12  Motifs: common in peaks, or changed between conditions

A motif is a short DNA pattern that a transcription factor prefers to bind. If a motif is found more often in your regions than in other DNA, the factor that binds it may act there. This chapter shows that a motif analysis can ask two different questions, and that the answer depends on what the regions are compared with.

12.1 The question

Which transcription factors could bind my regions? There are two versions of this question:

1. Which motifs are common in my peaks? This describes the regions themselves.
2. Which motifs are more common in the regions that changed than in the regions that did not? This describes the experiment.

The two questions need two different comparisons, and they give different answers.

12.2 How it works

Motifs and their database

A motif is stored as a table of how often each base occurs at each position. JASPAR is an open database of such tables for known transcription factors (Rauluseviciute et al. 2024). Factors of one family, such as MEF2A, MEF2B, MEF2C and MEF2D, have almost the same motif. In a result, several motifs of one family count as one signal, not several.

The comparison set decides the question

A motif test counts how many of your sequences contain the motif, and compares this with a control set.

• Shuffled sequences. The control is your own sequences with their bases shuffled. This keeps the base composition but destroys any motif. The test then asks: which motifs are common in my peaks? AME is a tool that works this way (McLeay and Bailey 2010).
• Other regions of the same experiment. The control is regions that did not change. The test asks: what is special about the regions that changed?

Matching the GC content

Many motifs are rich in G and C. Regions with much GC, such as promoters, contain such motifs by chance. HOMER, a widely used motif tool, warns that “CpG Islands are the single biggest source of sequence content bias in mammalian genomes, and are unfortunately found near transcription start sites, where all the action is!” (HOMER 2026; Heinz et al. 2010). So a good control set has the same GC content as the tested set. HOMER builds its control this way, in bins of GC content (HOMER 2026).

How much sequence to test

Transcription factors bind short sites, but histone marks cover wider regions. For histone regions, HOMER suggests that “500-1000 bp is probably a good idea” (HOMER 2026). The factor sites lie between the nucleosomes that carry the mark, so the window must be wide enough to include them. For ATAC-seq and transcription factors, a short window around the peak summit is enough.

Two tests for changed regions

• SEA compares the gained, or the lost, regions with a control set and counts how many sequences contain each motif (Bailey and Grant 2021).
• monaLisa sorts all tested regions into groups (bins) by their fold change, and tests each group against all the others (Machlab et al. 2022). This shows how a motif follows the change: for example, more common the more a region was lost.

Power

A motif present in 30 % of the changed regions and 20 % of the controls is easy to see in 2,000 regions, but hard to see in 100. Small sets have little power. A weak result in a small set does not mean that no factor binds there.

12.3 In EpiNexus

NoteIn EpiNexus 1.2

The Transcription Factor Analysis page has two parts:

• Motifs in peaks (vs shuffled sequences): AME on the called peaks, against shuffled copies of the same sequences. This answers the first question.
• Motifs in changed regions: for each comparison, SEA on the gained and the lost regions, and on gained against lost (and lost against gained). The control is regions of the same comparison that did not change, matched for GC content in 5 % bins. The window is the region’s middle ± 250 bp for histone regions, and the summit ± 100 bp for transcription factors and ATAC-seq. Below, Along fold change (monaLisa) shows monaLisa’s result.

All tests use the JASPAR 2024 motifs of the genome’s group of species (vertebrates for mouse and human). Results are grouped by transcription-factor family.

TipBest practice

Correct for testing hundreds of motifs at once, and read related motifs together: motifs of one transcription-factor family are one signal (Rauluseviciute et al. 2024). EpiNexus calls a motif significant at q < 0.05 with an enrichment above 1. No publication sets a minimum number of regions: EpiNexus tests sets of 100 or more, marks sets of 50–99 as “low power”, and does not test smaller sets.

12.4 Reading the result

Motifs common in all peaks

For the H3K27ac peaks of the heart dataset (Chapter 14), AME found GC-rich zinc-finger motifs at the top: YY2, ZNF93, ZNF281 and members of the KLF and SP families. These motifs are common at promoters, which are rich in GC. They say little about the heart, and nothing about the experiment.

Motifs in the changed regions

Table 12.1: Motifs in changed regions in the two case studies (EpiNexus SEA results).

Dataset, region set	Regions	Significant motifs	Top families (best motif, share of regions vs control)
H3K27ac, lost	2,267	48	MEF2 (MEF2D, 23 % vs 13 %); ESRR (ESRRB)
H3K27ac, gained	194	1	Hairy-related (HES2, q 0.03)
ATAC-seq, lost	6,561	329	OCT4–SOX2 and SOX (Pou5f1::Sox2, 42 % vs 11 %)
ATAC-seq, gained	834	200	Nuclear receptors (NR1H3, NR2F1, ESRRB)

Figure 12.1 shows the lost H3K27ac regions on the TF Analysis page.

Figure 12.1: Motifs in changed regions on the TF Analysis page of the heart example (run cafc042f), set Lost in aMHC-Cre;Arid1a f/f cKO (2,267 regions). The grey badges list the significant TF families, each with its best motif: “Regulators of differentiation” is the MEF2 family, and “Steroid hormone receptors (NR3)” includes ESRRB. The table shows the first of the 48 significant motifs: the share of regions with the motif, against the GC-matched control. EpiNexus 1.2 interface; not covered by the book’s licence.

Four lessons come from Table 12.1:

• The control changes the answer. Against shuffled sequences, the H3K27ac peaks showed GC-rich promoter motifs. Against unchanged regions of the same GC content, the lost regions showed MEF2.
• Families are one signal. In the H3K27ac lost regions, MEF2D, MEF2B and MEF2A are all significant. They are one finding: a MEF2-family factor may act there.
• Effects differ in size. In the ATAC-seq lost regions, the top motif is almost four times as common as in the control. In the H3K27ac lost regions, MEF2D is less than twice as common. Histone regions are much wider than the site of any one factor, so each motif covers a smaller part of each region.
• Small sets show little. The 194 regions that gained H3K27ac show one weak motif. The 834 regions that gained accessibility show 200. Power depends on the number of regions and on how strong the effect is.

monaLisa tells the same story in more detail. In the ATAC-seq data, the Pou5f1::Sox2 motif is most enriched among the regions that lost the most (Chapter 16). In the H3K27ac data, the MEF2 motifs are most enriched among regions with moderate losses (Chapter 14).

12.5 Pitfalls

WarningPitfall: a motif is not binding

A motif shows which factors could bind. It does not show that they do, or that they cause the change. To show binding, profile the factor itself with ChIP-seq or CUT&RUN.

WarningPitfall: comparing with the wrong control

Peaks compared with shuffled or random DNA show the motifs of active DNA in general, such as promoter motifs. To learn what changed, compare the changed regions with unchanged regions of the same GC content.

WarningPitfall: counting family members as separate findings

Factors of one family share a motif. Five significant SOX motifs are one SOX signal, and the motif alone cannot tell which family member binds.

12.6 Further reading

• Rauluseviciute et al. (2024): JASPAR 2024.
• Heinz et al. (2010) and HOMER (2026): HOMER and its GC-matched backgrounds.
• McLeay and Bailey (2010): motif enrichment with AME.
• Bailey and Grant (2021): SEA.
• Machlab et al. (2022): monaLisa.

Bailey, Timothy L., and Charles E. Grant. 2021. SEA: Simple Enrichment Analysis of Motifs. bioRxiv preprint. https://doi.org/10.1101/2021.08.23.457422.

Heinz, Sven, Christopher Benner, Nathanael Spann, Eric Bertolino, Yin C. Lin, et al. 2010. ‘Simple Combinations of Lineage-Determining Transcription Factors Prime Cis-Regulatory Elements Required for Macrophage and B Cell Identities’. Molecular Cell 38 (4): 576–89. https://doi.org/10.1016/j.molcel.2010.05.004.

HOMER. 2026. Finding Enriched Motifs in Genomic Regions (findMotifsGenome.pl). Http://homer.ucsd.edu/homer/ngs/peakMotifs.html.

Machlab, Dania, Lukas Burger, Charlotte Soneson, Filippo M. Rijli, Dirk Schübeler, and Michael B. Stadler. 2022. ‘monaLisa: An R/Bioconductor Package for Identifying Regulatory Motifs’. Bioinformatics 38 (9): 2624–25. https://doi.org/10.1093/bioinformatics/btac102.

McLeay, Robert C., and Timothy L. Bailey. 2010. ‘Motif Enrichment Analysis: A Unified Framework and an Evaluation on ChIP Data’. BMC Bioinformatics 11: 165. https://doi.org/10.1186/1471-2105-11-165.

Rauluseviciute, Ieva, Rafael Riudavets-Puig, Romain Blanc-Mathieu, Jaime A. Castro-Mondragon, Katalin Ferenc, et al. 2024. ‘JASPAR 2024: 20th Anniversary of the Open-Access Database of Transcription Factor Binding Profiles’. Nucleic Acids Research 52 (D1): D174–82. https://doi.org/10.1093/nar/gkad1059.