Oncofusion

start / graph / ana-q108-h3k9me3-at-the-kmt2a-bcr-measured

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q108-h3k9me3-at-the-bcr

RUN: the eight unreported H3K9me3 samples of GSE210480, read at the KMT2A CTCF element - H3K9me3 does not move where H3K9me1 moves 3.9-fold, and the me1/me2 positive controls passed

also called the GSE210480 H3K9me3 re-analysis, executed · the missing arm, measured · H3K9me3 at the KMT2A exon-11 CTCF site in RPE

A 2023 paper measured a third silencing tag on the KMT2A gene and never showed the result; the files have sat in a public archive since. We read them. The tag the paper showed jumps almost four-fold at the key spot; the tag it hid does not move there at all.

The same thing, for a scientist

The eight deposited-but-unreported H3K9me3 bigWigs of GSE210480 (PMID 37788669, human RPE) were read by HTTP range request over the *KMT2A* CTCF element, with the matched H3K9me1 and H3K9me2 arms in the same cells as internal positive controls and a declared readout written to disk first [E1, E2]. **The positive controls passed all four pre-registered tests.** H3K9me1 in control cells is depleted over the breakpoint cluster region relative to its 20 kb flanks (0.54x versus 1.28x the track's chr11 mean); siKDM3B raises it **2.67x** over the BCR, **3.74x** over the 1,048 bp CTCF/RAD21/SMC3 element and **3.85x** over canonical exon 11, with both replicates of siKDM3B above both replicates of control; siG9A alone does not raise it (0.90x); and the double knockdown lands between the two (2.09x at the element) [E2]. In a chr11 null binned at the BCR's own width, the BCR is rank **72 of 18,082** evaluable bins for the siKDM3B/control H3K9me1 log ratio - **top 0.40%** - which is the quantitative form of the paper's claim that the locus was one of the strongest events in the genome [E2]. **Against that validated pipeline, H3K9me3 does not move at the element: siKDM3B/control = 1.01, siG9A/control = 0.98, double/control = 1.19, and every arm's value lies inside the control replicate spread (0.127-0.235x).** In log terms the H3K9me3 response at the element is **1.1%** of the H3K9me1 response. Over the wider 7,466 bp BCR and at the DCAL cut point H3K9me3 does rise modestly and reproducibly (1.34x and 1.55x, both replicates), but it rises **equally in the siG9A arm** (1.38x, 1.74x), which is the wrong shape for a KDM3B-versus-G9a balance. **Everything here is retinal pigment epithelium.**

Go deeper
The measurement, with its controls beside it

All values are the interval's mean divided by the same track's own chr11 mean. The deposited bigWigs are already deepTools ChIP-over-Input ratio tracks, so no separate input division is possible or needed - and the Input samples in this series carry no supplementary file at all 1.

Arm / control fold ratios at each interval (mean of two replicates; OPP marks the KDM3B-versus-G9a balance signature, i.e. the two single knockdowns pushing opposite ways):

interval (GRCh38)H3K9me1 3B/G9A/dblH3K9me2 3B/G9A/dblH3K9me3 3B/G9A/dbl|me3|/|me1| in log2
CTCF element, chr11:118,488,159-118,489,2063.74 / 0.90 / 2.09 OPP0.44 / 0.85 / 0.661.01 / 0.98 / 1.190.011
CTCF summit +/-250, chr11:118,488,444-118,488,9443.85 / 0.91 / 2.17 OPP0.44 / 0.86 / 0.661.04 / 1.01 / 1.220.031
canonical exon 11, chr11:118,488,614-118,488,7603.85 / 0.91 / 2.17 OPP0.44 / 0.86 / 0.661.04 / 1.01 / 1.220.031
DCAL cut point +/-250, chr11:118,484,612-118,485,1122.37 / 0.84 / 1.09 OPP0.62 / 0.84 / 0.461.55 / 1.74 / 1.060.509
BCR core (A+B+C), chr11:118,482,422-118,489,8872.67 / 0.92 / 1.48 OPP0.69 / 0.89 / 0.581.34 / 1.38 / 1.100.300
20 kb centromeric flank1.01 / 1.12 / 1.151.26 / 0.87 / 0.91 OPP1.01 / 0.82 / 0.841.361
20 kb telomeric flank1.85 / 0.76 / 1.47 OPP1.06 / 1.02 / 0.931.23 / 0.89 / 0.920.334

Per-replicate, for the three claims that matter. H3K9me1 at the element: control [0.148, 0.474], siKDM3B [1.240, 1.085] - both siKDM3B replicates above both controls, no overlap. H3K9me3 at the element: control [0.235, 0.127], siKDM3B [0.177, 0.190], siG9A [0.179, 0.175], double [0.277, 0.152] - every arm falls inside the control range, and the control-to-control spread (1.85x) is larger than any arm difference. H3K9me3 at the cut point: control [0.249, 0.152], siKDM3B [0.362, 0.260], siG9A [0.342, 0.357] - both replicates of both single knockdowns above both controls, so the modest rise there is replicate-consistent even though its shape is not the balance's.

The chr11 null, binned at the BCR's own 7,466 bp width, 18,082 evaluable bins:

markBCR log2(siKDM3B/control)rankpercentile
H3K9me1+1.16072top 0.40%
H3K9me3+0.3973,877top 21.4%
H3K9me2-0.03310,276top 56.8%

Corrected for the chromosome-wide shift: chr11's own median log2 ratio is +0.114 for H3K9me3 and +0.004 for H3K9me1, so the locus-specific me3 change over the BCR is 1.22x, not 1.32x, while H3K9me1's is essentially unaffected at 2.23x.

Mark-identity control on the tracks themselves. In control cells the H3K9me3 tracks read 3.76x chr11 mean over the ZNF274 body - the canonical SETDB1/H3K9me3 domain - against 0.31x at the GAPDH promoter, a 12.2x ratio; the H3K9me1 and H3K9me2 tracks are not enriched at ZNF274 (0.38x and 0.78x). The me3 files are measuring H3K9me3. NaN fraction at the element is 0.0000 in all 24 histone samples, so no interval mean here is inflated by uncovered bases.

What the numbers do and do not license

They license this. At the element whose CTCF occupancy is the mechanism's terminal step, and at the 147 bp of canonical exon 11 inside it, the KDM3B-G9a axis moves H3K9me1 by nearly four-fold with the balance's own signature - up on siKDM3B, unmoved on siG9A, reset by the double - and moves H3K9me3 by 1% of that in log terms, with arm values indistinguishable from replicate noise. On the pre-registered reading R1, that supports mark identity as the resolution of this graph's H3K9 conflict, at this element, in this cell type.

They do not license this. (a) A 1 kb floor: every histone track in this series is 1,000 bp binned (measured by run-length encoding, see absence-gse210480-processed-tracks-are-1-kb-binned), so the 147 bp exon-11 query and the 501 bp summit query return the same value and neither is independent of its neighbours. A sub-kilobase H3K9me3 statement is not available from the deposited files at any price. (b) No spike-in, so only within-track relative comparisons are interpretable. (c) n = 2, with a control-to-control spread at the element of 1.85x for me3 and 3.2x for me1; effect sizes only, no p-values. (d) RPE. Retinal pigment epithelium, not a blood cell and not a KMT2A-rearranged cell.

And they refuse a tidy conclusion in one place. Over the wider BCR and at the DCAL cut point, H3K9me3 does rise ~1.3-1.6x, replicate-consistently. It would be easy to call that noise because the tidy answer is a flat me3. It is not noise; it is the wrong shape. A mark under the KDM3B-G9a balance must move in opposite directions in the two single knockdowns. H3K9me1 does exactly that at five of seven intervals. H3K9me3 at the cut point moves the same way in both (1.55 and 1.74) and returns to control in the double. Whatever is producing that, it is not the balance this mechanism describes - and it is recorded here rather than rounded away.

Two defects in my own pre-registration, named

(1) The magnitude clause was declared and not coded. design.py states reading R2 as me3 moving "in the SAME direction AND comparable magnitude as me1". run.py's classifier tested direction and a 1.25x flatness band and never tested comparability, so it printed R2 - me3 moves with me1; mark identity refuted on the same line as the numbers that refute R2 as written (me3 1.34x against me1 2.68x at the BCR). The fix is in followups.py, labelled post-hoc, and the adjudication above follows the pre-registered text, not the buggy code. This is the class of error this project has hit repeatedly: the last line is the one that gets quoted, and here the last line was wrong.

(2) The balance-signature test was not pre-registered at all. It should have been - it is the single test that distinguishes 'a mark that moves' from 'a mark the mechanism controls'. It is in followups.py, flagged post-hoc, and it is the reason the wider-BCR me3 rise is reported as un-shaped rather than as support for R2.

(3) The S1 tolerance was set in ignorance of the bin size. design.py allowed +/-150 bp for calling Gray's CTCF summit the same object as the measured summit. The CTCF tracks are 200 bp binned, so the tolerance was narrower than one bin and run.py printed S3 for a result that is S1 at the resolution the files support. See genome-kmt2a-exon-11-ctcf-summit-rpe-gray.

Facts and figures
scriptanalyses/q108-gse210480-h3k9me3-kmt2a-bcr/run.py
statusrun
run date2026-09-22
run byQ108-h3k9me3-at-the-bcr
normalisationinterval mean (NaN filled to 0) divided by the same track's mean over the TRUE hg19 chr11 length, excluding the 100 kb HOMER pad
bcr definition usedDCAL subregions A+B+C = Ensembl canonical exon 8 start to canonical exon 12 end = GRCh38 chr11:118,482,422-118,489,887, 7,466 bp. Chosen because it is the only BCR definition in this graph that reproduces Meyer's published subregion lengths to the base (1,761 / 679 / 5,026); all three identities re-derived from ENST00000534358 in lift.py.
pre registered reading reachedR1 at the CTCF element, canonical exon 11 and the summit: H3K9me3 is flat while H3K9me1 and H3K9me2 behave as published, which supports mark identity as the resolution of this graph's H3K9 conflict AT THIS ELEMENT IN THIS CELL TYPE. NOT R2 over the wider BCR, because the rise there fails both the magnitude clause of R2 as written (30-50% of me1's log response) and the balance signature (both single knockdowns push the same way).
localizationchromatin at the KMT2A locus, chr11q23.3; the measurement is of histone modification density over defined genomic intervals and says nothing about protein localisation
subnuclear localizationnot measured - a ChIP coverage track has no subnuclear coordinate. The element is a CTCF/cohesin loop anchor, which in other systems implies a TAD-boundary position, but that was not assayed here
cell typehTERT-immortalised human retinal pigment epithelial (RPE) cells - epithelial, non-haematopoietic, karyotypically near-diploid, no KMT2A rearrangement
cell stateasynchronous proliferating culture 72 h after siRNA transfection; no differentiation, drug or damage arm in the samples read here
epigenetic readoutH3K9me1, H3K9me2 and H3K9me3 ChIP-over-input coverage, plus CTCF ChIP, at 1 kb (histone) and 200 bp (CTCF) resolution over the KMT2A breakpoint region
shown for fusionNot shown for any fusion. Every value here is from unrearranged RPE cells; no KMT2A::AFF1, KMT2A::MLLT3 or other fusion-bearing cell was assayed, and the locus measured is the intact KMT2A gene before any rearrangement exists.
absence classmeasured-not-analysed
closesabsence-h3k9me3-at-the-kmt2a-locus-was-assayed-and-never-reported (partially: the RPE question is answered; the haematopoietic question is not)
chromosome wide shift correctionchromosome 11's MEDIAN siKDM3B/control log2 ratio is +0.114 for H3K9me3 against only +0.004 for H3K9me1, so about a quarter of the BCR's apparent me3 log-rise is a chromosome-wide shift with nothing to do with the locus. Median-corrected, the locus-specific me3 change over the BCR is 1.22x against H3K9me1's 2.23x. See ana-q108-chr11-rank-of-the-kmt2a-bcr-h3k9-response.
absence class noteMIGRATED 2026-09-23, mechanically and not by judgement: the class string's own first word is this token. The field previously held the whole verdict as free text and that string is preserved here verbatim because it says more than the token can: 'measured-not-analysed - CLOSED for H3K9me3 at this element in this cell type'
inputs
[
  "GSE210480 - 8 H3K9me3, 8 H3K9me1, 8 H3K9me2 bigWigs in four arms (control, siKDM3B, siG9A, siKDM3B+siG9A), 4 unique CTCF bigWigs, plus 8 exp1 me1/me2 tracks; Homo sapiens, RPE; hg19",
  "coords.json - nine KMT2A intervals in GRCh38 with a per-interval Ensembl lift to GRCh37, plus seven control loci looked up directly in GRCh37"
]
supporting files
[
  "analyses/q108-gse210480-h3k9me3-kmt2a-bcr/design.py",
  "analyses/q108-gse210480-h3k9me3-kmt2a-bcr/lift.py",
  "analyses/q108-gse210480-h3k9me3-kmt2a-bcr/followups.py",
  "analyses/q108-gse210480-h3k9me3-kmt2a-bcr/coords.json",
  "analyses/q108-gse210480-h3k9me3-kmt2a-bcr/results.json",
  "analyses/q108-gse210480-h3k9me3-kmt2a-bcr/results_followups.json"
]
acceptance test result
{
  "A1_control_me1_lower_at_bcr_than_flanks": "PASS - 0.539x versus 1.277x",
  "A2_siKDM3B_raises_me1_at_bcr": "PASS - 2.675x, both replicates above both controls",
  "A3_double_resets_me1_at_bcr": "PASS - 1.475x, between control and siKDM3B in both replicates",
  "A4_siKDM3B_raises_me2_on_the_flank": "PASS but WEAK - 1.26x centromeric, replicate ranges overlap (1.145 vs 1.011, 1.075 vs 0.751)",
  "all_pass": true
}
headline numbers
{
  "h3k9me1_siKDM3B_over_control_at_the_ctcf_element": 3.738,
  "h3k9me1_siKDM3B_over_control_at_canonical_exon_11": 3.852,
  "h3k9me1_siKDM3B_over_control_at_the_bcr_core": 2.675,
  "h3k9me3_siKDM3B_over_control_at_the_ctcf_element": 1.014,
  "h3k9me3_siG9A_over_control_at_the_ctcf_element": 0.978,
  "h3k9me3_double_over_control_at_the_ctcf_element": 1.185,
  "h3k9me3_log_response_as_fraction_of_h3k9me1_at_the_element": 0.011,
  "h3k9me3_siKDM3B_over_control_at_the_bcr_core": 1.344,
  "h3k9me3_siG9A_over_control_at_the_bcr_core": 1.377,
  "h3k9me3_siKDM3B_over_control_at_the_dcal_cut_point": 1.551,
  "h3k9me3_siG9A_over_control_at_the_dcal_cut_point": 1.744,
  "h3k9me2_siKDM3B_over_control_at_the_ctcf_element": 0.44,
  "h3k9me2_siKDM3B_over_control_on_the_centromeric_flank": 1.26,
  "bcr_rank_for_me1_among_18082_chr11_bins": 72,
  "bcr_rank_for_me3_among_18082_chr11_bins": 3877,
  "bcr_rank_for_me2_among_18082_chr11_bins": 10276,
  "me3_track_identity_control_znf274_over_gapdh_promoter": 12.168,
  "nan_fraction_at_the_element_all_24_histone_samples": 0,
  "chr11_median_log2fc_sikdm3b_over_control": {
    "H3K9me1": 0.0044,
    "H3K9me2": 0.0212,
    "H3K9me3": 0.1138
  }
}
effective resolution bp
{
  "histone_tracks": 1000,
  "ctcf_tracks": 200,
  "consequence": "the 147 bp exon-11 interval and the 501 bp summit interval return identical values in every histone track; no sub-kilobase histone statement is available from the deposited files"
}
hard limits
[
  "no spike-in normalisation anywhere in the series, so only within-track relative comparisons are interpretable",
  "n = 2 per arm; control-to-control spread at the element is 1.85x for me3 and 3.2x for me1",
  "the deposited tracks are input-ratio derivatives, so read depth and FRiP are not recoverable; NaN fraction (0.0 at the element) is the only coverage proxy available",
  "RPE - retinal pigment epithelium, not haematopoietic and not KMT2A-rearranged"
]
How this could bite us back
  • This entry supplies the reassurance that a G9a inhibitor given prophylactically would not touch H3K9me3 - and it supplies it from retinal pigment epithelium. The warning in `mark-h3k9me3` lives at fusion-target promoters in a leukaemic blast, which is a different mark location, a different cell and a different disease stage. A flat me3 here must not be quoted as a safety finding there.
  • The pipeline was validated on H3K9me1 and it passed emphatically, which makes it tempting to treat every number it produced as equally solid. The me3 numbers rest on a 1 kb bin, an n of 2, and a control replicate spread larger than any arm difference. 'Flat' here means 'no effect detectable at this resolution with this power', not 'zero'.
  • The 1.3-1.7x me3 rise over the wider BCR is the part of this result that could be right and inconvenient. If a better-powered experiment showed it is real and mechanistically connected, the mark-identity resolution weakens at the scale of the whole breakpoint region even while it holds at the element - and the BCR, not the element, is where breaks actually happen.
Still open
  1. Does H3K9me3 at the KMT2A CTCF element behave the same way in a haematopoietic cell? Nothing in this measurement is in blood, and the entry that motivated it warns explicitly that a null in epithelium licenses no statement about a leukaemic blast. This is now the only version of the question still open, and it needs new wet work.
  2. What produces the replicate-consistent 1.3-1.7x H3K9me3 rise over the wider BCR and at the DCAL cut point in BOTH single knockdowns but not the double? It is not the KDM3B-G9a balance, whose signature is opposite single-arm effects. Candidates: a shared siRNA-transfection or proliferation response, SETDB1 redistribution, or genuine convergence. All three are testable with the same deposited files plus a scrambled-siRNA-only arm, which this series does not contain.
  3. Why does H3K9me2 FALL at the CTCF element (0.44x, both replicates) when the paper's abstract says KDM3B depletion increases H3K9me1/2 at the locus? If the deposited me2 tracks are the ones behind the figures, then 'at the locus' must mean the flank - and the mono-to-di ratio at the element inverts on KDM3B loss, which no entry in this graph currently records.
  4. Would the raw reads in SRA support a 200 bp readout at exon 11? The deposited processed tracks are 1 kb binned, which is the resolution floor for every histone statement here. The CTCF tracks are 200 bp, so the raw data plainly permit finer binning than was deposited for the histones.
Every source behind this entry (3)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    GSE210480's per-sample supplementary files are deepTools ChIP-over-Input ratio coverage bigWigs aligned to hg19 with BWA; the Input samples carry no supplementary file; the eight H3K9me3 samples are GSM6430988 and GSM6430990 (control), GSM6431006 and GSM6431010 (siKDM3B), GSM6430992 and GSM6430994 (siG9A) and GSM6431004 and GSM6431008 (double), matched by H3K9me1 and H3K9me2 in the same four arms; and the bigWigs declare hg19 chromosome lengths plus exactly 100,000 bp on chr1, chr2, chr11 and chrX.

    What this evidence cannot show

    The processing description is the submitter's and was not independently verified against the reads; what WAS verified directly from the files is that every track's chr11 mean is close to 1.00 (0.99-1.15 across 40 tracks), which is what an input-ratio track looks like and not what a raw coverage track looks like. The build was verified from the chrom table rather than from GEO's genome_build field, because in a sibling series three human HCT116 samples carry !Sample_organism_ch1 = Mus musculus. GEO's organism field for GSE210480 does read Homo sapiens, correctly, but it was not used as evidence.

    Evidence that argues against this (1)
    • A ratio track is a processed derivative and compresses dynamic range: CTCF over this element reads 4.7x the chr11 mean in Gray's ratio tracks against 50x in Canela's raw-coverage HCT116 tracks for the same element. The magnitudes in this entry are therefore NOT comparable with any fold-enrichment computed from raw coverage, only with each other. Anyone comparing this entry's numbers with Q107's HCT116 numbers is comparing two different normalisations. Canela A et al. 2019
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    Read directly from the deposited bigWigs: H3K9me1 rises 3.74x at the KMT2A CTCF element and 2.67x over the BCR on KDM3B depletion with both replicates separated from control, is unchanged on G9a depletion (0.90x) and is partly reset by the double knockdown (2.09x), and the BCR ranks 72 of 18,082 chr11 bins of its own width for that log ratio; while H3K9me3 in the same cells and the same four arms reads 1.01x, 0.98x and 1.19x at the same element with every arm inside the control replicate range, its log response being 1.1% of H3K9me1's; H3K9me2 FALLS at the element (0.44x) while rising on the centromeric flank (1.26x); and the H3K9me3 tracks read 3.76x chr11 mean over the ZNF274 body against 0.31x at the GAPDH promoter, confirming they measure H3K9me3.

    GSE210480 per-sample bigWigs read at interval resolution over HTTP range requests with pybigtools 0.3.0, 2026-09-22; scripts, pre-registration and full output persisted at analyses/q108-gse210480-h3k9me3-kmt2a-bcr/ (design.py, lift.py, run.py, followups.py, coords.json, results.json, results_followups.json) · 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

    n = 2 per arm, no spike-in, no per-window significance test attempted. Every histone track is 1,000 bp binned, so the 147 bp exon-11 and 501 bp summit intervals return identical values and the effective resolution of every histone statement here is 1-2 kb. Values are fold over each track's own chr11 mean, a within-track denominator chosen because no absolute scale exists; the chr11 mean itself shifts between arms (1.057-1.145 for H3K9me1), so a rise measured this way is a rise RELATIVE to chr11, which is the only thing these files can support. RPE cells, epithelial, not KMT2A-rearranged. The classifier in run.py mis-stated its own pre-registered reading (see the section on defects); the adjudication above is by the pre-registered text.

    Evidence that argues against this (1)
    • The paper's own abstract states that KDM3B depletion 'increases H3K9me1/2 levels ... at the MLL/KMT2A locus' - me1 AND me2 together. This measurement reproduces that for H3K9me1 emphatically and CONTRADICTS it for H3K9me2 at the element itself, where me2 falls to 0.44x with both replicates below both controls; the me2 rise is confined to the centromeric flank (1.26x, and there the replicate ranges overlap). Either the deposited tracks are not what the figures were made from, or 'at the locus' in the abstract means the flank for me2 - which is what the paper's figures actually show and what this graph's mechanism entry already records. The discrepancy is against this entry's method and is stated rather than resolved. Gray ZH et al. 2023
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    The GRCh38-to-GRCh37 offset at this locus is exactly 129,285 bp and was re-derived per interval across nine intervals spanning 1,048 bp to the 90,341 bp KMT2A gene, each Ensembl mapping filtered on seq_region_name == '11' to exclude the HG122_PATCH mapping that returns identical coordinates; and the three base-pair identities that fix the DCAL exon frame as Ensembl canonical + 1 were re-derived from ENST00000534358 rather than quoted.

    What this evidence cannot show

    A constant offset across 90 kb of chr11q23 does not license the same arithmetic elsewhere on chr11 or on any other chromosome. One discrepancy is worth recording: lifting the GRCh38 KMT2A span gives GRCh37 chr11:118,307,207-118,397,547, while looking the gene up directly in the GRCh37 archive gives chr11:118,307,205-118,397,539. The 2 bp and 8 bp differences are ANNOTATION differences between the two gene builds, not assembly differences - which is why a lift is per-interval and why a gene-length span should be looked up, not lifted.

    Evidence that argues against this (1)
    • A constant offset is necessary but not sufficient for a correct lift: if both builds contained the same error at this locus the offset would still be constant. The independent check that the coordinates land on the right object is biological - CTCF reads 7.0x chr11 mean at the lifted exon-11 interval and 1.3x at the lifted cut point 3.8 kb away, which is the expected contrast and would not survive a 20 kb displacement. Ensembl REST API; GRCh37 archive 2026
Connected to