13  Super-enhancers and enhancer–gene links

Enhancers are DNA regions that raise the activity of genes, often from far away. This chapter covers two analyses of enhancers: finding the strongest ones, called super-enhancers, and predicting which genes each enhancer controls, with the Activity-by-Contact (ABC) model.

13.1 The question

Which enhancers are the strongest in my cells? And which genes does each enhancer act on?

13.2 Super-enhancers

How it works

Whyte and colleagues found that some enhancers in embryonic stem cells are much larger and carry much more signal than all others (Whyte et al. 2013). They called them super-enhancers. The method, often called ROSE, has three steps:

1. Join nearby enhancers. “We allowed regions within 12.5kb of one another to be stitched together” (Whyte et al. 2013).
2. Rank the joined regions by their signal, from weakest to strongest.
3. Find the bend. Plotted against their rank, the signals form a flat curve that rises steeply at the end: a hockey stick (Figure 14.2). In the original study, the enhancers “fell into two classes” at this point (Whyte et al. 2013). Regions above the bend are super-enhancers.

The original study used Mediator, a coactivator, to measure the signal. Later studies often use H3K27ac. Super-enhancers often lie at genes that define the cell type: in more differentiated cells, they “are also found at genes that define cell identity” (Whyte et al. 2013).

NoteIn EpiNexus 1.2

EpiNexus finds super-enhancers for each sample of H3K27ac, MED1 or BRD4, the marks used for them. It joins peaks within 12.5 kb, leaves out peaks within 2.5 kb of a gene start (promoters), ranks the joined regions by their signal and calls the regions above the bend. The Super-Enhancers page shows the hockey-stick plot and a table of regions with their nearest gene. For other marks and for ATAC-seq, the step is skipped and the run says why.

Reading the result

In the heart dataset (Chapter 14), each H3K27ac library has 715 to 977 super-enhancers among 13,171 to 16,254 joined regions (Table 14.4, Figure 14.2). The strongest lie near Gata4, a key transcription factor of the heart, and Rbm20, which controls the splicing of heart muscle genes. So, as in the original study, the strongest enhancers point to the identity of the cells. The Super-Enhancers page shows each library’s curve and counts (Figure 13.1).

Figure 13.1: The Super-Enhancers page for the control library H3K27Ac_Control_rep1 (run cafc042f): 14,292 joined regions, of which 767 lie beyond the bend (orange, right of the dashed cut-off) and are called super-enhancers. EpiNexus 1.2 interface; not covered by the book’s licence.

WarningPitfall: comparing super-enhancer counts between conditions

The two knockout libraries of the heart dataset have 715 and 977 super-enhancers. The difference between replicates is larger than the difference between conditions. The number of super-enhancers depends on depth and on where the bend falls. To compare conditions, use the differential analysis (Chapter 9).

13.3 Enhancer–gene links: the ABC model

How it works

An enhancer acts on a gene if it is active and if it touches the gene’s promoter in the folded DNA. The ABC model scores each pair of enhancer and gene as activity × contact, divided by the sum of the same product over all elements near the gene (Fulco et al. 2019). So the score is the share of the gene’s regulatory input that comes from this enhancer.

• Activity is measured with chromatin accessibility (ATAC-seq or DNase-seq) and, ideally, H3K27ac. ABC “minimally requires: (i) a measure of chromatin accessibility (DHS or ATAC-seq)” (Fulco et al. 2019).
• Contact is best measured with Hi-C. Without Hi-C, it is estimated from the distance: the further apart, the less contact (Fulco et al. 2019).

A link is kept when its score passes a threshold. Published thresholds, such as 0.015 (Nasser et al. 2021), were set on human cells, with activity scaled to a reference cell line.

How good are such links? A large test against CRISPR experiments in human cells compared several versions (Gschwind et al. 2026). Of the ABC versions, the one with DNase-seq found true enhancer–gene pairs best. ABC with H3K27ac alone did only a little better than distance alone. In that test, each gene had a median of three predicted enhancers (Gschwind et al. 2026).

TipBest practice

Use a threshold calibrated for your kind of data. Published ABC thresholds were set against CRISPR tests, with activity scaled to a reference human cell line (Nasser et al. 2021). EpiNexus does not scale activity that way, so no published threshold fits. It keeps links with a score above 0.02, its own choice.

NoteIn EpiNexus 1.2

EpiNexus follows the rules of the reference ABC code: candidate elements are the peaks plus every gene’s start site, scores are normalised over all elements within 5 Mb, contact comes from a distance power law (no Hi-C), and each gene’s own promoter counts as fully linked. Activity is ATAC-seq or DNase-seq accessibility, or H3K27ac. From EpiNexus 1.2.3, samples of other marks are not linked, and the run says why. The Peak-Gene Links page lists the links with their scores and distances.

Reading the result

Table 13.1: Enhancer–gene links of one library in each case study (runs cafc042f and 1e7862c4).

	H3K27ac, heart (control rep 1)	ATAC-seq, stem cells (untreated rep 1)
Links to distal elements	113,797	155,442
Genes with at least one link	32,416	37,608
Links per gene, median	3	4
Distance from gene start, median	79 kb	22 kb
Links more than 100 kb away	45 %	13 %

Both runs give a median of three or four enhancers per gene, close to the large test above (Table 13.1). But the H3K27ac links reach much further than the ATAC-seq links. One reason may be the elements: for ATAC-seq, EpiNexus uses 500 bp around each summit; for H3K27ac, whole peaks, which can be several kilobases wide. Fulco and colleagues also warn that H3K27ac alone can mark regions “close to strong enhancers but [that] do not themselves have enhancer activity, due to the spreading H3K27ac signal” (Fulco et al. 2019).

The Peak-Gene Links page shows these totals for every library, and says when the links rest on H3K27ac alone (Figure 13.2).

Figure 13.2: The Peak-Gene Links page of the heart example (run cafc042f). The text at the top says how the links were made. The yellow box warns that with H3K27ac as the only activity measure the links are “ABC-style”. The table Per sample gives each library’s links, the genes with links, and the genes scored. EpiNexus 1.2 interface; not covered by the book’s licence.

Chapter 14 follows one link, near the gene Atp1a2, from the control to the knockout.

WarningPitfall: reading links as proof

An ABC link is a prediction. With H3K27ac alone, it is a rough one. Use the links to choose which enhancers to test, for example with CRISPR, not as the result.

13.4 Further reading

• Whyte et al. (2013): super-enhancers and cell identity.
• Fulco et al. (2019): the ABC model.
• Nasser et al. (2021) and Gschwind et al. (2026): ABC maps across many cell types, and how well they work.

Fulco, Charles P., Joseph Nasser, Thouis R. Jones, Glen Munson, Drew T. Bergman, et al. 2019. ‘Activity-by-Contact Model of Enhancer–Promoter Regulation from Thousands of CRISPR Perturbations’. Nature Genetics 51 (12): 1664–69. https://doi.org/10.1038/s41588-019-0538-0.

Gschwind, Andreas R., Kristy S. Mualim, Alireza Karbalayghareh, Maya U. Sheth, Kushal K. Dey, et al. 2026. ‘An Encyclopedia of Human Enhancer–Gene Regulatory Interactions’. Nature 657 (8130): 179–90. https://doi.org/10.1038/s41586-026-10781-4.

Nasser, Joseph, Drew T. Bergman, Charles P. Fulco, Philine Guckelberger, Benjamin R. Doughty, et al. 2021. ‘Genome-Wide Enhancer Maps Link Risk Variants to Disease Genes’. Nature 593 (7858): 238–43. https://doi.org/10.1038/s41586-021-03446-x.

Whyte, Warren A., David A. Orlando, Denes Hnisz, Brian J. Abraham, Charles Y. Lin, et al. 2013. ‘Master Transcription Factors and Mediator Establish Super-Enhancers at Key Cell Identity Genes’. Cell 153 (2): 307–19. https://doi.org/10.1016/j.cell.2013.03.035.