Step 5 — The output program: HOXA9/HOXA10/MEIS1/PBX3 plus MEF2C, MYB, BCL2, FLT3 and PROM1/CD133
also called HOXA9-MEIS1 program · MLL fusion target genes · self-renewal signature · HOX code
The genes the fusion keeps switched on are the ones a blood stem cell normally uses only briefly during development. Chief among them are HOXA9 and MEIS1, plus a survival gene (BCL2), a growth-signal receptor (FLT3) and a stem-cell surface marker (CD133).
The same thing, for a scientist
KMT2A-rearranged leukemias carry a highly distinct expression profile consistent with an early haematopoietic progenitor co-expressing multilineage markers and select HOX genes. The core module is HOXA9 with HOXA10, MEIS1, PBX3 and MEF2C; HOXA9 suppression co-downregulates all of them, defining a single coordinated program rather than independent targets. Layered on top are directly bound effectors with clear therapeutic handles: BCL2 (via a 3' H3K27ac enhancer and gene-body H3K79me3), MCL1, FLT3, MYB and PROM1/CD133 (via an intragenic H3K79me2/3 enhancer element). A KMT2A::AFF1 gene-regulatory network built from ChIP-seq, patient RNA-seq and CRISPR screens shows the fusion works through core transcription factors such as RUNX1 in feed-forward-loop and cascade motifs, so many 'targets' are network outputs rather than direct binding events.
One program, not a list of genes2 deeper
The founding observation is that KMT2A-translocated ALL has a characteristic, highly distinct expression profile consistent with an early haematopoietic progenitor expressing select multilineage markers and individual HOX genes — separable by clustering from both conventional ALL and AML, and robust enough to classify cases 1.
The internal structure of that profile was then resolved by knocking down its apex. HOXA9 suppression co-downregulated HOXA10, MEIS1, PBX3 and MEF2C — a program also highly expressed in human KMT2A-rearranged AML and murine KMT2A-fusion leukemia stem cells. HOXA9 depletion across 17 human AML/ALL lines induced proliferation arrest and apoptosis specifically in the 7 KMT2A-rearranged lines (p = 0.007), the same selectivity held in primary AMLs (p = 0.01), and HOXA9-depleted t(4;11) SEMK2 xenografts carried significantly lower leukemia burden 2.
So HOXA9 is the hub. But note the word co-downregulated: this is one regulatory module with shared control, which is why knocking down any single member gives a partial phenotype (see mechanism-hoxa9-meis1-dose-requirement).
The directly bound effectors with drug handles
Beyond the HOX module, three targets matter because each has an approved or advanced drug pointing at it.
BCL2. t(4;11) patient cells express high BCL-2 and are highly sensitive to venetoclax/ABT-199. KMT2A::AFF1 specifically upregulates BCL2 — but not other BCL2-family members — via DOT1L-mediated H3K79me2/3, a t(4;11) line is sensitive to venetoclax plus DOT1L inhibitor, and venetoclax synergises with induction-type chemotherapy in a xenograft 3. Mechanistically the fusion binds a BCL-2-specific enhancer made of two H3K27ac clusters at the 3' end of the gene; fusion loss reduces both gene-body H3K79me3 and 3' enhancer H3K27ac. The same work shows the fusion directly activates BCL2 and MCL1 while repressing BIM through recruitment of the PRC1 component CBX8 4.
PROM1/CD133. Aberrant PROM1 expression is essential for leukemic cell growth and is driven by direct fusion binding at an intragenic H3K79me2/3 enhancer element (KEE), which raises enhancer-promoter contacts between PROM1 and the neighbouring gene TAPT1. In non-expressing cells the PROM1 locus is instead repressed by PRC2 5. This is the field's cleanest single-locus picture of fusion-driven enhancer activation, and CD133 is a surface antigen — i.e. a CAR-T / antibody handle rather than a small-molecule one.
FLT3. FLT3 is part of the KMT2A-r signature and, notably, in a quantitative calibrated-ChIP study of DOT1L inhibition in MV4;11, FLT3 downregulation was the major pathway accounting for the proliferation defect, ahead of polycomb effects 6.
Most 'targets' are network outputs, not binding events
Integrating fusion ChIP-seq, patient RNA-seq and CRISPR essentiality screens yields an explicit KMT2A::AFF1 gene-regulatory network in which core transcription factors — prominently RUNX1 — relay the fusion's output through feed-forward-loop and cascade motifs. A core node set is shared between KMT2A-r ALL and AML, and several network factors mediate venetoclax response 7.
This reframes the whole step. Direct-binding lists systematically over-attribute causation: a gene can be strongly fusion-dependent while never being bound, because it sits downstream of RUNX1 in a cascade. It also predicts that indirect targets should be the ones that rebound first when the fusion is inhibited, because they are buffered by network motifs — a concrete, testable prediction for menin-inhibitor time-course data.
Why the program is developmental, not novel
Nothing in this list is a cancer-specific gene. HOXA9/MEIS1/PBX3 are the normal posterior-HOX self-renewal circuit of fetal and early postnatal progenitors; PROM1 marks normal stem cells across tissues; FLT3 and BCL2 are physiological. The fusion does not invent an oncogene — it fails to let a normal developmental program switch off. Supporting this directly: at HOX loci, H3K79me2 is normally converted to H3K79me1 as haematopoietic cells mature, coinciding with declining HOX expression, and AF10 inactivation reverses the leukemia-associated profile and precludes abnormal HOXA expression 8. The mark is not created de novo by the fusion; its normal resolution is prevented.
That is the deepest reason the menin-interface strategy is attractive and also the deepest reason it may be insufficient: removing the fusion returns the cell to a decision point, it does not delete an oncogene.
| step | 5 |
|---|---|
| chain | mechanism-kmt2a-fusion-causal-chain |
| developmental nature | normal fetal/early-postnatal posterior-HOX self-renewal circuit that fails to switch off |
| rate limiting | downstream; HOXA9 is a maintenance hub but single-gene knockdown is only partially effective |
core hox module
[ "gene-hoxa9", "HOXA10", "gene-meis1", "PBX3", "MEF2C" ]
direct effectors with drug handles
[
{
"gene": "gene-bcl2",
"mechanism": "3' H3K27ac enhancer + gene-body H3K79me3",
"drug": "drug-venetoclax"
},
{
"gene": "MCL1",
"mechanism": "direct activation",
"drug": "MCL1 inhibitors (investigational)"
},
{
"gene": "BIM/BCL2L11",
"mechanism": "repressed via CBX8/PRC1 recruitment",
"drug": "none (de-repression target)"
},
{
"gene": "gene-flt3",
"mechanism": "signature gene; major DOT1L-dependent survival pathway in MV4;11",
"drug": "FLT3 inhibitors"
},
{
"gene": "PROM1",
"mechanism": "intragenic H3K79me2/3 KEE; PRC2-repressed when silent",
"drug": "CD133 CAR-T / antibody"
},
{
"gene": "MYB",
"mechanism": "fusion-dependent",
"drug": "none direct"
}
]network relay factors
[ "RUNX1" ]
- Every gene in the program is a normal developmental gene, so targeting the output directly (HOXA9, MEIS1, FLT3, BCL2) hits normal progenitors and has no fusion-specific window.
- Network buffering: because most targets are relayed through core TFs in feed-forward loops, partial upstream inhibition can leave the output largely intact — the reason transcriptional 'response' and clinical response diverge.
- BCL2 dependence invites venetoclax, but MCL1 is co-activated by the same fusion, giving the cell a built-in escape route from BH3 mimetics.
- HOXA9 is dispensable for leukemia initiation in the mouse, so an anti-HOXA9 strategy may fail to prevent disease even where it suppresses established disease.
- Which members of the output program rebound first after menin inhibition, and are the rebounders the indirect (network-relayed) targets rather than the directly bound ones? A menin-inhibitor time course stratified by direct-binding status would answer this and would predict combination partners.
- Is CD133/PROM1 a viable surface target in KMT2A-r infant ALL, given it also marks normal stem cells across tissues? The therapeutic-index question has not been addressed.
- Is FLT3's prominence in the DOT1L-dependence data a general feature of KMT2A-r or an artefact of the FLT3-ITD-positive MV4;11 model? Repeating the calibrated-ChIP analysis in an FLT3-wild-type t(4;11) line is a cheap and decisive experiment.
- Can the BIM de-repression arm be exploited? The fusion actively represses BIM via CBX8/PRC1, so a CBX8 chromodomain inhibitor plus venetoclax has a specific mechanistic rationale that nobody appears to have tested.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
KMT2A-translocated ALL has a characteristic, highly distinct gene expression profile consistent with an early haematopoietic progenitor expressing select multilineage markers and individual HOX genes, separable by clustering from conventional ALL and AML and sufficient to classify cases.
MLL translocations specify a distinct gene expression profile that distinguishes a unique leukemia · Armstrong SA, Staunton JE, Silverman LB, et al. (Korsmeyer SJ, senior author) · Nature Genetics 30(1):41-7 2002What this evidence cannot show
Early-generation oligonucleotide arrays, modest cohort, and 'MLL' here mixes fusion partners; the profile is a class-level statement and does not resolve partner-specific or HOXA-low cases.
Evidence that argues against this (1)
- A 2002 microarray classification on a small cohort establishes that KMT2A-translocated leukaemia separates from other ALL, and the entry reads it as one output programme. This graph's own sibling entry records the opposite structure: within t(4;11) alone, patients divide into HOXA-high and HOXA-low groups. A 'distinct profile' at the disease level is compatible with substantial heterogeneity inside it. Nature Genetics 2002
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
HOXA9 suppression co-downregulates HOXA10, MEIS1, PBX3 and MEF2C, and HOXA9 depletion induces proliferation arrest and apoptosis specifically in KMT2A-rearranged human lines (7 of 17 lines tested, p=0.007) and preferentially in KMT2A-rearranged primary AMLs (p=0.01), with reduced leukemia burden in HOXA9-depleted t(4;11) SEMK2 xenografts.
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):2375-85 2009What this evidence cannot show
shRNA knockdown, so residual HOXA9 persists and the experiment cannot distinguish a threshold from an all-or-none requirement. Cell lines dominate; only one xenograft. Directly contradicted on the initiation question by the Hoxa9-null mouse result (see counter-evidence).
Evidence that argues against this (1)
- Mll-AF9 knock-in mice with or without Hoxa9 develop myeloid leukemia with similar incidence and time course; Hoxa9-null leukemias are merely more immature. The authors conclude no single Hoxa gene is likely necessary and propose a 'Hox code'. HOXA9 is therefore required for maintenance of established human disease but dispensable for murine initiation. Kumar AR et al. 2004
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
t(4;11) patient cells express high BCL-2 and are highly sensitive to venetoclax; KMT2A::AFF1 specifically upregulates BCL2 but not other BCL-2 family members via DOT1L-mediated H3K79me2/3, a t(4;11) line is sensitive to venetoclax plus DOT1L inhibition, and venetoclax synergises with induction-type therapy in a xenotransplant model.
MLL-Rearranged Acute Lymphoblastic Leukemias Activate BCL-2 through H3K79 Methylation and Are Sensitive to the BCL-2-Specific Antagonist ABT-199 · Benito JM, Godfrey L, Kojima K, et al. (Milne TA, Konopleva M, senior authors) · Cell Reports 13(12):2715-27 2015What this evidence cannot show
Combination data rest on a single t(4;11) line; BCL2 dependence in KMT2A-r AML is less consistent than in ALL, and venetoclax resistance via MCL1/BCL-xL switching is well documented in other settings.
Evidence that argues against this (1)
- High BCL-2 and venetoclax sensitivity in t(4;11) cells is a strong preclinical result whose clinical translation has been partial, and the dependency is not stable: cells moving toward a monocytic state lose BCL2 dependence and shift to MCL1, which this graph records elsewhere as a resistance route. An output programme that includes BCL2 therefore includes a node that the disease can abandon. Cell Reports 2015
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
KMT2A::AFF1 binds directly to a BCL-2-specific enhancer comprising two H3K27ac clusters at the 3' end of the gene; fusion loss reduces gene-body H3K79me3 and 3' enhancer H3K27ac. The fusion directly activates BCL2 and MCL1 while repressing BIM through recruitment of the PRC1 component CBX8.
MLL-AF4 binds directly to a BCL-2 specific enhancer and modulates H3K27 acetylation · Godfrey L, Kerry J, Thorne R, et al. (Milne TA, senior author) · Experimental Hematology 47:64-75 2017What this evidence cannot show
Capture-C and knockdown in t(4;11) lines; the enhancer is pre-existing and modulated rather than created, and the BIM/CBX8 repression arm is a single-locus observation not extended genome-wide in this paper.
Evidence that argues against this (1)
- Direct binding at a BCL-2 enhancer with modulated H3K27 acetylation is occupancy plus a chromatin correlate, and the entry uses it as a causal step in an output programme. The experiment that would close the gap is enhancer deletion with BCL2 measured, which this does not report. Experimental Hematology 2017
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Aberrant PROM1/CD133 expression is essential for leukemic cell growth and is driven by direct KMT2A::AFF1 binding at an intragenic H3K79me2/3 enhancer element (KEE), which increases enhancer-promoter interactions between PROM1 and the neighbouring gene TAPT1; in non-expressing cells the PROM1 locus is repressed by PRC2 binding.
H3K79me2/3 controls enhancer-promoter interactions and activation of the pan-cancer stem cell marker PROM1/CD133 in MLL-AF4 leukemia cells. · Godfrey L, Crump NT, O'Byrne S, et al. (Roy A, Milne TA, senior authors) · Leukemia 35(1):90-106 2021What this evidence cannot show
Cell-line-dominated with primary-sample expression correlation; the PROM1/TAPT1 dual-locus regulation is a single-locus mechanism and its generality across fusion targets is asserted rather than shown.
Evidence that argues against this (1)
- PROM1/CD133 is described by its own source as a PAN-CANCER stem cell marker, so its activation is not specific to this fusion's programme. The mechanism reported is H3K79me2/3 control of enhancer-promoter contact, which places it downstream of DOT1L rather than of the fusion directly. Leukemia 2021
- E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Internally calibrated ICeChIP-seq in MV4;11 (KMT2A::AFF1, FLT3-ITD) shows H3K79me2 is markedly depleted at pinometostat-downregulated fusion target loci with paradoxical increases in H3K4me3 and loss of H3K27me3; polycomb-component downregulation accounts for part of the proliferation defect, but FLT3 downregulation is the major pathway.
Non-canonical H3K79me2-dependent pathways promote the survival of MLL-rearranged leukemia · Richter WF, Shah RN, Ruthenburg AJ · eLife 10:e64960 2021What this evidence cannot show
Single cell line that is FLT3-ITD-positive, so the prominence of FLT3 may be model-specific rather than general to KMT2A-r; perturbation (DOT1L inhibitor) rather than steady-state biology.
Evidence that argues against this (1)
- The calibrated measurement is the most rigorous item here and it complicates the entry rather than supporting it: H3K79me2 depletion at pinometostat-downregulated loci came with PARADOXICAL increases in H3K4me3. An output programme read from mark levels has to account for marks moving in opposite directions under the same perturbation. eLife 2021
- E7moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
A KMT2A::AFF1 gene-regulatory network built from fusion ChIP-seq, patient RNA-seq and CRISPR essentiality screens shows that core transcription factors including RUNX1 relay fusion output through feed-forward-loop and cascade motifs, with a core node set shared between KMT2A-r ALL and AML and several factors mediating venetoclax response.
A KMT2A-AFF1 gene regulatory network highlights the role of core transcription factors and reveals the regulatory logic of key downstream target genes. · Harman JR, Thorne R, Jamilly M, et al. (Milne TA, senior author) · Genome Research 31(7):1159-1173 2021What this evidence cannot show
An integrative computational model: network motifs are inferred from correlated occupancy and expression, with selective experimental validation. Network edges should be treated as hypotheses, not measurements.
Evidence that argues against this (1)
- A gene-regulatory network integrating fusion ChIP-seq, patient RNA-seq and CRISPR essentiality is an inference, and its edges are statistical associations rather than measured regulatory events. Networks of this kind are also sensitive to the essentiality screen used — and this graph has since shown that the KMT2A-rearranged DepMap panel is myeloid-enriched, which biases any essentiality input. Genome Research 2021
- E8strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
At HOX loci, H3K79me2 is normally converted to H3K79me1 as haematopoietic cells mature, coinciding with declining HOX expression; AF10 inactivation reverses leukemia-associated epigenetic profiles, precludes abnormal HOXA expression and impairs transformation by MLL-AF9, MLL-AF6 and NUP98-NSD1.
AF10 regulates progressive H3K79 methylation and HOX gene expression in diverse AML subtypes · Deshpande AJ, Deshpande A, Sinha AU, et al. (Armstrong SA, senior author) · Cancer Cell 26(6):896-908 2014What this evidence cannot show
Mouse models and myeloid fusions; the me2-to-me1 conversion is inferred from population-level profiling across maturation stages, not from single-cell or lineage-traced data.
Evidence that argues against this (1)
- This is an AF10 result across diverse AML subtypes, and the entry is about KMT2A::AFF1. The progressive H3K79me2-to-me1 conversion during maturation is a general property of HOX loci rather than a fusion-specific mechanism, so it explains why HOX genes are poised to respond and not why this fusion's programme looks as it does. Cancer Cell 2014
the program whose cut set is being sought
the same output programme reached without KMT2A
- ← depends onThe differentiation block — what it actually is, operationally, in KMT2A::AFF1 leukaemia
HOXA-type clusters, including the KMT2A::AFF1 cluster, hold a stage-inappropriate HOXA module on
fusion knockdown lowers HOXA7, HOXA9 and MEIS1
the combinatorial test of the output-programme claim; the pair that reproduces the programme contains no HOX gene
output-program transcription factor, not a fusion-complex subunit
rate-limiting regulator of MLL leukemia stem cell potential
- ← depends onFLT3 in KMT2A-r leukemia — near-universal overexpression, but activating mutation in only ~3% of infants
FLT3 is part of the program the fusion switches on
- ← part ofmiR-196b — the HOXA-cluster microRNA that is a direct KMT2A-fusion target and a genuine double agent
encoded within the HOXA cluster and directly switched on by the fusion
- proposed by →Thomas Milne (MRC WIMM, Oxford) — MLL-AF4 target biology, enhancer architecture and PROM1/CD133
MLL-AF4 target biology and enhancer usage
HOXA9 and MEIS1 only appear as co-dependencies in leukemia lines, where the program runs
MEIS1 exceeds its matched null; the output program shows as a co-dependency once the comparison is fair
- ← related toHOTAIRM1 lncRNA — the myeloid-differentiation transcript at the 3' end of HOXA, with a KMT2A-r evidence gap
sits at the differentiation end of the same cluster the output program is read from
the mark sustains the program
- ← depends onThe differentiation block — what it actually is, operationally, in KMT2A::AFF1 leukaemia
the block is the program's output, not a separate event
how much of the program is actually required
long non-coding RNA at the HOXA locus
- ← regulatesCTCF/cohesin boundaries and the HOXA topological domain — the 3D scaffold the fusion program runs on
the loop architecture the program is wired through
- ← related toWhich step is rate-limiting, and what compensates when you block it — the bite-back map for the causal chain
which step is worth blocking, and what compensates
the circuit-level link: H3K14ac sustains the HOXA9/HOXA10 output the fusion depends on
HOXA9/MEIS1 suppression is the standard G2 readout for every gate in the register
RNA-seq readout is the HOXA9/MEIS1 output programme this graph already carries
step 5 - HOXA9/MEIS1 output; maintenance measured, initiation refuted
- ← related toInterfant-06 — 651 infants, and the definitive answer that myeloid-style consolidation does not help
two failed intensification randomisations argue the fusion output program, not chemosensitivity, is the limiting variable