Oncofusion

start / graph / ana-gothe-2019-bliss-deposition-audit

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

Gothe 2019's deposit is all-human, has base-resolution break lists, includes primary CD34+ marrow and a transcription-inhibition arm - and contains no CTCF or cohesin perturbation of any kind

also called GSE121742 audit · GSE121740 BLISS manifest · Q107 deposition audit, Gothe · the CD34+ break lists

The Gothe paper's data is entirely human, includes real bone-marrow progenitor cells, and lists every single break it found at single-base resolution - the richest deposit of the three. What it does not contain, anywhere, is an experiment that removes CTCF or cohesin.

The same thing, for a scientist

PMID 31202576 (Gothe et al., Mol Cell 2019) deposited SuperSeries **GSE121742** (20 samples, *Homo sapiens* only) with SubSeries **GSE121740** [BLISS-Seq, 18 samples] and **GSE121741** [RNA-Seq, 2 samples]. Every BLISS sample is deposited **twice** - as a `.bed.gz` list of DSB genomic positions with per-position unique-UMI counts, and as a `.bw` coverage track built by extending each UMI-DSB +/-100 bp - aligned with BWA-MEM to **hg19** at MAPQ > 30 with UMI-based duplicate filtering [E1]. **This is the only one of the three deposits in this dispute that ships break calls rather than coverage alone.** The arms are: TK6 DMSO/ETO x2 replicates; K562 DMSO/ETO; **TK6 +/- DRB x DMSO/ETO x2 replicates** (the transcription-elongation arm); and **primary CD34+ bone-marrow progenitor cells, DMSO/ETO x2 replicates**. **There is no CTCF, RAD21, SMC3, SMC1, NIPBL or WAPL perturbation sample, and no Hi-C** - loop anchors are imported from Rao et al. 2014 GM12878 [E1][E2]. Confirmed against the full text, not inferred from the manifest: the paper's only degron and siRNA arms are **TOP2A-mAID / TOP2A-mAID-TOP2B-/- HCT116**, a **CtIP auxin-degron TK6** line, and **siTOP2A / siTOP2B (ThermoFisher s14309, s108)** with **shRNA TOP2A-1/-2 and TOP2B-1/-2 (Origene TL308699A/C, TL308698A/C)** - the complete siRNA and shRNA catalogue in the methods is three siRNAs (no-target, TOP2A, TOP2B) and five shRNAs (two TOP2A, two TOP2B, scramble) [E2].

Go deeper
The manifest, with every sample title verbatim

GSE121742 (SuperSeries) - 20 samples, Homo sapiens. GSE121742_RAW.tar 2,147,235,840 bytes. Supplementary types: BED, BW 1.

SubSeriesassaysamplesprocessed types
GSE121740BLISS-Seq (sBLISS)18.bed.gz + .bw per sample
GSE121741RNA-Seq2.tsv (no filelist served)

The 18 BLISS samples 1:

GSM!Sample_titlesource.bed.gz bytes.bw bytes
GSM3444984TK6_DMSO_rep1TK66,812,79411,934,731
GSM3444985TK6_ETO_rep1TK663,404,037129,527,668
GSM3444986TK6_DMSO_rep2TK63,712,3895,535,419
GSM3444987TK6_ETO_rep2TK645,595,23497,462,428
GSM3444988K562_DMSOK56231,478,75848,564,381
GSM3444989K562_ETOK56262,737,121124,817,018
GSM3444990TK6_DRB_DMSO_rep1TK629,617,18047,229,129
GSM3444991TK6_untr_DMSO_rep1TK66,045,72310,749,349
GSM3444992TK6_DRB_ETO_rep1TK619,563,67431,811,876
GSM3444993TK6_untr_ETO_rep1TK645,344,20895,003,865
GSM3444994TK6_DRB_DMSO_rep2TK620,233,96533,607,720
GSM3444995TK6_untr_DMSO_rep2TK619,119,59231,463,599
GSM3444996TK6_DRB_ETO_rep2TK613,793,59123,798,707
GSM3444997TK6_untr_ETO_rep2TK651,474,874108,487,974
GSM3687236CD34_DMSO_rep1primary CD34+ bone marrow cells7,981,75113,259,943
GSM3687237CD34_ETO_rep1primary CD34+ bone marrow cells145,603,378271,755,799
GSM3687238CD34_DMSO_rep2primary CD34+ bone marrow cells12,145,28319,450,528
GSM3687239CD34_ETO_rep2primary CD34+ bone marrow cells160,374,955297,704,904

Plus TK6_WT_rep1 and TK6_WT_rep2 in GSE121741 [RNA-Seq]. GSE121741 has no suppl/filelist.txt - the FTP path returns a 404 Object-not-found page - so the two TSVs must be reached through the GSM records.

!Sample_data_processing for GSM3687237, verbatim 1:

"Reads were mapped to the human genome (GRCh37/hg19) using BWA-MEM and only reads with mapping quality scores greater than 30 were retained." ... "Bed files were generated containing a list of double-strand break (DSB) genomic locations associated with the number of unique UMIs identified." ... "First, one entry was created per UMI-DSB in the BED file and then each UMI-DSB was extended +/- 100bp."

The specific question the King asked: is there a CTCF or cohesin depletion arm at MLL/KMT2A?

No - and this is a paper-level answer, from the full text, not a deposit-level inference.

Token counts over the complete preprint text including the methods (179,145 characters): CTCF 84, cohesin 17, RAD21 4, SMC3 2, SMC1 0, NIPBL 0, degron 5, auxin 13, siRNA 10, shRNA 12, knockdown 2, depletion 5 2. Every one of those hits was read in context, and they resolve as:

tokenwhat it actually is
CTCF, cohesin, SMC3, RAD21ENCODE ChIP-seq occupancy used as annotation, and Rao 2014 GM12878 loop-anchor calls. "we found that ~40% of ETO-induced hotspots coincided with binding sites of CTCF and cohesin subunits (CTCF, SMC3, Rad21; p<10-6, Mann-Whitney-Wilcoxon test)"
degron, auxinTOP2A-mAID and TOP2A-mAID-TOP2B-/- HCT116 (Natsume 2016 parental), and a CtIP auxin-degron TK6 line from Hoa 2015
siRNAthe complete catalogue is three: "siRNA no target (Silencer Selected negative control, #4390843); siTOP2A (#4390824, siRNA ID #s14309); siTOP2B (#4390824, siRNA ID #s108)"
shRNAthe complete catalogue is five: "shRNA TOP2A-1 (TL308699A), shRNA TOP2A-2 (TL308699C), shRNA TOP2B-1 (TL308698A), shRNA Top2B-2 (TL308698C), shRNA scramble (TR30021)"

So the triad stated in this beat's dispatch brief stands as written: Gothe 2019 is not a second directional experiment on occupancy, and premise (i) remains untested by it. What Gothe does supply is the correlational half of premise (i) at single-element resolution, and - which the brief did not anticipate - a directional test of a different premise. See ana-gothe-measures-mll-breakage-and-fusion-in-the-same-cells.

And one thing the brief's framing of the dispute gets wrong. Gothe states in its own words that the MLL break hotspot is at a CTCF/cohesin site, and cites Canela 2017 for it: "This ETO-induced MLL DSB-hotspot coincides with DNase I hypersensitivity, CTCF, cohesin and TOP2B binding and activity sites (Canela et al., 2017; Cowell et al., 2012)" 2. The two literatures do not share a readout with Gray 2023, which is the real finding; but within the fragility side, the element's identity as a CTCF/cohesin site at MLL is explicit and cross-cited, not implicit.

Cell types, and what was measured in which

From the full text 2:

celltypewhat was done
TK6human lymphoblastoidsBLISS (2 reps), +/- DRB sBLISS (2 reps), C-Fusion 3D; isogenic panel: TDP2-/-, MRE11-/-, MRE11-/H129N, MRE11 loxP, CtIP-/- / CtIP auxin-degron, LigIV-/-, RAD54-/-, LigIV-/- RAD54-/-, and double TDP2-/-/MRE11-/H129N
K562erythroleukaemiasBLISS (1), C-Fusion 3D
primary CD34+ bone marrowhuman primary HSPCsBLISS (2 reps), C-Fusion 3D breakage AND fusion formation, synapsis measurement
HCT116colorectal carcinomaTOP2A-mAID and TOP2A-mAID-TOP2B-/-, C-Fusion 3D
HT1080fibrosarcomadox-inducible TOP2A loss, C-Fusion 3D
Cal51breast carcinomasiTOP2A/siTOP2B, C-Fusion 3D
U2OSosteosarcomagammaH2AX imaging under transcription/replication inhibition

Fusion FORMATION was measured, including in primary CD34+ cells. 4-color C-Fusion 3D scores both MLL breakage and specific fusions - MLL-AF4, MLL-AF6, MLL-AF9, MLL-ENL, with NPM1-ALK as a control partner pair. Quoted frequencies: "MLL fusions occurred rarely, at frequencies ranging from 6x10-4 to 2x10-3 in human hematopoietic cell lines and bone marrow progenitor CD34+ cells", against breakage frequencies of "approximately 14% of K562 cells" at 20 uM etoposide (5,731 of 41,318 cells) and "only 6% of cells had the MLL locus broken in CD34+ hematopoietic progenitor cells" even at 60 uM (2,994 of 48,912 cells) 2.

So hyp-ctcf-occupancy-splits-break-formation-from-rearrangement-fate's recorded status for Gothe - fusion formation measured "partly, not at this element" - needs amending. Fusion formation WAS measured, in five human systems including primary CD34+ marrow, with per-partner resolution. What is true is that it was never measured under an occupancy perturbation, and that C-Fusion 3D is an imaging proximity/separation assay at whole-gene probe resolution, so it cannot localise a fusion junction to the element. The element-level claim should be about assay resolution, not about whether formation was measured.

Facts and figures
scriptmanifest audit via eutils elink/esummary and GEO filelist.txt; full text retrieved as markdown and grepped locally; no committed script
statusrun
result summary20 samples, all human, hg19. BLISS deposited as both base-resolution BED break lists with UMI counts and +/-100 bp extended bigWigs - the only break-CALL deposit of the three studies. Arms: TK6 +/-ETO x2, K562 +/-ETO, TK6 +/-DRB x +/-ETO x2, primary CD34+ marrow +/-ETO x2. No CTCF/cohesin perturbation in the deposit and none in the full text; the only degron/siRNA/shRNA arms are TOP2A, TOP2B and CtIP. Fusion formation WAS measured by 4-color C-Fusion 3D in TK6, K562, HCT116, HT1080, Cal51 and primary CD34+ cells, for MLL-AF4/AF6/AF9/ENL.
genome build depositedhg19 (GRCh37)
verdictno-ctcf-or-cohesin-arm; the triad stands
inputs
[
  "GEO GSE121742 (SuperSeries, PMID 31202576), 20 samples",
  "GEO GSE121740 [BLISS-Seq], 18 samples, filelist.txt with 36 file entries",
  "GEO GSE121741 [RNA-Seq], 2 samples (no filelist served)",
  "GSM3687237 full record",
  "bioRxiv 10.1101/485763 v1 full text, 179,145 characters"
]
accessions
[
  "GSE121742",
  "GSE121740",
  "GSE121741"
]
How this could bite us back
  • Gothe's CD34+ arm is adult bone marrow. Importing it as a model of the cell of origin for infant KMT2A::AFF1, which arises in fetal liver, is exactly the lineage-and-stage substitution this project's rule 73 exists to prevent - and the substitution is tempting here precisely because the data is so much better than anything else available.
  • C-Fusion 3D scores fusion formation by imaging probe co-localisation at whole-gene resolution. A therapeutic claim about preventing fusions at a 147 bp element cannot be validated with an assay that cannot see 147 bp.
Still open
  1. Does the PUBLISHED July 2019 version contain a figure panel added in revision that the December 2018 preprint lacks? The CD34+ samples were deposited in March 2019, so at least one arm is post-preprint, and the published figure list has not been compared to the preprint's.
  2. How many CD34+ donors, of what age, and from what source? The deposit says only 'primary CD34+ bone marrow cells' and the age matters a great deal for a disease of infancy.
  3. The .bed.gz break lists carry per-position UMI counts. What is the actual UMI count at the KMT2A exon-11 element in each CD34+ replicate, and does it sit on the CTCF motif or beside it? That question needs the BED files (145-160 MB each for the etoposide replicates) and is the natural next compute.
Every source behind this entry (2)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    PMID 31202576 deposited GEO SuperSeries GSE121742 (20 samples, Homo sapiens only) with SubSeries GSE121740 [BLISS-Seq, 18 samples] and GSE121741 [RNA-Seq, 2 samples]; each BLISS sample is deposited both as a .bed.gz list of DSB positions with unique-UMI counts and as a .bw coverage track from +/-100 bp UMI-DSB extension, aligned with BWA-MEM to GRCh37/hg19 at MAPQ>30 with UMI-based duplicate filtering; the arms are TK6 DMSO/ETO x2, K562 DMSO/ETO, TK6 +/-DRB x DMSO/ETO x2, and primary CD34+ bone marrow cells DMSO/ETO x2; no sample in the series perturbs CTCF, RAD21, SMC3, SMC1, NIPBL or WAPL, and no Hi-C sample exists; GSE121741 has no supplementary filelist.txt at the FTP path.

    What this evidence cannot show

    A manifest audit, retrieved 2026-09-22. GSE121742_RAW.tar and GSE121740_RAW.tar are both reported as exactly 2,147,235,840 bytes, which is 2 GiB to the byte and is almost certainly a GEO display cap rather than the true size, so the archive sizes quoted here should not be used for a download budget. GSE121741's two TSVs were not enumerated because the FTP filelist path 404s; their existence rests on the esummary suppfile field (TSV) and the sample titles TK6_WT_rep1/rep2. The CD34+ donors' number, age and source are not in the fields quoted; the source field reads only 'primary CD34+ bone marrow cells' (with a double space).

    Evidence that argues against this (1)
    • The deposit's greatest apparent strength - base-resolution BED break lists - carries a hidden weakness that cuts against reading fine positional structure from it: the bigWigs extend every UMI-DSB by +/-100 bp, so the coverage tracks have ~200 bp effective resolution and a single break becomes a 201 bp plateau. Any argument about whether a break sits ON a CTCF motif versus 60 bp away must use the .bed.gz files, which are 145-160 MB compressed for the CD34+ etoposide replicates, not the bigWigs. NCBI Gene Expression Omnibus 2019
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.preprint

    The complete full text of PMID 31202576, including methods, contains no CTCF, cohesin, RAD21, SMC3, SMC1, NIPBL or WAPL perturbation: its only degron lines are TOP2A-mAID and TOP2A-mAID-TOP2B-/- HCT116 plus a CtIP auxin-degron TK6, its entire siRNA catalogue is 'siRNA no target (#4390843); siTOP2A (#4390824, ID s14309); siTOP2B (#4390824, ID s108)' and its entire shRNA catalogue is 'shRNA TOP2A-1 (TL308699A), shRNA TOP2A-2 (TL308699C), shRNA TOP2B-1 (TL308698A), shRNA Top2B-2 (TL308698C), shRNA scramble (TR30021)'; CTCF, cohesin, SMC3 and RAD21 appear only as ENCODE occupancy annotation and Rao 2014 GM12878 loop calls, and the paper states verbatim 'This ETO-induced MLL DSB-hotspot coincides with DNase I hypersensitivity, CTCF, cohesin and TOP2B binding and activity sites (Canela et al., 2017; Cowell et al., 2012)'; the human systems are TK6 (with isogenic TDP2, MRE11, CtIP, LigIV, RAD54 mutants), K562, primary CD34+ bone marrow cells, HCT116, HT1080, Cal51 and U2OS; and 4-color C-Fusion 3D measured both MLL breakage and MLL-AF4/AF6/AF9/ENL fusion formation, including in primary CD34+ cells, with fusion frequencies of 6x10-4 to 2x10-3.

    Spatial chromosome folding and active transcription drive DNA fragility and formation of oncogenic MLL translocations - complete preprint full text with methods (179,145 characters), read 2026-09-22 · Gothe HJ, Bouwman BAM, Gusmao EG, Piccinno R, Sayols S, Drechsel O, Petrosino G, Minneker V, Josipovic N, Mizi A, Nielsen CF, Wagner EM, Takeda S, Sasanuma H, Hudson DF, Kindler T, Baranello L, Papantonis A, Crosetto N, Roukos V · bioRxiv (published as Mol Cell 75:267-283.e12, PMID 31202576) 2018
    DOI
    What this evidence cannot show

    This is the DECEMBER 2018 PREPRINT, version 1, not the July 2019 published article - the published version is paywalled, has no PMC deposit and could not be retrieved (see ana-gothe-2019-full-text-the-route-that-worked). Mol Cell review added content: the CD34+ sBLISS samples were deposited in March 2019, three months after the preprint, so the preprint's figure set may not include them, and a reviewer-requested CTCF or cohesin experiment added between December 2018 and July 2019 would not appear in this text. The negative is therefore bounded to the preprint. That said, the GEO deposit - which was finalised in June 2019, at publication - contains no cohesin or CTCF sample either, so a revision-added occupancy-perturbation genomics experiment is excluded from both directions; only a non-sequencing one (an imaging or blot arm) could have slipped between them. Token counts were computed on the retrieved markdown rendering, which includes figure legends, reference anchor text and boilerplate, so the counts are upper bounds on body-text occurrences.

    Evidence that argues against this (1)
    • The genuine residual risk in this negative is the three-month gap, and it is not hypothetical for this paper: Mol Cell published Gothe and Canela 2019 back-to-back in the same issue (75:252-266 and 75:267-283), which is a co-review arrangement that very often produces cross-requested experiments. If a reviewer asked Gothe to test cohesin dependence, the natural response would have been to cite the companion paper's HCT116 RAD21-AID data rather than repeat it - which would leave no new sample and no new method text, and would be consistent with everything observed here. Canela A et al. 2019
Connected to