SUPERSEDED IN PART, CYCLE 56 - the MI-2-2 response is ~9x fusion-selective (that stands), but 'does not touch HOXA or MEIS1' was a fact about a weak compound: VTP-50469 collapses MEIS1 -5.25 log2 in the same line

The SEM arm is n=2 MI-2-2 versus n=2 DMSO because three of the submitter's four QC exclusions are SEM arrays (SEM-DMSO-1, SEM-MI22-2, SEM-COMBO-2); RS4;11 lost DMSO-1. No p-values are computed and none should be inferred - a single outlier array can move any SEM number here, and the RS4;11 arm exists in this analysis precisely because the SEM arm cannot carry the claim alone. A probe-count threshold of 1 log2 is a convention, not a test. SEM, RS4;11 and REH differ in far more than KMT2A status - different fusions, different culture media (SEM in IMDM, the others in RPMI), different growth rates and baseline transcriptomes - so a fusion-line-selective response is CONSISTENT with menin dependence and does not establish it; there is no isogenic arm anywhere in this series.

• The selectivity ratio is computed on counts of probes crossing a threshold, which is sensitive to each line's noise floor and dynamic range. REH's DMSO arm has n=3 and the fusion lines n=2, and a smaller replicate count usually INFLATES apparent fold changes by leaving noise unaveraged - which biases this comparison in the direction of the reported result. The EPZ and Combo ratios are included as internal controls for exactly this reason, and the fact that the EPZ ratio is much smaller in the same arrays argues the bias does not generate the effect, but it does not eliminate it.
• MI-2-2 is a first-generation tool compound used here at 10 uM for six days. At that exposure off-target transcriptional effects are expected, and nothing in this dataset measures menin target engagement. A response that is larger in fusion-positive lines is consistent with on-target action and equally consistent with fusion-positive lines being more dependent on something else the compound also hits at 10 uM.

This is a six-day endpoint on a microarray. It cannot exclude an early collapse of HOXA/MEIS1 that had recovered by day 6, nor selection of a drug-tolerant subpopulation during those six days - although note that selection of HOXA-low survivors would predict a DECREASE, and the measured values are flat or slightly positive. Affymetrix HOXA-cluster probes can cross-hybridise within a tandem gene family, and HOXA9 specifically has no unambiguous probe, so the single gene the hypothesis is most about is measured only indirectly. Transcript abundance is not protein, and neither is chromatin occupancy; a fusion could be evicted from the HOXA locus with mRNA unchanged over this window. Probe saturation is not the explanation: MEIS1 at 11.6 and HOXA10 at 11.2 sit well below ACTB at 14.0 on the same arrays.

• The strongest alternative reading is that MI-2-2 simply did not inhibit menin enough to collapse the programme, and that the 769 selective probe changes come from something else. Potency argues for this: MI-2-2 is far weaker than VTP-50469 or revumenib, and published menin-inhibitor HOXA9/MEIS1 collapse is generally reported with the later compounds. Against it: the response is 9x selective for the fusion lines and reproduced across two of them, which is a strange signature for a purely off-target effect. This analysis cannot decide between the two, and a revumenib or VTP-50469 arm in the same lines would.
• A published result in this graph points the other way and should be weighed against this one: menin-inhibitor treatment is reported to silence fusion target loci with PRC1.1 deposition rather than mere eviction (see mechanism-menin-inhibitor-silencing-is-prc1-1-deposition-not-only-eviction), which is a mechanism for target-gene suppression. That work uses different compounds, lines and timepoints. The disagreement is real and unresolved here, and this entry should not be cited as though it settled it.
• MEIS1 does fall in REH (-0.71) while staying flat in the fusion lines, which inverts the expected pattern. Its REH baseline is 5.10 - essentially off - so this is almost certainly noise at the low end of the array's range, but it is a reminder that the same threshold applied to a near-absent transcript produces a number that looks like a finding.

The top-30 lists are the extremes of an n=2-versus-n=2 contrast in SEM, which is exactly where noise concentrates; that is why every row is reported with its RS4;11 and REH value and why the replication figure is given as 34 of 60 rather than as a claim that the whole list reproduces. Gene identity comes from GPL570's 'Gene Symbol' field, so ambiguous probes appear in results.json as '(ambiguous or unmapped)' and are not credited to a gene. The differentiation reading is an interpretation of a marker set, not a measured phenotype: this is an expression array, and no immunophenotyping, morphology or viability data are analysed here. 'mir-223' is a probe annotation on an expression array and is not a proper miRNA measurement. ATF5's repression is a robust and unexpected observation, and this project has no entry for ATF5 and no mechanistic account of why it would be menin-inhibitor-sensitive.

• A differentiation programme rising while HOXA/MEIS1 stay flat can be read against this analysis's own framing: if the cells were genuinely differentiating, the fusion-maintained stemness programme would normally be expected to fall, and it does not. Either the marker rise is not differentiation, or differentiation in these lines does not require the HOXA programme to shut down, or the two populations are different cells - a mixed response in which a subpopulation differentiates strongly while the HOXA-high bulk persists would produce exactly this bulk-array signature. Nothing here distinguishes those, and a single-cell or sorted readout would.
• The source paper reports that the combination markedly enhances DIFFERENTIATION and cell killing, so the rise of differentiation markers is the authors' own finding and not a new one. What is new here is that it is measurable from the single menin-inhibitor arm, that it reproduces in a second fusion line, that it is near-absent in the wild-type line, and that it happens without HOXA/MEIS1 suppression.

Abstract-level for the publication: PMID, title, journal, year, DOI and first author verified live by efetch on 2026-09-22 and the abstract text read, but the full text, figures and supplementary tables were NOT retrieved. The paper may well report HOXA/MEIS1 behaviour in a figure, in which case this analysis reproduces something already published rather than finding it, and the 'nine entries cite it' point would be the only novel part. The phrase 'target genes of MLL-fusion proteins' is not defined in the abstract and the authors may mean a signature set that does not centre on HOXA/MEIS1 - in which case there is no disagreement between their claim and these numbers at all.

• This is the THIRD time in one cycle that this project has found itself citing a paper repeatedly while never reading its data: PMID 26766589 was cited ten times with its fusion-side interaction rows unread, PMID 24735925 once with its PAX5 rows unread, and PMID 27840424 nine times with GSE63664 unopened. The pattern suggests the graph's citation practice systematically privileges abstracts and conclusions over deposited data, which is a methodological weakness of this project and not a finding about menin inhibitors.

Cellosaurus's AG field records age at SAMPLING, not age at diagnosis, so a line taken at relapse could come from a patient diagnosed considerably earlier - line-sem-kmt2a-aff1 already carries this exact caveat and notes it is the single fact that would most change how the line should be read. So '5 years' bounds the sampling age and does not by itself prove the donor's disease was not infant-onset. What it does establish is that 'infant-derived' is not a statement the record supports, and that 'adult-derived', which six files in this graph assert of SEM, is wrong.

• This does not weaken the analysis above, and it is important not to let it: the fusion-selectivity and HOXA/MEIS1 results are statements about KMT2A::AFF1 lines and do not depend on donor age. What donor age changes is the reach of any inference to INFANT disease - and since infant and childhood MLL-AF4 ALL differ in whether the fetal programme is retained, a flat HOXA response in two non-infant lines is weaker evidence about infant biology than the same result in an infant-derived line would have been. No infant-derived KMT2A::AFF1 line appears anywhere in this series.

The MI-2-2 measurements in this entry are unchanged and were re-read from stored results rather than retyped; what is corrected is their interpretation. The two datasets differ in compound, platform, duration and normalisation, so only the cell line and the gene panel are held constant - but potency is the explanation this entry itself nominated in advance, and a MEIS1 swing from +0.21 to -5.25 is far outside what a platform change produces. See ana-the-potent-menin-inhibitor-collapses-what-the-weak-one-left-untouched-and-hoxa9-was-never-measured-by-either for the full comparison.

• The correction removes a result that read as a challenge to the menin-interface hypothesis, and that should be said plainly rather than buried: with a potent compound the target programme behaves as the field says it does. This graph carried the opposite reading for one cycle.
• The correction does NOT rescue the mechanism story wholesale, because the DOT1L inhibitor EPZ5676 moves the same genes in the same direction in the same experiment (MEIS1 -1.50, MEF2C -1.71 at day 7). MEIS1, MEF2C and PBX3 therefore report on the elongation complex as a whole and are not a specific readout of menin-interface engagement.