H3K9me1 depletion over the KMT2A breakpoint cluster replicates in a leukaemia line - as a broad-domain property of the 7.5 kb BCR, not of the 1 kb CTCF element, where the two tracks disagree in sign
also called K562 H3K9me1 KMT2A BCR · first link replicates in blood · ENCSR000AKW · broad-domain not element · no CD34+ H3K9me1 in ENCODE
The one part of this chain that is a property of a resting cell - that the mark is unusually scarce across the stretch of DNA where KMT2A breaks - holds up in a leukaemia cell as well as in the skin-type cell it was found in. But it holds for the whole stretch, not for the small element inside it, where the two available datasets point opposite ways.
The same thing, for a scientist
Q108 read the unreported H3K9me3 arm of GSE210480 and then named the cheapest unrun follow-up: ask whether the **resting-state H3K9me1 depletion** over the *KMT2A* BCR - the first link of the H3K9 brake chain, and the only link that is a property of an unperturbed cell - also holds in a haematopoietic cell. It said a negative would be the most important negative available in this territory. **It came back positive, with a specific limit.** **What is actually available, measured before designing around it.** ENCODE holds **four** H3K9me1 histone ChIP-seq experiments in total - IMR-90, keratinocyte, **K562**, and umbilical-vein endothelium - and **no CD34+ progenitor H3K9me1 anywhere**. H3K9me2 has 19 experiments and **none** in a blood cell. So the CD34+ experiment Q108 proposed cannot be done from ENCODE, and this runs the nearest substitute: **K562**, a CML-derived erythroleukaemia line. It is a blood cell and a leukaemia; it is **not** a CD34+ progenitor and **not** KMT2A-rearranged, and that substitution is this analysis's principal weakness. **It deposits GRCh38 fold-change-over-control tracks**, so nothing is lifted and that whole error class is removed. **THE RESULT, at the BCR scale, replicates.** Fold over each track's own chr11 mean, with the two 20 kb flanks as the local comparator: | | BCR core (7,466 bp) | mean flank | BCR/flank | |---|---:|---:|---:| | K562 rep a | 0.737 | 1.310 | **0.563** | | K562 rep b | 0.801 | 1.429 | **0.560** | | RPE control (Q108) | 0.539 | 1.277 | 0.422 | Both tracks agree in direction and land within 0.003 of each other. The depletion is shallower than RPE's but unmistakably present in a leukaemia cell. **AND IT DOES NOT HOLD AT THE ELEMENT.** At the 1,048 bp CTCF element the two tracks give **0.877** and **1.155**; at canonical exon 11 they give **0.836** and **1.056**. They disagree about whether the element is depleted at all, so **no element-level claim is available**. The DCAL cut point is 1.09 and 1.06 - not depleted in either. **That pattern is informative rather than merely disappointing**, and it agrees with Q108 from the other direction. Q108 found that H3K9me1 *rises* broadly on KDM3B depletion while CTCF falls at canonical exon 11 and nowhere else, and concluded the locus address sits on the **CTCF arrow, not the mark arrow**. Here the resting mark depletion is likewise **broad-domain** - a property of the 7.5 kb BCR - while the element itself carries no reproducible mark signature. Two independent datasets, two cell types, same conclusion: the mark is a regional property and the element-level specificity is CTCF's. **The bin check mattered and passed this time.** This project added a registry row this cycle for a binned track answering a finer query without error. Run-length encoding gives a **modal run of 1 bp** in both K562 tracks, so these are base-resolution and the 147 bp exon-11 interval is a real measurement - unlike GSE210480's 1 kb bins, where the same interval returns a single value.
| script | analyses/k562-h3k9me1-over-the-kmt2a-bcr/run.py |
|---|---|
| results | analyses/k562-h3k9me1-over-the-kmt2a-bcr/results.json |
| status | run |
| run date | 2026-09-22 |
| why this ran | Q108 named it as the cheapest unrun follow-up and did not take it: the resting-state depletion is the first link of the H3K9 brake chain and the only link that is a property of an unperturbed cell, so testing it in the right lineage is free and a negative would have been the most important negative available. |
| experiment | ENCSR000AKW, K562 H3K9me1, Bradley Bernstein lab, Broad; two fold-change-over-control bigWigs in GRCh38 (ENCFF271KWX, ENCFF654SLZ) |
| assembly | GRCh38 native - nothing lifted, which removes the 129,285 bp offset and the HG122_PATCH double-mapping as error sources entirely |
| absence class | measured-not-analysed |
what encode actually has
{
"H3K9me1 histone ChIP-seq experiments in all of ENCODE": 4,
"their biosamples": [
"IMR-90",
"keratinocyte",
"K562",
"endothelial cell of umbilical vein"
],
"CD34+ progenitor H3K9me1": "NONE - the proposed experiment cannot be done from ENCODE",
"H3K9me2 experiments": 19,
"haematopoietic among those": "NONE",
"so": "K562 is the only blood-cell H3K9me1 track in the archive"
}bin check ran first
{
"modal_run_bp": 1,
"median_run_bp": 1,
"verdict": "base-resolution tracks, so the 147 bp canonical exon-11 interval is a real measurement here",
"contrast": "GSE210480's H3K9 tracks are 1,000 bp binned, where the same 147 bp interval returns ONE distinct value and equals a 501 bp interval containing it",
"why_it_was_mandatory": "this project added a registry row this cycle for exactly this failure after paying for it twice"
}the bcr scale result
{
"rep_a": {
"bcr_core_fold": 0.7371,
"flank_centromeric": 1.424,
"flank_telomeric": 1.1963,
"mean_flank": 1.3102,
"bcr_over_flank": 0.5626
},
"rep_b": {
"bcr_core_fold": 0.801,
"flank_centromeric": 1.4383,
"flank_telomeric": 1.4207,
"mean_flank": 1.4295,
"bcr_over_flank": 0.5603
},
"rpe_control_equivalent": 0.4221,
"both_tracks_agree_on_direction": true,
"reading": "the resting depletion replicates in a leukaemia cell, shallower than RPE but replicate-consistent to within 0.003"
}the element scale result is NOT available
{
"ctcf_element_1048bp": {
"rep_a": 0.8767,
"rep_b": 1.1545
},
"canonical_exon_11_147bp": {
"rep_a": 0.836,
"rep_b": 1.056
},
"dcal_cut_point_plus_minus_250": {
"rep_a": 1.092,
"rep_b": 1.0611
},
"verdict": "the two tracks DISAGREE IN SIGN at the element and at the exon, so no element-level claim is available from this dataset at all",
"and_this_is_the_informative_part": "the depletion is a broad-domain property of the 7.5 kb BCR, not a property of the 1 kb element - which agrees with Q108's independent finding that the locus address sits on the CTCF arrow rather than the mark arrow"
}controls
{
"ZNF274_h3k9me3_site": {
"rep_a": 0.873,
"rep_b": 0.7818,
"reading": "an H3K9me3 site is NOT H3K9me1-enriched, as it should not be"
},
"GAPDH_promoter": {
"rep_a": 0.7269,
"rep_b": 0.8709
},
"ACTB_promoter": {
"rep_a": 1.1548,
"rep_b": 1.4292
},
"chr2_gene_desert": {
"rep_a": 1.2307,
"rep_b": 1.0731
},
"note": "the two flanks differ within rep a (1.424 centromeric vs 1.196 telomeric), so the flank normaliser has internal structure and the BCR/flank ratio is not a precision figure"
}KING CORRECTION - THE ZNF274 CONTROL IS MISLABELLED
{
"what_is_wrong": "analyses/k562-h3k9me1-over-the-kmt2a-bcr/run.py labels the interval chr19:57,996,000-58,000,000 as ZNF274_h3k9me3_positive, and every ZNF274 statement in this entry - including the ZNF274_h3k9me3_site row, rep_a 0.873 and rep_b 0.7818 - is about that window, not about ZNF274.",
"the_real_coordinates": {
"gene": "ZNF274, ENSG00000171606",
"GRCh38": "chr19:58,183,029-58,215,964",
"GRCh37": "chr19:58,694,396-58,724,928",
"distance_from_my_window_start_GRCh38_bp": 187029,
"distance_from_my_window_start_GRCh37_bp": 698396,
"source": "Ensembl REST lookup/symbol/homo_sapiens/ZNF274, both assemblies, re-resolved by the king 2026-09-22"
},
"it_is_wrong_in_BOTH_assemblies": "so this is not an hg19-versus-hg38 confusion - the window is simply not the gene, and no lift fixes it",
"WHY_THIS_IS_WORSE_THAN_A_WRONG_LABEL": "The window sits inside the chr19q13.43 zinc-finger cluster, which is constitutively H3K9me3-high. So the positive control PASSED - it just passed for the wrong reason. A control that succeeds on the wrong coordinate provides no assurance at all, and this project's standing lesson is to check what a checker walks rather than what it reports. Q110 found it because its own inherited copy of the control FAILED a peak test, and checking the coordinate instead of declaring instrument failure turned the failure into this defect report.",
"what_survives": "The heterochromatin reading probably survives, because a ZNF-cluster window is a legitimate H3K9me3-high region and the ratios measured there are real measurements of a real heterochromatic interval. What does NOT survive is the claim that the control was anchored on ZNF274. Read the ZNF274_h3k9me3_site row as 'a chr19q13.43 ZNF-cluster window'.",
"what_is_unaffected": "Q108 is clean: its lift.py resolves ZNF274 by symbol rather than by a hardcoded window, so its 3.76x ZNF274-body figure is about the real gene. The KMT2A-locus measurements in this entry are unaffected - the defect is confined to the control interval.",
"the_script_is_deliberately_not_edited": "analyses/k562-h3k9me1-over-the-kmt2a-bcr/run.py keeps the window unchanged so its results.json stays reproducible against what was published. The correction lives here and in the script's comments, not in a silent coordinate swap that would make the stored output unexplainable.",
"found_by": "Q110, verified independently by the king"
}- Nothing here supports or relaxes any intervention. The immediately tempting inference - that a resting depletion replicating in a leukaemia cell makes the H3K9-brake chain more solid, and therefore makes G9a or KDM3B inhibition more reasonable to contemplate - does not follow. This measures ONE link, in an unperturbed cell, in a line that is not KMT2A-rearranged, and the perturbation arms that would test the rest of the chain do not exist in any blood cell.
- The element-level result is a refusal, not a reassurance. Two tracks disagreeing in sign at the CTCF element must not be read as 'the mark is unchanged there', which would be the convenient reading for anyone arguing the mark is safe to manipulate.
- Is H3K9me1 depleted over the BCR in a CD34+ progenitor or a KMT2A-rearranged blast? Still unmeasured, and ENCODE cannot answer it - there is no CD34+ H3K9me1 track in the archive. BLUEPRINT is the next place to look and was not searched here.
- Is the depletion specific, or just what a transcribed gene body looks like for this mark? A matched control locus of comparable transcriptional status and length would settle it and was not run.
- Are ENCFF271KWX and ENCFF654SLZ independent replicates or a replicate and a pooled track? The element-level sign disagreement means the answer changes how much weight that disagreement carries.
- Does the element carry any reproducible H3K9me1 signature in any cell type? Two datasets in two cell types now decline to say, which is beginning to look like a property of the mark rather than of the datasets.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database
H3K9me1 is depleted over the KMT2A breakpoint cluster region in K562, a leukaemia line, replicating the direction found in RPE epithelium. Measured as fold over each track's own chr11 mean in two GRCh38 fold-change-over-control bigWigs from ENCSR000AKW: the 7,466 bp DCAL A+B+C core reads 0.737 and 0.801 while the mean of the two flanking 20 kb intervals reads 1.310 and 1.429, giving BCR/flank ratios of 0.563 and 0.560 against the RPE control equivalent of 0.422. Both tracks agree in direction and fall within 0.003 of each other. A run-length check gives a modal and median run of 1 bp in both files, so these are base-resolution tracks rather than binned ones.
What this evidence cannot show
K562 IS NOT A CD34+ PROGENITOR. It is a CML-derived erythroleukaemia line, aneuploid and BCR-ABL1 positive, and it is not KMT2A-rearranged - so this is a blood-lineage substitute for the experiment that was actually proposed, and that substitution is the main weakness. One experiment, one lab, two files, and whether those two GRCh38 files are independent replicates or a replicate plus a pooled track was NOT established, so they are reported separately and never averaged. Fold-change-over-control and the ChIP-over-Input ratio tracks used for RPE are not on the same scale, so the comparison with RPE is of direction and rough magnitude only. The two flanks differ from each other within one track (1.424 vs 1.196), so the BCR/flank ratio carries real uncertainty that two decimal places overstate.
Evidence that argues against this (2)
- A depletion of a heterochromatin-associated mark over a transcribed gene body is close to the expected default, not a finding. KMT2A is expressed in K562 and in RPE, and H3K9me1 tracks gene bodies poorly; a 0.56 BCR/flank ratio could reflect nothing more specific than that the BCR sits in a transcribed region while the 20 kb flanks include less-transcribed sequence. The analysis has no matched control locus of comparable transcriptional status, which is the control that would make this specific.
- The flank comparator is doing most of the work and it is not symmetric. In rep a the centromeric flank reads 1.424 and the telomeric 1.196 - a 19% difference between two intervals chosen only for being adjacent - so the denominator of every ratio here is a rough local average rather than a baseline, and a differently chosen flank would move the headline.
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
No element-level H3K9me1 claim is available, because the two tracks disagree in sign. At the 1,048 bp CTCF+RAD21+SMC3 element the two K562 tracks give fold-over-chr11-mean values of 0.877 and 1.155; at the 147 bp canonical exon 11 they give 0.836 and 1.056; and at the DCAL cut point plus or minus 250 bp they give 1.092 and 1.061. So one track reads the element as depleted and the other as enriched, and neither reads the cut point as depleted. The depletion that does replicate is a property of the 7,466 bp BCR core.
What this evidence cannot show
This is a negative about availability rather than a measurement of the element: two tracks disagreeing does not establish that the element is undepleted, only that this dataset cannot say. With n=2 and no error model, a sign disagreement is the weakest possible evidence of anything, and a third track could resolve it in either direction. The narrow intervals are legitimate here only because the bin check returned 1 bp.
Evidence that argues against this (1)
- The convergence with Q108 may be coincidence dressed as corroboration. Q108 concluded the locus address sits on the CTCF arrow because the KDM3B-induced me1 RISE was broad while the CTCF FALL was pinpoint; this analysis finds the RESTING me1 depletion broad while the element is unresolved. Those are different quantities in different cell types under different perturbations, and reading them as one story is an inference this analysis cannot support - the honest version is that neither dataset offers element-level mark specificity, which is weaker than saying both point the same way.
its results carry the cross-pipeline note bounding fold-change-over-control against ChIP-over-input to direction and rough magnitude
takes the follow-up that beat named as cheapest and did not run; the resting H3K9me1 depletion replicates in K562 at 0.563 and 0.560 against RPE's 0.422
agrees from the other direction: the resting MARK depletion is broad-domain over the 7.5 kb BCR while the element is unresolved, so element-level specificity is not the mark's - but the two datasets measure different quantities and that caveat is in the entry
the FIRST link of this chain - a resting-state property - now has a blood-lineage measurement. Only the first link: every perturbation arm remains epithelial
measured on this entry's own DCAL A+B+C core, 7,466 bp at GRCh38 chr11:118,482,422-118,489,887
narrows the cell-type gap on the MARK side without touching the occupancy side: ENCODE holds no CD34+ H3K9me1 at all and no haematopoietic H3K9me2, so K562 is the only blood-cell option in the archive
the resting depletion is the substrate KDM3B is proposed to act on; no KDM3B perturbation exists in any blood cell, so nothing here tests the enzyme
reports a control defect in that analysis: its 'ZNF274_h3k9me3_positive' window at chr19:57,996,000-58,000,000 is 187 kb from ZNF274 in GRCh38 and 698 kb in GRCh37, so it is a generic chr19 ZNF-cluster window and not ZNF274. The heterochromatin reading probably survives; the label does not
the resting-state depletion over the BCR now replicates in a primary CD34+ progenitor as well as in K562