Re-analyse the eight unreported H3K9me3 samples in GSE210480 at the KMT2A breakpoint cluster region - compute only, and it decides whether the rearrangement brake is mark-specific
also called the GSE210480 H3K9me3 re-analysis · is the KMT2A brake me1/me2-specific?
Eight public sequencing samples measure a silencing tag on the KMT2A gene that the paper reporting them never showed. Downloading and analysing them would settle whether the gene-breaking mechanism runs on one particular tag or on silencing in general.
The same thing, for a scientist
GSE210480 contains H3K9me3 ChIP-seq in RPE cells across four arms - control, siKDM3B, siG9A and siKDM3B+siG9A, two replicates each - matched to H3K9me1, H3K9me2, CTCF and Input, and PMID 37788669 reports **no** H3K9me3 result [E1, E2]. The analysis: align all three mark arms plus inputs to hg38, **reproduce the published H3K9me1 result over the *KMT2A* BCR as the pipeline's acceptance test**, then quantify H3K9me3 over the BCR, over the exon-11 CTCF site, and over the 17,077 CTCF peaks reported to lose intensity on KDM3B depletion. **Read-level resolution over the locus, not the paper's 10 kb genome-wide bins**, because a single 10 kb bin cannot separate the 8.3 kb BCR from its flank - which is the distinction that separates the paper's me1 result from its me2 result [E2]. Pre-registered readings: if H3K9me3 is flat across all four arms while me1 and me2 reproduce, the brake is me1/me2-specific and `mark-h3k9me3`'s direction and this beat's mechanism are reconciled on mark identity. If H3K9me3 tracks me1/me2, the brake is a general H3K9-methylation effect, the reconciliation fails, and `drug-unc0642`'s contraindication hardens.
Design, with the acceptance test first
Inputs. GSE210480: 8 H3K9me3, 8 H3K9me1, 8 H3K9me2 (main series), 8 Input, plus the 4 CTCF + 4 Input samples and the exp1 replicate set. Match case-insensitively - the deposit spells arms as H3K9ME3, H3K9me3 and CTRLsiKDm3B - and print every matched title, because a naive string match silently drops arms 2.
Acceptance test, run before any me3 number is believed. Reprocess the H3K9me1 arm and reproduce two published results: (a) H3K9me1 is lower across the BCR than in flanking sequence in control cells, and (b) siKDM3B raises it uniformly across the BCR while siKDM3B+siG9A resets it. If (a) and (b) do not reproduce, the pipeline is wrong and the me3 answer is uninterpretable. Report the acceptance-test figures whatever the outcome.
Measurements.
| region | why |
|---|---|
| the 8.3 kb KMT2A BCR, in ~500 bp windows | the locus; 10 kb bins are too coarse |
| the exon-11 CTCF site | the element whose occupancy is the brake's terminal step - and this analysis must also output its hg38 coordinates, which no publication gives |
| BCR flank, ±20 kb | me2 rose here and me1 did not; a geometry control |
| chromosome 11 centromere-proximal control region | the paper's own negative FISH control region |
| the TCF3 and AFF3 loci | the two other responsive loci |
| the MLLT1 and MLLT3 loci | the two non-responsive loci - the best internal negative controls available |
| all CTCF peaks, ±5 kb | the paper did this for me1; doing it for me3 is a free genome-wide comparison |
Statistics and the null. Two replicates per arm is the design, so this cannot support a per-window significance test. Report effect sizes with both replicates plotted, and state the null explicitly: the distribution of me3 fold-change across all 10 kb bins genome-wide, against which the BCR's rank is the actual quantity of interest - the same framing the paper uses for me1, where the BCR was 'one of the strongest events across the genome'.
What could go wrong, named in advance
- Depth. H3K9me3 is a broad-domain mark and a library adequate for me1 peaks may be too shallow for confident me3 quantification over 8.3 kb. Report per-sample depth and the FRiP or equivalent before any conclusion; if depth is inadequate, the honest output is 'the deposited data cannot answer this', which is itself worth recording.
- No spike-in. The paper describes none. So a global shift in me3 cannot be distinguished from a normalisation artefact, and only relative comparisons - BCR versus flank, BCR versus its genome-wide bin distribution - are interpretable. This is the single hardest limit on the analysis and it cannot be engineered away.
- An empty or unusable archive. This project has found a well-formed, correctly named, empty supplementary archive and a SuperSeries that denied having raw data in three different ways, all three wrong. Check that raw files exist and are non-trivial before scheduling compute.
- Confirmation pressure. The result that tidies this graph up is 'me3 flat'. The pipeline must be validated on me1 first, the me3 answer read second, and both reported regardless of which way the tidy conclusion falls.
- Cell type. RPE, epithelial. A null here does not license any statement about H3K9me3 at fusion-target promoters in a leukaemic blast, which is where
mark-h3k9me3's argument lives.
| script | analyses/gse210480-h3k9me3-kmt2a-bcr/run.py |
|---|---|
| status | planned |
| acceptance test | reproduce the published H3K9me1 result before believing any H3K9me3 number: (a) H3K9me1 lower across the BCR than flanking sequence in control cells; (b) siKDM3B raises it uniformly across the BCR and siKDM3B+siG9A resets it. Report these figures whatever the outcome. |
| null to report | the distribution of H3K9me3 fold-change across all 10 kb bins genome-wide, against which the BCR's RANK is the quantity of interest - the same framing the paper applied to me1, where the BCR was 'one of the strongest events across the genome' |
| n replicates | 2 |
| statistics note | two replicates per arm cannot support per-window significance testing. Report effect sizes with both replicates plotted and the genome-wide rank; do not report p-values per window. |
| hard limit | NO SPIKE-IN NORMALISATION was used in the original ChIPs, so a global shift in H3K9me3 cannot be distinguished from a normalisation artefact. Only relative comparisons - BCR versus flank, BCR versus the genome-wide bin distribution - are interpretable. This cannot be engineered away. |
| cost tier | low |
| timeline | days - download, align, quantify. No wet work, no cohort, no permissions. |
| closes | absence-h3k9me3-at-the-kmt2a-locus-was-assayed-and-never-reported |
| cell type limit | RPE, epithelial. A null result here licenses NO statement about H3K9me3 at fusion-target promoters in a leukaemic blast, which is where mark-h3k9me3's argument lives. |
| absence class | measured-not-analysed |
inputs
[ "GSE210480 - 8 H3K9me3, 8 H3K9me1, 8 H3K9me2, matched Inputs, 4 CTCF + Inputs (Homo sapiens, RPE)", "hg38 reference and a KMT2A BCR interval defined in hg38 base positions, NOT exon numbers" ]
regions
[ "the 8.3 kb KMT2A BCR in ~500 bp windows", "the exon-11 CTCF site (and OUTPUT its hg38 coordinates - no publication gives them)", "BCR flank +/- 20 kb", "the chromosome 11 centromere-proximal control region", "TCF3 and AFF3 (responsive loci)", "MLLT1 and MLLT3 (non-responsive loci - internal negative controls)", "all CTCF peaks +/- 5 kb, for the genome-wide comparison the paper made for me1" ]
pre registered readings
{
"me3_flat_while_me1_and_me2_reproduce": "the brake is me1/me2-specific; mark identity reconciles mark-h3k9me3 with this beat's mechanism; a G9a-SELECTIVE inhibitor becomes the right tool and drug-unc0642's dual activity becomes the wrong one",
"me3_tracks_me1_me2": "the brake is a general H3K9-methylation effect; mark identity does NOT reconcile the two entries; drug-unc0642's contraindication hardens into a genuine conflict and mark-h3k9me3's bite_back applies directly to the protective strategy",
"insufficient_depth": "record 'the deposited data cannot answer this' - a real and reportable outcome that upgrades the absence from measured-not-analysed to needing new wet work"
}design not obtainable cycle 59
{
"what_this_analysis_specified": "~500 bp windows and read-level resolution over the locus, with 8 Input samples among its inputs",
"what_the_deposit_actually_is": "1,000 bp binned ratio tracks with no Input files. The bins sit BETWEEN the 500 bp this design wanted and the 10 kb it feared, which is why nobody caught it: a binned track answers a finer query without raising an error.",
"verified_by": "king, independently - run-length encoding over 20 kb gives modal and median 1000, and the 147 bp exon-11 interval returns the same mean as a 501 bp interval containing it",
"what_survives": "the ACCEPTANCE TEST, which was excellent and passed exactly as written: me1 depleted at the BCR versus flanks in control, siKDM3B raising me1 at the BCR with no replicate overlap, the double knockdown resetting, and me2 rising on the centromeric flank.",
"status": "run at 1 kb; the sub-kilobase half of the design is unobtainable from any file in this series"
}- The result that tidies this graph up is 'me3 flat', and that is exactly why the pipeline must be validated on me1 first and both results reported. An analysis run to resolve a contradiction is an analysis with a preferred answer.
- A flat me3 profile from a shallow library looks identical to a genuine null. Depth and library-complexity metrics must be reported before the profile is interpreted, or this analysis will manufacture the reassurance it was run to test.
- What are the exon-11 CTCF site's hg38 coordinates? This analysis should output them as a by-product, because no publication gives them and three of this beat's entries are blocked on them.
- Does the G9a-knockdown-alone arm lower H3K9me3 at the BCR? If it does, then G9a - annotated as an me1/me2-only writer - is influencing me3 in cells, which would mean the enzymology-based resolution of this beat's central tension is wrong even before the KDM3B arm is considered.
- Do MLLT1 and MLLT3 show no H3K9 methylation change at all, or a change without a cytogenetic consequence? The paper reports them as non-responsive by FISH and never shows their chromatin. If their marks move and their loci do not, the mark is not sufficient.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
PMID 37788669 states it performed ChIP sequencing for H3K9me1/2/3 methylation marks in control and KDM3B siRNA transfected RPE cells and lists H3K9me3 antibody Abcam ab8898 in its STAR Methods, yet no H3K9me3 ChIP result appears in any figure, supplemental figure legend, result sentence or discussion passage - the only other occurrence of the string in the retrievable text is a Supplemental Fig. 2A in vitro enzyme western blot.
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; et al. · Cell 2023What this evidence cannot show
Established by retrieving the full text via efetch db=pmc on PMC10591855 (Europe PMC fullTextXML returned HTTP 500; <body> confirmed) and regex-searching the tag-stripped 80,638-character body, which returned exactly two hits. The bound is on the **retrievable text**: a result present only as an unlabelled figure panel, or in a supplementary file not carried in the PMC author manuscript, would not be seen. Recorded in full in `absence-h3k9me3-at-the-kmt2a-locus-was-assayed-and-never-reported`.
Evidence that argues against this (1)
- Depositing an assayed arm and not reporting it usually means the result was null, so the most probable content of the missing arm is the one that supports the paper's own framing. This analysis is worth running because the answer is cheap and load-bearing, not because the omission is suspicious. Gray ZH; Chakraborty D; Duttweiler RR; Alekbaeva GD; Murphy SE; et al. 2023
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
GSE210480 (Homo sapiens, 60 samples) contains eight H3K9me3 ChIP-seq samples across control, siKDM3B, siG9A and siKDM3B+siG9A arms with two replicates each, matched to H3K9me1, H3K9me2, CTCF and Input samples in the same arms, with sample titles inconsistently cased within the series (H3K9ME3, H3K9me3, CTRLsiKDm3B), and the paper's own genome-wide analyses are binned at 10 kb - too coarse to separate the 8.3 kb breakpoint cluster region from its flank.
NCBI GEO series GSE210480, brief sample listing retrieved 2026-09-22 with all 60 !Sample_title lines enumerated · NCBI Gene Expression Omnibus 2026What this evidence cannot show
A sample-title listing establishes submission, not usability: sequencing depth, library complexity and whether raw files are actually downloadable were not checked by this beat. The 10 kb bin size is taken from the paper's own text ('the list of all 10 Kb genomic bins'), which describes its genome-wide tables and not necessarily every locus-level panel.
Evidence that argues against this (1)
- If the deposited H3K9me3 libraries are too shallow for a broad-domain mark over 8.3 kb, this analysis returns 'cannot answer' rather than an answer - and a shallow library can still produce a plausible-looking flat profile, which is the failure mode to guard against by reporting depth first. NCBI Gene Expression Omnibus 2026
the compute that closes it: eight public H3K9me3 samples, quantified over the KMT2A BCR with the me1 arm as the pipeline's acceptance test
reproducing the published H3K9me1 BCR result is the acceptance test before any me3 number is believed
the H3K9me3 tracks open; that analysis has no remaining blocker
executes the pre-registered analysis; acceptance test A1-A4 passed
answers its third open question: MLLT3 shows no mark change, MLLT1 a change without a consequence
its design asks for ~500 bp windows and read-level resolution; the deposited histone tracks are 1 kb binned and cannot deliver either