Oncofusion

start / graph / exp-q173-znf384-c-terminal-truncation-alone-in-hspc

Experimentspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.q173-the-znf384-fusion-family

Express a partner-free ZNF384 C-terminal truncation in HSPCs alongside TCF3::ZNF384 - the one experiment that decides whether the fusion partner is necessary

also called ZNF384 p.480 truncation experiment · partner-free ZNF384 HSPC assay

Build the version of ZNF384 that one child actually had - the protein chopped off near its end, with no partner attached - and put it into blood stem cells next to the normal fusion. If it causes the same leukaemia, the partner is not the cause. If it does nothing, the partner is.

The same thing, for a scientist

**Design.** Express four constructs in human CD34+ HSPCs and in mouse HSPCs, in parallel, in the systems Dickerson et al. have already built and validated [E1]: (a) **ZNF384 truncated at residue 480** with the nine-residue SFQDYCACQ tail, reproducing the ZNF384::TEX41 protein exactly as annotated (ZNF384 p.481_516 > SFQDYCACQ) [E2]; (b) **ZNF384 p.393_397 KHNPP>RAG** and (c) **ZNF384 p.483_484 CS>YFEWDPND**, the two last-exon indels [E2]; (d) **TCF3::ZNF384** as the positive control, since it is the only construct shown to produce lethal disease from human CD34+ cells in NSG-SGM3 [E1]; plus **wild-type ZNF384** and empty vector as negatives. **Readouts:** survival and latency; CD19/CD33/MPO co-expression by flow; RNA-seq scored against the published ZNF384-rearranged signature; ZNF384 CUT&RUN scored against the 10,470 peaks reported in a TCF3-ZNF384 PDX and specifically at the z-FLT3 enhancer (hg38 chr13:28,124,365-28,124,868) [E3]. **Why now:** three human cases already carry these exact lesions and have the disease [E2], and the founding paper's both-halves-required control used **wild-type** CIZ, never a truncation [E4] - so the decisive construct has never been made in twenty-four years.

Go deeper
What each outcome means, pre-registered
resultinterpretation
truncation gives lethal disease with B/myeloid phenotype, TCF3::ZNF384 also doeshyp-znf384-oncogenesis-is-c-terminal-loss-not-partner-gain supported. The partner is dispensable for transformation.
truncation inert at 150 days, TCF3::ZNF384 lethalHypothesis refuted. The partner is necessary. This is the pre-registered falsifier and it is the more likely outcome on the prior.
both inertUninformative - the assay failed. Do not reinterpret. Report the failure and do not proceed to a mouse arm.
truncation gives the expression signature and occupancy but no lethal diseasePartial: the truncation is sufficient for the PROGRAMME and not for transformation, which would match Dickerson's finding that mouse HSPCs need a cooperating lesion. Add NRAS G12D and repeat before concluding.

The third row is the one that matters procedurally. This project has been caught before treating an assay failure as a negative result, and a construct that is inert in an assay whose positive control also failed says nothing at all.

Controls the design must carry, and why each
  1. Wild-type ZNF384, because Martini et al. showed wild-type CIZ alone is inert and that negative must be reproduced in this system for the truncation's result to mean anything 4.
  2. Expression-level matching by western blot, because Kusano et al. reported that one of their transfectants expressed weakly and had to be excluded, and an inert construct that is simply not expressed is the commonest way this experiment fails silently.
  3. The 5' partner fragment alone - TCF3(1-exon13) with no ZNF384 - because that is the other half of Martini's both-halves control and it has never been run for a non-FET partner.
  4. A cytokine-non-humanised NSG arm for at least the positive control, because NSG-SGM3 mice express human SCF, GM-CSF and IL-3 and bias toward myeloid output, which is the readout. Without this, a myeloid phenotype is partly a property of the host.
  5. At least 8 mice per arm, so that a null result has the power to be a null result rather than an absence of data.
Facts and figures
designParallel retroviral/lentiviral expression of six epitope-tagged constructs in human CD34+ HSPCs (transplanted into NSG-SGM3, with a non-cytokine-humanised NSG arm for the positive control) and in mouse HSPCs (with and without NRAS G12D): (a) ZNF384 p.481_516>SFQDYCACQ truncation; (b) ZNF384 p.393_397 KHNPP>RAG; (c) ZNF384 p.483_484 CS>YFEWDPND; (d) TCF3::ZNF384 positive control; (e) wild-type ZNF384 negative control; (f) empty vector. All constructs V5- or HA-tagged so that occupancy profiling reads the construct, not endogenous ZNF384.
readoutSurvival and latency; peripheral-blood human chimerism and CD19/CD33/MPO co-expression by flow cytometry; bulk RNA-seq scored against the published ZNF384-rearranged expression signature; anti-tag CUT&RUN scored for recovery of the 10,470-peak TCF3-ZNF384 reference set and for occupancy and accessibility at the z-FLT3 enhancer, hg38 chr13:28,124,365-28,124,868.
controlsWild-type ZNF384 and empty vector as negatives; TCF3::ZNF384 as the positive control in the human arm and EP300::ZNF384 in the mouse arm, matching each system's published positive; western blot for construct expression in every arm BEFORE transplant, because an unexpressed construct is the commonest silent failure; the TCF3 5' fragment alone, to complete Martini's both-halves control for a non-FET partner; at least 8 mice per arm.
cost tierhigh
timeline9-15 months: 2 months construct synthesis and validation, 1 month transduction and expression checks, 5-6 months in vivo observation to a 150-day endpoint, 3-4 months sequencing and analysis.
preregistered outcomes
{
  "truncation_lethal_and_control_lethal": "hypothesis supported; the partner is dispensable for transformation",
  "truncation_inert_at_150d_and_control_lethal": "hypothesis REFUTED - this is the pre-registered falsifier",
  "both_inert": "ASSAY FAILURE, uninformative; do not reinterpret as a negative",
  "truncation_gives_signature_and_occupancy_but_no_disease": "partial - sufficient for the programme, not for transformation; add NRAS G12D and repeat before concluding"
}
measurements
[
  {
    "quantity": "n_per_arm",
    "subject": "mice per experimental arm",
    "value": 8,
    "units": "mice",
    "n": 6,
    "universe": "six construct arms in the human CD34+ NSG-SGM3 experiment",
    "null": null,
    "note": "A CONVENTION, NOT A POWER CALCULATION. Dickerson et al.'s mouse numbers were not extracted by this beat, so the published assay's variance is unknown and this figure should be replaced by a real calculation before the experiment is run."
  },
  {
    "quantity": "endpoint",
    "subject": "observation period before calling a construct inert",
    "value": 150,
    "units": "days",
    "n": null,
    "universe": "set at roughly twice the 74-day median latency reported for TCF3::ZNF384 in this system",
    "null": 74,
    "note": "the null is the positive control's published median latency; an endpoint at twice that is the minimum that makes 'inert' a claim rather than an absence of follow-up"
  }
]
How this could bite us back
  • This experiment is a laboratory study with no direct patient intervention, so the bite-back is about what a positive result would license. If the truncation transforms, the immediate clinical consequence is diagnostic rather than therapeutic: ZNF384-signature cases would need read-level ZNF384 reanalysis as standard, which is a workflow change with a real false-positive cost in a gene where indel calling is already hard.
  • A negative result also has a cost that should be anticipated: it would be read as vindicating partner-directed therapy, and it would not. The partner being necessary for transformation in a mouse does not make it druggable, and EP300::ZNF384 - the commonest fusion - still contains no catalytic domain to aim at.
  • Animal-welfare cost is substantial: six arms, two species, at least 8 mice per arm, to a 150-day endpoint. The cheaper deposited-data reanalysis named in the open questions should be done first, and this experiment should be conditioned on it.
Still open
  1. Should the truncation be expressed from the endogenous ZNF384 locus by base or prime editing rather than from a vector? The indels are heterozygous at roughly 50% allelic frequency in patients, and over-expression from a vector is a different perturbation. Editing would be slower and much more convincing, and it is the version to run if the vector experiment is positive.
  2. Is the human CD34+ arm or the mouse arm the better model? They disagree about which fusion transforms. Running both is the conservative choice made here, but if they disagree again about the truncation, the experiment will not have settled anything - which is a design weakness worth naming before spending the money.
  3. Can a cheaper partial test be run first? Zaliova's three cases have RNA-seq, and Dickerson and Zhao both deposited occupancy data. Comparing the indel cases against canonical cases at the level of the known ZNF384-fusion targets - SALL4, ID2, RGS1, CREB5, FLT3 - costs nothing and would raise or lower the prior before any mouse is used.
  4. What is the right positive readout if the disease does not develop? The expression signature and z-FLT3 occupancy are proposed here, but neither has a published threshold for 'matches'. Defining that threshold in advance, against the between-patient variability in the canonical cases, is a prerequisite and has not been done.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The required system already exists and is validated: Dickerson et al. expressed ZNF384 fusion oncoproteins and wild-type ZNF384 in mouse and human CD34+ HSPCs, transplanted the human cells into NSG-SGM3 mice, and found that TCF3::ZNF384 - but not wild-type ZNF384 and not the other ZNF384 fusions - produced lethal disease at a median latency of 74 days with CD33, CD19 and MPO expression consistent with human B/myeloid MPAL; in mouse HSPCs a cooperating lesion such as NRAS G12D was required for fully penetrant leukaemia, except for EP300::ZNF384 which alone produced serially transplantable leukaemia.

    ZNF384 Fusion Oncoproteins Drive Lineage Aberrancy in Acute Leukemia. · Dickerson KM, Qu C, Gao Q, Iacobucci I, Gu Z, Yoshihara H, Backhaus EA, Chang Y, Janke LJ, Xu B, Wu G, Papachristou EK, D'Santos CS, Roberts KG, Mullighan CG · Blood Cancer Discov 2022
    PMID 35247902DOI
    What this evidence cannot show

    Read from the PMC rendering of PMC9414741; mouse numbers, statistics and construct titres were not extracted, so the power of the existing assay is unknown to this beat and the n = 8 per arm proposed here is a convention rather than a calculation from the source. NSG-SGM3 is cytokine-humanised and biases toward myeloid output. The wild-type ZNF384 negative control in that paper is exactly the control this experiment needs and is the reason the system is the right one.

    Evidence that argues against this (1)
    • The system's own internal inconsistency is a design risk for this experiment: EP300::ZNF384 is the unique positive in the mouse HSPC arm and is NEGATIVE in the human CD34+ arm, where TCF3::ZNF384 is the unique positive. An assay whose two arms disagree about which fusion transforms may also disagree about whether a truncation transforms, which is why this design runs both species in parallel and pre-registers 'both inert' as uninformative rather than negative. Dickerson KM et al. 2022
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The three constructs to be tested correspond to real human lesions with published annotations: ZNF384 p.481_516 > SFQDYCACQ (the out-of-frame ZNF384-TEX41 fusion, retaining all zinc fingers), ZNF384 p.393_397 KHNPP>RAG (c.1179_1192del14insGGGCAGGG, 47% allelic frequency), and ZNF384 p.483_484 CS>YFEWDPND (c.1448_1451del4insATTTTGAGTGGGACCCGAACGA, 52% in RNA-seq and 43% in whole-exome sequencing); all three patients had the ZNF384-rearranged expression signature and myeloid-marker expression, and two of three were classified as MPAL.

    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. The protein annotations are as printed in the Results; the paper's supplemental material, which it says holds the detailed molecular analysis, was NOT retrieved, so anyone building these constructs should read it before ordering synthesis. The three lesions are from three patients and none has had any functional work done.

    Evidence that argues against this (1)
    • These are three patients out of twelve ZNF384-aberrant cases in one national cohort, and the authors explicitly decline to infer causality from them. Building an experiment on them presumes they are primary drivers, which is the very thing the experiment is meant to test - so the design must be able to return 'these lesions are inert', and the pre-registered outcome table above says so. A less charitable reading of the whole programme is that three unreplicated cases are a thin basis for a mouse study, and the answer to that is that the constructs are cheap and the question is central. Zaliova M et al. 2021
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The occupancy readouts for this experiment are already defined by published data: ZNF384 CUT&RUN in a TCF3-ZNF384 PDX identified 10,470 binding peaks, and the fusion protein activates and loops a 500 bp intergenic enhancer at hg38 chr13:28,124,365-28,124,868, 25 kb 5' of the FLT3 promoter, which drives 12.7-fold luciferase activation in the EP300-ZNF384 line JIH5 and is closed by ATAC-seq in other ALL subtypes including KMT2A-AF4.

    PMID 36104354DOI
    What this evidence cannot show

    FULL TEXT READ via Europe PMC fullTextXML for PMC9474531. The 10,470-peak reference set comes from ONE PDX sample with ONE partner, so using it as a benchmark for a truncation construct compares across systems as well as across constructs; a same-experiment TCF3::ZNF384 arm is therefore essential and is in the design. THE PEAK-PARTITION ARITHMETIC IN THIS PAPER DOES NOT RECONCILE - promoter plus enhancer peaks sum to 11,470 against a stated total of 10,470, and the H3K27ac/H3K4me3 overlap percentages do not match the counts - so the total should be used as an order-of-magnitude benchmark and no derived fraction from that partition should be used at all.

    Evidence that argues against this (1)
    • The antibody problem is unresolved and it bites this design directly: no reagent distinguishes fusion ZNF384 from wild-type ZNF384, which Zhao et al. state explicitly. A CUT&RUN in cells expressing a truncation construct will pull down both the construct and the endogenous protein, so peak recovery cannot be attributed cleanly. The design should therefore epitope-tag every construct, as Dickerson et al. did with V5-HA on their knockin, and profile the tag rather than ZNF384. Zhao X et al. 2022
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The both-halves-required control in the founding paper tested wild-type CIZ/ZNF384, not a C-terminally truncated ZNF384: constructs expressing exclusively the EWSR1 or TAF15 part of the fusion, or wild-type CIZ, gave a normal phenotype in NIH3T3, demonstrating that transformation depends on the presence of both halves.

    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 AACR HighWire page dated 2009-09-29; no DOI exists for this pre-DOI article. The result is reported as 'not shown', i.e. the control data are not in the paper, so the strength of the negative cannot be assessed - no expression check, no n, no figure. expected_absent_terms names 'truncated' because the ABSENCE of a truncation construct is the point of citing this item.

    Evidence that argues against this (1)
    • A 'not shown' control is weak evidence in both directions. It is cited here to establish that the truncation was never tested, which is a claim about absence and is safe; but it also means the wild-type-CIZ negative that this experiment must reproduce may itself not be solid. If wild-type ZNF384 turns out to transform in a modern HSPC assay - which Dickerson's negative for wild-type ZNF384 in human CD34+ cells argues against, and which is the better-documented negative - then the 2002 control would be superseded rather than confirmed. Dickerson KM et al. 2022
Connected to