Oncofusion

start / graph / ana-q173-the-inverse-architecture-framing-fails-three-of-its-four-claims

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.q173-the-znf384-fusion-family

The ZNF384 family is not the inverse of KMT2A::AFF1 - it is the SAME grammar written in the opposite gene order, and three of the four claims in the inverse framing are false

also called ZNF384 inverse architecture test · ZNF384 versus KMT2A architecture

The idea being tested was that ZNF384 leukaemia is KMT2A leukaemia turned back to front: a DNA-binding protein with a chromatin regulator stuck on, rather than a chromatin protein with an elongation factor stuck on. Three of the four things that idea assumed turn out not to be true. What is true is less tidy and more interesting: both families keep the half of the protein that decides WHERE to go, and throw away everything else.

The same thing, for a scientist

This entry scores a hypothesis the coordinator wrote into this beat's brief and asked to have tested rather than confirmed. **The framing was: ZNF384 fusions are the inverse architecture of KMT2A::AFF1 - a transcription factor fused to a chromatin coactivator, with ZNF384 invariantly 3', retaining all eight zinc fingers, and the partner invariantly contributing a coactivator or IDR.** Scored against the literature, **three of its four component claims are false and the fourth is true but trivial**: | claim | verdict | what actually holds | |---|---|---| | ZNF384 is always the 3' partner | **REFUTED** | ZNF384::TEX41 is a sequenced, out-of-frame fusion with ZNF384 5'; and a whole class of cases has no fusion at all [E1] | | the fusion retains all eight zinc fingers | **NOT ESTABLISHED, AND UNMEASURABLE AS STATED** | the count is disputed 8 / 6 / 'five to eight'; exon-8 exclusion removes two fingers; no case-level census exists [E2] | | the partner always contributes a coactivator or IDR | **REFUTED** | about 45% of named partners are kinases, trafficking proteins, an ATPase, a helicase or a lncRNA - and the COMMONEST partner, EP300, contributes a fragment with the HAT domain deleted [E3][E4] | | ZNF384 is 3' where KMT2A is 5' | **TRUE, AND UNINFORMATIVE** | gene order differs; the functional grammar does not - see below | **The finding that replaces the framing: both families retain the ADDRESS module of the index gene and discard everything else.** KMT2A fusions keep the N-terminal AT-hooks and CXXC and lose the SET; ZNF384 fusions keep the C-terminal zinc-finger array and lose almost nothing else because there is almost nothing else. **They are not inverse architectures. They are the same architecture - retain the DNA-targeting half, replace the rest - and the 5'/3' difference is an accident of which end of each protein carries the address** [E5]. Where the two families genuinely converge is downstream and it is measured: **a stem-cell-resident founding lesion, lineage-ambiguous output, MPAL overlap, lineage switch under lineage-directed therapy, and FLT3 over-expression - the last of these through demonstrably DIFFERENT cis elements** [E6].

Go deeper
Test 1 - is ZNF384 always the 3' partner? REFUTED

Zaliova et al., in a consecutive Czech cohort of 643 children, report verbatim 1:

"Routine analysis of RNA-seq data using in-house bioinformatic pipelines for single-nucleotide variants, indels, and fusion gene detection revealed noncanonical ZNF384 fusion (ZNF384 as the 5' partner) in 1 patient"

The fusion is ZNF384::TEX41, out of frame, terminating nine residues into intronic sequence from a long non-coding RNA gene. The case carried the ZNF384-rearranged expression signature and clustered with canonical cases.

And the refutation is broader than one inverted fusion. The same paper's other two cases have no fusion at all - in-frame indels in the last exon of ZNF384. A family definition that turns on which side of a fusion ZNF384 sits cannot accommodate patients who do not have a fusion.

What survives: the canonical fusions do place ZNF384 3'. Zaliova state it plainly: "Within these 'canonical' fusions, the entire ZNF384 coding region is fused in-frame to the partner coding sequence and forms the 3' part of the fusion gene/C-terminus of the resulting fusion protein." The word doing the work is canonical, and it is doing it by definition.

Test 2 - are all eight zinc fingers retained? NOT ESTABLISHED, and the question is malformed

Three problems, in increasing order of seriousness 2:

  1. The count is disputed. UniProt Q8TF68 and the 2002 founding paper both say eight; Li et al. 2018 and Zaliova et al. 2021 say six; Dickerson et al. say five to eight. The 'six' is the number of fingers UniProt annotates in the paralogue ZNF362, whose coordinates lie within six residues of ZNF384's first six - the newer schematic appears to be the paralogue-shared core.
  2. The expressed number is not fixed even once the count is. Dickerson et al.: "Exclusion of exon 8 results in in-frame splicing with loss of zinc fingers 4 and 5." So alternative splicing removes fingers from the protein regardless of the breakpoint.
  3. Nobody has counted. This beat enumerated 112 PubMed records for ZNF384 with fusion or rearrangement terms in title and abstract and found no case-level census of retained zinc fingers for any cohort. 'All eight are retained' is not a finding that has been checked and failed; it is a statement nobody has tested.

The premise is therefore not refuted, it is unmeasurable as stated - which is a different and in some ways worse result, because it means the claim has been repeated without a denominator.

Test 3 - does the partner always contribute a coactivator or an IDR? REFUTED, twice over

First, by the partner list. Alongside EP300, CREBBP, TCF3, TAF15, EWSR1, ARID1B, SMARCA2/4 and NIPBL, the reported partners include BMP2K (a serine/threonine kinase, with its catalytic domain retained), CLTC (clathrin heavy chain), SYNRG (an AP-1 endosomal adaptor), ATP2C1 (a calcium ATPase), SEC24B (a COPII coat subunit), USP25 (a deubiquitinase), DDX42 (an RNA helicase), CCAR1, and TEX41 (a long intergenic non-coding RNA contributing nine out-of-frame residues) 34. That is roughly 45% of the named partner list.

Second, and far more damagingly, by the commonest partner. EP300::ZNF384 joins EP300 exon 6 to ZNF384 exon 3, producing a 1,027-residue protein that retains TAZ1 and discards the HAT domain and the bromodomain 4. EP300 is in this fusion as a ~450-residue protein-interaction fragment with no catalytic activity. The single commonest member of 'transcription factor fused to a chromatin coactivator' does not contain the coactivator's catalytic function.

Third, the functional evidence has pointed the same way since 2002 and has never been answered. Martini et al. found the TET-CIZ fusions transform NIH3T3 cells, bind the CIZ consensus in EMSA, and do not transactivate a CIZ-site reporter or induce any tested CIZ target, while EWSR1-FLI1 controls in the same experiment induced strongly 5. Their conclusion, verbatim: "suggesting a contribution to oncogenesis that is independent of the transactivating properties of the fusion protein."

The correct weaker claim, which does survive: most patients in this subtype have a chromatin- or transcription-associated partner, because EP300 and TCF3 alone are 74% of partner-typed cases. Partner-name frequency and patient frequency are different statistics and the framing conflated them.

And the partner is not nothing. Kusano et al. showed that TCF3::ZNF384 and EP300::ZNF384 both down-regulate CD10 but only TCF3::ZNF384 confers dexamethasone resistance and raises CCND2 7. The right model is a shared ZNF384-driven lineage programme plus partner-specific modifiers - not a partner that supplies the oncogenic module.

Test 4 - and the reframing that the evidence actually supports

5' versus 3' is the wrong axis. Line the two families up by function instead of by gene order:

KMT2A::AFF1ZNF384 fusions
index gene contributesAT-hooks + CXXC: a non-sequence-specific CpG address modulethe zinc-finger array: a sequence-specific [GC]AAAAA address module
index gene losesPHD, bromodomain, FYRN, TAD, SET - i.e. its catalytic and regulatory machineryits C-terminus - a region with no annotated function, and demonstrably enough to matter (mechanism-znf384-c-terminal-truncation-phenocopies-the-fusion)
partner contributesan elongation module (AFF1 / SEC)variable: an activation region, a TAZ1 fragment, a prion-like IDR, a kinase domain, or nine residues of nonsense
gene order of the retained address5'3'

Both families keep the address and throw away the machinery. That is one grammar, not two, and the 5'/3' difference falls out of where each protein happens to carry its DNA-targeting domain - KMT2A's CXXC is N-terminal, ZNF384's fingers are C-terminal. The 'inverse architecture' is a fact about protein topology, not about mechanism.

Where they differ, which is the part worth keeping. KMT2A fusions occupy invariant CpG islands; EWS::FLI1 creates its addresses at polymorphic GGAA microsatellites - a contrast this graph already records. ZNF384 fusions read a short sequence-specific consensus and, at FLT3, activate a distal element that is closed in KMT2A-AF4 samples 6. So three fusion families, three different classes of genomic address, converging on overlapping output.

Where they converge, and it is measured, not asserted:

  • Stem-cell-resident lesion. TCF3-ZNF384 is detectable in a subset of HSCs; this graph records the same cell-of-origin model for KMT2A.
  • Lineage ambiguity. ZNF384 rearrangement in 48% of paediatric B/myeloid MPAL, with ZNF384r MPAL and ZNF384r B-ALL transcriptionally indistinguishable; KMT2A-r is the other defining-genetic MPAL subtype.
  • Lineage switch under lineage-directed therapy, reported for both.
  • FLT3 over-expression, highest in exactly these two subtypes across 1,988 B-ALL transcriptomes - and through different cis elements, which is the mechanistically interesting part.

One measured contrast worth remembering because it runs the wrong way: CD10 is eight-fold HIGHER in ZNF384-rearranged than in KMT2A-rearranged adult B-ALL (median 36.16% vs 4.53%, p < 0.0001), even though ZNF384-r is CD10-low against BCP-ALL generally. Two subtypes both described as 'CD10-weak, myeloid-marker-positive' are strongly separable on that marker.

Facts and figures
statusrun
result summaryOf four component claims in the inverse-architecture framing, one is refuted by a sequenced counterexample (ZNF384 is not always 3'), one is refuted at the partner-list level and by the commonest partner's own domain content (the partner does not always contribute a coactivator or IDR), one is unmeasurable as stated because the zinc-finger count itself is disputed and no case-level census exists, and one is true but uninformative (the gene order differs). The framing is replaced by: both families retain the DNA-targeting module of the index gene and discard the rest, so the architectures are the SAME grammar with the address at opposite ends of the two proteins. The genuine cross-family results are downstream and convergent - stem-cell-resident lesion, lineage ambiguity, MPAL overlap, lineage switch, and FLT3 over-expression through demonstrably different cis elements.
measurements
proportion0.910proportion: 0.91count1.000count: 1ratio7.980ratio: 7.98residue449.0residue: 449
inputs
[
  "PMID 34529760 full text (PMC8579251)",
  "PMID 33692463 full text (PMC8550960)",
  "PMID 12359745 full text (Internet Archive snapshot of the AACR HighWire page)",
  "PMID 25943178 full text (University of Tsukuba repository PDF)",
  "PMID 29531323 full text (PMC5931087)",
  "PMID 35247902 (PMC9414741)",
  "PMID 36104354 full text (PMC9474531)",
  "PMID 30209392 (PMC6195459)",
  "PMID 40391410 abstract",
  "PMID 30487223 full text (PMC6294900)",
  "UniProtKB Q8TF68, Q5T0B9, Q09472, Q92793, P15923, Q92804, Q9NSY1, Q01844 (all retrieved live 2026-09-24)"
]
measurements — the numbers behind the chart
[
  {
    "quantity": "count",
    "subject": "component claims of the inverse-architecture framing that survive scoring",
    "value": 1,
    "units": "claims",
    "n": 4,
    "universe": "the four claims stated in the q173 dispatch brief: ZNF384 always 3'; all eight zinc fingers retained; partner always a coactivator or IDR; ZNF384 3' where KMT2A is 5'",
    "null": null,
    "note": "the one survivor is the gene-order claim, which is true and carries no mechanistic content. A second claim is not refuted but unmeasurable as stated."
  },
  {
    "quantity": "proportion",
    "subject": "ZNF384 partner-typed patients whose 5' partner is a chromatin- or transcription-associated protein",
    "value": 0.91,
    "units": "fraction",
    "n": 193,
    "universe": "Ponte di Legno partner distribution: EP300 83, TCF3 60, TAF15 17, CREBBP 15 = 175 of 193, plus chromatin members of the 18-case 'other' group (EWSR1 6, ARID1B 1, SMARCA2 1, NIPBL 1)",
    "null": 0.55,
    "note": "THE NULL IS THE SAME QUANTITY COUNTED BY PARTNER NAME RATHER THAN BY PATIENT, where roughly 55% of about 20 named partners are chromatin-associated. The framing's error was using the partner-name intuition to make a patient-level claim, or vice versa - the two differ by 36 percentage points."
  },
  {
    "quantity": "residue",
    "subject": "EP300 contribution to the commonest ZNF384 fusion",
    "value": 449,
    "units": "aa (approximate, derived)",
    "n": null,
    "universe": "1,027 aa predicted fusion minus 577 aa of complete ZNF384, less the 5-bp untranslated insertion; against EP300 Q09472, 2,414 aa",
    "null": 2414,
    "note": "DERIVED BY THIS BEAT, NOT PUBLISHED. The null is full-length EP300: the fusion carries 19% of it, and none of the catalytic or bromodomain-containing 81%."
  },
  {
    "quantity": "ratio",
    "subject": "CD10 positivity in ZNF384-rearranged versus KMT2A-rearranged adult BCP-ALL",
    "value": 7.98,
    "units": "fold",
    "n": 72,
    "universe": "43 ZNF384-rearranged versus 29 KMT2A-rearranged adult BCP-ALL; medians 36.16% and 4.53% of blasts, p < 0.0001",
    "null": 1,
    "note": "DERIVED RATIO of two published medians, 36.16/4.53. Recorded because the direction is the opposite of what 'ZNF384r is CD10-weak' predicts, and because this is the only head-to-head marker comparison of the two families this beat could obtain - from an abstract only."
  }
]
How this could bite us back
  • This analysis touches interventions indirectly but decisively. Every partner-directed therapeutic idea in this family - degrading EP300, blocking a bromodomain, disrupting a prion-like IDR, inhibiting a kinase - depends on the partner being the oncogenic module, and three lines of evidence say it may not be: the commonest partner contributes no catalytic domain, the founding functional experiment found no transactivation, and three human cases have the disease with no partner at all.
  • The corollary for patient selection is concrete. If a trial enrols on 'ZNF384 fusion detected', it excludes the last-exon-indel class and the out-of-frame 5'-ZNF384 class, who have the same expression signature and the same immunophenotype. If a trial enrols on the expression signature instead, it includes patients with no fusion for a partner-directed drug to act on. Neither selection rule is right until the partner question is settled.
  • And a warning about importing from Ewing sarcoma: EWSR1::ZNF384 appears to carry the identical EWSR1 exon-7 fragment as EWSR1::FLI1, which makes condensate-directed ideas superficially transferable. This graph's own Ewing comparator records that increasing low-complexity-domain interaction REPRESSES EWS::FLI1 targets, so the transfer could move a ZNF384 fusion the wrong way along a non-monotone curve.
Still open
  1. Does a ZNF384 protein truncated after its zinc-finger array, with NO fusion partner, transform a haematopoietic stem and progenitor cell? This is the experiment that decides between 'the partner supplies the oncogenic module' and 'the lesion is loss of the ZNF384 C-terminus'. Three natural human lesions - two indels and one out-of-frame fusion - already say the partner is dispensable; nobody has tested it on purpose.
  2. Repeat Martini's 2002 transactivation experiment with the targets discovered since: SALL4, ID2, RGS1, CREB5 and FLT3. A twenty-four-year-old negative on the wrong target list is currently the strongest evidence against the family's central mechanistic assumption, and it can be settled in a month.
  3. Does the same 'retain the address, discard the machinery' grammar hold across other fusion families this graph covers - NUP98, CBFA2T3::GLIS2, MEF2D, DUX4? If it does, it is a general principle worth stating; if it does not, the KMT2A/ZNF384 similarity is a coincidence of two families rather than a rule. This is a corpus-wide analysis this graph is uniquely positioned to run against its own domain payloads.
  4. Why is CD10 eight-fold higher in ZNF384-rearranged than in KMT2A-rearranged disease if both arrest at a similar point? Either the arrest points differ more than the shared 'CD10-weak B/myeloid' description suggests, or the comparison's KMT2A arm was partner-skewed. The discriminator is a paediatric replication with partner-resolved KMT2A arms.
  5. Is the framing's failure informative about how such framings are generated? All three failed claims share a shape: a true statement about the commonest cases, generalised into a universal. The partner-name-versus-patient-count confusion in particular moved a real 91% into an apparent 100% and made a counterexample look like an anomaly rather than a refutation.
Every source behind this entry (7)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    A noncanonical ZNF384 fusion with ZNF384 as the 5' partner (ZNF384-TEX41, out of frame) was identified in one patient, and two further patients in the same consecutive cohort of 643 children carried in-frame indels in the last exon of ZNF384 and no fusion at all; all three had the ZNF384-rearranged gene-expression signature. The paper states that within canonical fusions the entire ZNF384 coding region is fused in frame to the partner and forms the 3' part of the fusion gene.

    A novel class of ZNF384 aberrations in acute leukemia. · Zaliova M, Winkowska L, Stuchly J, Fiser K, Triska P, Zwyrtkova M, Hrusak O, Starkova J, Sramkova L, Stary J, Trka J, Zuna J · Blood Adv 2021
    PMID 34529760DOI
    What this evidence cannot show

    FULL TEXT READ via Europe PMC fullTextXML for PMC8579251. n = 1 for the 5'-ZNF384 fusion and n = 2 for the indel class, out of 12 ZNF384-aberrant cases in the cohort. No functional experiment was done on any of the three lesions. A single counterexample refutes a universal claim, which is the logical work this item does, but it does not establish how common the exception is - 2 of 12 consecutively diagnosed ZNF384-aberrant cases were noncanonical, with a Wilson 95% interval of roughly 4.7-45%.

    Evidence that argues against this (1)
    • The defender of the framing has a reasonable reply: the canonical fusions really are invariantly 3', the paper says so explicitly, and the exceptions are a different lesion class rather than counterexamples within the class. That reply is fair as far as it goes, and it is why this analysis grades the claim REFUTED rather than the family incoherent. What it cannot rescue is the inference the framing was built on - that the architecture is invariant BECAUSE the partner must supply something to the N-terminus - since the two indel cases and the TEX41 case have no N-terminal partner and produce the same disease. Zaliova M et al. 2021
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The number of ZNF384 zinc fingers is given as eight by UniProtKB Q8TF68 and by the 2002 founding paper, as six by Li et al. 2018 and Zaliova et al. 2021, and as 'five to eight' by Dickerson et al. 2022; Dickerson additionally report that exclusion of ZNF384 exon 8 results in in-frame splicing with loss of zinc fingers 4 and 5. UniProt annotates exactly six C2H2 fingers in the paralogue ZNF362 (Q5T0B9), at coordinates within six residues of ZNF384's first six, with no counterpart to ZNF384's fingers at 403-425 and 433-455.

    What this evidence cannot show

    This item aggregates four sources and two database records and is expanded in full in ana-q173-the-znf384-zinc-finger-count-is-not-agreed, which should be read before this one is cited. The paralogue coordinate comparison is a COORDINATE COMPARISON, NOT A SEQUENCE ALIGNMENT - no alignment was run. The exon-8 splicing statement is from a figure and its frequency was not extracted.

    Evidence that argues against this (1)
    • None of the four sources argues against the others - they simply state different numbers without engaging, which is why this is recorded as an unresolved disagreement rather than a controversy with a winner. The reading offered here, that 'six' is the paralogue-shared core and ZNF384's last two fingers are additional, is this beat's inference from two feature tables and is not stated by any source. An equally consistent alternative is that ZNF384's seventh and eighth are degenerate profile over-calls, in which case the papers are right and the database is generous. The discriminating experiment - a DNA-binding assay on ZNF384 truncated at 397 versus 455 - has not been done. Dickerson KM et al. 2022
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The reported ZNF384 5' partner list includes proteins that are not chromatin regulators or transcriptional coactivators: BMP2K (a serine/threonine protein kinase whose catalytic domain is retained in the fusion), CLTC (clathrin heavy chain), SYNRG (an AP-1 adaptor), ATP2C1 (a calcium ATPase), SEC24B (a COPII coat subunit), USP25 (a deubiquitinase), DDX42 (an RNA helicase), CCAR1, and TEX41 (a long intergenic non-coding RNA). In the Ponte di Legno series the partner distribution by PATIENT is dominated by chromatin-associated partners: EP300 83 of 193 (43%), TCF3 60 (31%), TAF15 17 (9%), CREBBP 15 (8%), other 18 (9%).

    Clinical characteristics and outcomes of B-ALL with ZNF384 rearrangements: a retrospective analysis by the Ponte di Legno Childhood ALL Working Group. · Hirabayashi S, Butler ER, Ohki K, Kiyokawa N, Bergmann AK, Moricke A, Boer JM, Cave H, Cazzaniga G, Yeoh AEJ, et al., Moorman AV, Manabe A · Leukemia 2021
    PMID 33692463DOI
    What this evidence cannot show

    FULL TEXT READ via Europe PMC fullTextXML for PMC8550960 for the partner counts; the wider partner list combines that source with the Zhu 2023 review (PMID 37306722) and with the ZNF384::TEX41 case (PMID 34529760). THE CLASSIFICATION OF PARTNERS AS 'NOT CHROMATIN' IS THIS BEAT'S, is debatable at the margins, and is offered to make the argument checkable rather than as a measurement. The consortium series is a retrospective collection across 16 groups and 26 years with heterogeneous detection methods, so the partner spectrum reflects ascertainment as well as biology.

    Evidence that argues against this (1)
    • The strongest counter is that this argument switches denominators and the switch must be declared. By PARTNER NAME roughly 45% of the list is not chromatin; by PATIENT the chromatin partners are about 91% of partner-typed cases. If the framing is read as a claim about what most patients have, it is substantially correct and this item does not refute it. The framing as written was universal - 'the partner list is almost entirely chromatin coactivators and remodellers, and the architecture is invariant' - and it is the universality that fails. A weaker, denominator-explicit version survives intact and should replace it. Zhu L et al. 2023
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    EP300::ZNF384, the commonest member of the family, joins EP300 exon 6 to ZNF384 exon 3 and produces a predicted 1,027-residue protein that retains the TAZ1 domain of EP300 but lacks the histone acetyltransferase domain; an independent series confirms the exon 6 to exon 3 junction in 8 of 9 patients and states that the truncation eliminates the HAT and the bromodomain.

    A novel recurrent EP300-ZNF384 gene fusion in B-cell precursor acute lymphoblastic leukemia. · Gocho Y, Kiyokawa N, Ichikawa H, Nakabayashi K, Osumi T, Ishibashi T, et al. · Leukemia 2015
    PMID 25943178DOI
    What this evidence cannot show

    FULL TEXT READ as the author-accepted PDF in the University of Tsukuba repository, the free location Unpaywall lists for this DOI. The bromodomain statement is from a second paper (McClure et al., PMID 29531323, read via Europe PMC). No source gives an explicit retained-or-lost verdict for EP300's KIX, TAZ2 or IBiD domains; the inference that they are lost follows from the fusion's stated total length and is arithmetic performed by this beat, recorded as derived in fusion-ep300-znf384.

    Evidence that argues against this (1)
    • A defender of the coactivator framing can point out that TAZ1 is a protein-interaction module of a coactivator and that the fusion may recruit wild-type EP300 in trans - and there is evidence for exactly that, since ZNF384-fusion proteins bind EP300 more strongly than wild-type ZNF384 does. But that evidence cuts the other way on inspection: the binding was mapped to the ZNF384 half of the fusion, not the EP300 half, and the GST bait was EP300 residues 1053-1860, a region the fusion does not contain. If the fusion recruits EP300 through its ZNF384 portion, the EP300 moiety is not what makes it a coactivator fusion. Yamamoto H et al. 2019
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The founding report of the family found that TAF15-CIZ and EWSR1-CIZ transform NIH3T3 cells and bind the CIZ DNA recognition sequence in EMSA, but show no enhanced or decreased transactivation of a CIZ-binding-site reporter and induce no transcription of mmp3, eat2 or manic fringe in NIH3T3 cells or mmp3 or mmp7 in Ba/F3 cells, while EWSR1-FLI1 controls in the same experiment induced strongly; transformation required both halves, since constructs expressing only the TET portion or only wild-type CIZ gave a normal phenotype. The authors concluded the contribution to oncogenesis is independent of the transactivating properties of the fusion protein.

    Recurrent rearrangement of the Ewing's sarcoma gene, EWSR1, or its homologue, TAF15, with the transcription factor CIZ/NMP4 in acute leukemia. · Martini A, La Starza R, Janssen H, Bilhou-Nabera C, Corveleyn A, Somers R, Aventin A, Foa R, Hagemeijer A, Mecucci C, Marynen P · Cancer Res 2002
    PMID 12359745
    What this evidence cannot show

    FULL TEXT READ from an Internet Archive snapshot of the original HighWire page dated 2009-09-29; aacrjournals.org is Cloudflare-blocked and no DOI exists for this pre-DOI AACR article. The experiments are in NIH3T3 and Ba/F3, neither a human B-cell precursor. Only two of the family's partners were tested and both are FET proteins - the result may not generalise to EP300 or TCF3 fusions. expected_absent_terms names ZNF384 because the gene is CIZ/NMP4 throughout this paper.

    Evidence that argues against this (1)
    • THE TARGET LIST IS THE LIMITATION AND IT IS SERIOUS. The reporter and the target genes tested were the CIZ targets known in 2002 - MMP3, MMP7, eat2, manic fringe - and UniProt still flags the MMP/COL1A1 targets as inferred 'By similarity' rather than experimentally established in human. Real ZNF384-fusion targets found later could not have been tested: SALL4 and ID2, RGS1 and CREB5, and FLT3 through a distal enhancer. Martini et al. name this possibility themselves - 'or that MMPs are not the true targets of CIZ'. So the founding negative may show the wrong targets were chosen rather than that transactivation is not the mechanism, and it has never been repeated with the right list. This is the single most fixable gap in the family's evidence base. Zhao X et al. 2022
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Across 1,988 B-ALL transcriptomes the highest FLT3 expression was in ZNF384-rearranged (n = 49) and KMT2A-rearranged (n = 136) cases; the ZNF384 fusion protein activates a 500 bp intergenic enhancer 25 kb 5' of FLT3 (hg38 chr13:28,124,365-28,124,868) and loops it to the promoter, and this element is completely void of ATAC-seq signal in ALL samples of other subtypes including KMT2A-AF4.

    PMID 36104354DOI
    What this evidence cannot show

    FULL TEXT READ via Europe PMC fullTextXML for PMC9474531. The KMT2A-AF4 comparison is an ATAC-seq track in one figure panel, not a quantified test across a KMT2A cohort, so 'closed in KMT2A-AF4' rests on a small number of samples. The paper also notes that FLT3 expression tracks B-cell differentiation stage, higher in the immature subtypes KMT2A-r and ZNF384-r, so the convergent over-expression is partly confounded with maturation. Genome-wide peak-partition arithmetic in this paper does not reconcile (see target-flt3-in-znf384-rearranged-leukaemia) and no derived quantity should be taken from it.

    Evidence that argues against this (1)
    • The convergence claim is weaker than it looks because the same paper says FLT3 over-expression is not specific to these two subtypes - 'FLT3 overexpression is also observed in ALL cases without KMT2A or ZNF384 rearrangements' - and gives hyperdiploid ALL with a locus microdeletion as an example of a third route. Two subtypes topping a ranking that also has other routes to the same phenotype, in a setting where the phenotype tracks differentiation stage and both subtypes are immature, is a weaker convergence than 'the same output by different cis elements' suggests. The different-element part is solid; the shared-output part is partly a stage effect. Zhao X et al. 2022
  7. E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Transduction of BCP-ALL cell lines with TCF3::ZNF384 and EP300::ZNF384 side by side showed that both down-regulate CD10, but only TCF3::ZNF384 confers resistance to dexamethasone-induced apoptosis and only TCF3::ZNF384 enhances CCND2 expression - supporting a model of common ZNF384-rearrangement functions producing the lineage-ambiguous phenotype plus fusion-partner-dependent functions layered on top.

    TCF3::ZNF384 induces steroid resistance in B-cell precursor acute lymphoblastic leukemia cells. · Kusano S, Ueno-Yokohata H, Hori M, Ishibashi T, Fujimura J, Shimizu T, Ohki K, Kiyokawa N · Pediatr Int 2025
    PMID 40391410DOI
    What this evidence cannot show

    ABSTRACT ONLY - cell lines, replicate numbers and effect sizes are not available to this beat, so this item supports a direction and not a magnitude. Retroviral over-expression in established BCP-ALL lines is not the cell of origin identified for this subtype. Only two of the family's twenty-plus partners were compared, so 'partner-dependent' means these two differ.

    Evidence that argues against this (1)
    • The abstract reports 'no growth advantage was observed' in the same sentence as the CCND2 result, which is not what a functional cyclin D2 increase predicts and leaves the proposed mechanism of the steroid resistance unsupported. There is also a missing control: TCF3::ZNF384 retains most of the TCF3 N-terminal activation region, and over-expressing a strong activation domain alone can confer apoptosis resistance without any contribution from the ZNF384 half. Martini et al. ran precisely that control for the TET partners in 2002 and found the partner fragment alone inert; whether Kusano et al. did is not visible in the abstract. Martini A et al. 2002
Connected to