Oncofusion

start / graph / target-flt3-in-znf384-rearranged-leukaemia

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

FLT3 in ZNF384-rearranged leukaemia - the fusion builds its own enhancer 25 kb upstream, and gilteritinib works in xenografts

also called z-FLT3 enhancer · FLT3 in ZNF384r ALL · gilteritinib in ZNF384-rearranged ALL

The ZNF384 fusion protein switches on a growth-signalling gene called FLT3 by activating a stretch of DNA 25,000 letters away and physically looping it round to the gene. Because the leukaemia then depends on FLT3, an existing FLT3-blocking drug slows it down in mice.

The same thing, for a scientist

Profiling 1,988 B-ALL transcriptomes, Zhao et al. found the highest FLT3 expression in **ZNF384-rearranged (n = 49) and KMT2A-rearranged (n = 136)** cases - the two subtypes side by side [E1]. The mechanism is not mutation: FLT3 mutation was present in only **5 of 66 (7.6%)** ZNF384-rearranged cases [E1]. Instead the fusion protein activates an **intergenic enhancer 25 kb 5' of the FLT3 promoter** - a 500 bp element at **hg38 chr13:28,124,365-28,124,868** marked by H3K27ac and H3K4me3, open by ATAC-seq only in ZNF384-rearranged samples, and driving **12.7-fold** luciferase activation in the EP300-ZNF384 line JIH5 - and ChIA-PET with a ZNF384 antibody shows the **enhancer-promoter loop is mediated by the fusion protein itself** [E2]. Globally, ZNF384 CUT&RUN in a TCF3-ZNF384 PDX gave **10,470 binding peaks**, with substantial overlap with H3K27ac [E2]. Therapeutically: gilteritinib LC50 was **3.9 nM in JIH5** and **44 nM in the KMT2A-rearranged line SEM**, the most sensitive two of eleven ALL lines, with LC50 negatively correlated with FLT3 mRNA at r = -0.9 (P = 0.0002); CRISPR knockout of the fusion cut FLT3 expression by 48% and reduced gilteritinib sensitivity roughly **100-fold** (P = 0.006); and oral gilteritinib at 10 mg/kg prolonged leukaemia-free survival in two ZNF384-rearranged xenografts (P = 0.0027 and P = 0.0026) [E3]. Independently, Dickerson et al. saw sensitivity of a ZNF384-rearranged xenograft to FLT3 inhibition in vivo [E4]. **The bite-back is severe and is recorded below: the closest clinical precedent, a FLT3 inhibitor added for KMT2A-rearranged infant ALL, failed.**

Go deeper
The element, and why it is the interesting part

Verbatim 2:

"Importantly, a prominent ZNF384 binding site was identified in the 5' distal region of the FLT3 gene, 25 kb from FLT3 promoter (Fig. 2c). This region was also marked with sharp H3K27ac and H3K4me3 modifications, characteristic of an active enhancer. Using the luciferase assay in JIH5 cells, this 500 bp segment (hg38, chr13:28,124,365-28,124,868) increased transcription activity by 12.7-fold compared to vector control"

"By contrast, this enhancer region was completely void of ATAC-seq signals in ALL samples of other molecular subtypes (TCF3-PBX1, DUX4 rearrangement, ETV6-RUNX1, and KMT2A-AF4), suggesting a causal effect of ZNF384 fusion protein... Notably, this z-FLT3 enhancer was not observed in other cancer cell lines in the ENCODE dataset"

Note the comparator list includes KMT2A-AF4, and the element is closed there. So KMT2A::AFF1 and ZNF384 fusions both over-express FLT3, and they do it through different cis elements. That is a genuine mechanistic dissociation between the two families, and it is recorded in ana-q173-the-inverse-architecture-framing-survived-one-of-four-tests.

The loop was demonstrated directly:

"To explore the involvement of ZNF384 fusion in the complex looping architecture, we also performed ChIA-PET using ZNF384 antibody and directly confirmed that this enhancer-promoter looping was mediated by the fusion protein in PDX-derived ALL blasts and in JIH5 cells."

The arithmetic in the genome-wide occupancy numbers does not close, and it is recorded here rather than smoothed

The paper reports, verbatim 2:

"a total of 10,470 ZNF384 binding peaks were identified, with 4820 peaks in promoter regions (i.e., +/-2 Kb of transcription start site) and 6650 in putative enhancer regions (i.e., intronic or intergenic). Of ZNF384 binding sites, 6750 (65.5%) and 4602 (34.5%) overlapped with H3K27ac and H3K4me3 marks, respectively"

4,820 + 6,650 = 11,470, not 10,470. And against the stated total of 10,470, 6,750 is 64.5% (not 65.5%) and 4,602 is 44.0% (not 34.5%); the two stated percentages also sum to exactly 100%, which cannot be right for two overlapping mark sets.

This does not touch the enhancer result, which rests on a named coordinate, a luciferase number and a ChIA-PET experiment. It does mean no derived quantity should be computed from the genome-wide peak partition, and this entry stores the raw counts with the discrepancy attached rather than a tidied version of them. Recording it is cheaper than a future beat re-deriving 34.5% and building on it.

Facts and figures
target geneFLT3
fusionZNF384 rearrangement (EP300::ZNF384 and TCF3::ZNF384 both tested)
modalitysmall-molecule type-I FLT3 tyrosine kinase inhibitor
druggilteritinib
mechanismfusion-protein-mediated activation of an intergenic enhancer 25 kb 5' of FLT3, looped to the FLT3 promoter
enhancer coordinates hg38chr13:28,124,365-28,124,868
enhancer length bp504
measurements
percent_change-48.0percent_change: -48correlation-0.900correlation: -0.9proportion0.076proportion: 0.076lc503.900lc50: 3.9fold_activation12.7fold_activation: 12.7lc5044.0lc50: 44fold_change100.0fold_change: 100
assays
[
  "RNA-seq",
  "CUT&RUN (ZNF384, H3K27ac, H3K4me3)",
  "ATAC-seq",
  "luciferase reporter",
  "ChIA-PET (CTCF, RNA Pol II, ZNF384)",
  "CRISPR/Cas9 fusion knockout",
  "MTT drug sensitivity",
  "xenograft"
]
cell lines
[
  "JIH5 (EP300-ZNF384, DSMZ ACC 788)",
  "SEM (KMT2A-rearranged, DSMZ ACC 546)",
  "Nalm6",
  "697",
  "UOCB-1",
  "REH",
  "RPMI8402",
  "CEM",
  "Jurkat",
  "DND41",
  "MOLT4",
  "GM12878 (fusion-negative control)"
]
pdx
[
  "TCZ (TCF3-ZNF384 PDX, Zhao et al.)",
  "SJBALL004031 (Dickerson et al.)",
  "SJMPAL046472 (Dickerson et al.)"
]
measurements — the numbers behind the chart
[
  {
    "quantity": "lc50",
    "subject": "gilteritinib in JIH5 (EP300-ZNF384 ALL line)",
    "value": 3.9,
    "units": "nM",
    "n": 1,
    "universe": "11 ALL cell lines spanning DUX4, KMT2A-r, TCF3-PBX1, TCF3-HLF, ETV6-RUNX1, ZNF384-r and T-ALL, 72 h exposure",
    "null": 44,
    "note": "the natural comparator is not vehicle but SEM, the KMT2A-rearranged line, at 44 nM - the second most sensitive of the eleven. An 11-fold gap between the two FLT3-high subtypes."
  },
  {
    "quantity": "lc50",
    "subject": "gilteritinib in SEM (KMT2A-rearranged ALL line)",
    "value": 44,
    "units": "nM",
    "n": 1,
    "universe": "same 11-line panel, 72 h exposure",
    "null": null,
    "note": "recorded because this graph's index fusion is KMT2A::AFF1 and SEM is a KMT2A::AFF1 line; the closest clinical test of this idea, lestaurtinib in KMT2A-r infant ALL, failed"
  },
  {
    "quantity": "correlation",
    "subject": "gilteritinib LC50 versus FLT3 mRNA across ALL cell lines",
    "value": -0.9,
    "units": "Pearson r",
    "n": 11,
    "universe": "11 ALL cell lines; P = 0.0002. Surface FLT3 protein gave r = -0.8364, P = 0.0013 over the same panel.",
    "null": 0,
    "note": "an r of -0.9 on n = 11 is being driven by the two FLT3-high lines at one end; with SEM and JIH5 removed the panel has little FLT3 range left"
  },
  {
    "quantity": "fold_activation",
    "subject": "500 bp z-FLT3 enhancer in a luciferase reporter in JIH5 cells",
    "value": 12.7,
    "units": "fold over empty vector",
    "n": null,
    "universe": "hg38 chr13:28,124,365-28,124,868 cloned into a reporter, assayed in the EP300-ZNF384 line JIH5",
    "null": 1,
    "note": "null is vector control by construction. The element was NOT tested in a ZNF384-fusion-negative line in what this beat read, so the reporter alone does not establish fusion-dependence; the ATAC-seq and ChIA-PET do."
  },
  {
    "quantity": "percent_change",
    "subject": "FLT3 expression after CRISPR/Cas9 knockout of EP300-ZNF384 in JIH5",
    "value": -48,
    "units": "% change",
    "n": null,
    "universe": "JIH5 cells, CRISPR/Cas9-mediated editing of the fusion",
    "null": 0,
    "note": "a 48% reduction, not abolition - FLT3 retains half its expression with the fusion knocked out, so the fusion is one input to this locus and not the only one"
  },
  {
    "quantity": "fold_change",
    "subject": "JIH5 sensitivity to gilteritinib after fusion knockout",
    "value": 100,
    "units": "fold decrease in sensitivity",
    "n": null,
    "universe": "JIH5 CRISPR fusion knockout versus parental; P = 0.006; the paper states '~100-fold'",
    "null": 1,
    "note": "this is the causal link between the fusion and the drug response and it is much larger than the 48% change in FLT3 mRNA, which is worth noticing rather than explaining away"
  },
  {
    "quantity": "proportion",
    "subject": "FLT3 mutation among ZNF384-rearranged ALL cases",
    "value": 0.076,
    "units": "fraction",
    "n": 66,
    "universe": "ZNF384-rearranged cases pooled across a 1,988-patient B-ALL cohort and a 377-patient paediatric Asian ALL cohort; 5 of 66",
    "null": null,
    "note": "this is the number that makes the mechanism epigenetic rather than genetic: FLT3 is high without being mutated"
  }
]
How this could bite us back
  • RESISTANCE BY LINEAGE, NOT BY MUTATION. This subtype is B/myeloid and already crosses lineage under pressure - the Ponte di Legno authors report lineage switch to myeloid leukaemia after CAR-T in a TCF3-ZNF384 patient. A fusion-maintained enhancer loop is exactly the kind of dependency that a state change can dissolve without any mutation, so relapse on gilteritinib may present with wild-type FLT3, a closed z-FLT3 element and a different phenotype. A resistance panel that sequences FLT3 will report nothing.
  • THE CLOSEST CLINICAL TEST ALREADY FAILED. COG AALL0631 added lestaurtinib to chemotherapy for KMT2A-rearranged infant ALL - the other FLT3-high immature B-lineage subtype named in the same paper - and it did not improve event-free or overall survival. Anyone proposing gilteritinib for ZNF384-rearranged disease has to say what is different, and 'a lower LC50 in one cell line' is not yet an answer.
  • ESSENTIAL-FUNCTION TOXICITY. FLT3 is required for normal lymphoid and dendritic-cell development from haematopoietic progenitors, and this subtype's founding lesion is present in HSCs and multiple progenitor populations, so the drug is being aimed at a receptor the patient's normal stem-cell compartment also uses. Myelosuppression on top of chemotherapy in a paediatric induction is the predictable cost.
  • PARTIAL TARGET COVERAGE. Knocking out the fusion removed only 48% of FLT3 expression, so half the FLT3 in these cells comes from somewhere else. A therapy premised on the fusion driving FLT3 is premised on a driver that accounts for about half the signal.
  • DELIVERY AND SELECTION. Gilteritinib is oral and approved in FLT3-mutated AML, so the pharmacology is not the obstacle - patient selection is. ZNF384-rearranged status is the proposed biomarker, and it is exactly the status that the last-exon-indel class in mechanism-znf384-c-terminal-truncation-phenocopies-the-fusion is negative for on every routine assay.
Still open
  1. Does the z-FLT3 enhancer exist in patients with the noncanonical ZNF384 lesions - the last-exon indels and the ZNF384::TEX41 truncation - where there is no fusion partner to supply an activation domain? If a truncated ZNF384 with no partner still opens this element, the element is a ZNF384 property rather than a fusion property, and the whole 'gained coactivator' account of the subtype weakens. If it does not, these patients should not be offered a FLT3 inhibitor. Nobody has looked, and the two patients exist.
  2. Why is JIH5 eleven-fold more gilteritinib-sensitive than SEM when both are FLT3-high? If the answer is FLT3 dependence rather than FLT3 level, that is the argument for trying in ZNF384-r what failed in KMT2A-r. If the answer is cell-line idiosyncrasy at n = 1 per subtype, it is not.
  3. Is FLT3 expression in ZNF384-rearranged leukaemia separable from differentiation stage? Zhao et al. report FLT3 tracking B-cell maturation, high in the immature subtypes and low in the mature ones, and ZNF384-r and KMT2A-r are both immature. A depth- and stage-matched comparison would say whether the fusion adds anything on top of the stage - and this project has already been caught once treating a stage-confounded ranking as a fusion effect.
  4. Does the enhancer survive treatment? Both xenograft experiments started drug within days of engraftment. If the fusion-maintained loop is reversible, resistance could arise by closing the element and switching to a different survival input rather than by mutating FLT3 - which would be invisible to a FLT3 sequencing panel at relapse.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Across 1,988 B-ALL transcriptomes (1,610 children and 378 adults), the highest FLT3 expression was observed in 49 cases with ZNF384 rearrangements and 136 with KMT2A fusion genes; FLT3 mutation was uncommon in ZNF384-rearranged ALL, found in 5 of 66 cases (7.6%) across two cohorts.

    PMID 36104354DOI
    What this evidence cannot show

    FULL TEXT READ via Europe PMC fullTextXML for PMC9474531. Cohort 1 is a published RNA-seq dataset of 1,988 B-ALL patients described as mostly of European descent; cohort 2 is 377 paediatric Asian ALL cases. The 5/66 mutation figure pools the two cohorts. FLT3 expression is mRNA from bulk RNA-seq, not surface protein, and the paper separately notes that FLT3 expression tracks B-cell differentiation stage - higher in the immature subtypes KMT2A-r and ZNF384-r, lower in ETV6-RUNX1, TCF3-PBX1 and MEF2D-r - so maturation stage is a live confounder for the between-subtype comparison.

    Evidence that argues against this (1)
    • The paper states plainly that FLT3 over-expression is NOT specific to these two subtypes: 'It is important to note that FLT3 overexpression is also observed in ALL cases without KMT2A or ZNF384 rearrangements (Fig. 1a). For example, microdeletion at this locus in hyperdiploid ALL creates enhancers that directly activate FLT3 transcription.' And because FLT3 expression tracks differentiation stage, the ZNF384-r/KMT2A-r ranking is partly confounded with immaturity rather than with the fusion. High FLT3 therefore does not identify the subtype and the subtype is not the only route to high FLT3. Zhao X et al. 2022
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    A 500 bp intergenic element at hg38 chr13:28,124,365-28,124,868, 25 kb 5' of the FLT3 promoter, is bound by ZNF384, carries H3K27ac and H3K4me3, is open by ATAC-seq only in ZNF384-rearranged samples (and closed in TCF3-PBX1, DUX4-rearranged, ETV6-RUNX1 and KMT2A-AF4), drives 12.7-fold luciferase activation in JIH5 cells, and loops to the FLT3 promoter in a contact shown by ChIA-PET with a ZNF384 antibody to be mediated by the fusion protein. Genome-wide, ZNF384 CUT&RUN in a TCF3-ZNF384 PDX identified 10,470 binding peaks.

    PMID 36104354DOI
    What this evidence cannot show

    FULL TEXT READ (PMC9474531). The CUT&RUN was performed in ONE TCF3-ZNF384 PDX sample (TCZ) and repeated in ONE EP300-ZNF384 cell line (JIH5), so the genome-wide occupancy map rests on n = 2 systems and one partner each. The ATAC-seq comparison used two EP300-ZNF384 primary samples. THE PEAK-PARTITION ARITHMETIC IN THIS PAPER DOES NOT CLOSE: 4,820 promoter plus 6,650 enhancer peaks is 11,470 against a stated total of 10,470, and the H3K27ac/H3K4me3 overlap percentages (65.5% and 34.5%) do not match the counts given (6,750 and 4,602) against that total and sum to exactly 100%. The enhancer result does not depend on any of these numbers; no derived quantity should be computed from them.

    Evidence that argues against this (1)
    • The ChIA-PET antibody cannot distinguish the fusion protein from wild-type ZNF384, because both carry the same C-terminal zinc-finger region the antibody recognises and the paper states elsewhere that the reagents to separate them do not exist: 'this needs to be verified by pull-down assays using antibodies specific to wildtype ZNF384 or the fusion protein, respectively. If these reagents were to become available in the future...' Alexander et al. did run the discriminating experiment in a different system - ChIP-seq of fusion versus wild-type ZNF384, finding 2,298 peaks with new or increased fusion binding and 495 reduced - which supports the general claim of perturbed binding but is not this locus and not these cells. Alexander TB et al. 2018
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Gilteritinib LC50 across eleven ALL cell lines was lowest in JIH5 (EP300-ZNF384, 3.9 nM) and SEM (KMT2A-rearranged, 44 nM), with LC50 negatively correlated with FLT3 mRNA (r = -0.9, P = 0.0002) and with surface FLT3 protein (r = -0.8364, P = 0.0013); CRISPR/Cas9 knockdown of the EP300-ZNF384 fusion in JIH5 decreased FLT3 expression by 48% and reduced gilteritinib sensitivity roughly 100-fold (P = 0.006); and oral gilteritinib at 10 mg/kg daily, started three days after engraftment, delayed leukaemia progression and prolonged leukaemia-free survival in two ZNF384-rearranged xenograft models (P = 0.0027 and P = 0.0026).

    PMID 36104354DOI
    What this evidence cannot show

    FULL TEXT READ (PMC9474531). THE IN VIVO RESULT HAS NO EFFECT SIZE: the paper reports only p-values for progression (P < 0.005) and leukaemia-free survival (P = 0.0027, P = 0.0026), with no median survival in days and no fold reduction in burden, so 'gilteritinib works' cannot be quantified from this paper. Treatment started three days after injection, i.e. against minimal disease, not established leukaemia. Gilteritinib is a monotherapy here and only two xenograft models were used, one of them the JIH5 cell line rather than a patient sample. The ex vivo primary-sample arm included only three EP300-ZNF384 cases among 47.

    Evidence that argues against this (1)
    • The clinical precedent in the adjacent subtype failed, and the paper says so itself: 'the COG AALL0631 trial tested the effects of lestaurtinib in infants with KMT2A-r ALL, and unfortunately the addition of this FLT3 inhibitor failed to improve event-free or overall survival.' KMT2A-rearranged infant ALL is the closest clinical analogue - high FLT3 by the same paper's own ranking, an immature B-lineage leukaemia, a FLT3 inhibitor added to chemotherapy - and it did not work. That is the strongest available prior against the ZNF384 proposal, and it is stronger than any preclinical in-vivo p-value in the other direction. Zhao X et al. 2022
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Independently of Zhao et al., a ZNF384-rearranged patient-derived xenograft was sensitive to FLT3 inhibition in vivo with gilteritinib: vehicle-treated mice were moribund at 5 weeks post-transplant whereas gilteritinib-treated mice had a reduced tumour burden.

    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; the result is tied to Figure 7C-E, hence support_location 'figure'. This is a second independent group reaching the same in vivo conclusion with the same drug, which is the reason it is cited separately - but it is the same institution's programme as neither paper's PDX panel is large, and again no survival effect size is quoted here. The named PDX lines in that paper are SJBALL004031 and SJMPAL046472.

    Evidence that argues against this (1)
    • Two independent xenograft results with the same drug in the same subtype is real corroboration, and this beat found no published negative in vivo FLT3-inhibitor result in ZNF384-rearranged leukaemia. The honest statement of the gap is that nobody has published a NEGATIVE, and that a preclinical literature with two positives and no negatives is exactly the publication pattern a subtype-plus-drug combination produces regardless of whether it will work in patients. A search of ClinicalTrials.gov for a gilteritinib trial selecting on ZNF384 status was not run by this beat. Zhao X et al. 2022
Connected to