The cohesin-degron break map was run in the one human cell line where the KMT2A element does not break - so premise (i) is not underpowered, it is unaddressed
also called the HCT116 mismatch · why the RAD21 degron cannot answer this · no break, no experiment
To ask whether cohesin causes a DNA break, you need a cell where that break happens. The one experiment that removed cohesin used a bowel-cancer cell line in which the KMT2A spot shows no drug-induced break at all - so the experiment could not have detected an effect either way.
The same thing, for a scientist
The only loss-of-cohesin DSB map in human cells is Canela 2019's `END_seq_HCT116__RAD21_AID_0_5h` (GSM3714904). In HCT116 the *KMT2A* exon-11 CTCF/cohesin element reaches the **94.004th** percentile of chr11 147-bp bins with 50 uM etoposide for 30 min, while the **same line's no-etoposide control reaches the 92.893rd** percentile - a depth-normalised etoposide/untreated ratio of **1.74x / 1.48x = 1.18**, against the paper's own peak-calling threshold of **>10-fold over the matched non-treated sample** [E1]. **There is no etoposide-induced peak at this element in HCT116 to lose.** In the arms where there is one - Nalm6 99.948th percentile (10.77x / 1.81x = 5.9-fold over its own NT), primary CD34+ marrow 99.658th and 99.796th (7.0-fold and 2.4-fold over DMSO), K562 99.944th (10.5-fold), TK6 99.610th (DMSO reads **0.00x**) - cohesin was never perturbed [E1][E2]. **This is not an n-of-1 problem and no amount of replication of the deposited design would fix it.** The failure is that the perturbation and the phenotype were measured in different cell types, and the graph should stop describing the RAD21-degron arm as 'the loss-of-occupancy experiment in this dispute' and start describing it as 'a loss-of-occupancy experiment in a cell type where the phenotype is absent'.
The mismatch in one table
| cell type | is the element an etoposide DSB hotspot? | was cohesin perturbed? |
|---|---|---|
| Nalm6 (human B-cell-precursor ALL) | yes - 99.948th chr11 percentile, 475 of 918,411 bins above it | no |
| primary CD34+ bone marrow | yes - 99.796th / 99.658th, 2 replicates | no |
| K562 (erythroleukaemia) | yes - 99.944th | no |
| TK6 (lymphoblastoid) | yes - 99.610th; DMSO signal is exactly zero | no |
| MCF7 (breast) | yes - 99.883rd | no |
| HCT116 (colorectal) | no - 94.004th with etoposide, 92.893rd without | yes - the only one |
Every cell in which the element is a top-0.4% break site lacks the perturbation; the only cell with the perturbation lacks the break.
Three candidate reasons, and which of them can be checked
1. Lineage. KMT2A is a haematopoietic master regulator, and Gothe's central claim is that fragility scales with transcriptional output. If KMT2A is lowly transcribed in a colorectal line, the element would not be loaded with elongating polymerase or trapped TOP2. Checkable, and not from these deposits - neither Canela series has an HCT116 transcription track (GSE129527's three nascent-RNA samples are mouse pre-B), so this needs an external HCT116 RNA-seq or Pol II track. It is the single cheapest open question arising from this beat.
2. Cell-cycle state. HCT116 was arrested in G1 with 20 uM lovastatin for 24 h before a 4 h auxin pre-treatment and a 30-min etoposide pulse; Nalm6, MCF7, TK6, K562 and CD34+ cells were treated without synchronisation and, in Gothe's case, for 4 h. TOP2A is cell-cycle-regulated and G1 arrest removes most of it. Partly checkable within the deposit - GSE129524 contains a 0.5 to 12.5 h etoposide time course, but only in mouse pre-B cells.
3. Pulse length. 30 min in HCT116 against 4 h in Gothe's arms. Canela 2019's own model separates trapping from conversion and makes conversion time-dependent, so a 30-min pulse may report trapping and not breaks. Partly checkable - and note the HCT116 arm was run only with the ExoVII+ExoT variant, which reports all TOP2 lesions rather than DSBs alone, so if anything it should be the more sensitive readout, which makes this explanation weaker.
What survives all three. The internal comparison is protocol-matched: GSM3714903 (WT + etoposide) versus GSM3714905 (WT, no etoposide) are the same line, same arrest, same processing, and the element does not rise. Whatever the reason, the phenotype is absent in that experiment's own hands, and a degron cannot remove an absent phenotype.
What this changes in the graph
hyp-ctcf-occupancy-splits-break-formation-from-rearrangement-fate and concept-ctcf-at-kmt2a-fragile-and-guarded-is-one-element both treat premise (i) as plausible but unreplicated. The sharper status is: premise (i) has a strong correlational leg in the right cell types and zero directional evidence in any of them, and the one directional experiment that exists is not a weak test of it - it is not a test of it.
That matters for how the triad is resolved, because the triad's whole force is that all three premises cannot hold. If (i) is unaddressed rather than supported, the cheapest way to break the triad is no longer to attack (iii); it is to run (i) in a haematopoietic cell, which is a single well-defined experiment - see exp-cohesin-degron-break-map-in-a-human-b-precursor-at-kmt2a.
| script | pybigtools 0.3.0 under CPython 3.12.13, HTTP range reads; chr11 147-bp-bin nulls computed per file |
|---|---|
| status | run |
| result summary | The element is a 99.6th-99.95th-percentile chr11 etoposide DSB site in Nalm6, CD34+, K562, TK6 and MCF7, and a 94.0th-percentile site in HCT116 whose own untreated control sits at the 92.9th percentile. Cohesin was perturbed only in HCT116. The perturbation and the phenotype are in disjoint cell types, so premise (i) is unaddressed rather than underpowered. |
| verdict | premise-(i)-unaddressed-by-a-cell-type-mismatch |
inputs
[ "GSE129524 GSM3714903 / GSM3714904 / GSM3714905 (HCT116 END-seq)", "GSE99194 GSM2635568/9 (MCF7), GSM2635574/5 (Nalm6)", "GSE121740 GSM3444985 (TK6 ETO), GSM3444989 (K562 ETO), GSM3687236-9 (CD34+)", "GSE99195 and GSE121740 sample listings for the perturbation inventory" ]
- This entry makes the RAD21-degron result look uninformative, which is a licence to ignore it. It is not: the same series' mouse Smc3f/f+CRE and Waplf/f+CRE MEF arms are genuinely directional, and a beat that dismisses the human arm should still read those two before concluding that no directional cohesin data exists anywhere.
- Is KMT2A transcribed in lovastatin-arrested G1 HCT116? If it is silent, the HCT116 negative is a transcription fact rather than a lineage fact, and Gothe's model predicts it exactly. No transcription track for HCT116 exists in either Canela deposit.
- Does the published PMID 31202577 show an HCT116 MLL panel that contradicts this readout? PMC8170508 is open access and was not read for this entry.
- What is the element's break rank in Gothe's HCT116, HT1080, Cal51 and U2OS arms? They have C-Fusion data but no sBLISS, so the question cannot be answered from any deposit and would require new sBLISS in HCT116 - which would also be the cheapest direct test of explanation 1.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
In Canela 2019's deposited HCT116 END-seq bigWigs, the KMT2A exon-11 element (GRCh37 chr11:118,359,329-118,359,475) reaches the 94.004th percentile of 918,411 chr11 147-bp bins in WT cells given 50 uM etoposide for 30 min after 24 h of 20 uM lovastatin G1 arrest (55,071 bins above it; 1.74x the file's genome mean) and the 92.893rd percentile in the same line without etoposide (1.48x genome mean), giving an etoposide-over-untreated enrichment ratio of 1.18 against the paper's own deposited peak-calling threshold of greater-than-10-fold over the matched non-treated sample.
GEO GSE129524 samples GSM3714903, GSM3714904 and GSM3714905 - deposited bigWig tracks read over HTTP range requests with pybigtools 0.3.0 on 2026-09-22, with the chr11 147-bp-bin null computed from the same files; growth and data-processing protocols quoted from the GSM records · Canela A, Maman Y, Huang SN, Wutz G, Tang W, Zagnoli-Vieira G, Callen E, Wong N, Day A, Peters JM, Caldecott KW, Pommier Y, Nussenzweig A · NCBI GEO (data accompanying Mol Cell 75:252-266.e8) 2019What this evidence cannot show
n = 1 per condition. The etoposide-over-untreated ratio is computed from depth-normalised region means, not from the paper's own fold-enrichment pipeline, which used MACS 1.4.3 with the NT library as control - so 1.18 is this beat's arithmetic and not the authors' number, and it could differ from what MACS would report. The comparison being protocol-matched (same line, same arrest, same processing) is the load-bearing part and does not depend on the exact ratio. The percentile null is chr11-only and includes a pericentromeric artefact region that dominates the top bins of every END-seq track here, which makes the percentile conservative.
Evidence that argues against this (1)
- The most serious challenge to this entry's conclusion is that the published paper may show an HCT116 panel at MLL that this readout does not reproduce - the full text of PMID 31202577 was NOT read for this entry, only the deposit. If Canela 2019 reports a cohesin-dependent END-seq peak at MLL in HCT116 in a figure, then either the deposited bigWigs are not the tracks behind that figure, or the reading here is wrong, and the discrepancy would itself be the finding. PMC8170508 is open access and this check is one fetch away; it is the highest-priority follow-up in this beat. Canela A et al. 2019
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Cohesin was perturbed in none of the human cell types in which the KMT2A exon-11 element is an etoposide DSB hotspot: GSE99194 (Canela 2017) contains Nalm6 and MCF7 END-seq with no cohesin arm and its ChIP series GSE99195 has no SMC3, SMC1, NIPBL or WAPL sample and no human sample at all; and GSE121740 (Gothe 2019) contains TK6, K562 and primary CD34+ sBLISS with no CTCF, RAD21, SMC3, SMC1, NIPBL or WAPL perturbation sample, consistent with the paper's full text, whose only degron, siRNA and shRNA arms target TOP2A, TOP2B and CtIP.
NCBI GEO GSE99194, GSE99195 and GSE121740 full sample listings, retrieved 2026-09-22, cross-checked against the complete full text of the Gothe study (bioRxiv 10.1101/485763 v1, 179,145 characters, token counts for CTCF/cohesin/RAD21/SMC3/SMC1/NIPBL/siRNA/shRNA/degron/auxin read in context) · NCBI Gene Expression Omnibus; Gothe HJ, Bouwman BAM, Gusmao EG, et al.; Roukos V · NCBI GEO / bioRxiv 2019What this evidence cannot show
For Canela 2017 this bounds the deposit only - its full text was not read, so an unpublished-to-GEO cohesin experiment cannot be excluded there. For Gothe it bounds both deposit and the December 2018 preprint text, leaving a seven-month revision window in which a non-sequencing occupancy experiment could have been added. Three other cell lines in Gothe (HCT116, HT1080, Cal51) and one (U2OS) have C-Fusion or imaging data but no sBLISS, so this beat did not measure the element's break rank in them.
Evidence that argues against this (1)
- Cohesin HAS been perturbed in a haematopoietic-lineage context in a separate literature this entry does not survey - cohesin-subunit mutations are recurrent in AML and STAG2/RAD21 knockdown studies in CD34+ cells exist. None of them maps DSBs, so they cannot test premise (i), but they mean the reagents and the cell systems for the missing experiment already exist and the gap is one of assay pairing rather than feasibility. Atkin ND et al. 2021
upgrades the objection from a cell-TYPE mismatch to a cell-type plus cell-cycle plus transcriptional-state mismatch: the HCT116 degron arm is lovastatin-arrested in G1 by the paper's own methods
rests on the percentile readout across all thirteen human tracks
premise (i) is unaddressed rather than underpowered: perturbation and phenotype are in disjoint cell types
amends the adjudication: the RAD21-degron arm is not a weak test of premise (i), it is not a test of it
exists because of the cell-type mismatch that entry establishes
the dissociation it is built on: strongest available occupancy, no break