Oncofusion

start / graph / ana-kmt2a-exon-11-ctcf-anchor-is-one-cohesin-bound-element

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q105-ctcf-adjudication

Run in coordinates: the protective CTCF site and the patient break-prone window are ONE element, it is cohesin-bound, and the 'functional cut point' is 3.75 kb away - two of the graph's three exon-11s were the same exon and the third was not

also called the exon-11 CTCF anchor, resolved in hg38 · coordinate adjudication of the KMT2A BCR CTCF site · DCAL numbering is canonical plus one

Three papers each described something at 'exon 11' of the KMT2A gene using a different numbering scheme, so nobody knew whether they meant one place or three. Putting all of them on the same genomic ruler shows that two of them mean exactly the same 1,000 letters of DNA - a spot where the clamp protein CTCF and its partner ring cohesin both sit - and the third means a spot 3,752 letters away.

The same thing, for a scientist

This runs the check that `ana-kmt2a-bcr-ctcf-site-coordinate-overlap` pre-registered and had not run, and the answer is not the outcome that entry called most likely. Working from the Ensembl canonical *KMT2A* transcript **ENST00000534358** in **both** GRCh38 and GRCh37 [E5]: **(1)** the DCAL/recombinome exon convention is **canonical numbering plus one**, proved by three exact base-pair identities rather than inferred - Meyer et al.'s subregions read 'exon 9 - intron 9 = 1761 bp', 'exon 10 - intron 10 = 679 bp', 'exon 11 - intron 11 - exon 12 - intron 12 and exon 13 = 5026 bp' [E3], and canonical exon8+intron8 = 74+1,687 = **1,761**, canonical exon9+intron9 = 132+547 = **679**, canonical exon10-to-exon12 = chr11:118,484,862-118,489,887 = **5,026** [E5]. **(2)** Atkin et al.'s break-prone window, hg38 chr11:118,488,500-118,488,800, **contains the whole of canonical exon 11** (118,488,614-118,488,760) and nothing else, so Atkin's 'intron 10 / exon 11' is canonical numbering [E2]. **(3)** The element is not a bare CTCF motif. ENCODE clustered ChIP-seq at this locus carries **CTCF chr11:118,488,159-118,489,206 at score 1000, RAD21 118,488,428-118,488,933 at score 1000 and SMC3 118,488,454-118,488,896 at score 1000** - the maximum score for all three - reproduced independently in GRCh37 as CTCF 118,359,222-118,359,582 (1000), RAD21 118,359,253-118,359,515 (1000), SMC3 118,359,255-118,359,540 (953) [E6]. It is a **cohesin-co-occupied CTCF site**: precisely the CTCF+RAD21 class that Canela et al. report as 48-fold enriched for double-strand breaks, 62-fold with TOP2B [E9]. **(4)** Since Gray et al. located their site by conserved ENCODE occupancy [E1], and the only maximum-score conserved CTCF cluster in the BCR spans canonical exon 11, **Gray's 'exon 11' is canonical exon 11 too** - so Gray's protective site and Atkin's break-prone window are the same DNA, 0 bp apart. **(5)** But the DCAL 'functional cut point', the intron-10/exon-11 border [E3], is canonical intron-9/exon-10 = chr11:118,484,861|118,484,862, which is **3,752 bp centromeric** of canonical exon 11. **The graph's claim that the CTCF site sits at the functional cut point is false and was manufactured by two different exon-11s.** What survives is weaker and still real: the anchor sits inside DCAL subregion C, the 39.5%-of-breaks subregion, and at **70.4%** along the Broeker BCR - the **telomeric** half, which carries 75% of therapy-related and 26% of de novo breakpoints [E4]. And the 8.3 kb everyone quotes is a **BamHI restriction fragment** from 1996 [E4], not an exon span, which is why no exon convention reproduces it: canonical exons 8-14 are 8,897 bp.

Go deeper
The table the corpus did not have

Ensembl canonical transcript ENST00000534358 (KMT2A-213, 36 exons, protein-coding), gene ENSG00000118058, plus strand, chr11 5. GRCh38 gene span 118,436,456-118,526,832; GRCh37 118,307,205-118,397,539.

canonical exonGRCh38GRCh37DCAL name (canonical+1)
7118,481,715-118,482,092118,352,430-118,352,807exon 8
8118,482,422-118,482,495118,353,137-118,353,210exon 9
9118,484,183-118,484,314118,354,898-118,355,029exon 10
10118,484,862-118,484,975118,355,577-118,355,690exon 11
11118,488,614-118,488,760118,359,329-118,359,475exon 12
12118,489,792-118,489,887118,360,507-118,360,602exon 13
13118,490,129-118,490,249118,360,844-118,360,964exon 14
14118,491,196-118,491,318118,361,911-118,362,033exon 15

Introns that matter: canonical intron 9 = 118,484,315-118,484,861 (547 bp); canonical intron 10 = 118,484,976-118,488,613 (3,638 bp). Canonical intron 10 is DCAL intron 11 - the interval Meyer et al. report as conferring worse outcome than upstream breakpoints 3.

Every coordinate in this table was read from a live Ensembl REST call on 2026-09-22, in both assemblies separately, not lifted over. 5

How the DCAL convention was pinned, and why arithmetic beats assertion here

The dangerous way to reconcile exon conventions is to assume an offset and check one landmark. Three independent lengths, published to the base pair, make that unnecessary 3:

Meyer et al. subregion, verbatimlengthcanonical equivalentcanonical length
(A) exon 9 - intron 91761 bpexon 8 + intron 874 + 1,687 = 1,761
(B) exon 10 - intron 10679 bpexon 9 + intron 9132 + 547 = 679
(C) exon 11 - intron 11 - exon 12 - intron 12 and exon 135026 bpexon 10 start to exon 12 end118,489,887 - 118,484,862 + 1 = 5,026

Three exact matches, including one five-feature composite. DCAL = canonical + 1, without residue.

The offset is not a curiosity. It means the two things this graph calls 'the intron-10/exon-11 boundary' are different boundaries 3,752 bp apart, and the graph joined them.

What is actually at the element: a maximum-score CTCF/RAD21/SMC3 anchor

UCSC ENCODE clustered transcription-factor ChIP-seq, queried over chr11:118,480,000-118,494,000 in GRCh38 and chr11:118,350,000-118,365,000 in GRCh37 6:

factorGRCh38 intervalscoreGRCh37 intervalscore
CTCF118,488,159-118,489,2061000118,359,222-118,359,5821000
RAD21118,488,428-118,488,9331000118,359,253-118,359,5151000
SMC3118,488,454-118,488,8961000118,359,255-118,359,540953
TRIM22118,488,579-118,488,7541000--
STAT1118,488,455-118,488,819607--
RB1118,488,459-118,488,869553--
BHLHE40118,488,544-118,488,886533--
ARID2118,488,451-118,488,791518--

There is a second, much weaker CTCF site in the BCR and it is the one a DCAL reading of 'exon 11' would land near: CTCF 118,485,122-118,486,000 at score 514 with RAD21 118,485,198-118,485,662 at score 196 (GRCh37: CTCF 118,355,972-118,356,402 at 394, RAD21 118,356,078-118,356,268 at 160) 6. It sits 147 bp telomeric of canonical exon 10. Two CTCF sites, not one, and the corpus held them as a single element.

Why the cohesin call is the load-bearing new fact. genome-kmt2a-exon-11-ctcf-site lists 'whether cohesin (RAD21/SMC3) is bound - no cohesin ChIP was done at it' among the things not known about this element. ENCODE answers it at maximum score in two assemblies. That moves the element from 'a CTCF site' into the exact category Canela et al. quantified: CTCF+RAD21 co-occupied sites are 48-fold enriched for DSBs over matched random regions, 62-fold when TOP2B is also present 9. Atkin et al. reached the same place from the patient side, reporting 404 CTCF, 517 RAD21 and 256 SMC3 sites preferentially break-prone in KMT2A-rearranged individuals 2.

The three (four) overlap answers, in base pairs

1. Gray's protective CTCF site vs Atkin's break-prone window: SAME ELEMENT, 0 bp apart. Atkin's hg38 chr11:118,488,500-118,488,800 (301 bp) lies wholly inside the ENCODE CTCF cluster 118,488,159-118,489,206 (1,048 bp), is centred on the RAD21 and SMC3 footprints, and contains all 147 bp of canonical exon 11 256. Nothing needs converting: the window is the cohesin footprint of the anchor.

2. Gray's exon 11 = canonical exon 11. This cannot be settled from Gray's prose, which gives no coordinates, but it can be settled from the data Gray used: the site was identified as 'a highly conserved occupancy for CTCF at exon 11 within the BCR' across multiple ENCODE cell lines and tissues 1, and the only maximum-score CTCF cluster in the BCR spans canonical exon 11 6. The residual possibility is stated in caveats and there is a named way to close it.

3. The DCAL functional cut point: 3,752 bp away, and the graph's convergence claim fails. DCAL intron-10/exon-11 = canonical intron-9/exon-10 = chr11:118,484,861|118,484,862 35. Distance to canonical exon 11 start = 3,752 bp; to the CTCF cluster's centromeric edge = 3,297 bp. mechanism-h3k9me1-2-ctcf-balance-gates-kmt2a-rearrangement and genome-kmt2a-exon-11-ctcf-site both assert that the CTCF site is at the BCR's functional cut point. It is not. This is the failure mode ana-kmt2a-bcr-ctcf-site-coordinate-overlap itself listed - 'CONTAINMENT REPORTED AS COINCIDENCE' - and it entered by a route that entry did not list: two conventions both saying 'exon 11'.

4. What survives, and it is not nothing.

  • The anchor lies inside DCAL subregion C (canonical exon 10 - exon 12, chr11:118,484,862-118,489,887), the subregion carrying 39.5% of 3,401 characterised breakpoints 3.
  • The CTCF cluster's centromeric edge (118,488,159) lies 455 bp inside the distal end of canonical intron 10 = DCAL intron 11, the poor-outcome breakpoint interval 3. The interval that predicts worse outcome ends inside the anchor.
  • Position within the Broeker BCR: canonical exons 8-14 span 118,482,422-118,491,318 = 8,897 bp; the CTCF cluster midpoint 118,488,682 sits at 70.4% of it. KMT2A is on the plus strand at 11q23.3 (114.6-121.3 Mb), distal to the chr11 centromere at 51.0-55.8 Mb, so increasing coordinate runs toward the telomere 7. Broeker et al. split the 8.3 kb fragment into a 4.57 kb centromeric half (74% of de novo, 25% of t-AML breakpoints) and a 3.87 kb telomeric half (26% de novo, 75% t-AML, P = 0.02) 4. At 70.4% the anchor is in the telomeric, therapy-related-skewed half - which is the half Gray's doxorubicin mechanism should occupy if it is a t-AML mechanism, and it does. That was recorded as the untested secondary prediction and it passes.
The 8.3 kb is a restriction fragment, which is why no exon convention reproduces it

Every entry in this graph quotes the BCR as 8.3 kb, and the mismatch against exon spans has been read as an exon-numbering problem. It is not. Broeker et al. mapped breakpoints 'within the 8.3-kb BamHI breakpoint cluster region', with a 4.57 kb centromeric and 3.87 kb telomeric half 4 - 8.44 kb of restriction fragment, defined by two BamHI sites, in 1996, before the current transcript annotation existed. Gray et al. inherit the number and relabel it 'the 8.3kb breakpoint cluster region (BCR) spanning exons 8-14', citing Broeker for it 1; Meyer et al. define BCR1 independently as 'between KMT2A intron 7 and KMT2A exon 13' and never say 8.3 kb 3 - that interval is canonical intron 6 to exon 12, 9,649 bp.

So there are three different intervals in circulation, all called the BCR: a 8.44 kb BamHI fragment, a 9,649 bp DCAL BCR1, and an 8,897 bp canonical exon-8-to-exon-14 span. State which, always. No exon relabelling can make 8.3 kb come out, because 8.3 kb was never an exon span.

What could still be wrong, and the one measurement that closes it

The chain has exactly one inferential link: that Gray's 'exon 11' is canonical exon 11 rather than DCAL exon 11 (= canonical exon 10). Everything downstream of finding 1 depends on it, and it rests on ENCODE occupancy strength rather than on a coordinate Gray published.

The definitive route exists and is unread. GSE210480 contains four CTCF ChIP-seq tracks from the Gray study - GSM6431019_siKDM3B_57-CTCF-High_rep1.bw, GSM6431021_siKDM3B_58-CTCF-High_rep2.bw, GSM6431023_siCT-CTCF-High_rep1.bw, GSM6431025_siCT-CTCF-High_rep2.bw - and Gray's ChIP-seq was aligned to hg19 18. Reading the siCT peak summit over chr11:118,354,000-118,362,000 in GRCh37 settles the numbering in one query. What the deposit does not contain is peak calls: the series supplement is a RAW tar plus bigWig/bw coverage only, with no narrowPeak or bed file 8. So this beat's statement is sharper than 'no coordinates were published' - no coordinates were deposited either, and the answer requires re-analysis of coverage, not a lookup.

If the summit is at ~chr11:118,359,400 (GRCh37) this analysis stands as written. If it is at ~chr11:118,355,600, then Gray's site is canonical exon 10 / the weak score-514 cluster, findings 1 and 2 invert, and the protective site is not the patient break-prone window - which would dissolve the strongest version of the contradiction and make it a two-elements story instead.

Facts and figures
scriptnone - this analysis is a set of live database calls and arithmetic, all of them recorded in full in the sections and evidence so that any step can be re-run by hand
statusrun
run date2026-09-22
assembly disciplineEvery coordinate in this entry carries its assembly. GRCh37 and GRCh38 values were fetched from SEPARATE endpoints, never lifted over, in both Ensembl and UCSC. The hg38-minus-hg19 offset for this locus is 129,285 bp at the canonical exon 11 start and is recorded only as a consistency check, never used to generate a coordinate.
result summaryDCAL/recombinome exon numbering equals Ensembl canonical numbering plus one, proved by three exact base-pair identities (1,761 / 679 / 5,026). Atkin's hg38 chr11:118,488,500-118,488,800 break-prone window and Gray's conserved exon-11 CTCF site are the SAME element - 0 bp apart, the window lying wholly inside an ENCODE CTCF cluster at chr11:118,488,159-118,489,206 that is co-occupied by RAD21 and SMC3 at maximum score in two assemblies, and containing all 147 bp of canonical exon 11. The DCAL 'functional cut point' is canonical intron-9/exon-10 at chr11:118,484,861|118,484,862, which is 3,752 bp away, so this graph's claim that the CTCF site sits at the functional cut point is FALSE. Surviving positional facts: the anchor lies inside DCAL subregion C (39.5% of 3,401 breakpoints), its centromeric edge lies 455 bp inside the distal end of the poor-outcome interval (DCAL intron 11 = canonical intron 10), and it sits at 70.4% along the Broeker BCR, in the telomeric half that carries 75% of therapy-related breakpoints - confirming the untested secondary prediction that a doxorubicin-driven mechanism should map to the therapy-related half. The 8.3 kb BCR is a BamHI restriction fragment from 1996, not an exon span, which is why no exon convention reproduces it.
inputs
[
  "Ensembl REST lookup/symbol/homo_sapiens/KMT2A?expand=1, GRCh38 and GRCh37 endpoints",
  "UCSC API getData/track encRegTfbsClustered hg38 chr11:118,480,000-118,494,000",
  "UCSC API getData/track wgEncodeRegTfbsClusteredV3 hg19 chr11:118,350,000-118,365,000",
  "UCSC API getData/track cytoBand hg38 chr11",
  "PMC10591855 full text (PMID 37788669)",
  "PMC10169636 full text (PMID 37019990)",
  "PMID 8634439 abstract",
  "PMID 34405474 abstract plus the hg38 window as stored in gene-ctcf",
  "GEO GSE210480 supplementary filelist.txt"
]
element
{
  "name": "the KMT2A exon-11 CTCF/cohesin anchor",
  "assembly": "GRCh38",
  "ctcf_cluster": "chr11:118,488,159-118,489,206 (1,048 bp), ENCODE clustered ChIP-seq score 1000",
  "rad21_cluster": "chr11:118,488,428-118,488,933, score 1000",
  "smc3_cluster": "chr11:118,488,454-118,488,896, score 1000",
  "canonical_exon_11": "chr11:118,488,614-118,488,760 (147 bp), ENST00000534358",
  "atkin_break_window": "chr11:118,488,500-118,488,800 (301 bp), wholly inside the CTCF cluster, containing all of canonical exon 11",
  "grch37_equivalents": "CTCF chr11:118,359,222-118,359,582 (1000); RAD21 118,359,253-118,359,515 (1000); SMC3 118,359,255-118,359,540 (953); canonical exon 11 118,359,329-118,359,475",
  "second_weaker_ctcf_site_in_the_bcr": "chr11:118,485,122-118,486,000 score 514 with RAD21 118,485,198-118,485,662 score 196 (GRCh37: 118,355,972-118,356,402 score 394 / 118,356,078-118,356,268 score 160) - 147 bp telomeric of canonical exon 10, and the site a DCAL reading of 'exon 11' would land near"
}
exon convention table
[
  {
    "convention": "Ensembl canonical ENST00000534358",
    "used_by": "this analysis; Atkin et al. (PMID 34405474); Gray et al. (PMID 37788669) by inference from ENCODE occupancy",
    "offset_vs_canonical": 0,
    "exon_11_is": "chr11:118,488,614-118,488,760 (GRCh38)"
  },
  {
    "convention": "DCAL / KMT2A recombinome",
    "used_by": "Meyer et al. (PMID 37019990); breakpoint-kmt2a-major-bcr",
    "offset_vs_canonical": 1,
    "exon_11_is": "canonical exon 10 = chr11:118,484,862-118,484,975 (GRCh38)"
  },
  {
    "convention": "historic 'exons 5-11' BCR labelling",
    "used_by": "breakpoint-kmt2a-bcr-architecture",
    "offset_vs_canonical": "approximately -3 at the BCR ends; not pinned by this analysis because no published base-pair length was available to test against",
    "exon_11_is": "not determined"
  },
  {
    "convention": "mouse Kmt2a exons",
    "used_by": "Canela et al. (PMID 28735753), which reports etoposide DSB peaks 'near exon 10' and 'at exon 12' in activated MOUSE B cells",
    "offset_vs_canonical": "not determined - a different species' annotation, not an offset",
    "exon_11_is": "not applicable"
  }
]
distances bp
{
  "gray_ctcf_site_to_atkin_break_window": 0,
  "ctcf_cluster_to_canonical_exon_11": "containment - the exon lies 455 bp inside the cluster's centromeric edge",
  "dcal_functional_cut_point_to_canonical_exon_11_start": 3752,
  "dcal_functional_cut_point_to_ctcf_cluster_centromeric_edge": 3297,
  "distal_end_of_poor_outcome_interval_into_ctcf_cluster": 455,
  "position_along_broeker_bcr_pct": 70.4
}
interval lengths all called the bcr
{
  "broeker_bamhi_fragment": "8.3 kb as named, 4.57 + 3.87 = 8.44 kb as decomposed (PMID 8634439)",
  "meyer_bcr1_intron7_to_exon13_dcal": "canonical intron 6 to exon 12 = chr11:118,480,239-118,489,887 = 9,649 bp",
  "gray_exons_8_to_14_read_as_canonical": "chr11:118,482,422-118,491,318 = 8,897 bp",
  "meyer_subregions_a_plus_b_plus_c": "chr11:118,482,422-118,489,887 = 7,466 bp"
}
what this changes in entries this beat does not own
[
  "genome-kmt2a-exon-11-ctcf-site: data.what_is_NOT_known_about_this_element lists 'its hg38 coordinates', 'whether cohesin (RAD21/SMC3) is bound', 'whether it is the same element as the hg38 chr11:118,488,500-800 break-prone region' and 'which HALF of the BCR the site sits in' - all four are now answered. Its claim that the element coincides with the functional cut point is contradicted and should be replaced by the 3,752 bp distance.",
  "mechanism-h3k9me1-2-ctcf-balance-gates-kmt2a-rearrangement: the sentence 'THE CTCF SITE IS AT THE FUNCTIONAL CUT POINT OF THE BCR' should become 'the CTCF site is 3,752 bp telomeric of the functional cut point, inside subregion C, and inside the telomeric t-AML-skewed half'.",
  "ana-kmt2a-bcr-ctcf-site-coordinate-overlap: status 'planned' should become 'run', and its pre-registered most-likely outcome ('scattered across subregion C') did not occur.",
  "breakpoint-kmt2a-major-bcr: data.exon_convention is null and should record DCAL = Ensembl canonical + 1, with the three base-pair identities that establish it."
]
failure modes encountered
[
  "An earlier draft asserted that Gray's 10 kb bin-to-exon labels exactly matched canonical numbering. They do not: canonical exons 6 and 7 fall in the bin Gray assigns to the central region, and the labels overlap at their boundary exons, so the bins cannot discriminate a one-exon offset. The claim was withdrawn and replaced by the ENCODE argument.",
  "Europe PMC's citedOrder field is NOT the reference number printed in the paper. For PMID 37788669 it gives Gothe as 2 and Atkin as 57; the article's own numbered list, read from the PMC JATS ref ids, gives Atkin as 56 and Gothe as 57.",
  "Two NCBI esearch calls returned count=null with querytranslation=null - an API rate-limit error object, not a zero. Re-run with pacing, they returned 1 and 2. A parsed zero requires a non-empty querytranslation."
]
How this could bite us back
  • This is a coordinate result and nothing in it is an intervention claim. It touches doxorubicin only because the element it locates sits in the half of the BCR that therapy-related breakpoints prefer. That is a statement about where breaks land in a 8.4 kb interval in 8 t-AML patients from 1996, not a statement about anthracycline risk in any individual, and it must not be read as one. Doxorubicin is in the infant-ALL backbone of this graph for good reasons and a locus-level positional correlation does not touch them.
  • If the anchor is a cohesin-co-occupied loop anchor, then any intervention that removes cohesin or CTCF to protect the locus is proposing to disable a genome-organising complex. This graph's own DepMap scan puts CTCF at a pan-essentiality of -1.827 across all lines; RAD21 and SMC3 are similar. The protective direction of Gray's result, taken as a drug hypothesis, points at INCREASING occupancy, not decreasing it, and the tractable end is the methyltransferase, not the anchor.
  • The strongest claim here - two of three descriptions are the same element - makes the contradiction between this graph's two CTCF models sharper, not softer. Anyone using this entry to argue that the dispute is resolved has read it backwards: it establishes that both sides are talking about the same DNA, which is what makes the directional disagreement real.
Still open
  1. Where is the summit of Gray et al.'s own CTCF ChIP-seq peak in GRCh37 - at ~118,359,400 (canonical exon 11) or ~118,355,600 (canonical exon 10)? This is the one inferential link in the chain and it is answerable from GSE210480's siCT CTCF bigWigs, which nobody has read.
  2. Is TOP2B bound at this anchor in any cell type? ENCODE's clustered set contains no TOP2B at this locus, so the 62-fold CTCF+RAD21+TOP2B fragility class cannot be asserted for this element, only the 48-fold CTCF+RAD21 class.
  3. Where do Broeker et al.'s two BamHI sites actually fall in GRCh38? Without them the centromeric/telomeric half assignment is a proportional proxy on an exon span, and the whole de novo versus therapy-related skew in this graph is anchored on that fragment.
  4. Does the second, weaker CTCF site at chr11:118,485,122-118,486,000 do anything? It sits 147 bp from canonical exon 10 and 260 bp telomeric of the functional cut point, and no paper on either side of this dispute mentions it.
  5. What is the historic 'exons 5-11' convention's transcript basis, and does it have a published base-pair length anywhere that would let it be pinned the way DCAL was?
Every source behind this entry (9)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Gray et al. state that both H3K9me1 and H3K9me2 increase upon KDM3B depletion across the KMT2A gene body, 'particularly H3K9me1 within the 8.3kb breakpoint cluster region (BCR) spanning exons 8-14, which is enriched for KMT2A rearrangements', citing Broeker et al. for that definition; that 'upon evaluating multiple cell lines and tissues from ENCODE, we observed a highly conserved occupancy for CTCF at exon 11 within the BCR of KMT2A, directly overlapping with KDM3B binding'; that their genome-wide 10 kb bins place 'the central region of KMT2A gene (containing exons 8-14: chr11 bins 118.35-118.36 and 118.36-118.37 Mbp)'; and that their ChIP-seq reads were 'aligned against the human hg19 reference genome using BWA' with peaks called by HOMER. No base-pair coordinate for the CTCF site is given anywhere in the paper.

    Epigenetic balance ensures mechanistic control of MLL amplification and rearrangement. · 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 · Cell 2023
    PMID 37788669DOI
    What this evidence cannot show

    Read in full text from PMC10591855 on 2026-09-22, body and figure legends. The exon-numbering convention is never stated by the paper; that it is the canonical one is this analysis's inference from ENCODE occupancy [E6], not the paper's own statement. Gray's 10 kb bin-to-exon labels overlap at their boundary exons (exons 1-8 / 8-14 / 14-36) and are therefore too coarse to discriminate a one-exon offset - an earlier draft of this analysis claimed they were an exact match and that claim was withdrawn after checking canonical exons 6 and 7 against the bin edges.

    Evidence that argues against this (1)
    • The same paper's own citation for the doxorubicin arm reports that CTCF genomic binding is largely unchanged under stress even as the protein pool falls, with the loss confined to an SC-35 nuclear-speckle species - so a coordinate for the site does not by itself establish that occupancy there moves with total protein. Lehman BJ et al. 2021
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Atkin et al., using a low-input adaptation of DSBCapture on normal (non-blast) cells from AML patients in remission, found DNA breakage at the KMT2A intron 10 / exon 11 region, given as hg38 chr11:118,488,500-800, to be 2- to 10-fold higher in each KMT2A-rearranged patient than in each non-fusion patient (group average 3.2-fold, significant), and identified 304 TOP2-sensitive, 404 CTCF, 517 RAD21 and 256 SMC3 binding sites preferentially break-prone in the fusion-driven patients, with low-dose etoposide ex vivo raising breakage at these regions in the KMT2A-rearranged patients only and no difference in genome-wide median breakage between samples.

    PMID 34405474DOI
    What this evidence cannot show

    n = 3 per arm, adult AML, myeloid. Abstract and metadata read directly on 2026-09-22; the hg38 window coordinate is taken as recorded in this graph's `gene-ctcf` from the paper and was not re-read from the figure by this beat. The window is 301 bp of scored interval, not a break position. That it contains canonical exon 11 exactly is consistent with the paper's own 'intron 10 / exon 11' description and is the basis for assigning it to canonical numbering.

    Evidence that argues against this (1)
    • The same paper's partner-side result argues against the two-sided loop-anchor reading: none of the three patients' own partner genes appeared among the 34 regions carrying all four break-prone features, and only the 3' end of AFDN - not its breakpoint region - carried a break-prone CTCF site. Atkin ND et al. 2021
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Meyer et al. state verbatim: 'The major breakpoint cluster region (BCR1) can be mapped between KMT2A intron 7 and KMT2A exon 13, and the minor BCR (BCR2) between intron 20 and exon 24. The majority of patients (n = 3336; 98%) showed breakpoints in BCR1'; 'we subdivided the KMT2A BCR1 into three subregions: (A) exon 9 - intron 9 = 1761 bp; (B) exon 10 - intron 10 = 679 bp; (C) exon 11 - intron 11 - exon 12 - intron 12 and exon 13 = 5026 bp. The functional cut is between regions A-B and C'; 'The observed mean distribution (MD) for these three KMT2A breakpoint regions was A = 37.2%, B = 19.8% and C = 39.5% for all 3401 patients'; and that 'the outcome of leukemia patients with breakpoints in KMT2A intron 11 was worse compared to those patients with upstream breakpoints. The cut point was exactly at the borderline between intron 10 and exon 11.'

    The KMT2A recombinome of acute leukemias in 2023. · Meyer C, Larghero P, Almeida Lopes B, Burmeister T, Groger D, Sutton R, Venn NC, Cazzaniga G, Corral Abascal L, Tsaur G, et al., Marschalek R · Leukemia 2023
    PMID 37019990DOI
    What this evidence cannot show

    Read in full text from PMC10169636 on 2026-09-22; the three subregion definitions and the functional-cut sentence were extracted verbatim rather than taken from this graph's stored summary. The paper does not state which transcript its exon numbering uses. The 'canonical + 1' mapping here is an arithmetic inference from three published base-pair lengths, all three of which match canonical features exactly; it is not a statement the paper makes. The poor-outcome claim about intron 11 is attributed by Meyer et al. to their reference 23, which this beat did not read.

    Evidence that argues against this (1)
    • Meyer et al.'s own definition of the interval is a diagnostic-PCR convenience rather than a mechanistic one, and the same paper records that KMT2A intron 9 (their numbering) contains four Alu elements of which three are transcriptionally active - so subregion break density may reflect repeat content rather than any chromatin element, and the 39.5% of subregion C is spread over 5,026 bp against 679 bp for subregion B. Meyer C et al. 2023
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Broeker et al. mapped breakpoints 'within the 8.3-kb BamHI breakpoint cluster region' in 31 de novo ALL/AML patients and 8 t-AML patients: 23 of 31 de novo breakpoints mapped to the centromeric half (4.57 kb) and 8 to the telomeric half (3.87 kb), whereas only 2 of 8 t-AML breakpoints were centromeric and 6 telomeric (P = 0.02 for the de novo distribution); a low- or weak-affinity scaffold attachment region maps just centromeric to the BCR and a high-affinity SAR within its telomeric half, with one topoisomerase II consensus site adjacent to the telomeric SAR and six within it; they conclude that 74% of de novo and 25% of t-AML breakpoints lie centromeric and 26% of de novo and 75% of t-AML breakpoints lie telomeric, and that the mechanisms leading to translocation may differ between de novo leukaemia and t-AML.

    PMID 8634439
    What this evidence cannot show

    Abstract read on 2026-09-22; no DOI is registered for this 1996 Blood article and the citation carries the PMID and PubMed URL instead. Eight t-AML patients is a very small denominator for a 75/25 split. The BamHI fragment's genomic coordinates are not given in the abstract, so the centromeric/telomeric boundary used here (54.1% of the fragment, from 4.57/8.44) is applied to the canonical exon-8-to-exon-14 span as a proxy; the anchor at 70.4% is far enough from 54% that a few hundred base pairs of fragment-boundary uncertainty does not change the assignment, but the assignment is a proxy.

    Evidence that argues against this (1)
    • A published model for the same locus argues that local topoisomerase II concentration is NOT what determines breakpoint clustering: Cowell et al. report that most etoposide-induced MLL breaks depend on TOP2B, but that TOP2 alpha and beta occupancy and etoposide-induced cleavage data 'suggest factors other than local topoisomerase II concentration determine specific clustering of MLL translocation breakpoints in t-AML' - which undercuts reading the telomeric SAR/TOP2-site enrichment as the cause of the t-AML skew. Cowell IG et al. 2012
  5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    Ensembl REST returns HGNC symbol KMT2A as ENSG00000118058, 'lysine methyltransferase 2A [Source:HGNC Symbol;Acc:HGNC:7132]', chromosome 11, plus strand, GRCh38 118,436,456-118,526,832, with canonical transcript ENST00000534358 (KMT2A-213, protein_coding, 36 exons, 118,436,492-118,526,832); the GRCh37 endpoint returns the same gene and transcript at 118,307,205-118,397,539; and canonical exon coordinates in GRCh38 are exon 7 118,481,715-118,482,092, exon 8 118,482,422-118,482,495, exon 9 118,484,183-118,484,314, exon 10 118,484,862-118,484,975, exon 11 118,488,614-118,488,760, exon 12 118,489,792-118,489,887, exon 13 118,490,129-118,490,249, exon 14 118,491,196-118,491,318, with GRCh37 equivalents exon 8 118,353,137-118,353,210 through exon 14 118,361,911-118,362,033.

    What this evidence cannot show

    Resolved by SYMBOL, not by a typed Ensembl id, and the returned description and HGNC accession were read back to confirm the record is KMT2A and not a neighbour - this graph has had an Ensembl id resolve to the wrong gene. Exon numbering is transcript-specific: ENST00000534358 is Ensembl's canonical choice and a different transcript would renumber these exons. The GRCh37 and GRCh38 coordinates were fetched from separate endpoints rather than lifted over, so a liftover error is excluded, but the two assemblies' annotations are independently curated and could differ in exon boundaries; here exons 8-14 have identical lengths in both.

    Evidence that argues against this (1)
    • A single canonical transcript is a convention, not a fact about the locus: UniProt models KMT2A as a full-length protein plus two taspase1 cleavage products (N320 and C180) filed under separate identifiers, and this graph has recorded that filtering on one identifier discards the fragment-scoped records. The same caution applies to exon numbering - the convention chosen here is Ensembl's, and DCAL's differs by one. UniProt Consortium 2026
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    UCSC Genome Browser API over ENCODE clustered transcription-factor ChIP-seq returns, at the KMT2A breakpoint cluster region in GRCh38 (track encRegTfbsClustered, dataTime 2019-05-16): CTCF chr11:118,488,159-118,489,206 score 1000, RAD21 118,488,428-118,488,933 score 1000, SMC3 118,488,454-118,488,896 score 1000, TRIM22 118,488,579-118,488,754 score 1000, STAT1 118,488,455-118,488,819 score 607, RB1 118,488,459-118,488,869 score 553, BHLHE40 118,488,544-118,488,886 score 533, ARID2 118,488,451-118,488,791 score 518, plus a second, weaker CTCF cluster at 118,485,122-118,486,000 score 514 with RAD21 118,485,198-118,485,662 score 196; and in GRCh37 (track wgEncodeRegTfbsClusteredV3): CTCF chr11:118,359,222-118,359,582 score 1000, RAD21 118,359,253-118,359,515 score 1000, SMC3 118,359,255-118,359,540 score 953, with the weaker pair at CTCF 118,355,972-118,356,402 score 394 and RAD21 118,356,078-118,356,268 score 160.

    What this evidence cannot show

    These are CLUSTERS across many ENCODE cell lines, not a measurement in any one cell type, and the score is a cluster-level summary (capped at 1000) rather than an occupancy level - a score of 1000 means reproducibly and strongly bound across the contributing experiments, not that CTCF is bound in a haematopoietic progenitor. No leukaemia or CD34+ experiment is identified here. The two tracks are different ENCODE releases (ENCODE 3 clusters for hg38, V3 for hg19) with different interval boundaries, so their agreement is a reproducibility check on the site's existence and approximate extent, not on its edges. TOP2B is not present in either cluster set at this locus, so the 62-fold class in [E9] cannot be asserted for this element from this source.

    Evidence that argues against this (1)
    • ENCODE-style clustered occupancy is exactly the quantity that cannot settle the dispute this element sits in, because a hotspot map built on CTCF-bound sites in the cells assayed cannot distinguish 'CTCF site' from 'CTCF-occupied site' in the condition of interest; and the one paper that measured CTCF genomic binding under the relevant stress reports it largely unchanged while the protein pool falls. Lehman BJ et al. 2021
  7. E7established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    UCSC cytoBand for GRCh38 chr11 places the acrocentric/centromeric bands p11.11 at 51,000,000-53,400,000 and q11 at 53,400,000-55,800,000, the band q23.3 at 114,600,000-121,300,000, and the terminal band q25 at 130,900,000-135,086,622 - so a KMT2A coordinate near 118.49 Mb lies in 11q23.3, distal to the centromere, and increasing coordinate runs toward the q-arm telomere.

    What this evidence cannot show

    This establishes only the direction of the coordinate axis relative to centromere and telomere. It does not establish where Broeker et al.'s BamHI fragment boundaries fall, which is the quantity the centromeric/telomeric half assignment actually needs.

    Evidence that argues against this (1)
    • Cytoband boundaries are staining-based approximations reassigned between assemblies and carry no base-pair authority; nothing in this claim depends on the band edges, only on the centromere lying below and the telomere above the locus. UCSC Genome Browser 2026
  8. E8established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    The GEO supplementary file list for GSE210480, the series accompanying PMID 37788669, contains exactly four CTCF tracks - GSM6431019_siKDM3B_57-CTCF-High_rep1.bw, GSM6431021_siKDM3B_58-CTCF-High_rep2.bw, GSM6431023_siCT-CTCF-High_rep1.bw and GSM6431025_siCT-CTCF-High_rep2.bw, plus later re-deposits GSM7594696 and GSM7594698 with identical file sizes - together with H3K9me1, H3K9me2 and H3K9me3 bigWigs across control, siKDM3B, siG9A and double-knockdown arms; the series supplement itself offers only GSE210480_RAW.tar and this filelist, and every processed CTCF file is a bigWig or bw COVERAGE track. No narrowPeak, broadPeak or BED peak-call file is deposited for CTCF or for any other target in the series.

    What this evidence cannot show

    This is the file listing only; no file was downloaded and no track was read, so this claim bounds what is available for lookup and says nothing about the data's content. Peak calls may exist inside GSE210480_RAW.tar (1.4 GB) or in a companion series not identified here; the paper states HOMER was used for peak calling [E1], so calls were made but this beat could not confirm they were deposited. 'CTCF-High' in the sample names is unexplained by the file list and its meaning was not established.

    Evidence that argues against this (1)
    • The same series is the subject of an existing finding in this graph that its eight deposited H3K9me3 samples were assayed and never reported, so an argument that this deposit is incomplete or under-analysed is already on record here and cuts both ways: absence from a file listing is weak evidence about what was measured. NCBI Gene Expression Omnibus 2024
  9. E9strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Canela et al. report that evolutionarily conserved chromosome loop anchors bound by CTCF and cohesin are vulnerable to TOP2B-mediated double-strand breaks; that relative to sites occupied by CTCF alone, regions also bound by RAD21 and TOP2B were more likely to form DSBs; that 'sites that were co-occupied by both CTCF and RAD21 were 48-fold enriched in DSBs relative to a similar number of randomly located regions' and '62-fold' when TOP2B was also present; that etoposide induced 'a small broad DSB peak in MLL near exon 10 and a larger sharp peak at exon 12, both sites overlapping with the highly conserved breakpoint cluster region' in activated mouse B cells; and that DSBs at loop anchors are largely transcription-, replication- and cell-type-independent, form continuously through interphase, and frequently occur at breakpoint clusters commonly translocated in cancer.

    Genome Organization Drives Chromosome Fragility. · Canela A, Maman Y, Jung S, Wong N, Callen E, Day A, Kieffer-Kwon KR, Pekowska A, Zhang H, Rao SSP, Huang SC, Mckinnon PJ, Aplan PD, Pommier Y, Aiden EL, Casellas R, Nussenzweig A · Cell 2017
    PMID 28735753DOI
    What this evidence cannot show

    Read in full text from PMC6133249 on 2026-09-22. The MLL peaks quoted are in MOUSE Kmt2a in activated mouse B cells, and 'exon 10' and 'exon 12' there are a FOURTH exon-numbering frame in this dispute - mouse exons, which this beat did not map to human canonical coordinates. The 48-fold and 62-fold enrichments are genome-wide class statements about CTCF/RAD21-co-occupied sites, not measurements at the human KMT2A element, and nothing in this paper measures the human exon-11 anchor.

    Evidence that argues against this (1)
    • The transcription-independence claim is contradicted by the other founding paper of the same model, which reports TOP2 DSBs enriched at loop anchors with HIGH transcriptional activity, break frequency correlating with transcriptional output and directionality, and MLL fusion formation dependent on transcriptional elongation as well as on both TOP2 isoforms. Gothe HJ et al. 2019
Connected to