How much HOXA9 and MEIS1 is actually required? Partial dependence, a Hox code, and why this caps what any single target can do
also called HOXA9 dose · MEIS1 dose · Hox code · Hoxa9-null MLL-AF9 · partial dependence
You might expect that removing HOXA9 would stop the leukemia, but mice without it still develop MLL-AF9 leukemia at the same rate, just with more immature cells. The program is buffered, so partial knockdowns only ever give partial answers.
The same thing, for a scientist
The dose question has a clean and uncomfortable answer. HOXA9 is required for survival of established human KMT2A-rearranged leukemia (proliferation arrest and apoptosis specifically in KMT2A-r lines, p=0.007) but is dispensable for murine Mll-AF9 leukemia initiation: Mll-AF9 knock-in mice with or without Hoxa9 develop myeloid leukemia with similar incidence and time course, with Hoxa9-null leukemias merely more immature. The authors' own conclusion is that no single Hoxa gene is likely necessary, and they propose a 'Hox code' minimally defined by the Hoxa5-a9 cluster. Meis1 knockdown delays rather than prevents leukemia. No allelic-dose titration study exists for either gene. The practical implication is that HOXA9/MEIS1 suppression is a graded, buffered readout, so partial engagement of any upstream target will yield partial program suppression.
The two results that do not agree, and how to hold them1 deeper
Maintenance: required. shRNA HOXA9 depletion across 17 human AML/ALL lines (7 KMT2A-r, 10 KMT2A-germline) caused proliferation arrest and apoptosis specifically in the KMT2A-rearranged cells (p = 0.007); in primary AMLs apoptosis was greater in KMT2A-r samples (p = 0.01); and HOXA9-depleted t(4;11) SEMK2 xenografts had significantly lower leukemia burden. HOXA9 suppression co-downregulated HOXA10, MEIS1, PBX3 and MEF2C 1.
Initiation: dispensable. Mll-AF9 knock-in mice with or without Hoxa9 both showed myeloid expansion at 8 weeks and developed myeloid leukemia with similar incidence and time course. Hoxa9-null leukemias were more immature in phenotype. Mll-AF9 overexpressed Hoxa5, a6, a7, a9 and a10, and the authors concluded that none of these as a single gene is likely to be necessary for leukemia development, proposing a 'Hox code' minimally defined by the Hoxa5-a9 cluster 2.
The reconciliation that fits both: HOXA9 is required to maintain established human disease but is dispensable for murine initiation, because the Hoxa cluster is internally redundant. A germline Hoxa9-null animal has a lifetime for the rest of the cluster to compensate; an acutely knocked-down human cell line does not. Both results are real and they answer different questions.
MEIS1 tells the same story in a third way. shRNA Meis1 knockdown in murine Mll-AF9 cells decreased proliferation, survival and clonogenicity, increased monocytic differentiation, and significantly delayed — but did not prevent — leukemia establishment in transplant recipients, with reduced cell-cycle-entry gene expression 3. Delay, not prevention, is the signature of a dose effect rather than a switch.
What is missing: nobody has done the titration
This is a striking hole for a question the field treats as settled. The verification pass for this cycle found no allelic-dose or haploinsufficiency titration study for either Hoxa9 or Meis1 in KMT2A-fusion leukemia. Everything available is a complete germline knockout (which permits developmental compensation) or an shRNA knockdown (which leaves an unmeasured residual and cannot report a dose-response curve).
The experiment that would settle it is cheap by modern standards: a degron-tagged HOXA9 or MEIS1 at endogenous levels, titrated with degrader dose, reading out self-renewal and differentiation. That gives an actual EC50 for program collapse, which is the number a drug-discovery programme needs and currently does not have.
A second, purely computational version: mine existing KMT2A-r patient RNA-seq for the relationship between HOXA9/MEIS1 expression level and outcome, and look for a threshold rather than a gradient. This is listed as an analysis node in this beat.
Why this sets the ceiling on interface disruption
This is the strategic point, and it deserves to be stated plainly because it cuts against the project's favoured hypothesis.
A menin-interface inhibitor does not delete the fusion; it reduces the fusion complex's chromatin occupancy. Occupancy reduction produces graded reduction of the output program. And the dose data above say the output program is buffered — redundant within the Hoxa cluster, relayed through network feed-forward motifs (mechanism-hoxa9-meis1-output-program), and capable of producing delay rather than prevention when partially reduced.
So the chain is: partial drug occupancy → partial complex eviction → partial output reduction → a buffered program that may be slowed rather than switched off. That is a mechanistic account of how a drug can show textbook target engagement and textbook pharmacodynamics and still fail to cure — and it matches the observation that heavily pretreated KMT2A::AFF1 samples resist menin inhibition with sustained on-target transcriptional effect (see mechanism-fusion-dependency-reversibility-kinetics).
The counterweight, and it is a real one: fusion removal (degron, tet-off) does collapse the program and does cause regression of established disease. So the ceiling is not set by the biology being untargetable — it is set by how completely the complex can be evicted. That reframes the medicinal-chemistry goal from affinity to fractional occupancy and residence time, and it argues that a degrader of the fusion, or a combination that evicts it globally (menin plus DOT1L, menin plus KAT6/KAT7), is mechanistically superior to a better single-interface inhibitor.
| chain | mechanism-kmt2a-fusion-causal-chain |
|---|---|
| hox code | minimally Hoxa5-Hoxa9 cluster; no single Hoxa gene necessary |
| allelic dose titration study | does not exist for HOXA9 or MEIS1 in KMT2A-fusion leukemia |
| ec50 for program collapse | unmeasured |
| strategic implication | partial occupancy gives graded suppression of a buffered program; the medicinal-chemistry goal is fractional occupancy and residence time, not affinity alone |
hoxa9
{
"required_for_maintenance_human": true,
"required_for_initiation_mouse": false,
"p_value_selectivity_lines": 0.007,
"p_value_selectivity_primary_aml": 0.01,
"n_lines_tested": 17,
"n_kmt2a_r_lines": 7
}meis1
{
"knockdown_effect": "delayed but did not prevent leukemia; increased monocytic differentiation"
}preferred strategies given dose biology
[ "fusion degrader", "menin + DOT1L combination", "menin + KAT6/KAT7 combination" ]
- The output program is internally redundant: no single Hoxa gene is necessary, so any therapy aimed at one output gene should be expected to fail, and partial upstream suppression may only slow the disease.
- Because a buffered program can be slowed without being switched off, a drug can display perfect target engagement and pharmacodynamics while failing clinically — and biomarkers built on HOXA9 suppression would look reassuring throughout.
- Hoxa-cluster compensation is an untested but plausible non-genetic escape route from menin inhibition, and nobody profiles the whole cluster during treatment.
- MEIS1 reduction increases monocytic differentiation, which in a lymphoid disease raises the lineage-switch concern rather than resolving it.
- What is the EC50 for collapse of the self-renewal program as a function of HOXA9 or MEIS1 protein level? A degron titration at endogenous levels would produce the number the field has been asserting without measuring for twenty years.
- In patient KMT2A-r RNA-seq, is the relationship between HOXA9/MEIS1 level and outcome a threshold or a gradient? A threshold would justify aiming for partial suppression; a gradient would not.
- Does the Hoxa-cluster redundancy that rescues Hoxa9-null mouse initiation also operate acutely in human cells under drug pressure? If HOXA10 or HOXA7 upregulates during menin-inhibitor treatment, that is a non-genetic resistance mechanism nobody is monitoring.
- Is the Hoxa9-dispensable-for-initiation result specific to the Mll-AF9 knock-in, or does it hold for KMT2A::AFF1 in a lymphoid context where the output program differs?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
shRNA HOXA9 depletion across 17 human AML/ALL lines caused proliferation arrest and apoptosis specifically in the 7 KMT2A-rearranged lines (p=0.007), with greater apoptosis in KMT2A-r primary AMLs (p=0.01) and significantly lower leukemia burden in HOXA9-depleted t(4;11) SEMK2 xenografts; HOXA9 suppression co-downregulated HOXA10, MEIS1, PBX3 and MEF2C.
HOXA9 is required for survival in human MLL-rearranged acute leukemias · Faber J, Krivtsov AV, Stubbs MC, et al. (Armstrong SA, senior author) · Blood 113(11):2375-2385 2009What this evidence cannot show
shRNA knockdown leaves residual protein and reports no dose-response curve, so this establishes requirement but not threshold. All lines used were HOXA-high, so the result does not extend to HOXA-low disease.
Evidence that argues against this (1)
- Mll-AF9 knock-in mice with or without Hoxa9 develop myeloid leukemia with similar incidence and time course, so HOXA9 is dispensable for initiation even though it is required for maintenance of established human disease. Kumar AR et al. 2004
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Mll-AF9 knock-in mice with or without Hoxa9 both showed myeloid expansion at 8 weeks and developed myeloid leukemia with similar incidence and time course, with Hoxa9-null leukemias more immature in phenotype; Mll-AF9 overexpressed Hoxa5, a6, a7, a9 and a10, and the authors concluded no single one of these genes is likely to be necessary, proposing a Hox code minimally defined by the Hoxa5-a9 cluster.
Hoxa9 influences the phenotype but not the incidence of Mll-AF9 fusion gene leukemia · Kumar AR, Hudson WA, Chen W, Nishiuchi R, Yao Q, Kersey JH · Blood 103(5):1823-1828 2004What this evidence cannot show
Germline Hoxa9-null, so lifelong developmental compensation within the Hoxa cluster is not excluded and is in fact the authors' own interpretation. Endpoints are incidence and latency, not leukemia-stem-cell frequency, so a quantitative LSC deficit could be missed. Myeloid Mll-AF9, not lymphoid KMT2A::AFF1.
Evidence that argues against this (2)
- This is murine MLL-AF9 in the myeloid compartment and the index disease of this graph is human infant KMT2A::AFF1 B-lymphoid leukaemia. The two differ in the partner, in the lineage, in the cell of origin and — on this beat's own evidence — in HOXA output, since infant t(4;11) is HOXA-LOW while the mouse models are built on HOXA activation. Retroviral MLL-AF9 also expresses the fusion far above the single rearranged allele a patient carries. The specific inference is also weaker than stated: paralogue redundancy across Hoxa5/6/7/9/10 means a Hoxa9-null leukaemia is not a HOXA-null leukaemia, so the experiment shows Hoxa9 is individually dispensable and says little about the pathway's dose requirement. Oncofusion knowledge graph (internal analysis) 2026
- The more-immature phenotype of the Hoxa9-null leukaemias is a real difference the entry files as incidental. A leukaemia that arises with similar incidence but a different phenotype has had its biology changed, which is not what "no single one is required" conveys. Oncofusion knowledge graph (internal analysis) 2026
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
shRNA Meis1 knockdown in murine Mll-AF9 cells decreased proliferation and survival, reduced clonogenicity, increased monocytic differentiation and significantly delayed — but did not prevent — leukemia establishment in transplant recipients, with reduced expression of cell-cycle-entry genes; MEIS1 knockdown also reduced growth of human KMT2A-fusion lines.
A role for MEIS1 in MLL-fusion gene leukemia · Kumar AR, Li Q, Hudson WA, et al. (Kersey JH, senior author) · Blood 113(8):1756-1758 2009What this evidence cannot show
Short report; shRNA knockdown with residual protein and no dose titration. Delay rather than prevention is consistent with a dose effect but also with partial knockdown efficiency, and the two cannot be distinguished in this design.
Evidence that argues against this (1)
- This is murine MLL-AF9 in the myeloid compartment and the index disease of this graph is human infant KMT2A::AFF1 B-lymphoid leukaemia. The two differ in the partner, in the lineage, in the cell of origin and — on this beat's own evidence — in HOXA output, since infant t(4;11) is HOXA-LOW while the mouse models are built on HOXA activation. Retroviral MLL-AF9 also expresses the fusion far above the single rearranged allele a patient carries. Delayed but not prevented is also the signature of incomplete knockdown rather than of a non-essential gene: shRNA leaves residual protein, and a transplanted population selects for the cells with the least effective knockdown. Oncofusion knowledge graph (internal analysis) 2026
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Degron-mediated removal of the KMT2A fusion protein itself does collapse the target program — elongation changes within 15 minutes — and ablation of an inducible fusion causes regression of established, genetically complex leukemia, indicating the limiting variable is completeness of complex eviction rather than intrinsic untargetability of the program.
MLL::AF9 degradation induces rapid changes in transcriptional elongation and subsequent loss of an active chromatin landscape · Olsen SN, Godfrey L, Healy JP, et al. (Armstrong SA, senior author) · Molecular Cell 82(6):1140-1155.e11 2022What this evidence cannot show
Cited here for the contrast between complete removal and partial inhibition; the regression claim belongs to the separate doxycycline-repressible MLL-ENL study cited in mechanism-fusion-dependency-reversibility-kinetics. Complete degradation is not achievable pharmacologically today for this target class.
Evidence that argues against this (1)
- Collapse of the target programme on degron removal establishes that the programme needs the fusion, and it is silent on the dose question this entry is about. A complete removal experiment cannot measure what fraction of occupancy is required, which is precisely the variable a menin inhibitor manipulates. Journal of Clinical Oncology 2025
- ← analogous toFault trees, minimal cut sets and common-cause failure — reliability engineering on a buffered oncogenic program
the buffered HOXA9/MEIS1 program read as a fault-tolerant system with a minimal cut set
turns the unmeasured dose question into a response-surface experiment
partially opposes the HOXA9/MEIS1 framing: HOXA9/MEIS1 transforms but does NOT reproduce the fusion transcriptome, while PBX3/MEIS1 does
dose requirement and the redundant Hoxa5-a10 code
partial inhibition at several nodes rather than full inhibition at one
recorded as the least-required gene in the KMT2A-r panel; a negative reference point for dose claims
- related to →Step 5 — The output program: HOXA9/HOXA10/MEIS1/PBX3 plus MEF2C, MYB, BCL2, FLT3 and PROM1/CD133
how much of the program is actually required
modifier, not a step - dose constraint on link 5
- ← supportsNecessary is not explanatory: four different questions that a knockout answers as if they were one
the rate-limiting column, where no dose axis exists