Subtraction is a partner class, not a curiosity — and the class is defined by compartment, not by sequence
also called subtraction partner class · loss-driven KMT2A fusions
We think losing something matters in several KMT2A leukemias, not just one. But the losses do not all look alike: sometimes a piece of the protein is missing, sometimes a normal protein is dragged to the wrong place, and sometimes a healthy gene copy is switched off. Only the first kind shows up if you look at the sequence.
The same thing, for a scientist
Proposes that loss-of-function contributions are a recurrent feature of KMT2A-rearranged leukemia across partners, and that they fall into three mechanistically distinct sub-types: discarded-module (MLLT10 PZP; candidate MLLT3 YEATS), sequestration (AFDN pulling endogenous AF6 out of the cytoplasm; ELL delocalising endogenous EAF1 from speckles), and allelic (KMT2A-PTD silencing the wild-type allele). The corollary that matters for this project is negative: only the first sub-type is detectable by domain-fate analysis, so a retained-module scan is structurally incapable of finding the other two, and both of those have at least as much evidence behind them as the domain-level cases.
Statement and current standing
Across the six partners audited in this beat, loss contributes in at least three and probably four, but by three different physical mechanisms.
| Partner | Sub-type | Loss shown causal? |
|---|---|---|
| MLLT10 | discarded module (PZP) | yes, in CALM-AF10 |
| MLLT3 | discarded module (YEATS) | not tested |
| AFDN | sequestration (cytoplasmic AF6) | loss-of-function shown; requirement for transformation not shown |
| ELL | sequestration (EAF1 speckles) + discarded N-terminal EAF site | no |
| KMT2A-PTD | allelic silencing | yes, in patient blasts |
| MLLT1 | none — reader retained and required | n/a, refuted |
| AFF1 | none — degron retained, no functional loss | n/a, refuted |
The hypothesis is not that most KMT2A fusions are loss-driven. It is that loss is a recurrent co-mechanism, it is heterogeneous in kind, and the project's search method can see only one of its three forms.
What would make this false
The hypothesis is falsifiable in a strong sense, because two of its three sub-types rest on a single result each.
- If in-cis PZP restoration fails to suppress KMT2A::MLLT10 (
exp-pzp-cis-resupply-kmt2a-mllt10) and YEATS restoration fails to suppress MLL-AF9 (exp-yeats-cis-resupply-kmt2a-mllt3), the discarded-module sub-type shrinks to a single non-KMT2A fusion and should not be generalised. - If a self-association-dead, nuclear-localised MLL-AF6 still sequesters endogenous AF6, then sequestration is downstream of the transforming event rather than a parallel mechanism, and it is a consequence rather than a cause.
- If allele-specific re-expression of wild-type KMT2A in PTD blasts does not induce death, the strongest patient-level subtraction result dissolves into hypomethylating-agent pleiotropy.
It also makes a prediction that could be checked without any experiment: wild-type allele silencing should not be confined to PTD. If the untranslocated KMT2A allele is intact and expressed in translocation-driven KMT2A-r blasts, the allelic sub-type is PTD-specific and the class is smaller than claimed.
What follows if it is true
Target selection in this project has been a search over retained protein modules and interfaces on them. If the hypothesis holds, that search has a systematic blind spot with a definite shape: it cannot see mechanisms whose physical substrate is a location (sequestration) or a transcriptional state (allelic silencing) rather than a sequence.
The practical fix is cheap and does not require new biology: extend the per-fusion dossier template with three questions that are currently never asked.
- What does the fusion protein discard, and is the discarded region functional in the intact protein?
- Does the fusion relocate any endogenous protein — most obviously its own wild-type counterpart — and does that relocation remove a function?
- Is the untranslocated allele of either fusion partner silenced?
Question 2 would have caught AFDN and ELL, and both were invisible to the current template.
Challenges already on record
Each of the three sub-types has a published challenge, and two of the three challenges came out before the results this hypothesis is built on were interpreted as subtraction.
Discarded-module. MLL-AF10 and CALM-AF10 transformation is Dot1l-dependent by conditional knockout and by EPZ004777, for initiation and for maintenance [E1 counter, Chen 2013]. The retained OM-LZ -> DOT1L axis is a sufficient addition mechanism, so the PZP-loss result has to earn its place by explaining something DOT1L dependence does not.
Sequestration. The AF6 RA1 domain is the minimal region sufficient for MLL-AF6 immortalisation and short-latency leukaemogenesis; Ras-binding-dead substitutions do not abrogate oncogenesis; heterologous RA domains and direct fusions to activated K-RAS, H-RAS or RAP1 do not substitute; and RA1 mediates self-association, leading to the conclusion that constitutive MLL self-association is the pathogenic mechanism [E2 counter, Liedtke 2010]. This is close to the falsifier already listed in this entry ('a self-association-dead, nuclear MLL-AF6 still sequesters endogenous AF6') approached from the other direction: the self-association arm is already shown to be the transforming one.
Allelic. Mll(PTD/-) mice die at birth; HoxA up-regulation and increased CFU-GM occur with and without the wild-type allele, which the authors read as dominant gain of function; and increased pluripotent CFU-GEMM progenitors occur ONLY in Mll(PTD/WT), so in that compartment the PTD requires cooperation with wild-type Mll [E3 counter, Dorrance 2008].
The hypothesis is not dead - loss is still documented in each case, and the claim about the project's SEARCH METHOD (that domain-fate analysis cannot see location or transcriptional state) is untouched by any of this. But the prior should come down, and the entry now carries it at 0.35 rather than 0.55.
| statement | Loss-of-function contributions recur across KMT2A-rearranged leukemia in three mechanistically distinct sub-types — discarded-module, sequestration and allelic silencing — of which only the first is detectable by the domain-fate analysis this project uses for target selection. |
|---|---|
| origin | literature-gap |
| prior probability | 0.35 |
| prior revision note | Lowered from 0.55 to 0.35 on 2026-09-19 after the counter-argument sweep: each of the three sub-types has a published challenge (Chen 2013 PMID 23138183; Liedtke 2010 PMID 20395419; Dorrance 2008 PMID 18617636). The corollary about the project's search method is unaffected. |
predictions
[ "In-cis restoration of AF9 YEATS to MLL-AF9 will reduce transformation efficiency and lengthen latency, with the effect abolished by aromatic-cage mutations.", "A self-association-dead but nuclear MLL-AF6 will still sequester endogenous AF6, showing that sequestration is separable from the transforming event.", "The ~16% of MLL-ENL patients with exon-6/7 breakpoints that exclude YEATS will resemble MLL-AF9 in lineage distribution and will be insensitive to SGC-iMLLT.", "Wild-type allele silencing will be found in a subset of translocation-driven KMT2A-r cases and not only in PTD.", "Menin-inhibitor benefit will be smaller, per unit of transcriptional response, in partners with a demonstrated subtraction component than in KMT2A::AFF1." ]
experiments
[ "exp-pzp-cis-resupply-kmt2a-mllt10", "exp-yeats-cis-resupply-kmt2a-mllt3" ]
analyses
[ "ana-partner-subtraction-triage" ]
falsifiers
[ "In-cis restoration fails to suppress transformation in BOTH KMT2A::MLLT10 and MLL-AF9, leaving no KMT2A fusion with a demonstrated discarded-module mechanism.", "A self-association-dead nuclear MLL-AF6 does not sequester endogenous AF6, making sequestration a downstream consequence of transformation rather than a parallel mechanism.", "Allele-specific re-expression of wild-type KMT2A in PTD blasts does not induce death, attributing the Whitman result to hypomethylating-agent pleiotropy.", "A systematic survey finds the untranslocated KMT2A allele intact and expressed across translocation-driven KMT2A-r blasts, confining allelic subtraction to PTD.", "Already partly satisfied: the AF6 RA1 domain is sufficient for MLL-AF6 transformation and acts by self-association with Ras binding dispensable (Liedtke 2010), which is the addition explanation the sequestration sub-type must displace.", "Already partly satisfied: Mll-PTD up-regulates HoxA dominantly, Mll(PTD/-) mice die at birth, and the PTD requires wild-type Mll for the pluripotent-progenitor phenotype (Dorrance 2008), which is the dominant-gain-of-function explanation the allelic sub-type must displace." ]
sub types
[
{
"name": "discarded-module",
"substrate": "protein sequence",
"examples": [
"MLLT10 PZP",
"MLLT3 YEATS (candidate)"
],
"visible_to_domain_fate_analysis": true
},
{
"name": "sequestration",
"substrate": "subcellular location",
"examples": [
"AFDN / endogenous AF6",
"ELL / endogenous EAF1"
],
"visible_to_domain_fate_analysis": false
},
{
"name": "allelic",
"substrate": "transcriptional state of the second allele",
"examples": [
"KMT2A-PTD"
],
"visible_to_domain_fate_analysis": false
}
]- Two of the three sub-types have no inhibitor-shaped therapeutic handle. Establishing that a mechanism is loss-driven may mean establishing that the partner is undruggable by the project's current modalities, which is useful triage and bad news.
- If the class is real, pooling partners in trial analysis hides it. But de-pooling costs statistical power in a disease where infant cohorts are already too small — the same tension `ana-target-safety-triage-v2` hit when it gated a constraint on its own statistical power.
- Is the untranslocated KMT2A allele expressed in translocation-driven KMT2A-r blasts? A single allele-specific expression analysis on existing RNA-seq would settle whether allelic subtraction is PTD-specific, and this graph already holds the dataset entries needed to try.
- Do fusions whose partner protein is normally cytoplasmic (AFDN, EPS15, GAS7, SEPT family) all sequester their endogenous counterparts? If so, sequestration is a property of a structural class and can be predicted rather than discovered case by case.
- Does menin-inhibitor benefit differ between addition-driven and subtraction-driven partners? Trial data are pooled and would not currently show it, but per-partner response depth in AUGMENT-101 could be re-cut.
- Is there any KMT2A fusion for which a loss contribution has been shown to be REQUIRED for transformation, as opposed to documented alongside it? After this sweep the answer appears to be no - the one in-cis restoration result is in CALM-AF10, and the disease it causes is already fully accounted for by DOT1L dependence.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Restoring the PZP domain in cis to CALM-AF10 prevents transformation and abrogates leukemogenesis, establishing at least one KMT2A-family partner protein in which loss of a discarded module is causally oncogenic.
The role of the PZP domain of AF10 in acute leukemia driven by AF10 translocations · Klein BJ, Deshpande A, Cox KL, Xuan F, Zandian M, Barbosa K, Khanal S, Tong Q, Zhang Y, Zhang P, Sinha A, Bohlander SK, Shi X, Wen H, Poirier MG, Deshpande AJ, Kutateladze TG · Nature Communications 2021What this evidence cannot show
CALM-AF10, not KMT2A::MLLT10. Verified against PubMed on 2026-09-18.
Evidence that argues against this (1)
- A complete addition mechanism for AF10 fusions is already published, so the discarded-module sub-type is not needed to explain the disease it rests on. Chen et al. showed by conditional knockout and by pharmacological inhibition that both MLL-AF10 and CALM-AF10 transformation is dependent on Dot1l: Dot1l disruption abolished in vitro transformation and in vivo initiation AND maintenance of both leukaemias, and EPZ004777 suppressed Hoxa-cluster and Meis1 expression and impaired proliferation of transformed cells. The retained OM-LZ recruits DOT1L and DOT1L is necessary. A hypothesis that loss is a recurrent CO-mechanism must explain what loss adds beyond a mechanism that is already sufficient, and for the founding member of the class it has not. Chen L et al. 2013
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
MLL-AF6-rearranged cells relocate endogenous AF6 from cytoplasm to nucleus with consequent RAS activation, and silencing the fusion restores cytoplasmic AF6 and reduces RAS-GTP — a loss of function produced by relocation rather than by deletion.
MLL-AF6 fusion oncogene sequesters AF6 into the nucleus to trigger RAS activation in myeloid leukemia · Manara E, Baron E, Tregnago C, Aveic S, Bisio V, Bresolin S, Masetti R, Locatelli F, Basso G, Pigazzi M · Blood 2014What this evidence cannot show
Does not show that the sequestration is required for transformation. Verified against PubMed on 2026-09-18.
Evidence that argues against this (1)
- The AFDN/AF6 sequestration positive - the single result that made question Q2 look productive - has a published alternative explanation that is an ADDITION, not a loss. Liedtke et al. identified the AF6 RA1 domain as the MINIMAL REGION SUFFICIENT for MLL-AF6-mediated myeloid progenitor immortalisation in vitro and short-latency leukemogenesis in vivo; single amino-acid substitutions known to abolish the AF6-Ras interaction did NOT abrogate MLL-AF6 oncogenesis; fusing MLL to heterologous RA domains of c-Raf1 or RalGDS, or directly to constitutively active K-RAS, H-RAS or RAP1, was not sufficient; and RA1 was shown to mediate efficient SELF-ASSOCIATION, leading the authors to conclude that constitutive MLL self-association is a more common pathogenic mechanism than previously appreciated. On that account the transforming event is oligomerisation of the MLL moiety, and the relocation of endogenous AF6 to the nucleus is a side effect of expressing a nuclear AF6 fragment rather than the mechanism. Both observations can be true; only one of them is shown to be required for transformation, and it is not the sequestration. Liedtke M et al. 2010
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
MLL(PTD/WT) blasts express the PTD allele but not the wild-type allele; silencing is reversed by DNMT and HDAC inhibitors and wild-type MLL induction is associated with selective sensitivity to cell death, supporting a recessive gain-of-function model in which loss of wild-type function contributes to the leukemic phenotype.
The MLL partial tandem duplication: evidence for recessive gain-of-function in acute myeloid leukemia identifies a novel patient subgroup for molecular-targeted therapy · Whitman SP, Liu S, Vukosavljevic T, Rush LJ, Yu L, Liu C, Klisovic MI, Maharry K, Guimond M, Strout MP, Becknell B, Dorrance A, Klisovic RB, Plass C, Bloomfield CD, Marcucci G, Caligiuri MA · Blood 2005What this evidence cannot show
DNMT and HDAC inhibitors are pleiotropic; the causal link to wild-type MLL specifically is not isolated. Verified against PubMed on 2026-09-18.
Evidence that argues against this (1)
- The 'recessive gain-of-function' reading of MLL-PTD, which is the sole basis of the allelic sub-type, is contradicted in the genetically cleanest test of it. Dorrance et al. generated mice expressing Mll-PTD in the ABSENCE of wild-type Mll and compared them with Mll(PTD/WT). Mll(PTD/-) mice die at birth, unlike Mll(PTD/WT), Mll(WT/-) and Mll(WT/WT) animals, so removing the wild-type allele is not simply permissive. Both Mll(PTD/-) and Mll(PTD/WT) fetal liver cells showed increased HoxA expression and increased CFU-GM, which the authors read as the PTD acting in a DOMINANT gain-of-function fashion; and only Mll(PTD/WT) cells showed increased pluripotent CFU-GEMM progenitors, i.e. in selected tissues the PTD REQUIRES cooperation with wild-type Mll to produce the abnormality. Wild-type-allele silencing in patient blasts is a real observation, but the inference that the silencing is what makes the lesion oncogenic does not follow, and in the compartment that matters most for stemness the wild-type allele appears to be needed rather than dispensable. Dorrance AM et al. 2008
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The YEATS reader is retained in 84.1% of MLL-ENL patients and is required for efficient MLL-ENL leukemogenesis, so reader loss is not a general property of the ENL/AF9 partner family.
The ENL YEATS epigenetic reader domain critically links MLL-ENL to leukemic stem cell frequency in t(11;19) Leukemia · Hu H, Saha N, Yang Y, Ahmad E, Lachowski L, Shrestha U, Premkumar V, Ropa J, Chen L, Teahan B, Grigsby S, Marschalek R, Nikolovska-Coleska Z, Muntean AG · Leukemia 2022What this evidence cannot show
Included deliberately as the strongest counter-evidence inside the hypothesis itself: it bounds the claim to specific partners rather than the class. Verified against PubMed on 2026-09-18.
Evidence that argues against this (1)
- The retention statistic that bounds this hypothesis is a literature survey rather than a typed cohort, and the largest systematic breakpoint series shows why that matters. Meyer et al. determined genomic breakpoints in 2,345 patients and found 135 distinct MLL rearrangements, of which only nine fusions account for more than 90% of cases - so per-partner denominators outside those nine are small - with the MLL-side breakpoint position itself varying by age and lineage. Two further problems are structural rather than statistical: a percentage of REPORTED cases is a percentage of what was published and typed, and retention in patients is survivorship, not requirement. A domain present in 84% of fusions may be retained because excluding it is structurally unlikely given where the partner's breakpoints fall, not because its presence is selected for. Meyer C et al. 2018
- depends on →MLLT10 (AF10): the discarded PZP reader — the one partner where loss is proven to be the oncogenic event
Counter-evidence (Chen 2013, PMID 23138183): MLL-AF10 and CALM-AF10 transformation is Dot1l-dependent for initiation and maintenance, so a sufficient addition mechanism already explains the founding member of the discarded-module sub-type
Counter-evidence (Liedtke 2010, PMID 20395419): constitutive MLL self-association via the AF6 RA1 domain is the demonstrated transforming mechanism of MLL-AF6, which is the addition explanation the sequestration sub-type must displace
Counter-argument sweep 2026-09-19: in none of the four 'loss' partners has the loss been shown to be required for transformation (Chen 2013 PMID 23138183; Liedtke 2010 PMID 20395419; Dorrance 2008 PMID 18617636); the claim about the project's search-method blind spot is unaffected
if geography decides, subtraction is not a designed partner class
- ← related toThe third category: modules lost from the fusion protein but supplied in trans by the wild-type KMT2A allele
the same two-axis test should be applied to the partner-side subtraction beat, where it has never been asked
the class claim and the audit both rest on breakpoint percentages that the 2,345-patient recombinome series has never been used to reproduce (Meyer 2018)
- ← testsRe-supply the AF10 PZP module in cis to KMT2A::MLLT10 and ask whether transformation is suppressed
a null result here is one of the two falsifiers of the discarded-module sub-type
the other falsifier of the discarded-module sub-type, plus a cheap human natural experiment
extends the audit from one counterexample to a three-sub-type taxonomy, and bounds it: the two commonest lymphoblastic partners are still addition
the desk analysis that applies the three questions across partners
bounding counter-evidence held inside the hypothesis itself
shares the tension between mechanistic stratification and statistical power in small infant cohorts
the audit that generates and bounds the hypothesis