One demethylase family sits at both ends of this disease - KDM3B gates whether KMT2A rearranges, JMJD1C is what the resulting fusion needs to survive - and a single compound hits both, so its therapeutic index runs in opposite directions in the two cell populations it reaches
also called the KDM3 family paradox · one compound, two ends of the disease · KDM3B upstream, JMJD1C downstream
Two closely related enzymes bracket this leukaemia: losing one makes the KMT2A gene break in the first place, and the leukaemia that results then depends on the other to stay alive. The problem is that the only drug-like molecule available hits both of them, so it might treat the leukaemia in a patient while creating new ones in their healthy stem cells.
The same thing, for a scientist
**Statement.** The KDM3/JHDM2 family occupies both ends of KMT2A-rearranged leukaemia. **Upstream:** KDM3B keeps H3K9me1/2 low across the *KMT2A* breakpoint cluster region and CTCF bound at the exon-11 site; its loss - genetically, chemically, or via doxorubicin/etoposide - produces copy gains and rearrangements in cells, in primary cord-blood CD34+ HSPCs and in mice [E1]. **Downstream:** JMJD1C/KDM3C is a dependency of the established fusion - the 23rd most KMT2A-selective dependency of 17,728 genes in this project's own DepMap scan, with the strongest published shRNA effect in **SEM**, the index fusion's own B-lineage line, and one of only **three** loci that lose fusion occupancy under a menin inhibitor [E2, E3]. **And the one available molecule, JDI-12, binds both paralogues** [E4]. **The prediction that makes this a hypothesis rather than an observation:** a KDM3-family inhibitor given systemically has **opposite signs in two compartments at once** - anti-leukaemic in the rearranged blast via JMJD1C, and rearrangement-inducing in the normal progenitor via KDM3B - so its therapeutic index cannot be expressed in viability units and must be expressed as a ratio of leukaemic kill to cytogenetic events in normal cells. **No published study measures both endpoints in any system**, and the two papers that measure one each do not cite one another on the point [E1, E4].
The two ends, side by side
| upstream: KDM3B | downstream: JMJD1C / KDM3C | |
|---|---|---|
| what it does here | keeps H3K9me1/2 low across the KMT2A BCR; keeps CTCF bound at exon 11 | required for leukaemia-stem-cell self-renewal; interacts with HOXA9; JmjC and zinc finger both required |
| disease stage | genesis, in a normal cell, before any fusion | maintenance, in an established leukaemia |
| effect of losing it | KMT2A copy gain and rearrangement | apoptosis, differentiation, loss of LSC frequency |
| DepMap KMT2A selectivity (this project's scan, same run) | rank 9,382 of 17,728; difference +0.0018; pan-essentiality +0.0226 | rank 23 of 17,728; difference -0.4535; pan-essentiality -0.0121 |
| index-fusion evidence | none - all chromatin work in epithelium, primary arms are FISH only | SEM (KMT2A::AFF1), strongest effect in the human panel; and the JMJD1C locus is one of three losing fusion occupancy under VTP-50469 |
| therapeutic direction | restore it (KDM3B over-expression rescues) | inhibit it |
| what one compound does | inhibits it | inhibits it |
That last row is the whole hypothesis. JDI-12 binds both by SPR; its discovery paper reports it killing MLL-rearranged leukaemia cells and sparing cord blood cells; PMID 37788669 uses it at 1 microM for 72 h to produce KMT2A copy gains in cord-blood CD34+ HSPCs 14. Those are not contradictory observations - viability and karyotype are different endpoints - but they are contradictory conclusions, and the safety claim rests on the endpoint blind to the hazard.
Predictions, in the order they should be tested
- Paralogue assignment. In isogenic KDM3B-null and JMJD1C-null cells, JDI-12's two phenotypes separate: the cytogenetic effect requires KDM3B, the cytotoxic effect requires JMJD1C. If both survive in both, the compound is acting through something else entirely and every phenotype attributed to either gene is unassigned.
- The missing endpoint. JDI-12 applied to an established KMT2A-rearranged line (SEM, RS4;11, MOLM-13) produces KMT2A FISH copy gains and break-apart in the surviving cells at sub-lethal concentrations. Nobody has looked; if it does, the compound rearranges the locus in the very cells it is proposed to treat.
- The index fusion's own dependency. JMJD1C CRISPR knockout in SEM and RS4;11 reproduces the 2014 shRNA result. This is the only B-lineage, index-fusion datum in the downstream arm and it rests on one abstract.
- Locus specificity of the downstream arm. If JMJD1C is a fusion dependency because the fusion sits on its locus, then JMJD1C should be a spreading target, and its dependency should be menin-inhibitor-sensitive. This graph's own entry already asks why MEIS1, MEF2C and JMJD1C are menin-dependent when HOXA and MYB are not.
- The mark. JMJD1C loss in a KMT2A-rearranged cell changes H3K9me1/2 at fusion-bound loci. UniProt doubts JMJD1C is a histone demethylase at all, and the domain-mapping study names H3K36me instead - so this prediction may fail and its failure would be the most interesting result in the list.
- The repair route. JMJD1C inhibition impairs MDC1-RNF8-dependent double-strand-break repair, and combining it with topoisomerase II inhibition raises KMT2A rearrangement more than either alone. Both halves of the genesis mechanism would then be pushed the same way by one molecule.
What kills this hypothesis
- JDI-12's cytogenetic effect surviving in a KDM3B-null cell. The upstream arm's pharmacology would then not be about KDM3B, the 'both ends' framing collapses to a coincidence of family membership, and the compound is simply a dirty tool.
- No KMT2A rearrangement in KMT2A-rearranged lines under sub-lethal JDI-12. The clinical tension dissolves: the compound would be cytotoxic where you want it and genotoxic only in cells you are not dosing, which is an ordinary and manageable profile.
- The JMJD1C dependency failing to replicate in SEM by CRISPR. The downstream arm's only index-fusion, B-lineage datum is a 2014 shRNA result known here from an abstract. If it does not replicate, the downstream end of the bracket rests on murine myeloid MLL-AF9 alone - which is this project's most-documented generalisation failure.
- The two ends being in different cells that never co-exist. If the normal progenitors at risk of rearrangement are not present or not cycling in a patient receiving therapy, the two signs never apply simultaneously and the hypothesis is a laboratory curiosity rather than a clinical tension. Note the mechanism requires ~3 cell divisions for an alteration to become inherited.
- KDM3B's upstream role being non-catalytic. If a JmjC-dead KDM3B rescues locus stability, an active-site inhibitor of the family would not phenocopy KDM3B loss at all, and the shared-target premise fails at the first step.
| statement | The KDM3/JHDM2 family occupies both ends of KMT2A-rearranged leukaemia - KDM3B gates whether the KMT2A locus rearranges, JMJD1C/KDM3C is a dependency of the fusion once it exists - and the one available small molecule binds both paralogues, so a systemic KDM3-family inhibitor has opposite signs in the leukaemic blast and in the normal haematopoietic progenitor at the same time. Its therapeutic index therefore cannot be expressed in viability units and must be expressed as leukaemic kill per cytogenetic event in normal cells, a ratio nobody has measured in any system. |
|---|---|
| origin | data-pattern |
| prior probability | 0.3 |
| how it was found | This beat was dispatched on the upstream mechanism and reached the downstream end through the compound: the paper's 'KDM3i' is JDI-12, whose discovery paper names JMJD1C, and JMJD1C turned out to be already named in three entries of this graph as a menin-inhibitor-sensitive fusion target with a rank-23 DepMap selectivity signal sitting unquoted in the project's own analysis directory. |
| what would make it important | A drug that treats a disease it can also cause needs its index expressed in the currency of the harm. If prediction 2 holds, no KDM3-family inhibitor can be developed for KMT2A-rearranged leukaemia without a cytogenetic safety endpoint from the first in vitro experiment - which is not how epigenetic-inhibitor programmes are currently run. |
| absence class | never-measured |
predictions
[ "In isogenic KDM3B-null and JMJD1C-null cells, JDI-12's two phenotypes separate: the CYTOGENETIC effect requires KDM3B and the CYTOTOXIC effect requires JMJD1C.", "JDI-12 at sub-lethal concentrations produces KMT2A FISH copy gains and break-apart in surviving cells of established KMT2A-rearranged lines (SEM, RS4;11, MOLM-13) - the endpoint nobody has measured in the cells the compound is proposed to treat.", "JMJD1C CRISPR knockout in SEM and RS4;11 reproduces the 2014 shRNA dependency, which is currently the only B-lineage index-fusion datum in the downstream arm and is known here only from an abstract.", "If JMJD1C is a fusion dependency because the fusion occupies its locus, JMJD1C should be a spreading target and its dependency should be menin-inhibitor-sensitive.", "JMJD1C loss in a KMT2A-rearranged cell changes H3K9me1/2 at fusion-bound loci - a prediction that may well FAIL, since UniProt doubts JMJD1C is a histone demethylase and the domain-mapping study names H3K36me instead.", "JMJD1C inhibition impairs MDC1-RNF8-dependent double-strand-break repair, so combining it with a topoisomerase II inhibitor raises KMT2A rearrangement more than either alone - both halves of the genesis mechanism pushed the same way by one molecule." ]
experiments
[ "exp-dijos-aml-kmt2a-status" ]
analyses
[ "ana-gse210480-h3k9me3-at-the-kmt2a-bcr", "ana-kmt2a-bcr-ctcf-site-coordinate-overlap" ]
falsifiers
[ "JDI-12's cytogenetic effect survives in a KDM3B-null cell - the upstream pharmacology is not about KDM3B and the 'both ends' framing collapses to coincidence of family membership", "sub-lethal JDI-12 produces NO KMT2A rearrangement in KMT2A-rearranged lines - the clinical tension dissolves into an ordinary, manageable profile", "the JMJD1C dependency fails to replicate in SEM by CRISPR - the downstream end rests on murine myeloid MLL-AF9 alone, which is this project's most-documented generalisation failure", "a JmjC-catalytically-dead KDM3B rescues locus stability - an active-site family inhibitor cannot phenocopy KDM3B loss, and the shared-target premise fails at the first step", "the two ends never co-exist in a treated patient: if the progenitors at risk are absent or non-cycling during therapy, and the mechanism needs about three divisions for an alteration to become inherited, the opposite signs never apply simultaneously" ]
what is NOT claimed
[ "that JDI-12 or any KDM3 inhibitor has been given to a human - none has", "that KDM3B and JMJD1C act through the same mark - UniProt doubts JMJD1C is a histone demethylase at all", "that the upstream arm has been shown in a B-lineage or infant cell - it has not; all chromatin work is epithelial", "that the two ends are mechanistically linked - they share a protein family and a compound, which is a pharmacological coincidence until prediction 1 is run" ]
- This hypothesis is a reason to be careful about a compound class, not a reason to dismiss it. Menin inhibitors were developed against exactly this kind of objection and they work; the right response to a two-sided pharmacology is to measure both sides, not to abandon the target.
- The proposal that follows - cytogenetic safety endpoints from the first in vitro experiment - would slow down every KDM3-family programme, including ones aimed at diseases where the upstream arm is irrelevant. The requirement should be scoped to agents intended for, or likely to reach, haematopoietic progenitors.
- The compound at the centre of this hypothesis has three PubMed records and no ChEMBL entry. Building a clinical caution on a reagent this thinly characterised risks the opposite error - attributing to a drug class a liability that belongs to one uncharacterised natural-product hit.
- If prediction 5 fails and JMJD1C is not a histone demethylase, the 'one family, one mark, two ends' story is largely rhetorical, and what remains is a narrower and less interesting claim: one promiscuous compound hits two paralogues with different jobs. The entry should be rewritten to that if the mark prediction fails.
- Which paralogue carries which phenotype? One isogenic pair and two assays - FISH and viability - would settle the central ambiguity, and until it is run every statement about 'KDM3B inhibition' in a leukaemia cell is unassigned between two proteins.
- Does sub-lethal KDM3-family inhibition rearrange KMT2A in the cells it is meant to treat? This is the prediction with clinical consequences and the cheapest of the six to run.
- Is there any KDM3B-selective or JMJD1C-selective chemical matter? KDM3B has 34 PDB fragment-screen structures from a crystallographic campaign and a ChEMBL target entry; a selective compound for either paralogue would dissolve the whole ambiguity and none exists.
- Does the upstream mechanism need catalysis? No catalytically dead KDM3B has been tested, and if the answer is no, an active-site inhibitor is the wrong instrument for the upstream arm entirely.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
KDM3B keeps H3K9me1 low across the KMT2A breakpoint cluster region and CTCF bound at the conserved exon-11 CTCF site; KDM3B siRNA depletion or chemical inhibition with JDI-12 raises H3K9me1 across the BCR, lowers CTCF occupancy there, and produces site-specific KMT2A copy gains and break-apart events - transient at 12 h and inherited after 72 h, approximately three cell divisions - in RPE cells, KG1a, a primary AML line, a primary AML organoid and primary CD34+ HSPCs enriched from human umbilical cord blood; doxorubicin and etoposide both lower KDM3B and CTCF protein; and G9a depletion or dual G9a/GLP inhibition prevents the events while G9a over-expression alone causes them.
Epigenetic balance ensures mechanistic control of MLL amplification and rearrangement. · Gray ZH; Chakraborty D; Duttweiler RR; Alekbaeva GD; Murphy SE; Chetal K; Ji F; Ferman BI; Honer MA; Wang Z; et al. · Cell 2023What this evidence cannot show
**Full text and STAR Methods read 2026-09-22 via efetch db=pmc on PMC10591855 (Europe PMC fullTextXML HTTP 500; <body> confirmed, 80,638 stripped characters).** All chromatin measurements are in RPE, an epithelial line; the haematopoietic arms are DNA FISH only. 'Break apart' is FISH probe separation - no junction was sequenced. The paper attributes every compound effect to KDM3B and does not report which paralogue is engaged in any cell. KG1a and HL60 carry KDM3B loss of heterozygosity. No spike-in normalisation. The KDM3B occupancy data are re-analysed public ChIP-seq from colorectal cancer stem cells (GSE71885).
Evidence that argues against this (1)
- The compound's own discovery paper reports that it does NOT kill cord blood cells and proposes it as a treatment for MLL-rearranged acute leukaemia - and PMID 37788669's discussion cites that paper approvingly for the proposition that KMT2A-rearranged cells are more susceptible to KDM3B inhibition. The paper using the compound as a genotoxin also uses it as evidence of a therapeutic opportunity, in the same discussion. Xu X; Wang L; Hu L; Dirks WG; Zhao Y; Wei Z; Chen D; Li Z; Wang Z; Han Y; et al. 2020
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
In this project's own prior-free genome-wide DepMap scan over 17,728 protein-coding genes, 11 Cellosaurus-annotated KMT2A-rearranged leukaemia lines and 43 other leukaemia lines, JMJD1C ranks 23rd (percentile 0.13, difference -0.4535, KMT2A-rearranged mean -0.4697, other leukaemia -0.0162, pan-essentiality across 1,208 lines -0.0121, empirical p versus scan 0.0016) while KDM3B ranks 9,382nd (percentile 52.9, difference +0.0018, pan-essentiality +0.0226).
analyses/depmap-kmt2a-selective-scan-full/results.json - this project's own genome-wide scan, generated 2026-09-19T10:12:06Z, seed 20260918, n_scanned 17,728, n_kmt2a_rearranged 11, n_other_leukemia 43 · Iqbal Lab oncofusion knowledge graph, internal analysis over Broad DepMap 2026What this evidence cannot show
Read directly from the stored results file 2026-09-22 rather than quoted from a memo. The scan's caveats apply: 11 KMT2A-rearranged lines, **none infant-derived**; Cellosaurus fusion annotation may be incomplete, biasing toward the null; the empirical p is the genome-wide background fraction, not a separate random draw; **the DepMap release is not stated by the API** (rule 83). Rule 84 applies to the KDM3B figure and is why it is reported only as a contrast - a rank at the median is what half of all genes have, and this graph's base rate for the anti-selective half is 45.3%. A cell-line dependency difference is not a mechanism, and a pooled viability screen is structurally blind to the cytogenetic phenotype in the upstream arm.
Evidence that argues against this (1)
- The downstream dependency has a published resistance genotype before any drug exists: MLL-rearranged AML bearing activating RAS mutations is resistant to JMJD1C loss, via IL-3 receptor upregulation with RAS/MAPK and JAK-STAT activation. RAS co-mutation is common in KMT2A-rearranged disease and this graph already holds RAS co-mutation entries. Izaguirre-Carbonell J; Christiansen L; Burns R; Schmitz J; Li C; Mokry RL; et al. 2019
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
An epigenetic shRNA library screen in murine MLL-AF9 AML, counter-screened against normal mouse bone marrow, identified Jmjd1c as one of the best candidate drug targets; depletion impaired growth, colony formation and leukaemia establishment after transplantation, and in human leukaemic lines the STRONGEST effect was observed in the MLL-rearranged ALL cell line SEM, primarily through increased apoptosis; separately, a conditional mouse showed JMJD1C loss substantially decreased leukaemia-stem-cell frequency and caused differentiation of MLL-AF9- and HOXA9-driven leukaemias with JMJD1C directly interacting with HOXA9, while causing only minor defects in blood homeostasis.
shRNA screening identifies JMJD1C as being required for leukemia maintenance. · Sroczynska P; Cruickshank VA; Bukowski JP; Miyagi S; Bagger FO; Walfridsson J; et al. · Blood 2014What this evidence cannot show
**Abstract only for both papers; neither full text retrieved.** The SEM result is the only B-lineage, index-fusion datum in the downstream arm of this hypothesis and its effect size, hairpin count, knockdown efficiency and rescue controls are uncharacterised - it should be read in full before being leaned on, and prediction 3 exists to replicate it. 2014 shRNA carries seed-based off-target risk. 'Strongest effect' is a within-panel comparison, not a systematic lineage comparison. The conditional-mouse work (PMID 26878175) is murine, myeloid, MLL-AF9 and HOXA9, with no KMT2A::AFF1 and no B-lineage arm.
Evidence that argues against this (1)
- Jmjd1c knockout mice show no significant change in peripheral blood composition and initiate JAK2V617F-driven myeloproliferative disease normally, which the authors attribute to redundancy among JmjC-domain family members - so the same gene is essential in one haematopoietic context and redundant in another, and the therapeutic-window claim is endpoint-dependent. Staehle HF; Heinemann J; Gruender A; Omlor AM; Pahl HL; Jutzi JS 2020
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
JDI-4 and JDI-12, identified by virtual screening against the catalytic Jumonji domain of JMJD1C, bind JMJD1C AND its family homolog KDM3B with modest affinity by surface plasmon resonance; JDI-4 reverses the H3K9 demethylation conferred by KDM3B in vitro; both induce a global increase of H3K9 methylation in cells; and both kill MLL-rearranged acute leukaemia and other malignant haematopoietic cells but NOT leukaemia cells resistant to JMJD1C depletion and NOT cord blood cells.
Small molecular modulators of JMJD1C preferentially inhibit growth of leukemia cells. · Xu X; Wang L; Hu L; Dirks WG; Zhao Y; Wei Z; Chen D; Li Z; Wang Z; Han Y; et al. · Int J Cancer 2020What this evidence cannot show
**Abstract only; full text not retrieved.** No Kd ('modest' is the authors' word), no IC50, no selectivity panel against KDM3A, the KDM4 family or other 2-oxoglutarate oxygenases, and no statement of which H3K9 methylation state the in vitro assay read out. Virtual-screening hits from natural-product libraries, not characterised chemical probes; no ChEMBL id was located for JDI-12 by this beat and no structure appears in the abstracts read. 'Cord blood cells' is unspecified as to enrichment, assay, concentration and duration - which matters, because the contradicting cytogenetic result used 1 microM for 72 h on immunomagnetically enriched CD34+ cells.
Evidence that argues against this (1)
- UniProt states outright that it is uncertain whether JMJD1C removes methyl groups from histone proteins as the other KDM3 proteins do, annotating instead non-histone demethylation of MDC1 for double-strand-break repair and of STAT3. So a compound screened against JMJD1C's JmjC domain may not be an H3K9-pathway agent at all, and the shared-mark premise linking the two ends of this hypothesis could be wrong even if the shared-target premise holds. UniProt Consortium 2026
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The JMJD1C locus is one of only three - with MEIS1 and MEF2C - at which fusion (MLL1n) occupancy falls more than 2-fold under the menin inhibitor VTP-50469 and under CRISPR deletion of MEN1, while HOXA genes, MYB and GAPDH retain occupancy; MEN1 deletion followed by CUT&RUN identified 39 genes losing more than 2-fold occupancy, 20 shared with the drug, with the same locus pattern.
A Menin-MLL Inhibitor Induces Specific Chromatin Changes and Eradicates Disease in Models of MLL-Rearranged Leukemia. · Krivtsov AV; Evans K; Gadrey JY; Eschle BK; Hatton C; Uckelmann HJ; et al. · Cancer Cell 2019What this evidence cannot show
**Imported from this graph's existing `mechanism-the-menin-inhibitor-loss-set-is-small-and-named`, where it is recorded as read; not re-read for this entry.** The systems there are MOLM13 and RS4;11. This item establishes that the JMJD1C LOCUS is a menin-dependent fusion target - a statement about DNA occupancy - and says nothing about whether the JMJD1C PROTEIN is required, which is E2 and E3's separate claim. Conflating the two would be the error to avoid, and the coincidence that both are true of the same gene is part of what makes this hypothesis worth testing rather than proof of it.
Evidence that argues against this (1)
- If JMJD1C both requires HOXA9 and is a menin-dependent fusion target while HOXA retains fusion occupancy under menin inhibition, the regulatory relationship is circular and its direction is not established by either dataset. This graph's own entry leaves 'what makes MEIS1, MEF2C and JMJD1C menin-dependent when HOXA and MYB are not' as an open question. Zhu N; Chen M; Eng R; DeJong J; Sinha AU; Rahnamay NF; et al. 2016
the only human test either end of this hypothesis has
the upstream end: KDM3B gates whether the locus rearranges
the downstream end: JMJD1C is what the fusion needs once it exists - DepMap rank 23 of 17,728, strongest shRNA effect in SEM
the molecule that binds both paralogues and so has opposite signs in the two compartments
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
brackets the index fusion at both ends: the locus that becomes it, and the dependency it acquires - with SEM the one index-fusion datum
the downstream end is already in that entry: JMJD1C is one of three loci losing fusion occupancy under a menin inhibitor
the upstream arm of the hypothesis is this mechanism