Oncofusion

start / graph / ana-q116-hp1a-is-depleted-at-the-exon-11-anchor-and-its-phosphomimetic-nte-mutant-reads-higher-than-its-phospho-dead-one-at-every-interval

Analysisweak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.Q116-h3k9-axis-at-the-locus

RUN: endogenous HP1a/CBX5 is DEPLETED at the exon-11 anchor (median 23.94th percentile) and has the lowest KMT2A gene-body reading of any protein in the panel (30.16) - while the phosphomimetic Bio-ChIP mutant 4E reads higher than the phospho-compromised 4A at all seven intervals AND that difference falls inside the genome-wide interquartile range of the same comparison, so it carries no information

also called HP1a at the KMT2A locus · the 4A/4E pair, reported separately from wild type · the reader is not at the anchor

The protein that reads the silencing tag is scarce across the whole KMT2A gene and scarcest at the spot where it breaks, agreeing with the tag itself. A version engineered to read the tag better did stick more, but by no more than two such experiments differ from each other anywhere in the genome, so that part means nothing.

The same thing, for a scientist

**Three different assays, reported separately and never pooled**, as the dispatch brief required. **1. Endogenous HP1a, antibody ChIP (GSM4133826 -E2, GSM4133827 +E2).** At the 146 bp exon-11 anchor: **47.89 and 0.00** of 56,171 same-cell-type CTCF summit windows, median **23.94** - **DEPLETED** by the pre-registered rule, the only protein antigen in the panel to score DEPLETED at the anchor. Over the 90 kb gene body: **35.76 and 24.56**, median **30.16** - the **lowest of any protein antigen in a 28-track panel whose median is 61.39** [E1]. **2. The CBX5 Bio-ChIP mutant pair (GSM4133790 = 4A, GSM4133791 = 4E), one track each, -E2 only.** The deposit calls these "mutant 4A" and "mutant 4E" and defines neither. The paper does: they are **phospho-compromised alanine** and **phospho-mimetic glutamic acid** substitutions at four phosphorylatable serines of HP1a's **N-terminal extension (NTE)**, and the paper's own figure legend states that the NTE phosphorylation **regulates the chromodomain-H3K9me3 tail interaction** [E2]. **4E reads higher than 4A at every single one of the seven intervals:** | interval | 4A (phospho-dead) | 4E (phosphomimetic) | difference | |---|---:|---:|---:| | exon11 (146 bp) | 65.60 | **76.13** | +10.53 | | kmt2a_gene (90 kb) | 44.04 | **60.98** | +16.94 | | kmt2a_tss_2kb | 96.76 | **99.75** | +2.99 | | median of the six non-anchor intervals | 49.79 | **72.37** | **+22.58** | **The direction fits the chemistry and the geometry says it is not about the anchor.** The largest difference is in the *non-anchor* median (+22.58), and every interval moves the same way. So the correct reading is **"4E is relatively more enriched over the whole *KMT2A* locus than 4A"**, not "4E binds the exon-11 anchor." By the pre-registered rule 4A is **ORDINARY** and 4E is **NEITHER** (76.13 clears the 75th percentile but with 1 of 1 track and a non-anchor median of 72.37, the specificity clause fails) [E1]. **A CORRECTION TO MY DISPATCH BRIEF.** The brief calls 4A and 4E "a chromodomain-binding-competence pair". They are **NTE phospho-mutants**, not chromodomain mutants. The brief's *effect* is right - the pair does differ in chromodomain-H3K9me3 binding competence - but its *mechanism* is wrong, and the distinction matters because an NTE phospho-mutant retains an intact chromodomain and can still be recruited by protein-protein contacts that a chromodomain-dead mutant could not [E2]. **And the thing that makes all of this weak: they are three different assays.** Endogenous antibody ChIP gives 30.16 over the gene body; Bio-ChIP of a BAP-tagged, BirA-biotinylated, **overexpressed** HP1a gives 44.04 and 60.98. **The wild-type and the mutants are not comparable to each other**, and the only comparison this entry makes between the two mutants is between two single tracks in the same assay format. **RUN THE SAME DAY, AND IT KILLS THE MUTANT COMPARISON.** The open question below asked for the genome-wide distribution of 4E-minus-4A percentile over all 56,171 anchor windows, and said the two vectors were cached and the computation was one line. It was. Result: the **genome-wide median 4E-minus-4A difference is -0.19 percentile points** and **48.79% of anchors have 4E above 4A** - so there is no global shift between the two constructs, which is good. But the **interquartile range of that difference is -22.36 to +27.72 percentile points**, and *KMT2A*'s differences (exon11 **+10.53**, gene body **+16.94**, non-anchor median **+22.58**) all sit **inside** it. **The 4E-over-4A difference at the *KMT2A* locus is not distinguishable from the anchor-to-anchor scatter of the same comparison genome-wide.** The direction fitting the chemistry was a coincidence at this n, and the ordered table above should be read as such.

Go deeper
Why the wild-type depletion is the more interesting half

HP1a is the reader of H3K9me3, and in the same cell line H3K9me3 over the same 90 kb reads the 2.56th-15.00th percentile across eight tracks (median 6.53). Reader and mark agree: both are depleted over the KMT2A gene body, and the reader is the only protein antigen in the panel that is.

That is the one place in this beat where two independent antigens agree in the direction the biology predicts, and it is worth more than the mutant comparison. Everything else in the panel - SUV39H1, EHMT2, TRIM28, KDM4B, KDM4C - reads the locus up, which the panel baseline says is a property of the locus. CBX5 and H3K9me3 read it down, together.

And it matches Q110's peak-level result in blood: zero called CBX5 peaks in the KMT2A gene or within 50 kb in K562 and GM12878, and zero for CBX1 and CBX3 as well. Three cell types, two assay formats, and the reader is absent from this gene in all of them.

What the mutant pair would need to be worth anything

One track each, one condition, and the difference is inside the genome-wide scatter. The honest cost list:

  1. Replicates. The deposit has one 4A and one 4E track. With an interquartile range of -22.36 to +27.72 percentile points for the 4E-minus-4A difference across 56,171 anchors, a single pair cannot resolve a 10-23 point effect at one locus.
  2. A domain-scale comparison. HP1a spreads over domains, and the two constructs differ in affinity rather than in address. A 100 kb-window comparison might separate them where a 146 bp one cannot - and the 90 kb gene-body delta (+16.94) is the largest of the interval deltas, which is consistent with that.
  3. A construct-level measurement. Nothing in the deposit shows that 4A and 4E express and biotinylate equally. The 4A track has a higher zero fraction over the anchor set (8.54% against 5.58%), which is what a less-abundant or less-efficiently-captured construct looks like as well as what a lower-affinity one looks like.

None of that is a criticism of the deposit. These samples were made to study oestrogen-induced pause release at ER target promoters, where the depositors had the replicates and the controls they needed. This beat is reading them somewhere else.

Facts and figures
scriptanalyses/q116-h3k9-axis-at-the-locus/run116.py, score.py and promoter_specificity.py
prereganalyses/q116-h3k9-axis-at-the-locus/PREREGISTRATION.md, section 8
statusrun
run date2026-09-23
run byQ116-h3k9-axis-at-the-locus
complete runtrue
mutant pair direction4E above 4A at ALL SEVEN intervals; largest difference in the six-non-anchor median (+22.58), smallest at the promoter (+2.99). This is a locus-wide relative-enrichment difference, NOT an anchor difference.
what 4A and 4E actually arephospho-compromised alanine (4A) and phospho-mimetic glutamic acid (4E) substitutions at phosphorylatable serines of HP1a's N-TERMINAL EXTENSION, which regulates the chromodomain-H3K9me3 interaction. NOT chromodomain mutations. Residue numbers are in the paper's figure panel and were not read.
correction to the dispatch briefThe brief calls 4A/4E 'a chromodomain-binding-competence pair'. They are NTE phospho-mutants with an intact chromodomain. The functional effect is on chromodomain-H3K9me3 binding; the mutations are elsewhere.
reader and mark agreeCBX5 is the only protein antigen in the 28-track panel that reads the KMT2A gene body DOWN (30.16 against a panel median of 61.39), and H3K9me3 - the mark it reads - is at 2.56-15.00 across eight tracks. Every other protein in the panel reads the locus up.
accession notedata/symbol-collisions.json was grepped for CBX5 before the first query. The CBX/HP1 row records that CBX1 (185 aa), CBX3 (183 aa) and CBX5 (191 aa) are within 8 residues, that no UniProt name readback distinguishes them, and that the prescription is to resolve by accession - CBX5 is P45973. The antigen here is taken to be CBX5 on the deposit's own label only; no clone is stated.
localizationchromatin at chr11q23.3 - the seven stored KMT2A intervals
subnuclear localizationnot measured
cell typeMCF-7 (ER+ breast adenocarcinoma). NOT haematopoietic, NOT KMT2A-rearranged.
cell statehormone-deprived culture and 40 min oestradiol for wild type; hormone-deprived only for both mutants, in cells stably overexpressing a BAP-tagged construct plus BirA
epigenetic readoutoccupancy of the H3K9me3 reader HP1a/CBX5, and of two NTE phospho-mutants of it, at the KMT2A locus
shown for fusionNO - MCF-7 carries no KMT2A rearrangement, and Q110's CBX5 zeros are in K562 and GM12878, neither of which is rearranged either
absence classmeasured-not-analysed
absence class noteFour tracks public since 2019, never read at this locus. The mutant pair is the only chromodomain-competence perturbation of an H3K9me3 reader in any archive and nobody had opened it.
three assays never pooled
Bio_ChIP_mutant_4A_phosph…49.8Bio_ChIP_mutant_4A_phospho_compromised_alanine: 49.79endogenous_antibody_ChIP51.8endogenous_antibody_ChIP: 51.76Bio_ChIP_mutant_4E_phosph…72.4Bio_ChIP_mutant_4E_phosphomimetic_glutamate: 72.37
results
[
  "analyses/q116-h3k9-axis-at-the-locus/results.json",
  "analyses/q116-h3k9-axis-at-the-locus/results_scored.json",
  "analyses/q116-h3k9-axis-at-the-locus/results_promoter_specificity.json"
]
three assays never pooled — the numbers behind the chart
{
  "endogenous_antibody_ChIP": {
    "tracks": [
      "GSM4133826 -E2",
      "GSM4133827 +E2"
    ],
    "exon11_percentiles": [
      47.89,
      0
    ],
    "exon11_median": 23.94,
    "verdict": "DEPLETED",
    "kmt2a_gene_percentiles": [
      35.76,
      24.56
    ],
    "kmt2a_gene_median": 30.16,
    "kmt2a_tss_2kb_percentiles": [
      96.32,
      96.97
    ],
    "kmt2a_promoter_pct_of_2000_active_promoters": [
      91.05,
      94.5
    ],
    "x_median_active_promoter": [
      1.53,
      1.71
    ],
    "six_nonanchor_interval_median": 51.76,
    "C1_ratios": [
      9.6,
      9.8
    ]
  },
  "Bio_ChIP_mutant_4A_phospho_compromised_alanine": {
    "track": "GSM4133790 -E2",
    "exon11_percentile": 65.6,
    "verdict": "ORDINARY",
    "kmt2a_gene_percentile": 44.04,
    "kmt2a_tss_2kb_percentile": 96.76,
    "six_nonanchor_interval_median": 49.79,
    "C1_ratio": 9.5,
    "promoter_specificity_pass": "NOT RUN for this track"
  },
  "Bio_ChIP_mutant_4E_phosphomimetic_glutamate": {
    "track": "GSM4133791 -E2",
    "exon11_percentile": 76.13,
    "verdict": "NEITHER - clears the 75th percentile but with 1 of 1 track and a non-anchor median of 72.37 the specificity clause fails",
    "kmt2a_gene_percentile": 60.98,
    "kmt2a_tss_2kb_percentile": 99.75,
    "six_nonanchor_interval_median": 72.37,
    "C1_ratio": 11.6,
    "promoter_specificity_pass": "NOT RUN for this track"
  }
}
genome wide 4E minus 4A test
{
  "n_anchors": 56171,
  "median_delta_percentile": -0.185,
  "mean_delta_percentile": 1.742,
  "frac_anchors_4E_above_4A": 0.4879,
  "IQR_delta_percentile": [
    -22.36,
    27.72
  ],
  "frac_zero_4A": 0.0854,
  "frac_zero_4E": 0.0558,
  "kmt2a_deltas": {
    "exon11": 10.53,
    "kmt2a_gene": 16.94,
    "kmt2a_tss_2kb": 2.99,
    "six_nonanchor_median": 22.58
  },
  "verdict": "NO global shift between the constructs (median -0.19, 48.79% above) AND the KMT2A differences all fall inside the genome-wide interquartile range of -22.36 to +27.72. The 4E-over-4A difference at this locus is NOT distinguishable from anchor-to-anchor scatter.",
  "caveat": "the comparison distribution is 146 bp anchor windows, so the exon11 delta (+10.53) is width-matched and the 90 kb gene-body delta (+16.94) is not; the width-matched one is the more clearly inside.",
  "results_file": "analyses/q116-h3k9-axis-at-the-locus/results_cbx5_4a_4e.json"
}
How this could bite us back
  • hyp-the-hoxa-meis-independent-cbx5-programme-is-an-unasked-menin-inhibitor-resistance-route asks whether CBX5's role in that programme is chromodomain-dependent, and says that if it is not, the hypothesis has nothing to do with H3K9me3. This entry does not answer that - it measures CBX5 at ONE locus in a breast line - but it does show that the chromodomain-competence pair exists as public data in a usable format, which lowers the cost of asking.
  • An intervention aimed at HP1a at the KMT2A locus has no target by this measurement: the endogenous reader is the most depleted protein antigen in the panel over this gene body and DEPLETED at the anchor. Anything that looks like an HP1-directed effect at this locus would more likely be acting through the mark's absence elsewhere, or through CBX1/CBX3, for which this graph holds no entry at all.
Still open
  1. ANSWERED THE SAME DAY, NEGATIVELY: the genome-wide 4E-minus-4A difference has a median of -0.19 percentile points and an IQR of -22.36 to +27.72 over 56,171 anchors, and every KMT2A difference falls inside it. What remains open is whether a DOMAIN-scale comparison (the mutants over 100 kb windows rather than 146 bp anchors) would separate them, since HP1a spreads over domains and the two constructs differ in affinity rather than in address.
  2. Which four serines? The paper's figure panel has the residue numbers and this entry did not read them. The HP1a NTE serine cluster is the standard site in the literature, but naming residues from a general expectation rather than from the figure would be exactly the kind of identifier error this project keeps catching.
  3. Do the two constructs express and biotinylate equally? Nothing in the deposit measures construct level, so a uniform relative-enrichment difference between 4A and 4E could be a construct-abundance difference. This is the main alternative explanation and it is untested.
  4. Is HP1a at the KMT2A gene body in ANY cell type? Three cell types and two assay formats now say no - MCF-7 at the 30.16th percentile, and zero called peaks in K562 and GM12878. If the reader is never there, the H3K9me3 story at this locus has a reader-shaped hole in it even where the mark is present.
Every source behind this entry (3)
  1. E1weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.dataset

    Endogenous HP1a/CBX5 antibody ChIP in MCF-7 reads the 47.89th and 0.00th percentile of same-cell-type CTCF summit windows at the 146 bp exon-11 anchor (median 23.94, DEPLETED) and the 35.76th and 24.56th over the 90 kb gene body (median 30.16, the lowest protein-antigen value in a 28-track panel whose median is 61.39). The Bio-ChIP mutants give exon11 65.60 (4A) and 76.13 (4E), gene body 44.04 and 60.98, and six-non-anchor-interval medians of 49.79 and 72.37 - 4E above 4A at all seven intervals.

    What this evidence cannot show

    Confidence is WEAK and the reasons are structural, not incidental. THREE DIFFERENT ASSAYS: endogenous antibody ChIP for wild type, and Bio-ChIP of an OVEREXPRESSED BAP-tagged BirA-biotinylated construct for each mutant - these are not comparable to each other and are reported separately here for that reason. ONE TRACK PER MUTANT, -E2 only, no replicate, so the 4E-versus-4A difference rests on two numbers. Wild type has one track per condition and its exon-11 values are 47.89 and 0.00, a 47.9-point spread. MCF-7 only, not KMT2A-rearranged, not haematopoietic. No input, no IgG, no antibody clone. The promoter-specificity pass (percentile among 2,000 active promoters) was run for the wild-type tracks (91.05 and 94.50) and NOT for the two mutants, so their promoter numbers are only against CTCF anchors and are not specificity-controlled. data/symbol-collisions.json was grepped for CBX5 first: the CBX/HP1 row warns that CBX1, CBX3 and CBX5 are within 8 residues of each other and that no name readback distinguishes them, so the antigen here is taken to be CBX5/P45973 on the deposit's own label and nothing more.

    Evidence that argues against this (2)
    • The 4E-over-4A difference is LARGEST in the non-anchor median (+22.58 points) and present at all seven intervals. That is the signature of a locus-wide or genome-wide difference in relative enrichment, not of anchor binding, and it could equally reflect a difference in how well the two constructs express, biotinylate or immunoprecipitate. Nothing in the deposit measures construct level. The direction fits the published chemistry, which is precisely why it should not be over-read. source
    • RUN AND NEGATIVE. The genome-wide 4E-minus-4A percentile difference over 56,171 anchors has a median of -0.19 and an interquartile range of -22.36 to +27.72, and every KMT2A difference (+2.99 to +22.58) is inside that range. So the mutant comparison at this locus carries no information, and this entry's mutant table is reported as a measurement that failed its own control rather than as a finding. source
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The CBX5 4A and 4E constructs are phospho-compromised alanine and phospho-mimetic glutamic acid substitutions at phosphorylatable serines of HP1a's N-terminal extension, and that NTE phosphorylation regulates the chromodomain-H3K9me3 tail interaction. Both were expressed in MCF-7 as BAP-tagged constructs with BirA and read by streptavidin Bio-ChIP.

    Shape of promoter antisense RNAs regulates ligand-induced transcription activation · Yang F, Tanasa B, Micheletti R, Ohgi KA, Aggarwal AK, Rosenfeld MG · Nature 2021
    PMID 34194047DOI
    What this evidence cannot show

    Read via PMC8439151 on 2026-09-23. The load-bearing sentence is a figure legend: 'Domain map of human HP1a. Phosphorylatable serine residues, and the corresponding phosphor-mimetic glutamic acid residues and phosphor-compromised alanine residues at the N-terminal extension (NTE) of human HP1a regulating its chromodomain (CD)-H3K9me3 tail interaction are in bold.' The RESIDUE NUMBERS are in the figure itself, which was not inspected, so this entry does not state which four serines they are. The methods confirm 'BAP-HP1a (4A), BAP-HP1a (4E) or BAP-dCas9-gRNA' with BirA. CORRECTS MY DISPATCH BRIEF, which describes the pair as a 'chromodomain-binding-competence pair' - the effect is chromodomain binding, the mutations are not in the chromodomain.

    Evidence that argues against this (1)
    • Because the mutations are in the NTE and not in the chromodomain, 4A retains a functional chromodomain and can still be recruited to H3K9me3 or by protein-protein contact through its chromoshadow domain. So a smaller 4A signal is a reduced-affinity statement, not a loss-of-reading statement, and no experiment here distinguishes reduced H3K9me3 reading from reduced overall chromatin residence. Bannister AJ et al. 2001
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    In the same MCF-7 cells, H3K9me3 over the 90 kb KMT2A gene body reads the 2.56th-15.00th percentile of width-matched CTCF anchors across eight tracks (median 6.53), and Q110 found zero called CBX5 peaks in the KMT2A gene or within 50 kb in K562 and GM12878, with zero also for CBX1 and CBX3.

    analyses/q116-h3k9-axis-at-the-locus/results.json and ana-q110-h3k9-writer-and-reader-occupancy-at-the-kmt2a-locus · Q116-h3k9-axis-at-the-locus and Q110-h3k9me3-writer-layer · oncofusion-knowledge-graph 2026
    What this evidence cannot show

    The reader-and-mark agreement is a direction shared by two antigens in one cell line, and both were measured by the same submitter with the same pipeline, so a shared pipeline effect is not excluded. Q110's peak-level zeros are in two blood lines, neither KMT2A-rearranged. The K562 CBX5 peak set has zero peaks in the ZNF274 positive control window too, which that entry flags.

    Evidence that argues against this (1)
    • CBX5's depletion over the KMT2A gene body could simply be the mirror of the panel-wide elevation: if every other antigen is up at this locus for a technical reason, an antigen that is genuinely down would read down, and one that is genuinely flat would also read down relative to the panel. The panel baseline cannot separate 'CBX5 is depleted' from 'CBX5 is normal and everything else is inflated'. The 30.16 figure should be read as a rank within this deposit, not as an absolute depletion. source
Connected to