The KMT2A exon-11 anchor is not uniquely TARGETED by the KDM3B-G9a axis - it is unusually SENSITIVE to an H3K9me1 rise that happens almost everywhere, and the sensitivity is a property of the element
also called hypersensitive anchor, not a selective enzyme · the address is in the element, not the enzyme
Turning off one enzyme raises a silencing tag across the whole genome, yet only the KMT2A gene loses its anchoring protein and only that gene breaks. The proposal is that KMT2A is not singled out by the enzyme - its anchor is just unusually easy to knock off, so a general change lands on it hardest.
The same thing, for a scientist
Measurement in the founding paper's own deposited data shows the H3K9me1 rise on KDM3B depletion is **broad** - 1.85x on the *KMT2A* BCR's telomeric flank, 1.76x at the *GAPDH* promoter, 1.64x at *ACTB*, 1.38x at the non-responsive *MLLT1* - while the CTCF loss is **narrow**, 0.30x at *KMT2A* canonical exon 11 against 0.89-1.21x at eleven other intervals [E1]. The paper's own genome-wide figures point the same way: megabase-scale H3K9me1/2 skews (81% of 109 Mb and 75% of 292 Mb reset by double knockdown) and **17,077 CTCF peaks losing intensity**, of which only 1,005 are KDM3B co-occupied [E2]. **The hypothesis:** the enzyme axis sets a global H3K9me1/2 tone, and which loci convert that tone into CTCF eviction is decided by properties of each anchor - motif score and degeneracy, cohesin load, CpG methylation state of the motif, replication timing, and whether the anchor sits inside a transcribed exon. The *KMT2A* exon-11 anchor would then be near the top of a **sensitivity ranking**, and the disease association would follow from the anchor's biophysics plus the locus's pre-existing fragility, not from enzyme targeting. **This reframes the therapeutic target.** If the axis is global and the selection is at the element, a G9a inhibitor is a blunt instrument whose specificity is borrowed from the anchor, and the more precise intervention would be at the anchor - which is also, uncomfortably, the intervention the fragility literature predicts would raise breakage.
What makes it more than a restatement, and the ranking it predicts
It is falsifiable in one pass over four files. Call CTCF peaks from Gray's own control tracks (or take ENCODE RPE CTCF peaks), rank every anchor by siKDM3B/control CTCF ratio, and ask where KMT2A exon 11 sits. The hypothesis predicts KMT2A is extreme but not alone: in the top percentile of a continuous distribution with hundreds of anchors below 0.5x. The rival - enzyme targeting - predicts KMT2A is an outlier separated from the bulk, with the bulk near 1.0x. These two predictions are visibly different and the analysis is an afternoon.
It predicts which anchor properties should correlate with the ratio. In descending order of how testable they are from existing public data: (a) CTCF motif score and the number of degenerate positions - a PWM scan, desk job; (b) cohesin load, from the RAD21/SMC3 signal already in ENCODE and in Canela's HCT116 deposit; (c) CpG methylation at the motif, from any RPE or ENCODE WGBS track, since methylated CTCF motifs are the known mechanism of methylation-sensitive CTCF eviction; (d) whether the anchor lies inside an exon - KMT2A's does, which is unusual for a strong anchor and couples its occupancy to transcription.
And it explains the two loci that broke the mark-to-event mapping. MLLT1's marks rise 1.38x and its locus does not respond 1; on the hypersensitivity model that is expected, because MLLT1 has no comparably sensitive anchor for the mark to displace - its CTCF reads 0.93x, i.e. unmoved. The enzyme-targeting model has to treat MLLT1 as a mark change that simply does nothing.
The prior, stated honestly
0.45, and the reasons for not putting it higher. (a) The measurement behind it compares one anchor with eleven non-anchor intervals - gene bodies and promoters - which is the wrong comparison class and the single biggest weakness. (b) The paper's 17,077 lost CTCF peaks are consistent with the hypothesis but were computed by the authors with their own peak calls and thresholds, and this beat did not reproduce them. (c) A broad mark change with a narrow CTCF consequence is also exactly what you would see if KDM3B occupancy were locus-selective and the broad mark change were a separate, indirect phenomenon - and the paper does report KDM3B binding across KMT2A with a peak in the BCR, from reused colorectal data. That is a real alternative and it is not excluded here.
Why it is worth holding anyway. Both this hypothesis and its rival make the same near-term prediction about the intervention (a G9a inhibitor helps), so the distinction is not immediately actionable - but they diverge sharply on off-target risk. If the axis is global, then prophylactic G9a inhibition is changing H3K9 methylation at thousands of loci to protect one, and the loci it also changes include fusion-target promoters where mark-h3k9me3 argues the mark is restraining the disease. The hypothesis, if right, makes the safety question worse, not better, which is the direction that argues for testing it.
| statement | The KDM3B-G9a axis sets a broadly distributed H3K9me1/2 tone; which loci convert that tone into CTCF eviction and locus instability is decided by properties of each CTCF anchor (motif score and degeneracy, cohesin load, CpG methylation of the motif, replication timing, exonic position), and the KMT2A exon-11 anchor is near the top of that sensitivity ranking rather than being uniquely targeted by the enzymes. |
|---|---|
| origin | data-pattern |
| prior probability | 0.8 |
| cost of the decisive test | low - one pass over four bigWigs already proven readable, plus a CTCF anchor list from ENCODE RPE or HOMER peaks called from the same files. An afternoon. |
| what it changes if true | the specificity of a prophylactic G9a inhibitor would be borrowed from the anchor rather than from the enzyme, which makes the off-target question WORSE: thousands of loci change H3K9 methylation to protect one, and those loci include the fusion-target promoters where mark-h3k9me3 argues H3K9 methylation is restraining the disease. |
| localization | chromatin - CTCF/cohesin anchors genome-wide, and specifically the anchor inside KMT2A canonical exon 11 at chr11q23.3 |
| subnuclear localization | not measured. The hypothesis concerns loop-anchor occupancy, which implies a TAD-boundary position, but no imaging, Hi-C or nuclear-compartment assay bears on it here |
| cell type | hTERT-immortalised human retinal pigment epithelial (RPE) cells for the supporting measurement; the hypothesis is about anchors in general and is untested in any haematopoietic cell |
| cell state | asynchronous proliferating culture 72 h after siRNA transfection; replication timing is one of the proposed sensitivity determinants and was not measured |
| epigenetic readout | H3K9me1/2/3 and CTCF ChIP coverage; the proposed determinants of anchor sensitivity include CTCF motif CpG methylation and cohesin load, neither of which was measured here |
| shown for fusion | Not shown for any fusion. The hypothesis concerns the intact KMT2A locus before any rearrangement exists, measured in unrearranged RPE cells; no KMT2A::AFF1 or other fusion-bearing system is involved. |
| prior history | 0.45 at creation (cycle 59) -> 0.80 on 2026-09-22 (cycle 60), after the decisive anchor-matched test was run and this hypothesis's first stated falsifier failed to fire. |
predictions
[ "Ranking every CTCF anchor in Gray's own control tracks by siKDM3B/control ratio puts KMT2A exon 11 in the top percentile of a CONTINUOUS distribution with hundreds of anchors below 0.5x - not as an outlier separated from a bulk near 1.0x.", "CTCF motif score, cohesin (RAD21/SMC3) load and CpG methylation state at the motif correlate with the CTCF-loss ratio across anchors; the KMT2A anchor should be unusual on at least one of these axes.", "Loci whose marks rise without a sensitive anchor show no cytogenetic response - which is what MLLT1 does (me1 1.38x, CTCF 0.93x, non-responsive by FISH).", "KDM3B occupancy in RPE is NOT strongly enriched at KMT2A relative to the other loci that gained H3K9me1; if it is, the rival enzyme-targeting model wins." ]
experiments
[ "exp-h3k9me3-and-ctcf-at-the-kmt2a-anchor-in-a-cd34-progenitor" ]
analyses
[ "ana-q108-h3k9me3-at-the-kmt2a-bcr-measured", "ana-q108-mark-response-versus-cytogenetic-response-at-four-loci" ]
falsifiers
[ "An anchor-matched ranking of Gray's own CTCF tracks in which the bulk of comparable-strength anchors sit near 1.0x and KMT2A exon 11 is a separated outlier. That is enzyme targeting, and this hypothesis is dead.", "KDM3B ChIP in RPE showing strong, selective occupancy at KMT2A relative to GAPDH, ACTB, MLLT1 and TCF3 - the loci that also gained H3K9me1. The enzyme would then have a genuine address.", "A CTCF motif at KMT2A exon 11 that is unremarkable on every measurable property (motif score, cohesin load, methylation, position) while the locus still responds uniquely. The sensitivity would then have no material basis and the hypothesis would be empty.", "A demonstration that CTCF eviction at this anchor requires H3K9me1 deposited IN CIS at the anchor itself - for instance by a dCas9-tethered G9a that evicts CTCF at KMT2A and a tethering elsewhere that does not. Cis-sufficiency at a single anchor is compatible with targeting and not with a global-tone model." ]
the decisive test was run and this hypothesis survived
{
"source": "analyses/q111-ctcf-anchor-matched-ranking/ (design pre-registered before any signal value was read)",
"the_falsifier_as_stated": "an anchor-matched ranking in which the bulk of comparable-strength anchors sit near 1.0x and KMT2A exon 11 is a separated outlier - which would be enzyme targeting",
"what_was_measured": "the matched median is 0.6327x, not near 1.0x; 308 of 1,625 strength-matched anchors fall to <=0.5x, 64 to <=0.4x and 5 to <=0.3x; genome-wide 84.6% of 12,443 anchors fall to <=0.9x with median 0.6957x and 1,436 at <=0.5x; the KMT2A anchor sits at the 3.57th percentile of a CONTINUOUS distribution with no separated outlier",
"verdict": "THE FALSIFIER DID NOT FIRE. The distribution has the shape this hypothesis predicted and not the shape enzyme targeting predicted.",
"and_the_replicate_ambiguity_STRENGTHENS_this_hypothesis_rather_than_weakening_it": "The beat's pre-declared bias control puts the anchor at the 31.9th percentile rather than the 3.6th, because the two control replicates disagree two-fold at this element. That bracket is a problem for any claim about the DEGREE of sensitivity - but both ends of it say the anchor is not uniquely targeted, and the unfavourable end says it is even less exceptional. This hypothesis is therefore robust to the ambiguity that damages the entry it was competing with.",
"king_verification": "every figure above was recomputed by the king from the beat's persisted anchors_S.npy and anchors_R.npy and matched exactly"
}- If this hypothesis is right the therapeutic picture gets worse rather than better: a G9a inhibitor given to protect one anchor is altering H3K9 methylation across megabases, including at the fusion-target promoters where this graph's mark-h3k9me3 entry argues H3K9 methylation is what restrains the leukaemic programme. The hypothesis sharpens drug-unc0642's contraindication instead of softening it.
- It also suggests a more precise intervention - editing or shielding the anchor itself - whose direction of effect is genuinely unknown. Gray 2023 predicts weakening the anchor RAISES rearrangement; the loop-anchor fragility literature predicts CTCF-bound anchors are where breaks happen. An anchor-directed intervention could increase leukaemogenic risk, and this hypothesis would be the argument that led someone there.
- The hypothesis is built on a comparison between one anchor and eleven non-anchors. It is written as speculative with a 0.45 prior and a named decisive test precisely because the tempting move - treating a locus-specific CTCF number as proof of a mechanism-level claim - is the move this project has published and retracted before.
- ANSWERED IN CYCLE 60, IN THIS HYPOTHESIS'S FAVOUR. The KMT2A exon-11 anchor ranks 58th of 1,625 anchors matched within +/-10% of its baseline strength - the 3.57th percentile - and 258th of all 12,443 anchors (2.07th percentile). The question correctly said this single number would decide between this hypothesis and enzyme targeting, and it decided for this hypothesis: the matched median is 0.633x and the distribution is continuous with no separated outlier. What remains open is the DEGREE, which is a bracket from the 3.6th to the 31.9th percentile because the two control replicates disagree two-fold at this element.
- Is the CTCF motif at KMT2A exon 11 CpG-methylated in any cell type, and does its methylation state change on KDM3B depletion? Methylation-sensitive CTCF eviction is a known mechanism and would be a concrete material basis for hypersensitivity. No WGBS track at this element has been read by this project.
- Does an anchor sitting inside a transcribed exon behave differently from an intergenic anchor under an H3K9me1 rise? KMT2A's anchor is exonic, which is unusual for a strong anchor, and this is a general chromatin question with a public-data answer.
- The second falsifier is still untested and is now the one that matters: does KDM3B ChIP in RPE show strong, selective occupancy at KMT2A relative to GAPDH, ACTB, MLLT1 and TCF3 - the loci that also gained H3K9me1? A positive there would give the enzyme a genuine address and would resurrect targeting despite the ranking. This is why the prior is 0.80 and not higher.
- Does the element's sensitivity have a material basis that can be measured rather than inferred? CTCF loss scales with baseline anchor strength across the genome (decile medians 0.779 -> 0.615) and this anchor is at the 91.7th strength percentile, so part of its sensitivity may be nothing but its strength. Whether anything about the element beyond its strength contributes is untested, and the strength-matched design was built precisely to hold strength constant - so the residual 3.57th-percentile position is what needs explaining.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
In Gray 2023's deposited RPE data, KDM3B depletion raises H3K9me1 at loci well beyond KMT2A (1.85x on the BCR's telomeric flank, 1.76x at GAPDH promoter, 1.64x at ACTB promoter, 1.46x over TCF3, 1.38x over MLLT1) while lowering CTCF occupancy only at the KMT2A exon-11 anchor (0.30x) and at none of eleven other intervals (0.89x-1.21x), and MLLT1 shows a mark rise with no CTCF change and no cytogenetic event.
GEO GSE210480 H3K9me1/2/3 and CTCF bigWigs read over twelve intervals with pybigtools 0.3.0, 2026-09-22; scripts and full output at analyses/q108-gse210480-h3k9me3-kmt2a-bcr/ · 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 · NCBI GEO (data accompanying Cell 186:4528-4545.e18) 2023What this evidence cannot show
Eleven of the twelve intervals are gene bodies or 1 kb promoter windows, not CTCF anchors, so the comparison class is wrong for the strong version of the claim; a gene body dilutes an anchor-level change. n = 2 per arm from a single CTCF experiment (four unique files under eight accessions). Histone tracks 1 kb binned, CTCF 200 bp. RPE cells, epithelial, no KMT2A rearrangement. This item supports the OBSERVATION that the mark response is broad and the CTCF response narrow at these intervals; the hypersensitivity INTERPRETATION is this entry's proposal, not a measurement.
Evidence that argues against this (1)
- The same observation is equally consistent with the rival model in which KDM3B occupancy IS locus-selective and the broad mark rise is indirect: the paper reports KDM3B bound across KMT2A with a strong peak in the breakpoint cluster region, which would give the enzyme a genuine address. That evidence comes from reused ChIP-seq in colorectal cancer stem cells (GSE71885), not from RPE, which weakens it - but it is not excluded, and if KDM3B occupancy in RPE is strongly enriched at KMT2A relative to the loci that also gained H3K9me1, this hypothesis loses its main motivation. Gray ZH et al. 2023
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The founding paper's own genome-wide figures describe a broad axis rather than a KMT2A-selective one: double knockdown resets H3K9me1 across 81% of the affected genome (88 of 109 Mb) and H3K9me2 across 75% (219 of 292 Mb), 6,386 of ~15,400 strong KDM3B peaks overlap RPE CTCF peaks, and of 17,077 CTCF peaks that lose intensity on KDM3B depletion only 1,005 are KDM3B co-occupied.
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 2023What this evidence cannot show
These figures are recorded in this graph's `mechanism-h3k9me1-2-ctcf-balance-gates-kmt2a-rearrangement` from the paper's text and were not re-derived by this beat. The 17,077 figure in particular depends on the authors' peak-calling and intensity-change thresholds, which are not restated here; a permissive threshold inflates such a count. Megabase rescue figures describe the fraction of AFFECTED genome reset, not the fraction of the genome affected, which is a different and smaller quantity.
Evidence that argues against this (1)
- That only 1,005 of 17,077 CTCF-losing peaks are KDM3B co-occupied cuts BOTH ways: it argues the CTCF loss is mostly indirect and broad, which supports this hypothesis - but it also means the paper's own strongest statistical association (P < 1.0e-216, Z = 143.38) is for the co-occupied minority, and KMT2A exon 11 may well belong to that minority, in which case it IS a directly targeted site and the hypothesis is wrong for the one locus that matters. Iqbal Lab oncofusion knowledge graph 2026
the broad-mark / narrow-CTCF measurement is the pattern this hypothesis explains
proposes the enzymes set a global tone and the element selects itself, against a locus-targeted first step
the haematopoietic test; a ChIP-seq version also yields the anchor ranking this hypothesis needs
the anchor whose sensitivity is at issue, now localised in four experiments across three cell types
if the axis is global, the inhibitor's specificity is borrowed from the anchor and the off-target question worsens
MLLT1's mark rise without a CTCF change or an event is the pattern the hypothesis predicts
its first prediction confirmed: a continuous distribution, 1,436 anchors below 0.5x, KMT2A at the 2.07th percentile, no separated outlier
the population-level form of that hypothesis: the axis is graded, and the anchor is the strong end of a continuum