Oncofusion

start / graph / gene-kdm3b

Genemoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q102-h3k9-rearrangement-brake

KDM3B - the H3K9me1/2 demethylase that sits on the KMT2A breakpoint cluster region and keeps it from rearranging, and the only leukaemia gene in this graph whose direction of effect flips between the myeloid and the lymphoid literature

also called KDM3B · JMJD1B · JHDM2B · 5qNCA · C5orf7 · KIAA1082 · lysine-specific demethylase 3B · lysine demethylase 3B

KDM3B is an enzyme that strips silencing tags off DNA-packaging proteins, and it parks itself right on the fragile stretch of the KMT2A gene where this leukaemia's break happens. Take it away and that stretch fills with tags, the protein that pins it shut falls off, and the gene starts copying and breaking itself.

The same thing, for a scientist

KDM3B (Q7LBC6, 1,761 aa, HGNC:1337, ENSG00000120733) is a JmjC-domain, 2-oxoglutarate-dependent H3K9 demethylase of the JHDM2/KDM3 family, annotated with the reaction H3K9me2 -> H3K9me0 (EC 1.14.11.65) and with the note 'may have tumor suppressor activity' [E1]. It sits in the frequently deleted 5q31 region - it was originally named **5qNCA** for that reason - and *MLL/KMT2A* copy gains often co-occur with 5q loss of heterozygosity [E2]. **Its mechanistic role here is new, specific, and measured in the wrong cell type.** Published KDM3B ChIP-seq shows KDM3B binding across the *KMT2A* gene with a strong peak inside the 8.3 kb breakpoint cluster region that is lost on KDM3B knockdown; in near-diploid human RPE cells KDM3B depletion raises H3K9me1 uniformly across that BCR, lowers CTCF occupancy at the conserved exon-11 CTCF site, and produces site-specific *KMT2A* copy gains and break-apart events that become **heritable** after 72 h of suppression - and G9a/EHMT2 co-depletion reverses all of it [E2]. Chemical inhibition reproduces the effect in KG1a, a primary AML-derived line, a primary AML organoid and **cord-blood CD34+ HSPCs** [E2]. **But the gene's direction in blood is genuinely contested, not merely uncertain:** it is underexpressed in AML including MLL-AF6 and MLL-AF9 lines and behaves as a tumour suppressor there [E3], and it is *up*-regulated in ALL-type patient blood, activates the oncogene *LMO2* and blocks differentiation [E4]. **And it is a dependency nowhere:** DepMap Chronos mean **+0.0232** across 1,258 screened lines, 0.4% of lines below -0.5, lymphoid mean +0.0060 [E5], with a KMT2A-selectivity rank of **9,382 of 17,728** in this project's own scan [E6] - while its paralogue JMJD1C ranks **23rd** [E6].

What it looks like
KDM3B1,761 aa
JmjC (catalytic demethylase domain)JmjC (catalytic demethylase domain) · residues 1498–1721Zinc finger, C6-type · residues 1031–1056LXXLL motif (nuclear-receptor box) · residues 1293–129705001,0001,500
loading AF-Q7LBC6-F1…
AlphaFold AF-Q7LBC6-F1 ↗
colourshape

This is a prediction, not a measurement. No one has seen this protein in this shape. The colours say how much the software trusts itself, and that is worth reading before anything else — the orange stretches are places the model is effectively saying this part has no fixed shape.

very high (90+)confident (70–90)low (50–70)very low (<50)
I have never read one of these pictures before

A protein is one long chain of amino acids that folds up. The ribbon traces that chain from one end to the other; it is not a wire, it is a simplification of where the backbone runs. Drag to rotate, scroll to zoom.

Domains are the stretches of chain that do a particular job, and they are what the coloured blocks above the picture mark out. Click one and the view flies to it. In this project the important question about any domain is usually whether the fusion still has it — the labels say lost where it does not.

Anything in yellow sticks is a bound small molecule, which is usually the drug.

Go deeper
What KDM3B does at the KMT2A locus, in the words of the one experiment

The chain, each link from the full text of PMID 37788669 2:

  1. Occupancy. Published KDM3B ChIP-seq (re-analysed from GSE71885) shows KDM3B across the KMT2A gene with a strong peak inside the BCR, lost on shRNA KDM3B depletion. KDM3B also binds across TCF3 and AFF3.
  2. The mark. In control RPE cells H3K9me1 at the BCR is lower than in adjacent regions. KDM3B siRNA raises H3K9me1 strongly and uniformly across the whole BCR; H3K9me2 rises on the flank. Genome-wide, the H3K9me1 rise across the middle of KMT2A is one of the strongest events in the genome.
  3. The pin. A highly conserved CTCF site sits in KMT2A exon 11, inside the BCR, directly overlapping the KDM3B peak across multiple ENCODE cell lines and tissues. KDM3B depletion does not change total CTCF protein but significantly reduces CTCF occupancy at that site. Genome-wide, 6,386 KDM3B peaks (41.5% of strong peaks) overlap the 46,340 RPE CTCF peaks; 17,077 CTCF peaks lose intensity on KDM3B depletion, and 1,005 of the 6,386 co-occupied sites are among them (P < 1.0e-216, Z = 143.38). The H3K9me1 rise at the KMT2A BCR CTCF site is among the strongest of all CTCF sites genome-wide.
  4. The consequence. KDM3B knockdown or KDM3 inhibition produces KMT2A copy gains and break-apart events by two-colour DNA FISH, with no change at an adjacent control probe or at chromosome-11 centromere. Depleting CTCF alone produces the same alterations and KDM3B co-depletion does not add to them - CTCF is downstream.
  5. Transience, then heredity. 12 h of KDM3 inhibition gives alterations that vanish on washout (extrachromosomal, transient); 72 h - about three divisions - gives alterations that persist into later passages and are visible on mitotic chromosomes. G9a depletion rescues the transient form and does not rescue the inherited form.
  6. Reversal. Pre-depleting G9a/EHMT2 - but not EHMT1 - prevents the alterations; dual G9a/GLP inhibition (UNC0642) rescues them completely with no effect on growth; G9a over-expression alone is sufficient to cause them.

What the paper does not do, in its own words: it does not show how oncogenic KMT2A fusions are generated leading to leukaemia, does not identify the second hit, and does not map a single integration site or sequenced junction. 'Break apart' throughout is FISH probe separation.

The direction problem, which is the honest headline

KDM3B's own literature contains both signs, and the papers say so:

systemdirectionsource
AML patients; MLL-AF6, MLL-AF9, PML-RARA linesunderexpressed; overexpression represses colony formation; acts through HOXA1 RAREs3
APLsuppresses progression, facilitates ATRA-mediated PML-RARA degradation7
5q31.2 MDS minimally deleted regiondeleted - the 5qNCA candidate tumour suppressor12
ALL-type patient blood cellsup-regulated; activates LMO2; represses differentiation4
IDH2- and TET2-mutant primary HSPCsrequired - Kdm3b loss reduces mutant-clone fitness; nominated as a therapeutic target in clonal haematopoiesis8
MLL-rearranged acute leukaemia cellskilled by KDM3-family small-molecule modulators9
every DepMap cell linenot a dependency at all (mean +0.0232)56

The 2018 paper states the resolution the field has: 'The function of KDM3B is likely cell-type dependent and its seeming functional discordance may reflect its phenotypic dependence on downstream targets' 3. This graph should not pick a side, and this entry does not.

But one row is decisive for how the locus mechanism should be read. This project's own genome-wide DepMap scan puts KDM3B at rank 9,382 of 17,728 for KMT2A selectivity, with a difference of +0.0018 between the 11 KMT2A-rearranged and 43 other leukaemia lines - noise at the median - and a pan-essentiality of +0.0226, i.e. losing it is if anything mildly favourable to a cell line. A gene whose loss no cell line minds is exactly the right shape for a permissive first hit and exactly the wrong shape for the therapeutic target the JDI compound paper proposes 69. Note the null (rule 84): about half of any gene set falls above the median by construction and this graph's base rate is 45.3%, so being near rank 9,382 is not the finding. The finding is the contrast with JMJD1C at rank 23 in the same scan, same denominator, same run 6.

The human natural experiment, and nobody has looked

Gray et al.'s model makes a prediction about people: an individual born with one working copy of KDM3B should be predisposed to KMT2A rearrangement. That individual exists. Germline KDM3B haploinsufficiency causes Diets-Jongmans syndrome (DIJOS, MIM 618846) - intellectual disability or developmental delay, short stature, feeding difficulties in infancy, joint hypermobility, and a characteristic facial gestalt - reported in 14 unrelated individuals plus three affected parents, from missense and truncating variants 10.

And two of them developed childhood cancer: acute myeloid leukaemia and Hodgkin lymphoma 10.

That is a one-line sentence in a 2019 human-genetics abstract, published four years before the mechanism, and it is the closest thing to a human test this mechanism has. What is missing is the karyotype. Nobody has reported whether the DIJOS AML carried a KMT2A rearrangement or an 11q23 amplification, and PMID 37788669 does not cite this paper at all - KDM3B[tiab] AND CTCF[tiab] returns exactly one record, and "Diets-Jongmans"[tiab] returns 8, with no overlap. One cytogenetics report would either supply the mechanism's first human evidence or weaken it substantially. See exp-dijos-aml-kmt2a-status.

Read it cautiously, and here is why. Two cancers in 17 individuals from a cohort assembled by gene matching on a developmental phenotype is not an incidence estimate; ascertainment in such cohorts is biased towards the severe and the investigated, the denominator is not a population, and childhood AML has other causes. It is a lead worth one phone call, not an epidemiological result.

The compound is not what its name says, and the graph already holds the reason

PMID 37788669 phenocopies KDM3B loss with a compound it calls KDM3i and cites as 'a KDM3 family inhibitor'. The compound is JDI-12, and its discovery paper is titled Small molecular modulators of JMJD1C preferentially inhibit growth of leukemia cells 9. JMJD1C is KDM3C, the third member of the family.

The founding paper's own account: JDI-4 and JDI-12 bind JMJD1C and its family homolog KDM3B with modest affinity by SPR; JDI-4 reverses the H3K9 demethylation conferred by KDM3B in vitro; both induce a global increase in H3K9 methylation in cells; and both kill MLL-rearranged acute leukaemia cells but not leukaemia cells resistant to JMJD1C depletion and not cord blood cells 9. So the family claim is supported and the selectivity is JMJD1C's.

Why that matters more here than it would elsewhere. This graph already holds mechanism-the-menin-inhibitor-loss-set-is-small-and-named: under a menin inhibitor, fusion occupancy is lost at MEIS1, MEF2C and JMJD1C and nowhere else. And this project's own DepMap scan ranks JMJD1C 23rd of 17,728 for KMT2A selectivity. JMJD1C is a fusion-maintenance dependency in this graph; KDM3B is a locus-genesis gate. They are paralogues, and one compound hits both. The sentence in Gray et al.'s discussion - 'Consistent with a prior study, these data suggest that KMT2A-rearranged cells have increased susceptibility to KDM3B inhibition' - attributes to KDM3B a selectivity that its cited source measured, and DepMap independently assigns, to JMJD1C. See gene-kdm3c and drug-jdi-12.

Facts and figures
uniprotQ7LBC6
gene symbolKDM3B
length AA1761
hgncHGNC:1337
ensemblENSG00000120733
entrez gene ID51780
omim gene609373
omim disease618846
alphafold IDAF-Q7LBC6-F1
cytoband5q31.1
accession resolutiongene_exact:KDM3B AND organism_id:9606 AND reviewed:true returned exactly 1 row on 2026-09-22; primary gene name read back as KDM3B and length read back as 1,761 from that row. Not taken from recall. Synonyms on the record: C5orf7, JHDM2B, JMJD1B, KIAA1082.
ec number1.14.11.65
catalytic reaction as annotatedN(6),N(6)-dimethyl-L-lysyl(9)-[histone H3] + 2 2-oxoglutarate + 2 O2 = L-lysyl(9)-[histone H3] + 2 formaldehyde + 2 succinate + 2 CO2 (Rhea RHEA:60188) - note the substrate is the DIMETHYL state; the record does not annotate monomethyl demethylation
familyJHDM2/KDM3 - KDM3A (Q9Y4C1, 1,321 aa), KDM3B (Q7LBC6, 1,761 aa), JMJD1C/KDM3C (Q15652, 2,540 aa)
domains omitted and whyInterPro predicts a PWWP domain (IPR054504, Pfam PF22988) and a Tudor domain (IPR054503, Pfam PF22987) specific to KDM3A/KDM3B, and the UniProt feature table gives NO start/end for either. They are deliberately left out rather than given invented coordinates, because a domain lane with a guessed boundary is worse than an absent one and a null boundary is invisible to every figure.
PDB noteUniProt cross-references 36 PDB entries for Q7LBC6, of which 34 are 5R-series depositions from a crystallographic FRAGMENT SCREENING campaign against the catalytic domain. Only 4C8D and 6RBJ are listed here to keep the array meaningful; the fragment series is recorded in `chemical_matter` because a fragment screen is a tractability statement, not 34 separate structures worth drawing.
chemical matterChEMBL target CHEMBL3784906. Thirty-four PDB 5R-series structures from a crystallographic fragment screen against the catalytic domain, i.e. an open, published fragment-hit starting point. The only compound used against KDM3B in the KMT2A context is JDI-12, a virtual-screening natural-product hit published as a JMJD1C modulator that also binds KDM3B with modest affinity - see drug-jdi-12. No selective KDM3B chemical probe exists.
localizationNUCLEUS, AND THE ANNOTATION IS NOT A MEASUREMENT. UniProt Q7LBC6 carries exactly one subcellular location, 'Nucleus', on **ECO:0000250 - inferred by sequence similarity, with no experimental citation attached at all**. That is weaker than the ECO:0000305 curator-inference cases this graph has flagged, and it means the compartment of a chromatin enzyme has never been measured in its own right in any published record this beat could reach. What IS measured is its DNA address, not its compartment: KDM3B ChIP-seq shows binding across the KMT2A gene with a peak inside the breakpoint cluster region, and 6,386 strong KDM3B peaks (41.5%) overlap CTCF peaks genome-wide [E2]. Following the distinction gene-ctcf draws, a ChIP peak is a sequence address and is NOT counted here as localisation.
subnuclear localizationNEVER MEASURED, in any cell type, and the database cannot help. There is no nucleolar, speckle, lamina, PML-body or chromocentre placement for KDM3B in UniProt (the single location value is 'Nucleus' on sequence similarity alone) and none was found in the literature reached by this beat. The nearest thing to a subnuclear statement is inferential and worth recording as a lead rather than a fact: KDM3B's strongest KMT2A-locus peak coincides with a CTCF site, and CTCF's own measured subnuclear placements in this graph are nucleolar tethering (interphase) and pericentromere (mitosis), with nuclear speckles added by PMID 33411704 - so if KDM3B travels with CTCF, those are the compartments to image first. No imaging of KDM3B in any leukaemic cell exists. Absence class: never-measured.
localization shown for fusionNO for the protein's own compartment, in any KMT2A fusion, in any lineage - KDM3B has never been imaged in a KMT2A-rearranged cell. What HAS been shown in a fusion-relevant context is a DNA address and a phenotype, not a compartment: KDM3B occupies the KMT2A breakpoint cluster region (ChIP-seq re-analysed from a colorectal cancer stem-cell dataset, GSE71885) and its depletion or chemical inhibition produces KMT2A copy gains and break-apart events in RPE cells, KG1a, a primary AML line, a primary AML organoid and cord-blood CD34+ HSPCs [E2]. Filing either of those as localisation would be the conflation gene-ctcf's entry exists to prevent.
epigenetic readoutH3K9me1 and H3K9me2 ChIP-seq and ChIP-qPCR across the KMT2A breakpoint cluster region and at the exon-11 CTCF site, plus CTCF ChIP-seq/ChIP-qPCR at the same site, in control versus siKDM3B versus siG9A versus double-knockdown RPE cells - deposited as GSE210480 (60 samples). KDM3B's own occupancy is read from published KDM3B ChIP-seq (GSE71885, colorectal cancer stem cells). Antibodies, per rule 74: H3K9me1 Abcam ab8896-100, H3K9me2 Abcam ab1220, H3K9me3 Abcam ab8898, CTCF clone D31H2 CST #3418, anti-KDM3B Cell Signaling, anti-G9a Sigma.
eraser of that markH3K9me1 and H3K9me2 (annotated reaction begins at the dimethyl state); weak or no activity on H3K9me3 by family analogy with KDM3A
analogous biologyThe same KDM3B/G9a H3K9me1/2 balance controls copy gain and rearrangement of TCF3/E2A and AFF3/LAF4 - both B-ALL rearrangement loci, and AFF3/LAF4 is itself a KMT2A fusion partner - while MLLT1/ENL and MLLT3/AF9 did NOT change. TCF3 copy gains were inherited; AFF3 copy gains were not [E2]. So the locus specificity is real but is not KMT2A-exclusive, and the three responsive loci are all lymphoid-rearrangement genes.
absence classnever-measured
domains
Zinc finger, C6-type1,031Zinc finger, C6-type: 1031LXXLL motif (nuclear-rece…1,293LXXLL motif (nuclear-receptor box): 1293JmjC (catalytic demethyla…1,498JmjC (catalytic demethylase domain): 1498
domains — the numbers behind the chart
[
  {
    "name": "JmjC (catalytic demethylase domain)",
    "start": 1498,
    "end": 1721,
    "pfam": "PF02373"
  },
  {
    "name": "Zinc finger, C6-type",
    "start": 1031,
    "end": 1056,
    "pfam": null
  },
  {
    "name": "LXXLL motif (nuclear-receptor box)",
    "start": 1293,
    "end": 1297,
    "pfam": null
  }
]
PDB IDS
[
  "4C8D",
  "6RBJ"
]
cell type
{
  "normal_tissue_distribution": "Ubiquitous, highly expressed in placenta, skeletal muscle, kidney, heart and liver (UniProt TISSUE SPECIFICITY, ECO:0000269 PubMed:11087669 and PubMed:11687974). **Neither blood nor bone marrow is named in the high-expression list**, which is worth stating in a beat about a haematopoietic mechanism - and the underlying sources are two expression surveys from 2000 and 2001 whose method and cohort the record does not state, so this is a database assertion rather than a characterised measurement.",
  "cell_types_in_which_the_kmt2a_phenotype_was_produced": "immortalised near-diploid human RPE (all chromatin work); KG1a (AML, KDM3B LOH); HL60 (APL, KDM3B LOH); U937 (myeloid); a primary patient-derived AML case; a primary patient-derived AML ORGANOID on irradiated OP9 stroma; **cord-blood CD34+ HSPCs** isolated by immunomagnetic enrichment from fresh human umbilical cord blood; and mouse spleen in vivo [E2].",
  "the_cell_type_that_matters_here": "cord-blood CD34+ HSPCs are the closest available proxy for the fetal/neonatal progenitor in which infant t(4;11) is thought to arise in utero, and they are the one primary human haematopoietic cell in which KDM3 inhibition was shown to raise KMT2A copy gains. They are also, in the compound's own discovery paper, the cells reported to be SPARED by the same compound [E9]. No B-lineage-committed cell was tested in either paper.",
  "not_measured": "No KDM3B measurement of any kind exists in an infant KMT2A::AFF1 sample, in a B-lineage progenitor, or in fetal liver."
}
cell state
{
  "proliferation_dependence": "The phenotype is division-dependent and that is its most important state property: 12 h of KDM3 inhibition gives transient extrachromosomal KMT2A gains that disappear on washout, while 72 h - approximately three cell divisions - gives gains and rearrangements that are inherited into later passages and visible on mitotic chromosomes. G9a depletion rescues the transient form and does NOT rescue the inherited form [E2].",
  "differentiation_state": "KDM3B is down-regulated during ATRA-induced differentiation of HL-60 cells through corepressor recruitment, and represses leukaemia-cell differentiation when present [E4] - so its level is itself a function of differentiation state, in the one lineage where that was measured.",
  "genotype_dependent_state": "Kdm3b loss reduces fitness specifically in Idh2- and Tet2-mutant primary HSPCs and not in wild-type ones, via loss of Mpl and other cytokine receptors [E8]. So 'is KDM3B needed' has no state-free answer.",
  "selection_state": "In RPE lines already carrying inherited KMT2A gains, a short re-exposure to KDM3 inhibition raised copy gains further at 3-6 h and returned them to baseline by 12 h with increased Annexin V - the authors read this as negative selection of the aberrant cells alongside emergence of a new aberrant population [E2].",
  "bounded": "No KDM3B cell-state measurement exists in any lymphoid cell or in any KMT2A-rearranged blast."
}
mark erased
[
  "mark-h3k9me1",
  "mark-h3k9me2"
]
depmap
{
  "chronos_mean_all_lines": 0.0232,
  "n_screened_lines": 1258,
  "pct_lines_below_minus_0_5": 0.4,
  "lymphoid_mean": 0.006,
  "n_lymphoid_lines": 97,
  "fetched": "2026-09-22 via Open Targets GraphQL depMapEssentiality, approvedSymbol read back as KDM3B",
  "release": "NOT STATED BY THE API - rule 83; only the fetch date is on record",
  "how_to_read": "A POSITIVE mean means knockout is neutral-to-mildly-favourable in the average line. This is the expected shape for a candidate tumour suppressor and is NOT support for the locus mechanism, because a pooled viability screen cannot detect a cytogenetic rearrangement in a surviving cell at all."
}
depmap kmt2a selectivity
{
  "gene": "KDM3B",
  "difference": 0.0018,
  "mean_kmt2a_rearranged": 0.0068,
  "mean_other_leukaemia": 0.0051,
  "pan_essentiality_all_lines": 0.0226,
  "selectivity_rank": 9382,
  "rank_denominator": 17728,
  "percentile": 52.922,
  "paralogue_for_contrast": {
    "gene": "JMJD1C",
    "difference": -0.4535,
    "mean_kmt2a_rearranged": -0.4697,
    "mean_other_leukaemia": -0.0162,
    "pan_essentiality_all_lines": -0.0121,
    "selectivity_rank": 23,
    "percentile": 0.13,
    "empirical_p_vs_scan": 0.001636
  },
  "provenance": "Read 2026-09-22 directly from analyses/depmap-kmt2a-selective-scan-full/results.json (generated 2026-09-19T10:12:06Z, seed 20260918, 11 KMT2A-rearranged and 43 other leukaemia lines, none infant-derived). NOT re-fetched and not independent of that scan; every caveat it carries applies.",
  "rule_84_note": "KDM3B's own rank of 9,382 of 17,728 is at the median and carries no information - about half of any gene set is above the median by construction and this graph's base rate for the anti-selective half is 45.3%. The informative quantity is the CONTRAST with JMJD1C at rank 23 in the same run."
}
germline syndrome
{
  "name": "Diets-Jongmans syndrome (DIJOS)",
  "omim": "618846",
  "mechanism": "KDM3B haploinsufficiency (missense and truncating variants, de novo and inherited)",
  "cohort": "14 unrelated individuals plus 3 affected parents (PMID 30929739)",
  "phenotype": "intellectual disability or developmental delay, short stature, infant feeding difficulties, joint hypermobility, facial gestalt (wide mouth, pointed chin, long ears, low columella)",
  "childhood_cancers_reported": "2 of 17 - one acute myeloid leukaemia, one Hodgkin lymphoma",
  "what_is_missing": "no karyotype, cytogenetics or KMT2A status is reported for the AML case, which is the single fact that would test the locus mechanism in a human",
  "not_described_as": "a cancer-predisposition syndrome - it is classified as an intellectual-disability syndrome and the cancers are reported as a notable incidental observation"
}
identifier trap
{
  "token": "KDM3B -- the approved symbol is the better operand but misses a fifth of the literature",
  "class": "a_gene_a_fifth_of_whose_literature_is_written_under_a_retired_alias",
  "registry_row_existed_before": false,
  "registry_grepped_first": true,
  "measured_2026_09_22": {
    "KDM3B[tiab]": 92,
    "JMJD1B[tiab]": 22,
    "5qNCA[tiab]": 5,
    "JHDM2B[tiab]": 0,
    "union": 114,
    "reachable_only_via_alias": 22,
    "alias_only_pct": 19.3,
    "jmjd1b_only": 18,
    "5qnca_only": 5,
    "overlap_of_those_two": 1,
    "KDM3B AND CTCF": 1,
    "KDM3B AND leukemia": 9,
    "\"Diets-Jongmans\"": 8
  },
  "id_sets_compared_not_counts": true,
  "note": "JHDM2B returns zero WITH errorlist phrasesnotfound [\"JHDM2B\"] - the honest failure mode, not a silent well-formed zero. 5qNCA is the historical name and all five of its records predate 2008.",
  "separate_trap_worth_the_registry": "The compound named for this gene in the KMT2A literature ('KDM3i') is JDI-12, published as a JMJD1C modulator; JMJD1C is KDM3C, the paralogue, and it ranks 23rd of 17,728 for KMT2A-selective dependency in this project's own scan while KDM3B ranks 9,382nd. A drug name that carries a gene family's name is not a target assignment."
}
How this could bite us back
  • Two 2023-2024 papers propose opposite interventions on this one gene in the same cell type. One nominates KDM3B inhibition as a genotype-specific therapy for clonal haematopoiesis in primary HSPCs; the other shows KDM3-family inhibition of primary cord-blood HSPCs producing KMT2A copy gains and break-apart events. Neither cites the other, and anyone developing a KDM3B inhibitor for pre-leukaemic clones needs the cytogenetics endpoint that the first paper did not measure.
  • The protective interventions with the most direct evidence here - KDM3B over-expression and CTCF over-expression - are not deliverable modalities in a patient. What is deliverable is a G9a/GLP inhibitor, which acts on the other side of the balance and carries its own liabilities (see drug-unc0642). An entry that reads as 'restore KDM3B' is describing a rescue experiment, not a therapy.
  • The whole locus mechanism rests on chromatin measurements in retinal pigment epithelium. The haematopoietic arms are cytogenetics only. Any claim that this gene gates KMT2A rearrangement in a blood cell is currently an inference across a cell-type boundary, and the graph should say so in the claim and not only the caveat.
  • KDM3B's direction of effect in leukaemia is contested in the published literature, with the tumour-suppressor reports in the myeloid lineage and the oncogenic report in the lymphoid one - which is this graph's lineage. A therapeutic strategy premised on either direction is premised on the lineage in which it was measured.
  • The inherited form of the phenotype is not reversible by removing the upstream cause: G9a depletion rescued transient KMT2A gains and did not rescue gains already inherited. So a prophylactic G9a inhibitor has a window that closes, and a patient already carrying rearranged cells is outside it.
Still open
  1. Did the Diets-Jongmans AML carry a KMT2A rearrangement or an 11q23 amplification? One cytogenetics report would supply this mechanism's first human evidence or weaken it substantially, and the mechanism paper does not cite the syndrome paper at all.
  2. Is KDM3B expressed, and does it occupy the KMT2A breakpoint cluster region, in a fetal or cord-blood B-lineage progenitor? Every occupancy datum here comes from colorectal cancer stem cells and every chromatin datum from retinal pigment epithelium, and UniProt's high-expression tissue list names neither blood nor marrow.
  3. Is the locus phenotype catalysis-dependent? Nothing in the published work uses a catalytically dead KDM3B, so 'KDM3B protects the locus' and 'H3K9 demethylation protects the locus' are not yet separated - and KDM3B carries an LXXLL nuclear-receptor motif and a zinc finger that could act without the JmjC domain.
  4. Do KDM3B and CTCF interact physically, or only through the mark? The paper shows KDM3B occupancy overlapping CTCF occupancy at 6,386 sites and shows the mark as the intermediate, but no co-immunoprecipitation is reported, and a direct contact would make CTCF displacement a competition rather than a chromatin-state consequence.
  5. Which KDM3 family member does JDI-12 actually act through in a KMT2A-rearranged cell? The compound binds both, the published selectivity is JMJD1C's, and DepMap assigns the KMT2A-selective dependency to JMJD1C - so every phenotype attributed to 'KDM3B inhibition' in a leukaemia cell is currently unassigned between two paralogues.
Every source behind this entry (11)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    KDM3B is UniProt Q7LBC6 (KDM3B_HUMAN), 1,761 aa, primary gene name KDM3B with synonyms C5orf7, JHDM2B, JMJD1B and KIAA1082; a histone demethylase that specifically demethylates Lys-9 of histone H3 with an annotated reaction N(6),N(6)-dimethyl-L-lysyl(9)-[histone H3] -> L-lysyl(9)-[histone H3] in two 2-oxoglutarate/O2-dependent steps (EC 1.14.11.65), described as possibly having tumor suppressor activity; it carries a JmjC domain at 1498-1721, a C6-type zinc finger at 1031-1056 and an LXXLL nuclear-receptor motif at 1293-1297; subcellular location is annotated as Nucleus on ECO:0000250 (sequence similarity, NOT experimental); tissue specificity is 'Ubiquitous. Highly expressed in placenta, skeletal muscle, kidney, heart and liver'; and germline variants cause Diets-Jongmans syndrome (MIM 618846).

    What this evidence cannot show

    **Accession resolved live 2026-09-22: `gene_exact:KDM3B AND organism_id:9606 AND reviewed:true` returned exactly ONE row, and the primary gene name (KDM3B) and length (1,761) were read back from that row rather than assumed.** Three separate weaknesses in the record itself, all recorded rather than smoothed: (1) **the nuclear localisation is ECO:0000250 - inferred by sequence similarity, with no experimental citation at all.** That is weaker than the ECO:0000305 curator-inference case this graph has flagged before, and it means the compartment of an enzyme whose whole mechanism is chromatin has never been measured in its own right. (2) FUNCTION and CATALYTIC ACTIVITY both rest on one citation, PubMed:16603237, not retrieved here - rule 91. (3) **The annotated reaction starts from the DImethyl substrate**, so the record does not itself establish demethylation of H3K9me1, although 'H3K9me1/2 demethylase' is how the leukaemia literature and PMID 37788669's experiments use it. InterPro additionally predicts PWWP_KDM3B and Tudor_KDM3B domains (IPR054504, IPR054503) for which the UniProt feature table gives **no boundaries**, so they are deliberately omitted from `data.domains` rather than given invented coordinates.

    Evidence that argues against this (1)
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    KDM3B binds across the KMT2A gene with a strong peak inside the 8.3 kb breakpoint cluster region that is lost on KDM3B depletion; in near-diploid human RPE cells KDM3B siRNA depletion or KDM3-family chemical inhibition (JDI-12) raises H3K9me1 uniformly across the BCR and H3K9me2 on its flank, significantly reduces CTCF occupancy at the conserved KMT2A exon-11 CTCF site without changing total CTCF protein, and produces site-specific KMT2A copy gains and break-apart events with no change at an adjacent control probe; 12 h of inhibition gives transient alterations that vanish on washout while 72 h gives alterations inherited into later passages and visible on mitotic chromosomes; the same chemical inhibition raises KMT2A gains and rearrangements in KG1a, a primary AML-derived line, a primary AML organoid model and cord-blood CD34+ HSPCs; depleting CTCF alone is sufficient and is not enhanced by KDM3B co-depletion; and pre-depletion of G9a/EHMT2 but not EHMT1, or dual G9a/GLP inhibition with UNC0642, prevents the alterations.

    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; Myers C; Sun R; Kaniskan HU; Toma MM; Bondarenko EA; Santoro JN; Miranda C; Dillingham ME; Tang R; Gozani O; Jin J; Skorski T; Duy C; Lee H; Sadreyev RI; Whetstine JR · Cell 2023
    PMID 37788669DOI
    What this evidence cannot show

    **Full text read 2026-09-22 via `efetch db=pmc` on PMC10591855 (Europe PMC `fullTextXML` returned HTTP 500; a `<body>` element was confirmed present and the stripped body is 80,638 characters). This closes the gap `gene-ehmt2` E2 named, where the same paper was cited from its abstract alone.** **Every chromatin measurement is in RPE - immortalised retinal pigment epithelium, chosen because it is near-diploid, genomically stable and carries no cancer mutations.** It is epithelial, not haematopoietic. The haematopoietic and primary arms - KG1a, HL60, U937, a primary AML line, a primary AML organoid, cord-blood CD34+ HSPCs, and mouse spleen - are scored **only** by DNA FISH for copy gain and break-apart; **no H3K9 ChIP was performed in any blood cell.** HL60 and KG1a both carry KDM3B loss of heterozygosity, so their baseline is already perturbed. 'Break apart' is two-colour FISH probe separation - a cytogenetic proxy, not a sequenced junction; the paper states that integration sites and sequence maps remain to be determined. The KDM3B ChIP-seq is re-analysed **public data from a different cell line** (colorectal cancer stem cells, GSE71885), which the authors acknowledge. ChIP antibodies: H3K9me1 ab8896-100, H3K9me2 ab1220, H3K9me3 ab8898, CTCF clone D31H2 (CST #3418); no spike-in normalisation described. The compound called 'KDM3i' is JDI-12, published as a JMJD1C modulator - see E9.

    Evidence that argues against this (1)
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    KDM3B is underexpressed in AML patients and in AML cell lines carrying MLL-AF6, MLL-AF9 or PML-RARA translocations; overexpression of KDM3B repressed colony formation of an AML cell line with 5q deletion; KDM3B binds the retinoic-acid response elements but not the promoter of HOXA1, and KDM3B knockdown increased H3K9 mono-methylation while DECREASING H3K9 di-methylation at those elements - and the authors state in their own first sentence that KDM3B reportedly shows both tumour-suppressive and tumour-promoting activities in leukaemia and that its function is likely cell-type dependent.

    KDM3B shows tumor-suppressive activity and transcriptionally regulates HOXA1 through retinoic acid response elements in acute myeloid leukemia. · Xu X; Nagel S; Quentmeier H; Wang Z; Rosenwald A; Meyer C; Drexler HG; MacLeod RAF · Leuk Lymphoma 2018
    PMID 28540746DOI
    What this evidence cannot show

    **Abstract only; full text not retrieved**, so the ChIP antibodies (rule 74), element coordinates, replicate counts and effect sizes behind the opposite-sign me1/me2 result are uncharacterised - and that is the part of this item that contradicts E2. Myeloid lines only; no B-lineage arm, no KMT2A::AFF1, and the KMT2A-fusion relevance is expression level rather than a chromatin measurement in a fusion cell. Causality is tested by overexpression in a 5q-deleted line, so the role of endogenous KDM3B at HOXA1 is inferred. Note the connection this beat did not pursue: HOXA1 is a HOXA-cluster gene and this graph's central fusion output is the HOXA cluster, but HOXA1 is not HOXA9 and no link is claimed.

    Evidence that argues against this (1)
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    KDM3B displays H3K9me1/2 demethylase activity, induces expression of the leukaemogenic oncogene LMO2 via synergy with CBP, represses leukaemia-cell differentiation, is down-regulated during ATRA-induced differentiation of HL-60 cells through corepressor recruitment, and is UP-regulated in blood cells from ALL-type leukaemia patients - the opposite direction from the myeloid literature, in the lineage of this graph's index disease.

    KDM3B is the H3K9 demethylase involved in transcriptional activation of lmo2 in leukemia. · Kim JY; Kim KB; Eom GH; Choe N; Kee HJ; Son HJ; Oh ST; Kim DW; Pak JH; Baek HJ; et al. · Mol Cell Biol 2012
    PMID 22615488DOI
    What this evidence cannot show

    **Abstract only; full text not retrieved.** The genome-wide arm is ChIP-chip rather than ChIP-seq, performed in HL-60, which is one of the two KDM3B-LOH lines used in E2 - so the system's baseline KDM3B level is already abnormal. 'Up-regulated in blood cells from ALL-type leukaemia patients' carries no cohort size, assay or KMT2A status in the abstract, so it cannot be read as a statement about KMT2A-rearranged ALL; it is recorded because it is the only lymphoid-lineage direction available and because this graph's index disease is infant B-ALL. Critically, this measures KDM3B in ESTABLISHED blasts while E2 measures its role in a normal cell BEFORE any fusion exists - the two are not the same time point and the apparent contradiction may be entirely a stage effect.

    Evidence that argues against this (1)
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    KDM3B is not a dependency in any DepMap lineage: mean Chronos gene effect +0.0232 across 1,258 screened cell lines, with only 0.4% of lines depleted below -0.5, and a lymphoid-tissue mean of +0.0060 over 97 lines.

    What this evidence cannot show

    Rule 83: **the API does not return the DepMap release**, so only the fetch date is on record, and 1,258 is KDM3B's own gene-specific screen count and must not be read as a corpus-wide denominator. A positive mean gene effect means knockout is neutral-to-mildly-favourable in the average line - which is the expected reading for a candidate tumour suppressor and is NOT evidence for the mechanism in E2, because a pooled viability screen cannot detect a locus-rearrangement phenotype at all. That is the deeper limitation: DepMap scores proliferation, and the phenotype here is a cytogenetic event in a surviving cell, so this assay is structurally blind to the thing KDM3B is claimed to do.

    Evidence that argues against this (1)
    • Neutrality in immortalised cell lines does not transfer to primary cells: a CRISPR screen in primary hematopoietic stem and progenitor cells found that Kdm3b loss specifically reduces the fitness of Idh2- and Tet2-mutant HSPCs. The dependency is genotype- and cell-state-specific and a pan-line mean cannot see it. Waarts MR; Mowla S; Boileau M; Martinez Benitez AR; et al. 2024
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    In this project's own genome-wide DepMap scan over 17,728 protein-coding genes, 11 Cellosaurus-annotated KMT2A-rearranged leukaemia lines and 43 other leukaemia lines, KDM3B ranks 9,382 (percentile 52.9) with a difference of +0.0018 (KMT2A-rearranged mean +0.0068, other leukaemia +0.0051) and pan-essentiality +0.0226, while its paralogue JMJD1C ranks 23 (percentile 0.13) with a difference of -0.4535 (KMT2A-rearranged mean -0.4697, other leukaemia -0.0162), pan-essentiality -0.0121 and empirical p versus the scan of 0.0016.

    What this evidence cannot show

    Read directly from the stored results file 2026-09-22 rather than quoted from a memo. The scan's own caveats apply in full and are not reduced here: **11 KMT2A-rearranged lines, none infant-derived**, Cellosaurus fusion annotation may be incomplete which biases toward the null, the empirical p is the genome-wide background fraction rather than a separate random draw, and **the DepMap release is not stated by the API** (rule 83). Rule 84 applies to the KDM3B row and is the reason it is reported as a contrast rather than a result: about half of any gene set sits above the median by construction, this graph's own base rate for the anti-selective half is 45.3%, so 'KDM3B is at rank 9,382' carries no information by itself. The JMJD1C row is the informative one - rank 23 of 17,728 with near-zero pan-essentiality - and it is a difference in cell-line dependency, not a demonstration of a fusion-driven mechanism.

    Evidence that argues against this (1)
    • A rank in a 54-line scan can move with the line set: this graph has recorded a stored analysis figure inverting between cycles, and 11 rearranged lines drawn from nine individuals is a small and non-infant sample. The contrast between rank 23 and rank 9,382 is large enough to survive modest instability, but neither number should be quoted as a fixed property of either gene. Iqbal Lab oncofusion knowledge graph 2026
  7. E7weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    KDM3B suppresses acute promyelocytic leukaemia progression by restricting chromatin accessibility and facilitating the ATRA-mediated degradation of PML/RARalpha.

    PMID 31592194DOI
    What this evidence cannot show

    **Title and bibliographic record verified via PubMed esummary; neither the abstract nor the full text was read for this entry, so the assay, system, effect size and whether 'restricting chromatin accessibility' is an ATAC measurement are all uncharacterised.** Confidence is therefore `weak` and the claim is recorded as a pointer, not as evidence anything should rest on. It is included only because it is one of the two independent tumour-suppressor reports in a fusion-driven leukaemia and because `fusion-pml-rara` exists in this graph, making it a genuine cross-family link worth someone reading. Cancer Cell International is a journal whose editorial standards over this period warrant the same caution this graph applies to Oncotarget.

    Evidence that argues against this (1)
  8. E8strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In an ex vivo co-culture CRISPR/Cas9 screening platform using primary hematopoietic stem and progenitor cells on bone-marrow endothelial cells, loss of Kdm3b AND of its paralogue Jmjd1c specifically reduces the fitness of Idh2- and Tet2-mutant HSPCs; Kdm3b loss in mutant cells decreases expression of critical cytokine receptors including Mpl, rendering mutant HSPCs preferentially susceptible to inhibition of downstream JAK2 signalling - and the authors nominate KDM3B as a genotype-specific therapeutic target in clonal haematopoiesis and myeloid malignancy.

    PMID 38819218DOI
    What this evidence cannot show

    **Abstract only; full text not retrieved**, so the screen depth, the number of guides, the effect sizes and whether the KDM3B effect is catalysis-dependent are uncharacterised. The genotypes are IDH2- and TET2-mutant clonal haematopoiesis - not a KMT2A fusion and not a lymphoid context - and murine gene symbols are used for the screen, so the system is at least partly mouse. This item is recorded not as evidence about KMT2A biology but because it is the sharpest available statement of the therapeutic direction OPPOSITE to this beat's mechanism, in the same cell type (primary HSPCs) and within a year of it.

    Evidence that argues against this (1)
    • The same cell type gives the opposite therapeutic reading in the other paper: chemical KDM3-family inhibition of cord-blood CD34+ HSPCs produced KMT2A copy gains and break-apart events. So 'inhibit KDM3B in primary HSPCs' is simultaneously a proposed therapy for clonal haematopoiesis and a demonstrated inducer of KMT2A rearrangement, and neither paper cites the other. Gray ZH; Chakraborty D; Duttweiler RR; Alekbaeva GD; Murphy SE; et al. 2023
  9. E9strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The compound PMID 37788669 calls 'KDM3i' is JDI-12, discovered in a virtual screen of 149,519 natural products and 33,765 Chinese-medicine components against the catalytic Jumonji domain of JMJD1C; JDI-4 and JDI-12 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 2020
    PMID 31271662DOI
    What this evidence cannot show

    **Abstract only; full text not retrieved**, so the SPR affinities ('modest' is the authors' own word, with no Kd in the abstract), the selectivity panel, the cell-line list and the concentrations are uncharacterised, and the abstract does not state which H3K9 methylation state the in vitro demethylation assay read out. The compound is a virtual-screening hit from a natural-product library, not an optimised chemical probe, and it has no reported selectivity data against KDM3A, the KDM4 family or other 2-OG oxygenases - so 'KDM3 family inhibitor' is a claim about two measured family members and not a selectivity profile. **The direct conflict to hold in view: this paper's headline is that these compounds spare cord blood cells, and PMID 37788669 uses the same compound at 1 microM for 72 h to generate KMT2A copy gains in cord-blood CD34+ HSPCs. Sparing viability and inducing rearrangement are compatible, and their therapeutic readings are not.**

    Evidence that argues against this (1)
    • The selectivity this compound is named for belongs to the paralogue, and this project's own DepMap scan says so independently: JMJD1C ranks 23rd of 17,728 for KMT2A-selective dependency while KDM3B ranks 9,382nd. Attributing the anti-MLLr activity of JDI-12 to KDM3B, as PMID 37788669's discussion does, is not supported by either the compound's discovery paper or the dependency data. Iqbal Lab oncofusion knowledge graph, internal analysis over Broad DepMap 2026
  10. E10strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    De novo and inherited pathogenic KDM3B variants - missense and truncating, consistent with haploinsufficiency - cause Diets-Jongmans syndrome in 14 unrelated individuals and three affected parents, comprising varying intellectual disability or developmental delay with short stature, infant feeding difficulties, joint hypermobility and a facial gestalt of wide mouth, pointed chin, long ears and low columella; and NOTABLY, two of those individuals developed cancer in childhood - acute myeloid leukaemia and Hodgkin lymphoma.

    De Novo and Inherited Pathogenic Variants in KDM3B Cause Intellectual Disability, Short Stature, and Facial Dysmorphism. · Diets IJ; van der Donk R; Baltrunaite K; Waanders E; et al. · Am J Hum Genet 2019
    PMID 30929739DOI
    What this evidence cannot show

    **Abstract only; full text not retrieved.** The cohort was assembled by exome sequencing plus gene matching on a developmental phenotype, so ascertainment is biased towards investigated and severely affected individuals; 17 individuals is not a population denominator and two cancers in 17 is not an incidence estimate. **Crucially, the abstract gives no karyotype, no cytogenetics and no KMT2A status for the AML case** - which is exactly the fact that would test the mechanism in E2, and it is why this item supports an experiment rather than a conclusion. The syndrome is a constitutional haploinsufficiency, i.e. lifelong partial loss, whereas E2's mechanism is driven by acute depletion; a constitutional state may be compensated in ways an siRNA is not. Published four years before the mechanism paper, which does not cite it.

    Evidence that argues against this (1)
    • If constitutional KDM3B haploinsufficiency were a strong KMT2A-rearrangement predisposition, the syndrome would be expected to present as a cancer-predisposition syndrome and it is not described as one - it is described as an intellectual-disability syndrome, with the two cancers reported as a notable incidental observation. Constitutional loss producing a mild or absent cancer phenotype is evidence against the acute-depletion result transferring to a whole organism over a lifetime. Diets IJ; van der Donk R; Baltrunaite K; Waanders E; et al. 2019
  11. E11established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    As of 2026-09-22, PubMed reaches 92 records by the approved symbol KDM3B[tiab], 22 by JMJD1B[tiab], 5 by 5qNCA[tiab] and 0 by JHDM2B[tiab]; the union is 114, so 22 records (19.3%) are reachable ONLY through a retired alias, with 18 of them JMJD1B-only and 5 5qNCA-only (one overlapping); KDM3B[tiab] AND CTCF[tiab] returns exactly 1 record, KDM3B[tiab] AND leukemia[tiab] returns 9, and "Diets-Jongmans"[tiab] returns 8 with no overlap with the leukaemia set.

    What this evidence cannot show

    **`data/symbol-collisions.json` was grepped for KDM3B, JMJD1B, JHDM2B, 5qNCA and KDM3A before the first query and returned no row (rule 88) - this is a new trap, not a rediscovery.** Titles and abstracts only. `JHDM2B` returning zero with `phrasesnotfound` is the reassuring case: the instrument reported that it could not find the phrase rather than silently returning a well-formed zero. The alias-only fraction is a floor, because the alias list is not exhaustive (KIAA1082 and C5orf7 were not run) and because a paper using both the symbol and an alias counts in the reachable set. Unlike the EHMT2/G9a row already in the registry, here the approved symbol is the BETTER operand - the failure is recall of the remaining fifth, not a wrong-first-hit collision.

    Evidence that argues against this (1)
    • A count of alias-only records overstates the practical loss if the alias-only papers are old and superseded: all five 5qNCA records predate 2008, and the 5qNCA literature is the MDS deletion-mapping work whose conclusion - KDM3B as a 5q31.2 candidate tumour suppressor - is already reachable through the modern symbol. The 18 JMJD1B-only records are the ones that matter, and they span 2005 to 2026. US National Library of Medicine 2026
Connected to