Oncofusion

start / graph / ana-the-serine-arm-replicates-in-a-third-aff1-line-from-another-lab-but-the-stress-output-genes-do-not

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.

The serine arm replicates in a third KMT2A::AFF1 line from an independent laboratory - PHGDH -2.77, PSAT1 -2.17, GPT2 -2.34, ASNS -1.16 - but the ISR output genes do not, and it survives PCGF1 and BCOR knockout

also called MV4;11 replication · GSE260456 · ASNS replication independent lab · PRC1.1 knockout menin response

A finding from one leukaemia cell line was retested in a third line, measured by a different laboratory asking a different question, and the amino-acid-making machinery switched off there too. The stress-response genes that had looked like part of the same story did not repeat, so that half of the original reading was probably wrong.

The same thing, for a scientist

`hyp-menin-inhibition-collapses-the-serine-asns-programme-and-should-resensitise-kmt2a-r-all-to-asparaginase` rested on ONE line (RS4;11) in ONE dataset with a panel that was not pre-declared. GSE260456 breaks all three dependencies: a different line (**MV4;11**, KMT2A::AFF1, already `line-mv4-11-kmt2a-aff1-myeloid` here), a different laboratory and paper (PMID 41634341, Neville/Gilan, *Nature Cell Biology* 2026), and a different question entirely (PRC1.1 and transcriptional memory). The panel was copied verbatim from the hypothesis entry's own data block. **The serine/one-carbon arm replicates.** Safe3 control sgRNA, VTP-50469 500 nM versus same-genotype DMSO, n=3, with the median housekeeping drift (**-0.33**) subtracted in the second column because ACTB is -0.73 here and library-size CPM inflates repressions: | gene | MV4;11 raw | minus housekeeping | RS4;11 d7 | |---|---:|---:|---:| | PHGDH | **-2.77** | -2.44 | -3.24 | | GPT2 | **-2.34** | -2.01 | -1.54 | | PSAT1 | **-2.17** | -1.84 | -2.40 | | CBS | **-1.89** | -1.56 | -4.07 | | SLC1A5 | -1.61 | -1.28 | -0.99 | | EIF4EBP1 | -1.37 | -1.04 | -1.39 | | **ASNS** | **-1.16** | **-0.83** | **-1.72** | | MTHFD2 | -1.03 | -0.70 | -1.16 | | ATF4 | -0.99 | -0.66 | -0.86 | **The positive control works**, which is what makes the rest interpretable: MEIS1 **-3.52**, MEF2C **-5.64**, PBX3 **-3.04**, HOXA10 **-1.53**. And the DOT1L inhibitor SGC0946 at 5 uM is much weaker on the same genes (PHGDH -0.88, ASNS -0.70, MEIS1 -0.63), which is a cleaner separation than RS4;11 gave at day 7. **The ISR output genes do NOT replicate, and that half of the original reading should be dropped.** TRIB3 -0.52 against -3.10; ATF3 **+0.22** against -1.84; SESN2 -0.52 against -1.42; DDIT3 **+0.18** against -1.50; STC2 not detected. Only CHAC1 (-1.17) and ATF5 (-1.70) hold up. This is consistent with the earlier timing result - those genes were the LATE, secondary part of the response - and it means the "integrated stress response collapse" framing was over-read from one line. **It is not routed through PRC1.1.** In PCGF1-KO and BCOR-KO clones the menin response is essentially preserved (ASNS -1.27 and -1.28, PHGDH -1.36 and -1.79, MEIS1 -3.21 and -2.75), so removing the non-canonical PRC1.1 complex or BCOR does not abolish the serine collapse. **HOXA9 fails again, in a third dataset:** -0.66 here, -0.76 in RS4;11, and HOXA7 flat at 0.00. Three datasets, three assays, and the marker gene this field runs on has never been measurable in any of them.

Facts and figures
scriptanalyses/mv411-third-aff1-line-independent-lab-asns-replication/run.py
resultsanalyses/mv411-third-aff1-line-independent-lab-asns-replication/results.json
testshyp-menin-inhibition-collapses-the-serine-asns-programme-and-should-resensitise-kmt2a-r-all-to-asparaginase
seriesGSE260456, RNA-seq subseries of PMID 41634341 (Neville D, Ferguson DT, Heikamp EB, et al., 'DOT1L provides transcriptional memory through PRC1.1 antagonism', Nature Cell Biology 2026, doi 10.1038/s41556-025-01859-8)
matrixGSE260456_MV411_PRC1.1_KO_counts.tab.gz, 48 samples, 63,157 Ensembl gene rows
why this is a real independent testDifferent cell line, different laboratory, different paper, different scientific question (PRC1.1 / transcriptional memory rather than metabolism), and a panel copied verbatim from the hypothesis entry rather than chosen for this matrix.
armsSafe3 safe-harbour control sgRNA, PCGF1-KO clone CL19, BCOR-KO clones CL18 and CL8, each x DMSO / SGC0946 5 uM / VTP-50469 500 nM, n=3; plus parental and pooled DMSO controls
housekeeping driftmedian of ACTB/GAPDH/TBP/RPL13A in the Safe3 menin arm = -0.33 (ACTB alone -0.73). Reported and subtracted as a conservative second reading because library-size CPM forces the total to a constant and a real global shutdown therefore appears as a uniform downward drift.
positive control workedMEIS1 -3.52, MEF2C -5.64, PBX3 -3.04, HOXA10 -1.53. Without this the run would be uninterpretable.
dot1l inhibitor is weaker hereSGC0946 5 uM: PHGDH -0.88, PSAT1 -1.19, ASNS -0.70, MEIS1 -0.63, MEF2C -1.46. A cleaner menin-versus-DOT1L separation than RS4;11 gave at day 7, where EPZ5676 moved MEIS1 -1.50.
not routed through prc1 1PCGF1-KO menin response: ASNS -1.27, PHGDH -1.36, PSAT1 -1.60, MEIS1 -3.21. BCOR-KO: ASNS -1.28, PHGDH -1.79, PSAT1 -1.38, MEIS1 -2.75. The serine collapse survives loss of either.
hoxa9 fails a third timeHOXA9 -0.66 in MV4;11 and -0.76 in RS4;11; HOXA7 0.00 in both. Three datasets, three assay types (Affymetrix, 3'-end RNA-seq, this count matrix) and HOXA9 has not been measurable in any.
concordance with rs41110 of 12 comparable panel genes agree in sign (83%); 4 agree and both exceed 1 log2 (MEIS1, MEF2C, PBX3, HOXA10).
an identifier collision the readback caughtThe Ensembl REST symbol endpoint's FIRST hit for `CTH` was ENSG00000019102, whose display_name is **VSIG2** - CTH is simultaneously the approved symbol for cystathionine gamma-lyase and a registered alias of VSIG2. Taking the first hit would have substituted an immunoglobulin-domain membrane protein for a transsulfuration enzyme. The script reads every candidate back and compares display_name, so it rejected that id and resolved CTH to ENSG00000116761. DDIT3's first hit, ENSG00000285133, returned no display_name and was also rejected. This is the third identifier trap this project has logged of the form 'a plausible id resolves to a different gene'.
shown for fusiontrue
absence classmeasured-not-analysed
replicated serine one carbon
{
  "PHGDH": "-2.77 (adj -2.44); RS4;11 -3.24",
  "GPT2": "-2.34 (adj -2.01); RS4;11 -1.54",
  "PSAT1": "-2.17 (adj -1.84); RS4;11 -2.40",
  "CBS": "-1.89 (adj -1.56); RS4;11 -4.07",
  "SLC1A5": "-1.61 (adj -1.28)",
  "EIF4EBP1": "-1.37 (adj -1.04)",
  "SLC7A5": "-1.20 (adj -0.87)",
  "XBP1": "-1.20 (adj -0.87)",
  "ASNS": "-1.16 (adj -0.83); RS4;11 -1.72",
  "MTHFD2": "-1.03 (adj -0.70)",
  "ATF4": "-0.99 (adj -0.66)",
  "SHMT2": "-0.94 (adj -0.60)"
}
did not replicate the isr output
{
  "TRIB3": "-0.52 here against -3.10 in RS4;11",
  "ATF3": "+0.22 against -1.84",
  "SESN2": "-0.52 against -1.42",
  "DDIT3": "+0.18 against -1.50",
  "STC2": "0.00 - not detected in this matrix",
  "CTH": "+0.24 against -1.57",
  "what_did_hold": "CHAC1 -1.17 (adj -0.83) and ATF5 -1.70 (adj -1.37)"
}
How this could bite us back
  • This strengthens a hypothesis about a drug combination and must not be read as evidence for giving anyone a menin inhibitor with asparaginase. No asparaginase was given to any cell in any of the three datasets, every number is transcript abundance, and the gene the combination argument turns on - ASNS - is the weakest replicated member at -0.83 after housekeeping adjustment.
  • The replication is across a lineage boundary: MV4;11 is myeloid/biphenotypic and the index disease is infant pro-B ALL. None of the three lines tested is infant-derived.
  • The global housekeeping drift means every repression quoted here, including the strong ones, may be partly arithmetic. Both raw and adjusted values are given precisely so that a reader cannot quote the larger number without seeing the smaller one.
  • The PCGF1/BCOR result is weak-confidence, single-clone for PCGF1, and should not be cited as showing the serine response is PRC1.1-independent as a mechanism.
Still open
  1. Is the global shutdown real? ACTB at -0.73 with library-size normalisation is the single largest interpretive problem across all three of these datasets, and none of them has a spike-in. A drug-treated arm with ERCC spike-ins, or total-RNA-per-cell measurement, would convert every number in this beat from relative to absolute.
  2. Does the serine collapse happen in an INFANT-derived KMT2A::AFF1 line? Three lines tested across three datasets - SEM (5 years), RS4;11 (32 years), MV4;11 (10 years) - and none is infant-derived, while this graph names seven infant-derived AFF1 lines that appear in none of the datasets.
  3. Is ASNS protein reduced? Every number in this beat is transcript, and after housekeeping adjustment ASNS is only -0.83 in this line. The asparaginase hypothesis needs protein and ideally intracellular asparagine, not mRNA.
  4. Why do TRIB3, ATF3 and DDIT3 reverse sign between the two lines? The timing explanation is available and testable: GSE260456's exposure duration is not stated in the matrix, and if it is shorter than seven days these are exactly the genes expected to differ.
  5. The washout question. GSE260456 also contains washout count matrices (230606_F3_Washout, 230904_MOLM13_Washout_K562), and this graph carries exp-dot1l-washout-rechallenge-tests-the-memory-claim asserting that nobody has done a washout-rechallenge experiment. That entry now needs the same check that overturned the H3K27me3 timecourse entry.
  6. Does PCGF1 or BCOR knockout change the response of the CANONICAL targets in a way that matters? MEIS1 goes -3.52 in control, -3.21 in PCGF1-KO and -2.75 in BCOR-KO, which is a modest blunting that this analysis is not powered to call.
Every source behind this entry (3)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    In MV4;11, a KMT2A::AFF1 line, under VTP-50469 at 500 nM against same-genotype DMSO with a safe-harbour control sgRNA and n=3, the serine/one-carbon and amino-acid-transport programme falls: PHGDH -2.77, GPT2 -2.34, PSAT1 -2.17, CBS -1.89, SLC1A5 -1.61, EIF4EBP1 -1.37, SLC7A5 -1.20, XBP1 -1.20, ASNS -1.16, MTHFD2 -1.03, ATF4 -0.99, SHMT2 -0.94. Subtracting the median housekeeping drift of -0.33 gives PHGDH -2.44, GPT2 -2.01, PSAT1 -1.84, CBS -1.56 and ASNS -0.83. The canonical positive control collapses in the same samples (MEIS1 -3.52, MEF2C -5.64, PBX3 -3.04, HOXA10 -1.53), and the DOT1L inhibitor SGC0946 at 5 uM is markedly weaker on the same genes (PHGDH -0.88, ASNS -0.70, MEIS1 -0.63). This reproduces the direction and approximate magnitude measured independently in RS4;11 from GSE127506.

    DOT1L provides transcriptional memory through PRC1.1 antagonism. · Neville D, Ferguson DT, Heikamp EB, et al. · Nature Cell Biology 2026
    PMID 41634341DOI
    What this evidence cannot show

    Housekeeping is NOT flat in this matrix - ACTB is -0.73 against -0.35 in the RS4;11 comparison - so library-size CPM is absorbing what may be a genuine global transcriptional shutdown and inflating every repression. Both the raw and housekeeping-adjusted values are given for that reason, and the adjustment is a blunt instrument rather than spike-in normalisation, which the deposited counts cannot support. The exposure duration is not stated in the matrix filename, so magnitudes are not expected to match RS4;11's day 7. MV4;11 is KMT2A::AFF1 but MYELOID/biphenotypic where RS4;11 and SEM are B-lymphoid, so this replicates the axis across a lineage boundary rather than within the index disease. No p-values. Counts as deposited, no realignment. The panel was copied from the hypothesis entry and not re-declared, so it is confirmatory in the same sense the original was.

    Evidence that argues against this (2)
    • ASNS is the gene the hypothesis turns on and it is the WEAKEST replicated member: -1.16 raw and -0.83 after housekeeping adjustment, against -1.72 in RS4;11. The serine enzymes upstream of it replicate far more strongly. So what reproduces robustly is the serine/one-carbon pathway, and the asparaginase-relevant step reproduces only modestly. Any citation of this entry as support for the asparaginase hypothesis should quote the adjusted ASNS number, not PHGDH's.
    • A global downward drift of the housekeeping set is itself evidence for the arrest explanation this whole line of work has been unable to exclude. If VTP-50469 at 500 nM is shutting the cell down by the harvest point, then a coordinated fall in biosynthetic transcripts is expected whatever the mechanism, and the fact that the positive control falls harder than the metabolic panel is consistent with either reading.
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    The integrated-stress-response OUTPUT genes do not replicate, so that part of the original interpretation is not supported. Against their RS4;11 values: TRIB3 -0.52 (was -3.10), ATF3 +0.22 (was -1.84), SESN2 -0.52 (was -1.42), DDIT3 +0.18 (was -1.50), CTH +0.24 (was -1.57), and STC2 is undetected in this matrix. Only CHAC1 (-1.17) and ATF5 (-1.70) hold. Across the 12 comparable panel genes, 10 agree in sign with RS4;11 (83%) but only 4 agree while both exceed 1 log2, and all four are canonical menin targets (MEIS1, MEF2C, PBX3, HOXA10) rather than stress-response genes.

    What this evidence cannot show

    A non-replication across two platforms, two laboratories, two cell lines and an unknown difference in exposure duration is weak evidence of absence: TRIB3 and ATF3 are fast, transient stress genes, so a timing difference alone could explain their behaviour. The earlier day-2 versus day-7 analysis in RS4;11 found exactly these genes to be the LATE part of that response (TRIB3 carrying 13% of its day-7 effect at day 2, ATF3 79%), so the two results are compatible with a timing explanation rather than a contradiction. What is not defensible any longer is calling the whole thing an 'ISR collapse' on the strength of one line.

    Evidence that argues against this (1)
    • This partially undercuts the entry that named the ATF4 signature in the first place. That entry's identification rested on CHAC1, TRIB3, ATF3, SESN2 and STC2 together; here two of those five reverse sign and one is undetected. CHAC1 survives, and CHAC1 plus ATF5 plus the serine enzymes is a narrower and less tidy story than 'the integrated stress response'. The honest label for what replicates is the serine/one-carbon/amino-acid-transport programme, not the ISR.
  3. E3weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.database

    The serine collapse does not require non-canonical PRC1.1 or BCOR. In PCGF1-knockout clone CL19 the menin-inhibitor response is ASNS -1.27, PHGDH -1.36, PSAT1 -1.60, CBS -1.77, GPT2 -1.92, MEIS1 -3.21; in BCOR-knockout clones CL18 and CL8 pooled it is ASNS -1.28, PHGDH -1.79, PSAT1 -1.38, CBS -1.30, GPT2 -1.89, MEIS1 -2.75. Each is computed against its own genotype's DMSO arm.

    What this evidence cannot show

    The PCGF1 arm is a SINGLE clone (CL19), so a clone-specific effect cannot be separated from a genotype effect; the BCOR arm has two clones, which is the only internal check available and they agree. Each knockout's response is measured against its own DMSO, so a knockout that already shifted the baseline will show a smaller apparent drug effect for arithmetic rather than biological reasons - and PRC1.1 loss is expected to shift baselines at exactly the loci this drug acts on. Knockout efficiency was not verified from the data here beyond noting that PCGF1's own transcript moves little (-0.14 in the Safe3 menin arm), which is not a validation of the knockout. This is a weak-confidence observation included because a negative on the obvious mediator is worth recording, not because the design supports a strong claim.

    Evidence that argues against this (1)
    • The natural objection is that this is the wrong test of PRC1.1 involvement: the depositors' own claim is that PRC1.1 mediates the SILENCING that follows menin or DOT1L inhibition, which concerns the repression of canonical targets rather than the metabolic programme, and a knockout that blunts silencing at HOXA loci need not affect serine-pathway transcripts at all. So 'not routed through PRC1.1' should be read narrowly as 'the serine response is not abolished by PCGF1 or BCOR loss in single clones', which is much less than a mechanism.
Connected to