far more similar than that observed in EpiLCs.

To understand the functional role of ERK-activated genes in each context, we extracted genes lists annotated as a specific GO Term and compared their fold changes upon 4 h of ERK activation in each cell type. We found that genes annotated as endoderm related (GO:000792) were induced by ERK in both naïve and primed (Fig. 3D), consistent with the ability of these cell types to make primitive and definitive endoderm (Anderson et al., 2017). However, genes related to the neural lineage (GO:0061351) and stem cell maintenance (GO:0019827) were better induced by ERK in primed cells (Fig. 3D). We found genes related to placenta development repressed in RSCs and related to angiogenesis in EpiLCs (Fig. S3H). These data support the idea that ERK activates genes related to lineage progression but it also has an important role in suppressing non lineage-specific gene expression.

To determine the in vivo ERK response in naïve versus primed stages, we sequenced the transcriptome from pre- and post-implantation epiblast before and after ERK activation (Fig. 4A; Table S3). PCA of all uninduced samples suggested that major alterations in gene expression were due to in vitro adaptation to stem cell culture, but also illustrated a clear separation of the naïve (ESCs and pre-implantation epiblast) from the primed (EpiLCs and post-implantation epiblast) states (Fig. 4B). To investigate the differences driving the separation of in vivo and in vitro states, we extracted the most significant genes driving variation across PC1, but found them mostly related to biosynthesis and metabolism as opposed to developmental processes (Fig. S4A; Table S4). Moreover, E3.5 epiblast cells expressed naïve pluripotency markers (Fig. 4C), similar to ESCs (Fig. 2C), whereas E5.5 epiblast cells expressed primed pluripotency markers (Fig. 4C), similar to EpiLCs (Fig. 2C).

We then investigated which genes were upregulated in vivo with ERK induction (Table S3). We found fewer differentially expressed genes with ERK than in vitro. In E3.5 and E5.5, the genes upregulated with ERK showed enrichment in GO Terms related to metabolism (Fig. S4B,C; Table S4), which we did not observe in vitro. This suggests a generic link between MAPK signalling and metabolism in vivo, already reported by others (Huddleston, 2011; Papa et al., 2019). Some differentially expressed genes at E3.5 showed the same trend in ESCs (24 out of 235 genes), although the naïve ESC signalling response appears to reflect in vivo signalling less clearly than that observed in EpiLCs when compared with E5.5 (18 out of 43 genes; Fig. 4D). Although the lack of similarity between signalling in naïve ESCs and E3.5 embryos was initially surprising, we believe this is because a large component of the ERK network was already expressed at the time in which we induced signalling in ex vivo culture (Fig. S4D).

We also observed that the E3.5 epiblast upregulated 58% of endoderm-related genes upon ERK induction (Fig. 4E) and that the E5.5 epiblast showed 54% of stem cell-annotated genes being upregulated upon ERK induction (Fig. 4E), which recapitulated the effect observed in vitro (Fig. 3D).

Although we saw ERK inducing a global endoderm signature in E3.5 epiblast (Fig. 4E), we saw more robust Sox17 induction compared with Gata4 or Gata6 after 4 h of ERK (Fig. 5A,B; Fig. S5A-C). Similar to the vast majority of naïve ESC ERK targets, the lack of Gata6 induction at an mRNA level is likely because Gata6 is already robustly expressed before 4OHT addition, but this did not explain the lack of a direct impact on Gata4 RNA levels (Fig. 5B; Fig. S5A-C). We therefore sought to validate the capacity of intracellular ERK induction to functionally promote GATA6⁺ PrE in the blastocyst. We isolated E3.5 CRAFR26 embryos and treated them with either 4OHT or FGF4 for 4 h. Embryos were left to develop until late blastocyst stage (E4.5) in the presence of an FGFRi to prevent paracrine signalling (Fig. 5C). In ERK-induced blastocysts, we observed that 96% of the ICM cells expressed the PrE marker GATA6 at E4.5 (Fig. 5D,E), and this induction was more efficient than the same length of treatment (4 h) with FGF4 (Fig. 5D,E). Although there is some persistence of ERK activity following 4OHT removal (Fig. S5D), it is consistent with the reported activity of ERK on GATA6 protein levels (Bessonnard et al., 2014; Kelly et al., 2014; Meng et al., 2018). These surprising observations suggest that Sox17 is induced independently by ERK, rather than downstream of GATA4 and GATA6. Whereas GATA6 is believed to be an early marker of mouse PrE at a protein level (Artus et al., 2011), the induction of Sox17 mRNA may be an earlier response to ERK than GATA6 protein. Although 4 h of ERK induction induced Sox17 mRNA, it was not enough to induce upregulation of the SOX17 protein in E3.5 blastocysts (Fig. 5F-H). We observed a slight increase in SOX17⁺ cells in the ICM after 4 h of ERK (Fig. 5H), but this was not significant.

To determine whether chromatin context established signalling competence or was rearranged as a consequence, we performed Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq) (Buenrostro et al., 2013; Grandi et al., 2022) (Fig. 6A), and assessed which regions were differentially accessible between cellular states (Table S5) and upon ERK induction (Table S6).

We confirmed whether the changes observed by ATAC-seq were detecting putative regulatory regions by interrogating binding of cofactors with ERK using our existing dataset in ESCs (Hamilton et al., 2019). We found that opening regions gained cofactor occupancy [EP300 and Mediator (Med24)] (Fig. S6A), whereas less accessible regions in response to ERK were losing EP300 and Med24 (Fig. S6A). Super enhancers, large cooperative DNA elements known to regulate transcription in pluripotent cells (Hnisz et al., 2013), were not remodelled upon ERK activation (Fig. S6B), consistent with our previous observations (Hamilton et al., 2019). We also assessed changes in the histone modification H3K27ac, generally associated with activated transcription, using CUT&RUN and our published ChIP-seq dataset for H3K27ac in ESCs (Hamilton et al., 2019; GSE132444). In all cell types, opening regions also gained H3K27ac with ERK (Fig. S6C). In naïve ESCs we observed some decrease in H3K27ac in closing regions, whereas in RSCs and EpiLCs we observed little change in H3K27ac (Fig. S6C). This suggests that in naïve pluripotency, signalling-mediated alteration in accessibility involves changes in histone acetylation, whereas in primed pluripotency there appears to be no association of histone deacetylation with repression.

To validate cell-type-specific changes in our dataset, we analysed the ATAC-seq signal at ESC- and EpiLC-specific enhancers (Buecker et al., 2014; GSE56138). We found that ESC-specific enhancers were accessible in ESCs and closed down progressively in RSCs and EpiLCs, whereas EpiLC enhancers were more accessible in EpiLCs than in ESCs (Fig. S6D). Consistently, enhancer sets exhibited intermediate levels of accessibility in RSCs (Fig. S6D), although RSCs expressed mostly naïve genes (Fig. 2C). Finally, we assessed changes at the set of ERK-activated and repressed enhancers we previously defined based on EP300 and H3K27ac changes (Hamilton et al., 2019). We found that, although these enhancers were defined in naïve ESCs, ERK-activated enhancers appeared to become more accessible, with signalling in all three cell types (Fig. 6B). In contrast, although ERK-repressed enhancers became less accessible in naïve ESCs, this phenomenon was negligible in RSCs and EpiLCs (Fig. 6B), further supporting the notion that cell-type-specific responses depend on repression.

To assess the uptake of different regulatory regions, we compared the sets of peaks that were opening and closing with ERK induction in each cell type. We observed considerable overlap in regions that were opening in response to ERK, with 307 shared regulatory regions between all three cell types (Fig. 6C), with RSCs and EpiLCs sharing more regulatory regions than transcriptional responses (Fig. 6C versus Fig. 3A). Consistent with gene expression, ERK closed regions appeared to be almost mutually exclusive (Fig. 6C). Similar overlap was also found at regions that gained or lost H3K27ac in each cell type (Fig. 6D). H3K27ac followed the ATAC-seq signal in individual loci, as depicted in the Prdm14 enhancer (Fig. S6E).

ERK-responsive functional commonalities were determined based on enriched motifs in opening regions and we found that most motifs were shared between the three cell states and were related to canonical ERK factors (ETS–AP-1 motifs, i.e. JUN/FOS, ERG and ETV) (Fig. 6E; Table S7). Although RSCs and ESCs express similar transcription factors (TFs) (Fig. 2B,C), we observed strong clustering of certain immediate early TFs, mainly ETV and JUN/FOS, that are common to RSC and EpiLC (Fig. 6E), suggesting that downstream response to signalling in these two cell types could illicit different phenotypes in differentiation. Motifs in the closing regions were more enriched in specific contexts (Fig. 6F; Table S7). We found motifs related to naïve pluripotency being closed in ESCs and RSCs (TCF, OCT motifs), and motifs related to later development stages being decommissioned in EpiLCs (FOX, SOX motifs) (Fig. 6F; Table S7). To further confirm that repression appears to be cell-type specific, whereas activation seems relatively ubiquitous, we determined the genes nearest to opening and closing regulatory regions. We found a large number of opening regions mapped to the same genes in different cell types, but a high degree of specificity in negative regulation (Fig. S7A).

To distinguish between remodelling and TF occupancy, we assessed the capacity of TFs to block the activity of Tn5 transposase and create a footprint in the ATAC-seq signal. We exploited the computational tool TOBIAS (Bentsen et al., 2020) to extract TF footprints from ATAC-seq data and predict changes in TF occupancy upon ERK induction, obtaining a differential binding score. We found that the ERK-canonical TFs (ETS–AP-1) showed similar increased TF-binding scores upon ERK in all cells (Fig. 6G; Table S8). Consistent with this, we observed 53% of footprints to be shared in all cell types in the 307 commonly opening regions, with a core set of ETS–AP-1 TFs in 38% of these in all cell types (Fig. S7B; Table S8).

When we look for global TF binding release in response to ERK, we see little change in all three cell types (Fig. S7C). Notably, the small number of TFs that are significantly unbound show some cell-type specificity, with nuclear hormone receptors being unbound in ESCs and FOX/SOX TFs losing binding in EpiLCs (Fig. 6H). However, consistent with our previous observations (Hamilton et al., 2019), the great majority of TFs remain associated with binding sites in cis-regulatory elements following ERK-mediated repression (Fig. S7C). In summary, although ERK-mediated activation appears to be triggered by canonical ETS–AP-1 TFs, cell-type-specific repression appears to be mediated by decommissioning in the absence of robust reduction to lineage-specific TF binding.

By combining the RNA-seq with the ATAC-seq, we could distinguish between two strategies: either ERK is regulating genes in the proximity of already accessible regions, or factors downstream of ERK are directly influencing accessibility to produce changes in gene expression.

To distinguish accessibility before signalling, we extracted regions that were accessible in each cell type before ERK induction (at 0 h) and removed the regions that were being remodelled by ERK. We found that a significant number of these regions were specifically accessible in one cell type but not the others (Fig. S8A,B; Table S5), and these regions were enriched for different motifs (Fig. S8C; Table S7). When we analysed the LFC with ERK induction of the genes closest to these peaks, we found that in EpiLCs the genes found next to already accessible regions were strongly upregulated, whereas in ESCs and RSCs these regions appeared to be less associated with signalling-dependent transcription (Fig. 7A; Table S9). This suggested that signalling acts on accessible regulatory sequences in EpiLCs, as exemplified in the Spred3 enhancer, progressively accessible in RSCs and EpiLCs, while its transcription is activated by ERK solely in EpiLCs (Fig. 7B). Regions accessible in RSCs also show higher induction in transcription compared with ESCs but not as pronounced as in EpiLCs (Fig. 7A), suggesting that RSCs are an intermediate population in which some ERK-responsible regions are already available.

The observation that available regions do not appear to be exploited by ERK in naïve ESCs suggested a requirement for remodelling in naïve ESCs. We therefore investigated changes in gene expression next to remodelled regions and found a strong correlation between changes in accessibility and transcription at the nearest gene in naïve ESCs, but not in the other two cell types (Fig. 7C; Table S9). This was measured as a significant difference in transcriptional change from nearest genes to opening versus closing regions (Fig. 7C). This is exemplified by the Zic3 enhancer: its transcript was upregulated at a comparable fold change in all three cell types, while its enhancer was opening upon ERK induction in naïve cells but did not correlate with transcript levels in primed cells (Fig. 7D). Taken together, our observations suggest that the chromatin landscape already present before the signal arrives can produce a significant impact in the signalling output, given that ERK exploits a preconfigured set of targets in EpiLCs, whereas in naïve ESCs it drives the rearrangement of its target loci.

How the same signal can produce different effects depending on the cellular or tissue context is a long-standing question in developmental biology. Although FGF/ERK is a fundamental mediator of diverse processes (Dorey and Amaya, 2010; Turner and Grose, 2010) in numerous tissues across development (Corson et al., 2003), it is difficult to fathom how it achieves this diversity of regulation through a set of conserved immediate early TFs. To investigate the context-dependent responses to ERK, we have developed a tool that works in vivo and in vitro and allows for precise, homogeneous and controlled activation of ERK and found that context-dependent transcription appears to depend largely on the capacity of this pathway to suppress cell-type-specific transcription.

One of the advantages of our model is that it overcomes heterogeneity and feedback inhibition. Historically this has complicated the assessment of responses to signalling, as the immediate early responses involve rapid inhibition by phosphatases. This approach might seem somewhat artificial, but long pulses of active ERK have been observed in the embryo, more specifically in the PrE precursor cells in the ICM (Pokrass et al., 2020; Simon et al., 2020). How our system compares with other models for FGF/ERK induction, such as those that rely on optogenetics (Arekatla et al., 2023; Toettcher et al., 2013), remains to be elucidated. Our previous findings suggest that the range of enhancer regulation and phosphorylation induced in naïve ESCs is the same as that induced by FGF4 in Fgf4 mutant cells, but that ERK phosphorylation occurs synchronously throughout the culture. This is likely the reason that 4 h of ERK induction via 4OHT converts the ICM to entirely GATA6⁺ PrE over a subsequent 24-h period, similar to that achieved when FGF4 is present for the full 24 h. What does this suggest about the network involved in epiblast-endoderm segregation, that has always been modelled around GATA6-FGF-NANOG? Perhaps that the induction of GATA6 may not be a direct transcriptional response to FGF/ERK, but rather its levels are stabilised downstream of the signal. We have also observed that Gata6 mRNA levels are already high before ERK induction (Fig. 5B), suggesting that post-transcriptional regulation may be coupled to an ERK-induced cascade before E3.5.

In naïve ESCs, the regions losing accessibility upon ERK induction are characterized by the presence of TCF motifs (Fig. 6F). ESCs and RSCs are cultured in nearly identical conditions, with the only difference being that canonical WNT signalling is activated in ESCs and inhibited in RSCs (see Materials and Methods). In naïve ESCs, canonical β-catenin signalling exploits TCF3 to support pluripotency by associating with these regions and counteracting TCF3-mediated repression (Athanasouli et al., 2023; Watanabe and Dai, 2011). Yet canonical WNT signalling also induces PrE differentiation in ESCs and can do so in virtually identical media with one key difference, the removal of the FGF block and reduced insulin (Anderson et al., 2017). How does the same pathway support both activities? In response to ERK activation in naïve ESCs, we found decommissioned regulatory regions enriched in TCF binding, suggesting an FGF-dependent alteration in WNT activity preventing it from activating the TCF-dependent pluripotency network. As a result, WNT activation now triggers endoderm differentiation based on an ERK-dependent set of TCF targets.

In the epiblast lineage, where WNT is inhibited in RSCs, repression by ERK once again targets naïve ESC enhancers, but this time in regions enriched in POU/OCT motifs such that signalling promotes the transition into primed pluripotency (Fig. 6F). These regions could reflect the acquisition of bivalency at pluripotency enhancers in RSCs, enabling them to both move forward developmentally, while retaining the capacity to revert to naïve ESCs (Neagu et al., 2020). Finally, in EpiLCs, in which FGF/ERK supports self-renewal, we identify the FOX factors prominent in gastrulation stage differentiation.

Historically, it has been difficult to reconcile the role of FGF/ERK in neural differentiation with so-called default neural induction. We previously observed that anterior neural identity can be achieved in the presence of ERK inhibition (Hamilton and Brickman, 2014), and similar results were obtained when BMP receptor mutants were cultured in FGF receptor block (Di-Gregorio et al., 2007). However, this contradicts experiments in the chick embryo (Linker and Stern, 2004; Lunn et al., 2007) and ESC differentiation (Kunath et al., 2007; Stavridis et al., 2007), where it was argued that the naïve-to-primed transition requires ERK. Our findings provide a means to reconcile these observations, suggesting that context-dependent repression by ERK drives endoderm differentiation in the presence of WNT signalling, but epiblast differentiation in its absence. The presence of POU motifs in ERK-repressed RSC enhancers could also suggest that FGF/ERK represses the overlapping pluripotent/neural programmes regulated by POU and SOX proteins, such that signalling supports the maintenance of epiblast identity over its differentiation towards anterior neural.

At a regulatory level, the notion that cell-type-specific response is largely about repression has a certain logic. Immediate early genes are widely studied and likely to be context independent, whereas repression depends on the regulatory programme transcribed in the target cell type. We have previously shown that MED24 phosphorylation plays an important role in ERK transcriptional regulation in ESCs (Hamilton et al., 2019). MED23 phosphorylation has also been shown to be an important coactivator for MAPK transcription (Balamotis et al., 2009; Stevens et al., 2002) and ERK phosphorylation of MED1 results in enhanced association with MED7 and transcriptional activation (Belakavadi et al., 2008; Pandey et al., 2005). As a result, we favour a model in which phosphorylation of different mediator subunits and associated cofactors triggers disengagement of the RNAPII complex from key regulatory sequences, leading to the repression of transcription.

Although this work focuses on FGF/ERK, the question of context is universal to all signalling pathways. Here, we show that these ubiquitous factors can act identically in different cell types, but their capacity to induce transcription is not always correlated with alterations in chromatin structure. We found that the cis-regulatory strategy by which signalling generates a transcriptional output requires remodelling in naïve ESCs but not in EpiLCs, suggesting that the same signal can employ different strategies depending on the context. This signal adaptability to the cellular context and chromatin has also been observed in other settings during development (Delás et al., 2023), but the means by which these pathways can effect changes in remodelling in one cell type, but not require it in another, remains unknown. However, it does suggest that the canonical view of signalling specificity being governed by chromatin context and accessibility is not correct, as here we observe the same immediate early response, but in one context this requires remodelling and in another it does not, suggesting that ERK induction can overcome restricted chromatin accessibility.

Overall, our data provide a framework for signal-dependent context specificity where both transcriptional and chromatin changes in response to ERK are more universal in activation than repression. As we found that context-dependent repression is mediated mainly by deactivation of enhancers, leaving cell-specific TFs bound to their targets, this response is, by definition, context-dependent. Activation of ERK-responsive genes relies largely on commonly activated genes triggered as immediate early response to ETS–AP-1 canonical factors, but this can be propagated through existing cell-specific states. Whether activation exploits chromatin remodelling or not would depend on the existence of prebound core TFs, which could render cells plastic to respond to signalling and differentiate in specific contexts (Hamilton et al., 2019; Knudsen et al., 2023; Linneberg-Agerholm et al., 2024; Redó-Riveiro et al., 2024).

The CRAFR26 construct was engineered from two previous versions of the BXB plasmid (Hamilton and Brickman, 2014; Hamilton et al., 2019) that were digested and ligated to obtain a BXB-T2A-mCherry plasmid. Then, this fragment was introduced downstream of a lox-stop-lox cassette using the BigT plasmid (Srinivas et al., 2001) and then between the Rosa26 locus homology arms using the R26 plasmid (Srinivas et al., 2001).

The CRAFR26 construct was transfected into mESCs using a published gRNA directed to the Rosa26 locus (Gu et al., 2018) and the Cas9 plasmid PX458 (Addgene plasmid #48138). Clonal colonies were initially screened by PCR, and then further validated by Southern blot (see probe sequence in Table S10), Sanger sequencing and in house karyotyping. For in vitro ERK induction experiments, the cells were lipofected with a pCAG-CRE plasmid to remove the STOP codon. The cells were then validated by flow cytometry, western blot and immunostaining (see Fig. S1A-E).

To generate the CRAFR26 mouse line, correctly targeted CRAFR26 mESCs (Agouti background) were injected into C57BL/6 mouse embryos and chimera contribution was assessed by fur colour. Male chimeras were crossed with C57BL/6 females to obtain heterozygous F1s. Pups were genotyped from ear snips by PCR. Details of primers and genotyping PCR are in Table S12.

Mouse (Mus musculus) lines were maintained under 12 h light/12 h dark cycles in designated facilities at the University of Copenhagen. Natural mating was set up in the evening, and females were checked for plugs the following morning, which was established as E0.5. Mouse heterozygous Sox2Cre (Hayashi et al., 2002) females were mated with homozygous CRAFR26 males to produce constitutive CRAFR26 expression in pre-implantation embryos, and Sox2Cre males were mated with homozygous CRAFR26 females to produce expression of CRAFR26 in the epiblast lineage in post-implantation embryos (Hayashi et al., 2003). Sox2Cre females were superovulated to obtain a larger number of embryos per animal. Superovulation was performed by intraperitoneal injection of 5 IU pregnant mare serum gonadotropin (PMSG; Sigma-Aldrich) followed by 5 IU human chorionic gonadotrophin (hCG; Chorulon, Intervet) 42-50 h later. The females were mated immediately following hCG injection. Animal work was authorized by the Danish National Animal Experiments Inspectorate under project license 2018-15-0201-01520 and 2023-15-0201-01513, performed in accordance with Danish national guidelines and European legislation.

CRAFR26 mESCs were routinely cultured on 0.1% gelatine-coated flasks in 2i/LIF media (Ying et al., 2008) consisting of N2B27 media supplemented with 3 μM of the GSK3 inhibitor Chir (Axon, CHIR99021), 1 μM of the MEKi PD03 (PD0325901, Sigma-Aldrich, PZ0162) and 1000 U/ml of LIF (prepared in house). N2B27 media was prepared as previously described (Knudsen et al., 2023). For passaging, confluent flasks were washed with PBS, incubated for 3 min in accutase (Sigma-Aldrich, A6964) and tapped to release the cells from the flask. Accutase was inactivated by adding medium and the cells were centrifuged (500 g for 3 min), resuspended and replated onto new gelatine-coated flasks. All mESCs lines used in this study were routinely karyotyped and tested for mycoplasma.

For differentiation into RSCs, mESCs were plated into N2B27 media supplemented with 2 µM IWP2 (Millipore, 681671), 1 µM of the MEKi PD03 (PD0325901, Sigma-Aldrich, PZ0162) and 1000 U/ml of LIF (prepared in house). After 3 days cells were passaged as RSCs (Neagu et al., 2020).

For EpiLC differentiation, RSCs were plated in clumps at high confluency (50,000 cells/cm²) on Fibronectin-coated plates (Millipore, FC010), into N2B27 media supplemented with 12 ng/ml bFGF (Peprotech, 450-33-100µG), 20 ng/ml Activin A (Peprotech, 120-14E-200µG) and 1 µM XAV939 (Sigma-Aldrich, X3004). Medium was changed after 24 h and EpiLCs were collected after 48 h.

For ERK induction in vitro, mESCs and RSCs were cultured for 48 h before the experiment in 250 nM of the FGFRi PD17 (PD173074, Sigma-Aldrich, P2499) instead of PD03. For EpiLCs, PD17 incubation was reduced to 2 h before induction. Then, the c-RAF kinase construct was induced by adding 250 nM 4OHT (4-hydroxytamoxifen, Sigma-Aldrich, H7904) to the media.

Cells were cultured for immunostaining in eight-well slides (Ibidi, 80826) and then stained and imaged on the plate. Cells were fixed for 10 minutes with 4% formaldehyde (Thermo Fisher Scientific, PI-28906) at room temperature, then washed three times with PBS, then treated with ice-cold methanol for 10 min at −20°C and washed again three times with PBS. Cells were blocked for 2 h at room temperature in 10% donkey serum (Sigma-Aldrich, D9663), 0.3% Triton X-100 (Sigma-Aldrich, T8787) in PBS. Primary antibodies (see Table S11) were incubated overnight at 4°C in 1% bovine serum albumin (BSA; Sigma-Aldrich, A2153), then washed three times. Secondary antibodies were incubated for 2 h at room temperature also in 1% BSA, and further washed three times. DAPI (Molecular Probes, D1306, 1 μg/ml) was then added for 5 min and washed twice with PBS. Cells were imaged using a Zeiss 780 confocal microscope.

For quantification of immunostainings, we used Fiji (ImageJ) (Schindelin et al., 2012). We created a binary mask using the signal from the DAPI staining and applied the watershed algorithm to delimit individual nuclei. This mask was used to measure intensity in all other channels, and the median intensity for each nuclei was extracted. To ensure enough sampling, we imaged five z-stacks separated by 0.5 µm and performed maximum intensity projection before quantification.

Timed pregnant females were euthanised by cervical dislocation. E2.5 embryos were flushed from the oviducts using M2 medium (Sigma-Aldrich, M7167), and E5.5-E6.5 embryos were dissected from the uterus. For Sox2Cre; CRAFR26 embryos, they were briefly imaged in a Deltavision Widefield Screening microscope to confirm their genotype by mCherry expression.

For longer culture of E2.5 embryos the zona pellucida was removed by brief incubation in acidic Tyrode's solution (Sigma-Aldrich, T1788) and then the embryos were incubated in KSOM drops (Millipore, MR-101-D) covered in mineral oil (NidOil, Nidacon, NO-400K) at 37°C. To induce ERK, 4OHT was added at 250 nM for the times indicated. FGFRi (PD173074, Sigma-Aldrich, P2499) was added at 250 nM. FGF4 supplemented with heparin was added at 500 ng/ml. IVC2 media was used for short term ex vivo culture of E5.5-E6.5 embryos (Bedzhov et al., 2014).

E4.5 embryos were fixed in 4% formaldehyde (Thermo Fisher Scientific, PI-28906) at room temperature for 15 min, then washed twice in PBS. Embryos were permeabilized for 15 min in 0.25% Triton X-100 (Sigma-Aldrich, T8787) in PBS/PVP [3 mg/ml polyvinylpyrrolidone (PVP; Sigma-Aldrich, PVP40) in PBS]. Blocking was performed for 1 h in 2% donkey serum (Sigma-Aldrich, D9663), 0.1% BSA (Sigma-Aldrich, A2153), 0.01% Tween20 (Sigma-Aldrich, P1379) in PBS/PVP. Primary antibodies (see Table S11) were added to blocking buffer and incubated overnight at 4°C. Embryos were washed three times for 10 min in blocking buffer, and then secondary antibodies were added to blocking buffer and incubated for 2 h. Secondary antibodies were washed three times for 10 min in blocking buffer. DAPI (Molecular Probes, D1306, 1 μg/ml) was added to blocking buffer and incubated for 15 min at room temperature before washing once more in blocking buffer and mounting for imaging in PBS/PVP in 18-well slides (Ibidi, 81816). Imaging was performed using a Leica Stellaris confocal microscope.

E6.5 embryos were fixed in 4% formaldehyde (Thermo Fisher Scientific, PI-28906) at room temperature for 30 min, and then washed three times for 10 min in PBS. Embryos were permeabilized for 30 min in 0.5% Triton X-100 (Sigma-Aldrich, T8787) in PBS. Blocking was carried out overnight in 2% donkey serum (Sigma-Aldrich, D9663), 0.1% BSA (Sigma-Aldrich, A2153) and 0.1% Tween20 (Sigma-Aldrich, P1379) in PBS. Primary antibodies (see Table S11) were added to blocking buffer and incubated for 48 h at 4°C. Embryos were washed three times for 15 min in blocking buffer and then washed overnight. Secondary antibodies were added to blocking buffer and incubated overnight. Secondary antibodies were washed three times for 15 min and washed overnight in blocking buffer. DAPI (Molecular Probes, D1306, 1 μg/ml) was added to blocking buffer and incubated for 2 h at room temperature before washing three times for 15 min in blocking buffer and mounting for imaging in PBS in eight-well slides (Ibidi, 80826). All overnight incubations were carried out at 4°C with gentle shaking in the dark. Imaging was performed using a Leica Stellaris confocal microscope.

For quantification of immunofluorescence, we used Fiji (ImageJ) (Schindelin et al., 2012). A region of interest was manually delimited. Then, the intensity of the pixels in that area was measured for each channel, and we took the median intensity as the intensity measure for each embryo.

RNA was extracted using the RNeasy Mini Kit (Qiagen, 74106) following the manufacturer's instructions. Four biological replicates were collected per sample (ESCs versus RSCs versus EpiLCs) and per condition (0 h versus 2 h versus 4 h versus rev), defined as two different passages of two independent clones. RNA was quantified by Nanodrop and 1 µg of RNA per sample was treated for ribosomal RNA depletion using the NEBNext rRNA Depletion Kit (New England Biolabs, E6350) and then immediately followed by Library prep using NEBNext Ultra II kit (New England Biolabs, E7770). Libraries were pooled and sequenced using an H75 kit from Illumina in a NextSeq500 sequencer following the manufacturer's instructions.

To obtain an ICM with only epiblast cells, E2.5 Sox2Cre; CRAFR26 embryos were cultured in KSOM supplemented with 250 nM PD17 (PD173074, Sigma-Aldrich, P2499) for 24 h. Then, to induce ERK, the media was supplemented with either 250 nM 4OHT for ERK induction, or ethanol at the same dilution for control conditions.

To collect the ICM, the trophectoderm layer was removed by blastocyst immunosurgery as previously described (Linneberg-Agerholm et al., 2024; Solter and Knowles, 1975). Four biological replicates were collected per condition (treatment versus control), each containing a group of ten ICMs. The samples were placed into a 1.5 ml tube with lysis buffer from the Arcturus™ PicoPure™ RNA Isolation Kit (Applied Biosystems, KIT0204) and followed with RNA extraction according to the manufacturer's instructions. RNA was quantified using a Qubit™ RNA High Sensitivity (HS) kit (Invitrogen, Q32852). Libraries were prepared from 2 ng of RNA per sample using the NEBNext^® Single Cell/Low Input RNA Library Prep Kit for Illumina (New England Biolabs, E6420). Libraries were pooled and sequenced using a P3-100 kit from Illumina in a NextSeq2000 sequencer following the manufacturer's instructions.

To induce ERK in E5.5 Sox2Cre; CRAFR26 embryos, they were cultured in IVC2 media (Bedzhov et al., 2014) supplemented with either 250 nM 4OHT for ERK induction, or ethanol at the same dilution for control conditions. The TE and parietal endoderm were removed by dissection using forceps. Then, the epiblast was separated from the visceral endoderm by 8 min incubation on ice in a solution of 2.5% pancreatin (Sigma-Aldrich, P3292) and 0.5% trypsin (Sigma-Aldrich, T4799). Four biological replicates were collected per condition (treatment versus control), each containing a group of ten isolated epiblasts. The samples were immediately placed into a 1.5 ml tube with 350 µl of lysis buffer, as per manufacturer's instructions of the Micro RNeasy Kit (Qiagen, 74004), used for RNA extraction. After extraction, RNA was quantified using a Qubit™ RNA High Sensitivity (HS) kit (Invitrogen, Q32852). Libraries were prepared from 7 ng of RNA per sample using the NEBNext^® Single Cell/Low Input RNA Library Prep Kit for Illumina (New England Biolabs, E6420). Libraries were pooled and sequenced using a P3-100 kit from Illumina in a NextSeq2000 sequencer following manufacturer's instructions.

Raw reads were processed with bcl2fastq (v2.19.1) and STAR (v2.5.3a) was used to map sequencing reads and to generate the count table (Dobin et al., 2013). Genes with less than ten counts were discarded from downstream analysis. One sample of 0 h EpiLCs and one sample of 0 h E3.5 were discarded at quality control. PCA and differential expression analysis were performed using the DESeq2 (Love et al., 2014) and factoextra (https://rpkgs.datanovia.com/factoextra/index.html) packages in R, significance was defined as abs(LFC)>1.5 and Padj<0.05 for in vitro samples, and abs(LFC)>0.75 and Padj<0.05 for in vivo samples. P-value was calculated using the Wald test in the DESeq2 package. PCA was computed using the prcomp function. Hierarchical clustering was performed using the hclust function. Enrichment of GO terms was determined using the clusterProfiler package (Wu et al., 2021; Yu et al., 2012). Heatmaps were created using ComplexHeatmap (Gu et al., 2016).

ATAC-seq or Omni-ATAC was performed as described in Grandi et al. (2022). Briefly, 50,000 cells were collected in a 1.5 ml DNA LoBind tube (Eppendorf, 0030108051) and lysed for 3 min on ice in 0.1% IGEPAL, 0.1% Tween-20, 0.01% Digitonin (Promega, G9441). Nuclei were washed and pelleted for 10 min at 500 g, and then treated for 30 min at 37°C with Tn5 transposase (Illumina, 20034197). The DNA was extracted using DNA Clean and Concentrator 5-kit (Zymo Research, D4014) and libraries were prepared according to Buenrostro et al. (2013). Libraries were quantified using the NEBNext Library Quant Kit (New England Biolabs, E7630S), then pooled at same molarity and paired-end sequenced with a P2-100 kit from Illumina in a NextSeq2000 sequencer following the manufacturer's instructions.

Raw reads were processed with bcl2fastq, adapters were trimmed with cutadapt (Martin, 2011) and reads were mapped to mm10 using bowtie2 (Langmead and Salzberg, 2012). ChrM, duplicate reads and reads in blacklisted regions were removed. Peaks were called using macs2 (v2.2.7.1) (--broad -f BAMPE --qvalue 0.001) and consensus peaks per condition were defined from two replicates per condition using the DiffBind package (https://bioconductor.org/packages/release/bioc/html/DiffBind.html). We removed promoter regions using the publicly available Eukaryotic Promoter Database (Dreos et al., 2015; Périer et al., 1998). Reads within defined peaks were counted using subread featureCounts -p v2.0.3 (Liao et al., 2014). Counts were used to generate a count table for downstream differential enrichment analysis using DESeq2 (Love et al., 2014), to obtain the LFC enrichment per peak before and after ERK induction. Significance was defined as abs(LFC)>1.5 and Padj<0.01. Overlap of peaks between conditions was determined using bedtools intersect (v2.31.0) (Quinlan and Hall, 2010). Motif analysis and peak annotation was performed using HOMER (v4.11, findMotifsGenome.pl -size 200; and annotatePeaks.pl) (Heinz et al., 2010). Heatmaps and profile plots were created using deeptools (v3.5.1) (Ramírez et al., 2016). TF footprinting and differential binding score was determined using TOBIAS (Bentsen et al., 2020). Correction and scoring were carried out using merged bam files with both replicates, against all consensus peaks. BINDetect was performed in the peak set of interest. TF binding motifs for footprinting analysis were downloaded from the latest version of JASPAR (Rauluseviciute et al., 2023; Sandelin et al., 2004).

CUT&RUN (Cleavage Under Targets and Release Using Nuclease) (Skene and Henikoff, 2017) was performed on cells attached to a 24-well culture plate as previously described (Miura and Chen, 2020). Briefly, cells were permeabilised with Perm buffer [0.1% Triton X-100, 20 mM HEPES-KOH pH 7.5, 150 mM NaCl, 0.05 mM Spermidine and proteinase inhibitors (Roche, 05056489001)] for 15 min at room temperature and incubated with the corresponding primary antibody (see Table S11) overnight at 4°C. Recombinant pAG-MNase was bound for 1 h at room temperature, then activated in 5 mM CaCl₂ at 4°C for 30 min. The reaction was stopped with STOP buffer [680 mM NaCl, 40 mM EDTA, 8 mM EGTA, 100 μg/ml RNase A (Invitrogen, 12091021) and 0.1% Triton X-100] and DNA fragments were released by incubating at 37°C for 30 min. DNA was extracted from the collected supernatant using the Qiagen DNeasy Blood & Tissue Kit (Qiagen, 69504) and 6 ng of DNA were used for library prep with the NEBNext Ultra II DNA Library Prep Kit for Illumina (New England Biolabs, E7645) following the manufacturer's instructions. Libraries were pooled and paired-end sequenced using a P2-100 kit from Illumina in a NextSeq2000 sequencer following the manufacturer's instructions.

Reads were pre-processed using the nf-core CUT&RUN pipeline v3.2.2 (https://nf-co.re/cutandrun/3.2.2/; Ewels et al., 2020). Peaks were called using SEACR (Meers et al., 2019), and DiffBind (https://bioconductor.org/packages/release/bioc/html/DiffBind.html) was used to determine consensus peaks (using minOverlap 3/4 replicates) and to extract differentially bound peaks [Padj<0.01 and abs(LFC)>0.5]. Peaks were determined against a negative IgG control generated for each cell type. Bedtools intersect (v2.31.0) (Quinlan and Hall, 2010) was used to determine shared and specific regions between cell types. Profile plots were generated using deeptools (v3.5.1) (Ramírez et al., 2016).

We thank Heike Wollmann, Magali Michaut, Adrija Kalvisa and the reNEW Genomics Platform for technical expertise, support and the use of instruments. We thank Jutta Bulkescher, Anup Shrestha and the reNEW Imaging Platform for training, technical expertise, support and the use of microscopes. We thank Javier Martin Gonzalez, Ricardo Alonso Laguna Barraza and the Core Facility for Transgenic Mice for technical expertise and support. We thank the entire Brickman lab, especially Molly P. Lowndes and Rita S. Monteiro, for critical comments on this manuscript.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.202842.reviewer-comments.pdf