Oncofusion

start / graph / ana-canela-2019-rad21-degron-end-seq-deposition-audit

Analysisestablished confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.Q107-break-machinery-deposition

Canela 2019's RAD21-degron END-seq IS deposited and IS readable today - one sample per arm, no replicates, no minus-auxin control, hg19, bigWig coverage only

also called GSE129529 audit · GSE129524 END-seq manifest · the RAD21-AID END-seq deposit · Q107 deposition audit, 2019

The one experiment that removes the cohesin ring from human cells and then maps DNA breaks genome-wide is public, and its processed files can be read over the internet without downloading anything big. But the deposit contains a single sample for each condition, with no repeat and no matching untreated control for the depleted cells.

The same thing, for a scientist

PMID 31202577 (Canela et al., Mol Cell 2019) deposited SuperSeries **GSE129529** (73 samples) with four SubSeries: **GSE129524** [END-seq, 48 samples], **GSE129526** [ChIP-seq, 12], **GSE129527** [Nascent RNA, 3], **GSE129528** [HTGTS, 10]. The human RAD21-degron END-seq arm is **three samples, one per condition**: `END_seq_HCT116_0_5h` (GSM3714903, genotype WT), `END_seq_HCT116__RAD21_AID_0_5h` (GSM3714904, genotype `RAD21 AID`), `END_seq_HCT116_NT (control)` (GSM3714905, genotype WT) [E1]. Every processed file in the END-seq and ChIP-seq series is a **bigWig coverage track** built by `bedtools genomecov` + `bedGraphToBigWig`; **no narrowPeak, broadPeak or BED peak call is deposited for any END-seq or ChIP-seq sample**, although the methods state peaks were called with MACS 1.4.3 at >10-fold over the matched non-treated sample [E1]. Deposited human coordinates are **GRCh37/hg19**, confirmed twice - by the `Genome_build` field and, independently, by the bigWig chromosome table (chr1 = 249,250,621; chr11 = 135,006,516; chrX = 155,270,560) [E2]. **Verdict: the break half can be read out of the deposited processed files, and this beat read it (see `ana-kmt2a-exon-11-element-is-a-top-0-05-percent-etoposide-dsb-site`). It cannot be ANSWERED from them**, because the deposited design has n = 1 per arm, compares two different genotypes rather than one line +/- auxin, and provides no untreated control for the depleted arm [E1].

Go deeper
The manifest, verbatim

GSE129529 (SuperSeries) - 73 samples, Mus musculus; Homo sapiens. GSE129529_RAW.tar 9,921,464,320 bytes. Supplementary types: BW, TXT.

SubSeriesassaysamplesdeposited processed typesRAW.tar bytes
GSE129524END-seq4844 x .bw only5,931,765,760
GSE129526ChIP-seq1212 x .bw only3,570,708,480
GSE129527Nascent RNA-seq33 x .bw only414,013,440
GSE129528HTGTS1010 x .txt.gz (153 KB - 1.5 MB)4,997,120

The three human END-seq samples and their exact titles 1:

GSM!Sample_title verbatimgenotype fieldfilebytes
GSM3714903END_seq_HCT116_0_5hWTGSM3714903_END_seq_HCT116_0_5h.bw242,770,124
GSM3714904END_seq_HCT116__RAD21_AID_0_5hRAD21 AIDGSM3714904_END_seq_HCT116_RAD21_AID_0_5h.bw246,881,391
GSM3714905END_seq_HCT116_NT (control)WTGSM3714905_END_seq_HCT116_NT.bw88,109,321

Note the double underscore in GSM3714904's sample title (HCT116__RAD21) against a single underscore in its filename. A title-to-filename join on the string will fail. Also note that GSM3714905, the only untreated sample, is the WT line, so the depleted arm has no untreated baseline.

The cohesin-loss arms in the same series are three, not one 1:

GSMtitlegenotypespecies
GSM3714904END_seq_HCT116__RAD21_AID_0_5hRAD21 AIDhuman
GSM3714900END_seq_MEF_Smc3_FF_0_5hSmc3f/f+CREmouse
GSM3714902END_seq_MEF_Wapl_FF_CRE_0_5hWaplf/f+CREmouse

With WT/floxed controls GSM3714899 (Smc3_PP, genotype WT) and GSM3714901 (Wapl_FF, genotype Waplf/f). All n = 1.

Protocol, quoted from the deposit

GSM3714904's !Sample_growth_protocol_ch1 1:

"HCT116 cells were obtained from (Natsume et al., 2016), arrested in G1 by the treatment with 20uM of lovastatin for 24 hours and treated with auxin for 4 hours before 50uM etoposide treatment for 30 minutes."

So the readout is a 30-minute etoposide pulse in G1-arrested cells after a 4 h auxin pre-treatment. !Sample_description for both GSM3714903 and GSM3714904 is control: END_seq_HCT116_NT.fastq.gz - i.e. both etoposide arms are normalised against the same single WT untreated library.

The library is END-seq with a two-enzyme option: Exonuclease VII then Exonuclease T "to detect all TOP2 lesions", or Exonuclease T alone ("ExoT END-seq") "to detect only DSBs". In the mouse pre-B arms both variants were run for every condition (18 paired samples). For HCT116 only the ExoVII+ExoT variant was run - there is no ExoT_HCT116 sample - so the human arm cannot separate trapped TOP2 cleavage complexes from irreversible DSBs, which is the very distinction the paper's own model turns on 1.

Two metadata defects in the deposit that would corrupt a downstream analysis

1. The three HCT116 ChIP-seq samples are annotated as mouse, and they are human. GSM3714939 (ChIP_seq_RAD21_HCT116), GSM3714940 (ChIP_seq_CTCF_HCT116) and GSM3714941 (ChIP_seq_INPUT_HCT116) all carry !Sample_organism_ch1 = Mus musculus and Genome_build: GRCm38p2/mm10, while !Sample_characteristics_ch1 = cell line: HCT116 2. The files themselves are unambiguously human hg19: 24 chromosomes, chr1 = 249,250,621, chr11 = 135,006,516, chrX = 155,270,560 2. The error propagates into the series metadata - esummary db=gds returns taxon: Mus musculus for GSE129526 even though a quarter of its samples are human. Anyone selecting human samples by the organism field loses the only human occupancy tracks in the paper.

2. One antibody catalogue number is used for all three ChIP samples. chip antibody vendor: Millipore, chip antibody cat. #: 07-729 appears identically on the RAD21 sample, the CTCF sample and the INPUT sample 2. Millipore 07-729 is an anti-RAD21 reagent, so the CTCF row's antibody metadata is a copy-paste and cannot be used to identify the CTCF reagent. The two tracks are nevertheless different libraries - different summits (RAD21 chr11:118,359,392 vs CTCF chr11:118,359,409), different maxima (10.18 vs 6.85) and different genome-wide means - so the labels are not simply swapped files.

What this deposit CAN and CANNOT decide about premise (i) of the triad

Can: whether END-seq signal exists at a named 1-kb interval, and whether it changes when RAD21 is degraded. A bigWig is genome-wide and range-indexed; the question is a lookup, not an alignment. No raw reads are required and no bulk download is required.

Cannot:

  1. n = 1 per arm. Two numbers, no variance, no test.
  2. The contrast is genotype, not auxin. WT vs RAD21 AID - not RAD21-AID minus auxin vs RAD21-AID plus auxin. A clonal difference between the parental line and the tagged clone is indistinguishable from the auxin effect.
  3. No untreated control for the depleted arm. The paper's own peak-calling rule (>10-fold over matched NT) cannot be applied to GSM3714904, because its matched NT is a different genotype's library.
  4. The cell type is wrong for the question - see ana-hct116-is-the-wrong-cell-line-for-the-cohesin-arm-at-kmt2a, which is the load-bearing limitation and is a fact about the biology, not the deposit.

Cost, if raw reads were nevertheless wanted: GSE129524_RAW.tar is 5.93 GB and GSE129526_RAW.tar 3.57 GB; the three HCT116 END-seq SRA experiments are SRX5657095, SRX5657096, SRX5657097. Alignment would need bowtie 0.12.7 (the deposit's version) plus an hg19 index (~3.5 GB) and MACS. This box has 26 GB free on C: and 14 GB on D:, no samtools, no bedtools and no UCSC binaries; the raw route is feasible on disk but not on tooling, and it is unnecessary because the processed route answered the question.

Facts and figures
scriptread over HTTP range requests with pybigtools 0.3.0 under CPython 3.12.13; no analysis script is committed because the audit is of a manifest
statusrun
result summaryDeposited and readable. Human loss-of-cohesin END-seq arm = 3 samples, n = 1 each, hg19, bigWig coverage only, no peak calls, no minus-auxin control, no untreated control for the depleted genotype, and only the ExoVII+ExoT variant (so trapped TOP2ccs cannot be separated from DSBs in the human arm). Two metadata defects found: the HCT116 ChIP samples are annotated mouse/mm10 but are human/hg19, and one antibody catalogue number serves all three ChIP samples.
genome build depositedGRCh37/hg19 for human, GRCm38p2/mm10 for mouse
verdictreadable-from-processed-files-but-underpowered-by-design
inputs
[
  "GEO GSE129529 (SuperSeries, PMID 31202577), 73 samples",
  "GEO GSE129524 [END-seq] filelist.txt - 44 bigWig + GSE129524_RAW.tar (5,931,765,760 bytes)",
  "GEO GSE129526 [ChIP-seq] filelist.txt - 12 bigWig + GSE129526_RAW.tar (3,570,708,480 bytes)",
  "GEO GSE129527 [Nascent RNA] filelist.txt - 3 bigWig",
  "GEO GSE129528 [HTGTS] filelist.txt - 10 .txt.gz junction tables",
  "GSM records GSM3714899-GSM3714905, GSM3714939-GSM3714941"
]
accessions
[
  "GSE129529",
  "GSE129524",
  "GSE129526",
  "GSE129527",
  "GSE129528",
  "SRX5657095",
  "SRX5657096",
  "SRX5657097"
]
How this could bite us back
  • If a therapeutic argument is ever built on 'cohesin loss reduces breakage at KMT2A', it currently rests on two unreplicated libraries from two different clones in a colorectal cell line. That is not a foundation for a clinical claim, and the deposit is the reason to say so out loud rather than a reason to stop looking.
  • The mouse-annotated human samples are a live hazard for any automated pipeline in this project that selects samples by organism: it would silently drop the only human CTCF and RAD21 occupancy tracks in the paper and return a clean, wrong answer.
Still open
  1. Does GSE129524_RAW.tar (5.93 GB) contain MACS peak-call files alongside the FASTQs? The filelist cannot see inside the archive and the methods say peaks were called.
  2. Were the HCT116 arms ever replicated? The paper reports a single experiment per condition in the deposit; whether unreplicated data underlie the published human panel, or whether replicates exist and were not deposited, is answerable only by asking the authors.
  3. Why was the ExoT-only END-seq variant run for every mouse condition and for no human condition? Without it, the human arm cannot distinguish a trapped cleavage complex from an irreversible break - the paper's central mechanistic distinction.
Every source behind this entry (2)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    GEO SuperSeries GSE129529 (73 samples) accompanies PMID 31202577 and comprises SubSeries GSE129524 [END-seq, 48 samples], GSE129526 [ChIP-seq, 12], GSE129527 [Nascent RNA, 3] and GSE129528 [HTGTS, 10]; the human RAD21-degron END-seq arm is exactly three samples with the verbatim titles END_seq_HCT116_0_5h (GSM3714903, genotype WT), END_seq_HCT116__RAD21_AID_0_5h (GSM3714904, genotype 'RAD21 AID', auxin 4 h then 50 uM etoposide 30 min after 24 h lovastatin G1 arrest) and END_seq_HCT116_NT (control) (GSM3714905, genotype WT); every deposited processed END-seq and ChIP-seq file is a bigWig coverage track made with bedtools genomecov and bedGraphToBigWig, no peak-call file of any kind is deposited, and human reads were aligned to GRCh37/hg19 with bowtie 0.12.7 and peaks called with MACS 1.4.3 at >10-fold over the matched non-treated sample.

    What this evidence cannot show

    This is the deposited file manifest and sample metadata, retrieved 2026-09-22. No RAW.tar was opened, so a peak-call file could in principle sit inside one of the four RAW archives; the filelist is the authoritative statement of what is served as an individual supplementary file and it lists none. The 48 END-seq samples map to 44 bigWig files because two mouse activated-B arms are deposited as merged replicate pileups (GSM3714906, GSM3714908) and two samples are marked '(re-used)' from an earlier series. The genotype and treatment fields are the submitter's, not re-derived from the data.

    Evidence that argues against this (1)
    • The absence of deposited peak calls is a weaker limitation than it first appears, and arguing from it would be wrong: a bigWig is a complete quantitative record at base resolution, so peak calls are a convenience, not the measurement. The genuine limitation of this deposit is the n = 1 design, which no file format would fix. Equally, the same series DOES deposit junction-level tables for the HTGTS arm (10 x .txt.gz), so the submitters were not uniformly withholding processed calls - they deposited coverage where coverage was the product. NCBI Gene Expression Omnibus 2019
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    The three HCT116 ChIP-seq samples of GSE129526 (GSM3714939 RAD21, GSM3714940 CTCF, GSM3714941 INPUT) are annotated 'Sample_organism_ch1 = Mus musculus' with 'Genome_build: GRCm38p2/mm10' and an identical antibody entry (Millipore cat. # 07-729) on all three, yet the deposited bigWig files are human GRCh37/hg19 - 24 chromosomes with chr1 = 249,250,621, chr11 = 135,006,516 and chrX = 155,270,560, matching hg19 exactly and excluding mm10 (mm10 chr1 = 195,471,971).

    What this evidence cannot show

    The chromosome-size evidence proves the build is hg19 and therefore that the sample is human-aligned; it does not by itself prove the cells were HCT116 rather than another human line, which rests on the submitter's 'cell line: HCT116' characteristic. The identical antibody catalogue number across RAD21, CTCF and INPUT means the CTCF track's reagent is undocumented; the two tracks are distinguishable as separate libraries by summit position (chr11:118,359,392 vs 118,359,409), maximum (10.18 vs 6.85) and genome-wide mean (0.07896 vs 0.06706), so they are not duplicate files, but the CTCF identity rests on the sample title alone.

    Evidence that argues against this (1)
    • An alternative reading is that these samples are genuinely mouse files mislabelled HCT116 - but that is excluded by the chromosome sizes, which are hg19 and not mm10, and by the fact that the RAD21/CTCF summit falls inside human KMT2A canonical exon 11 at an hg19 coordinate. A mouse track could not produce a peak at a human exon coordinate. A weaker alternative - that the organism field is right and the build field wrong - is also excluded for the same reason. UCSC Genome Browser 2013
Connected to