Oncofusion

start / graph / absence-no-human-ctcf-or-cohesin-occupancy-measurement-in-a-blood-cell-at-kmt2a

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q107-break-machinery-deposition

Not one of the four studies in this dispute measured CTCF or cohesin occupancy in a human blood cell - so this beat read ENCODE instead, and the element is a STRONGER CTCF site in a B-lymphoblastoid line than in HCT116, with the summit at the same base

also called the occupancy gap, closed in-run · GM12878 puts the summit at chr11:118,488,694 too · a reading gap, not a literature gap

The whole argument turns on a protein sitting on one short piece of the KMT2A gene in a blood cell, and none of the four papers measured it in a blood cell. So we looked it up in a public atlas: in a human B-cell line the protein sits there more strongly than in the bowel-cancer cells the papers used, and at the very same base.

The same thing, for a scientist

Enumerated across all four deposits, **the human CTCF/cohesin occupancy measurements at this element are three HCT116 ChIP samples (colorectal) and four RPE ChIP samples (retina), and nothing haematopoietic**: Canela 2017's GSE99195 has **zero human samples**, and Gothe 2019 generated **no ChIP at all**, importing occupancy from ENCODE and loop anchors from Rao 2014 GM12878 Hi-C [E1][E2]. **That absence is real and it is a reading gap, not a literature gap - so this beat closed it.** Two independent ENCODE CTCF ChIP-seq experiments in **GM12878**, a human B-lymphoblastoid line, place the signal maximum over the element at **GRCh38 chr11:118,488,694** (ENCFF734CUT, ENCSR000AKB, reps 1+2) and **chr11:118,488,671** (ENCFF800WUV, ENCSR000DKV, reps 1-3), with the 147-bp exon reading **113.3x** and **135.3x** the respective file's genome-wide mean; ENCODE RAD21 in the same line (ENCFF571ZJJ) reads **35.1x** with its maximum at chr11:118,488,664 [E3]. **The first of those coordinates is the identical base this beat measured independently as the CTCF summit in HCT116.** For scale in the same files, the canonical H19/IGF2 imprinting control region reads 5.0x and 16.2x - **the *KMT2A* exon-11 element is 7 to 27 times stronger than the textbook CTCF site** - while the functional cut point reads 1.5-2.0x and a gene desert 1.0x [E3]. **Conclusion: occupancy at this element is not a colorectal artefact. It is at least as strong in B lineage, and the summit is reproducible to the base across two cell types and three independent experiments.**

Go deeper
The absence, enumerated, so it is bounded rather than asserted
studyChIP serieshuman samplestargetscell types
Canela 2019 (31202577)GSE1295263RAD21, CTCF, inputHCT116 (colorectal); the other 9 are mouse pre-B and MEF
Canela 2017 (28735753)GSE991950CTCF, RAD21, TOP2A, TOP2Bmouse activated B, resting B, T, MEF; Mus spretus activated B
Gothe 2019 (31202576)none generated--occupancy from ENCODE; loops from Rao 2014 GM12878 Hi-C
Gray 2023 (37788669)GSE2104804CTCF (siCT and siKDM3B, 2 reps each)RPE (retinal pigment epithelium)

Human CTCF/cohesin ChIP samples in a haematopoietic cell, across the entire dispute: zero. Gray's RPE choice is explicit and defensible - near-diploid, genomically stable, no cancer mutations, which is what a FISH copy-number readout needs - and its haematopoietic and primary arms (KG1a, HL60, U937, primary AML, a primary AML organoid, cord-blood CD34+ HSPCs, mouse spleen) are scored only by two-colour FISH with no H3K9 or CTCF ChIP in any blood cell 2.

Closed in-run: ENCODE GM12878, and it agrees to the base

All values GRCh38, region means divided by that file's own genome-wide mean from its bigWig header 3:

region (GRCh38)CTCF ENCFF734CUTCTCF ENCFF800WUVRAD21 ENCFF571ZJJ
canonical exon 11, chr11:118,488,614-118,488,760113.27x135.25x35.08x
the 1,048-bp ENCODE cluster27.92x24.38x9.80x
weak cluster at canonical exon 10, chr11:118,485,122-118,486,0004.10x5.91x2.58x
cut point +/- 500 bp, chr11:118,484,361-118,485,3621.96x1.46x1.11x
positive control H19/IGF2 ICR, chr11:2,148,000-2,152,0005.01x16.18x3.54x
negative control GAPDH promoter, chr12:6,534,000-6,535,0002.62x3.52x1.40x
negative control chr2 gene desert0.97x1.01x1.21x
argmax within the clusterchr11:118,488,694chr11:118,488,671chr11:118,488,664

Three things follow, and they are the load-bearing content of this entry.

  1. ENCFF734CUT's maximum, GRCh38 chr11:118,488,694, is the exact base this beat measured as the CTCF summit in HCT116 (GRCh37 chr11:118,359,409 + 129,285). Two cell types, two laboratories, two decades apart in reagent terms, one base. The three maxima across the two CTCF experiments and the RAD21 experiment span 30 bp.
  2. The element is 7 to 27 times stronger than the H19/IGF2 imprinting control region in the same files. That is the most-cited CTCF site in the genome. This is not a marginal peak in B lineage; it is an exceptionally strong one.
  3. Lineage does not rescue the cut point. In a B-lymphoblastoid cell the functional intron-9/exon-10 cut point reads 1.46-1.96x for CTCF and 1.11x for RAD21 - i.e. the element and the cut point are 3.8 kb apart and one of them is 60-90x more occupied than the other, in the lineage that matters. Q105's central geometric finding survives a lineage-matched test.

What the closure does NOT cover. GM12878 is an EBV-transformed adult B-lymphoblastoid line, not a B-cell precursor, not CD34+, and not fetal. No CTCF ChIP in a CD34-positive common myeloid progenitor exists in ENCODE GRCh38 with fold-change-over-control output (queried: 0 results). A fetal-liver or fetal-HSC measurement remains genuinely absent, and that is the cell the index disease originates in.

The methodological point, which is the reason this entry is worth its length

This entry was written as a bounded absence with the sentence "the absence is that nobody in this argument has done it, not that the data does not exist... which makes it the cheapest thing on this beat's list to close" - and then it was closed, in the same session, by four API calls and three interval reads.

That is the failure mode: a caveat that names a one-query gap is not a caveat, it is an unexecuted query. The first draft of this entry listed "read ENCODE CTCF in GM12878 at this element" under what_would_close_it and assigned it to a future reader. There was no reason for that other than momentum. Anything in a what_would_close_it field that costs less than writing the field should be done before the field is written.

It also changes the sign of the risk this entry carried. Its original bite-back read: "If ENCODE GM12878 shows no CTCF peak at exon 11, the protective-site half of the dispute loses its element and the graph's adjudication would need rewriting from the coordinate up." That was the right thing to worry about and the answer is the reassuring one - and it was one query away from being either a crisis or a confirmation, sitting unasked while an entry chain several nodes deep rested on it.

Facts and figures
absence classmeasured-not-searched
still absentCTCF or cohesin occupancy at this element in a B-cell PRECURSOR, in a CD34-positive progenitor, or in any fetal haematopoietic cell. ENCODE has no CTCF ChIP-seq in CD34-positive common myeloid progenitor with GRCh38 fold-change-over-control output (queried 2026-09-22, 0 results).
statusrun
absence class noteMIGRATED 2026-09-23. This field previously held the whole verdict as free text: 'reading gap, not a literature gap - CLOSED IN-RUN by reading ENCODE GM12878'. That string was out of the controlled vocabulary, so a checker could not read it - but it said more than any token can, and it has been kept here rather than deleted. The original free text read 'reading gap, not a literature gap - CLOSED IN-RUN by reading ENCODE GM12878'. The signal was in an archive and the literature search space excluded it, which is the definition of measured-not-searched.
human ctcf or cohesin chip samples in the dispute
GSE99195_human0.000GSE99195_human: 0Gothe_2019_generated0.000Gothe_2019_generated: 0haematopoietic0.000haematopoietic: 0GSE129526_HCT1163.000GSE129526_HCT116: 3GSE210480_RPE_CTCF4.000GSE210480_RPE_CTCF: 4
human ctcf or cohesin chip samples in the dispute — the numbers behind the chart
{
  "GSE129526_HCT116": 3,
  "GSE99195_human": 0,
  "Gothe_2019_generated": 0,
  "GSE210480_RPE_CTCF": 4,
  "haematopoietic": 0
}
closure
{
  "files_read": [
    "ENCFF734CUT - CTCF, GM12878, ENCSR000AKB, reps 1+2, GRCh38 fold change over control",
    "ENCFF800WUV - CTCF, GM12878, ENCSR000DKV, reps 1-3, GRCh38 fold change over control",
    "ENCFF571ZJJ - RAD21, GM12878, ENCSR000EAC, reps 1+2, GRCh38 fold change over control"
  ],
  "exon11_x_genome_mean": {
    "CTCF_ENCFF734CUT": 113.27,
    "CTCF_ENCFF800WUV": 135.25,
    "RAD21_ENCFF571ZJJ": 35.08
  },
  "summit_grch38": {
    "CTCF_ENCFF734CUT": "chr11:118488694",
    "CTCF_ENCFF800WUV": "chr11:118488671",
    "RAD21_ENCFF571ZJJ": "chr11:118488664",
    "CTCF_HCT116_independent": "chr11:118488694"
  },
  "summit_spread_bp": 30,
  "ratio_to_h19_icr": "7x to 27x stronger than the canonical H19/IGF2 imprinting control region in the same files",
  "cut_point_x_genome_mean": "1.46 to 1.96 for CTCF, 1.11 for RAD21 - unoccupied"
}
How this could bite us back
  • The reassuring result raises the stakes rather than lowering them: an exceptionally strong, lineage-conserved CTCF site inside a coding exon of a leukaemia driver is now a plausible base-editing or CRISPRi target, and the two literatures predict opposite signs for weakening it - Gray says rearrangement goes UP, the fragility side says breakage goes DOWN. Confirming the site's strength makes an intervention more tempting and no safer.
  • This entry closed its own gap and the next one may not. An absence entry whose closure route is four API calls should be treated as a task queue, not as a finding, and the graph should expect some of its bounded absences to be of exactly this kind.
Still open
  1. Is CTCF bound at this element in a FETAL haematopoietic cell? ENCODE has no CD34-positive common myeloid progenitor CTCF track with GRCh38 fold-change output, and the index disease originates in fetal liver. BLUEPRINT and the Roadmap fetal collections were not searched by this beat.
  2. The two GM12878 CTCF experiments disagree by 20% on the exon mean and 23 bp on the summit. Is that ordinary inter-experiment variation, or does the element have two adjacent CTCF sites that different antibodies weight differently? A PWM scan of the footprint would answer it and is undone.
  3. What is the element's rank among all GM12878 CTCF peaks genome-wide? A rank, not a fold-change, is what makes 'exceptionally strong' a defensible phrase, and this beat computed ranks only for the HCT116 and break tracks.
Every source behind this entry (3)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    Across the three fragility-side deposits there are exactly three human CTCF/cohesin ChIP samples and all three are HCT116: GEO GSE129526 contains GSM3714939 (RAD21, HCT116), GSM3714940 (CTCF, HCT116) and GSM3714941 (input, HCT116) alongside nine mouse samples; GEO GSE99195 contains 18 samples of CTCF, RAD21, TOP2A and TOP2B in mouse activated B, resting B, T and MEF plus Mus spretus activated B with zero Homo sapiens samples; and PMID 31202576 generated no ChIP-seq at all, its GSE121742 SuperSeries comprising only BLISS-Seq and RNA-Seq, with CTCF and cohesin occupancy taken from ENCODE and chromatin loop anchors taken from Rao et al. 2014 GM12878 Hi-C.

    DOI
    What this evidence cannot show

    This bounds the deposits, not the papers: for Canela 2017 and Canela 2019 the full texts were not read, so a ChIP shown as a figure without a GEO deposit is not excluded. The Gothe statement is from the December 2018 preprint. The three HCT116 samples of GSE129526 are mis-annotated as Mus musculus/mm10 in GEO and are human hg19, so an organism-field query returns zero human ChIP samples across all three deposits - a more extreme version of the same absence and a metadata artefact rather than a fact.

    Evidence that argues against this (1)
    • This absence was a reading gap and not a data gap, and E3 of this entry closes it: ENCODE CTCF and RAD21 ChIP in GM12878 is public, readable, and puts a maximum at the same base as HCT116. So the enumeration remains true and the worry it was written to express does not survive - which is the right outcome for a bounded absence and the reason absences should be written with their closure route attached. ENCODE Project Consortium 2020
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    The only CTCF ChIP-seq in the protection-side study (PMID 37788669) is in RPE cells - four deposited CTCF tracks in GEO GSE210480, GSM6431019/GSM6431021 (siKDM3B rep1/rep2) and GSM6431023/GSM6431025 (siCT rep1/rep2), as bigWig coverage with no peak calls in the series supplement - while that study's haematopoietic and primary arms (KG1a, HL60, U937, a primary AML line, a primary AML organoid, cord-blood CD34+ HSPCs and mouse spleen) are scored only by two-colour DNA FISH with no H3K9 or CTCF ChIP performed in any blood cell.

    NCBI GEO series GSE210480 supplementary filelist.txt and sample listing (60 samples, Homo sapiens, RPE), as recorded in this graph's Q102 and Q105 entries from the full text of PMID 37788669 read via efetch db=pmc on PMC10591855 · Gray ZH, Chakraborty D, Duttweiler RR, Alekbaeva GD, Sansano SS, Kaur S, Tsai HP, Vo N, Sun Y, Sarkar T, Mahmoud L, Tang Y, Islam M, Gaballa AA, Roman J, Bhattacharjee S, Lorenz IC, Rieger H, Gujja S, Karnoub AE, Whetstine JR · Cell 186:4528-4545.e18 2023
    PMID 37788669DOI
    What this evidence cannot show

    This beat verified the GSE210480 filelist and confirmed that one H3K9me3 bigWig opens with a padded hg19 chrom.sizes (every chromosome length is hg19 + exactly 100,000 bp). The sample enumeration and the claim that no H3K9 or CTCF ChIP was performed in a blood cell are carried from this graph's Q102 and Q105 entries, which read the full text at PMC10591855; this beat did not re-read that full text, so those clauses are second-hand from internal entries and are flagged as such.

    Evidence that argues against this (1)
    • RPE was a deliberate and defensible choice for Gray's phenotype rather than an oversight: the readout is copy-number gain and break-apart by FISH, which needs a near-diploid, genomically stable, non-cancer background to score at all. Criticising the tissue is criticising the only design in which that phenotype is measurable. The valid criticism is narrower - the chromatin mechanism was established in RPE, the disease relevance was established by FISH in blood cells, and no chromatin measurement links them. Gray ZH et al. 2023
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Two independent ENCODE CTCF ChIP-seq experiments in GM12878, a human B-lymphoblastoid line, place the signal maximum over the KMT2A exon-11 element at GRCh38 chr11:118,488,694 (ENCFF734CUT, experiment ENCSR000AKB, biological replicates 1 and 2) and chr11:118,488,671 (ENCFF800WUV, experiment ENCSR000DKV, replicates 1-3), with the 147-bp canonical exon 11 reading 113.27x and 135.25x the respective file's genome-wide mean, and ENCODE RAD21 in the same line (ENCFF571ZJJ, ENCSR000EAC, replicates 1 and 2) reading 35.08x with its maximum at chr11:118,488,664; in the same three files the canonical H19/IGF2 imprinting control region reads 5.01x, 16.18x and 3.54x, the GAPDH promoter 2.62x, 3.52x and 1.40x, a chr2 gene desert 0.97x, 1.01x and 1.21x, the weak CTCF cluster at canonical exon 10 4.10x, 5.91x and 2.58x, and the functional intron-9/exon-10 cut point plus or minus 500 bp 1.96x, 1.46x and 1.11x; and ENCODE contains no CTCF ChIP-seq in CD34-positive common myeloid progenitor with GRCh38 fold-change-over-control output (0 results).

    What this evidence cannot show

    GM12878 is an EBV-transformed ADULT B-lymphoblastoid line - not a B-cell precursor, not CD34-positive, not fetal - so this closes the lineage question one step and not all the way to the cell of origin. The values are fold-change-over-control tracks, so the x-genome-mean figures are ratios of a ratio and are not directly comparable with the raw-coverage x-genome-mean figures quoted for HCT116 elsewhere in this beat; only the argmax positions and the within-file rank orderings are comparable across the two. GRCh38 H19 ICR and GAPDH promoter windows were chosen by lifting the hg19 windows used for the HCT116 controls and are approximate to a few kb; the H19 window in particular is 4 kb and contains only a subset of CTCF sites, so 5.01x understates the local peak. No replicate-level percentile null was computed for these ENCODE files.

    Evidence that argues against this (1)
    • A fold-change-over-control track can exaggerate a peak wherever the control is locally low, and this beat separately measured the HCT116 input as reading 2.67x its own genome mean over this exon - an elevated, not depressed, local baseline. An elevated control should DEFLATE a fold-change track, so the 113x and 135x figures are, if anything, conservative for that reason; but the two ENCODE experiments disagree with each other by 20% on the exon mean and by 23 bp on the summit, which is the honest measure of this readout's precision and is larger than the apparent agreement with HCT116 suggests. Q107 beat; data from Canela et al. 2019 2026
Connected to