Fig 2: CKM-Mediator has a limited role in 3D genome organisation but is essential for Polycomb domain interactions.a, A representative Western blot analysis of CDK8 immunoprecipitation (n = 2) from nuclear extracts from Med13/13lfl/fl (WT) and Med13/13l-/- (CKM-Mediator KO) ESCs, probed with the indicated antibodies. b, Quality control metrics of the Hi-C data, showing total sequenced read-pairs in millions, total valid contacts in millions and percentages in cis contacts for WT and CKM-Mediator KO ESCs. c, Aggregate analysis of super enhancer interactions in WT and CKM-Mediator KO ESCs. The difference between WT and KO is shown. d, Aggregate analysis of Hi-C signal (10 kb resolution) at pairs of Polycomb domains at the indicated distance ranges in Med13/13lfl/fl (WT) and Med13/13l-/- (CKM-Mediator KO) ESCs, with 200 kb flanking regions. Interactions of inactive non-Polycomb gene promoters subsampled to match regions as in Fig. 1f (n = 2096), are included as a negative control (bottom). The difference between WT and KO is shown. e, Capture-C interaction scores for interactions between Polycomb domains in WT and CKM-Mediator KO ESCs (number of promoters = 51, number of interactions = 148). f, Boxplot analysis of Capture-C interaction scores from WT and CKM-Mediator KO ESCs showing interactions between Polycomb gene promoters and other Polycomb-domains (left), or non-Polycomb gene promoters with active sites (H3K27ac, right). Number of promoters (P) and interactions (int) is shown. Boxes show interquartile range, center line represents median, whiskers extend by 1.5x IQR or the most extreme point (whichever is closer to the median), while notches extend by 1.58x IQR/sqrt(n), giving a roughly 95% confidence interval for comparing medians. Source data

Fig 3: CKM–Mediator has a limited role in 3D genome organization but is essential for Polycomb domain interactions.a, A schematic of Med13/13lfl/fl ESCs. 4-Hydroxytamoxifen (TAM) induces conditional disruption of the CKM–Mediator complex (CKM–MED). b, A representative western blot analysis (n = 6) of nuclear extracts from Med13/13lfl/fl (wild type, WT) and Med13/13l-/- (CKM–MED KO) ESCs showing depletion of MED13 and MED13L proteins. HDAC1 is shown as a loading control. c, Hi-C contact matrices of WT and CKM–MED KO ESCs at 10 kb resolution. Genomic coordinates are indicated. d, Aggregate analysis of TADs and loops in WT and CKM–MED KO ESCs at 10 kb resolution. e, Hi-C contact matrices of WT and CKM–MED KO ESCs at 5 kb resolution. Interactions between Polycomb domains are indicated with a red circle. The blue track shows binding of PRC1 (RING1B ChIP–seq). Genomic coordinates are indicated. f, Aggregate analysis of Hi-C signal (10 kb resolution) at pairs of Polycomb domains in Med13/13lfl/fl (WT) and Med13/13l-/- (CKM–MED KO) ESCs, with 200 kb flanking regions. The difference between WT and KO is shown. g, A snapshot showing Capture-C read count signal in WT and CKM–MED KO ESCs. Interactions between the Nkx2-1 promoter bait (triangle) and surrounding Polycomb-bound sites are shown with arrowheads. PRC1 binding (RING1B ChIP–seq) is shown as a reference. h, Boxplot analysis of mean normalized read counts from WT and CKM–MED KO ESCs showing interactions between Polycomb gene promoters and other Polycomb domains (left), or non-Polycomb gene promoters and active sites (H3K27ac, right). Interactions were not distance-matched due to differences in the interaction ranges for the two promoter types. Boxes show IQRs, center line represents median, whiskers extend by 1.5 × IQR or the most extreme point (whichever is closer to the median), while notches extend by 1.58 x IQR/sqrt(n), giving a roughly 95% confidence interval for comparing medians.Source data

Fig 4: FBXL19-responsive gene promoter-interacting sites are required for gene activation during differentiation (A) A schematic illustrating the workflow of the experiment. (B andD) A screenshot showing the distal interaction site for Fli1 (B) and Hoxb3 (D). To highlight the properties of deleted regions, FBXL19, CDK8 and H3K27ac ChIP-seq as well as accessibility (ATAC-seq) are illustrated for each locus. The regions deleted by CRISPR/CAS9 based editing are illustrated with scissors. Below the deleted region the line plots depict the effect on Capture-C signal when FBXL19 or MED13/13L are removed, or when cells are treated with RA. (C) Relative mRNA-levels for Fli1, Rad23b and Nr2f1 in RA-treated wild-type ESC clones (grey, n = 2) and in ESC clones in which the Fli1 promoter interacting site has been deleted (blue, n = 4) as determined by RT-qPCR. Shown is the mean signal normalized internally to a control gene and to expression in wild-type clones treated by RA. Error bars indicate standard error of mean across four biological and three technical replicates. P-values: Student's t-test. (E) Relative mRNA-levels of Hoxb3, Rad23b and Nr2f1 in RA-treated wild-type ESC clones (grey, n = 4) and in ESC clones in which the Hoxb3 promoter interacting site has been deleted (blue, n = 3) as determined by RT-qPCR. Shown is the mean signal normalized internally to a control gene and to expression in wild-type clones treated by RA. Error bars indicate standard error of mean across three biological and two technical replicates. P-values: Student's t-test.

Fig 5: CKM-Mediator enables gene induction via recruitment of the Mediator complex.a, A schematic illustration of the generation of the T7-MED14 expressing Med13/13lfl/fl ESC line. b, PCR showing amplification of homozygously-tagged T7-Med14 alleles (n = 2). c, A representative Western blot analysis of nuclear extracts from the T7-MED14 Med13/13lfl/fl ESC line, following tamoxifen (TAM) treatment (n = 3). Extract from an untagged ESC line was used as a control. HDAC1 was used as a loading control. d, A representative immunoprecipitation (IP) of endogenously T7-tagged MED14 with T7 antibody using nuclear extracts from Med13/13lfl/f ESCs before (UNT) and after tamoxifen (TAM) treatment (n = 2). The IPs were probed with the indicated antibodies. An IP from an untagged ESC line was performed as a negative control and a Western blot for SUZ12 was included as a control protein that does not interact with Mediator. e, Heatmaps of CDK8 and T7-MED14 ChIPseq signal at Polycomb domains (n = 2097) and H3K27ac peaks (n = 4037), sorted by decreasing CDK8 signal. f, Boxplots showing gene expression change (log2FC) of CKM-Mediator-dependent (n = 631) and CKM-Mediator-independent (n = 2689) RA-induced genes following RA differentiation of WT ESCs. Boxes show interquartile range, center line represents median, whiskers extend by 1.5x IQR or the most extreme point (whichever is closer to the median), while notches extend by 1.58x IQR/sqrt(n), giving a roughly 95% confidence interval for comparing medians. g, Boxplots showing T7-MED14 ChIPseq signal at the TSS (1000 bp) of the different classes of RA-induced gene classes as defined in e in ESCs and RA-induced cells (WT and CKM-Mediator KO). Boxes are defined as in f. Signal is an average from three independent biological experiments. Source data