CTCF loss on KDM3B depletion is a graded property of anchor strength, not a set of targeted events: decile medians fall monotonically 0.78 to 0.62 across 12,443 RPE anchors, and if that gradient is biological then every strong anchor in the genome is a candidate instability site
also called the strength gradient · strong anchors are more CTCF-labile · 0.78 to 0.62 across ten deciles
When one enzyme is removed, loop-anchor protein falls off sites all over the genome - and the stronger a site was to begin with, the larger the proportional loss. If that is real biology rather than a statistical artefact, then the KMT2A gene is not being singled out; it is simply one of the strongest anchors, and every other strong anchor is at the same kind of risk.
The same thing, for a scientist
Measured across **12,443 CTCF anchors** called from Gray 2023's own RPE control tracks, the siKDM3B/control occupancy ratio declines **monotonically** with baseline anchor strength - decile medians 0.779, 0.751, 0.731, 0.731, 0.714, 0.694, 0.672, 0.653, 0.640, **0.615** [E1]. **84.6% of anchors fall to 0.9x or below and the population median is 0.696**, so this is a genome-scale graded response and not a set of discrete targeted events. **The hypothesis:** the KDM3B-G9a axis does not select anchors. It lowers CTCF occupancy across the anchor population with an effect size that scales with how much CTCF was bound, and the *KMT2A* anchor's apparent specialness (0.396x, 3.6th percentile of its strength class) is the joint consequence of (a) being in the top strength decile and (b) sitting in the upper tail of a broad within-class distribution whose median is already 0.633. **If true, the locus address is not on the mark arrow and not on the CTCF arrow** - it is downstream, in whatever converts CTCF loss into a cytogenetic event, and *KMT2A*'s pre-existing fragility is the obvious candidate for that step. **The competing reading, which these data cannot exclude, is that the gradient is regression to the mean.** Anchors are selected and matched on the same control signal that forms the ratio's denominator, so the strongest deciles are enriched for upward noise and their ratios are pushed down by construction. The pre-declared split-replicate control - anchor calling and strength from control replicate 1, denominator from control replicate 2 alone - is exactly the test for that, and it shows the artefactual component is **substantial**: the *KMT2A* anchor moves from the 3.6th to the 31.9th percentile [E1]. **It does not show the gradient disappears**, because a one-replicate denominator adds independent noise to every anchor and the design cannot separate the two effects at n = 2 per arm. **So the prior here is genuinely near even**, and the value of the hypothesis is that it names a discriminating experiment that costs nothing new.
Why it matters even if the gradient is partly artefactual
Because the discriminating experiment is the same either way. An independent-replicate design - call anchors and measure baseline strength in one pair of CTCF replicates, compute the ratio from a different pair - removes the shared-noise component entirely. It needs n >= 4 control replicates per arm, which GSE210480 does not have: it has two, deposited twice under eight accessions. So the test is a new ChIP in RPE, or an existing CTCF ChIP dataset with four or more replicates and a KDM3B or G9a perturbation - and no such dataset exists, because there is no siG9A and no double-knockdown CTCF arm anywhere in this series (concept-one-deposited-file-under-two-accessions-doubles-a-replicate-count).
And because the therapeutic reading turns on it. If the gradient is biological, a KDM3B or G9a intervention destabilises the strongest loop anchors in the genome preferentially - which is a systematic, predictable, and rather alarming off-target profile that could be screened for. If it is statistical, the intervention's effect on any individual anchor is smaller and noisier than the ranking suggests, and the whole KMT2A-selectivity argument loses its quantitative footing without gaining a new hazard. Both readings argue against prophylactic use and for different reasons, which is worth stating plainly because the published framing argues for it.
The cheap tests that do not need new data
- Overlay the ranking on Gray's own amplified and rearranged regions. If CTCF loss were sufficient for a cytogenetic event, the 57 anchors that lose more CTCF than KMT2A should be enriched for the copy-number changes the paper reports. The anchor table is persisted (
anchors.tsv, 12,443 rows with coordinates); the paper's copy-number data are in its figures and supplements. Seeexp-overlay-the-ctcf-loss-ranking-on-grays-own-amplified-regions. - Test the same gradient in the H3K9me1 arm. If the H3K9me1 rise on KDM3B depletion also scales with baseline H3K9me1 level, then the chr11 bin ranking that places the KMT2A BCR at 72 of 18,082 needs the same matching correction, and both arms of the published chain become graded rather than targeted. The tracks have already been read by this project at 1 kb.
- Join the matched anchors against public annotation. Motif score, motif orientation, CpG methylation, replication timing and exonic position for the 1,625 strength-matched anchors, asking whether any of them predicts the ratio. A positive answer would give the sensitivity a material basis; a null would say the within-class variation is noise, which would itself resolve this hypothesis against the biological reading.
| statement | The KDM3B-G9a axis lowers CTCF occupancy across the CTCF anchor population with an effect size graded by baseline anchor strength, rather than selecting particular anchors; the KMT2A exon-11 anchor's response is the joint product of being in the top strength decile and sitting in the upper tail of a broad within-class distribution, and the locus address of the published rearrangement brake therefore lies downstream of CTCF, in whatever converts occupancy loss into a cytogenetic event. |
|---|---|
| origin | data-pattern |
| prior probability | 0.5 |
| what it changes if true | The specificity of any KDM3B or G9a intervention would be borrowed entirely from downstream fragility, not from the enzyme and not from the anchor's response. A prophylactic G9a inhibitor would be altering loop occupancy at the strongest anchors genome-wide - a predictable, screenable, and large off-target surface - to protect one locus. It also relocates the interesting question from 'why this anchor' to 'why this locus breaks when its anchor weakens, while thousands of others do not', which is a question about replication, transcription and repair at 11q23.3 rather than about H3K9 methylation. |
| what it changes if false | If the gradient is pure regression to the mean, the KMT2A anchor's rank is bracketed between the 3.6th and 31.9th percentile and cannot be narrowed from this deposit, the eleven-interval contrast is still disqualified, and the honest position becomes that the CTCF step's locus specificity is UNMEASURED rather than refuted. |
| cost of the decisive test | low for the overlay against Gray's own copy-number data and for the H3K9me1 gradient (both are re-analysis of data already read); med for a four-replicate CTCF ChIP in RPE; the opposing-perturbation test is not available because the series has no siG9A CTCF arm |
| localization | chromatin, the CTCF anchor population of the human genome; no cellular or subnuclear localisation is asserted |
| subnuclear localization | not measured and not claimed. Loop anchors imply TAD-boundary positions but no imaging or Hi-C supports any statement here. |
| cell type | hTERT-immortalised human retinal pigment epithelial (RPE) cells - the only cell type in which the gradient has been measured. Nothing in this hypothesis is supported in a haematopoietic or KMT2A-rearranged cell. |
| cell state | asynchronous proliferating culture 72 h after control or KDM3B siRNA transfection |
| epigenetic readout | CTCF ChIP-over-control occupancy at 200 bp across 12,443 anchors, stratified by baseline occupancy decile |
| shown for fusion | Not shown for any fusion. The gradient is measured in unrearranged RPE and no fusion-bearing cell has been assayed for CTCF occupancy anywhere in this series. |
predictions
[ "An independent-replicate design (anchors and baseline strength from one pair of control replicates, ratio from a different pair) retains a monotone decline of the ratio across baseline-strength deciles. If the gradient is regression to the mean it flattens.", "The anchors that lose the most CTCF are NOT enriched for the copy-number changes Gray reports, because CTCF loss is graded and common while cytogenetic events are rare - so CTCF loss is necessary and not sufficient.", "Within the 1,625 anchors matched to KMT2A on strength, the ratio is predicted by measurable material properties of the anchor - motif score and degeneracy, CpG methylation of the motif, cohesin load, replication timing - rather than being noise.", "The same graded relationship appears for the H3K9me1 RISE: the fold increase on KDM3B depletion scales with baseline H3K9me1 level across bins, which would make the chr11 rank-72-of-18,082 figure for the KMT2A BCR require the same matching correction.", "A G9a inhibitor given to RPE cells raises CTCF occupancy preferentially at the strongest anchors, mirroring the loss - the opposing-perturbation test, which cannot be run from this deposit because no siG9A CTCF arm exists." ]
experiments
[ "exp-overlay-the-ctcf-loss-ranking-on-grays-own-amplified-regions" ]
analyses
[ "ana-q111-anchor-matched-ranking-of-the-kmt2a-ctcf-anchor" ]
falsifiers
[ "An independent-replicate design in which the baseline-strength gradient flattens to nothing. The gradient would then be entirely regression to the mean, this hypothesis is empty, and the correct statement is only that an unmatched contrast cannot support a specificity claim.", "An overlay showing that the most CTCF-losing anchors ARE enriched for amplification or rearrangement in Gray's own data. CTCF loss would then be close to sufficient, the graded response would predict widespread instability, and a model in which the axis is non-selective could not explain why only KMT2A is reported.", "A demonstration that CTCF eviction at an anchor requires H3K9me1 deposited IN CIS at that anchor - for instance a dCas9-tethered G9a that evicts CTCF where it is tethered and nowhere else. Cis-requirement at single anchors is compatible with targeting and not with a graded global response.", "A measurement showing the KMT2A anchor's ABSOLUTE CTCF occupancy falls below some threshold that its strength-matched peers do not cross, despite a similar fold change. A threshold model would make the rank irrelevant and this hypothesis the wrong frame rather than false.", "KDM3B ChIP in RPE showing strong selective occupancy at KMT2A relative to the other loci that respond. The enzyme would then have a genuine address and the non-selective premise fails at the first step." ]
- If this hypothesis is right, the intervention case gets worse in a specific and quantifiable way: a KDM3B or G9a perturbation moves CTCF at 84.6% of anchors, and the strongest anchors - which include the boundaries of many developmentally regulated domains - move most. That is a mechanism for broad, structured, dose-dependent harm in a developing infant, and it is not addressed anywhere in the published framing.
- If this hypothesis is wrong because the gradient is statistical, the graph must not treat that as reassurance. A flat gradient would leave the KMT2A anchor's rank unresolved between the 3.6th and 31.9th percentile, which is not a licence for the specificity claim - it is a statement that the deposit cannot decide, and the entry that made the claim would still need its 'only' removed.
- This graph's two H3K9 entries already predict opposite signs for the same manipulation, and this hypothesis adds a third frame in which the sign at any single anchor is partly noise. Any safety statement must say that the magnitude and, in a blood cell, the sign of the effect at this element are both unresolved - not pick whichever of the three numbers supports the argument being made.
- Is there any published CTCF ChIP-seq dataset with four or more control replicates and a KDM3B, G9a or H3K9-pathway perturbation in the same cells? That single question decides whether the discriminating test is a re-analysis or a wet experiment, and it has not been asked of GEO, ENCODE or the IHEC federation.
- Does the strength gradient appear for CTCF loss under OTHER perturbations - heat shock, replication stress, cohesin depletion, DNA-methylation change? If a gradient of the same shape appears whenever CTCF occupancy is perturbed at all, it is a property of the measurement rather than of this axis, and that would be the strongest evidence for the regression-to-the-mean reading.
- What fraction of the 20 anchors at or below 0.3x lie in genes with reported focal amplification in any cancer? A quick join against a focal-amplification catalogue would say whether the extreme tail of this ranking is enriched for instability at all, and it needs no new data.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Across 12,443 CTCF anchors called from the two siCT control CTCF tracks of GSE210480 in hTERT-RPE cells, the median siKDM3B/control occupancy ratio declines monotonically with baseline anchor strength from 0.779 in the weakest decile to 0.615 in the strongest, the population median is 0.696, 10,521 anchors (84.6%) sit at or below 0.9x, 1,436 at or below 0.5x, 280 at or below 0.4x and 20 at or below 0.3x; the KMT2A exon-11 anchor is at the 91.7th percentile of baseline strength with a ratio of 0.396x, rank 58 of 1,625 strength-matched anchors whose median is 0.633; and a pre-declared split-replicate control that removes the shared-noise component of the selection moves that anchor from the 3.57th to the 31.9th percentile, showing the artefactual component of the gradient is substantial without showing it is the whole of it.
GEO GSE210480 - four unique RPE CTCF ChIP bigWigs read genome-wide at 200 bp with pybigtools 0.3.0, anchors called from the control tracks only, pre-registered design with a split-replicate bias control and a threshold sweep, 2026-09-22; analyses/q111-ctcf-anchor-matched-ranking/design.py, run.py, results.json, anchors.tsv · 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
One dataset, one cell type, one perturbation, n = 2 versus n = 2, no spike-in, no input file, and no FDR on the anchor calls. The gradient's biological and statistical components cannot be separated in this design - that is the hypothesis's central uncertainty and the reason its prior is set near even rather than high. Anchor strength is a RATIO-track magnitude and would not be numerically comparable with a raw-coverage strength measure. The monotonicity is in decile MEDIANS; the within-decile spread is wide (population IQR 0.58-0.83), so the gradient is a shift in a broad distribution and not a tight relationship.
Evidence that argues against this (1)
- The paper reports a phenotype this hypothesis does not explain. KDM3B depletion produces copy gain and break-apart at KMT2A and, by FISH, at TCF3 and AFF3 but not at MLLT1 or MLLT3 - a discrete, locus-patterned outcome. A purely graded CTCF response predicts no such pattern unless something downstream supplies it, and the hypothesis therefore requires an unmeasured second factor to be doing all the locus-selection work. That is a real weakness: a model that explains the CTCF data and none of the cytogenetics is not obviously better than one that explains the cytogenetics and overstates the CTCF specificity. Gray ZH et al. 2023
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Depleting CTCF is by itself sufficient to generate MLL/KMT2A amplifications and rearrangements, and KDM3B depletion increases H3K9me1/2 and reduces CTCF occupancy at the locus while CTCF overexpression rescues the doxorubicin-induced alterations - so CTCF occupancy is the step the published mechanism treats as causal, which is what makes its genome-wide behaviour decisive for the specificity question.
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
Taken from the paper's abstract, which is what establishes that CTCF depletion alone suffices; the figure-level detail of which cells, what degree of depletion and what amplification frequency was not re-read for this entry. CTCF depletion being sufficient at KMT2A says nothing about whether it is sufficient elsewhere - the paper does not report a genome-wide survey of which loci amplify on CTCF loss, and that absence is precisely the gap the overlay experiment proposed here would fill.
Evidence that argues against this (1)
- Global CTCF depletion is not the same perturbation as partial occupancy loss at one anchor. A whole-cell CTCF knockdown dissolves loop architecture everywhere and could produce KMT2A amplification through an indirect route - replication stress, transcription-replication conflict, or loss of a boundary many kilobases away - so the sufficiency result does not license treating a 0.40x local occupancy change as the causal quantity. That cuts against this hypothesis and against the entry it argues with equally. Gray ZH et al. 2023
the strength gradient measured across 12,443 anchors is what generated this hypothesis
the overlay decides whether CTCF loss is sufficient or merely necessary
the population-level form of that hypothesis: the axis is graded, and the anchor is the strong end of a continuum
the perturbation whose CTCF footprint is proposed to be graded rather than targeted