DUX / DUXA
DUX4 and its relatives DUXA and DUXB carry two homeodomains in tandem within a single polypeptide, an architecture found only in placental mammals and derived from a retrotransposed copy of an ancestral homeobox gene. The two domains bind side by side, head to head, and together read an eleven-base site, TAATCTAATCA, that is far longer and more specific than a single homeodomain site. These genes are normally expressed only for a few cell divisions after fertilisation, where they help switch on the embryonic genome and a set of endogenous retroviral promoters. Misexpression of DUX4 in skeletal muscle causes facioscapulohumeral muscular dystrophy, and DUX4 rearrangements define a subtype of B-cell acute lymphoblastic leukaemia, so this motif marks one of the better understood pathogenic regulatory programs.
The factors
DUX genes exist only in eutherian mammals 21110847. Each encodes two complete homeodomains in tandem, joined by a short linker, in a protein of a few hundred residues. Human DUX4 is encoded inside the D4Z4 macrosatellite repeat at 4q35, with a nearly identical array at 10q26 and many DUX4L pseudogenes elsewhere 40855454; DUXA and DUXB are single-copy genes. The mouse counterpart is Dux. DUX4 has a C-terminal transactivation domain that binds p300/CBP, so it is a strong activator 26951377; DUXA is transcriptionally inert but binds an overlapping set of sequences and competes with DUX4, and because DUXA is itself a DUX4 target this constitutes a negative feedback loop that limits the window of DUX4 activity 37744032. The ancestral state is a single homeodomain: the platypus sDUX protein has one, binds DNA as a homodimer, and is already able to activate zygotic-genome-activation genes and long terminal repeats 37744032. Expression of the human genes is confined to the cleavage-stage embryo, testis and thymus.
The motif
The site is TAATCTAATCA, about 11 informative positions. The crystal structure of the DUX4 tandem homeodomains bound to DNA explained its shape. The two domains bind head to head, and the linker between them makes anchoring minor-groove contacts plus protein-protein contacts that fix their relative positions, so the spacing between the two half sites is not variable the way it is for two independent homeodomain monomers. The surprise from that structure is that the two homeodomains, although tandem duplicates of each other, do not read the same core sequence: an arginine-to-glutamate substitution unique to primates repositions a key arginine side chain in one recognition helix 30540931. That is why the consensus is not a clean palindrome but an asymmetric TAATC-TAATCA. The same divergence explains why mouse DUX and human DUX4 have different motifs, and the difference tracks with which retrotransposon promoters each one can activate 28459454. In the assembled data, DUX4 and DUXA models are similar but not identical: DUX4 models read TAATTTAATCA or TGATTAGGTTA, DUXA models TGATTAAATTA or TGATTTAATCA, with within-family similarities from 0.84 to 1.0 and overall cohesion of 0.89. One low-information entry, a 5.8-bit CIS-BP DUXA model reading only TGATTG, is flagged in the data as belonging with Q50 homeodomains instead: a six-base TGATTA fragment is simply a generic paired-like homeodomain half site and carries no double-homeodomain information.
Where it acts
DUX4 sites are overwhelmingly repeat-resident. DUX4 occupies MERVL/HERVL long terminal repeats, ERVL-MaLR elements and other retrotransposon promoters, along with conventional promoters of cleavage-stage genes including ZSCAN4, TRIM43, MBD3L family members, PRAMEF family members, KHDC1L and LEUTX 22209328,28459457. Because the motif sits inside repeat families, the number of genomic matches is large, in the tens of thousands, and DUX4 ChIP-seq peaks are heavily biased toward repeats 22209328,26951377. Where DUX4 binds it recruits p300/CBP and deposits H3K27 acetylation, so the sites behave as bona fide enhancers and promoters rather than as passive footprints 26951377. In adult somatic tissue these sequences are silent and the motif is not occupied, because the factors are not expressed.
Biology and disease
Facioscapulohumeral muscular dystrophy is the clearest example of a disease caused by a single transcription factor being switched on in the wrong tissue. In FSHD1, contraction of the D4Z4 array to between one and ten units on a permissive 4qA haplotype, which supplies a polyadenylation signal downstream of the last repeat, allows a stable DUX4 transcript to accumulate in muscle 20724583. In FSHD2 the same end is reached through mutations in chromatin repressors such as SMCHD1 23143600. DUX4 expressed in myoblasts activates germline and cleavage-stage genes, retroelements and immune mediators, and is cytotoxic 22209328,37744032. In normal development, mouse Dux and human DUX4 activate cleavage-stage genes and MERVL/HERVL elements at zygotic genome activation 28459457,28459456, and DUX4 also has roles in priming the human embryonic genome beyond transcriptional activation 35402882. Rearrangements that place DUX4 under the IGH enhancer, and related DUX4 fusions, define a distinct subtype of B-cell acute lymphoblastic leukaemia in adolescents and young adults; the retained portion is the double homeodomain 27019113.
Reading a hit
An 11-base double-homeodomain match is a distinctive shape, but three cautions apply. First, it does not tell you which DUX protein. DUX4 and DUXA have overlapping specificity and opposite activity, so the same site can be an activated enhancer or a competitively blocked one. Second, half-site-only matches, TAATCA or TGATTA, fall straight back into the large generic paired-like homeodomain pool and are not DUX evidence. Third, one JASPAR entry in this family is the mouse Dux motif, which is genuinely different from the human one, and the JASPAR DUX4 model and one HOCOMOCO DUX4 model are ChIP-derived from overexpression systems rather than from in vitro selection on the purified protein 22209328. Finally, a match in an ERVL or MaLR repeat is expected by composition and says little on its own.
How this was worked out
Sequencing the partially deleted D4Z4 locus of an FSHD patient identifies an open reading frame with two homeoboxes inside each 3.3-kb repeat unit, the gene later named DUX4.
Still open then: The repeat was widely regarded as non-coding, and there was no evidence that the predicted gene was transcribed or that its product bound DNA.
- 2010highhaplotype and SNP analysis of patient chromosomes plus transfection assays of transcript stability20724583
A unifying genetic model shows that repeat contraction causes disease only on a 4qA haplotype carrying a polymorphism that creates a polyadenylation signal distal to the last repeat, stabilizing the DUX4 transcript, which explains why contraction alone is not sufficient.
Still open then: It remained unclear what DUX4 does to muscle once the transcript is stabilized.
- 2012mediuminducible expression in myoblasts with ChIP-seq and expression profiling of patient muscle22209328
DUX4 is shown to activate germline and early stem-cell genes, MaLR retrotransposon long terminal repeats and innate-immune mediators, and these targets are detectable in FSHD muscle but not controls, giving direct support for the misexpression model.
Still open then: Which of these induced programs is responsible for muscle damage was not resolved.
FSHD2, which occurs without repeat contraction, is shown to arise from mutations in the chromatin repressor SMCHD1 inherited together with a permissive D4Z4 allele, bringing both forms of the disease under a single mechanism of DUX4 derepression.
Insertion of D4Z4 repeats containing DUX4 into the IGH locus is found recurrently in Philadelphia-negative B-cell acute lymphoblastic leukemia of adolescents and young adults, and DUX4-IGH expression in pro-B cells generates leukemia in mice.
- 2017highexpression profiling, ChIP-seq and knockout or knockdown in embryonic stem cells and embryos28459457 28459456 28459454
Three studies establish mouse Dux and human DUX4 as activators of the cleavage-stage program and of MERVL/HERVL elements at zygotic genome activation, and show that the two orthologs have diverged in sequence preference, with motif divergence tracking which retrotransposon promoters each can activate.
Still open then: Whether human DUX4 is required for zygotic genome activation in human embryos, rather than merely sufficient in cultured cells, was untested.
The crystal structure of the DUX4 tandem homeodomains on DNA shows head-to-head binding with the linker making minor-groove and protein-protein contacts, and reveals that the two duplicated homeodomains read different core sequences because of a primate-specific arginine-to-glutamate change that repositions an arginine in one recognition helix.
Still open then: Explains the asymmetric site but not how DUX4 and DUXA, which bind overlapping sequences, are partitioned in cells.
Comparison of the three human DUX proteins with the single-homeodomain platypus sDUX shows that cytotoxicity is a property of DUX4 and of ancestral sDUX but not of DUXA or DUXB, and that DUXA, itself a DUX4 target, competes for DUX4 sites and antagonizes DUX4 in FSHD patient cells.
Still open then: The relative occupancy of DUX4 and DUXA at shared sites in intact tissue has not been measured.
Still unresolved
- What makes DUX4 cytotoxic is unsettled: in human myoblasts the C-terminal p300/CBP-binding region is required for both target activation and cell death, yet the ancestral single-homeodomain platypus sDUX protein is also cytotoxic, so toxicity may not depend on the double-homeodomain architecture or on the primate C-terminus.
- Whether human DUX4 is required for embryonic genome activation or only contributes to it is disputed: mouse Dux depletion impairs zygotic genome activation and the 2-cell-like state, whereas DUX4 knockdown in human zygotes altered the activation transcriptome without terminating the embryos, which fits a priming role rather than a switch.
- Nothing in the sequence distinguishes a DUX4-occupied site from a DUXA-occupied one, and because the two proteins have opposite transcriptional output, how occupancy is partitioned between them in cells that express both is unknown.
- The usability of the mouse Dux motif for human sequence is contested: motif collections list it alongside human DUX4 models, while the comparative work shows that the primate-specific change in one recognition helix gives the two orthologs genuinely different preferences and different retrotransposon targets.
The motif shape is good but the label is not resolvable: DUX4, DUXA and DUXB bind overlapping sequences, and DUXA represses what DUX4 activates, so a hit does not predict direction of effect. One JASPAR entry here is mouse Dux, whose motif differs from human DUX4 because of a primate-specific change in one recognition helix; do not transfer it to human sequence. The best DUX4 models are ChIP-derived from forced expression, not from in vitro selection. Because the site lives inside MERVL/HERVL and MaLR repeats, matches are numerous and largely explained by repeat content; outside the cleavage-stage embryo, testis, thymus, FSHD muscle or DUX4-rearranged leukaemia, no DUX protein is present to occupy them.
References
- Lee JK, Bosnakovski D, Toso EA, et al. Crystal Structure of the Double Homeodomain of DUX4 in Complex with DNA. Cell Rep 2018;25:2955-2962.e3. PMID 30540931
- Lemmers RJ, van der Vliet PJ, Klooster R, et al. A unifying genetic model for facioscapulohumeral muscular dystrophy. Science 2010;329:1650-3. PMID 20724583
- Geng LN, Yao Z, Snider L, et al. DUX4 activates germline genes, retroelements, and immune mediators: implications for facioscapulohumeral dystrophy. Dev Cell 2012;22:38-51. PMID 22209328
- Hendrickson PG, Doráis JA, Grow EJ, et al. Conserved roles of mouse DUX and human DUX4 in activating cleavage-stage genes and MERVL/HERVL retrotransposons. Nat Genet 2017;49:925-934. PMID 28459457
- De Iaco A, Planet E, Coluccio A, et al. DUX-family transcription factors regulate zygotic genome activation in placental mammals. Nat Genet 2017;49:941-945. PMID 28459456
- Whiddon JL, Langford AT, Wong CJ, et al. Conservation and innovation in the DUX4-family gene network. Nat Genet 2017;49:935-940. PMID 28459454
- Bosnakovski D, Toso EA, Ener ET, et al. Antagonism among DUX family members evolved from an ancestral toxic single homeodomain protein. iScience 2023;26:107823. PMID 37744032
- Yasuda T, Tsuzuki S, Kawazu M, et al. Recurrent DUX4 fusions in B cell acute lymphoblastic leukemia of adolescents and young adults. Nat Genet 2016;48:569-74. PMID 27019113
- Leidenroth A, Hewitt JE. A family history of DUX4: phylogenetic analysis of DUXA, B, C and Duxbl reveals the ancestral DUX gene. BMC Evol Biol 2010;10:364. PMID 21110847
- Vuoristo S, Bhagat S, Hydén-Granskog C, et al. DUX4 is a multifunctional factor priming human embryonic genome activation. iScience 2022;25:104137. PMID 35402882
- Belayew A, Rosa AL, Zammit PS. DUX4 at 25: how it emerged from "junk DNA" to become the cause of facioscapulohumeral muscular dystrophy. Skelet Muscle 2025;15:24. PMID 40855454
- Gabriëls J, Beckers MC, Ding H, et al. Nucleotide sequence of the partially deleted D4Z4 locus in a patient with FSHD identifies a putative gene within each 3.3 kb element. Gene 1999;236:25-32. PMID 10433963
- Lemmers RJ, Tawil R, Petek LM, et al. Digenic inheritance of an SMCHD1 mutation and an FSHD-permissive D4Z4 allele causes facioscapulohumeral muscular dystrophy type 2. Nat Genet 2012;44:1370-4. PMID 23143600
- Choi SH, Gearhart MD, Cui Z, et al. DUX4 recruits p300/CBP through its C-terminus and induces global H3K27 acetylation changes. Nucleic Acids Res 2016;44:5161-73. PMID 26951377
8 motifs in this family — JASPAR 3 · HOCOMOCO 2 · CIS-BP 2 · Vierstra 1
| motif | database | consensus | bits |
|---|---|---|---|
| DUX4.H14CORE.0.P.BPB | HOCOMOCO | TGATTAGgTTA | 18.4 |
| Dux (MA0611.3) | JASPAR | TGATTbAATCA | 16.3 |
| DUXA.H14CORE.0.SM.BSMB | HOCOMOCO | TGATTAAaTTA | 16.1 |
| DUXA (MA0884.2) | JASPAR | TgATTtAATcA | 15.4 |
| DUX4 (MA0468.1) | JASPAR | TAAtttAATCA | 15 |
| M08135_2.00 | CIS-BP | TAAtttAATCA | 15 |
| AC0009:DUXA/DUX:Homeodomain | Vierstra | TAAtcTAATCA | 14.1 |
| M01247_2.00looks like HD_Q50 | CIS-BP | TGATTg | 5.8 |
HOCOMOCO tags read as assay + quality grade: P ChIP-seq, S HT-SELEX, M methyl-HT-SELEX, G GHT-SELEX, I SMiLE-seq, B PBM; grade A means two or more assay types agreed. A P-only matrix is a record of where the protein was crosslinked, which need not be a sequence it binds itself.
The factors, elsewhere
These entries are written from the primary literature and every citation is checked against PubMed, but this one can still be out of date, too confident, or simply wrong — especially where a motif’s published site and its database matrices disagree. Corrections and additions are welcome, and they are the fastest way to make this page better for everyone.