Long-Range Enhancer Interactions Are Prevalent in Mouse Embryonic Stem Cells and Are Reorganized upon Pluripotent State Transition

Abstract

Transcriptional enhancers, including super-enhancers (SEs), form physical interactions with promoters to regulate cell-type-specific gene expression. SEs are characterized by high transcription factor occupancy and large domains of active chromatin, and they are commonly assigned to target promoters using computational predictions. How promoter-SE interactions change upon cell state transitions, and whether transcription factors maintain SE interactions, have not been reported. Here, we used promoter-capture Hi-C to identify promoters that interact with SEs in mouse embryonic stem cells (ESCs). We found that SEs form complex, spatial networks in which individual SEs contact multiple promoters, and a rewiring of promoter-SE interactions occurs between pluripotent states. We also show that long-range promoter-SE interactions are more prevalent in ESCs than in epiblast stem cells (EpiSCs) or Nanog-deficient ESCs. We conclude that SEs form cell-type-specific interaction networks that are partly dependent on core transcription factors, thereby providing insights into the gene regulatory organization of pluripotent cells.

Keywords: chromatin looping; differentiation; epigenetics; gene regulation; genome organization; pluripotency; promoter capture Hi-C.

Graphical abstract

Figure 1

Promoter-Capture Hi-C Reveals that SEs Form Spatial Networks in Mouse ESCs (A) From the 231 previously defined SEs in ESCs (Whyte et al., 2013), 80 overlapped baited Hi-C fragments and were not considered further in this analysis. Significant promoter-SE interactions were identified at 138 of the remaining SEs, and a comparison of SE target genes between PCHi-C and computational assignments is shown. (B) Example of a computationally predicted SE-target gene (Dppa3) without significant PCHi-C interactions (top). Examples of a newly identified SE-target gene (Sox15; middle) and a computationally predicted SE-target gene that was confirmed by PCHi-C (Klf4; bottom). Tracks show HindIII sites (black lines), baited HindIII sites (purple lines), ESC SEs (maroon), and ESC OSN (OCT4, SOX2, and NANOG) binding sites (teal). Arcs show significant interactions. Red and gray vertical bars highlight the location of SEs and promoters, respectively. The upper genome browser screenshot also exemplifies SE promoter skipping, where the SE close to Phc1 contacts the Nanog promoter, skipping over six genes. (C) Stacked bar chart showing the number of SEs interacting with 1, 2–5, 6–10, or >11 gene promoters. (D) Stacked bar chart showing the number of gene promoters interacting with 1 or 2–5 SEs. (E) Schematic proposing an updated model of SE interactions from a binary promoter-SE interaction (i) to a more complex network where SEs can contact multiple promoters, and promoters can interact with more than one SE (ii). Red and gray rectangles represent SEs and genes, respectively. See also Figure S1 and Table S1.

Figure 2

Pluripotent State-Specific Wiring of SE Interactions (A) The promoters in close physical proximity to ESC SEs were largely maintained in EpiSCs, but subsets of significant SE interactions were detected uniquely in each cell type. (B–G) Differences in the SE interactome between ESCs and EpiSCs are shown for the Gata1 (B), Zfp281 (C), Klf4 (D), Nanog (E), Nodal (F), and Klf13 (G) loci. Arcs indicate significant interactions colored by cell-type specificity (ESC only, green; EpiSC only, blue; shared, gray). (H) Boxplot showing the log2 fold change in gene expression between EpiSCs and ESCs, for genes interacting with SEs only in ESCs, in ESCs and EpiSCs, and only in EpiSCs. ^∗∗p < 0.01 (Mann-Whitney U test). (I) Model illustrating the rewiring of interactions between pluripotent states at an individual SE. The red arrows represent the physical contacts between promoters detected by PCHi-C, creating a spatial hub for potential SE regulation (green/blue circles). Upon transitioning to EpiSCs, some contacts are lost (gray arrow), while new ones are established (green arrow). See also Figure S2 and Table S1.

Figure 3

Long-Range Promoter-SE Interactions Are Prevalent in ESCs (A) After ROSE analysis, we selected three enhancer categories: the top 250 H3K27ac-ranked regions (comprising a subset of SEs), the first 250 enhancers immediately below the SE-threshold (middle), and 250 enhancers with low levels of H3K27ac. Red boxes highlight the selected enhancers within each category. (B) Proportion of PCHi-C interactions detected for the three enhancer categories and control regions in ESCs (left) and in EpiSCs (right). Control regions were randomly selected and size-matched to SEs. (C) Number of interactions after concatenation of the fragments spanning an individual SE, normal enhancer (NE), low enhancer (LE), or randomly selected control regions (CTRL) in ESCs and EpiSCs. (D) Expression (log2 reads per kilobase per million mapped reads [RPKM]) of genes interacting with the three classes of enhancers and control regions in ESCs and EpiSCs. ^∗∗∗∗p < 0.0001; ^∗∗p < 0.01 (Mann Whitney U test). (E) ESCs have a high proportion (25%) of promoter-SE interactions spanning >800 kb (red) when compared to interactions in EpiSCs or to size matched NEs and control regions (CTRL; Benjamini-Hochberg adjusted chi-square p < 0.0001). (F) A significant proportion of LRIs between promoters and SEs in ESCs (green, 25%) compared to EpiSCs (blue, 7%) or to size- and distance-match random control regions (light green, 13%; Benjamini-Hochberg adjusted p < 0.0001). NE and LE are not significantly different from controls. See Table S2 for a list of LRIs. (G) Classification of concatenated SE fragments according to the genomic distance of their target promoters. In ESCs, 58% of interacting SE regions only engage in short-range (<800 kb), 26% in long-range only (>800 kb), and 16% in both short- and long-range promoter interactions. In EpiSCs, most interacting SEs (89%) engage in short-range interactions only. Note that the proportion of interactive SE regions is likely to be underestimated due to the stringent statistical analysis applied to the concatenated regions, and also due to the inability to detect interactions when the baited promoter and SE are within neighboring HindIII fragments. See also Table S2.

Figure 4

Long-Range Promoter-SE Interactions Are Enriched for NANOG Binding (A) CODEX database was integrated to screen for the binding of transcription factors (TFs) at the promoter of genes involved in long- and short-range SE interactions in ESCs. The heatmap shows the top 25 TFs bound at promoters engaged in LRIs with SEs, normalized by column. The side panel shows the Benjamini-Hochberg adjusted chi-square p values for the proportion of long- versus short-range interactions bound/not bound by NANOG, POU5F1, SOX2, and ESRRB. (B) Either end of the promoter-SE/CTRL contacts was classified according to the binding of NANOG. The majority (68%) of promoter-SE interactions in ESCs were bound by NANOG, particularly at the SE end (56%). The percentage of interactions within each category is shown. OE, other end of interaction (either SE or matched control). ^∗∗p < 0.01; ^∗p < 0.05 (Mann Whitney U test). (C) LRIs were significantly enriched for NANOG occupancy when compared to short-range (<800 kb) interactions in ESCs (green dots, Benjamini-Hochberg adjusted chi-square p = 2E-18; Figure S3A). This enrichment was moderately significant in EpiSCs (blue dots, Benjamini-Hochberg adjusted chi-square p = 8E-03) and not significant at control regions (gray dots). (D) Genome browser illustrating an example of LRIs between promoters and SEs in ESCs (green dashed arcs) and in EpiSCs (blue dashed arcs). (E) Gene expression levels (log2 RPKM) of genes engaged in long- and short-range SE interactions in ESCs (Mann-Whitney U test; p = 0.01). See also Figure S3 and Tables S2 and S3.

Figure 5

Long-Range Promoter-SE Interactions Are Depleted in Nanog-Deficient ESCs (A) Comparison of the SE-interactome between WT and Nanog^−/− ESCs. The Venn diagram shows the overlap of genes that significantly interact with the 231 SEs (Whyte et al., 2013) in wild-type (WT) ESCs and Nanog^−/− ESCs. A similar overlap was detected for ROSE-called WT SEs (not shown). (B) Examples of significant promoter-SE interactions in Nanog^−/− ESCs (top, Elf3) and in WT ESCs (bottom, Gata1). Significant interactions common to WT and Nanog^−/− ESCs are shown as gray arcs, while orange or green arcs represent interactions detected only in the knockout or in the WT ESCs. (C) The proportion of long-range promoter-SE interactions was significantly reduced in Nanog^−/− ESCs compared to WT ESCs (57 versus 330 LRIs, respectively; Benjamini-Hochberg adjusted chi-square p < 0.0001). (D) The proportion of long-range versus short-range promoter-SE interactions in WT ESCs is significantly higher compared to Nanog^−/− ESCs (Benjamini-Hochberg adjusted chi-square p < 1E-05). No significance in the proportion of long- versus short-range was detected between promoter-SE and promoter-CTRL in Nanog^−/− ESCs. (E) Promoter-SE interactions at a large region encompassing the Nanog locus. Grey arcs show promoter-SE interactions common to both WT and Nanog^−/− ESCs, and orange and green arcs represent interactions detected in each, respectively. Dashed arcs indicate LRIs. Vertical gray lines denote SE interactions at sites bound by the pluripotency transcription factors OCT4, SOX2, and NANOG (OSN). See also Figure S3 and Tables S1 and S3.