The resting-state H3K9me2 depletion over KMT2A DOES replicate in a human CD34+ haematopoietic progenitor - measured in 3 of 3 progenitor ChIPs, where the locus sits in the 7th, 11th and 37th longest H3K9me2-free gap on chr11 - but it is not a progenitor property: granulocytes, K562 and 6 of 10 primary AML samples show the same gap
also called the CD34+ H3K9me2 test, run · GSE71809 read over the KMT2A BCR · the progenitor half of the H3K9 brake, closed
The first step of the published KMT2A-breakage brake is that a silencing tag is unusually scarce across the stretch of DNA where the gene breaks, and until now that had only been shown in an eye-derived cell line. It is now measured in real human blood stem cells and it holds - but it also holds in mature white cells, in a leukaemia line and in most leukaemia patients' cells, so it is a property of the place rather than of the cell.
The same thing, for a scientist
`exp-h3k9me3-and-ctcf-at-the-kmt2a-anchor-in-a-cd34-progenitor` named a cheaper first step than its own wet experiment: *"ENCODE and BLUEPRINT both carry CD34+ histone panels. If H3K9me1 or H3K9me2 tracks exist for CD34+ cells, the BASELINE half of the question ... is answerable in an afternoon with no wet work."* **Neither archive carries them** - measured, see `absence-no-h3k9me1-or-h3k9me2-in-a-cd34-progenitor-in-the-ihec-federation` [E3]. **GEO does.** GSE71809 (PMID 28301528) deposits H3K9me2 ChIP-seq in **normal human CD34+ bone-marrow haematopoietic progenitors**, two donors and three ChIPs, alongside granulocytes, K562 +/- the G9a inhibitor UNC0638, **ten primary AML myeloblast samples**, matched inputs and CD34+ H3K9me3/H3K27me3/H3K27ac/H3K4me2 - public in GEO since 2017-03-20, no DAC [E1]. **THE RESULT.** In all three CD34+ H3K9me2 ChIPs the *KMT2A* BCR (hg19 chr11:118,353,137-118,360,602), the 1,048 bp CTCF/RAD21/SMC3 element, canonical exon 11, both 20 kb flanks and the whole 90 kb gene span carry **zero** called H3K9me2 domain coverage, while **40.0-49.3% of chr11 and 38.8-46.6% of the genome is inside a called domain** [E2]. The gene span plus its 20 kb flanks (47,466 bp) sits inside a contiguous domain-free gap of **317,700 / 274,500 / 179,300 bp**, which is the **7th, 11th and 37th longest of 2,076 / 2,116 / 1,749 such gaps on chr11** - the 98.1st to 99.7th percentile [E2]. **AND THE DEFLATION, which is the more useful half.** The same gap is present in **4 of 4 granulocyte ChIPs, 2 of 2 K562 ChIPs, both UNC0638-treated K562 ChIPs, CD34+ H3K9me3, CD34+ H3K27me3 and CD34+ H3K4me2, and in 6 of 10 primary AML myeloblast ChIPs** [E2]. So the depletion is not a progenitor property and not a normal-cell property: it is a property of the locus that survives lineage, maturation, transformation and G9a inhibition. Four of the ten AML samples DO place H3K9me2 over the locus, which is the direction the paper's AML-specific H3K9me2 blocks predict [E1]. **The null matters: ~17.8-21.5% of same-width chr11 windows are entirely domain-free in these samples**, so "entirely free" alone is a 1-in-5 event and only the gap-length percentile is a strong statement [E2]. And **CD34+ H3K27ac covers the BCR and the element completely** in the one CD34+ H3K27ac ChIP, so the locus is not silent chromatin in a progenitor [E2].
What was measured, and at what resolution
GSE71809's supplementary files are RSEG 0.4.8 RC ENRICHED-domain calls in BED, one per sample - not coverage tracks. So this cannot produce the BCR/flank signal ratio that ana-h3k9me1-depletion-over-the-kmt2a-bcr-replicates-in-a-blood-cell-but-not-at-the-element produced from ENCODE K562 bigWigs (0.563 and 0.560 against RPE's 0.422). What it produces is a domain-level presence/absence readout, which is the right shape for a question about a broad-domain mark and the wrong shape for the 1 kb element 2.
The resolution was measured, not assumed, because a domain call answers a narrow query without error exactly as a binned coverage track does (absence-gse210480-processed-tracks-are-1-kb-binned). Domain boundaries in these files are multiples of 100 bp; the modal called-domain width is on the order of a kilobase and the median tens of kilobases, varying by sample. A 147 bp exon is therefore inside or outside a domain and nothing finer is available 2.
| CD34+ sample | ChIP | domains | chr11 in a domain | KMT2A gene+20kb free | containing gap | gap rank on chr11 |
|---|---|---|---|---|---|---|
| GSM1846171 | H3K9me2 | 45,778 | 44.9% | yes | 317,700 bp | 7 of 2,076 |
| GSM1846172 | H3K9me2 | 46,839 | 40.0% | yes | 274,500 bp | 11 of 2,116 |
| GSM2152593 | H3K9me2 | 39,515 | 49.3% | yes | 179,300 bp | 37 of 1,749 |
| GSM2152597 | H3K9me3 | 34,161 | 44.9% | yes | 180,000 bp | 56 of 1,516 |
| GSM2152594 | H3K27me3 | 9,251 | 32.1% | yes | - | - |
| GSM2152595 | H3K27ac | 14,502 | 35.6% | NO - 100% covered | - | - |
| GSM2152596 | H3K4me2 | 75,441 | 15.2% | yes | - | - |
The two donors are independent: CD34+ 1 (female) gives GSM1846171 and GSM1846172, CD34+ 2 (male) gives GSM2152593. Both are cryopreserved bone-marrow CD34+ progenitors from AllCells, fixed immediately after thawing 1.
The controls, including the one that fails
The negative control that matters is the input. RSEG run on ChIP input calls 8,056-19,686 domains covering 35.1-46.4% of the genome, which is not a meaningful enrichment - it is a segmentation of coverage noise. Two of the three CD34+ inputs call nothing over the locus and one (GSM1846173) calls the entire locus ENRICHED 2. That one input is the single most useful number here, and it cuts both ways:
- It proves the region is not a deadzone. RSEG was run with
deadzones-k41-hg19.bed, and a mappability hole would make every sample blank there. One sample calls a domain, so the zeros in the H3K9me2 ChIPs are not an artefact of unmappability. That is the check that makes the result interpretable at all. - It also means the input tracks cannot serve as a quantitative comparator, because they disagree with each other at this locus. The honest reading is: the inputs establish callability and nothing else.
The mark-identity control. ZNF274 body, the H3K9me3 positive control this graph uses, and the GAPDH and ACTB promoters are all measured in the same pass and are in results.json per sample.
The pharmacological arm is present and it does nothing here. K562 treated for 5 days with 0.5 uM UNC0638, a G9a/GLP inhibitor, still shows zero H3K9me2 domain coverage over KMT2A - which is expected, since the locus was already free, and is a reminder that a floor cannot fall further 12.
What this does and does not do to the chain
It closes the cheaper first step of exp-h3k9me3-and-ctcf-at-the-kmt2a-anchor-in-a-cd34-progenitor for H3K9me2 at domain resolution, and it closes it POSITIVELY: the baseline depletion is not an RPE artefact.
It does not close the experiment. Four things remain untouched. (1) H3K9me1 in a CD34+ progenitor does not exist in any archive checked 3, and me1 is the mark the published chain moves. (2) No perturbation. GSE71809 has no KDM3B arm in any cell type, so the change half of the chain is untested in blood. (3) Element resolution. A domain call cannot say anything about the 1,048 bp CTCF element, which is exactly where the K562 bigWig comparison already disagreed in sign (0.877 against 1.155). (4) No CTCF. GSE71809 has no CTCF ChIP, so the step that carries the locus address in mechanism-kdm3b-depletion-lowers-ctcf-only-at-the-kmt2a-anchor is still unmeasured in any blood cell, as absence-no-human-ctcf-or-cohesin-occupancy-measurement-in-a-blood-cell-at-kmt2a records.
And it weakens the specificity reading. If the H3K9me2-free gap over KMT2A is present in progenitors, granulocytes, K562, UNC0638-treated K562 and most AML blasts, then baseline mark depletion at this locus is not what distinguishes the cells in which KMT2A rearranges. Whatever selects the locus has to be something that varies between those cells, and this measurement says the resting mark state is not it.
| script | analyses/q111-cd34-h3k9me2-over-the-kmt2a-bcr/run.py |
|---|---|
| second script | analyses/q111-cd34-h3k9me2-over-the-kmt2a-bcr/nulls.py |
| status | run |
| run date | 2026-09-22 |
| series | GSE71809 |
| n samples read | 31 |
| deposit format | RSEG 0.4.8 RC ENRICHED-domain calls in BED, hg19; NOT coverage tracks |
| cd34 donors | 2 |
| input control | 2 of 3 CD34+ ChIP inputs call nothing at the locus, 1 (GSM1846173) calls it entirely ENRICHED - which establishes the region is not a mappability deadzone and simultaneously disqualifies the inputs as a quantitative comparator |
| absence class | measured-not-analysed |
| archive check run | true |
| archive queried | NCBI GEO (elink pubmed->gds, then acc.cgi per-sample SOFT records); ENCODE portal search API; IHEC Data Portal metadata API, all 50 releases; BLUEPRINT DCC current_release file index; EGA metadata API |
| localization | chromatin at chr11q23.3, the KMT2A gene body and its flanks; a domain-occupancy comparison that asserts nothing about protein localisation |
| subnuclear localization | not measured. A large H3K9me2-free block is the shape usually associated with the nuclear interior rather than the lamina, but no imaging, Hi-C or DamID was read here and no subnuclear position is claimed |
| cell type | normal human CD34-positive bone-marrow haematopoietic progenitors (two donors, cryopreserved, AllCells), with normal human peripheral-blood granulocytes, K562 erythroleukaemia and ten primary AML myeloblast samples as comparators - the first haematopoietic H3K9me2 measurement at this locus in this graph |
| cell state | resting / unperturbed for the progenitor and granulocyte samples; K562 asynchronous culture with and without 5 days of 0.5 uM UNC0638; AML samples are diagnostic or presentation marrow, state not further specified |
| epigenetic readout | H3K9me2 ChIP-seq ENRICHED-domain occupancy over the KMT2A BCR against chromosome- and genome-wide domain fraction, with H3K9me3, H3K27me3, H3K27ac and H3K4me2 in the same CD34+ donor as within-cell mark controls |
| shown for fusion | Not shown for any fusion. The CD34+, granulocyte and K562 cells carry no KMT2A fusion; the ten AML samples have no karyotype or fusion status recorded in GEO, so whether any is KMT2A-rearranged is unknown and must not be assumed either way. |
| cost tier | low - 31 domain-call BED files, a few megabytes, no alignment |
| absence class note | MIGRATED 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, and it CONVERTS a claim this graph had recorded as measured-not-searched: the CD34+ H3K9me2 data existed and the search space (ENCODE plus BLUEPRINT) excluded it' |
results
[ "analyses/q111-cd34-h3k9me2-over-the-kmt2a-bcr/results.json", "analyses/q111-cd34-h3k9me2-over-the-kmt2a-bcr/results_nulls.json" ]
cd34 h3k9me2 samples
[ "GSM1846171", "GSM1846172", "GSM2152593" ]
kmt2a domain coverage fraction cd34 h3k9me2
{
"bcr_core_7466bp": [
0,
0,
0
],
"element_ctcf_1048bp": [
0,
0,
0
],
"exon11_canonical_147bp": [
0,
0,
0
],
"flank_centromeric_20kb": [
0,
0,
0
],
"flank_telomeric_20kb": [
0,
0,
0
],
"gene_90kb": [
0,
0,
0
]
}chr11 domain fraction cd34 h3k9me2
[ 0.449, 0.4, 0.493 ]
genome domain fraction cd34 h3k9me2
[ 0.435, 0.388, 0.466 ]
containing gap bp
[ 317700, 274500, 179300 ]
gap rank on chr11
[ 7, 11, 37 ]
n chr11 gaps
[ 2076, 2116, 1749 ]
gap length percentile
[ 0.9971, 0.9957, 0.9806 ]
the null
{
"fraction_of_same_width_chr11_windows_entirely_domain_free": [
0.1889,
0.2148,
0.1781
],
"n_windows_tested": 13496,
"why_it_is_here": "40-49% of chr11 is inside a domain, so about half the chromosome is domain-free by construction; 'entirely free' for a 47 kb window is a 1-in-5 event and only the gap-length percentile carries information"
}generalisation across cells
{
"granulocyte_H3K9me2": "4 of 4 samples domain-free at the locus",
"K562_H3K9me2": "2 of 2 domain-free",
"K562_plus_UNC0638_H3K9me2": "2 of 2 domain-free",
"primary_AML_myeloblast_H3K9me2": "6 of 10 domain-free; 4 of 10 carry coverage",
"CD34_H3K9me3": "domain-free",
"CD34_H3K27me3": "domain-free",
"CD34_H3K4me2": "domain-free",
"CD34_H3K27ac": "100% COVERED - the locus is in active chromatin in a progenitor"
}resolution measured not assumed
{
"median_called_domain_width_bp_cd34_h3k9me2": [
15300,
12750,
17000
],
"median_called_domain_width_bp_cd34_inputs": [
63600,
63600,
23400
],
"boundary_modulus_differs_between_samples_of_the_same_mark": "100% of GSM1846171's domain boundaries are multiples of 300 bp; only 24.8% of GSM1846172's and GSM2152593's are multiples of 100 bp",
"the_mark_identity_control_cannot_discriminate": "in CD34+ H3K9me3 the ZNF274 body, the GAPDH promoter, the ACTB promoter and the MLLT1 and TCF3 gene bodies are ALL 100% covered while KMT2A is 0% - a 15 kb domain swallows the neighbourhood, so the positive control passes for the wrong reason and only the zero is informative",
"what_is_therefore_supported": "domain-scale statements only: the gap, its length, and its rank among chr11 gaps. NOT a ratio, NOT the 1,048 bp element, NOT the 147 bp exon.",
"see": "concept-an-enriched-domain-call-is-not-a-coverage-track"
}what is still not measured
[ "H3K9me1 in any CD34+ progenitor - absent from every archive checked", "any KDM3B or G9a perturbation in a haematopoietic cell", "CTCF or cohesin occupancy in any blood cell at this element", "anything at 1 kb resolution, because the deposit is domain calls" ]
- This entry makes the resting-state link look more secure than it is, and in a direction that matters for an intervention. A prophylactic G9a inhibitor is supposed to work by keeping H3K9me1/2 low at KMT2A. If the locus is ALREADY free of H3K9me2 domains in progenitors, in granulocytes, in K562 and in most AML blasts, then there is no H3K9me2 there to remove, and the drug's proposed action at this locus has no substrate in the resting state. The measurement is consistent with the published baseline and simultaneously removes the obvious reason to expect a locus-specific drug effect.
- This graph now holds three H3K9 measurements at this locus that predict different things for the same manipulation - a broad H3K9me1 rise on KDM3B loss in RPE, a K562 element-scale pair that disagrees in sign, and this domain-level H3K9me2 gap that is present in every haematopoietic cell measured. Any safety statement about a G9a or KDM3B intervention has to contend with all three, and the honest summary is that the sign of the effect at this element in a blood cell is unknown.
- The four AML samples that DO carry H3K9me2 over the locus are the uncomfortable rows. If H3K9me2 accumulation at KMT2A were a feature of leukaemic cells rather than of normal ones, a G9a inhibitor given to a patient would be acting where the mark exists - and the paper's own model says the AML-enriched blocks are the G9a-dependent ones. That is an argument for the intervention in established disease and against it as prophylaxis, which is the opposite of how this graph has framed it.
- Does the H3K9me2 depletion survive at coverage resolution in a CD34+ progenitor? The raw reads are in SRA (every sample carries an SRX, e.g. GSM1846171 -> SRX1136002), so a BCR/flank signal ratio directly comparable with the K562 bigWig figures of 0.563 and 0.560 and RPE's 0.422 is an alignment job on 31 libraries - medium cost, and it is the single cheapest way to make this result numerically commensurable with the two that already exist.
- Which of the ten AML samples in GSE71809 are KMT2A-rearranged? Four of the ten AML H3K9me2 ChIPs place a domain over the KMT2A locus and six do not, and if that split tracked KMT2A rearrangement status it would be the first haematopoietic evidence bearing on the mark arm of this chain in the disease. The karyotypes are not in GEO; they may be in the paper's supplementary tables or obtainable from the Penn State group.
- Is the KMT2A locus inside a large H3K9me2-free block in EVERY human cell type, or only in these? A gap that is invariant is architecture; a gap that varies is a candidate mechanism. Answering it needs H3K9me2 in more cell types, and the ENCODE 19 - nine of which are haematopoietic-lineage cell lines including the T-ALL line Loucy - are the obvious next set and were not read here.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
GSE71809 deposits H3K9me2 ChIP-seq in normal human CD34+ bone-marrow haematopoietic progenitors (two donors, three ChIPs: GSM1846171, GSM1846172, GSM2152593), plus four granulocyte H3K9me2 ChIPs, K562 with and without 0.5 uM UNC0638 for 5 days, ten primary AML myeloblast samples, matched ChIP inputs, and CD34+ H3K9me3, H3K27me3, H3K27ac and H3K4me2; reads were aligned to hg19 with bowtie 1.0.0 and domains called with RSEG 0.4.8 RC using deadzones-k41-hg19.bed; the series has been public in GEO since 2017-03-20 with no data-access committee.
Genome-wide mapping of histone H3K9me2 in acute myeloid leukemia reveals large chromosomal domains associated with massive gene silencing and sites of genome instability. · Salzberg AC, Harris-Becker A, Popova EY, Keasey N, Loughran TP, Claxton DF, Grigoryev SA · PLoS One 2017What this evidence cannot show
The CD34+ cells are CRYOPRESERVED commercial bone-marrow CD34+ progenitors (AllCells, Emeryville CA), thawed and fixed immediately - not freshly isolated, not cord blood, and not sorted beyond CD34. n = 2 donors. The ten AML samples carry no karyotype, no fusion status and no FAB subtype in the GEO characteristics, so NONE of them is known to be KMT2A-rearranged and none can be assumed not to be. Chromatin was digested with micrococcal nuclease to 150-200 bp rather than sonicated, which is unusual for a broad-mark ChIP and changes the fragment-size bias relative to most H3K9me2 datasets. Two sequencing platforms are mixed across the series (Illumina HiSeq 2500 and AB SOLiD 4), and for male samples the chrX/chrY reads were DUPLICATED before domain calling, which is why several filenames carry 'XYdouble' - this analysis uses only autosomal chr11, chr19, chr9, chr2, chr12 and chr7 intervals, so the duplication does not enter, but any future chrX claim from these files must account for it.
Evidence that argues against this (1)
- The paper's own framing is that H3K9me2 UNDERGOES DRAMATIC REARRANGEMENT between progenitors and granulocytes and accumulates in AML on >100 kb blocks enriched for translocation sites. If the mark is that mobile between these cell types, a single domain-free gap shared by all of them is more likely a property of chromatin architecture at 11q23 - gene density, transcription, replication timing - than a mechanism-bearing feature. The paper also reports that the AML-enriched blocks are the G9a-dependent ones, so a locus that is free in progenitors AND in most AML is, on the paper's own model, in the part of the genome the mechanism does not act on. Salzberg AC et al. 2017
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Read from GSE71809's deposited RSEG domain calls: in all three CD34+ H3K9me2 ChIPs the KMT2A BCR, the 1,048 bp CTCF element, canonical exon 11, both 20 kb flanks and the 90 kb gene span carry zero called H3K9me2 domain coverage while 40.0-49.3% of chr11 is inside a called domain, and the 47,466 bp gene-plus-flanks span lies inside a contiguous domain-free gap of 317,700, 274,500 and 179,300 bp - the 7th, 11th and 37th longest of 2,076, 2,116 and 1,749 chr11 gaps (98.1st-99.7th percentile); the same span is entirely domain-free in 4 of 4 granulocyte, 2 of 2 K562, 2 of 2 UNC0638-treated K562 and 6 of 10 primary AML H3K9me2 ChIPs, and in CD34+ H3K9me3, H3K27me3 and H3K4me2, but is 100% covered in CD34+ H3K27ac; 17.8-21.5% of same-width chr11 windows are entirely domain-free in the CD34+ H3K9me2 samples, and one of the three CD34+ ChIP inputs calls the whole locus ENRICHED, which establishes that the region is not a mappability deadzone.
GSE71809 RSEG ENRICHED-domain BED files for 31 samples, parsed over hg19 intervals with per-sample genome-wide and chr11-wide domain-coverage nulls and a gap-length ranking, 2026-09-22; code and full per-sample output at analyses/q111-cd34-h3k9me2-over-the-kmt2a-bcr/run.py, nulls.py, results.json and results_nulls.json · Salzberg AC, Harris-Becker A, Popova EY, Keasey N, Loughran TP, Claxton DF, Grigoryev SA · NCBI GEO (data accompanying PLoS One 12:e0173723) 2017What this evidence cannot show
A DOMAIN CALL IS NOT A COVERAGE TRACK. Zero domain coverage means 'RSEG did not call an enriched domain here', not 'there is no H3K9me2 here'; a sub-threshold enrichment is invisible and no ratio is computable, so this cannot be compared numerically with the K562 bigWig BCR/flank figures of 0.563 and 0.560 or with RPE's 0.422. AND THE MARK-IDENTITY CONTROL CANNOT DISCRIMINATE AT THIS RESOLUTION, which is measured rather than feared: the called domains have a median width of 12,750-17,000 bp in the CD34+ H3K9me2 samples and 23,400-63,600 bp in the inputs, and in the CD34+ H3K9me3 sample the ZNF274 body (the intended positive control) AND the GAPDH promoter, the ACTB promoter and the whole MLLT1 and TCF3 gene bodies are all 100% covered. A 15 kb domain swallows the neighbourhood, so only a domain-scale statement - the gap and its length rank - is supported, and the per-interval covered_fraction values for intervals narrower than a domain are binary membership and not occupancy. The caller's bin size also differs BETWEEN samples of the same mark (100% of GSM1846171's boundaries are multiples of 300 bp; only 24.8% of GSM1846172's and GSM2152593's are multiples of 100), so resolution is a per-file property here. Full measurement in concept-an-enriched-domain-call-is-not-a-coverage-track. The domain calls are a single segmentation per sample with no replicate-consistency filter and no FDR. Gap-length percentile depends on the chromosome's own domain density, which varies 40.0-49.3% across three samples of the same cell type - a 23% relative spread that bounds how precise any of these percentiles can be. The 100 bp mask resolution in nulls.py rounds every boundary outward, which can only make a gap look SHORTER and a coverage look LONGER, so the domain-free calls are conservative. hg19 coordinates come from this project's GRCh38 values minus the 129,285 bp offset re-derived independently in analyses/q111-ctcf-anchor-matched-ranking/run.py across six intervals from 2 bp to 90 kb; they were not re-lifted here.
Evidence that argues against this (1)
- The null undercuts the headline. 17.8-21.5% of same-width chr11 windows are entirely domain-free in these samples, so an interval of this width being free is roughly a 1-in-5 event and is not by itself surprising. Only the GAP LENGTH percentile (98.1-99.7%) is a strong statement, and gap length is a property of a ~200-300 kb neighbourhood, not of the 7.5 kb BCR - so the measurement may be reporting that 11q23.3 sits in a large euchromatic block rather than anything about the breakpoint cluster. The CD34+ H3K27ac result points the same way: the locus is in active chromatin, and active chromatin is H3K9me2-poor by construction. Iqbal Lab oncofusion knowledge graph 2026
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
No H3K9me1 or H3K9me2 dataset in a CD34-positive haematopoietic progenitor exists in the IHEC Data Portal federation: across all 50 data releases and 49,663 datasets the portal carries 178 distinct assay labels, of which H3K9me2 is not one at all and H3K9me1 accounts for 9 datasets in four biosamples (IMR-90, HUVEC, K562, NHEK keratinocyte), and BLUEPRINT's own current_release file index (11,278 file rows, 15 distinct EXPERIMENT_TYPE values) contains zero H3K9me1 and zero H3K9me2 rows anywhere in the project.
IHEC Data Portal metadata API (getReleases.py and getDataHub.py, all 50 releases) and the BLUEPRINT DCC current_release file index 20160816.data.index, retrieved 2026-09-22 · International Human Epigenome Consortium Data Portal; BLUEPRINT Data Coordination Centre, EMBL-EBI 2026What this evidence cannot show
Full detail, every archive named and every trap recorded, is in absence-no-h3k9me1-or-h3k9me2-in-a-cd34-progenitor-in-the-ihec-federation. The IHEC portal is an aggregator and does not ingest everything its member consortia hold - ENCODE's 19 H3K9me2 experiments appear in ENCODE's own API and in NONE of the portal's releases - so the portal bounds the portal, not the consortia. That is why the ENCODE API and the BLUEPRINT file index were queried separately.
Evidence that argues against this (1)
- This absence is an absence in curated consortium archives, and GSE71809 is the proof that it does not bound the literature: a fully public CD34+ H3K9me2 dataset had existed in GEO for nine years while this graph was writing H3K9me2 absences bounded by ENCODE and BLUEPRINT. The correct lesson is that a consortium archive is a poor instrument for a non-core mark, because IHEC's six-mark core panel excludes H3K9me1 and H3K9me2 by design. NCBI Gene Expression Omnibus 2017
closes that experiment's own named cheaper first step for H3K9me2, positively and at domain resolution
the resting-state depletion over the BCR now replicates in a primary CD34+ progenitor as well as in K562
first haematopoietic H3K9me2 measurement at the KMT2A locus in this graph - CD34+ progenitors, granulocytes, K562 and eight primary AML samples
- expressed in →CD34 - named in 132 entries of this graph and written in none, and it does not mean what the graph has been using it to mean
normal human CD34-positive bone-marrow haematopoietic progenitors, two donors
- related to →KMT2A / MLL1 — the gene that breaks
the KMT2A BCR, element, exon 11 and 90 kb gene span carry zero H3K9me2 domain coverage in 3 of 3 CD34+ ChIPs
the consortium absence is real and does not bound the literature - GEO has had the CD34+ H3K9me2 data since 2017
the resolution bound that decides what that measurement can and cannot say
the locus-level record of that measurement, with block coordinates and chr11 gap ranks