Oncofusion

start / graph / ana-the-washout-arm-cannot-answer-the-durability-question-and-meis1-does-not-recover

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

The washout data cannot say whether the serine/ASNS collapse is durable - the washout control never recovers, ASNS is undetectable in the line, and the housekeeping drifts in opposite directions in the two matrices

also called washout durability · GSE260456 washout · MEIS1 does not recover · serine ASNS persistence

We asked whether the drug's effect on the cell's amino-acid machinery lasts after the drug is removed, using public data that removes it. The data cannot answer: the genes that should bounce back as a check never bounce back, and the one gene the question turns on is not even switched on in that cell line.

The same thing, for a scientist

This analysis was set up to decide something the cycle-56 ASNS hypothesis leaves open: whether the serine/one-carbon collapse under menin inhibition is **durable reprogramming** or **reversible pharmacology**. The two answers imply different combination schedules. GSE260456 contains the washout arms — found while correcting a different entry's novelty claim — and the answer is that **they cannot decide it**, for three independent reasons. The refusal is the result. **1. The washout control never recovers.** MOLM13 NT, VTP-50469 then 48 h and 96 h washout, n=3 per arm. Recovery fraction = 1 − (WO − DMSO)/(VTP − DMSO), so 0 is fully suppressed and 1 is back to baseline. Median recovery of the canonical targets (MEIS1, MEF2C, PBX3, HOXA9, HOXA10, JMJD1C) at 96 h: **0.208**. MEIS1 goes **further down** (−3.69 at drug → −4.30 at 96 h washout); MEF2C −2.58 → −2.94. If the genes that must recover do not, a serine panel that also does not recover says nothing about durability — only that the cells are still drug-affected. Median serine/ASNS recovery was 0.303, and that number is *not reportable* as a finding. **2. The gene the hypothesis turns on is not expressed in the line.** ASNS DMSO baseline in MOLM13 is **log2CPM 0.10**. CBS is 0.00 and SESN2 0.00. So even with a working washout, this matrix could not test ASNS. **3. The mouse arm fails differently, which is worse.** The F3 (FLAG-MLLAF9) matrix has the denser timecourse — V8, V8+24W, V48, V48+24W, +48W, +72W — and there the serine programme is **barely engaged by the drug at all**: Asns **−0.45**, Phgdh **−0.18**, Atf4 −0.05 at 48 h. There is no suppression to recover from. Meis1 meanwhile shows **zero recovery over 72 h** (−4.69 → −4.66). **And the normalisation is unsafe in both, in opposite directions.** Median housekeeping drift in MOLM13 NT: −0.258 (drug), −0.375 (48 h), **−0.601 (96 h)**. In mouse F3, Actb drifts the other way: **+0.38 → +1.38** across the washout. Library-size CPM is absorbing large global shifts whose sign is not even consistent between the two matrices. **ONE OBSERVATION SURVIVES, and it survives precisely because of the drift.** MEIS1/Meis1 does not recover in either line — and in the mouse arm Actb drifts **upward** by 1.38 log2 while Meis1 sits at −4.66, so *relative to housekeeping* Meis1 is more suppressed at 72 h of washout than at drug. A global artefact cannot produce that, because it points the wrong way. So: **the canonical menin-inhibitor target does not rebound within 72–96 h of drug withdrawal in either an MLL-AF9 mouse line or MOLM13** — which is congruent with the source paper's own transcriptional-memory framing and directly relevant to `exp-dot1l-washout-rechallenge-tests-the-memory-claim`.

Facts and figures
scriptanalyses/does-the-serine-asns-programme-recover-after-menin-inhibitor-washout/run.py
resultsanalyses/does-the-serine-asns-programme-recover-after-menin-inhibitor-washout/results.json
testshyp-menin-inhibition-collapses-the-serine-asns-programme-and-should-resensitise-kmt2a-r-all-to-asparaginase
seriesGSE260456, RNA-seq subseries of GSE260742 / PMID 41634341
matricesGSE260456_230904_MOLM13_Washout_K562_counts.tab.gz (human) and GSE260456_230606_F3_Washout_counts.tab.gz (mouse FLAG-MLLAF9)
human designMOLM13 non-targeting sgRNA and PCGF1/BCOR double knockout, each DMSO / VTP-50469 / VTP then 48 h washout / VTP then 96 h washout, n=3 per arm. K562 DMSO and SGC0946 columns excluded by construction.
mouse designF3 FLAG-MLLAF9, NT and a PCGF1-KO clone, DMSO / V8 / V8+24W / V48 / V48+24W / V48+48W / V48+72W, n=3.
the reported quantityrecovery fraction = 1 - (WO - DMSO)/(VTP - DMSO), computed only where the drug moved the gene by at least 0.5 log2, with raw arm means printed beside it so the ratio cannot be read without its numerator.
the one surviving observationMEIS1/Meis1 does not recover within 72-96 h of menin-inhibitor withdrawal in either line, and in the mouse arm this is drift-robust because Actb drifts UP 1.38 log2 while Meis1 stays at -4.66 - so relative to housekeeping it is more suppressed at washout than on drug. A uniform global artefact cannot produce that.
serine panel numbers not to be quoted as a findingMedian serine/ASNS recovery 0.109 at 48 h and 0.303 at 96 h; ISR 0.036 and 0.261. Recorded so nobody re-derives them and reports them, because the washout control forbids their interpretation.
the instrument failure this shipped with firstBoth matrices are MALFORMED: the MOLM13 file has 34 header fields with a trailing empty, and 35 data fields with empties at irregular positions 32 and 34 - a spurious empty column mid-row that shifts every later value. A positional slice parts[1:1+len(names)] therefore hits an empty string, float() raises, and every row is discarded: the loader returned 0 of 44,430 gene rows while parsing all 32 sample columns correctly. The fixed-in-advance panel is what made it loud - all 32 genes reporting 'not in matrix' at once is absurd in a way a wrong number is not. Fix: drop empty fields from BOTH header and data and align the cleaned lists, with a guard that refuses to analyse fewer than 20,000 rows.
shown for fusiontrue
absence classmeasured-not-analysed
why the answer is a refusal
{
  "1_washout_control_fails": "Median recovery of MEIS1/MEF2C/PBX3/HOXA9/HOXA10/JMJD1C at 96 h = 0.208. MEIS1 -3.69 -> -4.30 and MEF2C -2.58 -> -2.94, i.e. further suppressed after withdrawal.",
  "2_asns_undetectable": "ASNS DMSO log2CPM 0.10 in MOLM13; CBS 0.00; SESN2 0.00.",
  "3_mouse_arm_not_engaged": "Asns -0.45, Phgdh -0.18, Atf4 -0.05 at V48 in F3 - no suppression to recover from - while Meis1 shows zero recovery across 72 h.",
  "4_normalisation_unsafe_in_both": "Housekeeping median drift MOLM13 NT: -0.258 / -0.375 / -0.601. Mouse F3 Actb: +0.38 -> +1.38. Opposite signs between matrices."
}
How this could bite us back
  • Nothing here supports or refutes a dosing schedule. The analysis exists to say that the public washout data cannot decide whether the menin-inhibitor effect on amino-acid metabolism is durable, and a scheduling argument built on either answer would be built on nothing.
  • The surviving observation - the canonical target not rebounding within 96 h - must not be read as 'the drug keeps working after it is stopped'. Residual intracellular compound explains it equally well, nothing here measures drug clearance, and this is transcript in cell lines.
  • Do not quote the serine-panel recovery figures (0.109 at 48 h, 0.303 at 96 h). They are recorded in the entry only so that nobody re-derives and reports them: the washout control forbids their interpretation.
  • Neither line carries the index fusion, and ASNS is not even expressed in the human one.
Still open
  1. Is there any menin-inhibitor washout dataset with spike-in normalisation? Every number in this beat is hostage to library-size CPM absorbing a global shift whose sign differs between the two matrices, and no washout claim in either direction is safe without one.
  2. Was drug actually cleared? A free-compound or target-engagement measurement at the washout timepoints would separate durability from residual occupancy, and its absence is the single largest weakness of the surviving observation.
  3. Why does Hoxa9 recover while Meis1 does not, in the same samples? A split within a co-regulated programme is either informative about the mechanism or evidence that one of the two measurements is unreliable.
  4. Does an AFF1 line behave the same way? Neither matrix contains a KMT2A::AFF1 model, and this is now the fourth beat in two cycles to hit that wall - the seven infant-derived AFF1 lines this graph names appear in none of these datasets.
  5. Is ASNS expressed in any KMT2A-rearranged line with a washout arm? It is undetectable in MOLM13, which means the asparaginase hypothesis cannot be tested for persistence in the one human washout dataset available.
Every source behind this entry (2)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    The washout arms of GSE260456 cannot decide whether the serine/one-carbon collapse persists. In MOLM13 non-targeting cells the median recovery fraction of the canonical menin-inhibitor targets at 96 h of washout is 0.208, with MEIS1 moving from -3.69 on drug to -4.30 after 96 h without it and MEF2C from -2.58 to -2.94. ASNS is not expressed in this line (DMSO log2CPM 0.10), as are CBS (0.00) and SESN2 (0.00). In the mouse F3 FLAG-MLLAF9 matrix the serine programme is barely engaged at 48 h of drug (Asns -0.45, Phgdh -0.18, Atf4 -0.05), so there is no suppression to recover from, while Meis1 shows no recovery across 72 h (-4.69 to -4.66). Median housekeeping drift in MOLM13 NT is -0.258, -0.375 and -0.601 at drug, 48 h and 96 h; in mouse F3, Actb drifts from +0.38 to +1.38 across the same washout.

    PMID 41634341
    What this evidence cannot show

    This is a refusal, and a refusal inherits the weaknesses of what it refuses on. Neither line is KMT2A::AFF1 - MOLM13 is KMT2A::MLLT3 and F3 is murine MLL-AF9 - so even a clean result would have transferred only as far as shared elongation biology does, and the hypothesis is about AFF1 disease. Exposure concentrations and the precise meaning of the '+Nw' labels are the submitter's; 'washout' means drug removed from medium and does not establish that intracellular compound is gone, so a non-recovering control is as consistent with residual drug as with durability. No p-values; the recovery fraction is a ratio of two differences of means and is withheld below a 0.5 log2 drug effect for that reason. 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)
    • A sceptic could say the washout control did not 'fail' at all - that MEIS1 not recovering IS the finding, and the analysis is refusing to read a real result as one. That reading is available and is why the surviving observation is reported separately in E2. What the control genuinely forbids is the specific comparison this analysis was built for: you cannot say the serine arm is durable BECAUSE it did not recover, when nothing recovered.
    • The drift undermines the refusal as well as the result. If library-size normalisation is absorbing a global shift of 0.6 log2 in one matrix and 1.4 in the other, then the recovery fractions for every gene including the controls are computed on a moving baseline, and the 0.208 figure that triggers the refusal is itself unreliable. The honest position is that this series needs spike-in normalisation for any washout claim, in either direction.
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    One observation survives the normalisation problem because of its direction. MEIS1/Meis1 does not rebound within 72-96 h of menin-inhibitor withdrawal in either system. In the mouse F3 arm this is drift-robust: Actb rises from +0.38 to +1.38 log2 across the washout series while Meis1 remains at -4.66, so relative to the housekeeping trend Meis1 is MORE suppressed after 72 h without drug than it was on drug. A uniform global artefact would have to move Meis1 upward with Actb, and it does not.

    What this evidence cannot show

    A single gene in two lines, one of them murine, at transcript level, with no protein or occupancy measurement, and with 'washout' meaning medium change rather than verified drug clearance. It is congruent with the source paper's transcriptional-memory framing but does not test it: memory in the inducible sense requires a SECOND exposure and the rate of re-silencing, which this series does not contain - a point established separately in exp-dot1l-washout-rechallenge-tests-the-memory-claim. Hoxa9 by contrast recovers substantially in the mouse arm (-1.64 to -0.65), so non-recovery is not a property of the whole programme and picking MEIS1 out of the panel is a choice the panel did not pre-specify.

    Evidence that argues against this (2)
    • The simplest explanation is not durability but residual drug. VTP-50469 binds menin with long residence and nothing here measures free intracellular compound, so a target that stays suppressed for 96 h after a medium change may simply still be engaged. That explanation also accounts for the canonical targets as a group and should be preferred until a compound-clearance measurement exists.
    • Hoxa9 recovering while Meis1 does not is awkward for any single story. If the effect were durable chromatin-level reprogramming one would expect the co-regulated posterior-HOXA programme to behave coherently; if it were residual drug, likewise. A split within the programme points at something the design cannot resolve.
Connected to