In murine MLL-AF9, knocking out PCGF1 changes the menin inhibitor's effect on H3K79me2 by 0.0166 of a share ratio and the DOT1L inhibitor's by 0.0101 - both smaller than the 0.0584 two control cultures of the same line already differ by, so the mark's drug response runs outside PRC1.1
also called the PCGF1 knockout arm · H3K79me2's drug response is PCGF1-independent · hyp-q117 prediction 5, the clause that could be scored
Deleting PCGF1 - a component of the repressive complex that moves into the spots the fusion protein vacates - does not change how either drug rearranges the H3K79 chemical tag on histones. The effect of removing the gene is smaller than the difference between two untouched cultures of the same cells, so whatever the drugs do to this tag, they do not do it through that complex.
The same thing, for a scientist
**In murine MLL-AF9 (KMT2A::MLLT3) myeloid cells - not the infant B-lineage KMT2A::AFF1 disease this graph is about - a full 2 x 3 design deposited in GSE260742 but never analysed gives the first genotype test of H3K79me2's drug response in this corpus** [E1]. A non-targeting-sgRNA control culture (`NT`) and an isogenic PCGF1-knockout clone (`C1`) were each treated with DMSO 72 h, SGC0946 5 uM 72 h and VTP-50469 500 nM 48 h. Scored at `S_FLAG`, the tagged fusion's own top-0.1% sites - **a set defined in neither genotype**, from a FLAG track in a third culture, so the genotype comparison is not confounded with the set definition: | within-sample share ratio at `S_FLAG`, 1,228 bins of 1,227,708 | `NT` control | PCGF1-KO `C1` | \|difference\| | |---|---:|---:|---:| | **VTP-50469 500 nM 48 h / DMSO 72 h** (`V`) | **0.8156** | **0.7990** | **0.0166** | | **SGC0946 5 uM 72 h / DMSO 72 h** (`G`) | **0.1415** | **0.1314** | **0.0101** | | **F = (1 - V)/(1 - G)**, lower bound | **0.2148** | **0.2314** | 0.0166 | **Both pre-registered criteria fire and both refutations fail.** **P1** (*the menin inhibitor's effect is PCGF1-independent*, fires if `|V_KO - V_NT| < 0.0584`) **FIRES at 0.0166**. **P3** (*direct DOT1L inhibition is PCGF1-independent*, same threshold) **FIRES at 0.0101**. **P2** (PCGF1-dependence, fires at `>= 0.2089`) does not fire, and **RP1** - *PRC1.1 is upstream of the mark and the depositing paper's one-way frame is incomplete* - **DOES NOT FIRE**. **The thresholds are measured quantities, not chosen ones.** 0.0584 is Q122's own F3-against-F3_NT absolute difference at this exact set in this exact line, read from its committed `results.json`; 0.2089 is Q120's RS4;11-against-MOLM13 difference, the effect size `hyp-q120` is built on. So P1 asks *is the genotype effect bigger than the noise this line already showed us*, and the answer does not depend on a number anybody invented. **The knockout is not inert in every respect, which is what makes the null interpretable.** The two genotypes' H3K79me2 landscapes differ substantially at baseline - the Jaccard overlap of their DMSO top-0.1% sets is **0.6915** (1,004 of 1,452) and the Spearman of their DMSO share vectors over the whole universe is **0.5040**. **RP3** - *the knockout is not a knockout*, declared in advance so the cleanest-looking result would get the most scepticism - **DOES NOT FIRE**: the genotypes are distinguishable, they are simply indistinguishable in how they respond to either drug.
What the pre-registered criteria did
Scored exactly as written in analyses/q124-the-timecourse-and-the-pcgf1-knockout/PREREGISTRATION.md section 5, before any bin file for this beat existed on disk.
| id | criterion | verdict |
|---|---|---|
| I0 | no track is one-artefact dominated (largest bin < 5% of non-blacklisted total) | PASS - worst is 0.00013 |
| I1 | build called from each file's own extents against mm10/mm39/hg38/hg19 | PASS - all 14 files call mm10, chr1-19+X, no chr20 |
| I2 | the positive control is visible to a within-sample instrument (G < 0.95) | PASS - 0.1283 (NT), 0.1248 (KO) |
| I3 | the universe reproduces Q122's exactly | PASS - 1,227,708 bins and k = 1,228, asserted in code against its committed JSON |
| P1 | the menin inhibitor's effect is PCGF1-independent | FIRES (0.0166 against 0.0584) |
| P2 | the effect is PCGF1-dependent | does not fire |
| P3 | direct DOT1L inhibition is PCGF1-independent | FIRES (0.0101 against 0.0584) |
| P4 | Q122's F range replicates in a third and fourth culture | FIRES on S_FLAG; AT THRESHOLD on the stricter reading - see the correction below |
| P5 | the three-culture spread is smaller than the cross-partner difference | FIRES (0.0584 against 0.2089) |
| P6 | the knockout's baseline landscape is comparable | PASS (Jaccard 0.6915 against 0.30) |
| RP1 | PRC1.1 is upstream of the mark | DOES NOT FIRE |
| RP3 | the knockout is not a knockout | DOES NOT FIRE |
A correction to my own scoring, found by re-reading the criterion text against the output rather than by a number looking wrong. P4's text says 'S_FLAG and each genotype's own S_K79'; my code walked S_FLAG only. On S_FLAG alone F_NT = 0.2148 and F_KO = 0.2314, both inside the accepted [0.0447, 0.3000], so P4 fires. Adding each genotype's own S_K79, F_NT at S_K79nt is 0.3260, which exceeds the interval by 0.0260, so on the stricter reading P4 is AT THRESHOLD. Both readings are in results.json and the verdict carries both.
Why this is scored at S_FLAG and not at the mark's own sites
A genotype difference scored on a set defined inside one genotype confounds the comparison with the set definition, and here that is not hypothetical: the two genotypes' own H3K79me2 top-0.1% sets overlap at Jaccard 0.6915, so a third of each set is genotype-specific. S_FLAG is defined by a FLAG-DMSO track from a third culture and appears in neither scored contrast.
The own-set numbers are reported for completeness and they say the same thing more loosely: V_NT at S_K79nt is 0.7158 and V_KO at S_K79ko is 0.7926, a difference of 0.0768 - above P1's 0.0584 threshold but far below P2's 0.2089, i.e. UNDECIDED in the declared indifference region. The cross-scored values are tighter: at S_K79ko the two genotypes give 0.7942 and 0.7926, a difference of 0.0016. So the apparent 0.0768 at each genotype's own set is a set-definition effect and not a response difference, which is exactly what scoring at S_FLAG was pre-registered to avoid.
What this does NOT establish, stated as flatly as the result
- This is H3K79me2, not fusion occupancy and not silencing. There is no FLAG, MLL1 or MENIN track in the NTvKO arm of GSE260742, so nothing here says whether the fusion is evicted normally in the knockout.
- The knockout is unverified in the files analysed. The paper's key-resources sheet lists exactly one mouse guide -
Mouse PCGF1 sgRNA1 F CACCGCTGTGGATTTAGCATTAGT/R AAACACTAATGCTAAATCCACAGC- one clone (C1), no second clone and no rescue construct, and no PCGF1 protein or transcript measurement is deposited for these samples. A null result in a single unverified clone is the weakest form of a null. - A null in one channel is not a null in the biology, and the published literature says so directly - see the counter-evidence on E1.
- 48 h against 72 h.
F_NTandF_KOare lower bounds in the same sense and by the same mechanism as Q122's, because VTP is 24 h shorter than its own vehicle. - The null is what the published model predicts, not a surprise.
mechanism-menin-inhibitor-silencing-is-prc1-1-deposition-not-only-evictionrecords that the PRC1.1-dependent H2AK119ub under a menin inhibitor arises from progressive loss of DOT1L-mediated H3K79 methylation - the mark upstream, PRC1.1 downstream - and a downstream knockout cannot move an upstream quantity. So this measurement confirms an ordering the corpus already held rather than establishing a new one, and its value is the quantitative bound (0.0166 and 0.0101 against a 0.0584 two-culture spread) and the biomarker consequence, not the direction of the arrow. Found by runningcheck-citation-reuse.pyon this entry's own PMIDs, which returned 54 citations for PMID 39158067 and 43 for PMID 36635503.
| script | analyses/q124-the-timecourse-and-the-pcgf1-knockout/run.py |
|---|---|
| status | run |
| result summary | V 0.8156 (NT) against 0.7990 (PCGF1-KO), difference 0.0166; G 0.1415 against 0.1314, difference 0.0101; F 0.2148 against 0.2314. P1 and P3 fire against a 0.0584 threshold that is Q122's own measured two-culture spread. RP1 and RP3 do not fire. Jaccard of the two DMSO top-0.1% sets 0.6915, so the genotypes are distinguishable at baseline and indistinguishable in drug response. |
| cell line | murine MLL-AF9, GEO cell line field 'MLLAF9', cell type 'Myeloid', tissue 'Blood', Mus musculus taxid 10090, GPL24247 |
| fusion | KMT2A::MLLT3 (MLL-AF9), FLAG-tagged and retrovirally over-expressed |
| lineage | myeloid (AML model); NOT the infant B-lineage KMT2A::AFF1 disease |
| drug | VTP-50469 500 nM 48 h and SGC0946 5 uM 72 h, each against DMSO 72 h |
| genotypes | non-targeting sgRNA control (NT) and PCGF1-knockout clone C1, isogenic |
| genome build | mm10, verified from each of the 14 files' own per-chromosome declared extents against mm10, mm39, hg38 and hg19 |
| bin bp | 2000 |
| n bins universe | 1227708 |
| n held | 1228 |
| antibody | H3K79me2 ABCAM ab3594-100UG rabbit polyclonal; FLAG NEB 14793S clone D6W5B |
| epigenetic readout | H3K79me2 (ab3594) at the tagged fusion's own top-0.1% sites has a within-sample share ratio of 0.8156 under VTP-50469 in a non-targeting control culture and 0.7990 in an isogenic PCGF1 knockout; under SGC0946 it is 0.1415 and 0.1314 |
| mark written | H3K79me2, by DOT1L |
| writer of that mark | gene-dot1l |
| shown for fusion | KMT2A::MLLT3 in a murine FLAG-tagged over-expression model; nothing in this arm is human and nothing is KMT2A::AFF1. The H3K79me2 antibody cannot attribute the mark to the fusion at all, and the arm deposits no fusion-occupancy track, so the fusion enters only through the FLAG-defined set S_FLAG |
| absence class | measured-not-analysed |
| absence class note | The gap this entry closes is an ANALYSIS gap, not a data gap. GSE260742's NTvKO H3K79me2 2 x 3 design has been public since 2025-11-23; hyp-q117's prediction 5 named 'the depositing paper's NTvKO arm' as the relevant deposit in cycle 68 and recorded that the beat did not touch it; Q122 named the six accessions in its section 7 and declined them. Cost to close: eight downloads totalling 0.86 GB, one at a time with deletion after binning, and about ninety seconds of compute. |
| absence class assigned | 2026-09-23 (Q124) |
inputs
[ "GSM8123438 NTvKO Control - DMSO H3K79me2, 72hrs, ChIP_K79_NT_DMSO.tdf", "GSM8123439 NTvKO Control - SGC0946 5uM H3K79me2, 72hrs, ChIP_K79_NT_SGC.tdf", "GSM8123440 NTvKO Control - VTP50469 500nM H3K79me2, 48hrs, ChIP_K79_NT_VTP.tdf", "GSM8123435 NTvKO PCGF1 KO DMSO H3K79me2, 72hrs, ChIP_K79_C1_DMSO.tdf", "GSM8123436 NTvKO PCGF1 KO SGC0946 5uM H3K79me2, 72hrs, ChIP_K79_C1_SGC.tdf", "GSM8123437 NTvKO PCGF1 KO VTP50469 500nM H3K79me2, 48hrs, ChIP_K79_C1_VTP.tdf", "GSM8123416 NTvKO Control - DMSO input3 (robustness universe)", "GSM8123415 NTvKO PCGF1 KO DMSO input (robustness universe)", "GSM8549823 input batch 2, reused from C:/Temp/q117 - defines U_m, identical to Q117/Q120/Q122", "GSM8123359 FLAG DMSO, reused from C:/Temp/q117 - defines S_FLAG", "mm10-blacklist.v2.bed, 3,435 intervals, re-fetched and diffed byte-identical against Q117's copy", "mm10 chromosome sizes re-fetched live from api.genome.ucsc.edu, 0 mismatches against Q117's stored file" ]
- If H3K79me2's response to a menin inhibitor is unchanged in a PRC1.1-deficient cell, then the mark cannot report the resistance state that PRC1.1 deficiency creates. Zhou et al. show PCGF1 or BCOR loss confers menin-inhibitor resistance; this measurement says such a cell's H3K79me2 would look identical to a sensitive cell's at the fusion's own sites. Any trial using H3K79me2 as a pharmacodynamic or response biomarker would read a resistant patient as engaged.
- The same result cuts the other way for combination design. If removing a PRC1.1 subunit does not change what either drug does to H3K79me2, then a PRC1.1-directed agent and a DOT1L inhibitor are acting on non-overlapping chromatin axes, which is an argument FOR combining them rather than against - and PCGF1-deficient disease is exactly where a menin inhibitor is reported to fail.
- PCGF1 and BCOR loss-of-function are recurrent in myeloid malignancy, so a subset of patients starts with the genotype tested here. This entry says their H3K79me2 pharmacology is normal; Zhou et al. say their clinical pharmacology is not. Anyone reading the chromatin result as reassurance would have the wrong answer.
- A null in a single unverified CRISPR clone is the weakest form of a null, and this one is being used to score a standing prediction. If the C1 clone retains PCGF1 function, P1 and P3 are measuring nothing and the prediction remains untouched - and nothing in the deposit can distinguish those cases.
- Is the C1 clone a complete PCGF1 knockout? Nothing in the 158 sample records or the key-resources workbook measures PCGF1 in these samples, and the paper's MOESM1 is image-only with zero extractable characters, so the validation panel - if it exists - is outside this project's reach. Until it is read, this null rests on a single unverified clone from a single guide.
- Does H2AK119ub behave the same way? The same 2 x 3 design is deposited for H2AK119ub in THREE replicate conventions - GSM8123403/405/406 and 399/401/402 (paired), 410/411/412 and 407/408/409 (rep1), 420/421/422 and 417/418/419 (rep2) - so a PRC1.1 beat on this design could carry a spread that this beat cannot, and it is the only route to prediction 5's silencing clause.
- Zhou et al. report that PCGF1 loss confers menin inhibitor resistance through reactivation of bivalent noncanonical menin targets including MYC, and THOSE LOCI ARE PUBLISHED AS COORDINATES: Zhou supplemental Table 4, BLOOD_BLD-2023-023644-mmc4.xlsx, chr|start|end|gene id|group over 9,292 rows of which 1,733 are the Menin+H2AK119ub shared set, already located by ana-the-locus-list-existed-at-the-publisher-not-in-geo. Does H3K79me2 move at those loci in the knockout even though it does not move genome-wide at the fusion's sites? A locus-restricted re-score of these same bins would answer it with no new download, needing only a human-to-mouse ortholog map because Zhou's coordinates are hg38 and these bins are mm10. This beat did not attempt it.
- If the mark's response is PCGF1-independent and the transcriptional response is not, which chromatin channel IS the PRC1.1-dependent one? H2AK119ub is the obvious candidate and it is deposited; H3K27me3 (GSM8123423-428) and H3K4me3 (GSM8123429-434) are deposited too and neither has been binned by anyone.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
In murine MLL-AF9 (KMT2A::MLLT3) myeloid cells, 2 kb genome-wide bins from the deposited .tdf files, universe 1,227,708 bins (GSM8549823 input > 0, mm10 blacklist v2 removed, mm10 verified from each file's own per-chromosome declared extents), top 0.1% = 1,228 bins: at the FLAG-DMSO top-0.1% set the within-sample H3K79me2 share ratio under 500 nM VTP-50469 48 h against DMSO 72 h is 0.8156 in the non-targeting-sgRNA control culture (GSM8123438/GSM8123440) and 0.7990 in the isogenic PCGF1-knockout clone C1 (GSM8123435/GSM8123437), a difference of 0.0166; under 5 uM SGC0946 72 h against DMSO 72 h it is 0.1415 and 0.1314, a difference of 0.0101; F = (1-V)/(1-G) is 0.2148 and 0.2314. The two genotypes' own H3K79me2 DMSO top-0.1% sets overlap at Jaccard 0.6915 (1,004 of 1,452) and their DMSO share vectors correlate at Spearman 0.5040 over the universe.
Q124 re-analysis of GSE260742: the PCGF1-knockout 2 x 3 arm, eight files downloaded one at a time, binned at 2 kb and deleted. analyses/q124-the-timecourse-and-the-pcgf1-knockout/run.py, deviation1_enrichment_null.py, deviation2_common_support.py, deviation3_four_statistic_steps.py and results.json, run 2026-09-23 · Iqbal Lab oncofusion knowledge graph 2026What this evidence cannot show
THE KNOCKOUT IS UNVERIFIED IN THE FILES ANALYSED: no PCGF1 protein or transcript measurement is deposited for GSM8123435/436/437, the paper's key-resources sheet lists ONE mouse guide and the deposit has ONE clone (C1) with no second clone and no rescue, so a difference would be confounded with clonal drift and a null cannot exclude an incomplete knockout. THE VTP ARM IS 48 h AND ITS VEHICLE AND SGC ARM ARE 72 h, so F is a LOWER BOUND on the 72-hour effect. THIS ARM MEASURES H3K79me2 AND NOTHING ELSE - not fusion occupancy, not H2AK119ub, not transcription - so it reaches only the DOT1L-axis clause of hyp-q117's prediction 5. NO SPIKE-IN NORMALISATION IS POSSIBLE. Q117 verified zero hits for the entire spike-in vocabulary across all 419,824 characters of GSE260742's 158 sample records and no scale factor is deposited, so every cross-condition number here is a WITHIN-SAMPLE SHARE, which cancels the unknown per-sample scale factor exactly and IS BLIND TO A UNIFORM CHANGE BY CONSTRUCTION. Nothing here measures the mark's abundance; everything measures its distribution. n = 1 per antibody per condition per culture; no number carries a spread and none is precise. REAGENTS AND WHAT EACH CANNOT DISTINGUISH: H3K79me2 is ABCAM ab3594-100UG, 'Rabbit polyclonal to Histone H3 (di methyl K79) - ChIP Grade', read verbatim from the paper's key-resources sheet 41556_2025_1859_MOESM6_ESM.xlsx this session - a POLYCLONAL with no specificity document in the sheet, so it cannot distinguish me2 from me3, and it reads a MARK, so it cannot distinguish halted deposition from active demethylation from replicative dilution and cannot attribute any of them to the fusion. `antibod` occurs ZERO times in the 158 GEO records, so the spreadsheet is the only route to any reagent in this series. S_FLAG is defined by DYKDDDDK Tag (D6W5B) Rabbit mAb, NEB 14793S, on a RETROVIRALLY OVER-EXPRESSED tagged MLLAF9 construct, so it is the over-expressed molecule's sites and not an endogenous fusion's, and it cannot see a fusion whose tag has been proteolysed. VTP50469 disrupts an INTERFACE, so its H3K79me2 effect is at least four steps removed; SGC0946 inhibits DOT1L's CATALYSIS, so its effect says nothing about DOT1L occupancy. SYSTEM: murine MLL-AF9 (KMT2A::MLLT3) myeloid blood cells, NOT the infant B-lineage KMT2A::AFF1 disease this graph is about.
Evidence that argues against this (3)
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The PCGF1 knockout in GSE260742 was made with a single guide: the depositing paper's key-resources workbook, sheet Oligos, lists Mouse PCGF1 sgRNA1 F = CACCGCTGTGGATTTAGCATTAGT and R = AAACACTAATGCTAAATCCACAGC and no second mouse guide, against two guides for the human arm (Human PCGF1 sgRNA1 and sgRNA2); the H3K79me2 knockout samples are a single clone whose files are named C1 (GSM8123435/436/437) and whose H2AK119ub siblings are named CLONE1 (GSM8123417-419).
Neville D et al., supplementary key-resources table 41556_2025_1859_MOESM6_ESM.xlsx, sheets Antibody and Oligos, parsed from the workbook's own sharedStrings.xml on 2026-09-23 · Nature Cell Biology (supplementary information) 2026What this evidence cannot show
A key-resources table lists reagents, not which reagent produced which deposited sample, so the attachment of `Mouse PCGF1 sgRNA1` to GSM8123435/436/437 is an inference from the workbook plus the samples' murine organism field and their C1/CLONE1 naming, not a statement the deposit makes. The workbook is the ONLY route to any reagent in this series: `antibod` occurs zero times in the 419,824 characters of sample records. Whether the single clone is a complete knockout, a partial one or a hypomorph is not measured anywhere in the deposit or the workbook.
Evidence that argues against this (1)
cashes one of prediction 5's three clauses: in a PCGF1 knockout the menin inhibitor's H3K79me2 effect is unchanged (0.8156 against 0.7990, difference 0.0166 against a 0.0584 two-culture spread). The 'fusion is evicted normally' and 'vacated sites are not silenced' clauses are unreachable from an H3K79me2 arm
knocking out PCGF1 in murine MLL-AF9 changes H3K79me2's share ratio under a menin inhibitor by 0.0166 and under a DOT1L inhibitor by 0.0101, both below the 0.0584 that two control cultures of the same line differ by
H3K79me2's response to both a menin inhibitor and a DOT1L inhibitor runs outside PRC1.1: removing its PCGF1 subunit leaves both responses unchanged, while the published literature reports the same loss confers menin-inhibitor resistance transcriptionally
the first genotype test of this mark's drug response in this corpus: PCGF1-independent for both VTP-50469 and SGC0946 at the tagged fusion's own sites in murine MLL-AF9
F replicates in a third and fourth murine culture at 0.2148 and 0.2314, inside Q122's 0.0947-0.2500 range, with the same 48 h/72 h lower-bound qualifier
if silencing after eviction is PRC1.1 deposition, this arm says the H3K79me2 half of the same drug's chromatin action does not need PRC1.1 - the two halves are separable
two of the four cultures in the floor table are this beat's NT and PCGF1-KO arm, and the sign reversal is identical in both genotypes (VTP 0.7230/0.7620, SGC 1.9478/1.8905)
the measurement that scores the one reachable clause; this entry is the map of what that measurement does and does not settle
identifies the NT and C1 cultures the PCGF1 arm compares and supplies the third-culture value (NT, 0.8156) that makes the three-culture spread 0.0584
the single most likely way that null is wrong: no PCGF1 protein, transcript or locus measurement is deposited for GSM8123435/436/437, one guide, one clone, no rescue - so 'PCGF1 loss changes nothing' cannot be separated from 'PCGF1 was not lost'
the measurement the hypothesis is built on: 0.0166 and 0.0101 against a 0.0584 two-culture spread
the published resistance mechanism this null sits beside: PCGF1 or BCOR loss confers menin-inhibitor resistance, and H3K79me2's drug response is unchanged in the same genotype - so the mark cannot report that resistance state
predicts that null is a property of the aggregate statistic; UNTESTED, falsifier F2
first priority: six bin files already on disk, zero downloads