RUN: KDM4B and KDM4C both read the 76.6th-79.7th percentile of MCF-7 CTCF anchors over the whole 90 kb KMT2A gene body - above the 25-track panel median of 61.39, LOCUS-WIDE ELEVATED - and both are ORDINARY at the exon-11 anchor with replicate spreads of 92.3 and 61.5 percentile points
also called the erasers' occupancy · KDM4B and KDM4C at the KMT2A locus · SUV39H1's signature, in two more proteins
The two enzymes that remove the silencing tag are spread across the whole KMT2A gene at a slightly raised level, and neither is concentrated at the spot where the gene breaks. Several unrelated proteins now show the same pattern, which usually means the pattern is about the region rather than the proteins.
The same thing, for a scientist
**The occupancy half of the eraser question, and it is the half that needed the panel.** Four KDM4B and four KDM4C ChIP-seq tracks in MCF-7, ±oestradiol, all C1-passing (8.2x-40.5x dynamic range over a printed non-zero denominator) [E1]. | | exon11 (146 bp) | median | spread | kmt2a_gene (90 kb) | median | promoter (2 kb) | |---|---|---:|---:|---|---:|---| | **KDM4B** | 84.78 / 0.00 / 92.26 / 0.00 | **42.39** | **92.26** | 81.32 / 69.20 / 83.38 / 78.06 | **79.69** | 62.28-99.82 | | **KDM4C** | 86.06 / 35.31 / 56.28 / 24.53 | **45.79** | **61.53** | 78.52 / 74.82 / 78.42 / 72.72 | **76.62** | 97.71-99.37 | **Verdict, by the pre-registered rule: ORDINARY at the anchor for both** (medians 42.39 and 45.79, inside [25, 75]; KDM4B has 2 of 4 tracks above 75, below the 60% clause, and KDM4C 1 of 4). **And LOCUS-WIDE ELEVATED for both**: the 90 kb gene-body medians of **79.69** and **76.62** exceed both the 75th percentile and the **61.39** panel median across 25 C1-passing tracks [E1]. **This is SUV39H1's signature, and that is the finding.** Q114 measured SUV39H1 at the **81.65th-86.79th** percentile over the same 90 kb in all four of its MCF-7 tracks and could not say whether it was the protein or the locus. **A writer (SUV39H1), two erasers (KDM4B, KDM4C), a reader's tagged mutants and a scaffold (TRIM28) now all read this 90 kb up in the same cell line, with six-non-anchor-interval medians of 49.79 to 72.37.** Five to eight unrelated antigens agreeing is a property of the region, its accessibility, its copy number or the input-free normalisation - not of five unrelated proteins [E2]. **The replicate spreads make the anchor numbers unusable on their own.** KDM4B reads **92.26 and 0.00** at the same 146 bp window in two tracks of the same antibody, and **84.78 and 0.00** in the two -E2 replicates. In the same four files the 90 kb gene body spans **69.20 to 83.38** - a 14-point range. **The wide interval is the one that can be reported; the narrow one is not** [E1]. **And the biological reading, such as it is.** An eraser sitting over a domain that is already poor in its substrate is not obviously doing anything - H3K9me3 over the same 90 kb is at the **2.56th-15.00th** percentile in the same cell line. Either the erasers keep it that way, or they are there for another reason, or they are not specifically there at all. **The knockdown decides it and says they do not keep it that way**: removing both leaves the gene body relatively poorer, not richer. **AND THE PROMOTER NUMBERS COLLAPSE AGAINST THE RIGHT NULL, which is this entry's most useful result.** Against 56,171 CTCF summit windows both erasers look strongly promoter-bound at *KMT2A*: **KDM4B 62.28-99.82** and **KDM4C 97.71-99.37**, with per-antigen medians of **99.37** and **98.47** - above the 96.87 panel median. Against the promoters of the **2,000 highest-expressed MCF-7 genes**, the same -E2 tracks read **KDM4B 58.60** (1.05x the median active promoter, i.e. **background**) and **KDM4C 78.35** (1.61x). **KDM4B's promoter signal is entirely explained by the promoter being active**, and that makes it the case that validates the null the module entry rests on: TRIM28 98.8-99.4, CBX5 91.1-94.5, EHMT2 87.6-88.7, KDM4C 78.4, KDM4B 58.6. **Neither eraser belongs to the promoter module.**
Oestradiol, and a protocol split that is not oestradiol
The depositors' own result is that KDM4B and KDM4C are recruited to oestrogen-responsive promoters after 40 min of E2 3. At KMT2A:
| -E2 | +E2 | |
|---|---|---|
| KDM4B exon11 | 84.78, 92.26 | 0.00, 0.00 |
| KDM4B kmt2a_gene | 81.32, 83.38 | 69.20, 78.06 |
| KDM4C exon11 | 86.06, 56.28 | 35.31, 24.53 |
| KDM4C kmt2a_gene | 78.52, 78.42 | 74.82, 72.72 |
Both antigens read LOWER at the anchor and at the gene body after oestradiol, in every comparison. That is the opposite of the published direction at ER target promoters, and it would be a result if the anchor numbers were reliable - which the 92-point replicate spread says they are not. At the 90 kb scale, where the spreads are 2-14 points, the direction survives: four of four -E2 tracks read higher than their +E2 partners.
I am not calling that a redistribution. KMT2A is not an ER target, the effect is 5-14 percentile points, and Q114 measured this deposit's across-condition correlation exceeding its within-condition correlation for SUV39H1 - the signature that says a ±E2 contrast in this deposit cannot be separated from batch.
| script | analyses/q116-h3k9-axis-at-the-locus/run116.py and score.py |
|---|---|
| prereg | analyses/q116-h3k9-axis-at-the-locus/PREREGISTRATION.md, section 8 |
| status | run |
| run date | 2026-09-23 |
| run by | Q116-h3k9-axis-at-the-locus |
| complete run | true |
| panel median kmt2a gene percentile | 61.39 |
| n panel tracks | 27 |
| the shared signature | SUV39H1 81.65-86.79 (Q114), KDM4B 69.20-83.38, KDM4C 72.72-78.52, TRIM28 49.90-75.12, EHMT2 60.92-70.10 over the same 90 kb in the same cell line |
| oestradiol | both antigens read LOWER at the anchor and at the gene body after 40 min E2 in every comparison (4 of 4 pairs at the gene body), which is the opposite of the published direction at ER target promoters. Not called a redistribution: the effect is 5-14 percentile points, KMT2A is not an ER target, and Q114 measured this deposit's across-condition correlation exceeding its within-condition correlation. |
| localization | chromatin at chr11q23.3 - the seven stored KMT2A intervals |
| subnuclear localization | not measured |
| cell type | MCF-7 (ER+ breast adenocarcinoma). NOT haematopoietic, NOT KMT2A-rearranged. |
| cell state | hormone-deprived culture and 40 min 17-beta-oestradiol |
| epigenetic readout | occupancy of the two KDM4 H3K9me3 demethylases at the KMT2A locus |
| eraser of that mark | KDM4B (O94953) and KDM4C - both measured here, and both removed in the knockdown arm |
| shown for fusion | NO - MCF-7 carries no KMT2A rearrangement. KDM4C has a KMT2A-fusion literature (recruited by MLL fusions, required in mouse models) and none of it is occupancy data at this locus. |
| absence class | measured-not-analysed |
| absence class note | Eight tracks deposited publicly in 2019 and never read at this locus. |
| panel median provenance | C5/C5b PANEL MEDIANS ARE QUOTED AT THE FINAL 28-TRACK COUNT (2026-09-23). Earlier drafts of this entry quoted them at intermediate counts (12, 18 or 25 tracks) while the run was still going. The gene-body panel median moved 61.80 -> 61.39; the PROMOTER panel median moved 82.36 -> 96.87, i.e. UP, and the direction matters: at 96.87 the CTCF-anchor percentile at the KMT2A promoter is even less informative than the first draft said, and the promoter claim rests ENTIRELY on the active-promoter null of deviation D9. |
results
[ "analyses/q116-h3k9-axis-at-the-locus/results.json", "analyses/q116-h3k9-axis-at-the-locus/results_scored.json" ]
KDM4B
{
"tracks": [
"GSM4133832 -E2",
"GSM4133834 +E2",
"GSM4133833 -E2",
"GSM4133835 +E2"
],
"exon11_percentiles": [
84.78,
0,
92.26,
0
],
"exon11_median": 42.39,
"exon11_spread": 92.26,
"kmt2a_gene_percentiles": [
81.32,
69.2,
83.38,
78.06
],
"kmt2a_gene_median": 79.69,
"kmt2a_gene_spread": 14.18,
"kmt2a_tss_2kb_percentiles": [
62.28,
99.11,
99.82,
99.64
],
"six_nonanchor_interval_median": 70.31,
"C1_ratios": [
8.6,
17.2,
8.2,
11.4
],
"verdict": "ORDINARY at the anchor; LOCUS-WIDE ELEVATED, ANCHOR NOT PICKED OUT"
}KDM4C
{
"tracks": [
"GSM4133836 -E2",
"GSM4133838 +E2",
"GSM4133837 -E2",
"GSM4133839 +E2"
],
"exon11_percentiles": [
86.06,
35.31,
56.28,
24.53
],
"exon11_median": 45.79,
"exon11_spread": 61.53,
"kmt2a_gene_percentiles": [
78.52,
74.82,
78.42,
72.72
],
"kmt2a_gene_median": 76.62,
"kmt2a_gene_spread": 5.8,
"kmt2a_tss_2kb_percentiles": [
97.71,
98.12,
98.81,
99.37
],
"six_nonanchor_interval_median": 67.79,
"C1_ratios": [
17.5,
40.5,
21.4,
21.6
],
"verdict": "ORDINARY at the anchor; LOCUS-WIDE ELEVATED, ANCHOR NOT PICKED OUT"
}C5 final 28 track values
{
"panel_median_kmt2a_gene_percentile": 61.39,
"panel_median_kmt2a_promoter_percentile": 96.87,
"per_antigen_median_promoter_percentile_vs_CTCF_anchors": {
"TRIM28": 99.8,
"CBX5_mut4E": 99.75,
"KDM4B": 99.37,
"KDM4C": 98.47,
"CBX5_mut4A": 96.76,
"CBX5": 96.65,
"EHMT2": 95.83,
"H3K9me3": 29.43
},
"reading": "every PROTEIN antigen reads 95.8-99.8 at the KMT2A promoter against CTCF anchors and only the histone mark (29.43) does not - so that statistic cannot discriminate between proteins at all, and every promoter claim in this beat rests on the ACTIVE-PROMOTER null instead"
}promoter against the active promoter null
{
"KDM4B_minusE2": {
"pct_of_2000_active_promoters": 58.6,
"x_median_active": 1.05,
"reading": "background"
},
"KDM4C_minusE2": {
"pct_of_2000_active_promoters": 78.35,
"x_median_active": 1.61,
"reading": "modest"
},
"for_comparison_vs_CTCF_anchors": {
"KDM4B_median": 99.37,
"KDM4C_median": 98.47
},
"verdict": "neither eraser belongs to the promoter module; KDM4B's apparent promoter occupancy is entirely explained by the promoter being active, and it is the discriminating case that validates the active-promoter null",
"not_run_for": "the +E2 tracks and the second replicates of either eraser"
}- gene-kdm4c records that KDM4C is recruited by MLL fusions and that the KDM4C inhibitor SD70 preferentially kills MLL-fusion lines including SEM. This entry places both KDM4 erasers over the KMT2A gene body in a non-leukaemic line at a level shared by five unrelated antigens, which is not evidence for a KMT2A-locus mechanism for that drug. A KDM4 inhibitor's therapeutic rationale in this disease should not be attached to occupancy at this locus on the strength of these numbers.
- Both erasers read above the 75th percentile over the 90 kb gene body and ORDINARY at the anchor with replicate spreads above 60 percentile points. Anyone quoting the single 92.26th-percentile KDM4B anchor value as occupancy would be quoting one of two tracks whose partner reads 0.00 at the same coordinates.
- Do the KDM4B and KDM4C antibodies cross-react? The two catalytic cores are closely related, no clone is stated for either, and the two antigens' agreement at this locus is one of the main reasons this entry calls the elevation locus-level. If they cross-react, that agreement is circular.
- Is the whole-locus elevation a sub-threshold copy gain? The segment-level log2 ratio at 11q23 in MCF-7 has never been retrieved by any beat, and a discrete copy-number caller reports 0 for exactly the size of gain that would produce a 1.1-1.3x elevation.
- Why do both erasers read LOWER at KMT2A after oestradiol when the depositors report them being recruited at ER target promoters? Either KMT2A is being vacated as those promoters are occupied - which would be a redistribution story - or the deposit's +/-E2 contrast is batch, which Q114's correlation structure suggests. Nothing here separates the two.
- Do the erasers sit over the gene body because there is nothing to erase there? H3K9me3 over the same 90 kb is at the 2.56th-15.00th percentile in the same cell line. An eraser on an empty substrate is either the cause of the emptiness or irrelevant to it, and the knockdown says it is not the cause.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
In four KDM4B and four KDM4C MCF-7 ChIP-seq tracks, the 90 kb KMT2A gene-body window reads the 81.32/69.20/83.38/78.06 and 78.52/74.82/78.42/72.72 percentile of width-matched same-cell-type CTCF summit windows (medians 79.69 and 76.62, both above the 61.39 panel median of 25 C1-passing tracks), while the 146 bp exon-11 window reads 84.78/0.00/92.26/0.00 and 86.06/35.31/56.28/24.53 (medians 42.39 and 45.79, ORDINARY by the pre-registered rule). All eight tracks pass C1 at 8.2x-40.5x over a printed non-zero denominator.
GSE139196: ChIP-seq on KDM4B (GSM4133832-35) and KDM4C (GSM4133836-39) in MCF-7 with and without estradiol, hg38 bedGraph · Yang F, Tanasa B, Micheletti R, Ohgi KA, Aggarwal AK, Rosenfeld MG · NCBI GEO 2019What this evidence cannot show
MCF-7, an ER+ breast adenocarcinoma, not KMT2A-rearranged and not haematopoietic. No input, no IgG, no antibody clone stated for either target - and KDM4B and KDM4C are 51% identical in their catalytic cores, so cross-reactivity between the two antibodies is a real and unexcluded possibility that would make the two antigens' agreement circular. The C4 instrument control was declared FAILED in advance for this cell type. Replicate spreads at the 146 bp window are 92.26 and 61.53 percentile points, so no single-track anchor number here means anything; the gene-body numbers, with spreads of 14.18 and 5.80, are the reportable ones.
Evidence that argues against this (1)
- A LOCUS-WIDE ELEVATED verdict is a statement about the locus, not about the protein. The 61.39 panel median it is measured against comes from 28 tracks of eight antigens IN THE SAME DEPOSIT, all of them H3K9-axis proteins; if the KMT2A locus carries a genuine shared H3K9 platform then this verdict mislabels real biology as an artefact. An input track resolves it and does not exist. See concept-q116-a-panel-of-many-antigens-in-one-cell-type-is-the-input-track-the-deposit-did-not-provide. source
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
In the same MCF-7 deposit, SUV39H1 reads 81.65-86.79 over the 90 kb KMT2A gene body (Q114), KDM4B 69.20-83.38, KDM4C 72.72-78.52, TRIM28 49.90-75.12, EHMT2 60.92-70.10, and the six-non-anchor-interval medians run 49.79-72.37 for every protein antigen against 15.41 for H3K9me3 - so the locus-wide elevation is shared across a writer, two erasers, a reader's tagged mutants and a scaffold.
analyses/q116-h3k9-axis-at-the-locus/results_scored.json (control C5) and ana-q114-in-mcf7-the-whole-kmt2a-locus-is-elevated-and-the-anchor-is-not-picked-out · Q116-h3k9-axis-at-the-locus and Q114-suv39h1-at-the-locus · oncofusion-knowledge-graph 2026What this evidence cannot show
Q114's SUV39H1 figures are quoted from its entry and were not recomputed here; this beat did not open the four SUV39H1 bedGraphs. All the other figures are from this run's results.json. One deposit, one pipeline, one cell line.
Evidence that argues against this (1)
- Copy number would produce exactly this pattern and has never been excluded at the resolution required. CCLE calls KMT2A copy-neutral in MCF-7, but Q114 recorded that a discrete caller reports 0 for precisely the size of segmental gain that would give a 1.1-1.3x signal elevation, and the segment-level log2 ratio at 11q23 has not been retrieved by any beat. Until it is, 'locus-level property' includes 'sub-threshold copy gain' as a live candidate. source
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The depositors report that KDM4B and KDM4C are recruited at increased levels to oestrogen-responsive promoters after 40 min of E2, in these same cells, and that the interaction requires a compact stem-loop in promoter antisense RNA.
Shape of promoter antisense RNAs regulates ligand-induced transcription activation · Yang F, Tanasa B, Micheletti R, Ohgi KA, Aggarwal AK, Rosenfeld MG · Nature 2021What this evidence cannot show
Full text read via PMC8439151 on 2026-09-23. The recruitment is reported for ER-responsive ('PR') promoters; KMT2A is not among them and the word KMT2A does not appear in the paper. So the published direction is not a prediction about this locus, and the fact that KMT2A moves the other way is not a contradiction of the paper.
Evidence that argues against this (1)
- If KDM4 recruitment in this system is driven by promoter antisense RNA structure, then occupancy at a gene that is not an ER target carries no information about the mechanism the paper describes, and reading these tracks at KMT2A is reading them outside the regime they were generated to probe. That applies to every number in this entry. source
first KDM4B occupancy measurement at the KMT2A locus in any cell type: 79.69th percentile median over the 90 kb, ORDINARY at the anchor with a 92.3-point replicate spread
first KDM4C occupancy measurement at the KMT2A locus: 76.62nd percentile median over the 90 kb, ORDINARY at the anchor
two more antigens with SUV39H1's locus-wide signature; the shared elevation now spans a writer, two erasers, a reader's mutants and a scaffold
the occupancy half; both erasers are over the gene body and removing both does not raise the mark there, so their presence is not what keeps it clear
the signature that entry found for SUV39H1 replicates in KDM4B (79.69) and KDM4C (76.62) against a 27-track panel median of 61.39