Oncofusion

start / graph / ana-depmap-men1-dependency-calm-af10-u937

Analysisweak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.Q55-calm-af10

MEN1, DOT1L and LEDGF dependency in the only CALM-AF10 cell line DepMap has screened

also called U-937 MEN1 dependency · CALM-AF10 DepMap

We asked a large public screen whether the leukaemia cells carrying the CALM-AF10 fusion need the menin protein to survive. Only one such cell line has been screened, and in it menin barely matters while the DOT1L enzyme clearly does - but one cell line is one cell line.

The same thing, for a scientist

Cellosaurus holds **two** DepMap-linked PICALM::MLLT10 lines - **U-937 (ACH-000406)** and **TUR (ACH-002314)** - and only U-937 carries CRISPR gene-effect data. Against 42 non-KMT2A-rearranged leukaemia screens and 11 KMT2A-rearranged ones, U-937 shows the pattern *DOT1L-dependent, menin- and LEDGF-independent* [E1][E2]. **This is n=1 and no selectivity statistic is computed**, because a difference of means against a group of one is not a measurement. It is reported because it is the only quantitative evidence bearing on the headline question of this beat, and because its direction is checkable by anyone in an afternoon.

Go deeper
Result

DepMap CRISPR (Chronos) gene effect, leukaemia screens only, fetched 2026-09-20:

geneU-937 (CALM-AF10)KMT2A-r mean (n=11)other leukaemia mean (n=42)other leukaemia SD
MEN1-0.229-1.263-0.4010.406
DOT1L-0.542-0.604-0.3690.405
PSIP1 (LEDGF)+0.135-0.553+0.0090.240
KMT2A-0.375-0.811-0.2680.355
MEIS1-0.168-0.429-0.1220.257
HOXA9-0.280-0.271-0.2210.183
PICALM-0.345-0.381-0.2610.324
MLLT10+0.081-0.096+0.0520.172
ACTB (control)-0.874-0.949-0.7650.380

Reading: U-937's MEN1 effect is weaker than the average non-KMT2A leukaemia line, and more than a full unit weaker than the KMT2A-rearranged mean. Its DOT1L effect sits essentially at the KMT2A-rearranged level. ACTB confirms the screen resolves a real dependency in this line.

Four things this cannot show, stated before anyone quotes the table
  1. n=1. No p-value, no effect size, no selectivity claim. Per the brief's DepMap rules, a z is not an effect size and a group of one has no spread.
  2. U-937 is not a clean instrument. It is a long-passaged histiocytic-lymphoma-derived line with many co-occurring lesions, and it is myeloid, whereas 65% of PICALM::MLLT10 patients have T-ALL. The lineage of the only available line is the minority lineage of the disease.
  3. MLLT10 at +0.081 does not mean AF10 is dispensable for the fusion. A gene-level CRISPR score sums guides across the locus; guides in 5' exons cut a region the fusion does not contain, and cutting the endogenous allele does not remove the chimeric transcript. The same applies to PICALM at -0.345.
  4. Floor and ceiling. HOXA9 moves almost nothing in any arm, including the KMT2A-rearranged one (-0.271 versus -0.221), so this screen has essentially no power to detect HOXA-programme dependency in any line - which is itself worth knowing before anyone uses DepMap to argue about HOXA.
What would upgrade it

Two more screened CALM-AF10 lines would make a group. The T-ALL side is the priority, since the disease is T-lineage-skewed and DepMap's only instance is myeloid. Failing that, the direct experiment in exp-menin-inhibitor-dose-response-calm-af10-panel is cheaper than waiting for DepMap.

Bound by a prior analysis in this graph that already contained this line

This is not the first time U-937's MEN1 dependency has been computed here. ana-menin-programme-vs-rearrangement (2026-09-19) built a three-arm split whose PROGRAMME arm is exactly {OCI-AML3, U-937} - the entire supply of non-KMT2A HOXA/MEIS1-driven lines in DepMap's leukaemia panel.

The two agree numerically. That analysis reports mean_index = -1.2629 for the KMT2A-rearranged arm, which is the figure quoted here, and mean_negative = -0.3528 (sd 0.3763, n=44) against the -0.401 (sd 0.406, n=42) quoted here - the small difference is line filtering, not disagreement.

And it supplies the bound this entry lacked. For MEN1 the programme arm reproduces 0.548 of the index effect, but the minimum difference detectable at n=2 is 0.762 against an observed programme-versus-negative difference of 0.498 — so programme_diff_is_detectable is false. The arm is not powered even for the effect it appears to show, and n=2 is the whole supply rather than a sampling choice.

One thing in that analysis strengthens this entry from an independent direction. For DOT1L, the programme arm's programme_recapitulation is 1.661 - the KMT2A-free arm is more DOT1L-dependent than the index arm (mean -0.7906 against -0.6042) - while MEN1 recapitulates only 0.548. That is the same DOT1L-dependent, menin-independent separation this entry reports, arrived at in aggregate rather than from U-937 alone. It is underpowered too (min_detectable_diff_at_n2 0.8205), so it corroborates the direction and settles nothing.

Facts and figures
scriptrun ad hoc from the beat scratchpad on 2026-09-20; not committed to analyses/ because this beat's write permissions are limited to its node folder, edges file and memo. The query is fully specified below so it can be reproduced or promoted to analyses/ by the coordinator.
statusrun
result summaryU-937, the only screened CALM-AF10 line, is DOT1L-dependent at the KMT2A-rearranged level (-0.542 vs -0.604) and menin-independent below the non-KMT2A leukaemia average (-0.229 vs -0.401, SD 0.406), with LEDGF positive (+0.135 vs KMT2A-r -0.553). n=1; descriptive only.
n calm af10 lines in depmap with data1
n calm af10 lines annotated2
power noteWith the KMT2A-r versus other-leukaemia MEN1 difference of -0.862 and a pooled SD near 0.41, the effect is very large (Cohen's d above 2). Even so, a single line cannot be tested; two to three additional screened CALM-AF10 lines would give a testable group for an effect of that size.
lineage confoundThe only screened line is myeloid; the disease is 65% T-ALL. Any conclusion drawn here is drawn in the minority lineage.
inputs
[
  "Cellosaurus search var:PICALM AND dr:DepMap and var:MLLT10 AND dr:DepMap",
  "Open Targets GraphQL: query($ids:[String!]!){targets(ensemblIds:$ids){approvedSymbol depMapEssentiality{tissueName screens{depmapId diseaseFromSource geneEffect}}}}",
  "Ensembl ids: MEN1 ENSG00000133895, DOT1L ENSG00000104885, KMT2A ENSG00000118058, HOXA9 ENSG00000078399, MEIS1 ENSG00000143995, PICALM ENSG00000073921, MLLT10 ENSG00000078403, PSIP1 ENSG00000164985, ACTB ENSG00000075624"
]
How this could bite us back
  • A near-zero MEN1 score in one line is exactly the shape of result that gets quoted without its n. It is reported here with n=1 stated in the claim, the caveat and the data block precisely so that cannot happen quietly.
Still open
  1. Are there CALM-AF10-positive lines in DepMap that Cellosaurus has not annotated as such? A fusion-call-based rather than annotation-based search would answer it.
  2. Why does HOXA9 barely score in ANY leukaemia arm including KMT2A-rearranged? If DepMap cannot detect HOXA9 dependency where it is best established, HOXA-programme conclusions drawn from DepMap anywhere in this graph are suspect.
Every source behind this entry (2)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    Cellosaurus records exactly two DepMap-linked cell lines with a PICALM-MLLT10 / CALM-AF10 gene fusion: U-937 (ACH-000406) and TUR (ACH-002314).

    What this evidence cannot show

    Queried 2026-09-20 in both directions (var:PICALM and var:MLLT10, each filtered on the partner appearing in the fusion description), per the resolution discipline used by this project's other DepMap analyses. Resolution is by DepMap accession, never by cell-line name. Cellosaurus annotation is not exhaustive: a line whose fusion has not been curated will be missed.

    Evidence that argues against this (1)
    • THE POWER BOUND, from this graph's own prior analysis of the same line. ana-menin-programme-vs-rearrangement computed that for MEN1 the minimum difference detectable with the n=2 programme arm is 0.762, against an observed programme-versus-negative difference of 0.498 - so the arm is NOT powered even for the effect it appears to show, and n=2 is the entire supply of such lines in DepMap rather than a sampling choice. Any single-line figure inherits that bound. The same analysis notes U-937 additionally carries a CDKN2A deletion and TP53 alteration, so line-specific confounds are present on top of the power problem. Oncofusion knowledge graph (internal analysis) 2026
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    In DepMap CRISPR (Chronos) leukaemia screens, U-937 shows MEN1 -0.229, DOT1L -0.542, PSIP1 +0.135, KMT2A -0.375, MEIS1 -0.168, HOXA9 -0.280, PICALM -0.345, MLLT10 +0.081 and ACTB -0.874, against KMT2A-rearranged means (n=11) of -1.263, -0.604, -0.553, -0.811, -0.429, -0.271, -0.381, -0.096 and -0.949.

    What this evidence cannot show

    One CALM-AF10 line with data. Leukaemia restriction is by disease/tissue string match, as in this project's other DepMap analyses. The DepMap release version is not stated by the Open Targets API, so the numbers are release-dependent and should be re-pulled before publication. TUR has no screen data at all.

    Evidence that argues against this (1)
    • The same table shows HOXA9 barely scoring in ANY arm, including the KMT2A-rearranged one (-0.271 against an other-leukaemia -0.221). A screen that cannot detect HOXA9 dependency where it is best established is a weak instrument for any HOXA-programme claim, including this one. DepMap / Open Targets Platform 2026
Connected to