Oncofusion

start / graph / mechanism-enhancer-hijacking

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

Mechanism class 5 — enhancer hijacking (inv(3) GATA2/EVI1 — and DUX4-r, which only half belongs here)

also called enhancer adoption · regulatory rearrangement class · no-fusion-protein class

Sometimes the rearrangement does not build a new fusion protein: it moves a powerful genetic on-switch next to a gene that should be silent. But the cleanest example of that is inv(3) AML — in DUX4-rearranged leukaemia the relocation also truncates the gene, and the truncation turns out to matter.

The same thing, for a scientist

In this class the pathogenic consequence of the rearrangement is regulatory rather than structural. A strong lineage-specific enhancer is relocated into the topological neighbourhood of a gene that is normally silent, and drives its ectopic expression; the coding sequence of the target may be intact or only slightly truncated. inv(3)/t(3;3) AML is the clean case: it relocates a GATA2 distal haematopoietic enhancer to activate EVI1 while leaving one GATA2 allele functionally deprived, with no coding change at all. **IGH-DUX4 is only half a member.** It places the immunoglobulin heavy-chain enhancer upstream of a DUX4 whose C terminus is altered by the rearrangement -- and that alteration is load-bearing, because full-length DUX4 is pro-apoptotic [E4]. Downstream, the truncated DUX4 initiates transcription of ERGalt, a neomorphic ERG isoform that is itself transforming and dominant-negative over wild-type ERG [E5]. So the DUX4 subtype's output is a coding lesion plus a neomorphic protein, not pure regulatory deregulation. The class as a whole is still the furthest from conventional druggability and its therapeutic handle is still the downstream program rather than the lesion -- but the inference 'enhancer hijacking, therefore no protein to drug' does not hold for DUX4-r.

Go deeper
The cleanest proof: one enhancer move, two genes deregulated

In inv(3)(q21q26) and t(3;3) AML, a distal GATA2 haematopoietic enhancer is repositioned into the EVI1 locus. Allele-specific analysis showed that the single relocated enhancer simultaneously activates EVI1 and leaves the remaining GATA2 allele haploinsufficient, and CRISPR deletion of the repositioned enhancer in patient-derived cells collapsed EVI1 expression and restored differentiation 1. No fusion protein is produced at all. That paper is the reference standard for what 'enhancer hijacking' means and is why the mechanism is now searched for routinely in fusion-negative leukaemias. Two later results qualify the word 'hijacking' itself, and they are on this entry's counter-evidence: the element that actually drives EVI1 is non-functional at its endogenous GATA2 locus and is bound by MYB only on the translocated allele, so the enhancer gains an activity when it moves rather than carrying one across; and in an allele-tracking mouse model EVI1 activation and GATA2 haploinsufficiency are cooperating requirements separated by a pre-leukaemic interval, not a single simultaneous event.

DUX4 rearrangement in B-ALL: a fusion transcript whose point is the enhancer1 deeper

RNA sequencing of 195 consecutive paediatric B-cell-precursor ALL cases defined a subtype with recurrent IGH-DUX4 or ERG-DUX4 rearrangements in about 4% of cases, causing DUX4 overexpression and frequently co-occurring with intragenic ERG deletions 2. The transcript is chimeric, so it is catalogued as a fusion, but the mechanism is enhancer adoption: the IGH enhancer, active by definition in a B-lineage cell, drives an otherwise silent and normally germline-restricted transcription factor.

The co-occurring intragenic ERG deletion is itself DUX4-driven and produces a truncated ERG isoform, so the subtype is doubly defined by regulatory consequence rather than by a new protein surface. Clinically it matters because these children do unusually well, so the class includes the best-prognosis fusion subtype in paediatric B-ALL.

Why lineage context is the whole mechanism

Enhancer hijacking is only oncogenic if the hijacked enhancer is active in the cell where the break happens. IGH enhancers are active in B progenitors; the GATA2 enhancer is active in myeloid progenitors. This makes the class the most explicitly developmental-stage-dependent of the five, and it is the reason these lesions cluster tightly by lineage and age (see concept-haematopoietic-stage-ordering-of-fusions). It also means the same rearrangement in a different cell would be silent, which is a testable statement and a reason to be careful about modelling these lesions in the wrong progenitor.

Therapeutic implication — stated plainly

This class has yielded no targeted drugs and probably will not yield one against the lesion. There is no fusion protein interface, the hijacked enhancer is a stretch of normal regulatory DNA, and the overexpressed target (DUX4, EVI1) is a transcription factor with no pocket. Genome-editing the enhancer works in the laboratory 1 and is the only true on-mechanism intervention demonstrated, which puts this class in the gene-therapy column rather than the small-molecule column.

But the class carries the most encouraging clinical message in the atlas for a different reason: DUX4-rearranged B-ALL is cured at high rates by conventional chemotherapy. Not every fusion needs a targeted drug; some need accurate classification so that children are not over-treated. Any comparative strategy should separate 'fusions we must drug' (KMT2A-r, TCF3::HLF, CBFA2T3::GLIS2, NUP98-r) from 'fusions we must correctly identify and then leave alone' (DUX4-r, ETV6::RUNX1).

Correction (cycle 33): DUX4-r is a mixed lesion, not a clean enhancer hijack

Withdrawn: 'there is no novel protein-protein interface and often no novel protein at all' as applied to the DUX4 row.

  • The rearrangement alters the DUX4 C terminus, and full-length DUX4 is pro-apoptotic 4, so an untruncated DUX4 under IGH control would kill the cell. Enhancer relocation alone does not produce this disease.
  • The truncated DUX4 initiates ERGalt from a non-canonical ERG first exon; ERGalt is transforming and acts as a dominant negative over wild-type ERG, and ERG is separately deleted in many cases 5. Two neomorphic protein species are in the causal chain.
  • Frequency understated: the row says about 4%; the genomic characterisation says up to 7% of B-ALL 5.

Unaffected: the inv(3)/t(3;3) GATA2-to-EVI1 row, which remains the clean exemplar of the class, and the general conclusion that this class is the furthest from conventional druggability.

Why this matters for the beat. Mechanism-class assignment is being used in this graph to score druggability, including in ana-comparative-interface-druggability-scan, where classes 'without a discrete partner interface' are predicted to have no chemical matter. If a class member turns out to carry a coding lesion and a neomorphic downstream protein, the scan is scoring the wrong object for that row.

Narrowing 2026-09-19 — 'relocation' is not the whole mechanism, and the class is not undruggable

Two bounds from the Q11 counter-argument sweep.

The enhancer changes when it moves. A CRISPR enhancer scan across the translocated GATA2 enhancer found a single short element, carrying a MYB motif, that is required for EVI1 expression and dispensable for GATA2; MYB binds it only on the rearranged allele, and blocking MYB or mutating the motif drives differentiation and death. The authors' own framing is that a chromosomal aberration can expose a regulatory element that is non-functional at its endogenous locus. That is a gain-of-function lesion in regulatory DNA, not a relocation of an existing function, and it means the claim at 3 - that a relocated element alone reproduces the oncogenic output - is too strong even for the exemplar.

The class is less undruggable than this entry says. Proteomics of the hijacked 3q super-enhancer complex identified PARP1 and IKZF1 among its constituents; PARP inhibition reduced EVI1 expression, reduced the frequency of the enhancer-promoter contact, and killed 3q-AML cells with differentiation and CD34+ depletion. The conclusion 'no fusion protein interface, therefore no chemical matter' skips the assembled complex on the enhancer, which is protein and has ligands. ana-comparative-interface-druggability-scan scores this class on the lesion; it should score the complex.

Neither bound touches the entry's clinical conclusion that DUX4-rearranged B-ALL is chemo-curable and should be identified rather than drugged.

Facts and figures
mechanism classenhancer-hijack
invariant modulea lineage-active enhancer relocated into a new topological neighbourhood
variable modulethe silent target gene that is switched on (DUX4, EVI1), often intact or only truncated
druggable nodenone on the lesion; only the downstream program, or enhancer editing as a gene-therapy modality
druggability classundrugged
why not druggedFor inv(3)/t(3;3) GATA2-to-EVI1 this is exact: there is no novel protein and no novel interface, the pathogenic element is normal regulatory DNA in the wrong place, and the overexpressed target is a pocketless transcription factor. For DUX4-r it is WRONG as written and was corrected in cycle 33: the rearrangement alters the DUX4 C terminus (and full-length DUX4 is pro-apoptotic, PMID 17588759), and the truncated DUX4 initiates ERGalt, a neomorphic and transforming ERG isoform (PMID 27776115). DUX4-r is undrugged for want of a tractable target on either protein, not because no protein exists.
clinical counterpointDUX4-rearranged B-ALL has among the best outcomes in paediatric B-ALL, so for this class accurate classification and de-escalation matter more than new drugs.
developmental dependencehighest of the five classes — the hijacked enhancer must be active in the cell of origin for the lesion to do anything at all
member fusions
[
  "fusion-dux4-igh"
]
analogous non fusion lesions
[
  "inv(3)/t(3;3) GATA2-enhancer to EVI1",
  "IGH-CRLF2",
  "IGH-MYC"
]
drugs yielded
[]
drugs failed or stalled
[
  "no targeted agent has ever been developed against a hijacked enhancer"
]
class assignment corrections
[
  {
    "member": "DUX4-r B-ALL (IGH-DUX4, ERG-DUX4)",
    "original_assignment": "pure enhancer hijacking, no novel protein",
    "corrected": "mixed: enhancer relocation PLUS a C-terminal coding alteration that is required (full-length DUX4 is pro-apoptotic), plus a neomorphic downstream protein (ERGalt) that is transforming and dominant-negative over wild-type ERG",
    "pmids": [
      "17588759",
      "27776115",
      "27019113"
    ],
    "confidence": "the coding alteration is established; that it is REQUIRED for transformation is inferred from DUX4's pro-apoptotic activity in a non-haematopoietic context"
  }
]
frequency correction
{
  "row": "DUX4-r B-ALL",
  "stored": "about 4%",
  "upper_estimate": "up to 7% of B-ALL",
  "pmid": "27776115",
  "title": "Deregulation of DUX4 and ERG in acute lymphoblastic leukemia.",
  "authors": "Zhang J, McCastlain K, Yoshihara H, et al.",
  "venue": "Nat Genet",
  "year": 2016,
  "title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
}
How this could bite us back
  • If enhancer hijacking can be acquired as a secondary lesion, it offers the leukaemia a fusion-independent route to re-express the same oncogenic program after targeted therapy, which would look like non-mutational resistance.
  • Enhancer editing as therapy means editing haematopoietic stem cells, inheriting all the delivery, mosaicism and insertional-risk problems of gene therapy.
  • Because some members of this class are chemo-curable, over-eager targeted-therapy development here could expose children with an excellent prognosis to unnecessary novel-agent toxicity.
  • If ERGalt is the transforming species rather than DUX4 itself, then therapies aimed at DUX4 in this subtype are aimed one step upstream of the effector, and the 5-year EFS of 88.3% already achieved with chemotherapy leaves very little room to demonstrate benefit.
Still open
  1. How many apparently fusion-negative paediatric leukaemias are actually enhancer-hijack cases that short-read RNA sequencing misses, and would long-read or Hi-C screening reclassify them?
  2. Is enhancer hijacking ever a *secondary* event in KMT2A-rearranged leukaemia that could rescue the cell when the fusion program is blocked by a menin inhibitor?
  3. Since enhancer deletion works in the laboratory, is there a realistic base- or prime-editing route to inactivating a hijacked enhancer in a haematopoietic stem cell, and what fraction of cells would have to be corrected?
  4. Has anyone compared full-length DUX4 with the leukaemic C-terminally altered DUX4 side by side for transformation of B-cell progenitors? The class assignment for this subtype turns on that experiment and it does not appear to have been published.
  5. If the oncogenic activity of a hijacked enhancer depends on a trans factor that is dispensable at the enhancer's native locus, is there an equivalent allele-specific trans dependency at any KMT2A-rearranged breakpoint - and would that be a cleaner therapeutic target than the fusion itself?
Every source behind this entry (5)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In inv(3)/t(3;3) myeloid leukemia a single repositioned GATA2 distal haematopoietic enhancer simultaneously activates EVI1 and leaves GATA2 functionally haploinsufficient, and deleting the repositioned enhancer collapses EVI1 expression and restores differentiation.

    PMID 24703711DOI
    What this evidence cannot show

    A relatively small patient series with cell-line and patient-derived editing experiments; inv(3) AML is an adult-predominant entity, so its direct relevance to paediatric fusion biology is by analogy.

    Evidence that argues against this (2)
    • 'A single repositioned enhancer' understates what the translocated allele has become. An unbiased CRISPR/Cas9 enhancer scan across the translocated GATA2 enhancer pinpointed one short regulatory element that is critical for EVI1 transcription, containing a DNA-binding motif for MYB. MYB occupies that site SPECIFICALLY at the translocated allele, and is dispensable for GATA2 expression; MYB knockout or peptidomimetic blockade of CBP/p300-dependent MYB function downregulated EVI1 but not GATA2, and mutating the motif caused myeloid differentiation and death. The authors state the general point directly: a chromosomal aberration can reveal a regulatory element that is non-functional at its endogenous locus. The enhancer is therefore not doing at EVI1 what it did at GATA2, and 'repositioned' is not a complete description of the lesion. Smeenk L et al. 2021
    • The 'simultaneously' in this claim hides a requirement for both arms and an interval. In an allele-tracking mouse model of 3q-rearranged AML, a population co-expressing high EVI1 and Gata2 appeared in the marrow BEFORE frank leukaemia and had acquired serial colony-forming potential, i.e. there is a pre-leukaemic expansion rather than a single-step conversion. Dissected separately, inv(3)-driven EVI1 accumulates megakaryocyte- and myeloid-biased progenitors, and Gata2 heterozygous deletion enhances that skewing; only the two together provoked leukaemia with the megakaryocyte and platelet features seen in human 3q AML. EVI1 activation and GATA2 haploinsufficiency are cooperating requirements, not one event described twice. Yamaoka A et al. 2020
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Recurrent IGH-DUX4 or ERG-DUX4 rearrangements define a subtype of about 4% of paediatric B-cell-precursor ALL, cause DUX4 overexpression and frequently co-occur with intragenic ERG deletions.

    Identification of ETV6-RUNX1-like and DUX4-rearranged subtypes in paediatric B-cell precursor acute lymphoblastic leukaemia · Lilljebjörn H, Henningsson R, Hyrenius-Wittsten A, et al. · Nature Communications 2016
    PMID 27265895DOI
    What this evidence cannot show

    A single-region consecutive cohort of 195 cases, so the 4% frequency has wide confidence limits and DUX4 detection is sensitive to the sequencing and alignment approach used against a repetitive locus.

    Evidence that argues against this (4)
    • The DUX4 row is misclassified as 'no novel protein at all'. The lesion produces a C-terminally altered DUX4, and that alteration is not incidental: full-length DUX4 is a pro-apoptotic protein, so an intact DUX4 driven by the IGH enhancer would kill the cell rather than transform it. The rearrangement is therefore doing two things -- relocating an enhancer AND truncating the coding sequence -- and only the first is enhancer hijacking. Kowaljow V et al. 2007
    • Downstream, the class is further from 'purely regulatory' than the entry states. DUX4 binds a non-canonical first exon of ERG and initiates transcription of ERGalt, a novel ERG isoform that retains ERG's DNA-binding and transactivation domains, inhibits wild-type ERG transcriptional activity and IS ITSELF TRANSFORMING; ERG is otherwise deleted in many of the same cases. So the subtype's oncogenic output includes a neomorphic protein and a dominant-negative protein, neither of which is captured by 'a normal gene being shouted at by the wrong amplifier'. Zhang J et al. 2016
    • The frequency stated here (about 4%) is the lower bound. The larger genomic characterisation of this subtype puts DUX4/ERG deregulation at up to 7% of B-ALL, so the row understates the size of the class. Zhang J et al. 2016
    • 'Frequently co-occur with intragenic ERG deletions' is an assay statement more than a biological one, and the deletions are mostly late. Across the same lesion, ERG-deletion detection was 34% by SNP array, 66% by genomic PCR and 78% by ultra-deep amplicon sequencing, so the reported co-occurrence rate is set by the method; even at maximum sensitivity 22-34% of DUX4-rearranged cases have no ERG deletion at all, and SNP-array false negatives misclassified five patients. Decisively for causal order, ultra-deep sequencing found distinct CO-EXISTING ERG deletions in 22 of 24 patients: the lesion is polyclonal, so it arises repeatedly and late rather than forming part of the initiating event. ERG deletion is a recurrent secondary consequence of the DUX4 program, not a component of the founding lesion. Zaliova M et al. 2019
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Enhancer-driven overexpression can substitute mechanistically for a fusion protein: in inv(3)/t(3;3) a relocated regulatory element reproduces the oncogenic transcriptional output with no fusion protein produced. (Claim narrowed 2026-09-19: it previously said the relocated element 'alone' reproduces the output, and the sufficiency implied by that word does not hold - see the counter-evidence on this item and the 2026-09-19 narrowing section.)

    PMID 24703711DOI
    What this evidence cannot show

    Generalising from one locus; for most enhancer-hijack candidates in paediatric leukaemia the causal enhancer has not been deleted experimentally, so the mechanism is inferred from topology and expression. The word 'alone' was withdrawn on 2026-09-19: the critical element is non-functional at its endogenous GATA2 locus, is bound by MYB only on the translocated allele, and the assembled complex on the relocated enhancer requires CEBPA and RUNX1 binding sites and PARP1.

    Evidence that argues against this (2)
    • This is the generalisation the flagship case does not support. If a relocated regulatory element alone reproduced the oncogenic output, then the element would work the same way at both loci and would need no new trans input. It does not. The critical element in the translocated GATA2 enhancer is non-functional at its endogenous locus, is bound by MYB only on the translocated allele, and requires CBP/p300-dependent MYB activity that GATA2 expression does not require. 'Enhancer-driven overexpression substitutes for a fusion protein' therefore needs the qualifier that the enhancer acquires a neomorphic, allele-specific activity when it moves - which makes it structurally more like a gain-of-function lesion and less like a relocation than the term 'hijacking' suggests. Smeenk L et al. 2021
    • The relocated enhancer is also not self-sufficient in trans. Chromatin proteomics of the hijacked 3q-AML super-enhancer complex found its function depends on enriched CEBPA and RUNX1 binding sites and on a protein network including PARP1 and IKZF1; PARP inhibition reduced EVI1 expression AND reduced the frequency of the EVI1-enhancer interaction, and 3q-AML cells were highly sensitive to PARP inhibition with differentiation, apoptosis and CD34+ depletion. So the oncogenic output requires an assembled transcription-factor complex, which both weakens the 'element alone' claim and, usefully, supplies the druggable node this entry says the class lacks. Kiehlmeier S et al. 2021
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The leukaemic DUX4 protein is C-terminally altered, and the truncation matters because full-length DUX4 is pro-apoptotic; an intact DUX4 driven by the IGH enhancer would be cytotoxic rather than transforming.

    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 for leukaemia: the pro-apoptotic property of full-length DUX4 was established in the facioscapulohumeral dystrophy context, in myoblasts, not in B-cell progenitors. The leukaemia reports (PMIDs 27019113, 27776115) describe the aberrant C terminus but do not, in what was read here, present a side-by-side full-length-versus-truncated transformation assay in pro-B cells. The class assignment is nonetheless wrong as written, because a coding-sequence change is part of the lesion.

    Evidence that argues against this (1)
    • The pro-apoptotic result is real but the death machinery it runs through is cell-type-specific, and the entry's inference assumes it is wired the same way in a B progenitor. An siRNA screen in an inducible-DUX4 RD rhabdomyosarcoma line, confirmed in human myoblasts, showed DUX4-induced apoptosis is mediated by the MYC-mediated apoptotic pathway and by the double-stranded-RNA innate immune response, with DUX4 expression raising MYC mRNA and producing nuclear dsRNA foci associated with EIF4A3 aggregation. MYC set-point and dsRNA-sensing tone differ profoundly between skeletal muscle and a pro-B cell, so 'full-length DUX4 under IGH control would be cytotoxic rather than transforming' is an extrapolation across the two tissues in which it has NOT been tested. This is the experiment this entry's own open question already asks for, and it identifies what would settle it: express full-length and leukaemic C-terminally altered DUX4 side by side in B-cell progenitors and read out MYC induction, dsRNA response activation and apoptosis. Shadle SC et al. 2017
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    DUX4 binds a non-canonical first exon of ERG and initiates ERGalt, a novel ERG isoform that retains the DNA-binding and transactivation domains, inhibits wild-type ERG and is itself transforming; the subtype comprises up to 7% of B-ALL.

    Deregulation of DUX4 and ERG in acute lymphoblastic leukemia. · Zhang J, McCastlain K, Yoshihara H, et al. · Nature Genetics 2016
    PMID 27776115DOI
    What this evidence cannot show

    Abstract only. Transformation was shown for ERGalt in experimental systems; the relative contributions of DUX4 truncation, ERGalt and ERG deletion to the leukaemia are not partitioned, and 'up to 7%' spans several cohorts.

    Evidence that argues against this (2)
    • The 'up to 7%' figure is the top of a cohort- and assay-dependent range, and the range is wide because the lesion is cryptic. In an independent prospective front-line trial cohort (DFCI ALL Consortium Protocol 16-001) with whole-transcriptome sequencing of 173 patients, DUX4-rearranged cases were 11 of 173, i.e. 6.4%, sitting between the 4% of the discovery cohort cited at [E2] and the 7% cited here. The same report makes the detection point plainly: fusion calls were 100% concordant with conventional cytogenetics but RNA-seq found 56 ADDITIONAL fusions, and DUX4-r is a subtype that only gene-expression classification recovers. Any frequency quoted for this subtype is a frequency for the assay it was measured with. Tran TH et al. 2022
    • The ERG arm of this claim is weaker than its position in the causal chain suggests. ERG deletions in DUX4-rearranged ALL are absent in 22-34% of cases even by ultra-deep amplicon sequencing, and where present they are typically polyclonal, with distinct co-existing deletions in 22 of 24 patients - a late, repeatedly-acquired secondary event. ERGalt initiation by DUX4 can be real while the ERG axis is not the obligatory effector for the subtype, and a therapeutic strategy aimed at ERGalt would by construction miss the quarter to third of patients with no ERG lesion. Zaliova M et al. 2019
Connected to