Oncofusion

start / graph / ana-q114-the-h3k9me3-writer-and-its-own-mark-barely-coincide-in-the-same-nuclei

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q114-suv39h1-at-the-locus

The rarest thing in the deposit, measured: across 59,821 CTCF anchors in the SAME nuclei, SUV39H1 occupancy and H3K9me3 correlate at Spearman 0.148 and 0.078 - and a quarter to two-fifths of the enzyme's own strongest sites carry no more of its mark than the median anchor

also called writer versus mark coincidence · Q114 question C · SUV39H1 off its own mark

This dataset is unusual because it measured the enzyme and the chemical tag it makes in the very same cells. They barely track each other: where the enzyme is strongest, the tag is often no higher than average, which means the enzyme is doing something at many sites other than leaving its mark - or the antibody is not finding only the enzyme.

The same thing, for a scientist

**The question this answers was written into the corpus by the entry that catalogued the deposit** - `exp-suv39h1-and-h3k9me3-cutandrun-in-two-patient-derived-diffuse-midline-glioma-lines`: *"Do SUV39H1 and H3K9me3 coincide in these cells, and what fraction of SUV39H1 sites carry no H3K9me3? A writer found off its own mark is either a scaffolding role or an antibody artefact."* [E1] **Measured over 59,821 primary-assembly CTCF-summit windows, in SU-DIPG13 and SU-DIPG36, writer and mark from the same nuclei and the same pipeline** [E1][E2]: | | SU-DIPG13 | SU-DIPG36 | |---|---:|---:| | Spearman rho, SUV39H1 vs H3K9me3, over 59,821 CTCF anchors | **0.148** | **0.078** | | Spearman rho over 200,000 random 147 bp tiles | 0.159 | 0.085 | | CTCF anchors with **zero** SUV39H1 | **55.9%** | **63.3%** | | CTCF anchors with **zero** H3K9me3 | 25.5% | 47.2% | | median H3K9me3 percentile at the top 1% of SUV39H1 anchors (n=598) | 76.6 | 61.1 | | top-1% SUV39H1 anchors whose H3K9me3 is **below the 50th percentile** | **25.4%** | **39.1%** | **So the writer's strongest sites are modestly but not strikingly mark-enriched, and a quarter to two-fifths of them carry no more H3K9me3 than a median CTCF anchor.** A rank correlation of 0.08-0.15 between an enzyme and its own catalytic product, in one nucleus, is not what a simple write-and-stay model predicts. **Three readings, and this measurement does not choose between them.** (i) SUV39H1 has non-catalytic residency - it is at sites for scaffolding or complex reasons and deposits nothing there. (ii) The CST #8729S antibody is pulling something other than, or in addition to, SUV39H1; **the deposit has no IgG and no knockout track**, so specificity is unverifiable from these files. (iii) The mark is long-lived and mobile relative to the enzyme, so a snapshot of enzyme and a snapshot of product need not coincide. The corpus entry that posed the question already said only a knockout control can separate (i) from (ii), and it is right. **And the correlation is essentially the same over random tiles as over CTCF anchors** (0.159 vs 0.148; 0.085 vs 0.078), so the weak coupling is a genome-wide property of these tracks and not something peculiar to anchors. **Lineage, first clause: this is paediatric H3K27M glioma.** It says nothing about a blood cell, and in a cell type whose defining lesion rewrites the H3K27 landscape it is not safe to assume the H3K9 machinery behaves as it does elsewhere.

Go deeper
Why this is worth an entry even though it is a correlation

This graph's whole H3K9 argument has been writer-to-phenotype with the reader and the direct writer-mark link missing - mark-h3k9me3 says so in its own open questions, and absence-no-h3k9-writer-or-reader-localised-at-the-kmt2a-locus says every published statement assigning the mark at a locus to a particular enzyme is an inference from enzyme perturbation, never from enzyme occupancy.

This is the first direct measurement of the writer-mark relationship in a human cell in this corpus, and it comes out weak. That does not overturn the perturbation literature - a small mobile fraction of enzyme can maintain a large stable domain - but it removes the intuition that occupancy and mark are interchangeable readouts, which several entries here have leaned on.

It is reported as a correlation and it is not written as a hypothesis test. The pre-registration declared Question C descriptive before running, precisely so that a large n could not be mistaken for a strong effect: at n = 59,821 a rho of 0.078 is overwhelmingly 'significant' and almost entirely uninformative about any individual locus.

The same dissociation, at the one locus this project cares about

At the KMT2A exon-11 CTCF anchor in SU-DIPG13, H3K9me3 reads 1.6301 - the 57.8th percentile of class-matched CTCF anchors and 2.29x the random-tile median - while SUV39H1 reads 0.0000, an explicit zero record. In SU-DIPG36 the mark reads 0.8356 (63.1st percentile) and the writer 1.6660 (74.2nd).

So at this element, in one of the two lines, the mark is present at an unremarkable-to-middling level and the enzyme is absent entirely. That is a single-locus instance of the genome-wide pattern above and it is reported as such - one interval, one sample, no replicate.

Facts and figures
scriptanalyses/q114-suv39h1-at-the-kmt2a-locus/run_bigwig_posthoc.py
statusrun
run date2026-09-22
run byQ114-suv39h1-at-the-locus
result summarySpearman rho between SUV39H1 and H3K9me3 over 59,821 same-nucleus CTCF-anchor windows is 0.148 (SU-DIPG13) and 0.078 (SU-DIPG36); 25.4% and 39.1% of the top 1% of SUV39H1 anchors have H3K9me3 below the 50th percentile.
n top1pct598
statistic typedescriptive - declared descriptive in the pre-registration before running; no hypothesis test is performed and none should be read in
cell typeSU-DIPG13 and SU-DIPG36, patient-derived H3K27M diffuse midline glioma neurospheres
cell stateresting, unirradiated neurosphere culture
lineagepaediatric glioma - NOT haematopoietic
epigenetic readoutSUV39H1 occupancy and H3K9me3 occupancy, same nuclei, same CUT&RUN pipeline, same deposit
shown for fusionNO - nobody has looked in any KMT2A fusion system
resolvesthe first open question of exp-suv39h1-and-h3k9me3-cutandrun-in-two-patient-derived-diffuse-midline-glioma-lines, and the GSE225028 coincidence question of ana-q113-geo-holds-human-suv39h1-occupancy-data-that-the-encode-only-census-could-not-see
inputs
[
  "GSM7037531",
  "GSM7037529",
  "GSM7037532",
  "GSM7037530",
  "ENCFF685KTA"
]
spearman ctcf anchors
{
  "SU-DIPG13": 0.1481511100832043,
  "SU-DIPG36": 0.07817718947070826
}
spearman random tiles
{
  "SU-DIPG13": 0.1590448387575201,
  "SU-DIPG36": 0.08501189289418844
}
frac anchors zero writer
{
  "SU-DIPG13": 0.5585329566540178,
  "SU-DIPG36": 0.6329549823640528
}
frac anchors zero mark
{
  "SU-DIPG13": 0.25467645141338324,
  "SU-DIPG36": 0.4720750238210662
}
median mark percentile at top1pct writer sites
{
  "SU-DIPG13": 76.61,
  "SU-DIPG36": 61.09
}
frac top1pct writer sites with mark below median
{
  "SU-DIPG13": 0.254,
  "SU-DIPG36": 0.391
}
How this could bite us back
  • If a writer is routinely found off its own mark, then targeting the writer and targeting the mark are not the same intervention, and a drug that removes the enzyme may leave a long-lived mark in place - or remove a non-catalytic function nobody has assayed. Chaetocin trials in glioma are being argued on a catalytic rationale that this measurement does not support.
  • The same logic runs the other way for safety: an agent that lowers H3K9me3 globally is not exonerated by showing that SUV39H1 is absent from a locus, because the mark at that locus is present without the enzyme.
Still open
  1. Is the weak writer-mark coupling a property of SUV39H1 or of this antibody? Only a SUV39H1-knockout or -degron CUT&RUN with the same CST #8729S clone separates them, and the deposit contains neither. Until it exists, every SUV39H1 occupancy number in this corpus - including this beat's negative at the KMT2A anchor - carries an unverifiable specificity assumption.
  2. Does the same dissociation hold in MCF-7, where SUV39H1 and H3K9me3 are also in one deposit? The Q114 MCF-7 arm computes the identical statistic; if the two systems agree, the dissociation is a property of the enzyme rather than of a glioma line.
  3. Would the correlation rise at the scale SUV39H1 actually works on? H3K9me3 domains are tens to hundreds of kilobases; this statistic is 147 bp windows. A domain-scale recomputation is cheap and was not run, and a low correlation at 147 bp is compatible with a high one at 100 kb.
Every source behind this entry (3)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    GSE225028 deposits four hg38 bigWigs from patient-derived H3K27M diffuse midline glioma neurospheres - SUV39H1 (CST #8729S) and H3K9me3 (CST #13969S), each in SU-DIPG13 and SU-DIPG36 - aligned with Bowtie2 v2.3.4.1, and no peak file is deposited.

    What this evidence cannot show

    Per-sample SOFT records fetched 2026-09-22. There is no IgG and no SUV39H1-knockout track in the deposit, so antibody specificity cannot be checked from these files, and each antibody was run once per line - there are no replicates. The two H3K9me3 bigWigs carry a 455-entry chrom table and the two SUV39H1 bigWigs a 195-entry one, in the same deposit.

    Evidence that argues against this (1)
    • GEO metadata states a pipeline; it does not certify that the pipeline ran. The build was therefore re-verified from each file's own chrom table rather than from these lines, and all four are GRCh38 across chr1, chr2, chr11, chr19 and chrX. source
  2. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    The null, the threshold, the controls and a NEGATIVE prior were written to disk before any signal value at the KMT2A locus was read from either deposit, and every deviation from them is appended to the same file rather than edited into it.

    Q114 pre-registration - SUV39H1 occupancy at the KMT2A locus, in the only two human systems where it has ever been measured · Q114-suv39h1-at-the-locus · oncofusion-knowledge-graph (analyses/q114-suv39h1-at-the-kmt2a-locus/PREREGISTRATION.md) 2026
    What this evidence cannot show

    The pre-registration discloses one prior read - ChIP-Atlas's hg38 SUV39H1 target-gene list, which derives from the MCF-7 samples and does not touch this deposit - and it declares before running that C4 cannot be run in this arm at all. Six deviations (D1-D6) are appended, three of them failures of the author's own controls.

    Evidence that argues against this (1)
    • A pre-registration written by the agent that then runs the analysis is weaker than an external one. What it buys is only that the threshold could not move after the numbers appeared; the appended deviations are the audit trail, and two of them record controls that failed and were not re-tuned. source
  3. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The class-matched null is 59,821 primary-assembly CTCF point-source summits from ENCODE optimal-IDR CTCF peaks in SK-N-SH (ENCFF685KTA, ENCSR541AMF, GRCh38) - a neural line, because no CTCF experiment exists in any DIPG or H3K27M glioma line in any archive.

    What this evidence cannot show

    SK-N-SH is a neuroblastoma line, not a diffuse midline glioma. 71 of 59,892 peaks fall on unlocalised or alt scaffolds and were dropped so that the two arms query an identical universe; that 0.12% of the set contributed a large share of the highest SUV39H1 values, which is itself reported below.

    Evidence that argues against this (1)
    • A cross-cell-type null may include anchors that are not anchors in DIPG, which would lower the comparison distribution and INFLATE the target's percentile. That bias runs against the negative reported here, not in its favour. source
Connected to