An H3K4me3-reading PHD finger fused to NUP98 makes a potent leukaemia oncoprotein; the H3K4me3-reading PHD finger of KMT2A is deleted from every KMT2A fusion and reinstating it abolishes transformation. Same module class, opposite sign - and the one NUP98 fusion that carries KMT2A's PHD cassette is the weakest member of its family.
also called PHD finger polarity · the reader that transforms and the reader that suppresses · KMT2A PHD3 versus KDM5A PHD3 · NUP98-PHD
A small protein module called a PHD finger reads a chemical tag on chromatin that marks genes as switched on. Bolt that module onto the NUP98 partner and you get a fast, aggressive leukaemia, and breaking its ability to read the tag stops the leukaemia entirely. But the very similar module belonging to KMT2A is thrown away by every KMT2A fusion, and putting it back destroys the fusion's ability to transform cells - so the same kind of reader is required in one family and forbidden in the other.
The same thing, for a scientist
Wang et al. showed that fusing an **H3K4me3/me2-binding PHD finger** - from either PHF23 or JARID1A/KDM5A - to NUP98 produces potent oncoproteins that arrest haematopoietic differentiation and induce AML in mice, and that **PHD-finger mutations abrogating H3K4me3 binding abolish leukaemic transformation**; mechanistically these fusions act as chromatin boundary factors, dominating over Polycomb silencing to lock HOX, GATA3, MEIS1, EYA1 and PBX1 into an active H3K4me3-plus-acetylation state [E1]. **KMT2A has an H3K4me3-reading PHD finger too - PHD3 - and it is deleted from every KMT2A translocation product.** Chen et al. showed the deletion is not incidental: inserting the third PHD finger into MLL-ENL restores CyP33 and HDAC1 recruitment, drives CyP33-dependent downregulation of HOXC8, and **blocks the fusion's haematopoietic stem cell immortalisation**, while inserting PHD fingers 1 and 2 does nothing [E2]. **The same module class is therefore load-bearing in opposite directions in two fusion families**, and this graph already holds both halves without ever having put them side by side. The determinant is not the mark and not the fold: it is what else docks on the finger. KMT2A PHD3 carries a CyP33/PPIE RRM-binding surface that competes with H3K4me3 and converts the reader into an HDAC-recruiting repressor; the KDM5A and PHF23 fingers have no such switch. **And there is a natural experiment already in this graph that nobody has read this way.** NUP98::KMT2A retains KMT2A PHD1-4 along with CXXC, the WIN motif and an intact SET domain, and it is the **weakest NUP98 fusion on record** - median latency 80 weeks in an inducible mouse, **low** rather than high HOXA cluster expression, menin- and BET-inhibitor resistant, and interfering with cell-cycle progression rather than primarily blocking differentiation [E3]. A NUP98 fusion carrying an H3K4me3-reading PHD finger ought to be fast; this one is not, and the finger it carries is the one with the off-switch attached.
The 2x2 that has never been completed
| PHD finger | on a NUP98 scaffold | on a KMT2A fusion scaffold |
|---|---|---|
| KDM5A C-terminal PHD (H3K4me3 reader, no CyP33 site) | potent AML in mice; H3K4me3-binding mutation abolishes transformation 1 | never tested |
| PHF23 C-terminal PHD (H3K4me3 reader) | potent AML in mice 1 | never tested |
| KMT2A PHD3 (H3K4me3 reader with a CyP33/PPIE RRM-binding surface) | never tested in isolation; present as part of the whole PHD1-4 cassette in NUP98::KMT2A, which has 80-week latency, low HOXA and menin resistance 3 | reinsertion abolishes immortalisation; PHD1+2 reinsertion does not 2 |
Two cells of this table are empty and both are cheap. Fuse the KMT2A PHD3 alone to the NUP98 FG repeats and ask whether it transforms like the KDM5A finger or fails. Insert the KDM5A C-terminal PHD into MLL-ENL and ask whether it suppresses like KMT2A PHD3 or is inert. Those two constructs would settle whether the suppressive activity is a property of the finger (and therefore of CyP33 docking) or a property of the scaffold (KMT2A fusions cannot tolerate any reader in cis).
That distinction is not academic for this project. If the suppression is a property of the finger, then a molecule that restores the PHD3-CyP33 module in trans - or that mimics its HDAC-recruiting output at KMT2A-fusion target loci - is a fusion-selective mechanism entirely orthogonal to menin, and it would be expected to work in exactly the cells where a MEN1 interface mutation has removed the menin option.
Why NUP98::KMT2A is the most informative row and also the most confounded
NUP98::KMT2A is the closest thing to the missing experiment that exists in nature. It carries the NUP98 FG repeats and KMT2A PHD1-4, and in an inducible transgenic mouse it produced disease in over 90% of animals but with a median latency of 80 weeks, anaemia and reduced marrow cellularity, multilineage dysplasia, LSK expansion with a competitive repopulation advantage, and only 6 of 22 mice developing transplantable AML; the blasts did not overexpress the HOXA cluster, were resistant to MI-2-2 and to JQ1 at concentrations where KMT2A-AF9 cells arrest, and the authors concluded the fusion interferes with cell-cycle progression rather than primarily blocking differentiation 3.
Set that against the NUP98-PHD fusions of Wang et al., which arrest differentiation and induce AML by locking HOX loci open 1. Two NUP98 fusions, both carrying an H3K4me3-reading PHD finger, and opposite HOXA output.
The confound is large and must be stated first. NUP98::KMT2A deletes KMT2A exon 1 and therefore the menin-binding motif at residues 6-25 and the LEDGF IBM motifs, and it retains CXXC, the WIN motif and an intact SET domain. Any of those differences could explain the phenotype on its own, and this graph's own fusion-nup98-kmt2a reads the result primarily as evidence for the menin-interface model - correctly, because losing the menin motif is the most parsimonious explanation for menin-inhibitor resistance and low HOXA.
So this entry does not claim the PHD cassette causes the weakness. It claims something narrower and testable: the PHD-transformation rule stated by Wang et al. - an H3K4me3-reading PHD finger on NUP98 gives potent AML - has an apparent exception sitting in this corpus, and the exception carries the one H3K4me3-reading PHD finger in the haematopoietic literature that is known to be switchable to a repressor. Deciding between the menin explanation and the PHD explanation requires the isolated construct, and it has not been built.
What this adds to entries that already exist
All three components are already in this graph and none of them touches the others:
fusion-nup98-kdm5astates the Wang result and draws the reader-versus-writer contrast within the NUP98 family (KDM5A PHD versus NSD1 SET). It does not mention KMT2A's own PHD cassette.gene-ppiestates the CyP33/PHD3 mechanism precisely - RRM binding at about 2 uM by ITC, competitive with H3K4me3 at a distinct site on the same domain, converting PHD3 into an HDAC-recruiting repressor, deleted from every KMT2A translocation - and calls it 'an interactome entry about a lost contact'. It does not mention NUP98.fusion-nup98-kmt2arecords the 80-week latency and the menin resistance and reads them through the menin motif. It does not note that the fusion also carries PHD1-4, or that this makes it an exception to the NUP98-PHD rule.
The analogy is the join, and the join generates an experiment none of the three entries implies on its own. Chen et al. (PMID 18676843) is cited in 11 node files and Wang et al. (PMID 19430464) in one; a reader-reuse check would have flagged the first as load-bearing, which is why this entry is framed as a juxtaposition rather than as a new fact about either paper.
| domain | engineering |
|---|---|
| where the analogy breaks | NUP98::KMT2A is not a clean test of the PHD hypothesis, because it simultaneously loses the KMT2A menin-binding motif and retains CXXC, the WIN motif and the SET domain. The menin-motif loss alone predicts both its low HOXA output and its menin-inhibitor resistance, and this graph's existing entry reads it that way. The analogy therefore breaks at the point where it would be most useful - it cannot currently attribute the weak phenotype to the PHD cassette, and the literature contains no construct that isolates the variable. |
| what it predicts here | If PHD3's suppressive activity is a property of the finger (CyP33 docking) rather than of the KMT2A scaffold, then restoring the PHD3-CyP33 repressive module in trans at KMT2A-fusion target loci is a fusion-selective mechanism orthogonal to menin - and it should retain activity in cells carrying MEN1 interface mutations, which is the population menin inhibitors currently cannot serve. |
| epigenetic axis | H3K4me3 reading; the same mark is the address in both families and the consequence of reading it is opposite |
| relationship to existing entries | fusion-nup98-kdm5a holds E1 and contrasts reader versus writer WITHIN the NUP98 family. gene-ppie holds the CyP33/PHD3 mechanism and calls it a lost contact. fusion-nup98-kmt2a holds E3 and reads it through the menin motif. No entry joins them, and the join is what produces the missing constructs. Chen et al. is cited in 11 node files and Wang et al. in one, checked on 2026-09-23. |
mapping
[
{
"bio": "KDM5A or PHF23 C-terminal PHD finger fused to NUP98: potent AML, and abolishing H3K4me3 binding abolishes transformation",
"analog": "KMT2A PHD3, also an H3K4me3 reader, is deleted from every KMT2A fusion, and reinserting it abolishes immortalisation"
},
{
"bio": "NUP98-PHD fusions act as chromatin boundary factors that dominate over Polycomb and lock HOX, MEIS1, GATA3, EYA1, PBX1 active",
"analog": "KMT2A fusions enforce the same HOXA9/MEIS1 programme, but do it through menin/LEDGF tethering and elongation-factor recruitment, with no reader module in the fusion at all"
},
{
"bio": "the transforming NUP98 PHD fingers have no known repressive co-factor docking site",
"analog": "KMT2A PHD3 carries a CyP33/PPIE RRM-binding surface, competitive with H3K4me3, that converts the reader into an HDAC-recruiting repressor - the off-switch the fusion throws away"
},
{
"bio": "prediction from the NUP98 rule: any H3K4me3-reading PHD finger on a NUP98 scaffold should transform",
"analog": "APPARENT EXCEPTION - NUP98::KMT2A carries PHD1-4 and has 80-week latency, LOW HOXA and menin/BET resistance; confounded by loss of the menin-binding motif"
},
{
"bio": "the decisive constructs: KMT2A PHD3 alone fused to NUP98 FG repeats; KDM5A C-terminal PHD inserted into MLL-ENL",
"analog": "NEITHER HAS BEEN BUILT - the 2x2 is half empty and both missing cells are single constructs"
}
]generated hypotheses
[]
untested constructs
[ "NUP98 FG repeats fused to KMT2A PHD3 alone - does it transform like the KDM5A finger, or fail?", "KDM5A C-terminal PHD inserted into MLL-ENL in place of KMT2A PHD3 - does it suppress, or is it inert?", "NUP98::KMT2A with a CyP33-binding-deficient PHD3 - does latency shorten and HOXA rise?" ]
- Any therapeutic strategy built on restoring the PHD3-CyP33 switch is restoring a repressive HDAC-recruiting module. HDAC recruitment is not selective for fusion target loci, and this graph already records that no HDAC-inhibitor result in this disease attributes to a specific enzyme - the delivery problem is the whole problem.
- The orthogonality claim cuts both ways: a mechanism that works where menin inhibitors fail is also a mechanism with no clinical precedent, no chemical matter and no biomarker, whereas the menin axis has an approved-path drug and a patient-derived resistance map.
- PHD fingers are among the least tractable reader modules; fusion-nup98-kdm5a already records that the one experimentally required interaction in NUP98::KDM5A has no chemical probe and no experimental structure. A programme aimed at PHD-finger chemistry starts from a worse position than the menin pocket did.
- If the suppression turns out to be a scaffold property rather than a CyP33 property, this entry's therapeutic implication evaporates entirely and only the comparative observation survives. That outcome is one construct away and should be run before any chemistry is contemplated.
- Does the KMT2A PHD3, fused alone to the NUP98 FG repeats, transform haematopoietic progenitors? Wang et al. tested PHF23 and JARID1A fingers; the one H3K4me3-reading PHD finger with a documented repressive switch was not among them, and it is a single construct.
- Does inserting the KDM5A C-terminal PHD into MLL-ENL suppress immortalisation the way KMT2A PHD3 does? If it does, the suppression is a scaffold property and this analogy is wrong. If it does not, the suppression is CyP33-dependent and the therapeutic route opens.
- Is the weak NUP98::KMT2A phenotype attributable to the menin-motif loss alone, or does the retained PHD1-4 cassette contribute? A CyP33-binding-deficient PHD3 version of the same transgene would separate them, and nobody has made it.
- Can the PHD3-CyP33 repressive module be delivered or mimicked in trans? The KMT2A fusion has thrown away its own off-switch, and unlike menin this route would not be defeated by a MEN1 interface mutation - which makes it the most interesting orthogonal mechanism this analogy generates.
- Does NUP98::KMT2A, which carries PHD1-4, show CyP33/PPIE association that the other NUP98 fusions cannot? A simple co-immunoprecipitation would say whether the off-switch is physically present on that fusion, and that is the cheapest experiment in this entry.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Fusing an H3K4me3-binding PHD finger - the C-terminal PHD finger of PHF23 or of JARID1A/KDM5A - to NUP98, as found in human leukaemias, generates potent oncoproteins that arrest haematopoietic differentiation and induce acute myeloid leukaemia in murine models; a PHD finger that specifically recognises H3K4me3/me2 is essential for leukaemogenesis and mutations in the PHD finger that abrogate H3K4me3 binding abolish leukaemic transformation; the fusions prevent differentiation-associated removal of H3K4me3 at loci encoding lineage-specific transcription factors (HOX genes, Gata3, Meis1, Eya1, Pbx1) and act as chromatin boundary factors that dominate over Polycomb-mediated silencing to lock those loci in an active H3K4me3-plus-acetylation state defining the leukaemia stem cell.
Haematopoietic malignancies caused by dysregulation of a chromatin-binding PHD finger. · Wang GG, Song J, Wang Z, Dormann HL, Casadio F, Li H, Luo JL, Patel DJ, Allis CD · Nature 2009What this evidence cannot show
Abstract read in full from the PubMed record on 2026-09-23. Murine haematopoietic stem/progenitor cells and murine leukaemia models; the human disease correlate is paediatric AMKL and AML, not a KMT2A rearrangement and not B-lineage. The PHD fingers tested are those of PHF23 and JARID1A/KDM5A; the abstract does not report a KMT2A PHD3 construct, and this entry treats that cell of the comparison as untested rather than as negative.
Evidence that argues against this (1)
- NUP98::KMT2A carries KMT2A PHD fingers 1-4 and is the weakest NUP98 fusion characterised, with 80-week median latency, low rather than high HOXA expression and menin/BET-inhibitor resistance - an apparent exception to the rule that an H3K4me3-reading PHD finger on a NUP98 scaffold produces potent HOXA-high leukaemia. The exception is heavily confounded by the concurrent loss of the KMT2A menin-binding motif, which is why this entry treats it as a question rather than a refutation. Fisher JN et al. 2020
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The three-PHD-finger cassette of MLL/KMT2A is absent from all MLL fusion proteins; inserting the third PHD finger into MLL-ENL allows recruitment of Cyp33 and subsequently HDAC1 to the fusion protein, mediates Cyp33-dependent downregulation of HOXC8, and blocks the haematopoietic stem cell immortalisation potential of the fusion in serial replating colony assays, whereas insertion of the first and second PHD fingers alone has no such effect; the authors conclude that CyP33 binding to PHD3 switches MLL from transactivation to repression and that loss of the PHD fingers combined with gain of a partner activation domain makes the fusion a constitutive transactivator.
Loss of MLL PHD finger 3 is necessary for MLL-ENL-induced hematopoietic stem cell immortalization. · Chen J, Santillan DA, Koonce M, Wei W, Luo R, Thirman MJ, Zeleznik-Le NJ, Diaz MO · Cancer Res 2008What this evidence cannot show
Abstract read in full from the PubMed record on 2026-09-23. The fusion is MLL-ENL, not KMT2A::AFF1, and the assay is serial replating of murine haematopoietic cells, which is an immortalisation surrogate rather than leukaemogenesis in vivo. Per this project's standing rule, the first clause of any claim built on this must say murine MLL-ENL. This paper is cited in 11 node files in this corpus, so its conclusion is not new here; what is new is placing it beside E1.
Evidence that argues against this (1)
- If the suppressive effect were a property of ANY reader module placed in cis on a KMT2A fusion rather than of PHD3's CyP33 switch specifically, then the Wang result would not be in tension with it at all and the analogy would collapse into a scaffold effect. The PHD1+2 negative control in the same paper argues against the generic-module reading, but the decisive test - the KDM5A PHD inserted into MLL-ENL - has not been done. Wang GG et al. 2009
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In inducible iNUP98-KMT2A transgenic mice, over 90% of animals developed disease after a median latency of 80 weeks with anaemia, reduced bone marrow cellularity, increased white cell numbers, extramedullary haematopoiesis and multilineage dysplasia; lineage-negative Sca-1+ c-Kit+ cells expanded and outcompeted wildtype cells in repopulation assays; six mice developed transplantable AML; as reported in patients, the leukaemic blasts did NOT express increased levels of the HOXA cluster, and in contrast to KMT2A-AF9 cells they were resistant to pharmacological targeting of menin and of BET proteins by MI-2-2 and JQ1 respectively; and the fusion accumulated cells in G1 and abrogated replicative senescence, leading the authors to conclude it interferes with cell-cycle progression rather than primarily blocking differentiation.
Transforming activities of the NUP98-KMT2A fusion gene associated with myelodysplasia and acute myeloid leukemia. · Fisher JN, Thanasopoulou A, Juge S, Tzankov A, Bagger FO, Mendez MA, Peters AHFM, Schwaller J · Haematologica 2020What this evidence cannot show
Abstract read in full from the PubMed record on 2026-09-23; the PubMed rendering strips several gene names from this abstract, and the retained-domain architecture (PHD1-4, CXXC, WIN, SET, with exon 1 and the menin-binding motif lost) is taken from this graph's fusion-nup98-kmt2a entry rather than re-derived here. The human disease is two adult AML patients, so this is a mouse model of a very rare lesion. Most importantly, the comparison to E1 is confounded: NUP98::KMT2A differs from NUP98::KDM5A in far more than its PHD finger, and the loss of the menin-binding motif is the more parsimonious explanation for both the low HOXA and the menin-inhibitor resistance.
Evidence that argues against this (1)
- This graph's fusion-nup98-kmt2a entry reads this result as a positive control for the menin-interface model and a negative control for the FG-repeat condensate model, which is a sound reading and is not disputed here. Under that reading the PHD cassette is irrelevant to the phenotype and the apparent exception to Wang et al. is an artefact of the menin motif, not a fact about PHD fingers. This entry does not claim otherwise; it claims the two readings are not currently distinguishable and names the construct that would distinguish them. Wang GG et al. 2009
- related to →NUP98::KDM5A (NUP98-JARID1A), t(11;12)(p15;p13)
the family where an H3K4me3-reading PHD finger is required for transformation
- related to →NUP98::KMT2A, inv(11)(p15q23) — the NUP98 fusion that contains KMT2A and is menin-inhibitor RESISTANT
the apparent exception - a NUP98 fusion carrying KMT2A PHD1-4 with 80-week latency and low HOXA
CyP33 docking on PHD3 is the proposed determinant of the opposite sign
- related to →KMT2A / MLL1 — the gene that breaks
the PHD cassette deleted from every KMT2A translocation product
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
the index fusion, which like every KMT2A fusion lacks PHD1-3
the source of the transforming PHD finger, whose catalytic domain is lost in the fusion