Med12 cooperates with multiple differentiation signals to facilitate efficient lineage transitions in embryonic stem cells Open Access

Cell differentiation during development is regulated by a suite of extracellular signaling systems that trigger expression changes for large gene modules. To execute such complex transcriptional responses, the activity of intracellular signaling effectors must be integrated with that of general transcriptional regulators that direct the activity of RNA polymerase II. How the signaling systems associated with pluripotency and differentiation work together with general transcriptional regulators during early lineage transitions is still not fully resolved.

The earliest cell differentiation events of mammalian embryogenesis first segregate the extraembryonic trophoblast from the inner cell mass (ICM). In a second step, the ICM further differentiates into extraembryonic primitive endoderm (PrE) and the pluripotent embryonic epiblast (Epi), which ultimately forms the fetus (Chazaud and Yamanaka, 2016). Subsequently, epiblast cells transition from the naïve pluripotent state at pre-implantation to formative and then primed pluripotency as they prepare for germ layer differentiation (Kalkan and Smith, 2014; Nichols and Smith, 2009; Smith, 2017). Embryonic stem cells (ESCs) allow modeling both the differentiation towards an extraembryonic PrE identity, as well as transitions between different pluripotent states (Kalkan et al., 2017). Mouse ESCs (mESCs) can either be maintained in medium containing serum and the cytokine LIF, or in a ground state of pluripotency using serum-free N2B27 medium supplemented with LIF and two small-molecule inhibitors that activate Wnt/β-catenin signaling and inhibit FGF and ERK (FGF/ERK) signaling, respectively (2i+LIF; Ying et al., 2008). Efficient PrE differentiation from mESCs can be achieved from ground state pluripotency by the forced expression of GATA transcription factors together with active FGF/ERK signaling (Fujikura et al., 2002; Schröter et al., 2015; Wamaitha et al., 2015). Transitions between pluripotency states in contrast can be triggered by the removal of small-molecule inhibitors from the culture medium alone (Kalkan et al., 2017; Neagu et al., 2020). Together with experiments in the mouse embryo, these stem cell models have provided a comprehensive picture of the signaling control of the early lineage transitions in the mammalian embryo. Both PrE differentiation and exit of epiblast cells from naïve towards formative and primed pluripotency require FGF/ERK signaling as well as Wnt signaling inhibition (Athanasouli et al., 2023; Chazaud and Yamanaka, 2016; Kang et al., 2013; Nichols et al., 2009). PrE differentiation further benefits from LIF signaling (Morgani and Brickman, 2015), whereas the progression of epiblast cells is promoted by the Notch signaling effector RBPJ (Kalkan et al., 2019), and the mammalian target of rapamycin (mTOR) signaling effector TFE3 (Betschinger et al., 2013; Villegas et al., 2019).

Developmental signaling systems often culminate in the activation or deactivation of sequence-specific transcription factors. In eukaryotes, transmission of such transcription factor activity changes into altered RNA polymerase activity at specific promoters requires large multiprotein assemblies such as the Mediator complex, which physically bridges between transcription factors and the basal transcriptional machinery (Soutourina, 2018). The mammalian Mediator complex is formed by up to 30 subunits. It can be subdivided into a head, a core and a tail domain, and the transiently associated CDK8 module, which consists of four subunits – the CDK8 kinase, CCNC, MED12 and MED13 (Luyties and Taatjes, 2022; Soutourina, 2018). Unperturbed Mediator function is required for expression of most protein-coding genes (Soutourina, 2018). Still, individual Mediator subunits have been linked to transcriptional changes in response to specific signaling systems. Deletion of Sur2, which encodes the MED23 subunit, for example abrogates transcriptional activation downstream of ERK–MAPK signaling in mESCs (Stevens et al., 2002). The CDK8 module in particular has been implicated in directing rapid changes in gene expression patterns in response to various stimuli, such as serum stimulation (Donner et al., 2010; Luyties and Taatjes, 2022). Furthermore, CDK8 inhibition in mouse and human ESCs impairs pluripotency exit, mirroring the effects of MEK/ERK inhibition (Lynch et al., 2020). MED12, which activates CDK8 function in the kinase module (Knuesel et al., 2009; Park et al., 2018), is essential for axis elongation and the activation of Wnt target genes during mouse development (Rocha et al., 2010). Together, these works suggest that the transmission of specific developmental signals to RNA polymerase II activity can be mapped to specific Mediator subunits. How general these mappings are or whether they are context dependent is, however, not clear. It is also not known how interference with Mediator activity in pluripotent cells affects differentiation efficiency of different lineages.

Here, we aim at identifying factors that mediate transcriptional changes in response to signaling events during early mammalian cell differentiation, using the expression of a Spry4^H2B-Venus reporter allele as read-out in a genome-wide CRISPR screen. This reporter is an established indicator of developmental FGF/ERK signals, and its expression is switched on in both Epi and PrE cells during preimplantation development (Morgani et al., 2018). Our screen returns both known and new signaling inputs into Spry4^H2B-Venus reporter expression, as well as several components of the Mediator and Elongator complexes. Using epistasis analysis, we demonstrate that Med12, one of the strongest hits in the screen, functions independently of and in parallel to the FGF and mTOR signaling systems in pluripotent cells. Functional assays showed that, although not strictly required for lineage transitions, loss of Med12 leads to impaired signal responsiveness during pluripotency transitions. Collectively, these results point to new signal-independent functions of Med12 that help cells to efficiently execute lineage transitions in early development.

The differentiation of pluripotent cells towards different lineages rests on transcriptional changes triggered by extracellular signals. To identify effectors of signal-regulated gene expression during cellular differentiation, we performed a genome-wide CRISPR screen using the expression of a Sprouty4 reporter as a read-out (Hanna and Doench, 2020; Morgani et al., 2018; Fig. 1A). We chose Sprouty4 because it is a known target gene of developmental FGF signals, and shows strong expression changes upon both Epi and PrE differentiation (Morgani et al., 2018; Fig. S1A). In serum+LIF medium, the Spry4 reporter is expressed due to paracrine FGF signals. We generated Cas9-expressing Spry4:H2B-Venus reporter cells and transduced them with the Brie gRNA library targeting protein-coding genes (Doench et al., 2016). To identify positive and negative regulators of Spry4 reporter expression, we sorted cells, by flow cytometry, with decreased and increased fluorescence at 6 and 9 days after transduction, and determined enriched guides in sorted fractions (Fig. 1A). We first analyzed perturbations leading to reduced reporter expression. Gene-targeting gRNAs were more strongly enriched in the sorted fractions compared to non-targeting controls (Fig. 1B; Fig. S1B–G, Table S1). We used the robust rank algorithm (RRA; Li et al., 2014) to combine information from multiple gRNAs and to rank genes in each of the four conditions (1% and 5% gate at both day 6 and day 9, Table S1). This analysis revealed up to 17 individual genes with a false discovery rate (FDR)≤0.05, and up to 26 genes with an FDR≤0.2 (Fig. 1C; Fig. S2). We compiled a list of hits by selecting genes that were detected with an FDR≤0.05 in one condition, or with an FDR≤0.2 in at least two conditions (Fig. 1D). Protein–protein interaction network analysis with String-DB revealed that most of our hits fell into a small number of groups that were highly connected and associated with specific molecular functions (Szklarczyk et al., 2015; Fig. 1E). One of these groups contained the FGF signaling genes Fgfr1, Grb2, Sos1 and Ptpn11, as would be expected given the strong regulation of Spry4 by FGF signaling. Another group contained several genes involved in protein glycosylation and specifically the synthesis of heparan sulfates, such as Slc35b2, Ext1, Ext2 and Extl3. Heparan sulfates are crucial co-factors for efficient FGF signaling (Ornitz and Itoh, 2015). Given that pooled CRISPR screens mainly detect cell-autonomous functions of gene perturbations, the appearance of these hits indicates that surface-tethered heparan sulfates determine the responsiveness of a cell to FGF signaling. Three further groups contained genes associated with ribosome biogenesis and translation (Rpl9, Rpl18, Eif3i and Dhx37), as well as genes of the Elongator (Elp2–6, Ikbkap and Kti12) and Mediator complexes (Med10, Med12, Med16, Med24 and Med25; Fig. 1E), raising the possibility that these factors could have gene- or signal-specific functions in mESCs.

Next, to validate selected hits in an independent experimental setting, and to evaluate their effect size, we knocked out individual candidate genes using the most enriched gRNA from the screen. Given that we were interested in mechanisms of transcriptional regulation, we focused on hits related to the Mediator and Elongator complexes, but also included Grb2 and Ptpn11 as a reference to evaluate the effects of perturbing FGF signal transduction, and Sox2 and Fam98b as candidates that could not clearly be linked to a functional group (Fig. 1E). Flow cytometry showed that knockout of all tested candidates led to a reduction of mean Spry4:H2B–Venus fluorescence levels, albeit to a different degree – knockout of components of the Elongator complex as well as Fam98b and Sox2 affected Sprouty4:H2B–Venus expression only mildly, whereas knockout of the FGF signaling genes Grb2 and Ptpn11 had the strongest effect, although they did not reach the reduction achieved via pharmacological inhibition of FGF signal transduction with the MEK inhibitor PD0325901 (PD03) (Fig. 1F,G). Knockout of Mediator components reduced Spry4:H2B–Venus levels to different degrees, with Med12 having the strongest effect, reducing Spry4:H2B–Venus levels to 35.3±1.6% (mean±s.e.m.) compared to control (Fig. 1F,G). Thus, our screen and validation establish Med12 as a strong candidate regulator of signal-dependent gene expression in mESCs.

We next sought to use our screen to identify negative regulators of Spry4 expression, by looking at perturbations that led to increased reporter expression (Fig. 1A, right side of histogram). Also here, gene-targeting gRNAs were enriched over control guides in the sorted fractions (Fig. S2A–D; Table S1). Ordering according to RRA scores revealed up to 12 and 24 genes with FDR values of ≤0.05 and ≤0.2, respectively, in each of the conditions (Fig. 2A; Fig. S2E–G, Table S1). Using the same criteria as for the positive regulators, we compiled a list of 29 potential negative regulators of Spry4 transcription (Fig. 2B). Analysis with String-DB again showed that many of these hits were highly connected and associated with specific molecular functions (Fig. 2C). A large group of genes encoded proteins that localized to mitochondria or were otherwise associated with mitochondrial functions. Several genes had signaling functions – Lztr1 is a negative regulator of RAS-MAPK and hence FGF signaling, in line with the strong representation of genes promoting FGF signaling being among the positive regulators of Spry4 expression. We detected four genes that were related to mTOR signaling (Tsc1, Tsc2, Flcn, and Lamtor). Finally, we found a big group that contained genes involved in chromatin modification and transcription, amongst them three genes encoding SWI/SNF-related proteins. Thus, similar to the collection of positive hits identified above, our screen yields both signaling and transcriptional regulators that negatively control Spry4 expression. To select hits for validation, we focused on genes associated with mTOR signaling, because these genes have also been implicated in the maintenance of pluripotency (Betschinger et al., 2013; Li et al., 2018; Villegas et al., 2019). We also included Lztr1 as an FGF signaling gene, and Smarcc1 as a representative of the group of chromatin modifiers. Knockout of all six individual candidate genes with single gRNAs led to an increase of mean Spry4:H2B–Venus fluorescence levels in reporter cells. The effect of knocking out the mTOR signaling genes was stronger than that of knocking out Lztr1 or Smarcc1, and almost doubled reporter expression levels compared to mock-transfected cells (Fig. 2D,E).

We then wanted to know how the signaling and transcriptional regulators identified in the CRISPR-screen were functionally related. We focused on Med12, given that of all transcriptional regulators, it had the strongest effect on Spry4 reporter expression. Furthermore, MED12 is a component of the kinase module of Mediator, which could couple the activities of specific signaling systems to transcriptional activity (Lynch et al., 2020; Rocha et al., 2010). To analyze Med12 functions in mESCs, we first generated monoclonal Med12-mutant cell lines. We generated multiple independent clonal lines lacking part of exon 7, a region that was also targeted by sgRNAs used in the CRISPR-screen, and confirmed the loss of MED12 protein expression by immunoblotting (Fig. S3A,B). Med12-mutant cells grew normally in both serum+LIF and 2i+LIF medium, and showed a reduced increase in Spry4 reporter expression upon switching from 2i+LIF to the N2B27 base medium (Fig. S3C,D). p-ERK (phosphorated ERK1 and ERK2) levels were indistinguishable between wild-type and Med12-mutant lines (Fig. S3E,F), indicating that reduced Spry4 reporter expression is not due to deregulated FGF/ERK signaling. Fluorescent in situ hybridization revealed lower numbers of Spry4 mRNAs in Med12-mutant compared to wild-type cells (Fig. S3G–I), suggesting that Med12 regulates reporter expression transcriptionally. Intriguingly, we also found slightly lower numbers of Nanog mRNAs in Med12 mutants in 2i medium, indicating that Med12 might be quantitatively involved pluripotency gene expression.

To assess gene expression differences between wild-type and Med12-mutant cells more globally, we performed bulk RNA sequencing of cells in 2i medium and upon 24 h of differentiation in N2B27 (Fig. 3A). Principal component analysis separated samples in pluripotency and differentiation medium along PC1 (32.1% of variance), and wild-type and Med12-mutant cells along PC2 (16.8% of variance, Fig. 3B). Fewer genes were differentially expressed between pluripotency conditions and 24 h in N2B27 in Med12-mutant compared to wild-type cells. Most genes that were differentially expressed in the mutant were also differentially expressed in the wild type (Fig. 3C, left; Table S2). When comparing between wild-type and Med12-mutant cells within each culture condition, we found that more genes were differentially expressed between the two genotypes in N2B27 than in 2i (Fig. 3C, right; Table S2). Thus, loss of Med12 impairs gene expression changes at the exit from pluripotency. Med12 has a closely related paralogue named Med12l. Med12l expression was upregulated in Med12-mutant cells (Fig. S3J), and Spry4 reporter expression in Med12-mutant lines could be further reduced by simultaneously knocking out Med12l (Fig. 3, Fig. S3K). Thus, it is possible that Med12l partially compensates for loss of Med12 function in the mutant lines.

Exit from pluripotency is regulated by the interplay of a set of signaling systems, such as mTOR, Notch, Wnt, and FGF (Betschinger et al., 2013; Kalkan et al., 2019; Ying et al., 2008). We next asked whether gene expression downstream of any of these signaling systems was affected by loss of Med12, using previously defined sets of target genes specific for each signaling system (Lackner et al., 2021). A strong expression change of such a set of target genes between wild-type and mutant cells can be considered a footprint of the perturbation of the corresponding signaling system. We quantified the footprint for the loss of Med12 by comparing the gene expression changes of the defined list of target genes during 24 h of differentiation between the wild-type and the mutant. This gene expression footprint was strongest for mTOR target genes, less pronounced for Notch and Wnt target genes, and virtually absent for FGF target genes (Fig. 3D, top row). In comparison to previously described examples (Fig. 3D, lower rows; Lackner et al., 2021) for signal-specific perturbations, the effects of the loss of Med12 appeared less specific for a single pathway. We functionally tested the relationship between Med12 and mTOR signaling by knocking out the mTOR signaling genes Flcn, Lamtor4, Tsc1 and Tsc2 in Med12-mutant Spry4^H2B-Venus/+ cells. Similar to the situation in the wild type, knockout of these genes led to increased reporter expression, albeit from a lower baseline level (Fig. 3E). Therefore, Med12 regulates Spry4 expression independently from mTOR signaling. In addition to mTOR signaling genes and Mediator subunits, our CRISPR screen revealed a large number of genes involved in FGF signal transduction, but the footprinting analysis suggested that FGF inputs into gene expression were independent of Med12. To further corroborate this result, we generated Med12 mutations in the background of an Fgf4-mutant mESC line, which allowed us to specifically analyze the effects of FGF signaling upon the switch from pluripotency to differentiation medium. We wanted to focus on immediate gene expression changes triggered by FGF signaling, and therefore analyzed gene expression changes by bulk RNA sequencing after 6 h of transfer into N2B27 medium with or without addition of exogenous FGF (Fig. 3F). Again, principal component analysis showed that cells in pluripotency and differentiation medium were separated along PC1 (33.2% of variance), and wild-type and Med12-mutant cells separated along PC3 (9.3% of variance), both when analyzing independent Med12-mutant clonal cell lines and experimental replicates (Fig. 3G). If expression of FGF target genes was generally dependent on Med12, we would have expected that their fold-change would be lower in Med12-mutant compared to wild-type cells, and that the number of differentially expressed genes would be higher in the wild-type than in the Med12-mutant. However, when plotting the fold expression-change for each gene upon 6 h of FGF stimulation for wild-type versus Med12-mutant cells, we found that the majority of genes were induced to a similar degree in both genotypes (2006 genes), whereas similar numbers of 926 and 935 genes were differentially expressed in the wild-type or the Med12-mutant only (Fig. 3H; Table S3). Furthermore, the ratios of fold-change values of FGF target genes between Med12-mutant and wild-type cells showed a unimodal distribution with a mode of 1.068 for downregulated and 0.927 for upregulated genes (Fig. 3J). Although this slight deviation of the mode from 1 leaves open the possibility that Med12 influences their expression magnitude, overall these results argue against a strong and specific role of Med12 in the regulation of FGF target genes.

Loss of the Mediator subunit Med24 compromises PrE differentiation (Hamilton et al., 2019). We therefore asked whether Med12 was likewise required for the segregation of Epi and PrE identities, using their differentiation following transient expression of GATA6–mCherry as a model system (Fig. 4A; Schröter et al., 2015). A mix of SOX17-positive PrE cells and NANOG-positive Epi cells differentiated irrespective of Med12 status, but the fraction of PrE cells was lower, and that of Epi cells was higher in mutant compared to wild-type cultures (Fig. 4B,C). Those differences in cell type ratios could not be rescued by supplementing the differentiation medium with FGF4 (Fig. S4A,B). However, similarly high proportions of PrE cells differentiated from both wild-type and Med12-mutant cells when cells sorted for high GATA6–mCherry levels were treated with FGF4 during differentiation (Fig. S4C,D), indicating that the main cause for these ratio differences were lowered expression levels of the GATA6–mCherry transgenes in Med12-mutant cell lines compared to that in wild type (Fig. 4D). This conclusion was further corroborated when we connected GATA6–mCherry expression levels with differentiation outcome in individual cells through live-cell imaging followed by immunostaining for fate markers (Schröter et al., 2015). GATA6–mCherry levels rose more slowly and reached lower peak levels in the mutant relative to wild type, even when using longer induction times (Fig. S4F; Movie 1). Still, PrE differentiation could be predicted based on GATA6–mCherry expression levels with similar precision in both wild-type and Med12-mutant cells (Fig. S4G–I), and GATA6–mCherry threshold values required to trigger PrE differentiation appeared even slightly lowered in Med12-mutant cells (Fig. S4J). Thus, loss of Med12 impairs transgene induction levels but not general PrE differentiation potential.

To test whether loss of Med12 had more subtle effects on PrE differentiation, we performed single-cell RNA sequencing (scRNAseq) of wild-type and Med12-mutant cells in pluripotency conditions, and after GATA6–mCherry induction for 4 h (wild type) or 8 h (Med12 mutants) followed by 20 h of differentiation (Fig. 4E). Cells from 2i+LIF (pluripotency) and N2B27 medium (differentiation) formed two distinct groups in UMAP space that were marked by expression of the Wnt/β-catenin target Sp5 (pluripotency) or the differentiation marker Dnmt3l, respectively (differentiation) (Fig. 4F,G). Within both the pluripotency and the differentiation group, wild-type and Med12-mutant cells segregated from each other, indicating that loss of Med12 affected the single-cell transcriptional state in both conditions (Fig. 4F). In the differentiation group, the PrE markers Sox17, Dab2, and Cubn, and the Epi markers Nanog and Fgf4 showed a mutually exclusive expression pattern in both wild-type and Med12-mutant cells (Fig. 4G), providing an opportunity to compare lineage-specific effects of the loss of Med12 between Epi and PrE cells. To consistently identify Epi and PrE cells across genotypes, we integrated the five differentiated samples and clustered them to separate two major groups that were separated along PC1 in principle component space (Fig. 4H). Gata6 and Nanog expression identified these two clusters as PrE (cluster 0) and Epi (cluster 1), respectively (Fig. 4H,I; Table S4). The proportion of PrE cells in this dataset was again lower in Med12 mutants compared to wild-type, consistent with immunostaining results (Fig. 4J). To evaluate the roles of Med12 for lineage-specific gene expression changes upon differentiation, we first selected genes that had a log2 fold change ≥0.5 between pluripotency and differentiation in wild-type cells and plotted their fold-change in the mutant (Fig. 4K). These fold-changes showed a unimodal distribution in all four conditions (up- and down-regulation, and Epi and PrE differentiation). Furthermore, when plotting the fold change of all genes upon differentiation against each other for wild-type and mutant cells, we found that data points uniformly clustered around the line of unity (Fig. S4K,L). Finally, many of the top differentially expressed genes between wild type and mutant were the same in the pluripotent, Epi and PrE cells (Fig. S4M; Table S5). Taken together, these results suggest that loss of Med12 does not affect specific gene modules during differentiation, but rather results in gene expression changes that are shared between differentiated states.

Prompted by the observation that GATA6–mCherry induction levels were consistently lowered in Med12-mutant cells, we asked whether loss of Med12 had more global effects on transcriptional output and mRNA levels in cells. The multiplexed nature of our dataset allowed us to compare mRNA levels between conditions, revealing that differentiation led to an increase in captured mRNAs in both wild-type and mutant cells (Fig. S5N). Furthermore, fewer mRNAs were captured from differentiated Med12-mutant cells compared to wild-type cells [median number of unique molecular identifiers (UMIs) was 28227 and 25649, respectively; Fig. S5N]. Whereas EU labelling of newly synthesized RNAs suggested that the higher mRNA content in differentiated compared to pluripotent cells was associated with increased transcriptional output, we could not detect significant differences in EU labelling between wild-type and Med12-mutant cells (Fig. S5O–R). This leaves open the possibility that reduced mRNA levels in Med12-mutant cells arise from effects other than reduced transcriptional output.

Even though the results above indicate that Med12 is not strictly required for lineage segregation in embryonic stem cells, it might still shape the dynamics cell differentiation. To test this possibility, we focused on transitions between different pluripotency states. We first probed exit from naïve pluripotency with a colony formation assay (Kalkan et al., 2017), where we differentiated cells in N2B27 for 48 h, followed by seeding at clonal density in 2i+LIF medium (Fig. 5A). Because we grew cells in the presence of LIF before transfer into N2B27, wild-type cells still formed a large number of pluripotent colonies (Lackner et al., 2021). In Med12-mutant cells, in contrast, the number of pluripotent colonies was significantly reduced (Fig. 5B, left; Fig. S5A, Table S6).

To test how stable this phenotype is under different signaling conditions, we carried out the same colony assay in an Fgf4 mutant background. Upon supplementation of N2B27 with FGF4, the number of pluripotent colonies formed by Fgf4;Med12 double mutant cells was much reduced, both compared to the Fgf4 single mutant supplemented with FGF4, and the Fgf4;Med12 double mutant in the absence of FGF4 (Fig. 5B, right, Fig. S5B). Although we cannot rule out the possibility that the use of fetal bovine serum (FBS) required for the proliferation of Fgf4 mutant cells in 2i+LIF medium (Raina et al., 2021) had selective effects on the Med12-mutant cells, these results suggest that loss of Med12 impairs the ability of single cells to form naïve pluripotent colonies in 2i+LIF medium following transient differentiation in N2B27.

The ability to form pluripotent colonies in the colony formation assay depends on the pluripotency state of a cell at the beginning of the differentiation period, and its ability to react to changing culture conditions. We used our bulk RNA sequencing data comparing wild-type and Med12-mutant cells after 24 h in N2B27 to evaluate which of these properties were affected upon loss of Med12. We first determined differentiation delays of the two genotypes using a published high time-resolution reference dataset (Lackner et al., 2021; GEO: GSE145653). Surprisingly, this analysis indicated that Med12-mutant cells differentiated more slowly than the wild-type (Fig. 5C). This could be a consequence of stronger expression of naïve genes in pluripotency conditions, or reflect slower downregulation of these genes in N2B27 in the Med12 mutant. To distinguish between these possibilities, we plotted the expression levels of seven selected pluripotency marker genes in 2i and after 24 h in N2B27 in wild-type versus Med12-mutant cells. In this plot, data points above and below the unit line indicate higher and lower expression in the mutant relative to the wild type, respectively, and the slope of the connecting line between expression in 2i and N2B27 indicates the dynamics of downregulation (Fig. 5D, left). We found that in 2i medium, all marker genes were less strongly expressed in the mutant compared to in the wild type. Consistent with our results from bulk RNA sequencing, expression levels of the pluripotency genes Klf4, Zfp42 and Tbx3 were reduced in the single-cell RNA sequencing dataset of the mutant cells (Fig. 4E), whereas levels of Prdm14, Nanog, Esrrb and Tcf21l1 were similar between the two genotypes (Fig. S5C). Furthermore, their slopes during differentiation were consistently smaller than one (Fig. 5D). Such a systematic reduction in the slope of downregulation was also observed when we analyzed the top 100 downregulated genes in wild-type or Med12-mutant cells (Fig. 5E, left). Upregulation of genes was similarly affected, as the top 100 upregulated genes also showed a slope lower than 1 (Fig. 5E, right). Given that a loss of Oct4 (POU5F1) can be sufficient to block the differentiation potential of mESCs (Radzisheuskaya et al., 2013), we checked the protein expression level by immunostaining (Fig. S5D). However, there was no significant difference in expression levels between Med12-mutant and wild-type cells (Fig. S5E). Taken together, these results indicate that Med12-mutant cells show an impaired pluripotency gene expression profile as well as a slowed reconfiguration of gene expression programs upon changing culture conditions, thereby implicating Med12 as a global regulator of pluripotency transition dynamics.

Here, we use a Spry4^H2B-Venus reporter to screen for regulators of developmental gene expression in pluripotent stem cells. This screen returned components of the FGF/ERK and the mTOR signaling systems that positively and negatively regulate reporter expression, respectively, as well as several members of the Mediator and Elongator complexes. Focusing on Med12, we show that it cooperates with multiple signaling systems to regulate gene expression in pluripotent cells. Med12 was not required for the differentiation of Epi and PrE identities, but dynamics of pluripotency transitions were impaired in Med12-mutant cells. Together, these results suggest that Med12 amplifies transcriptional changes in pluripotent cells.

Previous genome-wide screens have used retention of clonogenicity or the continued expression of pluripotency-associated reporter alleles in differentiation conditions as read-outs to identify regulators of pluripotency and lineage transitions (Betschinger et al., 2013; Kagey et al., 2010; Li et al., 2018; Villegas et al., 2019). In our study, the combination of a Spry4^H2B-Venus allele with flow sorting constituted a highly sensitive read-out that was focused on the activity of specific signaling systems in pluripotent cells, and reliably detected new regulators even if they only had small effect sizes. The specificity of the screen shows in the strong representation of genes involved in the FGF/ERK signaling cascade, from genes that encode synthetases for FGF co-factors, over FGF receptors, to intracellular signaling genes. Surprisingly, however, we did not find any sequence-specific transcription factors downstream of FGF/ERK signaling in our screen. This could be explained by the expression of multiple functionally redundant FGF/ERK signaling effectors in pluripotent cells, or through previously proposed transcription factor-independent regulation of RNA polymerase activity by ERK (Tee et al., 2014).

In addition to components of FGF and mTOR signaling systems, the screen returned several members of the SWI/SNF and the Mediator complexes. The core Mediator complex is thought to be required for the expression of most genes in eukaryotic genomes, but individual subunits have been suggested to regulate gene expression downstream of specific signaling systems such as the serum response network (Donner et al., 2010; Stevens et al., 2002) or Wnt (Rocha et al., 2010). However, when we tested this idea for Med12, we found that its loss did not phenocopy the effects of specific signaling perturbations. This finding suggests that previously reported functional connections, such as the link between Med12 and Wnt signaling (Rocha et al., 2010), are strongly context dependent.

Med12 encodes a crucial component of the Mediator-associated CDK8 module. In contrast to pharmacological inhibition of CDK8, which has been reported to boost pluripotency in a manner similar to ERK signaling inhibition (Lynch et al., 2020), we find that loss of Med12 leads to lower expression levels of pluripotency markers. This observation argues against a positive role of Med12 for the activity of the CDK8 module, which is further supported by the absence of any other CDK8 module components in our screen. Lynch et al. found that maintenance of pluripotency requires CDK8 to be present, but its enzyme activity has to be turned off. One possible explanation for the opposing phenotypes of CDK8 inhibition and Med12 loss-of-function therefore is that MED12 participates in assembling kinase-inactive CDK8 complexes that support pluripotency. Alternatively, they might arise from Med12 functions that are independent from the CDK8 module (Aranda-Orgilles et al., 2016).

Med12 has been found not only to preserve pluripotency in the epiblast, but also to maintain an undifferentiated state of the trophoblast (Halstead et al., 2024 preprint). Its role to maintain pluripotency gene expression reported here is supported by previous studies that found that MED12 and NANOG proteins interact, that MED12 and NANOG have similar DNA-binding profiles, and that MED12 promotes Nanog expression (Apostolou et al., 2013; Tutter et al., 2009). Several Mediator subunits, including Med12, have been identified in a screen for pluripotency maintenance that focused on transcription and chromatin regulators (Kagey et al., 2010). That study furthermore suggested that interactions between Mediator and cohesin contribute to genome folding and efficient enhancer–promoter interactions. When we probed Med12 functions in lineage transitions, we found that loss of Med12 did not compromise PrE differentiation in contrast to a previously described role of Med24 (Hamilton et al., 2019). Med12-mutant cells, however, show a slower downregulation of pluripotency genes and a decreased ability to revert to naïve pluripotency in a colony-forming assay. We speculate that these cellular phenotypes are a reflection of the function of Mediator in establishing promoter–enhancer contacts (Kagey et al., 2010), which might impair the ability of Med12-mutant cells to rapidly reconfigure their chromatin upon changing signaling environments. It is likely that the expression of individual genes is differentially sensitive to the loss of Med12. This might be the reason why expression from the Spry4 locus, which is the most strongly upregulated gene upon acute FGF stimulation, shows a particularly high sensitivity to loss of Med12. In line with this idea, requirements of Med12 for efficient induction are not exclusive to endogenous genes but extend to exogenous transgenes such as the inducible GATA6–mCherry construct used to trigger primitive endoderm differentiation in our study. Surprisingly, such differential quantitative defects in the regulation of single genes upon loss of Med12 do not lead to strong defects in acquiring early differentiated fates, such that transcriptomes of individual differentiated cells are not systematically different from each other in Med12-mutant and wild-type cells. This suggests that intracellular regulatory networks can buffer the composition of cellular transcriptomes against variable transcription efficiencies.

Routine culture of mESCs was performed at 37°C with 5% CO₂ in either serum+LIF medium [ESL; composed of GMEM with 10% FBS, 2 mM GlutaMAX, 1 mM sodium pyruvate, 0.1 mM β-mercaptoethanol (all from Thermo Fisher Scientific) and 10 ng/ml LIF (protein chemistry facility, MPI Dortmund)] on 0.1% gelatine-coated dishes, or in 2i+LIF medium on fibronectin-coated dishes. 2i+LIF is N2B27 supplemented with 1 µM PD0325901 (SelleckChem), 3 μM CHIR99201 (Chiron, Tocris) and 10 ng/ml LIF. N2B27 was prepared as a 1:1 mixture of DMEM/F12 and Neuropan Basal Medium (both from PAN Biotech), supplemented with 1× N2 and 1× B27 supplements, 1× L-glutamax, 0.0025% bovine serum albumin (BSA) and 0.2 mM β-mercaptoethanol (all from Thermo Fisher Scientific). Fgf4-mutant cell lines were cultured in 2i+LIF supplemented with 10% FBS. Cells were passaged every 2 to 3 days, and detached with trypsin (PAN Biotech) or Accutase (Sigma-Aldrich). All cells were regularly tested negative for mycoplasma contamination.

All cell lines generated in this study were derived from the E14tg2a wild-type line (Hooper et al., 1987). The GATA4–mCherry inducible line used for single-cell RNA sequencing has been described previously (Raina et al., 2021). The Spry4^H2B-Venus/+-reporter line was generated with a previously described targeting construct (Morgani et al., 2018) using Lipofectamine 2000 according to the manufacturer's instructions (Thermo Fisher Scientific). Correctly targeted clones were identified via long-range PCR as described in Morgani et al. (2018). GATA6–mCherry inducible lines were established as described for GATA4–mCherry inducible lines in Raina et al. (,2021,), but replacing the Gata4 with a Gata6 coding sequence in the PiggyBac vector for inducible gene expression. We established multiple clonal lines and tested them for GATA6–mCherry induction levels upon Dox treatment by flow cytometry. Three independent clones with induction levels similar or slightly higher than the previously established GATA4–mCherry inducible lines were selected for the experiment shown in Fig. 5D, and a single clonal line was chosen for all other experiments, including mutagenesis of Med12. Newly generated Spry4^H2B-Venus/+ reporter and GATA6–mCherry inducible cell lines were checked for karyotypic abnormalities. To label nuclei for timelapse imaging, cells were transfected with pCX-H2B-Cerulean-IRES-puro (Schumacher et al., 2024). Cell lines carrying PiggyBac transgenes were kept under appropriate selection to prevent transgene silencing over passaging.

For mutagenesis of individual genes via CRISPR/Cas9, gene targeting sgRNAs (Table S7) were cloned into pX459 (Addgene plasmid #48139) using BbsI (NEB) overhangs following the process described in Ran et al. (2013,). Clonal mutant lines were generated using a combination of sgRNAs with targeting sequences 100 to 200 bp apart in the genome. Single sgRNAs were used when generating polyclonal lines. For validation experiments of the CRISPR screen (Figs 1 and 2), we selected the most enriched sgRNA in sorted cells. A total of 1 µg of sgRNA-containing pX459 vectors was mixed with a final concentration of 0.04 µg/ml Lipofectamine 2000 (Thermo Fisher Scientific) in Opti-MEM (Gibco) according to the manufacturer's protocol. For the generation of clonal lines, cells were seeded at clonal density into 10 cm dishes after transfection, for polyclonal experiments ∼50,000 cells/cm² were seeded. To enrich for successfully transfected cells, selection with 1.5 μg/ml puromycin was started 24 h after transfection for 48 h. To establish clonal lines, single-cell-derived colonies were picked 4 to 6 days after transfection and expanded. For molecular characterization of genetic lesions, genomic DNA was purified with Terra™ PCR Direct Genotyping Kit (Takara), followed by PCR amplification and Sanger sequencing of specific genomic regions encompassing the target site.

To generate stably CAS9-expressing Spry4^H2B-Venus/+ reporter cells, cells were transduced with lentiCas9-Blast lentiviral particles (Addgene #52962 Lentiviral Prep) at a multiplicity of infection of ∼0.1. Transduction was performed with attached cells, 20 h after seeding at 20,000 cells/cm², in presence of 5 µg/ml Polybrene (Sigma-Aldrich) in ESL. Continuous blasticidin (15 µg/ml, Gibco) selection was started 24 h after transduction. Lentiviral particles of the genome-wide gRNA library Brie (Addgene #73633) were generated according to standard protocols (Doench et al., 2016). For library transduction, 150×10⁶ CAS9-expressing Spry4^H2B-Venus/+ reporter cells were detached and mixed with the virus library in ESL with 5 µg/ml Polybrene. The following day, the same number of cells was reseeded and put under selection with puromycin (1.5 µg/ml, Sigma-Aldrich). Comparing cell counts with and without selection indicated a transduction efficiency of 25%, resulting in a >400-fold coverage of transduced cells per gRNA. In all subsequent steps, at least 31×10⁶ cells were processed to maintain gRNA coverage.

To identify gRNAs enriched in cell populations with high and low Spry4:H2B–Venus expression, at least 0.5×0⁶ cells with the lowest or highest 1% of Spry4:H2B–Venus fluorescence, or 3×10⁶ cells with the lowest or highest 5% of Spry4:H2B–Venus fluorescence were sorted by flow cytometry and their DNA isolated by column-based genomic DNA purification (Monarch Genomic DNA Purification Kit, NEB). For reference, the genomic DNA of 31×10⁶ non-sorted control cells was purified in parallel. The integrated gRNA was PCR amplified using Pfu polymerase (prepared in house) with a sample specific, sequencing adapter and index containing primers (Table S7; Carlini et al., 2021) using the complete purified genomic DNA as template. PCR samples were purified with the SPRIselect reagent (Beckman Coulter) with double-sided size selection. Briefly, 0.5× SPRIselect was added to each sample, incubated for 5 min at room temperature (RT) and the SPRIselect removed with a magnet. This supernatant was again mixed with 1.2× SPRIselect, incubated and then discarded. After washing the beads, the DNA library was eluted from the beads and used for sequencing.

Paired-End Illumina Sequencing with a read length of 150 bp pairs was performed with at least 10⁷ reads per sorted sample and 30×10⁶ reads for the unsorted library controls. The raw reads were trimmed using Cutadapt (Martin, 2011) to remove the vector-binding sequence. The reads were mapped to individual gRNAs, counted using norm-method total and statistically tested on the targeted gene levels using gene-lfc-method alphamean with Mageck (Li et al., 2014). Hits were selected based on the false discovery rate.

Immunostaining was performed as previously described (Schröter et al., 2015), Briefly, cells were washed with PBS containing Ca²⁺ and Mg²⁺, followed by fixation with 4% paraformaldehyde (Histofix, Sigma-Aldrich) for 15 min. Cells were permeabilized and blocked by rinsing and washing three times with PBS with 0.1% Triton X-100 and 1.0% bovine serum albumin (PBT-BSA). Primary antibodies [anti-mouse NANOG (Affymetrix eBioscience, cat. no. 14-5761), anti-SOX17 (R&D systems, cat. no. AF1924), anti-Oct3/4 (POU5F1, Santa Cruz Biotechnology, cat. no. sc-5279)] were diluted 1:200 in PBT-BSA and incubated with the cells overnight at 4°C. The next day, cells were washed in PBT-BSA and incubated with Alexa Fluor-conjugated secondary antibodies at 4 µg/ml (Invitrogen, Life Technologies) and Hoechst 33342 at 1 µg/ml (Invitrogen) in PBT-BSA in the dark for 2 h. Finally, samples were rinsed and washed with PBS and imaged in a mounting medium consisting of 80% glycerol, 16% PBS and 4% n-propyl-gallate.

For western blot analysis of MED12 and p-ERK, cells were washed twice with ice-cold PBS, supplemented with 1 mM activated orthovanadate in case of p-ERK detection. Cells were mechanically detached in lysis buffer, based on commercially available lysis buffer (Cell Signaling, cat. no. 9803S) supplemented with benzonase (Sigma-Aldrich, cat. no. E1014), cOmplete EDTA-free protease inhibitor cocktail (Roche, cat. no. 04693116001) and phosphate inhibitors P2 and P3 (Sigma, cat. nos P5726 and P0044). The lysates were snap frozen in liquid nitrogen twice and centrifuged (12,000 g for 10 min). Protein concentration in the supernatant was measured with a micro-BCA assay (Thermo Fisher Scientific). For western blot analysis, 20 µg of protein per sample were denatured by adding 5× Laemelli buffer and incubation at 95°C for 5 min. The SDS-PAGE was run in 1× MOPS buffer (Thermo Fisher Scientific) with 5 mM sodium-bisulfate and immediately transferred onto methanol-activated PVDF membranes. Transfer was performed in transfer buffer (12 mM Tris-HCl, 96 mM glycine and 20% methanol; all from Sigma Aldrich, pH 8.3) at 40 V for 1.5 h in a NuPage transfer system (Thermo Fisher Scientific). Membranes were blocked at RT for 1 h in Intercept blocking buffer (LI-COR), which was also used for the dilution and incubation with the primary antibodies anti-tubulin (1:10,000, T6074, Sigma-Aldrich), anti-pERK1/2 (1:1000, 4370S, Cell Signaling), anti-total ERK1/2 (1:1000, ab36991, Abcam) or anti-MED12 (1:250, A300-774A, Bethyl Laboratories). Appropriate secondary antibodies (IRdyes, LI-COR) were used at a dilution of 1:5000. Blots were imaged on an Odyssey CLx (LI-COR). Images of uncropped blots are shown in Fig. S6 for transparency. Quantification of p-ERK bands was performed using the gel quantification tool in Fiji, using the combined intensities of ERK1 and ERK2 bands, and normalizing p-ERK by total ERK signals.

In situ hybridization probes against Spry4 and Nanog were designed by, and all commercial reagents for the staining procedure were obtained from, Molecular Instruments, if not indicated otherwise. mRNAs were stained in solution as previously reported (Choi et al., 2018). Specifically, cells from confluent six-well plate wells were washed with PBS and detached using trypsin. Following centrifugation for 5 min at 200 g, the specification for all further centrifugation steps, cells were fixed with 4% PFA for 1 h. After removal of the fixative via centrifugation, cells were washed four times with PBS with 0.1% Tween 20 (PBST, Bio-Rad). Cells were incubated overnight in ice-cold ethanol. Starting with 10⁶ cells, and two additional washes with PBST, cells were incubated in 400 µl probe hybridization buffer for 30 min at 37°C. 2 pmol of gene-specific hairpins were prepared in 100 µl preheated probe hybridization buffer and added to the cell suspension, which was subsequently incubated overnight at 37°C. Preheated wash buffer was used to remove residual hybridization oligonucleotides by washing three times with 10 min of intermediate incubation. One final wash and 5 min incubation were performed in saline-sodium citrate (SSC, Sigma-Aldrich) buffer with 0.1% Tween 20 (SCCT). Cells were resuspended in 150 µl amplification buffer. 10 pmol of each labeled hairpin were mixed, heated to 95°C for 90 s, and cooled back to RT. Hairpins were added into 100 µl of amplification buffer, mixed with the cell suspension and incubated in the dark for 1 h. A final five washes with SCCT were followed by mounting the sample onto a microscopy glass slide. Samples were resuspended in 20 to 100 µl SCCT to ensure high single-cell densities. 2 µl of the cell suspension were squished between the slide and a cover slip to maximize area per cell and distance between mRNA spots.

Newly synthesized RNA was labeled using the Click-iT™ RNA Alexa Fluor™ 594 Imaging Kit (Invitrogen) according to the manufacturer's protocol. Briefly, we incubated cells with 1 mM 5-ethynyl uridine for the last 30 min or 1 h of differentiation or during pluripotency. After washing, fixation with 4% PFA and permeabilization, the click reaction with the Alexa Fluor 594 azide was performed for 30 min, followed by washing and nuclear staining with SiR-DNA 647 (Spirochrome) for 15 min.

Tilescans of immunostainings were imaged with a Leica SP8 confocal microscope (Leica Microsystems) with a 63×1.4 NA oil immersion objective. Images were analyzed in Fiji software (Schindelin et al., 2012). For segmentation, StarDist 2D (Schmidt et al., 2018) was used using the versatile (fluorescent nuclei) model and default post processing parameters. Mean fluorescence intensity was measured in segmented cells in all acquired channels. Cells with a nuclear area smaller than 40 µm² were filtered out. To determine fluorescence intensity threshold values for the classification of cell types, we manually selected thresholds that best bisected the bimodal expression profiles of the lineage markers. The same thresholds were applied to different samples in a single experiment. In the RNA-labeling experiment, after segmentation based on the nuclear staining and intensity measurement, a gaussian mixture model was fitted to the intensity distributions of the H2B–Cerulean channel. Using the posterior probability, the fluorescence intensity thresholds at which a cell can be assigned to the correct genotype with an ≥85% certainty were determined, and genotype assignment was performed accordingly, not assigning cells in between the populations. The integrated intensity of the RNA–Alexa Fluor 594 signal within the nucleus region was used to determine the amount of newly produced RNA within the labeling period.

Images of live Spry4^H2B-Venus/+ reporter cells were acquired with an Olympus IX81 widefield microscope, equipped with a stage top incubator (ibidi), pE4000 illumination (CoolLED) and ORCA-Quest qCMOS camera (Hamamatsu), with a 63×1.35 NA oil immersion objective. Hardware was controlled by Olympus CellSens Software. Images of in situ hybridization chain reaction (HCR) for mRNA counting were acquired on the same system with a 100×1.4 NA objective with 1024×1024 pixels and at least 20 z-slices, separated by 0.4 µm. Single cells were manually segmented on a brightfield image, taken in the middle of the z-stack. Counting mRNA spots from in situ HCR was performed using RS-FISH software (Bahry et al., 2022). The anisotropy was determined from single spots and set to 1.2. No robust fitting was applied. Sigma was set to 0.994. The threshold, the image maximum and the minimal spot intensity threshold were set for each channel per replicate manually, to account for different (background) intensity levels. Timelapse imaging was performed with a 40×0.9 NA objective on an Olympus IX81 widefield microscope, equipped with an LED-based illumination system (pE4000, CoolLED) and an iXon 888 EM-CCD camera (Andor). MicroManager (Edelstein et al., 2010) was used to control the hardware. Images were taken every 10 min. Tracking was performed with the manual tracking function in Trackmate v7 (Ershov et al., 2022) and fluorescence intensity was measured as the mean intensity in a spot with a 4 µm radius within the nucleus. In R Project for Statistical Computing software, tracks were smoothed with a rolling average over seven frames. For receiver operating characteristic (ROC) analysis the R package pROC (Robin et al., 2011) was applied and the optimal threshold was defined by the Youden's J statistic (Youden, 1950).

Analysis of Spry4:H2B–Venus reporter expression in live or fixed cells was performed on a LSRII flow cytometer (BD Biosciences). Cell sorting and analysis of GATA6–mCherry expression was carried out using a FACS Aria Fusion (BD Biosciences). Primary data analysis including gating single cells based on SSC and FSC was undertaken with FlowJo version 9 (BD Biosciences).

Clonogenicity assays were performed according to Kalkan et al. (2017). Briefly, 10⁴ cells/cm² were seeded in 2i+LIF for 24 h, followed by differentiation in N2B27 for 48 h. Control wells for each parental cell line were kept in 2i+LIF. Cells were then detached with Accutase to single cells, and 500 cells were reseeded into six-well plates with 2i+LIF+10% FBS. 10% FBS were included to support survival of Fgf4-mutant cells.

After 5 days, the colonies formed were fixed and stained with an alkaline phosphatase assay kit (Sigma-Aldrich) to distinguish pluripotent and differentiated colonies. Tile scans of the wells were acquired with an Olympus IX81 widefield microscope with a 4×0.16 NA objective. We applied background subtraction, gaussian blurring, Otsu-thresholding and conversion of images into a binary mask in ImageJ, and then used the AnalyzeParticles function to set thresholds for size and circularity and to determine the number of colonies. Colony numbers were normalized to the number of colonies obtained in the control.

For bulk RNA sequencing, cells were seeded at a density between 3.5 and 5.5×10⁴ cells/cm² in 2i or N2B27 + Chiron + LIF, followed by stimulation under indicated conditions. Replicates were obtained either from independent biological experiments (Fig. 3A,B) or from both independent biological experiments and independent Med12-mutant lines (Fig. 3F,G). RNA isolation was performed with TRIzol (ambion) according to the manufacturer's instructions. Sequencing libraries were prepared on polyA-enriched RNAs, followed by paired-end sequencing at a read-length of 150 bp and depth of ∼30×10⁶ reads per sample. Strand-specific libraries were generated only for the FGF-titration experiment (Fig. S1) and the differentiation of the Med12 wild-type and mutant cells (Fig. 3A,B). Raw reads were mapped to the mouse genome (GRCm39, release 108 (both Med12 mutant experiments) or release 97 (FGF titration experiment) with hisat2 (v2.1.0; Kim et al., 2019). SeqMonk was used to quantify counts per gene, either as TPM or as raw counts as input for downstream DESeq2 analysis (Love et al., 2014) for identification of differentially expressed genes.

Differentiation delay in Med12 mutants was estimated according to Lackner et al. (2021). We first determined the expression change of the naïve marker genes Nanog, Esrrb, Tbx3, Tfcp2l1, Klf4, Prdm14 and Zfp4 in Med12-mutant and wild-type cells, and then plotted the Euclidean distance of this expression change to that of the time-resolved dataset from Lackner et al. (2021).

Signaling footprint analysis in Med12 mutants was performed similarly to Lackner et al. (2021). This study defined a specific set of target genes for each pluripotency associated signaling system based on gene expression changes in knockouts of signaling genes. A signaling footprint for a knockout line can then be determined from the difference in the expression of pathway footprint genes to the wild-type line after 24 h of differentiation. Measures for the signaling footprint are the Spearman correlation between each knockout line and the respective pathway defining knockout, and the ratio between the sum of expression fold changes between a knockout line and the respective pathway defining knockout, defined as pathway activity. To compare the Med12 mutant data from this study, the wild-type conditions were used for batch correction.

Cells for scRNAseq were seeded at a density of 3.5×10⁴ cells/cm² in six-well plates in 2i+LIF and grown overnight. The next morning, induction in 2i+LIF+Dox (Sigma-Aldrich) was first started in the mutant clones, and 4 h later in the wild-type lines. After 8 h and 4 h, respectively, induction was stopped by washing once with N2B27, followed by 20 h of differentiation in N2B27. Controls for each cell line were continuously kept in 2i+LIF. For sequencing, cells were washed three times with PBS and detached with Accutase. Accutase was removed by centrifugation (200 g for 5 min) and 10⁶ cells per sample were resuspended in PBS+0.04% BSA and immediately used for multiplexing labeling following the protocol of 10× Genomics for samples with a viability above 80% (Cell Multiplexing Oligo Labeling for Single Cell RNA Sequencing Protocols with Feature Barcode technology, CG000391). Briefly, cells were spun down as above, resuspended with individual cell multiplexing oligonucleotides (CMO no. 301 to 310) and incubated for 5 min at RT. Cells were washed twice with PBS+1% BSA and passed through a cell strainer (FALCON, mesh size 35 µm). A total of 1.2×10⁵ single cells from all samples were pooled at equal ratios, and 4×10⁴ were used for droplet generation, corresponding to a target number of 2.4×10⁴ cell-containing droplets. Droplet generation, lysis, mRNA and cell barcode capture, and generation of both the gene expression library as well as the cell multiplexing library was performed following the instructions by 10× genomics (Chromium Next GEM Single Cell 3′ Reagent Kits v3.1 (Dual Index) with Feature Barcode technology for Cell Multiplexing, CG000388). Specifically, we chose 11 PCR cycles for cDNA amplification and 10 cycles for the sample index PCR. Concentration and insert size distribution for both the gene expression library and the cell multiplexing library were determined with a BioAnalyzer High Sensitivity DNA Assay (Agilent). Sequencing was performed on a NovaSeq 6000 on multiple flowcells with a paired-end 150 bp configuration. In total 1.2×10⁹ and 2.3×10⁸ read pairs were obtained for the gene expression and multiplexing library, respectively.

Demultiplexing to the individual samples, based on the cell multiplexing barcode and alignment to the mouse genome mm10 (GENCODE vM23/Ensembl 98, obtained from 10× Genomics) was performed with CellRanger (version 7.1.0, 10× Genomics). Downstream analysis was performed in R software with Seurat v5 (Hao et al., 2024). We first filtered cells by removing barcodes with ≤2500 detected genes and ≥15% of reads aligned to mitochondrial genes, retaining between 1100 and 1700 cells per sample with median mRNA counts per cell between 23,233 and 27,890 in the different samples. mRNA counts for each gene were normalized by dividing its counts by the total number of counts per cell, multiplied by 10,000. Log1p transformation was applied before plotting expression data as violin plots. For downstream analysis and representation of gene expression as heatmaps, centering counts for each feature and scaling to its standard deviation was applied. Principal component analysis was performed on the 2000 most-variable features in the relevant subset of cells. The resolution of the Louvain clustering algorithm was set to 0.05 when clustering multiple samples. For annotation of the Epi and PrE fate, the cells of the differentiated samples were integrated with Seurat integration based on the rpca reduction. Differentially expressed genes between cell states and genotypes were identified with the FindMarkers function in Seurat with a minimal expression difference in the log1p transformed expression values of 0.5.

We thank Kristin Nowak for generating Fgf4-mutant E14tg2a mESC lines, members of the Imig lab for help with lentivirus production, and the protein chemistry facility of the MPI Dortmund for supply of LIF and enzymes for molecular biology. Michelle Marten provided technical and organizational support for tissue culture and flow sorting throughout the project. We thank members of the Schröter lab, the Department of Systemic Cell Biology, Jochen Imig and Cristina Pina for critical discussions and feedback on earlier versions of the manuscript.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.263794.reviewer-comments.pdf