Oncofusion

start / graph / exp-overlay-the-ctcf-loss-ranking-on-grays-own-amplified-regions

Experimentmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q111-ctcf-anchor-ranking

The decisive test that needs no new data: overlay the 12,443-anchor CTCF-loss ranking on the copy-number changes Gray 2023 reports, and ask whether the 280 anchors that lose CTCF as hard as KMT2A also amplify

also called is CTCF loss sufficient? · the overlay · 280 anchors, one reported amplification

We now know that thousands of loop anchors lose their anchor protein when this enzyme is removed, and that the KMT2A gene is one of the harder-hit ones but far from unique. The question that settles the mechanism is whether all those other hard-hit anchors also get copied and broken. If they do not, then losing the anchor protein is not enough on its own, and something else about KMT2A is doing the work.

The same thing, for a scientist

`ana-q111-anchor-matched-ranking-of-the-kmt2a-ctcf-anchor` establishes that **280 of 12,443 RPE CTCF anchors fall to 0.4x or below on KDM3B depletion and 1,436 to 0.5x or below**, with *KMT2A* at rank 258 [E1]. PMID 37788669 reports that **CTCF depletion alone is sufficient** to produce *MLL/KMT2A* amplification and rearrangement, and that KDM3B depletion produces those events at *KMT2A* [E2]. Those two statements are in tension: if a 0.4x local occupancy loss were the operative cause, hundreds of loci should behave like *KMT2A*, and the paper reports focal alterations at a handful. **The experiment is an overlay, and every input already exists.** Take the persisted anchor table (`analyses/q111-ctcf-anchor-matched-ranking/anchors.tsv`, 12,443 rows with hg19 coordinates, baseline strength, ratio and a replicate-consistency flag). Take the copy-number and rearrangement calls Gray reports on KDM3B depletion - its amplification analysis and its FISH panel. Ask three questions with pre-committed readouts. **The readouts, committed in advance:** | result | reading | |---|---| | the top 280 CTCF-losing anchors are **enriched** for amplified regions | CTCF loss is close to sufficient; the axis is broadly destabilising; *KMT2A* is one of many and the safety case against the intervention is much stronger | | the top 280 are **not enriched**, and *KMT2A* is amplified while equally-losing anchors are not | **CTCF loss is necessary and not sufficient.** The locus address is downstream, and `hyp-ctcf-loss-on-kdm3b-depletion-scales-with-baseline-anchor-strength` is supported while `mechanism-kdm3b-depletion-lowers-ctcf-only-at-the-kmt2a-anchor`'s locus-address conclusion is relocated rather than restored | | the paper's amplified regions have **unremarkable** CTCF ratios | CTCF occupancy is not the operative variable at all, and both this graph's framing and the paper's are wrong about the causal step | **The single cheapest version** is one locus: **hg19 chr11:8,241,000 = GRCh38 chr11:8,370,285**, baseline strength 4.95x against *KMT2A*'s 4.92x and ratio **0.281x** against *KMT2A*'s 0.396x - a near-exact strength match, on the same chromosome, losing MORE CTCF [E1]. **Does it amplify?** If Gray's copy-number data cover chr11 at all, that is a yes-or-no answer from one figure panel, and it is a better control than any of the eleven intervals the published entry used.

Go deeper
Design, controls, and the failure modes that would make it uninterpretable

Inputs. (1) anchors.tsv - 12,443 anchors, hg19, with strength, ratio and consistency. (2) Gray 2023's amplified-region calls on KDM3B depletion, and its FISH-scored loci (KMT2A, TCF3, AFF3 responsive; MLLT1, MLLT3 non-responsive). (3) A build check: the anchor table is hg19 and any published region list must be lifted or confirmed, per this project's own offset discipline.

Statistic. Permutation, not a chi-square: draw 280 anchors at random matched on baseline strength decile and count overlaps with amplified regions, 20,000 draws. Matching on strength is mandatory for the same reason it was mandatory in the ranking - strong anchors sit preferentially in gene-dense, early-replicating, amplification-prone regions, and an unmatched permutation would recover that instead of the effect.

Controls that must be reported.

  • A spike. KMT2A itself must be recovered as an overlap by whatever region-matching code is used. If the code cannot find the one locus everybody agrees is amplified, no count it produces means anything. This project has published a confident wrong number from a lost spike before.
  • The dropped count. How many amplified regions contain no called anchor at all, and how many anchors fall in regions the copy-number analysis did not cover. If the amplified regions are mostly anchor-free, the overlay is testing coverage rather than biology and must say so.
  • A negative locus set. MLLT1 and MLLT3, FISH-scored non-responsive, should not be enriched for low-ratio anchors. If they are, the FISH calls and the CTCF ratios are decoupled and the whole chain needs re-examining - which is a real possible outcome, since ana-q108-mark-response-versus-cytogenetic-response-at-four-loci already found the MARK response contradicts the cytogenetic call at two of four loci.

What would make it uninterpretable, stated in advance. If Gray's copy-number analysis is reported only as gene-level or arm-level calls rather than as genomic intervals, the overlay cannot be run at anchor resolution and the result must be withheld rather than computed at a coarser scale and reported as though it answered the question. A partial fetch must suppress the verdict, not shrink it.

Facts and figures
designJoin the persisted 12,443-anchor CTCF-loss ranking (analyses/q111-ctcf-anchor-matched-ranking/anchors.tsv, hg19) against the amplified and rearranged regions Gray 2023 reports on KDM3B depletion, and test whether the most CTCF-losing anchors are enriched for those regions by a permutation matched on baseline-strength decile. The single-locus version tests one anchor: hg19 chr11:8,241,000 = GRCh38 chr11:8,370,285, strength 4.95x, ratio 0.281x - a closer strength match to the KMT2A anchor than any of the eleven intervals the published entry used, and a larger CTCF loss.
readoutenrichment of the top 280 (ratio <= 0.4x) and top 1,436 (<= 0.5x) anchors in amplified regions, as a permutation p against 20,000 strength-decile-matched draws; plus the yes/no amplification status of chr11:8,370,285; plus the CTCF ratios at the paper's own amplified regions as the reciprocal test
controlsA SPIKE - the KMT2A anchor must be recovered as an overlap by the same region-matching code, or the result is withheld. The DROPPED COUNT - how many amplified regions contain no called anchor, and how many anchors sit outside the copy-number analysis's coverage, both named in the output. A NEGATIVE LOCUS SET - MLLT1 and MLLT3, FISH-scored non-responsive, must not be enriched for low-ratio anchors. A BLACKLIST CHECK on the top-ranked anchors before any of them is treated as a comparator, since no blacklist was applied when they were called. And a BUILD CHECK, because the anchor table is hg19 and any published region list must be confirmed or lifted with the offset discipline this project already uses.
cost tierlow
timelineone to two days, most of it reading the paper's figures and supplements to find out whether the amplified regions are published as genomic intervals; the join and the permutation are minutes
what it decideswhether CTCF occupancy loss is sufficient for a focal cytogenetic event or merely necessary - which is the difference between a genome-wide instability mechanism with an enormous off-target surface and a locus-specific one whose address lives downstream of CTCF, in the fragility of 11q23.3
why it has not been runthe anchor ranking that supplies its main input was computed for the first time on 2026-09-22; before that there was no list of comparably-losing anchors to overlay
if uninterpretableif the paper reports copy-number changes only at gene or arm level, the overlay cannot be run at anchor resolution and the verdict must be SUPPRESSED rather than recomputed at a coarser scale - a partial input suppresses the conclusion, it does not shrink it
localizationchromatin, genome-wide; an overlay of two interval sets, asserting nothing about cellular localisation
subnuclear localizationnot applicable and not claimed - no imaging or Hi-C is involved
cell typehTERT-immortalised human retinal pigment epithelial (RPE) cells, because both inputs come from that system. A positive or negative result licenses nothing about a haematopoietic or KMT2A-rearranged cell, and the experiment cannot be repaired on that axis without new data.
cell stateasynchronous proliferating culture 72 h after control or KDM3B siRNA transfection
epigenetic readoutCTCF ChIP-over-control occupancy loss per anchor, used as the predictor; copy-number and break-apart status as the outcome
shown for fusionNot applicable to any fusion. Both inputs are from unrearranged RPE cells; the outcome measured by the paper is focal amplification and break-apart of the intact KMT2A locus, which is a pre-fusion event.
KING CORRECTION - the GRCh38 lift of the comparator anchor
{
  "THE_STATED_GRCh38_COORDINATE_IS_WRONG": "This beat states 'hg19 chr11:8,241,000 = GRCh38 chr11:8,370,285'. That is wrong. 8,370,285 - 8,241,000 = 129,285, which is the GRCh37->GRCh38 offset at the KMT2A locus at 118.4 Mb; it was applied here as though the offset were constant along chromosome 11, and it is not.",
  "the_correct_lift_measured_by_the_king_2026_09_22": {
    "hg19 chr11:8,241,000": "GRCh38 chr11:8,219,453, offset -21,547",
    "controls proving the offset varies": {
      "hg19 chr11:8,000,000": "GRCh38 7,978,453, offset -21,547",
      "hg19 chr11:50,000,000": "GRCh38 49,978,464, offset -21,536",
      "hg19 chr11:118,358,874 (the KMT2A element)": "GRCh38 118,488,159, offset +129,285"
    },
    "so": "the offset is locally constant but varies by more than 150 kb along chromosome 11 and changes sign; a single locus's offset is never transferable to another locus",
    "source": "Ensembl REST map/human/GRCh37/11:start..end:1/GRCh38, filtered on seq_region_name == '11'"
  },
  "NOTHING_MEASURED_IS_AFFECTED": "The GSE210480 CTCF bigWigs declare exactly hg19 and the whole 12,443-anchor ranking was computed in hg19. Every strength, ratio, rank and percentile in this beat stands. The error is confined to a convenience coordinate quoted for the reader.",
  "AND_THE_COMPARATOR_IS_NOW_ANNOTATED_AT_THE_CORRECT_POSITION": {
    "why_this_is_here": "this beat's own caveat said 'the chr11:8,241,000 comparator has not been annotated: what gene or element it sits in was not looked up'. It has now been looked up - at the corrected position, because the first attempt used the wrong one.",
    "GRCh38 chr11:8,219,453 overlaps": "lncRNA ENSG00000307368, 8,202,758-8,224,376",
    "LMO1 (protein_coding, 8,224,309-8,268,807)": "4,856 bp away",
    "other features within 50 kb": "ENSG00000307401 lncRNA 9,812 bp; ENSG00000307386 lncRNA 12,986 bp; RIC3-DT lncRNA 39,206 bp",
    "THE_COMPARATOR_IS_NOT_INERT": "LMO1 is an established T-lineage oncogene and neuroblastoma susceptibility gene, and the anchor sits 4.9 kb from it inside a lncRNA. So this is NOT the featureless intergenic control the proposal implies. It remains a valid STRENGTH-matched comparator - that was the matching criterion and it holds - but 'a strong anchor that loses more CTCF than KMT2A and has no disease relevance' is not something this annotation supports.",
    "what_is_still_unchecked": "the beat's other caveat stands and is not addressed here: whether this anchor is in a blacklist region or a segmental duplication was not tested, and a strong anchor with a large apparent change is what a blacklist artefact looks like"
  },
  "found_by": "the king, while running this beat's own recommended next step"
}
How this could bite us back
  • The outcome that would be worst for the intervention is the one this experiment is most likely to produce cleanly: if the most CTCF-losing anchors ARE enriched for copy-number change, then a KDM3B or G9a perturbation is a broad-spectrum genome-destabilising agent, and giving it prophylactically to an infant to prevent one rearrangement would be trading one focal event for hundreds. That reading has to be stated before the experiment runs, not after, because it is the reading a proponent would be slowest to volunteer.
  • The opposite outcome is not reassuring either. If CTCF loss is necessary and not sufficient, the protective mechanism the published chain proposes is acting on a step that thousands of loci share, and its selectivity for KMT2A comes entirely from a downstream fragility nobody is drugging. A drug whose specificity is borrowed from an unmeasured downstream property is a drug whose specificity has not been demonstrated at all.
  • And the experiment could produce a false negative that looks decisive. Copy-number analysis in a 72-hour siRNA experiment detects events in a population, so a rare event at a strongly-losing anchor would be invisible while a frequent one at KMT2A is not - which means 'not enriched' may mean 'not frequent enough to detect', and the design must report its detection floor rather than treating absence of overlap as absence of instability.
Still open
  1. Does Gray 2023 publish its amplified regions as genomic intervals? Everything else about this experiment is settled and that one question decides whether it is a two-hour join or impossible. It is answerable by opening the paper's supplementary tables.
  2. What is at hg19 chr11:8,241,000? The strength-matched, harder-losing comparator has not been annotated - gene, element, blacklist status and segmental-duplication status are all unknown, and all four are one lookup each. Until they are known the single-locus version of this experiment has no comparator.
  3. Did the authors already run this overlay? A paper reporting both a genome-wide CTCF peak analysis and a copy-number analysis in the same cells was in a position to ask whether lost peaks predict amplification. If they asked and found nothing, that is a published negative this graph should cite rather than an experiment it should propose.
Every source behind this entry (2)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    Across 12,443 CTCF anchors called from Gray 2023's own RPE siCT control CTCF tracks, 1,436 fall to 0.5x or below and 280 to 0.4x or below on KDM3B depletion, with the KMT2A exon-11 anchor at rank 258 and a ratio of 0.396x; the coordinates, baseline strengths and ratios of all 12,443 are persisted, and the anchor at hg19 chr11:8,241,000 (GRCh38 chr11:8,370,285) has baseline strength 4.95x against KMT2A's 4.92x and a ratio of 0.281x with both siKDM3B replicates below both controls - a near-exact strength match on the same chromosome losing more CTCF than KMT2A.

    GEO GSE210480 - four unique RPE CTCF ChIP bigWigs read genome-wide at 200 bp with pybigtools 0.3.0, 2026-09-22; anchor table persisted at analyses/q111-ctcf-anchor-matched-ranking/anchors.tsv with pre-registration in design.py · 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) 2023
    PMID 37788669DOI
    What this evidence cannot show

    'Anchor' means a local maximum of a deposited ratio track above a fixed fold cut with a both-control-replicates filter - no input model, no FDR, and the series' Input samples have no deposited file. n = 2 versus n = 2. The chr11:8,241,000 comparator has not been annotated: what gene or element it sits in was not looked up, and it could be an artefact region, a segmental duplication or a blacklist site, none of which was excluded because no blacklist was applied. That lookup is the first step of this experiment, not a result of the ranking.

    Evidence that argues against this (1)
    • The comparison rests on anchors called without a blacklist. ENCODE's hg19 blacklist regions produce spurious high-signal peaks in exactly the shape a fold-over-control track would show, and an unannotated strong anchor with a large apparent CTCF change is precisely what a blacklist region looks like. Until chr11:8,241,000 and the other top-ranked anchors are checked against a blacklist and against segmental-duplication tracks, the tail of this ranking could be partly artefact - which would not rescue the specificity claim (the matched MEDIAN of 0.633 does not depend on the tail) but would undercut the single-locus version of this experiment. Iqbal Lab oncofusion knowledge graph 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    PMID 37788669 reports that KDM3B depletion increases H3K9me1/2 and reduces CTCF occupancy at the MLL/KMT2A locus and thereby promotes amplification and rearrangements, that depleting CTCF is by itself sufficient to generate these focal alterations, and that CTCF overexpression rescues the doxorubicin-induced events - so the paper's own causal chain makes CTCF occupancy the operative variable, and it also reports 17,077 CTCF peaks losing intensity genome-wide on KDM3B depletion.

    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

    The causal chain is taken from the abstract; the 17,077 figure is taken from this graph's existing record of the paper's figures rather than from a fresh reading of the figure itself, and should be re-confirmed against the paper before it is used as the comparator in the overlay. Whether the paper reports amplified regions as genomic INTERVALS - which the overlay requires - or only as gene-level calls is not known and is the first thing to check; if it is the latter, this experiment cannot be run at anchor resolution.

    Evidence that argues against this (1)
    • The paper may already contain the answer in a form that makes this experiment unnecessary or already-decided. It reports a genome-wide CTCF peak analysis and a copy-number analysis in the same cells, so the authors were in a position to ask whether lost CTCF peaks predict amplification, and the fact that the paper frames the result as locus-specific suggests either that they asked and found no genome-wide correspondence, or that they did not ask. Reading the relevant figures and supplements before running anything is therefore part of the design, and it might convert this from an experiment into a citation. Gray ZH et al. 2023
Connected to