Oncofusion

start / graph / fusion-dux4-igh

Fusionstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q11-comparative

DUX4::IGH (IGH-DUX4), t(4;14)/ins(14;4) — D4Z4 repeat insertion into the IGH locus

also called IGH-DUX4 · DUX4-IGH · DUX4-r ALL · DUX4-rearranged B-ALL · DUX4/ERG-rearranged subtype · IGH::DUX4

A gene that is normally switched off in every cell except the early embryo and testis gets physically dropped into the antibody gene factory of a young B cell, where the cell's most powerful gene switch is permanently on. That switch drives a truncated version of the embryonic gene, which then wrecks a second gene called ERG - and, unusually for such a strange accident, children with this leukaemia do well.

The same thing, for a scientist

This lesion is **enhancer hijacking in its purest form**. **D4Z4 macrosatellite repeats containing DUX4** (Q9UBX2) are inserted into the **IGH locus**, placing a **truncated DUX4** coding sequence under the control of the immunoglobulin heavy-chain enhancers — the strongest enhancer module a B-lineage cell possesses — producing high-level expression of DUX4 protein **with an aberrant C terminus**; DUX4-IGH expressed in pro-B cells generates B-cell leukaemia in mice [E1]. DUX4 is normally silent outside germline and early embryo, so this is not corruption of a resident programme but the **forced expression of an embryonic transcription factor in a B progenitor**. The second half of the mechanism is the **DUX4-ERG axis**: DUX4 rearrangement and overexpression is present in all cases and is accompanied by deregulation of **ERG**, including expression of a novel isoform **ERGalt** whose non-canonical first exon is **transcriptionally initiated by DUX4 binding**; ERGalt retains ERG's DNA-binding and transactivation domains but **inhibits wild-type ERG activity and is itself transforming**, and frequent intragenic **ERG deletion** achieves the same loss of ERG function by a different route [E2][E3]. The authors call this a **unique paradigm: DUX4 deregulation causes loss of ERG function, either by deletion or by induced expression of a dominant-negative isoform** [E3]. The subtype comprises up to **7% of B-ALL** [E3] (4% in a consecutive paediatric series [E2]), is enriched in **adolescents and young adults** [E1], is almost invariably **CD371-positive** (27/28 DUX4-fusion cases) with a transient myelomonocytic switch and slow early treatment response, yet has a **good outcome** (5-year EFS 88.3% in the CD371-positive group) [E4].

What it looks like
loading 6E8C…
PDB 6E8C ↗
shape

This is real experimental data — atoms placed by crystallography or cryo-EM, deposited by the people who did the experiment.

I have never read one of these pictures before

A protein is one long chain of amino acids that folds up. The ribbon traces that chain from one end to the other; it is not a wire, it is a simplification of where the backbone runs. Drag to rotate, scroll to zoom.

Domains are the stretches of chain that do a particular job, and they are what the coloured blocks above the picture mark out. Click one and the view flies to it. In this project the important question about any domain is usually whether the fusion still has it — the labels say lost where it does not.

Anything in yellow sticks is a bound small molecule, which is usually the drug.

Go deeper
What is kept, what is lost, and why 'fusion' is almost the wrong word1 deeper

This node is a deliberate outlier in the beat, and the differences are the point.

The mechanism is not a chimeric protein in the usual sense. In ETV6::RUNX1, BCR::ABL1, PML::RARA and the rest, two coding sequences are joined and the protein is the oncogene. Here, the pathogenic event is relocation of a gene next to an enhancer. What matters is the regulatory context, not a novel protein-protein juxtaposition.

Kept / gained from DUX4 (the relocated gene): the double homeodomain — two tandem homeodomains that bind DNA cooperatively (the only solved structure relevant to this node: PDB 6E8C, the DUX4 double homeodomain in complex with DNA) 5 — plus most of the protein, expressed at high level with an aberrant C terminus because the insertion truncates the coding sequence 1.

Lost from DUX4: the native C-terminal transactivation domain is replaced or truncated. That truncation is not incidental: the aberrant C-terminus is a consistent feature of the leukaemic allele 1, and truncated DUX4 is what is oncogenic here, whereas full-length DUX4 is strongly cytotoxic (which is the basis of its pathology in facioscapulohumeral muscular dystrophy).

Contributed by IGH: nothing coding. IGH supplies the enhancers — Emu and the 3-prime regulatory region — which in a B progenitor are constitutively and powerfully active. No IGH protein sequence is part of the oncoprotein.

So the "5-prime / 3-prime partner" framing breaks down. The gene5/gene3 fields in this node record DUX4 and IGH for dashboard consistency, but the honest description is: DUX4 is the cargo, IGH is the promoter/enhancer, and there is no fusion protein interface at all.

Why this matters for the beat. Enhancer hijacking is mechanistically distinct from every other fusion here in one decisive respect: there is no fusion-specific protein surface to target. The protein present in the cell is (truncated) DUX4, essentially a normal embryonic transcription factor in the wrong place at the wrong level. Even the most optimistic structure-based programme has nothing fusion-specific to bind. The comparison table:

typical fusion in this beatDUX4::IGH
pathogenic entitya novel chimeric proteina normal-ish protein at abnormal level in the wrong cell
what the partner contributesa protein module (kinase, corepressor, DBD, IDR)an enhancer, no protein
fusion-specific protein surfaceyes (the junction, the composite)none
therapeutic implicationin principle drug the compositemust target the programme, the enhancer, or the dependency
The IGH locus as a recurrent hijacking enhancer

IGH is the archetypal hijacking enhancer across B-lineage malignancy — MYC in Burkitt lymphoma, BCL2 in follicular lymphoma, CCND1 in mantle cell lymphoma, CRLF2 and EPOR in Ph-like B-ALL, and DUX4 here. The reason is mechanical: the machinery that deliberately breaks and rejoins DNA at IGH during V(D)J recombination and class-switch recombination is an error-prone, physiological double-strand-break generator in exactly the cell type where the IGH enhancers are maximally active. A B progenitor is therefore structurally predisposed to deliver any nearby sequence into a high-expression regulatory environment.

That gives a general comparative prediction for the Queen: enhancer-hijack fusions should be over-represented in lineages with programmed physiological DNA breakage (B and T lymphocytes) and under-represented in myeloid disease, where no analogous recombinase operates. The eleven fusions in this beat are consistent with that: the only enhancer-hijack lesion in the set is the one in a B progenitor at the IGH locus.

Note the contrast with KMT2A, where the recurrent breakage is attributed to topoisomerase-II cleavage and apoptotic nuclease activity rather than to a programmed recombinase, and where the product is a genuine chimeric protein rather than a relocated gene.

The DUX4-ERG axis (PI depth request)

The second, and arguably more interesting, half of the mechanism is what DUX4 does to ERG, and it is a genuinely novel paradigm.

The observations 3:

  1. DUX4 rearrangement and overexpression is present in ALL cases of the subtype.
  2. It is accompanied by transcriptional deregulation of ERG.
  3. A novel ERG isoform, ERGalt, is expressed. ERGalt uses a non-canonical first exon whose transcription is initiated by DUX4 binding — i.e. DUX4, acting as a transcription factor, creates a new promoter for ERG.
  4. ERGalt retains ERG's DNA-binding and transactivation domains but inhibits wild-type ERG transcriptional activity and is transforming. It is a dominant-negative that is simultaneously an oncogene.
  5. Frequent intragenic ERG deletion occurs in the same subtype 23, achieving loss of ERG function by an entirely different route.

The synthesis, in the authors' own framing: this 'illustrate[s] a unique paradigm of transcription factor deregulation in leukemia in which DUX4 deregulation results in loss of function of ERG, either by deletion or induced expression of an isoform that is a dominant-negative inhibitor of wild-type ERG function' 3.

Why this is important beyond DUX4:

  • Two mechanisms converging on one endpoint. ERG deletion and ERGalt induction are alternative solutions to the same requirement. When a subtype shows two structurally unrelated lesions converging on loss of one gene's function, that gene's loss is almost certainly required, and this is strong genetic evidence that ERG loss of function is a necessary step, not a bystander.
  • A fusion driver that works by creating an alternative promoter. DUX4 does not bind a corepressor, methylate a histone or phosphorylate a substrate; it initiates transcription from a cryptic first exon and thereby manufactures a dominant-negative isoform of a different transcription factor. That is a mechanism class the graph's vocabulary does not have a term for, and it is a hypothesis generator: are there other fusions whose real effector is an induced alternative-promoter isoform of a downstream gene rather than the fusion protein itself? RUNX1::RUNX1T1's AML1-ETO9a splice isoform (see fusion-runx1-runx1t1) is a related but distinct phenomenon — an isoform of the fusion rather than of a downstream target.
  • Diagnostic consequence. Because ERG deletion is a frequent secondary event and ERG deletion has been studied as an independent favourable marker in B-ALL, some of the 'ERG-deleted' favourable-risk literature is in fact describing DUX4-rearranged cases. Any analysis that treats ERG deletion as an independent variable without adjusting for DUX4 status is confounded 23.
Clinical phenotype, age and prognosis

Disease: B-cell precursor ALL, now recognised as a distinct WHO/ICC entity ('B-ALL with DUX4 rearrangement').

Frequency: up to 7% of B-ALL overall 3; 4% of a consecutive series of 195 paediatric B-cell precursor ALL cases 2. The difference reflects the age distribution — the subtype is enriched in older patients.

Age: adolescents and young adults. DUX4 fusions were 'preferentially detected in the AYA generation' (15-39 years), and the discovery cohort was Ph-negative AYA-ALL; the authors argue AYA-ALL 'may be a clinical entity distinct from ALL at other ages' 1. So among the eleven fusions in this beat, this is one of the latest-presenting — the opposite end of the age axis from CBFA2T3::GLIS2 (infants) and ETV6::RUNX1 (toddlers).

Cytogenetically cryptic and easily missed. The lesion is an insertion of repetitive D4Z4 macrosatellite sequence, which is invisible on karyotype and is also hard for standard RNA-seq fusion callers and short-read alignment to detect, because the inserted sequence is repetitive and multi-copy. This is arguably the hardest-to-detect lesion in the whole beat, and it was only found when RNA-seq was applied systematically to Ph-negative cases 12.

A distinctive and clinically consequential immunophenotype. In the iBFM Flow Network analysis of 1812 newly diagnosed paediatric B-cell precursor ALL patients on AIEOP-BFM ALL 2009:

  • 27 of 28 DUX4-fusion-positive cases (96.4%) were CD371-positive 4;
  • CD371-positive B-ALL showed a transient myelomonocytic switch during induction, in up to 65.4% of samples at day 15;
  • these cases had an inferior early treatment response (slow early response, p<0.001);
  • yet 5-year event-free survival was 88.3% 4.

That combination — looks bad early, does well in the end — is the single most important clinical fact about this subtype, and it has a direct operational consequence that 4 spells out: recognising the myelomonocytic switch is essential for proper flow-cytometric MRD assessment, because a shifting immunophenotype will otherwise be read as MRD clearance or as lineage switch and mislead risk stratification.

Prognosis: favourable. 5-year EFS 88.3% in the CD371-positive group that is overwhelmingly DUX4-rearranged 4, despite slow early response. DUX4-rearranged B-ALL is generally reported among the more favourable subtypes, and part of the historical 'favourable ERG deletion' literature is describing these cases.

The lineage-switch caveat is real. 4 titles its finding 'propensity to lineage switch', and CD371-positive cases show a transient myelomonocytic phenotype. In the era of CD19-directed immunotherapy that matters: a leukaemia with demonstrated myelomonocytic plasticity is a candidate for genuine, permanent lineage switch under CD19 pressure.

Druggability: undrugged, and the hardest case in this beat

Status: undrugged, and structurally the least tractable of the eleven.

Why it is worse than the other undrugged fusions:

  1. There is no fusion protein. The oncoprotein is truncated DUX4 — a near-normal embryonic transcription factor. There is no composite surface, no junction epitope on a chimeric protein, nothing that exists only in the leukaemic cell as a protein-protein juxtaposition.
  2. DUX4 is a homeodomain transcription factor. Homeodomains bind DNA through a short helical module (PDB 6E8C) 5; they have no catalytic site and no ligand pocket, and their DNA-binding surfaces are shared across a large family.
  3. The pathogenic element is a regulatory relocation. Targeting an enhancer hijack means either disabling the IGH enhancers (which are required for normal B-cell function and are active in every B progenitor) or breaking the specific enhancer-promoter contact — a class of intervention with no clinical precedent.
  4. The downstream effector is loss of ERG function 3. Restoring a lost function pharmacologically is much harder than blocking a gained one, and half the cases achieve ERG loss by outright deletion, which nothing can reverse.

Mechanism-derived hypotheses, all untested:

  • The aberrant C terminus of the leukaemic DUX4 protein 1 is the one genuinely fusion-specific sequence in the whole lesion. It is a candidate neoepitope for T-cell-receptor-based or peptide-vaccine immunotherapy — far more plausible than a small molecule. DUX4's normal restriction to germline and early embryo makes DUX4 itself an attractive tumour-associated antigen too, since there is little normal tissue to damage.
  • ERGalt is a defined, transforming, dominant-negative isoform created by a DUX4-initiated promoter 3. Blocking that specific transcription start, or degrading ERGalt selectively, would be mechanism-matched.
  • Synthetic lethality with ERG loss is the obvious systematic approach: run a dependency screen in isogenic ERG-intact versus ERG-deficient B progenitors and look for a druggable partner.

What actually works clinically is conventional MRD-directed chemotherapy 4, and the practical clinical priorities are (i) detecting the lesion at all, given how cryptic it is, and (ii) not mis-reading the myelomonocytic switch as treatment failure or as MRD clearance 4.

Correction, 2026-09-19: IGH contributes translated read-through sequence, not enhancers alone

Withdrawn. This entry asserted in seven places that "IGH contributes only enhancer sequence" and, in evidence item E6, that it contributes "only enhancer sequence to this lesion rather than any coding sequence". The sibling statements in gene_order_note, uniprot3_note, pdb_note, product_type_note and the retained-domains list said the same thing.

Why. Zhang et al., describing this subtype, write: "In each case of this subtype, [DUX4] is inserted adjacent to the [IGH] enhancer region ... with variable truncation of the C-terminus of DUX4 and appending of a variable number of amino acids from read-through into the IGH locus" (PMID 27776115, doi 10.1038/ng.3691). The aberrant C terminus that this entry already treats as the one fusion-specific protein sequence in the lesion is therefore IGH-locus-derived. Lilljebjorn et al. corroborate the truncation independently: 7 of 8 genomically characterised cases expressed truncated DUX4 transcripts encoding between 312 and 420 amino acids, against a full-length 424 (PMID 27265895).

What the entry now says. IGH contributes no annotated coding exon and no protein domain; its only translated contribution is the short junction read-through that forms part of DUX4's aberrant C terminus. Everything that rested on the absence of an IGH protein domain survives unchanged: there is still no folded fusion interface to solve, still no IGH UniProt accession to record (read-through sequence has none either), and the lesion is still enhancer hijacking. What does not survive is the stronger phrasing, which made the entry's own E5 counter-evidence - already noting that 'the coding change is not a by-product of the insertion' - argue with the entry's claim sentences.

Consequence for the mechanism story. If the residues that detoxify full-length DUX4 come from IGH, then IGH is not a purely regulatory partner: it supplies both the promoter environment and, incidentally, the coding change. That is a sharper version of enhancer hijacking than the entry told, and it is testable - the appended residues are variable between patients, so their length and sequence can be correlated with DUX4 transactivation strength.

Analogy: two good-prognosis B-ALL subtypes whose cooperating lesion is loss of an ETS-family transcription factor

Closest analogue, and the axis. fusion-etv6-runx1, on the axis of the class of the cooperating lesion: in both subtypes the secondary event is loss of function of an ETS-family transcription factor. In ETV6::RUNX1 ALL it is deletion of the second, untranslocated ETV6 allele, found as subclonal populations in every fusion-positive patient in the series that looked 8. In DUX4-rearranged B-ALL it is loss of ERG function, by intragenic deletion or by the dominant-negative ERGalt isoform whose non-canonical first exon is transcriptionally initiated by DUX4 binding 3. ETV6 and ERG are both ETS-family factors; the two subtypes are otherwise mechanistically unrelated - one a corrupted RUNX1 transcription factor, the other pure enhancer hijacking of an embryonic homeobox gene - and nobody in this graph had drawn the line between them.

The clinical half of the analogy is already measured. Both are good-prognosis, non-infant B-ALL: ETV6::RUNX1 with 5-year EFS around 89%, DUX4-rearranged with 5-year EFS 88.3% in the CD371-positive group 4, and DUX4-rearranged enriched in adolescents and young adults. Neither has a fusion-directed drug and neither needs one. They sit together as the contrast class for infant KMT2A::AFF1, which needs almost no cooperating lesion and is not survivable at the same rates.

What the analogy predicts, and what has been tested. If the ETS lesion is genuinely secondary in both, it should be subclonal and dependent on the primary lesion rather than an independent initiating event. That is what is observed on both sides: ETV6 deletion appears as subpopulations within fusion-positive cases 8, and the ERG lesion occurs only on a DUX4-rearranged background, with ERGalt transcription physically initiated by DUX4 itself 3. The prediction survives.

Where it breaks, and this is the half worth keeping. The two ETS losses are not the same kind of event.

  • In ETV6::RUNX1 the ETS loss is a stochastic postnatal second hit. That is why the fusion is detectable in cord blood at roughly 100 times the lifetime incidence of the leukaemia, why latency runs to years, and why monozygotic twin concordance is low.
  • In DUX4-rearranged disease the ERG loss is written by the driver. DUX4 binds and initiates ERGalt, so the dominant-negative is part of the primary lesion's own programme rather than an independent hit 3.

Two subtypes that 'both need ETS loss' therefore make opposite predictions about natural history, and only the ETV6::RUNX1 one predicts a long silent interval. This graph records no evidence that DUX4-rearranged B-ALL has a prenatal origin at all, and that absence is now a specific, cheap question: is the DUX4 insertion present in the neonatal blood spot of a child who later develops this subtype? If it is, the driver-encoded second event should make the latency shorter than ETV6::RUNX1's; if the insertion is absent, the subtype is postnatal in origin and the analogy with ETV6::RUNX1 is limited to its clinical behaviour.

And the limit on the analogy itself. The ERG lesion is frequent, not universal, while the DUX4 rearrangement is present in all cases (counter-evidence on 8). The shared axis is the class of the cooperating lesion; it is not a claim that ETS loss is rate-limiting here.

Facts and figures
gene5DUX4
gene3IGH
gene order noteThe 5-prime/3-prime framing is misleading for this lesion and is recorded only for dashboard consistency. D4Z4 repeats containing DUX4 are INSERTED INTO the IGH locus, so IGH is the recipient regulatory environment and DUX4 is the relocated cargo; some publications write the lesion as IGH-DUX4 and some as DUX4-IGH. There is no chimeric protein in the usual sense: IGH contributes enhancers plus a short junction read-through translated onto DUX4's aberrant C terminus, and no IGH protein domain [E1][E6].
cytobandins(14;4)(q32;q35) / t(4;14)(q35;q32) - insertion of D4Z4/DUX4 from 4q35 into the IGH locus at 14q32
cytoband noteWritten variously as an insertion or a translocation depending on the report; the underlying event is relocation of D4Z4 macrosatellite repeat sequence containing DUX4 from the subtelomeric 4q35 region into IGH at 14q32 [E1]. An ERG-DUX4 variant also occurs, in which DUX4 is relocated to the ERG locus instead [E2].
frequency noteUp to 7% of B-ALL overall as an upper bound assembled across cohorts [E3]; 4% in a consecutive paediatric B-cell precursor ALL series of 195 cases (about 8 patients, so imprecise) [E2]. The subtype is enriched in adolescents and young adults [E1], which is why the all-ages figure exceeds the paediatric one. All estimates are probably conservative because the repetitive D4Z4 insertion is hard for short-read fusion callers to detect.
prognosisFavourable, but with a misleading early course. 5-year event-free survival 88.3% for the CD371-positive group that is overwhelmingly DUX4-rearranged, DESPITE a significantly inferior early treatment response (slow early response, p<0.001) and a transient myelomonocytic immunophenotypic switch in up to 65.4% of samples at day 15 [E4]. The operational consequence is that the switch must be recognised for flow-cytometric MRD to be interpreted correctly. Note the 88.3% figure is for the whole CD371-positive cohort (n=160), not specifically for the 28 DUX4-positive cases.
uniprot5Q9UBX2
uniprot3 noteIGH has NO single UniProtKB accession: the immunoglobulin heavy-chain locus is a multi-gene region assembled by somatic V(D)J recombination, and in this lesion it contributes enhancers plus, at the insertion junction, a short read-through stretch translated onto DUX4's aberrant C terminus - no annotated IGH coding exon and no IGH protein domain - so there is still no partner protein to accession [E6]. DUX4 = Q9UBX2 was verified (gene name DUX4, 424 aa, Homo sapiens).
PDB note6E8C is the DUX4 DOUBLE HOMEODOMAIN in complex with DNA - the DNA-binding module of wild-type DUX4 [E5]. There is NO structure of the leukaemic, aberrant-C-terminus DUX4 protein, and there CANNOT be a structure of a fusion interface, because IGH contributes no protein domain - only enhancers plus the short read-through stretch that forms DUX4's aberrant C terminus - so this lesion has no chimeric protein in the usual sense [E1][E6]. Ten further wild-type DUX4 homeodomain entries exist (5Z2S, 5Z2T, 5Z6Z, 5ZFW, 5ZFY, 5ZFZ, 6A8R, 6DFY, 6U81, 6U82) and are omitted as redundant.
mechanism class notePrimary and unambiguous class: enhancer-hijack - the only lesion in this beat where the pathogenic event is regulatory relocation rather than creation of a chimeric protein. The second label captures the DOWNSTREAM effector: DUX4 initiates transcription of the dominant-negative ERGalt isoform from a non-canonical first exon, causing loss of wild-type ERG function, which frequent intragenic ERG deletion achieves independently [E3]. The vocabulary has no term for 'fusion-induced alternative-promoter dominant-negative isoform of a downstream gene', which is the actual novelty here and is flagged for the Queen as a fifth gap.
druggability classundrugged
druggability noteStructurally the least tractable of the eleven fusions: there is no fusion protein, the oncoprotein is a near-normal embryonic homeodomain transcription factor, the pathogenic element is an enhancer relocation, and the downstream effector is LOSS of ERG function - which is much harder to correct pharmacologically than a gain. The most plausible routes are immunological (the aberrant DUX4 C terminus is a genuine neoepitope sequence, and DUX4 is normally restricted to germline and early embryo, making it a low-normal-tissue-risk antigen) or synthetic lethality with ERG deficiency. Nothing is clinical.
stage of hematopoiesisPro-B / early B-cell precursor. DUX4-IGH expressed specifically in PRO-B cells generated B-cell leukaemia in mice [E1], and the lesion arises in a cell with active IGH enhancers and active V(D)J/class-switch recombination machinery - i.e. a B progenitor is not merely the permissive cell but the mechanistically necessary one, since the recombinase-associated breakage at IGH is what enables the insertion. Clinically the disease presents in adolescents and young adults [E1], with a transient myelomonocytic-switch phenotype (CD371-positive) during induction [E4]
age peakadolescents and young adults (15-39 y); preferentially detected in the AYA generation [E1], making this one of the latest-presenting fusions in this beat
developmental order rank7
gene3 contributes codingfalse
product typeenhancer hijack
product type noteThis lesion produces no chimeric protein. IGH contributes no protein DOMAIN, so no IGH domain can be retained or lost; its only translated contribution is the junction read-through that forms part of DUX4's aberrant C terminus. The join checker previously reported 'no protein entry for 3' gene IGH' as a defect; it is not one, and the flag above tells the checker so. The only fusion-specific protein sequence in the entire lesion is DUX4's aberrant C terminus.
diseases
[
  "B-cell precursor ALL with DUX4 rearrangement (WHO/ICC entity)",
  "adolescent and young adult B-ALL",
  "pediatric B-ALL",
  "adult B-ALL"
]
frequency
{
  "b_all_pct": 7,
  "pediatric_b_all_pct": 4
}
breakpoints
[
  {
    "id": "bp-d4z4-dux4-insertion-igh",
    "gene5_exon_last": null,
    "gene3_exon_first": null,
    "frequency_pct": 85,
    "population": "B-cell precursor ALL with DUX4 rearrangement (IGH-DUX4 form)",
    "domains_retained_5": [
      "DUX4 double homeodomain (two tandem homeodomains, cooperative DNA binding)",
      "most of the DUX4 coding sequence"
    ],
    "domains_lost_5": [
      "native DUX4 C-terminal transactivation domain - replaced by an ABERRANT C TERMINUS created by the insertion, which is the only fusion-specific protein sequence in the entire lesion"
    ],
    "domains_retained_3": [
      "IGH Emu enhancer",
      "IGH 3-prime regulatory region enhancers - regulatory elements; no annotated coding exon, though junction read-through into IGH-locus sequence is translated onto DUX4's C terminus"
    ],
    "domains_lost_3": [
      "not applicable: no IGH protein sequence is part of the product"
    ],
    "phenotype_notes": "Enhancer hijacking: D4Z4 repeats containing DUX4 are inserted into IGH, placing truncated DUX4 under the strongest enhancers a B progenitor possesses, giving high-level expression of DUX4 with an aberrant C terminus [E1]. Exon-level breakpoints are recorded as null because the event is an insertion of repetitive macrosatellite sequence into a somatically rearranged locus, not a junction between two defined coding exons - any exon number would be meaningless here.",
    "gene5_exon_last_note": "Null by necessity: the insertion truncates DUX4 within its single-exon-dominated coding region and the recipient IGH locus is somatically rearranged, so conventional exon-pair breakpoint notation does not apply.",
    "frequency_pct_note": "85% is an approximation of the predominance of the IGH-DUX4 form over the ERG-DUX4 form and other DUX4 relocation partners [E2]; we did not verify a cohort-level breakdown."
  },
  {
    "id": "bp-d4z4-dux4-insertion-erg",
    "gene5_exon_last": null,
    "gene3_exon_first": null,
    "frequency_pct": 15,
    "population": "B-cell precursor ALL with DUX4 rearrangement (ERG-DUX4 variant form)",
    "domains_retained_5": [
      "DUX4 double homeodomain",
      "most of the DUX4 coding sequence"
    ],
    "domains_lost_5": [
      "native DUX4 C-terminus"
    ],
    "domains_retained_3": [
      "ERG locus regulatory sequence - recipient environment only"
    ],
    "domains_lost_3": [
      "not applicable: no ERG protein sequence is part of the product"
    ],
    "phenotype_notes": "The ERG-DUX4 variant, in which DUX4 is relocated to the ERG locus rather than IGH, producing the same DUX4 overexpression and clustering in the same expression subtype [E2]. Mechanistically elegant given that the downstream effector of the subtype is disruption of ERG function [E3]. Frequency is an approximation.",
    "frequency_pct_note": "Approximate; the IGH:ERG partner split is not something we verified in a cited cohort."
  }
]
PDB IDS
[
  "6E8C"
]
mechanism class
[
  "enhancer-hijack",
  "tf-repression-corruption"
]
How this could bite us back
  • Slow early response invites over-treatment. These patients look like treatment failures at day 15 - significantly inferior early response, plus a shifting immunophenotype - yet end up at 88.3% 5-year EFS [E4]. A risk-stratification algorithm that escalates on early response alone will intensify therapy, with all its toxicity, in a group that did not need it.
  • The myelomonocytic switch corrupts MRD measurement in both directions. A blast population that transiently loses B-lineage markers can read as MRD clearance (falsely reassuring, leading to de-escalation) or as lineage switch to AML (falsely alarming, leading to a change of protocol). [E4] states explicitly that recognising the switch is essential for proper flow MRD assessment - i.e. the measurement itself is the hazard.
  • Demonstrated myelomonocytic plasticity is a specific risk factor for CD19-directed immunotherapy escape. A leukaemia that can transiently shed B-lineage identity under chemotherapy is a plausible candidate for permanent lineage switch under blinatumomab or CD19 CAR-T, the failure mode already well documented in KMT2A-rearranged disease.
  • Detection is the first bite-back. A repetitive macrosatellite insertion is invisible to karyotype and poorly handled by standard short-read fusion pipelines [E1][E2], so the subtype is systematically under-diagnosed and its true frequency and prognosis are both measured on ascertained - and therefore biased - cases.
  • Nothing to drug, and a lost function to restore. There is no fusion protein, no pocket, and the downstream effector is ERG LOSS [E3] - half of it by outright deletion, which no agent can reverse. Any therapeutic programme here must be immunological or synthetic-lethal; a small-molecule inhibitor programme against this lesion would be a predictable failure, and this node exists partly to say so before the money is spent.
  • The ERG-deletion confounder contaminates the literature. Because ERG deletion is frequent in this subtype and has separately been studied as a favourable marker in B-ALL, published analyses treating ERG deletion as an independent prognostic variable without adjusting for DUX4 status are confounded [E2][E3] - a data-hygiene bite-back that affects meta-analyses, not patients directly.
  • If ERGalt rather than DUX4 is the transforming effector, any DUX4-directed therapy is one step upstream of the lesion that matters, and the subtype's 5-year EFS of 88.3% on conventional therapy leaves almost no margin in which to show benefit.
Still open
  1. Is ERG loss of function strictly REQUIRED, or merely strongly selected? Two structurally unrelated lesions - intragenic ERG deletion and DUX4-initiated ERGalt expression - converge on the same endpoint [E2][E3], which is the strongest available genetic argument for necessity. Testing it directly (restore wild-type ERG in a DUX4-rearranged patient-derived model and see whether the leukaemia dies) would be decisive, and if ERG loss is required then ERG-deficiency synthetic lethality becomes the rational therapeutic programme.
  2. Are there other fusions whose real effector is a fusion-INDUCED alternative-promoter isoform of a downstream gene, rather than the fusion protein itself? ERGalt is the only clear example in this beat [E3]. Systematically re-analysing existing fusion-positive RNA-seq for non-canonical first exons at candidate target genes is a cheap, purely computational experiment with a clear precedent, and it is the single most transferable idea in this node.
  3. How much DUX4-rearranged B-ALL is being MISSED? The lesion is an insertion of repetitive D4Z4 macrosatellite sequence, which short-read alignment and standard fusion callers handle poorly [E1][E2]. Re-analysing existing paediatric and AYA B-ALL cohorts with repeat-aware or long-read methods could raise the apparent frequency substantially and would change how much clinical attention the subtype deserves.
  4. Does the transient myelomonocytic switch [E4] predict genuine, permanent lineage switch under CD19-directed immunotherapy? These cases are already the flow-cytometry hard case; if their plasticity is real rather than an artefact of antigen modulation, they are a defined at-risk group for blinatumomab and CAR-T escape, and that is directly testable in existing immunotherapy cohorts.
  5. Is the aberrant DUX4 C terminus [E1] immunogenic? It is the only leukaemia-specific protein sequence in the entire lesion, and DUX4's normal restriction to germline and early embryo means the whole protein is nearly a tumour-specific antigen. HLA-binding prediction followed by T-cell-reactivity screening against patient material is a well-trodden, tractable path that nobody appears to have taken.
  6. What does the C-terminally altered leukaemic DUX4 protein actually look like, and does it acquire an interaction surface the full-length protein lacks? The entry currently records that no structure exists; the more useful statement is that nobody has looked.
  7. Is DUX4-rearranged B-ALL prenatal in origin? Every other good-prognosis B-ALL subtype this beat compares it with - ETV6::RUNX1, high hyperdiploidy - has backtracking evidence from neonatal blood spots, and the DUX4 subtype has none in this graph. The insertion is a structural rearrangement into IGH with a patient-specific junction, so it is exactly the kind of clonotypic marker backtracking was built for, and the answer decides whether the ETS-loss analogy with ETV6::RUNX1 extends to natural history or stops at the clinic.
Every source behind this entry (8)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Insertion of D4Z4 repeats containing DUX4 into the IGH locus is frequent in B-cell ALL of adolescents and young adults, producing high-level expression of DUX4 protein with an aberrant C terminus; expression of DUX4-IGH in pro-B cells generated B-cell leukaemia in transplanted mice, and DUX4 fusions were preferentially detected in the AYA generation.

    Recurrent DUX4 fusions in B cell acute lymphoblastic leukemia of adolescents and young adults · Yasuda T, Tsuzuki S, Kawazu M, Hayakawa F, Kojima S, Ueno T, Imoto N, Kohsaka S, Kunita A, Doi K, Sakura T, Yujiri T, Kondo E, Fujimaki K, Ueda Y, Aoyama Y, Ohtake S, Takita J, Sai E, Taniwaki M, Kurokawa M, Morishita S, Fukayama M, Kiyoi H, Miyazaki Y, Naoe T, Mano H · Nat Genet 2016
    PMID 27019113DOI
    What this evidence cannot show

    Discovery cohort of only 73 Ph-negative AYA-ALL specimens, so the frequency estimate within AYA-ALL is imprecise. The mouse transplantation assay used enforced DUX4-IGH expression in pro-B cells, i.e. it demonstrates sufficiency in an engineered context rather than recapitulating the physiological insertion. Detection of a repetitive D4Z4 insertion by short-read RNA-seq is technically difficult, so frequency estimates from this era may be underestimates.

    Evidence that argues against this (1)
    • The transforming species may be downstream of DUX4 rather than DUX4 itself: DUX4 binds a non-canonical ERG first exon and initiates ERGalt, which retains ERG's DNA-binding and transactivation domains, inhibits wild-type ERG and is itself transforming, while ERG is separately deleted in many of the same cases. Attributing the phenotype to DUX4 overexpression alone skips the effector. Zhang J et al. 2016
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In a consecutive series of 195 paediatric B-cell precursor ALL cases analysed by RNA sequencing, a subtype characterised by recurrent IGH-DUX4 or ERG-DUX4 fusions represented 4% of cases, led to overexpression of DUX4 and frequently co-occurred with intragenic ERG deletions.

    Identification of ETV6-RUNX1-like and DUX4-rearranged subtypes in paediatric B-cell precursor acute lymphoblastic leukaemia · Lilljebjorn H, Henningsson R, Hyrenius-Wittsten A, Olsson L, Orsmark-Pietras C, von Palffy S, Askmyr M, Rissler M, Schrappe M, Cario G, Castor A, Pronk CJH, Behrendtz M, Mitelman F, Johansson B, Paulsson K, Andersson AK, Fontes M, Fioretos T · Nat Commun 2016
    PMID 27265895DOI
    What this evidence cannot show

    4% of 195 cases is only about 8 patients, so the frequency estimate is imprecise; the cohort is paediatric, which understates the frequency relative to AYA and adult series [E1][E3]. RNA-seq-based fusion detection of a repetitive D4Z4 insertion is technically challenging and may miss cases.

    Evidence that argues against this (1)
    • The 4% is population-specific, quite apart from resting on about eight patients. Lee et al. assigned molecular subtype by RNA sequencing in 2428 children and adolescents with ALL enrolled on frontline trials in the United States, Singapore/Malaysia and Guatemala and found that the frequency of somatic DUX4 rearrangement varies with genetic ancestry: positively associated with East Asian ancestry (OR 1.30, 95% CI 1.16-1.45, P<0.001) and negatively associated with African ancestry (OR 0.70, 95% CI 0.48-0.93, P=0.01), while both African and Native American ancestry independently predicted worse event-free and overall survival even after adjusting for subtype. A single-country paediatric series therefore estimates a quantity that is not constant across the populations this graph will be read in. Lee SHR et al. 2022
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Deregulation of DUX4 and ERG is the hallmark of a B-progenitor ALL subtype comprising up to 7% of B-ALL; DUX4 rearrangement and overexpression is present in all cases and is accompanied by ERG deregulation, expression of a novel isoform ERGalt whose non-canonical first exon is transcriptionally initiated by DUX4 binding, and frequent ERG deletion; ERGalt retains ERG's DNA-binding and transactivation domains but inhibits wild-type ERG transcriptional activity and is transforming - a paradigm in which DUX4 deregulation causes loss of ERG function either by deletion or by a dominant-negative isoform.

    Deregulation of DUX4 and ERG in acute lymphoblastic leukemia · Zhang J, McCastlain K, Yoshihara H, Xu B, Chang Y, Churchman ML, Wu G, Li Y, Wei L, Iacobucci I, Liu Y, Qu C, Wen J, Edmonson M, Payne-Turner D, Kaufmann KB, Takayanagi SI, Wienholds E, Waanders E, Ntziachristos P, Bakogianni S, Wang J, Aifantis I, Roberts KG, et al., Downing JR, Mullighan CG · Nat Genet 2016
    PMID 27776115DOI
    What this evidence cannot show

    The 'up to 7%' figure is an upper bound assembled across cohorts of differing age composition rather than a single measured prevalence. ERGalt's transforming activity and dominant-negative effect were shown in experimental systems with enforced expression; the relative contribution of ERGalt versus outright ERG deletion in patients, and whether either is strictly required, is not resolved. Whether ERG loss of function is necessary or merely strongly selected is inferred from convergence of two lesion types, which is compelling but indirect.

    Evidence that argues against this (1)
    • The same entity was defined, and its clinical behaviour established, by the ERG lesion alone and by a mechanism that does not require DUX4. Clappier et al. characterised a recurrent V(D)J-recombinase-mediated intragenic ERG deletion by array CGH, built a breakpoint-specific PCR assay, and screened 897 unselected children aged 1-17 on the EORTC-CLG 58951 trial: the deletion was present in 29 (3.2%), mutually exclusive of every known classifying lesion, associated with older age (median 7.0 vs 4.0 years) and aberrant CD2 expression (43.5% vs 3.7%) and with frequent IKZF1 deletions (37.9% vs 5.3%), yet 8-year event-free and overall survival were 86.4% and 95.6%. A V(D)J-mediated deletion is a sufficient and independent origin for the ERG lesion, so convergence of ERG deletion and ERGalt does not by itself show that DUX4 deregulation CAUSES the loss of ERG function; and 3.2% in an unselected consecutive cohort sits below the 'up to 7%' upper bound assembled across cohorts of differing age composition. Clappier E et al. 2014
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In 1812 newly diagnosed paediatric B-cell precursor ALL patients on AIEOP-BFM ALL 2009, 27 of 28 (96.4%) DUX4-fusion-positive cases were CD371-positive; CD371-positive B-ALL showed a transient myelomonocytic switch in up to 65.4% of samples at day 15 and an inferior early treatment response (p<0.001), yet 5-year event-free survival was 88.3%; recognising the switch is essential for correct flow-cytometric MRD assessment.

    CD371-positive pediatric B-cell acute lymphoblastic leukemia: propensity to lineage switch and slow early response to treatment · Buldini B, Varotto E, Maurer-Granofszky M, Gaipa G, Schumich A, Bruggemann M, Mejstrikova E, Cazzaniga G, Hrusak O, Szczepanowski M, Scarparo P, Zimmermann M, Strehl S, Schinnerl D, Zaliova M, Karawajew L, Bourquin JP, Feuerstein T, Cario G, Alten J, Moricke A, Biffi A, Parasole R, Fagioli F, Valsecchi MG, Biondi A, Locatelli F, Attarbaschi A, Schrappe M, Conter V, Basso G, Dworzak MN · Blood 2024
    PMID 38215390DOI
    What this evidence cannot show

    The 88.3% EFS is for the whole CD371-positive group (n=160), NOT specifically for DUX4-rearranged cases, and the DUX4 analysis rests on only 28 fusion-positive cases within a 420-patient validation subset. The relationship is one-directional in strength: nearly all DUX4 cases are CD371-positive, but CD371 positivity is not specific to DUX4. Single trial protocol with centralised flow cytometry, so reproducibility across laboratories is not established.

    Evidence that argues against this (1)
    • The favourable outcome is not uniform inside DUX4-rearranged disease, and the disagreement is about the same modifier. Benjamin et al. subtyped 394 B-ALL patients (301 children, 93 adults) by whole-transcriptome sequencing combined with FISH, cytogenetics and RT-PCR; DUX4-R was favourable overall and was placed in their favourable paediatric risk group alongside hyperdiploidy and ETV6::RUNX1, but they report that among the DUX4-R B-ALL patients, those with IKZF1 deletion had worse event-free and overall survival. That contradicts the ERG-related literature, in which this subtype was reported to have a remarkably favourable outcome DESPITE a high incidence of IKZF1 aberrations. A single subtype-level survival figure - here 88.3% attached to a 160-patient CD371-positive group of which only 28 cases in a 420-patient subset were DUX4-positive - therefore conceals a genotype that does badly. Benjamin M et al. 2025
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The DUX4 double homeodomain binds DNA cooperatively and is structurally characterised (PDB 6E8C and related entries for Q9UBX2), but no structure exists of the truncated, aberrant-C-terminus leukaemic DUX4 protein, and there is no fusion protein interface to solve because IGH contributes no protein domain - only enhancers plus a short read-through C-terminal extension.

    What this evidence cannot show

    Verified by fetching the RCSB entry record and querying the RCSB search API on the reference-sequence accession field. All entries cover the DNA-binding homeodomain region of wild-type DUX4; none covers the C-terminus, which is precisely the part altered in the leukaemic allele [E1]. The claim that there is no fusion interface is a structural deduction from the enhancer-hijack mechanism, not an experimental result.

    Evidence that argues against this (1)
    • 'No fusion protein interface to solve because IGH contributes only enhancer sequence' is true of the IGH side and misleading about the lesion. The DUX4 the rearrangement produces is C-terminally altered, and full-length DUX4 is pro-apoptotic -- so the coding change is not a by-product of the insertion, it is what makes an IGH-driven DUX4 survivable. There is a protein to characterise here; what is missing is a structure of it, not a reason for one not to exist. Kowaljow V et al. 2007
  6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    There is no single UniProt accession for the IGH partner, because the immunoglobulin heavy-chain locus is a multi-gene, somatically rearranged region rather than a single protein-coding gene, and it contributes no annotated coding exon and no protein domain to this lesion - its only translated contribution is the short junction read-through that forms part of DUX4's aberrant C terminus.

    What this evidence cannot show

    Verified by fetching the UniProt REST record for Q9UBX2 and confirming gene name DUX4 and length 424 aa. The statement about IGH is a description of how the locus is annotated rather than a claim requiring a specific accession; individual constant-region genes (for example IGHM, P01871) do have accessions, but none is part of this lesion's product.

    Evidence that argues against this (1)
    • IGH does contribute translated sequence, so the final clause of this claim was wrong. Zhang et al. state for this subtype that 'in each case of this subtype, [DUX4] is inserted adjacent to the [IGH] enhancer region ... with variable truncation of the C-terminus of DUX4 and appending of a variable number of amino acids from read-through into the IGH locus'. The novel residues that make the leukaemic DUX4 protein aberrant are therefore IGH-locus-derived, even though no ANNOTATED IGH coding exon and no IGH protein domain is present in the product; consistently, Lilljebjorn et al. found 7 of 8 genomically characterised cases expressing truncated DUX4 transcripts encoding between 312 and 420 amino acids against a full-length 424. The accession argument is untouched - read-through sequence has no UniProt accession either - but 'contributes only enhancer sequence rather than any coding sequence' has been corrected (see the 2026-09-19 correction section). Zhang J et al. 2016
  7. E7moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Full-length DUX4 is a pro-apoptotic protein, so the C-terminal alteration carried by the leukaemic DUX4 is a necessary part of the lesion rather than an incidental consequence of the insertion.

    The DUX4 gene at the FSHD1A locus encodes a pro-apoptotic protein. · Kowaljow V, Marcowycz A, Ansseau E, et al. · Neuromuscular Disorders 2007
    PMID 17588759DOI
    What this evidence cannot show

    Abstract only, and indirect: established in the FSHD1A/myoblast context, not in B-cell progenitors, and no published side-by-side transformation comparison of full-length versus truncated DUX4 in pro-B cells was found. The inference is strong but is an inference.

    Evidence that argues against this (1)
    • Toxicity and transactivation map to the same residues, so losing the C-terminus does not selectively detoxify DUX4 - it disables it. Mitsuhashi et al. compared mutant, deletion and fusion variants of DUX4 against DUX4-FL on five readouts - activation of a DUX4 promoter reporter, expression of an endogenous DUX4 target gene, cell viability, activation of endogenous caspases and level of protein ubiquitination - and found each construct produced a similarly sized effect (or lack of effect) in every assay: the ability to activate transcription determined the extent of change in all of them. Transactivation was mediated by the C-terminal 80 amino acids, with most activity in the C-terminal 20; and non-toxic constructs retaining both homeodomains acted as DOMINANT-NEGATIVE inhibitors of DUX4-FL, presumably by competing for promoter sites. A C-terminal truncation is therefore not free detoxification: whatever transcriptional output leukaemic DUX4 retains has to be reconciled with the residues it lost, and a construct that lost enough of them would inhibit DUX4 rather than mimic it. Mitsuhashi H et al. 2018
  8. E8moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The cooperating event in the other good-prognosis B-ALL subtype has the same class as this one: in childhood ALL with the TEL/AML1 (ETV6::RUNX1) fusion, every fusion-positive patient carried subpopulations of leukaemic cells with deletion of the normal, untranslocated ETV6 allele - so in both subtypes the secondary lesion is subclonal loss of function of an ETS-family transcription factor, ETV6 there and ERG here.

    Fluorescence in situ hybridization analysis of 12;21 translocation in Japanese childhood acute lymphoblastic leukemia. · Eguchi-Ishimae M, Eguchi M, Tanaka K, Hamamoto K, Ohki M, Ueda K, Kamada N · Jpn J Cancer Res 1998
    PMID 9738986DOI
    What this evidence cannot show

    Seven TEL/AML1-positive patients in a 74-patient Japanese paediatric series, by FISH with YAC and cosmid probes; 'all patients had subpopulations' rests on small numbers and on an assay whose sensitivity for a minor subclone is not stated. The cross-subtype framing is this graph's, not the paper's: ETV6 and ERG are both ETS-family transcription factors and both lesions are loss of function, but no study has compared the downstream consequences of the two losses in the same system, so the shared class is a structural observation rather than a demonstrated shared mechanism.

    Evidence that argues against this (1)
    • The ETS-loss axis cannot be a necessary cooperating event in DUX4-rearranged disease, because the ERG lesion is not universal while the DUX4 lesion is. In a consecutive series of 195 paediatric B-cell precursor ALL, the subtype defined by recurrent IGH-DUX4 or ERG-DUX4 fusions - 4% of cases - was described as FREQUENTLY, not invariably, co-occurring with intragenic ERG deletions; the founding DUX4/ERG study likewise found DUX4 rearrangement and overexpression in ALL cases while ERG deletion was frequent rather than universal. So the analogy links the two subtypes at the level of lesion CLASS and cannot be read as making ETS loss the rate-limiting step here; the necessary event in this subtype is DUX4 itself. Lilljebjörn H et al. 2016
Connected to