What we could hit, and what it would cost
A good target here has two properties at once. The leukemia cannot manage without it, and normal tissue can. Every entry records both sides, plus the ways hitting it could come back to hurt us.
Five places to break the chain — and what each one costs
Five points of attack on one causal chain, with what each one costs. Two are in children today; one is published and undeliverable; one is a verdict of not feasible, recorded as such rather than left open. The entries below are the evidence behind each row.
Two properties are left off the chart on purpose. Every one of these entries carries published counter-evidence, and every one records its own open questions — both are requirements of the schema rather than findings, so a bar for either could not move, and a statistic that cannot move cannot tell you anything. All of them also carry a bite_back note on how the idea could hurt someone.
Protein-protein contacts the fusion depends on, with measured affinities and real structures.
14-3-3 theta binds AF4 at Ser588 and rides into the MLL-AF4 complex - the phospho-gated contact the fusion really does carry
14-3-3 proteins only stick to partners that have been phosphorylated. One of them, 14-3-3 theta, sticks to AF4 at a single spot and is found in a complex with the MLL-AF4 fusion; removing it makes t(4;11) leukaemia cells stop growing and die.
A CD19 CAR that secretes a CLEC12A bridge: aimed squarely at lineage switch, and it inherits both antigens' failure modes
The hardest failure in treating this leukaemia with CD19-targeted cell therapy is that the cancer changes clothes: it stops being a B-cell leukaemia and comes back as a myeloid one, where the CD19 therapy cannot see it. One design tries to cover both at once - a CD19 CAR cell that also secretes a small adaptor protein letting it attack a myeloid marker. It is clever and it is preclinical, and the catch is that the myeloid marker it picks has since failed in patients for a reason this design does not fix.
ACADM is a mitochondrial matrix beta-oxidation enzyme with an experimentally coded matrix-only localisation, recovered on 8 spectra from a nuclear-extract pulldown whose specific tail is 14.5% mitochondrial — declined
ACADM is an enzyme that burns fat inside mitochondria. The experiment pulled proteins out of cell nuclei, so ACADM should not be there at all — and one in seven of the proteins said to be uniquely bound by MLL-AF4 is a mitochondrial enzyme like it. This is what contamination looks like, and we are declining it.
AF10 fusions recruit and stabilise JAK1 - a kinase bound to the fusion with no phosphosite on the fusion
The AF10 fusion protein physically grabs a signalling enzyme called JAK1 and protects it from being broken down, which switches on an inflammation programme. But nobody has shown that JAK1 puts a phosphate on the fusion itself - the kinase is a passenger, not an editor, as far as anyone has looked.
AFF1 AF4_int motif to AF9/ENL AHD: the index fusion's only documented elongation interface, and it competes with DOT1L
The infant fusion keeps a short piece of AF4 that sticks to a floppy tail on the AF9 and ENL proteins. That same floppy tail is where the H3K79-writing enzyme binds, so the fusion and the enzyme want the same spot.
AFF1-AFF4 heterodimerisation through the C-terminal homology domain: the one SEC interface KMT2A::AFF1 certainly keeps, defined by a 2.2 A structure of the wrong subunit
AFF1 and AFF4 clip together through a domain at their far end, and that domain is inside the leukaemic fusion protein. The structure of the clip was solved on AFF4; AFF1's copy has never been solved at all.
AFF1-NPM1 proximity: nucleophosmin is the only bait in any published proximity experiment whose neighbourhood contains AFF1, KMT2A and menin together
Nucleophosmin is a shuttling nuclear protein that is itself one of the commonest things broken in adult leukaemia. When it was tagged and its neighbours listed, the list contained AFF1, KMT2A and menin - the three proteins at the centre of this project - and no other tagged protein managed that.
AFF1-RBM15 proximity: the m6A-complex adaptor, itself a paediatric leukaemia fusion partner, has AFF1 in its BioID neighbourhood
RBM15 directs a chemical tag onto RNA and is itself broken and fused in a leukaemia of infants. Tagging RBM15 and listing its neighbours in a leukaemia cell turns up AFF1, KMT2A and DOT1L.
AFF1/AFF4-CAMKV - DECLINED, and it is a LABELLED REDISCOVERY: data/symbol-collisions.json already holds a CAMKV row saying BioPlex 2017 and 2021 are one programme and that CAMKV is deposited under split isoform accessions
A brain vesicle protein shows up with both AFF scaffolds, in three separate large surveys, in two different cell lines. It is the most repeated hit on the list and it is still the same kind of experiment done three times.
AFF4-BTK - DECLINED. One osteosarcoma AP-MS row, and it is the only partner in the uncovered prey-only tail with an approved drug against it, which is exactly why the decline is written down
Bruton's tyrosine kinase, the target of ibrutinib, turned up with the elongation-complex scaffold AFF4 in one bone-cancer-cell experiment. BTK belongs in B cells, this experiment was not in one, and one row is not enough.
AFF4-CDK15: AFF4 is prey in a systematic CMGC-kinase TAP interactome - a kinase adjacent to the complex CDK9 works in, and nobody has followed it up
A systematic survey of which proteins stick to each human CMGC-family kinase found AFF4, the scaffold of the elongation complex, attached to the poorly studied kinase CDK15. Nobody has looked at it since.
AFF4-FMR1 - DECLINED, and the provenance is the finding: the 'publication' behind this IntAct row is HENA, an Alzheimer's data-integration dataset, not an experiment on AFF4
A database lists FMRP, the fragile-X protein, as an AFF4 partner. The paper it points to is not an experiment at all - it is a project that merged other people's databases into one big network for Alzheimer's research.
AFF4-MED10: AFF4 is prey in a 2004 Mediator MudPIT purification - a lead with an independent mechanistic reason to be true, and no experiment that tests it
AFF4, the scaffold of the elongation complex that KMT2A fusions hijack, turned up as a passenger in a 2004 purification of the Mediator coactivator. There is a known route by which that could be real, and nobody has ever tested it directly.
AFF4-SDCBP (syntenin): the only prey-only row in this census typed 'direct interaction' - a proteomic phage-display hit on AFF4's C-terminal peptide, in vitro
A screen that displays the last few amino acids of every human protein on bacteriophage found AFF4's tail binding syntenin's peptide-binding module. It is a real biochemical measurement of a short motif, done entirely in a test tube with no cell.
AKAP95 IDR1 co-condenses with the KMT2A N-terminal fragment - the one AKAP95 contact that lands inside the region every KMT2A fusion retains
The part of KMT2A that survives in every leukaemia fusion is a floppy, shapeless stretch. AKAP95 has a floppy stretch of its own that sticks to it, and the two form droplets together inside the nucleus - which is how AKAP95 gets pulled onto the genes the fusion switches on.
AKAP95-DPY30: a direct contact whose binding determinant two groups map to OPPOSITE ends of AKAP95, and neither has been structurally resolved
AKAP95 grips DPY30, the small shared part of every machine that writes the activating H3K4 mark. Two labs have tried to find which piece of AKAP95 does the gripping, and they got answers at opposite ends of the protein - nobody has yet solved the structure that would settle it.
ASH2L SPRY and RBBP5 347-356: a phospho-reader contact gated by RbBP5 Ser350, and the only WRAD interface with a measured fold-change
Two of KMT2A's four helper proteins stick to each other only when a single serine on one of them carries a phosphate. Remove the phosphate and the pair falls apart, which shuts down the enzyme - in the normal protein, not in the fusion.
BET bromodomain inhibition in KMT2A-rearranged leukemia — spectacular preclinically, stalled clinically
BET proteins are the clips that hold the transcription machinery onto DNA at the genes this leukemia depends on. Drugs that pop those clips off work beautifully in mice and in dishes, but no BET inhibitor has become a leukemia treatment, mainly because of toxicity in blood and gut.
CD15 (Lewis-x, SSEA-1, 3-fucosyl-N-acetyllactosamine) - a sugar, not a protein: the most-named antigen in this graph that no gene encodes
CD15 is not a protein at all - it is a small branched sugar stuck onto the outside of proteins and fats by enzymes, so no single gene makes it. It appears in 24 entries of this knowledge graph and could never have had an entry written by our automatic tool, because that tool looks up gene names.
CD19 CAR-T in infants with KMT2A-rearranged ALL — it works, it is manufacturable, and it does not hold
CD19 CAR-T is a treatment that reprograms a child's own T cells to hunt leukemia. In babies it can be made and it works at first, but most of the babies with KMT2A-rearranged leukemia eventually relapse, often with a leukemia that has changed lineage and no longer carries the CD19 target.
CD22-directed therapy (inotuzumab ozogamicin) — the second B-lineage antigen, with high response and poor durability
CD22 is a second marker on B-cell leukemia that can be targeted when CD19 stops working. A CD22-targeted antibody carrying a chemotherapy warhead puts most children back into remission, but only a minority are alive a year later.
CD65 - a neutral glycolipid epitope (V3FucnLc8Cer), no gene, and the CD number that was split in two because its antibodies did not agree
CD65 is not a protein and not even a sugar on a protein - it is a sugar chain on a fat molecule in the cell membrane, so no gene makes it. It is also a cautionary tale: the antibodies originally grouped under this one CD number turned out to recognise two different things, which is why CD65 and CD65s are separate entries here.
CD65s (VIM-2, sialylated ceramide dodecasaccharide) - the myeloid antigen that, with CD15, makes NG2 a perfect predictor of KMT2A rearrangement
CD65s is a sugar chain on a fat molecule in the cell membrane, carrying a sialic acid on the end, and no gene makes it. When a leukaemia cell carries it together with CD15, a third marker called NG2 identifies a KMT2A rearrangement with perfect accuracy in the one large adult study that measured it.
CDK4/6 inhibition in KMT2A-rearranged leukemia — the class was already tried, and the trial never reported
Drugs that block the cell-cycle enzymes CDK4 and CDK6 are already approved in breast cancer, and laboratory work says KMT2A-fusion leukemia cells depend on CDK6 specifically. A trial of one of these drugs, restricted to patients with this exact leukemia, finished in 2019 and has never published a result.
CDK9 inhibition in KMT2A-rearranged infant leukemia — the most direct hit on the hijacked machinery, and the narrowest window
CDK9 is the engine the fusion protein steals to force cancer genes to be read continuously. Blocking it shuts down the two proteins this leukemia most depends on, MYC and MCL-1, but CDK9 is needed by every cell in the body, which is the problem.
CEP112 is a 955-residue centrosomal coiled-coil protein on 7 spectra with zero leukaemia literature — its only nuclear annotation is a single 2024 citation, and a long coiled-coil is the classic non-specific AP-MS passenger — declined
CEP112 is a long, rope-like protein normally found at the centrosome, the structure that organises cell division. Long rope-like proteins stick to things non-specifically in this kind of experiment. Seven detections, no leukaemia literature at all, and nothing that connects it to our fusion.
CK2 phosphorylation of KMT2A S136/S142 tunes the LEDGF grip - a second, non-menin, phospho-gated handle on the fragment every fusion keeps
The broken KMT2A protein is usually described as having only one grip a drug can attack, the one menin holds. There is a second grip, held by a protein called LEDGF, and a cell enzyme called CK2 tightens it by adding phosphate to two spots that every leukaemia fusion keeps.
Cks1 and Cks2 hold both halves of KMT2A and set its level through the cell cycle
Two small adaptor proteins, Cks1 and Cks2, grip both halves of KMT2A and control how much of it exists at each stage of the cell cycle. Both are overexpressed in KMT2A-rearranged leukaemia, and drugs that block the disposal machinery they work with slow those cells more than normal ones.
Correcting the fusion with genome editing: verdict NOT FEASIBLE for infant leukaemia, though intron-targeted disruption is a real published strategy
You cannot glue the chromosomes back the way they were. What you can do is cut the fused gene again so it stops working, and this has been shown to shrink tumours in mice. Getting that machinery into every leukaemia cell in a sick baby, without hitting healthy cells, is the part nobody can do.
CXXC to unmethylated CpG DNA: nine mapped contact residues, and an affinity that spans 44 uM to 33 nM depending on the construct, the DNA and the salt
The fusion protein finds its genes by gripping DNA at unmethylated CpG sites with a small zinc-containing module. How tightly it grips depends enormously on how it is measured - published values run from 44 micromolar to 33 nanomolar - so this entry gives the contact residues and points at the measurement series rather than quoting one number.
CXXC-DNA interface of the KMT2A fusion: a validated but hard target
Every cancer-causing KMT2A fusion keeps a small hand-like module that grips unmethylated CpG DNA, and if you break that grip the leukemia cannot form. The problem is that grabbing DNA is a shallow, charge-driven contact, which is exactly the kind of surface small-molecule drugs are worst at.
DAXX and ATRX as drug targets: six ChEMBL rows and zero of them an affinity, against menin's 1,776 - but three crystal structures of the interface that would have to be blocked
If a cell needs DAXX, the next question is whether anyone could make a drug against it. The honest answer is that nobody has tried: there is one chemistry record in the whole public database and it is not a binding measurement. But the shape of the join between DAXX and its partner has been solved to better than 2 angstroms, so the target is not structurally blind.
DPY30 D/D domain and the ASH2L SDI helix: four hydrophobic pockets, one load-bearing leucine - and a 94-residue numbering offset that relocates the whole motif
DPY30 works as a pair and grips a single short helix from ASH2L in a cradle made of four small pockets. One leucine does most of the work, and swapping it out breaks the grip and lowers the histone mark the complex writes.
ELL-EAF1: a phosphorylation-gated interface reported to drive phase separation - and the assays used cannot distinguish that from clustered binding
ELL and its partner EAF1 are reported to form liquid-like droplets that a damage kinase makes form more readily, which if true is a kinase-controlled droplet on a piece of protein the leukaemia fusion keeps. But the tests used to call something a droplet cannot tell a droplet apart from a lot of protein bound to nearby sites, which is a different thing.
FLT3 in ZNF384-rearranged leukaemia - the fusion builds its own enhancer 25 kb upstream, and gilteritinib works in xenografts
The ZNF384 fusion protein switches on a growth-signalling gene called FLT3 by activating a stretch of DNA 25,000 letters away and physically looping it round to the gene. Because the leukaemia then depends on FLT3, an existing FLT3-blocking drug slows it down in mice.
FYRN/FYRC: the clip that holds the cut reciprocal protein together, and the only fusion-adjacent interface anyone has tried to break with a decoy
After the cut, the two halves of the mirror-image fusion protein do not drift apart - a pair of matching surfaces clips them back together. Blocking that clip with a decoy piece destroys the whole oncoprotein in a dish.
GLUD1 is a hexameric mitochondrial matrix dehydrogenase recovered on 11 spectra with non-zero control signal from a nuclear pulldown — and this corpus has already declined one KMT2A partner whose sole source was a cryo-EM structure of GLUD1 — declined
GLUD1 is a very abundant enzyme that lives inside mitochondria and works as a six-part barrel. It turns up in a pulldown done on cell nuclei, alongside fifteen other mitochondrial enzymes, and it also appeared in the negative control. This graph has already met GLUD1 once before as a passenger in somebody else's experiment.
HIV-1 integrase on the LEDGF IBD: the same pocket as KMT2A, a thousandfold tighter, and a published competition that displaces the leukaemic complex at IC50 301 nM
The part of LEDGF that grips KMT2A is the same part HIV's integrase grips, and integrase grips it about a thousand times harder. Adding integrase to the leukaemic complex pulls LEDGF off it - which is both the clearest proof the pocket can be drugged and the reason a drug for it will be hard to make selective.
HLA-DR - a two-chain MHC class II heterodimer, so no single gene symbol denotes the surface marker a flow report calls HLA-DR
HLA-DR is the flag that immune cells use to show fragments of proteins to other immune cells, and it is made of two separate chains from two separate genes. When a flow-cytometry report says HLA-DR it means the assembled pair on the cell surface, which is why no single gene name can stand for it.
KMT2A 747-755 on HLA-B*07:02: a phosphoneoantigen from the half of KMT2A every fusion keeps, with seven crystal structures and a cognate TCR
A nine-amino-acid piece of KMT2A, carrying a phosphate, is displayed on the surface of leukaemia cells by a common tissue type and is recognised by a specific T-cell receptor. That piece sits in the part of KMT2A that every leukaemic fusion keeps.
KMT2A-ABT1 is one spoke-expanded BioPlex 3.0 AP-MS row in HEK293T in which ABT1 is the bait and KMT2A the prey, and ABT1's nucleolus is ECO:0000250 only - declined
A database lists ABT1, a small nucleolus protein, as a KMT2A partner. The experiment tagged ABT1 and pulled it out of kidney cells; KMT2A merely turned up in the catch, and nobody has ever done the reverse experiment.
KMT2A-AKAP8 (AKAP95): graded 'prey-only' by this project, and it is a direct positive modulator of MLL H3K4 methyltransferase activity from a dedicated MLL paper - ADMITTED
AKAP95 is a nuclear scaffolding protein that sits on the MLL machine and makes it better at putting the activating H3K4 mark on chromatin. Our own bookkeeping filed it as a low-grade 'passenger' hit, because the experiment that found it used a different protein as the hook.
KMT2A-ANOS1 is one spoke-expanded BioPlex 3.0 AP-MS row with ANOS1 - a secreted, cell-membrane axon-guidance protein - as the bait and KMT2A as the prey in HEK293T - declined
A database lists anosmin-1, a protein cells secrete to guide growing nerve fibres, as a KMT2A partner. The single supporting experiment tagged anosmin-1 in kidney cells and found KMT2A in the catch, which is not a place a secreted protein and a chromatin enzyme can plausibly meet.
KMT2A-CLOCK: one paper, six IntAct records, and the bait/prey grade splits it down the middle - CLOCK falls in the discard bin and SIRT1 does not, from the same experiments
A 2015 paper showed that the body clock controls the MLL1 enzyme by switching an acetyl group on and off, which makes the activating H3K4 mark rise and fall through the day. Our filing system kept one partner from that paper and threw the other away, purely because of which protein the antibody was pointed at.
KMT2A-DRC5 - DECLINED. A ciliary dynein-regulatory-complex protein, recovered by BioID in a centrosome-cilium map, and the only one of 135 uncovered partners never mentioned anywhere in this corpus
The one candidate in this whole list that nobody here has ever written a word about turns out to be a cilium protein, picked up as a neighbour of KMT2A in a map of the cell's microtubule-organising centre.
KMT2A-EN1 declined: one spoke-expanded BioID row from an ectopic neural homeobox construct in a kidney cell, and the only screen that ever looked found EN1 unexpressed in AML lines
EN1 is the gene that patterns the midbrain, and it showed up beside KMT2A in a proximity screen done in kidney cells. We are declining it as a partner, because a published screen that deliberately looked for EN1 in leukaemia cell lines found it switched off.
KMT2A-FGFBP1 is spoke-expanded BioPlex 3.0 AP-MS recovered in BOTH HEK293T and HCT116 - it passes BioPlex's own replication criterion and the partner is a secreted extracellular protein - declined
A database lists FGFBP1, a protein cells release into the space outside them to hand growth factors around, as a KMT2A partner. Two different cell lines gave the same answer - which is exactly the check that is supposed to confirm an interaction, and here it confirms one that cannot be real.
KMT2A-HMGN5: recovered by endogenous CRISPR tagging in OpenCell, not by over-expression - a chromatin-decompacting nucleosome binder with a real reason to be near KMT2A
HMGN5 is a small protein that wedges into nucleosomes and loosens chromatin. It came up with KMT2A in a study that tagged proteins at their own genes rather than flooding cells with extra copies, which removes the commonest reason to disbelieve this kind of hit.
KMT2A-KLHL22 - DECLINED. The single source is a cryo-EM structure of KLHL22-CUL3 bound to glutamate dehydrogenase, and KMT2A is a HEK293F pull-down passenger with no nuclear rationale
A database lists the ubiquitin-ligase adaptor KLHL22 as a KMT2A partner. The only paper is a structure of that ligase gripping a mitochondrial metabolic enzyme, and KMT2A just came along in the pull-down.
KMT2A-KRR1 is one spoke-expanded BioPlex 3.0 AP-MS row in HEK293T with KRR1 as bait and KMT2A as prey, sitting in a seven-member ribosomal block at the floor score - declined
A database lists KRR1, a protein that helps build ribosomes in the nucleolus, as a KMT2A partner. The experiment tagged KRR1 and pulled it out of kidney cells, and KMT2A was one of the things that came along - as it did for six other ribosome proteins in the very same screen.
KMT2A-OASL is one spoke-expanded BioPlex 3.0 AP-MS row with OASL as bait and KMT2A as prey; OASL is the only partner in the beat natively expressed in blood leukocytes, and its nucleolar annotation is uncoded - declined, with the compartment argument unavailable
A database lists OASL, an interferon-induced antiviral protein, as a KMT2A partner. The one supporting experiment tagged OASL in kidney cells and found KMT2A in the catch; unlike most partners on this list OASL really is made by blood cells, so the usual objection - wrong compartment, wrong tissue - does not apply and the decline rests only on how thin the evidence is.
KMT2A-PRKRA is one spoke-expanded BioPlex 3.0 AP-MS row with PRKRA (PACT) as bait and KMT2A as prey; PRKRA's only UniProt compartments are cytoplasm and perinuclear region, both uncoded - declined
A database lists PACT, a cytoplasmic protein that switches on the antiviral kinase PKR and helps make small RNAs, as a KMT2A partner. The one supporting experiment tagged PACT in kidney cells and KMT2A came along in the catch; PACT is not recorded as a nuclear protein and the pair has never been tested the other way round.
KMT2A-TMCO1 is a single DSSO crosslink in one detergent fraction of isolated U2OS nuclei with no bait on either side; TMCO1 is an ER-membrane calcium channel and the source paper attributes 4% of its links to nucleus-ER contamination - declined
A database lists TMCO1, a calcium channel that sits in the membrane of the endoplasmic reticulum, as a KMT2A partner. The single supporting measurement is one chemical crosslink found in a detergent fraction of isolated nuclei, in a paper whose authors say some of their nucleus-to-ER links come from contamination during nuclei purification.
KMT2A::AFF1–LEF1 is one AP-MS row (p 0.0261, 10|8 spectra) in a Jurkat T-ALL host, where LEF1 is a resident lineage factor; it is the only one of the six AF4-specific candidates with non-zero negative-control signal, and no Wnt co-factor was co-recovered — declined
LEF1 was the most promising of six proteins said to bind only the MLL-AF4 fusion. But the experiment was done in T-cells, LEF1 is a T-cell protein, and LEF1 was the only one of the six that also showed up in the negative control. We are declining it as evidence of a real partnership while keeping the Wnt question open.
LEDGF IBD to KMT2A: TWO motifs, not one - IBM1 at 124-133 and IBM2 at 147-152, and the Kd this graph recorded as missing is 14.7 uM
LEDGF does not grab the KMT2A fragment at one place but at two, and on its own the grip is weak; menin tightens it roughly tenfold. Both grips sit in the piece of KMT2A that every leukaemic fusion keeps.
LEDGF PWWP chemistry exists after all - 68 fragment hits, double-digit micromolar leads - and the pocket has a structural ceiling the NSD PWWPs do not
A 2024 fragment screen did find small molecules that sit in LEDGF's histone-reading pocket. They bind weakly, and the pocket is smaller than the related pockets where much stronger drugs were made, so there is a ceiling on how good they can get without a new idea.
LSD1/KDM1A inhibition — a differentiation therapy with real clinical pharmacodynamics and no randomised proof
LSD1 is an enzyme that keeps leukemia cells locked in an immature state. Blocking it makes the cells grow up and stop dividing, and in patients it visibly forces the blasts to differentiate, especially in KMT2A-rearranged disease.
MBM2, the second menin-binding motif of KMT2A (residues 23-40): 26-fold weaker than MBM1, never separately co-crystallised, and the only part of the menin grip a drug is not aimed at
KMT2A holds menin mainly with one short motif, but it has a second, much weaker one just downstream. Every menin drug is built against the strong motif, and nobody has shown what the weak one does when the strong one is blocked.
MCL1 plus BCL-2 plus dexamethasone: a triple combination that matched induction chemotherapy in infant PDX models
Blocking two survival proteins at once - BCL-2 with venetoclax and MCL1 with S63845 - alongside the steroid dexamethasone put nearly every mouse carrying an infant leukaemia graft into lasting remission, as well as three-drug chemotherapy did. On its own the MCL1 blocker did almost nothing in mice, and combined with venetoclax alone it killed two thirds of them from tumour lysis.
MEN1-AKT1 is ONE BRET well from a mutation screen, for the E17K allele only - declined
A database lists AKT1, a central growth-signalling kinase, as a menin partner. The entire basis is a single well in one automated screen, testing the mutant AKT1 E17K rather than the normal protein.
MEN1-BRAF V600E: four methods and an endogenous menin lane - the only one of the nine that is a real observation, and it is still a property of the allele
Of nine cancer genes a database lists as menin partners, this is the only one with real experiments behind it: menin's own protein was pulled down with mutant BRAF in a cancer cell. But the interaction is made by the BRAF V600E mutation, not by menin, and that mutation does not occur in the leukaemia this project studies.
MEN1-CD226 is two proteome-scale AP-MS maps, both spoke-expanded, both in cell lines that do not express CD226 - declined
A database lists CD226, an immune-cell surface receptor, as a menin partner. Both supporting experiments overexpressed CD226 in cell lines that never normally make it - kidney, colon and bone cancer lines - and neither measured the menin pair directly.
MEN1-EGFR is ONE BRET well from the extended-driver arm of a mutation screen, for L858R only - declined
A database lists EGFR, the lung-cancer receptor, as a menin partner. It rests on one well in an automated screen, using the mutant EGFR L858R, and EGFR sits in the cell membrane while menin sits in the nucleus.
MEN1-FBXW7 is four alleles of ONE BRET assay, all typed 'mutation with no effect' - declined, and it is the most insidious class
A database lists the ubiquitin-ligase adaptor FBXW7 as a menin partner. The basis is one automated screen run over four FBXW7 mutants, none of which changed anything, so the row looks like an ordinary interaction while resting on a single assay.
MEN1-PIAS4 - DECLINED. The single supporting publication is a PARP-1 heat-shock sumoylation study in which menin is never named, and the row is a HeLa co-IP passenger
A database lists the SUMO-attaching enzyme PIASy as a menin partner. The only paper behind it is about how heat shock changes PARP-1, and menin is not mentioned in it at all.
MEN1-PTEN is ONE BRET well from the extended-driver arm, for R173C only - declined
A database lists the tumour suppressor PTEN as a menin partner. It rests on one well in an automated screen, testing the mutant PTEN R173C rather than the normal protein.
MEN1-SMAD4 is four alleles of ONE BRET assay, all 'mutation with no effect' - declined, and the real menin-TGF-beta literature is elsewhere
A database lists SMAD4, the central TGF-beta signalling protein, as a menin partner. The basis is one automated screen run over four SMAD4 mutants, none of which changed the signal.
MEN1-SPOP is four alleles of ONE BRET assay, all 'mutation with no effect' - declined, with a nuclear-speckle confound
A database lists SPOP, a ubiquitin-ligase adaptor, as a menin partner. The basis is one automated screen over four SPOP mutants, none of which changed the signal, and both proteins concentrate in the same crowded nuclear bodies where a proximity assay is hard to interpret.
MEN1-ZC3HC1 is one spoke-expanded BioPlex row in HCT116, and UniProt records ZC3HC1 weakly or not expressed in leukocytes - declined
A database lists ZC3HC1, a nuclear-envelope protein, as a menin partner. It rests on a single colon-cancer-cell experiment that never measured the pair directly, and the protein is barely present in blood cells at all.
Menin binds RPA2 directly, the contact maps to RPA2 aa 43-171 across its low-affinity ssDNA site, and three MEN1 patient missense mutations abolish it - a 2003 finding that this graph had never cited despite menin being its central node
Menin is the protein that the leukaemia fusion in this project grabs hold of, and it is also the protein whose loss causes an inherited tumour syndrome. Twenty-three years ago it was shown to bind directly to RPA2, part of the clamp that covers exposed single-stranded DNA - and mutations found in patients with that syndrome break the binding. This graph, built around menin, had never recorded it.
Menin-FANCD2: one 2003 co-immunoprecipitation, a curated sufficient region of menin 214-390, and nothing since
In 2003 one laboratory showed that menin and the Fanconi repair protein FANCD2 can be pulled out of a cell together, and that the two stick together more after radiation. Nobody has repeated that experiment in a human blood cell, nobody has solved a structure of the pair, and the only later study designed to test whether the two proteins actually work together concluded that they do not.
Menin-JunD: the same pocket, the same drug, the opposite job
The groove on menin that leukaemia drugs plug is the same groove that holds JunD, a growth-driving protein that menin keeps switched off in hormone glands. One groove, two jobs, and a drug cannot tell them apart.
Menin-SETD2, mapped to the NTD and Palm and tooled with a peptide at menin 231-248: the third partner assigned to the same half of menin, and the writer arm of the HR scaffold
Menin holds SETD2, the enzyme that writes the H3K36me3 mark used to guide accurate DNA repair, and a short menin peptide covering residues 231 to 248 is used as a tool to break that hold. This is the enzyme half of the repair scaffold whose histone-binding half is menin's 371-400 motif.
Menin's nucleosome grip (8GPN): a three-patch surface at menin 353-362, 428-442 and 556-587 that holds H3, H4 and H2B - and touches H2A at exactly one residue
A cryo-electron microscopy structure shows how menin grips a whole nucleosome: it holds histones H3, H4 and H2B firmly but barely touches H2A. That matters because a 2026 paper says menin binds the damage-marked version of H2A, so whatever that contact is, it is not the grip seen here.
Menin's two nuclear localisation signals, 479-497 and 588-608: mapped in 1998, verified residue-for-residue today, annotated by UniProt nowhere - NLS-1 in no structure at all, NLS-2 resolved in one
Menin carries two separate address labels that tell the cell to send it into the nucleus, both near its far end, and no protein database records either of them. The first has never been seen in any 3D structure and the second has been seen in just one of 69, so anything that cuts menin short leaves the protein stranded outside the nucleus where it cannot work.
Mitochondrial complex I plus venetoclax in KMT2A-rearranged paediatric AML - a genotype-stratified in vivo PDX result, on a drug whose adult clinical development stopped for toxicity
In mice carrying real children's leukaemia samples, a drug that blocks the first step of the mitochondrial energy chain, given together with the approved drug venetoclax, slowed KMT2A-rearranged myeloid leukaemia at both diagnosis and relapse. The difficulty is that the same drug caused nerve damage and dangerous acid build-up when it was tried in adults.
Myeloid-directed antigens (CD33, CD123) — the only immunotherapy axis that survives lineage switch
When KMT2A-rearranged leukemia escapes CD19-targeted therapy by turning into a myeloid leukemia, it usually starts wearing myeloid markers like CD33 instead. Drugs aimed at those markers are the only immune option that follows the leukemia where it goes.
Newborn screening for leukaemia fusions: verdict NOT FEASIBLE TODAY, and the reason is arithmetic plus the absence of anything to offer a positive
In principle you could test every newborn's blood spot for the leukaemia fusion. In practice the disease is so rare that almost every positive result would be wrong, the test would have to be near-perfect to avoid that, and there is nothing useful to offer the families who tested positive. It fails on all three counts at once.
NG2/CSPG4 is the one antigen in the whole surfaceome that KMT2A::AFF1 itself drives - and it is a pericyte and vascular protein, at 124.5 nTPM in normal blood vessel and 246 TPM in tibial artery
NG2 is the only surface protein this leukaemia makes because of its fusion gene rather than because of the cell type it came from, which is exactly the kind of target that a leukaemia cannot escape by changing cell type. The problem is that the same protein coats the cells wrapped around blood vessels throughout the body, so a therapy aimed at it would also be aimed at the child's blood vessels.
Phase-2 dossier: CBX8 - the candidate this review would kill, and the reasons are all numeric
CBX8 looked promising on paper and even has a crystal structure with an inhibitor in it. But when we looked at the actual data, cancer cells do not need it, the two best papers about it disagree with each other, and the chemistry that exists is a hundred times too weak. We recommend not starting this one.
Phase-2 dossier: DOT1L - perfect chemistry, a completed paediatric trial, and a target that has already had its answer
DOT1L was the textbook case: a clean enzyme, a picomolar drug, and a paediatric trial that actually finished. It did not work well enough on its own, so the only honest phase-2 question left is whether it is useful in combination.
Phase-2 dossier: KAT7/HBO1 - the only enzyme here with a real active site, a co-crystal inhibitor and a co-dependency that survives lineage adjustment almost intact
KAT7 is an enzyme that puts a chemical tag on histones which leukaemia stem cells need to keep their key genes switched on. It has a proper drug-shaped pocket, a published inhibitor and a crystal structure showing exactly how that inhibitor binds - and unlike most of the others here, the evidence that leukaemia needs it more than normal cells do holds up when you correct for cell type.
Phase-2 dossier: the CXXC-DNA interface - two structures, a genetically validated requirement, and a selectivity problem that decides everything
The fusion protein carries a small module that grips unmethylated DNA and tells the complex where to sit. Deleting it stops the cancer in mouse experiments, and there are good pictures of it. The problem is that about thirty other human proteins carry the same module, so a drug that blocks one would very likely block them all.
Phase-2 dossier: the ENL YEATS pocket - two co-crystals, nanomolar chemistry, and a co-dependency that only appears inside leukaemia
ENL is a reader protein that recognises a chemical tag on histones and helps keep the fusion programme switched on. Unusually for this field it already has a published inhibitor with a solved structure, and a degrader exists too - so the question is not whether we can block it but whether blocking it matters.
Phase-2 dossier: the fusion junction - a negative feasibility finding, with the one surviving modality named
The join between the two halves of the fusion protein is the only thing that is unique to the cancer, so it is the most attractive target imaginable. The problem is that there is no picture of it, both halves are floppy and shapeless, and a small molecule needs a solid pocket to hold on to. The one approach that could still work is showing a fragment of the join to the immune system.
Phase-2 dossier: the LEDGF/PSIP1 IBD - the best therapeutic index in the interactome, a solved structure, two hotspot residues, and three compounds in the world
LEDGF is the protein that anchors the fusion complex to the right place on DNA. Cells that carry the fusion need it badly while normal cells barely notice losing it, which is exactly the profile a good drug target should have - and almost nobody has made chemistry against it.
Phase-2 dossier: the menin MBM1 pocket - the only candidate here that is already de-risked, and therefore the only one whose gate is clinical
The pocket on menin that grips the fusion protein is already blocked by an approved drug, so there is nothing left to prove about whether it can be drugged. The open question for this lab is whether that drug should be given to infants at diagnosis instead of after relapse, and that is a clinical question, not a chemistry one.
Phase-2 dossier: the non-small-molecule modalities - degraders first, and the honest limits of each
Instead of jamming a protein's pocket, you can tag the whole protein for destruction, silence the gene that makes it, or teach the immune system to find the cell. Each of these is a real option here, and each fails for a different reason that we can name in advance.
PHD3 and the CyP33 RRM: Kd 7-15 uM, and a 2023 ternary structure that settles the 2010 disagreement - H3K4me3 and the RRM CAN be held at once, and the real switch is RNA
A prolyl isomerase called CyP33 grabs the same small finger of KMT2A that reads the histone mark, and for thirteen years two papers disagreed about whether both can be held at once. A 2023 structure settles it: they can, and what actually throws the switch is RNA binding to CyP33.
PHD3 reading H3K4me3: Kd 4.3 uM for the PHD3-bromo cassette, 30 uM for PHD3 alone, two co-crystals - and the whole module is thrown away by the breakpoint
Normal KMT2A can read the very histone mark it writes, through a small finger near its middle, and the neighbouring bromodomain makes that reading about seven times better. The leukaemic fusion is cut before the finger and cannot read anything.
Prevention by exposure avoidance: verdict NOT FEASIBLE, and the transplacental-exposure literature is much weaker than its prominence suggests
There is a well-known idea that things a mother eats or is exposed to during pregnancy cause infant leukaemia. The evidence for it is thin: the studies are small, the risk estimates are wildly imprecise, and the best-conducted one mostly did not find the effect. No modifiable exposure has been established, and this is not a realistic prevention route.
PSIP1-FASTKD2 declined: the bait is a mitochondrial matrix and nucleoid protein with two experimental localisations, the prey is nucleoplasmic LEDGF, and the row is one spoke-expanded BioID2 proximity in K562
FASTKD2 lives inside mitochondria, where it helps build the mitochondrial ribosome. LEDGF lives in the nucleus on chromatin. A proximity screen put them together in one row, and we are declining it, because two experiments place FASTKD2 in a compartment that a nuclear protein cannot be in.
PSIP1-FOXK2: a single TAP row from the same TF atlas as the RBPJ decline - kept as a lead rather than declined, because FOXK2 is a chromatin-recruiting forkhead expressed in lymphoid cells and LEDGF is a chromatin-recruiting tether
FOXK2 is a DNA-binding protein that brings other complexes onto chromatin, and LEDGF does the same job for the MLL machinery. They came up together in one big survey, which is thin evidence for two proteins that do similar things.
PSIP1-GDF5 is a single DSSO crosslink in isolated U2OS nuclei with no bait on either side; GDF5 is a secreted cartilage growth factor, and the source paper itself attributes 30% of its cross-compartment links to contamination during nuclei isolation - declined
A database lists GDF5, a growth factor cells secrete to build cartilage and bone, as a partner of the nuclear protein LEDGF/PSIP1. The single supporting measurement is a chemical crosslink found inside isolated cell nuclei, and the paper that produced it says outright that some of its cross-compartment links come from contamination while the nuclei were being purified.
PSIP1-IGHMBP2 is a single DSSO crosslink in one detergent fraction of U2OS nuclei with no bait; unlike the other Fasci rows the compartments match, so this is declined on evidence depth alone - and IGHMBP2 is a fourth ribosome-biogenesis-adjacent protein on this list
A database lists IGHMBP2, a nuclear and cytoplasmic helicase that unwinds DNA and RNA, as a partner of the nuclear protein LEDGF/PSIP1. Both proteins really are in the nucleus, so the usual objection does not apply; the problem is that the whole claim rests on one chemical crosslink seen in one preparation.
PSIP1-PPP1R13B (ASPP1) is a single DSSO crosslink in unfractionated U2OS nuclei with no bait; both proteins are nuclear and both are stress-response regulators, so this is declined on evidence depth alone and is the most biologically coherent row in the beat
A database lists ASPP1, a protein that helps p53 decide whether a stressed cell should die, as a partner of the nuclear protein LEDGF/PSIP1. Both are nuclear stress-response proteins so the pairing is not far-fetched, but the whole claim is one chemical crosslink seen in one preparation and nobody has ever tested it another way.
PSIP1-RBPJ - DECLINED. A 214-purification transcription-factor TAP atlas that this beat's own atlas heuristic failed to flag, and the Notch connection is the reason to say so out loud
The Notch signalling DNA-binding protein RBPJ turned up with LEDGF in a big survey of what sticks to transcription factors. Notch matters enormously in leukaemia, which is exactly why this weak hit needs saying plainly rather than quietly promoting.
PSIP1-SDC1 - DECLINED as an interaction, kept for the splicing thread: nuclear syndecan-1 co-IP in a mesothelioma line recovers LEDGF among an RNA-splicing-enriched network
Syndecan-1 is normally a cell-surface molecule, but in mesothelioma cells some of it travels to the nucleus. When researchers pulled it down there, LEDGF came with it, inside a network full of RNA-splicing proteins.
PSIP1-TMEM214 is a DSSO crosslink in two preparations of isolated U2OS nuclei with no bait on either side; TMEM214 is an ER-membrane mediator of CASP4-dependent ER-stress apoptosis - declined
A database lists TMEM214, a protein embedded in the membrane of the endoplasmic reticulum, as a partner of the nuclear protein LEDGF/PSIP1. The supporting measurement is a chemical crosslink inside isolated nuclei, in a paper whose authors say part of their nucleus-to-ER signal comes from contamination during nuclei purification.
PSIP1-UTP3 is one spoke-expanded BioID2 proximity row in K562 with UTP3 as bait and PSIP1 as prey at miscore 0.27; PSIP1 is recovered by 9 of the 50 baits in that screen and only 2 of the 9 are nucleolar, which is the panel's own base rate - declined
A database lists UTP3, a nucleolus protein that helps assemble ribosomes, as a partner of the nuclear protein LEDGF/PSIP1. In the screen that produced it, fifty different bait proteins were tagged and nine of them picked up LEDGF, only two of those nine being nucleolus proteins - the same proportion as in the bait panel overall.
Raising HEXIM1 to brake the fusion's output without touching the fusion - one supporting experiment, one chemical series, and no therapeutic window
Instead of attacking the fusion protein, one could try to put more of the brake protein into the cell so the elongation enzyme is held back. It has been tried once in a KMT2A-rearranged leukaemia cell and it slowed growth - but every dividing cell in the body needs the same enzyme, so there is no obvious safe dose.
Repurposing approved IMPDH inhibitors (mycophenolate, mizoribine, ribavirin) in KMT2A-rearranged leukaemia - cheap, mechanistically attractive, and asking an immunosuppressant of a neutropenic infant
The drugs that block IMPDH are already approved - they are the anti-rejection pills given after a transplant - so testing them in leukaemia would be fast and cheap. The catch is exactly what they were approved for: they switch off part of the immune system, and the patients here are babies whose immune systems are already flattened by chemotherapy.
RGPD8 is the top-ranked MLL-AF4-specific prey on 3 spectra, and 96.5% of its tryptic peptides are shared with seven near-identical paralogues — only 4 peptides in the whole protein could ever have discriminated it — declined
RGPD8 is the strongest of the proteins said to bind only the MLL-AF4 fusion. It belongs to a family of eight almost-identical human proteins created by a duplicated stretch of chromosome 2, and almost every fragment a mass spectrometer would see from it is also a fragment of one of its siblings. With only three detections, we cannot say it was RGPD8 at all.
RNA-targeting the fusion transcript: verdict THE LEAST HOPELESS of the three fusion-level routes, blocked by delivery rather than by biology
The fusion makes a hybrid message that exists in no healthy cell, so a drug that destroys just that message would be as specific as anything in medicine. It works in the dish and in mice. The reason it is not a treatment is that we cannot reliably get such drugs into leukaemia cells in the body.
RPA2 is the only compartment-compatible candidate of the six and the corpus already has a KMT2A resection thread it would fit — but it is an abundant ssDNA-binding subunit on 6 spectra in a benzonase-treated lysate with no RPA1 or RPA3 co-recovered — declined, thread flagged
RPA2 coats single-stranded DNA during repair, and it is genuinely in the nucleus, unlike most of the other candidates. But its two partner subunits, which it is always found with, were not pulled down alongside it — and the DNA it normally sits on was digested away before the experiment. Six detections is not enough, so we decline it while flagging that the repair question it points at is a real one this graph is already asking.
SCF(Skp2) grips the retained fragment and the fusion escapes it - the clearest documented non-menin contact on that region
A disposal machine called SCF-Skp2 normally grabs the front end of KMT2A and destroys it once per cell cycle. Leukaemia fusions keep that front end but are not destroyed, so something about the grip has changed - and that grip is on the exact piece everyone says only menin can hold.
Taspase1 and its two KMT2A cleavage sites: an enzyme-substrate interface with four apo structures, one co-crystal with a small molecule, and no structure with MLL at all
A protease called taspase1 cuts normal KMT2A in two places, and that cut controls how long the protein lasts. The fusion protein is truncated long before either cut site, so taspase1 never touches it - and no structure of the enzyme holding its KMT2A substrate has ever been solved.
The AFF-family N-terminus on cyclin T1: Kd 10 nM, a 2.94 A crystal, and every number in it measured on AFF4 rather than on AFF1
The scaffold protein of the super elongation complex grips the CDK9 kinase through its first seventy amino acids, and the grip is very tight. The catch for this project is that the crystal and the affinity are for AFF4, the sister protein, not for AFF1, the one inside the infant leukaemia fusion.
The DOT1L–ubiquitin and DOT1L–acidic-patch interfaces: two non-catalytic surfaces that MV4;11 cells cannot do without
DOT1L grips the nucleosome in two places that have nothing to do with the chemistry it performs: the ubiquitin tag on H2B, and a negatively charged pocket on the nucleosome surface. Breaking either grip stops a KMT2A-fusion leukaemia cell growing, which makes them targets that are not the drug site that already failed in the clinic.
The ELLbow: AFF4 301-351 wrapped over the ELL2 occludin domain at 2.0 A, Kd 86 nM - and no structure of ELL itself, the partner in t(11;19)
The scaffold of the elongation complex folds a fifty-residue stretch into an elbow shape that lies across a curved surface on ELL2. The protein fused to KMT2A in t(11;19) leukaemia is ELL, not ELL2 - and ELL has never been solved with any scaffold at all.
The ENL AHD to AFF1 contact: asserted throughout this graph, structurally unsolved, and represented only by an AF9 chimera measured at KD 0.17 nM
The floppy tail of ENL is said to hold the AF4 scaffold, and much of what this field believes about t(11;19) leukaemia rests on it. No structure of that pair exists - all twenty-six ENL structures in the world are of a different domain.
The fusion junction itself as a small-molecule target: a candid verdict
The seam where KMT2A is glued to its partner is the one piece of the leukemia protein that exists nowhere else in the body, so it looks like the perfect drug target. Unfortunately the seam sits inside a floppy, shapeless stretch of protein with no pocket for a drug to grip, and its exact sequence differs from patient to patient.
The HOX-PBX interface, and its one peptide: HXR9/HTL001 has been tried in eight tumour types and in KMT2A-rearranged leukaemia exactly once
The one drug-like thing ever aimed at this biology is a short peptide that copies the piece of HOXA9 that plugs into its partner PBX, so it competes the two apart. It kills cancer cells in melanoma, ovarian, breast, mesothelioma, oral and B-cell tumours - and in KMT2A-fusion leukemia it has been tested in a single published paper. That breadth is the reason to be interested and also the reason to be careful.
The KMT2A transactivation domain on the CBP KIX domain: a cooperative allosteric site at 2842-2858, solved three times, and thrown away by every breakpoint
Normal KMT2A carries a short activation segment that docks on one face of the coactivator CBP, and docking there makes CBP grip a second transcription factor on the opposite face more tightly. The leukaemic fusion is cut long before that segment, so it loses a cooperative switch nobody in this field discusses.
The LEDGF IBD–KMT2A grip: the second half of the clamp, and nobody is drugging it
Besides menin, a second protein called LEDGF grabs a nearby piece of KMT2A using a small clamp-shaped domain. Two phenylalanines on KMT2A do most of the gripping, and a peptide that blocks this grip already slows leukemia cells down.
The LEDGF PWWP–H3K36me nucleosome contact: how the fusion learns its address
The far end of LEDGF is a small reader module that grips nucleosomes carrying a specific methyl tag, H3K36 methylation. That grip is what tells the whole menin-KMT2A machine which stretches of DNA to sit on, and there are six good cryo-EM pictures of it.
The LSD1–GFI1 SNAG interface — a second worked example of drugging an undruggable repressor through its partner
GFI1 is a switch-off protein that leukemia cells use to stay immature, and it has no pocket a drug can sit in. LSD1 inhibitors seem to work in this leukemia by prising LSD1 off GFI1 rather than by blocking LSD1's chemistry — which is the same trick that menin inhibitors pull on the KMT2A fusion.
The menin acidic shelf, Glu366-Asp370-Val371: the counter-charge that MBM2 binds, resolved in 3U88 after all - and the one patch where the drug pocket, the SvID and the gamma-H2A.X motif all touch
There is a small negatively charged ledge on menin, just outside the pocket the drugs sit in. Two arginines from KMT2A land on it, and the same ledge is the first part of two other menin regions that other papers describe as separate. It is the single place where four different stories about menin overlap.
The menin side of the LEDGF clamp: residues 91-105 and 209-212, on the opposite face of menin from the drug pocket and 26-38 A from every other surface in this beat
LEDGF, the protein that anchors the KMT2A fusion to chromatin, grips menin on a patch made of two short stretches near menin's start. That patch is on the far side of menin from where the drugs bind, which is why a menin drug is not expected to break the LEDGF half of the clamp directly.
The menin small-molecule pocket as a coordinated surface: four clinical-stage compounds, 21-26 contact residues each, and bleximenib sitting on a different part of it from the other three
Four menin drugs have been crystallised in menin's central cavity, and this entry lists, for each one, exactly which menin amino acids it touches. Three of the four reach into a part of menin that the natural KMT2A peptide barely touches - and the fourth, bleximenib, stops short of it, which makes the four compounds less interchangeable than the phrase 'menin inhibitor' suggests.
The menin-gamma-H2A.X motif, menin Val371-Gln400: necessary, sufficient as a peptide, and the nine residues at its centre have never been seen in any of the 69 menin structures
Menin grabs a damage-marked histone, gamma-H2A.X, using a thirty-residue stretch of itself, and that grab is what lets cells repair broken DNA accurately. The middle of that stretch is floppy and has never been seen in any menin crystal or cryo-EM structure, so nobody knows what it looks like when it is doing its job.
The menin–KMT2A grip (MBM1), the interface the drugs actually block
The first dozen amino acids of KMT2A plug into a pocket on menin like a key in a lock. Every approved and experimental menin drug works by filling that pocket so the key cannot go in.
The MLL-protected Hoxa9 CpG sites bind the CXXC domain 4- to 9-fold tighter than the unprotected ones, in one ITC series at fixed everything else
In one clean experiment, the fusion's CpG-gripper was titrated against five short DNAs taken from the Hoxa9 gene. The pieces that MLL actually keeps unmethylated in living cells bound several times more tightly than the pieces it does not, which is the only published hint that the module can tell one CpG site from another.
The MT/BCB domain, Q03164 1147-1244: the CXXC fold plus a 41-residue basic tail nobody has measured on its own
The famous CpG-gripping module of the fusion is usually drawn as a 57-residue zinc finger. The original 2002 work used a longer 98-residue piece that adds a tail loaded with positively charged residues, and nobody has ever measured what that tail contributes to gripping DNA.
The retained AF9 surface in residue detail: one disordered 70-mer, three DOT1L sites, two of them high-affinity and simultaneously occupiable
The only part of AF9 the cancer protein keeps is a floppy 70-residue tail that folds into shape only when something grabs it. DOT1L grabs it in three different places, and two of those are strong enough that two copies of the tail can hold one DOT1L at once.
The SvID, menin 360-445: a truncation boundary the field reads as a domain, whose functional residues D418 and W436 are 10-20 A from every menin drug - but whose first ten residues ARE the drug pocket
Menin hands the repressive enzyme SUV39H1 to DNA using a stretch of itself called the SvID. This graph has been using that stretch as proof that a menin drug cannot disturb the enzyme - but the first ten residues of the SvID are part of the drug pocket, so the proof needs to be about individual residues, not about the whole stretch.
There is no selective PRC1 catalytic inhibitor - the entire chemistry is one 2013 tool compound and its analogues, and it hits both paralogues by design
Several entries in this graph propose blocking the Polycomb enzyme that silences genes. It is worth knowing that no good drug for that exists: there is one laboratory tool from 2013, it hits both versions of the enzyme at once, and it was described as a way to study DNA repair rather than as a cancer drug.
USP1: an inhibitor, activity in B-ALL, and zero papers in the disease we care about
The one part of the repair module our scan implicated that a drug can actually reach is USP1. Drugs against it exist, and killing leukaemia cells with them has been published. Nobody has ever tried it in the leukaemia this project is about.
Volasertib against PLK1: a differential window claimed in infant-type cells, on a target every dividing cell needs
Volasertib blocks PLK1 and did little for adults with this leukaemia. One group reports that infant-type cells respond much better, which would make it a drug for the wrong age group tested in the wrong one.
WDR5 and RBBP5: the second arm of the WRAD hub, RBBP5 V375-D376-V377 into the same propeller face, co-crystallised with the MLL1 WIN peptide in one structure
WDR5 holds two things at once: the KMT2A enzyme on one side and a partner called RBBP5 on the other, and a single crystal shows both grips together. The fusion protein is not in that picture at all.
WDR5 WIN site and KMT2A Arg3765: measured at 1.7 uM, three co-crystals, and absent from every leukaemic fusion - plus the MLL3 peptide that binds it 51-fold tighter
Normal KMT2A docks onto WDR5 by pushing a single arginine into a hole on WDR5's surface, and that docking is what builds its enzyme. Every leukaemia fusion is cut about 2,400 amino acids before that arginine, so the fusion never docks at all - and the same hole binds a related protein, MLL3, fifty times more tightly.
Zoledronic acid against the osteoclast: a bone-marrow-niche target, but not tested in infant KMT2A-rearranged disease
Leukaemia makes bone-eating cells more active, and those cells in turn release factors that help the leukaemia grow. Blocking them with a bisphosphonate extended survival in mice and reversed the bone damage, and three children tolerated the drug safely during chemotherapy - but none of this work used infant KMT2A-rearranged leukaemia.
Essentiality, normal-tissue expression and human genetic constraint for each candidate.
ABI1 (Abl interactor 1 / e3B1) — WAVE-complex cytoskeletal adaptor, KMT2A partner, and an ENL-binding protein
ABI1 is a connector protein in the machinery that builds and steers the cell's actin skeleton. It is a rare KMT2A fusion partner, and unusually it also sticks to ENL — one of the common KMT2A partners — which could let it reach the gene-reading machinery indirectly.
ABL1 — domain architecture and residue coordinates (with the 1a/1b numbering trap)
ABL1 is an enzyme that adds phosphate tags to other proteins, and it normally keeps itself switched off by folding a greasy tail back against its own working face. The leukaemia fusion cuts that tail off, which is why the enzyme never switches off again, and because two rival ways of counting ABL1's residues differ by exactly nineteen, every number here has to say which one it uses.
ACIN1 (Acinus) - a splicing/apoptosis protein whose only domain is the 35-residue module the leukaemic fusion keeps, and whose entire localization record is uncoded
ACIN1 normally works on messenger RNA splicing and helps condense the DNA of a dying cell. A leukaemia-causing accident keeps only its small DNA-gripping tip and throws away everything else it does - so the leukaemia is not caused by ACIN1 doing its job in the wrong place, but by that tip being used as a postal address for a different protein.
ACSL4 - a CRISPR-screen-derived, functionally validated KMT2A-selective vulnerability in myeloid disease, and a gene whose loss normally protects cells from ferroptosis
ACSL4 activates long polyunsaturated fats so the cell can use them, and KMT2A-rearranged myeloid leukaemia needs it much more than other leukaemias do. Turning it off slows those cells and delays the disease in mice - and feeding the cells the fats back rescues them, which is how the authors showed the fat handling is really the reason.
AFDN / AF6 — domain architecture and residue coordinates
AFDN is a cell-junction protein that becomes a KMT2A fusion partner, and unusually for a partner it normally lives outside the nucleus. This entry pins down where its parts sit, and the surprise is that the small module at its very front end — not anything larger — is the piece that makes the fusion cancerous.
AFF1 / AF4 — domain architecture and residue coordinates
AFF1 is the partner protein most often stuck onto the broken KMT2A gene in infant leukemia, and this entry pins down where each of its working parts sits along the protein chain. Most of AFF1 is a floppy, shapeless string, so several of its "domains" are conventions drawn on a diagram rather than things anyone has measured.
AFF1 / AF4 — the 3' half of the fusion, and a target with almost no structural footing
AFF1 is the protein whose tail end gets welded onto KMT2A in the commonest infant leukemia fusion. Despite being half the cancer protein, almost nothing is known about its shape: the entire Protein Data Bank contains one small piece of it.
AFF1::KMT2A (AF4-MLL), the der(4) product: what it is made of, residue by residue
When chromosomes 4 and 11 swap arms, two hybrid genes are made, not one. This entry is the second one: a protein that throws away the front of KMT2A - including the piece the menin drug grabs - and keeps everything behind it, including the enzyme end.
AFF2 (FMR2) — the SEC-L2 scaffold, and the AFF-family member with no dependency anywhere
AFF2 is the fourth member of the family of scaffold proteins that build the cell's transcription-speeding machine. Unlike its relatives it is switched off in most cells and no cell line tested needs it, so it is the family member that tells you the family slot can be empty.
AFF3 (LAF4) — the AFF-family scaffold of SEC-L3, and a KMT2A fusion partner in its own right
AFF3 is one of four closely related scaffold proteins, any of which can build a version of the elongation machine the KMT2A fusion hijacks. AFF3 builds a different version from the one AFF1 builds, and those versions switch on different sets of genes.
AFF4 — the textbook SEC scaffold that KMT2A::AFF1 cells need LESS than other leukemias
AFF4 is the long, floppy protein usually drawn at the centre of the super elongation complex, holding the other parts together. But when you delete it, KMT2A-rearranged leukemia cells are slightly better off than other leukemia cells are, which is the opposite of what the textbook picture predicts.
ALK (anaplastic lymphoma kinase) domain architecture
ALK is the 3' partner of NPM1::ALK and the target of crizotinib, alectinib and lorlatinib. This entry lists where each of its functional parts sits along the protein, measured as residue numbers, so the fusion diagrams can draw which parts survive the break and which are thrown away.
AP2B1 (beta-2-adaptin) - the only one of AFF4's four two-hybrid partners that is not a hub, and the only one with an echo in this graph's endocytosis biology
AP2B1 is part of the machinery that pulls cargo into cells from the cell surface. It was reported as a partner of AFF4 in one large automated yeast screen, and unlike the other three proteins from that screen it does not stick to everything, which makes it the least dismissible of them.
ARID1A (BAF250A) - required in the MLL-AF9 screen, present at the Myc enhancers, and flat in every selectivity measure
ARID1A is the large subunit that defines the canonical flavour of SWI/SNF and is one of the most frequently mutated genes in all of cancer. MLL-rearranged mouse leukaemia cells need it, and it sits on the same enhancers the mechanism runs through - but there is no sign that KMT2A-rearranged cells need it more than any other leukaemia does.
ARID2 (BAF200) - the PBAF subunit whose loss ACCELERATES MLL-AF9 leukaemia, and which is nonetheless required to maintain it
ARID2 marks out the PBAF flavour of the SWI/SNF machine. In mice, deleting it early makes MLL-AF9 leukaemia worse and shortens survival - but deleting it once the leukaemia exists slows it down. The direction of the effect depends entirely on when you remove it.
ASH2L — the WRAD subunit that has DPY30's essentiality profile and none of its co-dependency signal
ASH2L is the protein DPY30 clamps onto inside the shared core of the H3K4 machinery. It is needed about as often as DPY30 across cancer cells, yet unlike DPY30 its requirement does not track menin at all — and nobody knows why.
ATXN7L3 — the shared adaptor that actually sets H2Bub levels, and a KMT2A-selectivity hit nobody in this field has looked at
ATXN7L3 is not an enzyme; it is the helper protein three different eraser enzymes have to borrow before any of them can strip ubiquitin off histone H2B. Removing it raises the tag far more than removing the enzyme everyone names, and in cancer cell-line data it is one of the more KMT2A-selective genes in the genome.
BCL2 — the one co-dependency in this beat that already has an approved drug pointed at it
BCL2 is the brake on the cell's self-destruct switch, and KMT2A::AFF1 leukemias turn it up, which is why they die when you block it. Venetoclax already blocks it and is approved, so this is the shortest path from a dependency measurement in this project to a medicine a child could actually be given.
BCOR — the other side of the CBX8 conflict, and a target you would be inhibiting in the direction of disease
BCOR is a large scaffolding protein that grips the tail end of AF9 and also assembles a non-standard Polycomb complex. One paper says that grip is the thing leukemia cells cannot do without; large-scale deletion data says BCOR is not needed at all. Awkwardly, in other blood cancers BCOR is a gene that gets broken — so a drug that switches it off would be pushing in the same direction as the disease.
BCORL1 — the BCOR paralogue that is in the only crystal structure of the PRC1.1 core, and is the one PRC1.1 subunit nobody has tested in KMT2A-rearranged leukaemia
BCORL1 is a near-twin of BCOR, and one of the two interchangeable arms that hold the PRC1.1 gene-silencing machine together. It is the version that appears in the only solved picture of that machine's core, yet no one has ever asked what it does in a KMT2A-rearranged leukaemia.
BCR — domain architecture and residue coordinates
BCR is a large signalling protein whose very first stretch is a sticky coil that makes copies of itself clip together. In chronic myeloid leukaemia that coil is welded onto the ABL1 enzyme, and because the coil forces four copies together the enzyme is switched permanently on.
BICRA (GLTSCR1) - the ncBAF scaffold, zero PubMed records in KMT2A leukaemia, and a lineage-independent dependency in KMT2A-rearranged lines
BICRA is the scaffold subunit that, together with BRD9, defines the third and smallest flavour of the SWI/SNF machine. No paper has ever named it in MLL or KMT2A leukaemia, yet in a dependency screen KMT2A-rearranged lines need it in both lineages.
BOD1L (BOD1L1) — the top SETD1A co-dependency, and the only candidate in this cycle tested by CRISPR tiling in a KMT2A::AFF1 line
BOD1L is the clamp that holds the SETD1A enzyme onto DNA. Cutting the gene in leukemia cells kills them, but only if you cut the one small part of it that does the clamping, which is why the standard genome-wide screens missed it entirely.
BRD4 — the best-drugged protein in this beat, and the one whose clinical result is already in and disappointing
BRD4 reads acetyl marks on chromatin and was one of the great hopes for KMT2A-fusion leukemia after two 2011 Nature papers. Drugs that block it were built, taken into patients with leukemia, and largely did not work.
BRD9 — dispensable in the average cell, rank 39 of 17,728 in KMT2A-rearranged lines, and with an on-target proof most epigenetic targets never get
BRD9 is a reader protein inside one flavour of the SWI/SNF chromatin-remodelling machine. Most cells do not need it at all, but KMT2A-rearranged leukaemia cells do, and there are inhibitors plus an unusually rigorous proof that those inhibitors work by hitting BRD9 and nothing else.
BRIP1 — downstream effector; rank 22 of 8,269 for KMT2A-selective dependency, and no KMT2A literature at all
BRIP1 is part of a cellular repair crew that fixes DNA when the copying machinery gets stuck. A screen run by this project found that leukaemia cells carrying the KMT2A fusion need it more than other leukaemia cells do — and nobody has ever published a paper on this gene in KMT2A leukaemia, so the finding is genuinely new and genuinely unconfirmed.
BUD13 - a RES-complex splicing factor 1.66 Mb from KMT2A on 11q23.3, whose one chromatin paper puts it in the soluble fraction
BUD13 is part of the machinery that cuts introns out of new RNA, including the rare minor spliceosome that handles a small special class of introns. It was the protein doing the tagging in the experiment that put both KMT2A and menin on the interactome reading list, and its gene happens to sit less than two million bases from KMT2A on the same chromosome arm.
CBFA2T3 (ETO2 / MTG16) domain architecture
CBFA2T3 is the 5' partner of the CBFA2T3::GLIS2 fusion of infant acute megakaryoblastic leukemia. This entry lists where each of its functional parts sits along the protein, measured as residue numbers, so the fusion diagrams can draw which parts survive the break and which are thrown away.
CBFB — domain architecture and residue coordinates
CBFB is a small helper protein that never touches DNA itself; it clamps onto the DNA-gripping module of RUNX1 and makes that grip much tighter. In one type of acute myeloid leukaemia its last seventeen residues are swapped for a long muscle-protein rod, and the resulting molecule drags RUNX1 away from the genes it should be switching on.
CBX8 — the one place in this interactome where two good papers flatly disagree, and DepMap takes a side
CBX8 is a Polycomb protein that reads a silencing mark on chromatin and also grips the tail end of AF9. One influential paper says leukemia cells cannot survive without that grip; another says breaking it changes nothing. Large-scale gene-deletion data sides with the second.
CCND3 (cyclin D3) — the scan's rank-4 hit, and the only one of the four with a drug class already in children
Cyclin D3 is a protein that pushes a cell from resting into dividing, working with the enzymes CDK4 and CDK6 that existing breast-cancer drugs already block. A computer scan run by this project found leukemia cells carrying the KMT2A fusion need cyclin D3 more than other leukemia cells do, and unlike the other three genes the scan found, there is real published work behind it.
CCNT1 / cyclin T1 — the regulatory half of P-TEFb, and the one part of the elongation core with any dependency variance
Cyclin T1 is the partner that switches on CDK9, the kinase that tells RNA polymerase to start moving. The KMT2A fusion pulls this pair in through its own scaffold protein, and unlike CDK9 itself, cyclin T1 is not needed by every cell — which makes it the more interesting half.
CCNT2 / cyclin T2 - expressed in every KMT2A-rearranged line and needed by none of them, so the cyclin T split is real rather than an expression artefact
Cyclin T2 is the alternative partner that can switch on CDK9 in place of cyclin T1. This graph has learned that a paralogue difference only means something if the quiet paralogue is actually being made, so the check was run: cyclin T2 is present in all seven KMT2A-rearranged cell lines tested and the cells still do not need it.
CDC73 / parafibromin — the PAF1C subunit that was genetically tested, and a second tumour suppressor required by the fusion
CDC73 is one of the four proteins in the PAF1 machine, and it is the only one that has actually been deleted in mice to prove the machine is needed for KMT2A-fusion leukemia. Like menin, it is also a tumour suppressor whose loss causes a parathyroid disease in humans, which is a pattern worth noticing.
CDK12 - a different kinase with a different cyclin, and in KMT2A-rearranged lines it is an anti-dependency
CDK12 is a cousin of CDK9 that also phosphorylates RNA polymerase, but it uses a different partner protein and controls a different set of genes, mostly DNA-repair genes. In KMT2A-rearranged cells, removing it is if anything slightly less harmful than in other leukaemias, so the CDK9 story does not transfer to it.
CDK13 - rank 171 of 17,728, equally deep in both lineage halves, and essentially unstudied in leukaemia
CDK13 is another kinase that helps RNA polymerase keep moving. In the public CRISPR screens, KMT2A-rearranged leukaemia cells need it noticeably more than other leukaemia cells do, and unusually the effect is the same size in the lymphoid and myeloid halves - yet almost nobody has studied it in leukaemia at all.
CDK9 - the kinase the whole SEC story runs through, which this graph held only as half of a two-protein summary entry
CDK9 is the enzyme that gives a stalled RNA polymerase permission to move, and every account of how KMT2A::AFF1 drives its target genes ends at CDK9. Until now this graph held it only inside a joint entry with its cyclin partner, so the best-evidenced interactor in the whole set had no page of its own.
CEBPA — ranks 25th on the KMT2A scan and sits at the median of random genes
C/EBP-alpha is another master switch of myeloid identity. On one way of measuring it looked like the 25th most important gene for this leukaemia. Split the cell lines by blood-cell type and the effect is entirely in the myeloid ones, and against random genes it scores no better than average.
CHUK (IKK-alpha) — the only kinase that sits in BOTH NF-kB arms, which makes it the one subunit whose inhibition has an ambiguous sign in this disease
IKK-alpha is a kinase that works in two different versions of the NF-kB switch at once. One of those versions keeps the leukaemia going and the other one shuts it down, so blocking IKK-alpha does two opposite things in the same cell and nobody has measured which wins.
CREBBP (CBP) — binds the KMT2A transactivation domain, which no fusion keeps
CBP is a coactivator that normal KMT2A grabs with a short stretch near its middle in order to switch genes on. Leukemia fusions cut KMT2A long before that stretch, so the fusion has to turn genes on some other way.
CSNK2A2 - CK2's other catalytic subunit, expressed in every KMT2A-rearranged line and, if anything, slightly favourable to lose
CK2 is an always-on enzyme with two interchangeable catalytic halves. This graph already covers the first one, CSNK2A1, which touches KMT2A biology three separate ways; the second one is present in these leukaemia cells but removing it does not hurt them and may help them slightly.
CTNNB1 (beta-catenin) — the Wnt link inside DotCom, and the one piece of KMT2A biology that is not chromatin
Beta-catenin is the main switch of the Wnt signalling pathway, and it was found sitting inside the complex that places the H3K79 mark the KMT2A fusion depends on. That is the only published thread connecting the fusion's chromatin machinery to a classical signalling pathway.
CTR9 — a PAF1C subunit the field assumes is a fusion partner, with no paper that tested it
CTR9 is one of the four proteins in the PAF1 machine that the KMT2A fusion latches onto, and it is often listed as a fusion partner. We could not find a single study that tested CTR9 itself in KMT2A-fusion leukemia — the evidence is entirely about the complex as a whole.
CXXC1 (CFP1) - a CpG-reading COMPASS subunit that binds MLL1 and recruits it to HoxA7, and whose CXXC domain cannot substitute for KMT2A's in a fusion
CFP1 recognises unmethylated CpG islands - the same kind of DNA address the KMT2A fusion uses - and it physically travels with MLL1 to switch on HOX genes. But when researchers swapped CFP1's CpG-reading domain into the leukaemia fusion in place of KMT2A's own, the fusion stopped causing leukaemia, even though it still bound DNA and still reached the right gene.
CXXC4 (IDAX) - TET2's CXXC domain, sitting on a separate gene after an evolutionary inversion, and annotated cytoplasmic only by similarity
CXXC4 is a tiny protein that is essentially a spare address-reader: it is the CG-reading module that TET2 used to carry, split off onto its own gene early in evolution, and it still delivers TET2 to unmethylated CG islands. Whether it is even in the nucleus of a human cell has never actually been measured - the database says cytoplasm, by guesswork from other species.
CXXC5 (RINF) - the one CXXC reader measurably down-regulated in KMT2A-rearranged AML, which is the only direction-of-change any family member has in this disease
CXXC5 reads unmethylated CG islands and escorts the demethylating enzyme TET2 to them. Its gene sits in the piece of chromosome 5 that is deleted in some myeloid cancers, and - the reason it matters here - its message is specifically turned down in leukaemias carrying a KMT2A rearrangement.
DDX10 domain architecture (UniProt Q13206)
DDX10 is the partner protein in the NUP98::DDX10 fusion. This entry records where each of its parts sits along the protein, in residue numbers, so the fusion diagram can draw what the break keeps and what it throws away.
DDX6 - the only protein ever proposed to do for a cell what Tat does for the virus, on the strength of one 2016 paper
A virus can pull the elongation enzyme out of its storage particle. Only one protein our own cells make has ever been proposed to do the same thing: an RNA helicase called DDX6, which was reported to bind the storage RNA and hand the enzyme over to the AF4 scaffold. It has been shown once, in one laboratory, in a low-profile journal.
DEK - a chromatin architectural protein that gatekeeps H3.3 and boosts HP1alpha, is a fusion partner in its own right, and is the substrate behind this graph's third CK2 route
DEK is an abundant nuclear protein that holds chromatin in shape: it helps heterochromatin stay compact and it controls which chaperones are allowed to deposit a particular histone variant. It is also half of a fusion protein that causes its own kind of leukaemia, and switching off a kinase that modifies it kills leukaemia stem cells while sparing normal ones.
DNMT1 - the antagonist: its CXXC domain reads unmethylated CpG in order to STOP methylation, and it is the only heterologous CXXC that runs an MLL-AF9 leukaemia
DNMT1 copies the chemical tags on DNA after it is duplicated, so it is the enzyme that creates the tagged CG letters the KMT2A fusion cannot read. Oddly, it carries the same untagged-CG reader as the fusion - and uses it as a brake on itself. It is also the only borrowed version of that reader that lets an MLL fusion still cause leukaemia.
DOT1L — the target where excellent chemistry met an indirect interaction
DOT1L is the enzyme that puts a specific methyl mark on histones and keeps KMT2A-fusion leukemia genes switched on. Someone made an 80-picomolar inhibitor for it, and the drug still underperformed in patients — which is the most useful failure in this whole field.
DPF2 (BAF45D) - a cBAF-specific reader required in MLL-AF9 cells, with one PubMed record in this disease and no selectivity signal
DPF2 is a small reader subunit that sits only in the canonical flavour of the SWI/SNF machine. Removing it stops MLL-rearranged leukaemia cells growing, but the same is true of leukaemia cells in general, and only one paper in PubMed has ever looked at it in this disease.
DPY30 — the smallest COMPASS subunit, and the only one that behaves like a selective dependency
DPY30 is a tiny 99-amino-acid protein shared by all six of the cell's H3K4 methyltransferase complexes, including the one KMT2A belongs to. Unlike its three partners in that shared core, knocking it out does not kill every cell, and in leukemia lines its requirement tracks menin's — which makes it the only part of the shared machinery with a plausible therapeutic window.
DUX4 (double homeobox 4) domain architecture
DUX4 is the 5' partner of DUX4::IGH B-ALL. This entry lists where each of its functional parts sits along the protein, measured as residue numbers, so the fusion diagrams can draw which parts survive the break and which are thrown away.
EAF1 — the interactor with an actual subnuclear phenotype: an MLL fusion evicts it from Cajal bodies, and the bodies themselves fall apart
EAF1 normally sits with its partner ELL in small dots inside the nucleus called Cajal bodies. When an MLL-ELL fusion protein is present, EAF1 is scattered out of those dots into the general nuclear space, and in leukaemia cells carrying the fusion the dots are gone altogether.
EAF2 (U19) — the EAF paralogue whose ELL-binding domain is a mapped tumour-suppressor module, described entirely in prostate cancer
EAF2 is the near-twin of EAF1 and binds the same partner, ELL, which is itself an MLL fusion partner. Its whole functional story was worked out in prostate cancer, where putting it back into cells kills them — and the part of the protein that does the killing is exactly the part that grips ELL.
EBF3 — a single large-scale MS hit against menin, in a B-lineage transcription factor family, which is the only reason it is worth recording
EBF3 is a relative of the master switch that makes a cell become a B cell. It showed up once, in a very large fishing expedition for protein partners, as something menin might stick to. One hit in one big screen is thin evidence, and this entry says so — but the family it belongs to is the family that decides B-cell identity, which is the one thing that makes it worth writing down at all.
EHMT2 / G9a - the H3K9me1/2 writer that sets whether the KMT2A LOCUS ITSELF rearranges, and whose loss delays MLL-AF9 AML in a supplementary figure nobody can find
G9a puts a small silencing tag on histone H3, and how much of that tag sits on the KMT2A gene decides how easily that gene breaks and rearranges - the event that starts this leukaemia. In leukaemias that already exist, removing G9a slows them down and pushes the cells to mature.
ELL — rank 7 on selectivity and the deepest common-essential in the top ten, with one NMR fragment to its name
ELL is the part of the super elongation complex that actually speeds up RNA polymerase. KMT2A-rearranged leukemia cells need it a bit more than other leukemias do, but every cell needs it enormously — and despite being central to the whole complex, the only human ELL structure ever solved is one small piece floating on its own.
ELL2 — a SEC subunit that almost no cell needs, next to a paralogue almost every cell needs
ELL2 is one of two interchangeable parts that slot into the elongation machine the KMT2A fusion hijacks. Removing ELL2 barely affects any cell, while removing its twin ELL1 kills almost all of them — which tells you the slot matters and this particular part does not.
ELL2 — domain architecture and residue coordinates
ELL2 is the part of the transcription machine that the KMT2A fusion's partner grabs hold of to make genes run faster. This entry gives the residue numbers for the surface it is grabbed by, which is one of the few in this project resolved to 2.0 angstroms.
ELL3 — the third ELL paralogue, which marks enhancers and recruits SEC rather than sitting inside it
ELL3 is the third member of a family of three near-identical elongation factors, and it is the odd one out: instead of joining the machine, it sits on enhancers ahead of time and calls the machine in when a signal arrives. It is not a part of the KMT2A fusion complex.
ELOVL1 - the fatty-acid elongase that infant MLL-AF4 ALL retains from its fetal origin, and the lipid enzyme on the opposite side of the divide from ACSL4
ELOVL1 lengthens saturated fats into very long ones, and it is one of the genes infant MLL-AF4 leukaemia keeps switched on from its origin before birth. Switching it off kills those leukaemia cells, which makes altered fat handling a feature of the infant disease rather than an afterthought.
EP300 — the acetyltransferase whose paralogue split is the finding: p300 is the 108th most KMT2A-selective gene, CBP is the 16,950th
p300 is an enzyme that tags histones with acetyl groups to switch genes on. Its near-identical twin CBP was already in this graph; adding p300 shows the two behave completely differently in KMT2A-rearranged leukaemia, and only p300 looks like something worth targeting.
EPB41L3 / DAL-1 - a membrane-skeleton tumour suppressor that binds and modulates two arginine methyltransferases, and whose own promoter methylation is a clinical test
EPB41L3 is a cytoskeletal scaffold that also acts as a tumour suppressor, and it grips two enzymes that put methyl groups on proteins - turning one of them down. Its own gene is switched off by DNA methylation in many cancers, which is why a methylation test for it is used in cervical-cancer screening.
EPS15 / AF1p — domain architecture, and the one coordinate nobody has published
EPS15 normally helps cells pull receptors in from their surface, and when it is fused to KMT2A the fusion keeps almost the whole EPS15 protein, including the three 'EH' grabbing modules. The single piece that makes the fusion cancerous is a rod-shaped stretch that makes two copies stick together, and that stretch is the one part of EPS15 that no database gives residue numbers for.
ETV6 — domain architecture and residue coordinates
ETV6 is a switch-off protein that normally sticks to DNA and quietens genes. Its front end is a self-stacking module that links copies of itself into a long chain, and in leukaemia that chain-making module is welded onto other proteins, which is what makes the fusions dangerous.
EWSR1 — domain architecture and residue coordinates
EWSR1's front half is a low-complexity, repetitive stretch that behaves like a molecular glue, able to gather protein machines around it. In Ewing sarcoma that glue is welded onto a DNA-gripping protein, and the combination drags chromatin-remodelling machines onto stretches of DNA they should never occupy.
FAAP100 — FA core complex; rank 36 of 8,269 for KMT2A-selective dependency, and no KMT2A literature at all
FAAP100 is part of a cellular repair crew that fixes DNA when the copying machinery gets stuck. A screen run by this project found that leukaemia cells carrying the KMT2A fusion need it more than other leukaemia cells do — and nobody has ever published a paper on this gene in KMT2A leukaemia, so the finding is genuinely new and genuinely unconfirmed.
FACT (SPT16-SSRP1) - the chromatin arm of transcription elongation, and a clean negative result for the KMT2A::AFF1 interactome
FACT is a two-protein machine that pulls nucleosomes apart in front of RNA polymerase and puts them back behind it, so genes packed into chromatin can be read. This entry exists to record, with numbers, that despite being central to transcription it has no documented role in the KMT2A fusion machinery - which is a useful thing for a knowledge graph to state plainly rather than leave as an unexplored gap.
FANCA — FA core complex; rank 20 of 8,269 for KMT2A-selective dependency, and no KMT2A literature at all
FANCA is part of a cellular repair crew that fixes DNA when the copying machinery gets stuck. A screen run by this project found that leukaemia cells carrying the KMT2A fusion need it more than other leukaemia cells do — and nobody has ever published a paper on this gene in KMT2A leukaemia, so the finding is genuinely new and genuinely unconfirmed.
FANCD2 — the FA effector menin binds, and the weakest FA gene in this graph's own selectivity scan
FANCD2 is the protein that the Fanconi anaemia repair crew switches on by tagging it with a single ubiquitin; once tagged it goes to the damaged DNA and organises the repair. It is the only Fanconi protein reported to bind menin directly, and it is also the one Fanconi protein that this project's own dependency scan says KMT2A-rearranged leukaemia cells barely need.
FANCF — FA core complex; rank 11 of 8,269 for KMT2A-selective dependency, and no KMT2A literature at all
FANCF is part of a cellular repair crew that fixes DNA when the copying machinery gets stuck. A screen run by this project found that leukaemia cells carrying the KMT2A fusion need it more than other leukaemia cells do — and nobody has ever published a paper on this gene in KMT2A leukaemia, so the finding is genuinely new and genuinely unconfirmed.
FANCG — FA core complex; rank 27 of 8,269 for KMT2A-selective dependency, and no KMT2A literature at all
FANCG is part of a cellular repair crew that fixes DNA when the copying machinery gets stuck. A screen run by this project found that leukaemia cells carrying the KMT2A fusion need it more than other leukaemia cells do — and nobody has ever published a paper on this gene in KMT2A leukaemia, so the finding is genuinely new and genuinely unconfirmed.
FANCI — ID2 module; rank 15 of 8,269 for KMT2A-selective dependency, and no KMT2A literature at all
FANCI is part of a cellular repair crew that fixes DNA when the copying machinery gets stuck. A screen run by this project found that leukaemia cells carrying the KMT2A fusion need it more than other leukaemia cells do — and nobody has ever published a paper on this gene in KMT2A leukaemia, so the finding is genuinely new and genuinely unconfirmed.
FANCL — FA core complex; rank 26 of 8,269 for KMT2A-selective dependency, and no KMT2A literature at all
FANCL is part of a cellular repair crew that fixes DNA when the copying machinery gets stuck. A screen run by this project found that leukaemia cells carrying the KMT2A fusion need it more than other leukaemia cells do — and nobody has ever published a paper on this gene in KMT2A leukaemia, so the finding is genuinely new and genuinely unconfirmed.
FBXL19 - a CpG-island reader that delivers CDK-Mediator and RNF20 to developmental promoters, and whose experimental localisation is mostly cytoplasm
FBXL19 reads unmethylated CG islands and brings in two machines the KMT2A fusion also depends on: the Mediator complex that primes genes for switching on, and the enzyme that tags histone H2B. It is also, awkwardly, mostly found in the cytoplasm when anyone has measured it.
FLI1 — domain architecture and residue coordinates
FLI1 is a DNA-gripping transcription factor with two working parts: a self-stacking module near the front and a DNA-reading module near the back. In Ewing sarcoma the self-stacking module is discarded and only the DNA-reading module is kept, welded onto a sticky region from EWSR1.
FLT3 / CD135 — the strongest non-tether co-dependency, and a target the field may have written off on a badly executed trial
FLT3 is a growth-signal receptor on the surface of young blood cells, and KMT2A-fusion leukemias switch it up to very high levels. Cells that need menin also need FLT3 more than almost any other gene we tested, which sits awkwardly with a large infant trial of a FLT3 drug that failed.
GAS7 — growth-arrest-specific F-BAR protein, the second pillar of the forced-oligomerisation model
GAS7 is a protein that neurons and immune cells use to bend their own membranes — to grow branches, or to wrap around something they are about to swallow. It has nothing to do with genes, yet the part of it that makes copies of itself stick together is enough, on its own, to turn KMT2A into a leukaemia gene.
GFI1 — the repressor that LSD1 inhibitors actually work through, and the scan hit whose literature points the opposite way
GFI1 is a switch-off protein that decides when a young blood cell stops renewing itself and becomes a mature one. It matters here for an unexpected reason: the leukemia drugs called LSD1 inhibitors seem to work not by blocking an enzyme but by prising LSD1 off GFI1 — which is the same kind of trick this project is trying to pull with menin.
GLIS2 (GLI-similar 2) domain architecture
GLIS2 is the 3' partner of CBFA2T3::GLIS2. This entry lists where each of its functional parts sits along the protein, measured as residue numbers, so the fusion diagrams can draw which parts survive the break and which are thrown away.
GOLGA2 (GM130) - a Golgi matrix protein with 249 partners in the one yeast screen that reported it binding AFF4, which had five
GOLGA2, better known as GM130, is a scaffolding protein that holds the Golgi apparatus together. It was reported as a partner of the nuclear protein AFF4 in one large yeast screen, and in that same screen it bound 249 different proteins.
GPHN (gephyrin) — synaptic scaffold and molybdenum-cofactor enzyme, KMT2A fusion partner in t(11;14)(q23;q24)
Gephyrin is a brain protein that acts like a peg board: it clusters the receptors that calm nerve cells, and in the rest of the body it helps build a chemical the liver needs. It has nothing to do with switching genes on, which is why it matters so much that a tiny 15-letter piece of it is enough to turn a leukaemia gene into a cancer driver.
HCFC1 (HCF-1) — a founding member of the wild-type KMT2A complex that the fusion cannot bind
HCF-1 is one of the proteins that sticks to the normal, unbroken KMT2A protein. The fusion version of KMT2A loses the exact four-amino-acid landing pad HCF-1 needs, so HCF-1 belongs to the healthy machine and not to the cancer machine.
HCFC2 — menin's other host cell factor, present in the founding paper by gel filtration and never mentioned again
HCF-1 is a well-known part of the menin-KMT2A machine and has its own entry here. It has a less-studied sibling, HCF-2, which turned up in the very paper that founded this field and has essentially not been looked at since. The evidence for it is weaker than for HCF-1 and comes from a different kind of experiment.
HEXIM1 - the brake pad, the only part of the reservoir anyone has deliberately raised in a KMT2A-rearranged cell, and the result was that the leukaemia grew less
HEXIM1 is the protein that actually switches P-TEFb off inside the 7SK particle - the RNA only holds the two together. When researchers forced KMT2A-rearranged AML cells to make more HEXIM1, the cells grew more slowly, and when they removed it the cells grew faster and made more MYC.
HEXIM2 - the paralogue that holds the bound P-TEFb pool constant when HEXIM1 is removed, which is why single-gene experiments on this axis read flat
HEXIM2 is HEXIM1's near-twin and does the same job of switching P-TEFb off. When HEXIM1 is taken away, HEXIM2 steps in and keeps exactly the same amount of the enzyme switched off, which is why removing either one alone tells you very little.
Histone H3.3 - one protein, two genes, and the reason this graph should not file it as gene-h3f3a
H3.3 is a replacement version of a core histone that gets slotted into genes while they are being read, rather than only when DNA is copied. Two separate genes make it, and the protein they make is exactly the same, which is why this entry is about the protein and not about either gene.
HLF (hepatic leukemia factor) domain architecture
HLF is the 3' partner of TCF3::HLF, a rare and historically near-uniformly fatal B-ALL. This entry lists where each of its functional parts sits along the protein, measured as residue numbers, so the fusion diagrams can draw which parts survive the break and which are thrown away.
HNF4A - a liver and pancreas nuclear receptor acting as the metabolic effector of TIFAB in KMT2A::MLLT3 AML, on a single paper
HNF4A is a switch that normally runs the metabolism of liver and pancreas cells, and it turns out to be switched on in a mouse model of KMT2A-rearranged myeloid leukaemia, where it restores the leukaemia cells' ability to take up sugar and run their mitochondria. It is an odd gene to find in blood cancer, which is exactly why it is worth an entry.
HOXA9 — required for survival of established MLL-rearranged leukemia, and dispensable for starting it
HOXA9 is the best-known gene the KMT2A fusion switches on, and human MLL-rearranged leukemia cells die when you take it away. But mice engineered with an MLL fusion still get leukemia at the same rate without HOXA9, so it is needed to keep the disease going rather than to start it.
HOXD13 domain architecture (UniProt P35453)
HOXD13 is the partner protein in the NUP98::HOXD13 fusion. Almost all of it is unstructured except one small DNA-gripping module near the end, and that module is what the fusion keeps.
ICE1 — the 2,266-residue scaffold of the little elongation complex, the subunit that acts at initiation, and zero papers connecting it to leukaemia of any kind
ICE1 is the large scaffold of a small complex that lets RNA polymerase transcribe the genes for spliceosome components. It explains why ELL, an MLL fusion partner, is required by every cell — and it has never been studied in leukaemia at all.
ICE2 — the fourth LEC subunit, the one with no assigned step, no structure, and no leukaemia literature
ICE2 is the second scaffold protein of the little elongation complex, which lets RNA polymerase make the genes for spliceosome parts. Unlike its partners, nobody has assigned it a specific job, and like them it has never been studied in leukaemia.
IDH1 - the same oncometabolite, the same assay, and NO enrichment in KMT2A-rearranged AML (6.7% vs 7.6%, p = 0.53): the control that makes the IDH2 result interpretable
IDH1 is the cytoplasmic twin of IDH2, and when mutated it makes exactly the same DNA-methylation-raising molecule. In the biggest series that can test it, IDH1 mutations are no more common in KMT2A-rearranged leukaemia than in any other kind - which is the reason to be careful about the claim that IDH2 is.
IDH2 - the hypermethylating co-mutation that IS enriched in KMT2A-rearranged AML (16.0% vs 10.4%), at p = 0.0004 and not the p < 0.0001 the abstract claims
IDH2 is an enzyme that, when mutated, makes a molecule that jams the machinery cells use to strip methyl marks off DNA, so the DNA ends up over-methylated. In large adult AML series IDH2 mutations turn up more often alongside a KMT2A rearrangement than without one - which matters here, because the KMT2A fusion reads DNA that is not methylated.
IKBKB (IKK-beta) — the kinase whose knockdown tracks MLL-leukaemia growth most tightly, and whose inhibitors have failed clinically for a reason
IKK-beta is the kinase that starts the classical inflammation switch by tagging the brake protein for destruction. In MLL-rearranged leukaemia cells it is needed for growth, and blocking the kinase complex knocks the leukaemia fusion protein off the genes it controls — but drugs against this kinase have repeatedly failed in people because healthy immune cells need it too.
IKBKG (NEMO) — the scaffold that makes the canonical complex a complex, and a symbol that returns ZERO records in this disease
NEMO is not a kinase; it is the clamp that holds the two IKK kinases together and lets incoming signals switch them on. It is needed for MLL leukaemia cells to grow, but that single result sits inside one figure of one paper and no study has ever been written about NEMO in this disease.
ILF3 / NF90 - a nucleolar dsRNA-binding coactivator whose chromatin residency was mapped in the same leukaemia line that proximity-labelled KMT2A
ILF3 is a double-stranded-RNA-binding protein that sits on active gene switches and in the nucleolus, and it works as a partner for the machinery that turns genes on. It was the protein doing the tagging in the experiment that put KMT2A on this list, not the protein being tagged.
IMPDH2 - the rate-limiting guanine-nucleotide enzyme whose inhibition lowers LEDGF and menin protein and dismantles the MLL-fusion complex itself
IMPDH2 is the bottleneck enzyme cells use to build one of the four letters of RNA and DNA, and leukaemia stem cells with a KMT2A fusion run it unusually hard. Blocking it does something unexpected: the cell makes less of the two scaffold proteins the fusion needs, so the cancer-driving machine falls apart and a menin inhibitor works better.
ING2 - the PHD finger that reads H3K4me3 and turns it into REPRESSION, and whose founding co-crystal is filed under the MOUSE accession
ING2 grips the same chromatin tag that this leukaemia's fusion protein is built to create - but instead of switching a gene on, ING2 uses the tag to hold a switching-off machine in place. It also grips a rare fatty molecule in the nucleus, which competes for the same finger.
JMY - one AP-MS row with AFF4, in HCT116 only, from a paper whose whole argument is that interactomes differ between cell lines
JMY does two unrelated jobs: in the nucleus it helps the tumour suppressor p53 switch genes on, and in the cytoplasm it builds actin filaments. It was found once as a co-purifying protein of AFF4, in colon cancer cells and not in the kidney cells the same study also used.
KANSL1 - the NSL scaffold, and the protein through which WDR5 binding to NSL and to MLL/COMPASS is mutually exclusive
KANSL1 is the backbone of the NSL complex, which uses the enzyme MOF to acetylate histones. It matters here for a specific reason: KANSL1 and the KMT2A complex both need the same shared part, WDR5, and a structural study showed they cannot hold it at the same time - so the two complexes compete for a component the leukaemia complex needs.
KANSL2 - the second NSL subunit in the WDR5 structure, with one leukaemia record and a dependency profile that is the mirror image of a selective target
KANSL2 is one of the three proteins in the only solved structure of the NSL complex's WDR5-binding core. It is included here because that structure is the basis of the claim that NSL and the KMT2A-type complexes compete for WDR5 - and a claim about a structure should have entries for all the proteins in it.
KAT7 / HBO1 — an enzyme, a leukaemia-stem-cell dependency, and an inhibitor already in a crystal
KAT7 is an enzyme that tags histones with an acetyl group at H3K14, and leukaemia stem cells need it to keep their HOX genes switched on. It is an enzyme with a real active site, a published inhibitor and a crystal structure of that inhibitor bound — which is a much better starting position than most proteins in this list.
KAT8 (MOF) — a genuine KMT2A partner that binds the half of KMT2A the fusion discards
MOF is the enzyme that puts a specific acetyl tag on histones, and it forms a stable complex with normal KMT2A so the two marks land together. The contact is on the far end of KMT2A, which the leukemia fusion cuts off, so MOF stays behind with the healthy protein.
KDM1A (LSD1) — acts at the loci the fusion binds, spares normal stem cells, and still has never been shown to touch the fusion protein
LSD1 is an enzyme that strips activating marks off histones. In MLL-AF9 leukaemia it sits at exactly the places the fusion binds and keeps the oncogenic programme switched on, and blocking it lets the cells differentiate while leaving normal blood stem cells alone — one of the cleanest therapeutic windows in this whole field.
KDM2A - the CXXC reader that antagonises KMT2A leukaemogenesis, and displaces wild-type MLL rather than the fusion
KDM2A parks on the same unmethylated CG islands as the KMT2A fusion and strips off an activating chemical mark. It slows down KMT2A-driven leukaemia - but the published experiment shows it pushes the NORMAL MLL protein off chromatin while leaving the cancer fusion where it is.
KDM2B — 513 potent ChEMBL rows that are mostly patent bins, and no structure of the domain they claim to inhibit
KDM2B is the protein that parks a non-standard Polycomb complex on unmethylated CpG island DNA. On paper it looks like a well-drugged enzyme with hundreds of potent compounds, but almost all of those numbers are patent placeholders rather than real measurements, and leukemia cells barely need the gene at all.
KDM4C (JMJD2C/GASC1) — recruited by the retained KMT2A N-terminus, drugged in a KMT2A::AFF1 line, and invisible to CRISPR
KDM4C is an eraser enzyme that strips a silencing mark off DNA packaging proteins, and the leukemia fusion grabs it to keep its favourite genes switched on. A drug against it kills KMT2A::AFF1 leukemia cells, but deleting the gene outright does almost nothing, because three near-identical sibling enzymes cover for it.
KDM5A (JARID1A / RBP2 H3K4 demethylase) domain architecture
KDM5A is the 3' partner of NUP98::KDM5A. This entry lists where each of its functional parts sits along the protein, measured as residue numbers, so the fusion diagrams can draw which parts survive the break and which are thrown away.
KDM6A / UTX — a commonly asserted KMT2A partner that belongs to a different complex and points the opposite way
UTX is often listed as a partner of the KMT2A machinery. It is not: it belongs to the MLL3 and MLL4 complexes, which are different enzymes, and in leukemia it behaves as a brake that gets broken rather than an accelerator that gets hijacked.
KMT2A / MLL1 — the gene that breaks
KMT2A is a very large protein that acts as a foreman on DNA: it finds the right genes and marks them to be switched on during blood development. When the gene breaks in the middle, the surviving front half keeps the ability to find genes but loses the ability to stop, and gains a partner that jams the switch on.
KMT2A*: a menin-blind, LEDGF-blind, SET-competent KMT2A protein made from out-of-frame reciprocal alleles
KMT2A has a second start site hidden in the middle of the gene. When a translocation leaves the mirror-image allele out of frame, that hidden start still works and makes a short KMT2A protein - one that keeps the enzyme end but has lost every surface the menin drug and its partner protein use.
LARP7 - the clamp on the 3' end of 7SK, whose human loss-of-function phenotype is dwarfism and intellectual disability rather than cancer
LARP7 grips the tail of the 7SK RNA and keeps the whole particle from falling apart; without it the RNA is degraded. Children born with two broken copies have severe growth failure and intellectual disability, which is the clearest evidence that this brake matters in a living human.
LEO1 — the weakest-evidenced entry in the PAF1 complex, and the gene whose Ensembl id this project already got wrong once
LEO1 is the fourth protein in the PAF1 machine. Unlike the other three it has never been studied on its own in KMT2A-fusion leukemia, and it is also the gene where this project once pulled data for the wrong gene entirely because of a bad identifier.
LRRK2 - the AFF4 row is one of 78 hits on an in-vitro protein array, but the protein has a real AML literature that arrived through USP7 and interferon, not through the interactome
LRRK2 is the large kinase best known from Parkinson's disease. Its listing as an AFF4 partner comes from one glass-slide protein array, not from a cell - but separately, and much more interestingly, two papers put LRRK2 inside acute myeloid leukaemia biology.
MAML2 (mastermind-like 2) - the Notch co-activator that is a KMT2A partner in leukaemia and in thymoma
MAML2 is a helper protein for Notch, one of the signals that tells a cell which kind of cell to become. It is unusual among KMT2A partners because the same fusion appears both in leukaemia and in a solid tumour of the thymus.
MAP3K14 (NIK) — the one node in this pathway where ACTIVATION, not inhibition, is the proposed therapy, and the one that also damages normal stem cells
NIK is a kinase that is normally destroyed as fast as it is made; when you stop that destruction it switches on the second NF-kB circuit and, in mouse MLL leukaemia, the disease gets better rather than worse. The catch is that the same manipulation also harms normal blood stem cells, so it would have to be aimed at the leukaemia specifically.
MAT2A - the enzyme that makes the methyl donor, and the one metabolic dependency in this disease with two independent papers and a lymphoid PDX result
Cells turn the amino acid methionine into SAM, the universal chemical tag that gets stuck onto DNA and histones, and MAT2A is the enzyme that does it. KMT2A-rearranged leukaemia keeps its chromatin heavily tagged, so it needs more SAM than other leukaemias - which means starving it of methionine, or blocking MAT2A, kills it.
MBD1 - one protein with a methyl-CpG reader and three CXXC fingers, only the third of which binds unmethylated CpG, and whose CXXC fails the MLL-AF9 swap
MBD1 is unusual: it carries a module that reads chemically TAGGED CG letters and three more that are built like the KMT2A fusion's untagged-CG reader - but only one of the three actually grips untagged DNA. When researchers put MBD1's CG-reader into the leukaemia fusion in place of KMT2A's, the fusion stopped causing leukaemia, and the borrowed module turned out not to bind DNA at all on its own.
MCL1 - named in 34 entries of this graph with no entry of its own until now; a mitochondrial survival protein high in infant KMT2A-rearranged ALL, druggable, and so far undruggable in patients
MCL1 is a protein that sits on the mitochondria and stops a cell from killing itself, and infant KMT2A-rearranged leukaemia makes a lot of it - which is one reason steroids work badly in these babies. Drugs that block MCL1 kill these leukaemia cells beautifully in a dish, and in people they have so far mostly produced heart-muscle blood tests going up without much benefit.
MED12 - a named menin-associated protein that restrains menin's transcriptional output, and the anchor of the CDK8 module a KMT2A-r AML paper co-targets
MED12 is the part of the Mediator complex that holds the kinase CDK8. It is directly relevant here because it is one of a small set of proteins found sitting next to menin - the protein every menin inhibitor targets - and because a KMT2A-rearranged AML study used a drug hitting both RET and the MED12-associated kinase CDK8.
MED19 - the Mediator subunit curated against an engineered MLL-AFF4 fusion protein, and the one REST uses to deliver G9a
MED19 is a small piece of Mediator, the adaptor that carries instructions from gene switches to the machine that copies DNA into RNA. It is one of only two proteins in the whole curated interaction record that was ever tested against a laboratory-built MLL fusion protein rather than against normal MLL.
MED23 — rank 16 of 17,728, survives the lineage split with the B-lineage half stronger, and is not absent from the leukaemia literature after all
MED23 is a Mediator subunit that, in this project's dependency data, looks like one of the most KMT2A-selective genes in the entire genome - and unlike most such hits it holds up when the leukaemia cell lines are split into myeloid and B-lineage groups. Not one published paper mentions it together with KMT2A.
MED26 — a common-essential Mediator subunit with a solved 87-residue domain and no ChEMBL target entity at all
MED26 is a piece of Mediator, the machine that connects gene switches to RNA polymerase. Almost every cell needs it, it has a small, neatly solved domain that looks like a handle a drug could grab — and no one has ever registered a single compound against it.
MEF2C — an output-program transcription factor, the second-strongest leukemia co-dependency, and a gene humans cannot spare one copy of
MEF2C is a switch protein that the KMT2A fusion turns on downstream, and leukemia cells depend on it heavily. The catch is that people born with only one working copy have severe intellectual disability and epilepsy, so partially blocking it in a baby is a much more frightening proposition than the cell-line data alone suggests.
MEIS1 — rate-limiting for MLL leukemia stem cells, a surviving co-dependency, and embryonic-lethal when removed
MEIS1 partners with HOXA9 and sets how many leukemia stem cells a KMT2A fusion can make — turn it down and the disease loses its engine. Mice without it die before birth, which is the problem with treating it as a drug target.
MEN1 / menin — the adaptor the fusion cannot work without
Menin is a small protein that acts like a docking clamp, holding KMT2A onto DNA. Fusion versions of KMT2A keep the part that grips menin, so blocking that grip is the most advanced way we have of switching the cancer program off.
MEPCE - absent from this graph until now, and a deeper pan-dependency than BRD4
MEPCE puts a protective cap on the 5' end of the 7SK RNA and then stays bound to hold the particle together. The string MEPCE appeared nowhere in this knowledge graph before this entry, even though cells need it about as much as they need BRD4.
MLLT1 / ENL — the reader that beat its own paralogue, with a degrader already in hand
ENL is a small reader protein that grips acetylated histones and helps the fusion switch genes on. Two independent teams found that leukemia cells need ENL but not its near-identical twin AF9, and there is already a molecule that destroys ENL rather than merely blocking it.
MLLT10 / AF10 — domain architecture and residue coordinates
AF10 is the KMT2A partner that dominates infant myeloid leukemia, and it works by handing the fusion a grip on an enzyme called DOT1L. This entry gives the residue numbers for that grip, which is one of the few partner interfaces in this whole project with a real crystal structure behind it.
MLLT11 / AF1q — a 90-residue protein with no chromatin domain and no oligomerisation module, KMT2A partner in t(1;11)(q21;q23)
AF1q is one of the smallest proteins in the human genome that causes a cancer when it gets stuck onto another gene — just 90 building blocks long, with no part that touches DNA and no part that makes copies of itself stick together. It is the partner that none of the standard explanations for KMT2A leukaemia can account for.
MLLT3 / AF9 — a famous fusion partner that is not a dependency, and a hub domain that is
AF9 is one of the most common partners that KMT2A gets stuck to in leukemia, and its tail end is a famous meeting point for several other proteins. Yet when you delete the normal AF9 gene from leukemia cells, nothing much happens — the drug target here is the meeting point, not the gene.
MLLT6 (AF17) — the fourth KMT2A fusion partner inside DotCom, and the only one no cell needs
AF17 is one of four proteins that hold the H3K79 methylation enzyme together, and like its three relatives it can also be the partner in a KMT2A fusion. Unlike them, removing AF17 has no effect on any cancer cell line tested.
MLLT6 / AF17 — domain architecture, and where the leucine zipper actually is
A map of the AF17 protein showing which parts survive into the leukaemia fusion and which are cut away. The self-sticking coil that survives sits at residues 719 to 753; the three zinc-finger reading modules that are lost sit at the very front of the protein.
MSL1 - the MSL scaffold that carries MOF and MSL2 to HOXA9 and MEIS1, with zero records against KMT2A and a mapped, drawable domain architecture
MSL1 is the backbone of the MSL complex: it holds the acetylating enzyme MOF at one end and the ubiquitin ligase MSL2 at the other. Those two enzymes together switch on HOXA9 and MEIS1 - the exact two genes the KMT2A fusion depends on - yet not a single published paper puts MSL1 and KMT2A in the same study.
MSL2 — the second H2B ubiquitin ligase, sitting inside the MOF complex the KMT2A fragment contacts
MSL2 is a RING enzyme inside the same complex as MOF, and it puts a ubiquitin tag on a different spot of histone H2B than the better-known enzyme does. That tag switches on DOT1L too, and it was shown to matter at HOXA9 and MEIS1.
MTUS2 (TIP150) - a microtubule plus-end tracking protein with 153 partners in the one yeast screen that reported it binding AFF4
MTUS2 rides along the growing ends of microtubules in the cytoplasm. It was listed as a partner of the nuclear protein AFF4 on the strength of one yeast screen, in which it also bound 152 other proteins.
MYB — rank 17 of 17,728 for KMT2A selectivity, and the only node in this beat where partial, transient suppression eradicated leukaemia without stopping normal blood production
MYB is the transcription factor the MLL-AF9 fusion uses to keep leukaemia cells renewing themselves. Turning it down only partly, and only for a while, wiped out aggressive leukaemia in mice while normal blood production carried on — which is about as clean a therapeutic window as this field has ever produced.
MYH11 — domain architecture and residue coordinates
MYH11 is the motor protein of smooth muscle: a compact head that pulls on actin, and behind it a very long rigid rod that lets many copies of the molecule bundle into filaments. In one acute myeloid leukaemia the rod alone is welded onto CBFB, and it is the rod's habit of bundling that makes the fusion dangerous.
MYH9 (non-muscle myosin IIA) — the best-validated menin partner nobody in this field talks about, and the only one that maps to menin's N-terminus
Menin was found stuck to the motor protein that pinches a dividing cell in two, and the two were seen together in the pinch itself. Four different experiments in one paper agree, which is more than most menin partners have. It has been almost entirely ignored since, and it raises a question nobody asks: does a menin inhibitor do anything to cell division that has nothing to do with genes being switched on or off?
MYO1F — unconventional myosin IF, a KMT2A partner with neither a transactivation domain nor a dimerisation domain
MYO1F is a tiny motor protein that walks along the cell's actin cables — the kind of machine immune cells use to crawl and squeeze through tissue. When it is fused to KMT2A it makes leukaemia in babies, and it is one of only two such partners that the people who discovered it said outright has no gene-switching part and no self-sticking part either.
NELF — the brake that has to come off for elongation, absent from the KMT2A literature, and less required in KMT2A-rearranged lines than in any other leukaemia
NELF is the four-protein brake that holds RNA polymerase still at the start of a gene until a signal releases it. In leukaemia cells driven by a KMT2A fusion, two of its four parts are among the least-needed genes in the entire genome relative to other leukaemias — which is what you would expect if these cells are already running with the brake off.
NELFB (COBRA1) — the NELF scaffold, and the single least KMT2A-selective gene of the sixty least selective in the genome
NELFB is the scaffold that holds the pausing brake together. Of nearly eighteen thousand human genes, it is one of the sixty that KMT2A-rearranged leukaemia cells need least compared with other leukaemia cells — the strongest single anti-dependency in the whole pausing machinery.
NELFE (RDBP) — the RNA-binding subunit of the pausing brake, and the alias that inflated a record count fifty-fold before it was caught
NELFE is the part of the pausing brake that actually grips the new RNA as it emerges from the polymerase. Nothing connects it to KMT2A leukaemia — and finding that out required discarding a search that looked reasonable and was wrong by a factor of fifty.
NFKB1 (p105/p50) — the only menin-NF-kB subunit with a co-immunoprecipitation, and the arm the MLL leukaemia data points at
NFKB1 is made as a long protein, p105, that gets trimmed down to p50, the DNA-binding half of the classical NF-kB switch, and menin binds it. Nobody has checked whether removing menin with a drug changes that, even though the signalling that makes p50 is unusually active in MLL-rearranged leukaemia cells.
NFKB2 (p100/p52) — menin binds the non-canonical subunit too, and activating that arm SUPPRESSES MLL-AF9 leukaemia
NFKB2 belongs to the second, slower NF-kB circuit. Menin binds it, and in mouse MLL leukaemia switching that second circuit on actually shrinks the disease rather than feeding it. So the two NF-kB arms menin touches pull in opposite directions, and a drug that let go of both at once would be doing two contradictory things.
NFKBIA (IkB-alpha) — the switch itself. This graph has entries for everything the switch controls and none for the switch
IkB-alpha is the brake that holds NF-kB in the cytoplasm; destroying it is what actually flips the switch on. Everything upstream in this pathway exists to destroy it, and everything downstream happens because it was destroyed.
NPM1 (nucleophosmin) — the nucleolar shuttle whose mutant form binds chromatin with KMT2A and buys the second menin-inhibitor indication
NPM1 normally lives in the nucleolus, the cell's ribosome factory, and shuttles between compartments carrying other proteins. In about a third of adult AML a tiny four-letter insertion moves it into the cytoplasm instead, and that displaced protein switches on the same HOXA/MEIS1 gene programme that KMT2A fusions switch on — which is why the same menin-blocking drug works in both diseases.
NPM1 (nucleophosmin) domain architecture
NPM1 is the 5' partner of NPM1::ALK anaplastic large cell lymphoma. This entry lists where each of its functional parts sits along the protein, measured as residue numbers, so the fusion diagrams can draw which parts survive the break and which are thrown away.
NSD1 (nuclear receptor-binding SET domain protein 1) domain architecture
NSD1 is the 3' partner of NUP98::NSD1. This entry lists where each of its functional parts sits along the protein, measured as residue numbers, so the fusion diagrams can draw which parts survive the break and which are thrown away.
NSL complex - the MOF assembly that five NUP98 fusions bind alongside MLL1, and that competes with MLL/COMPASS for WDR5
NSL is a nine-protein machine built around the histone-acetylating enzyme MOF. It matters to this project twice over: five leukaemia fusion proteins that contain no KMT2A at all grab both NSL and the normal MLL1 complex to cause leukaemia, and NSL competes with MLL-type complexes for a shared component called WDR5.
NTRK3 (TrkC neurotrophin receptor) domain architecture
NTRK3 is the 3' partner of ETV6::NTRK3. This entry lists where each of its functional parts sits along the protein, measured as residue numbers, so the fusion diagrams can draw which parts survive the break and which are thrown away.
NUP98 — domain architecture and residue coordinates (with the 1817/920 numbering trap resolved)
NUP98 is part of the gate in the wall of the cell nucleus, and its front half is a floppy, sticky, repetitive tail that helps cargo slip through. In several leukaemias that sticky tail is welded onto a gene-regulating protein, and the tail's tendency to clump into droplets is what reorganises the cell's gene programme.
NUTM1 (NUT) - a testis-restricted, fully disordered protein that writes no mark and instead switches ON a writer, and the 3' partner of the KMT2A-wild-type infant B-ALL with the opposite prognosis
NUTM1 is a protein normally made only in maturing sperm cells, where it switches on a chromatin-modifying enzyme called p300 so that histones can be stripped out of the sperm genome. When a chromosome break fuses it to another gene in a baby's blood cell that switch gets turned on in the wrong place and causes leukaemia - but a leukaemia that, unlike the KMT2A-rearranged kind, responds very well to standard chemotherapy.
P-TEFb (CDK9 + cyclin T1) — the deepest dependency and the clearest therapeutic-index wall in the interactome
P-TEFb is the enzyme that releases stalled RNA polymerase so genes can actually be transcribed. KMT2A-rearranged leukemia cells need it more than almost anything else — but so does essentially every other cell in the body, which is why fifteen years of CDK9 drugs have never got past late-stage trials.
PAF1 — pan-essential subunit of a complex the fusion needs, where the window is in the interface and not in the protein
PAF1 is part of a machine that travels with RNA polymerase, and the KMT2A fusion grabs onto it using a piece of KMT2A that the break always keeps. Every cell needs PAF1, so you cannot block the protein — but you may be able to block only the handshake between it and the fusion, which is the same logic that makes menin a target.
PBX1 — domain architecture and residue coordinates
PBX1 is a DNA-gripping protein that works as a partner, holding HOX proteins in place so they land on the right genes. In one childhood leukaemia its own front end is cut away and replaced by a loud activation module from TCF3, turning a positioning device into a permanent amplifier.
PBX2 - an AFF4 co-purifying protein in two rounds of one AP-MS atlas, and the PBX paralogue the leukaemia literature explicitly excluded
PBX2 is one of three closely related proteins that partner HOX transcription factors; this graph's HOXA story is built on its sibling PBX3. PBX2 shows up as a co-purifying protein of AFF4 in a large automated survey of human protein interactions, and the one leukaemia paper that compared the three siblings directly said PBX2 is not the one that matters.
PCGF1 — the PRC1.1 scaffold whose LOSS causes menin-inhibitor resistance, so it is a restraint on the leukaemia rather than a dependency of it
PCGF1 is the central strut of a small protein machine (PRC1.1) that keeps certain genes switched off in leukaemia cells. Removing PCGF1 does not kill the leukaemia — it lets MYC come back on and makes the cells resistant to menin inhibitors, which is the opposite of what a drug target looks like.
PICALM / CALM - domain architecture, residue coordinates, and which parts the fusion keeps
This is the parts list for the PICALM protein, with the numbers of the amino acids each part occupies. It matters because the leukaemia fusion keeps almost all of PICALM and throws away only the last four amino acids - so the interesting question is which of the kept parts actually does the damage.
PIP4K2A - the lipid kinase at a genome-wide childhood-ALL susceptibility locus, whose product is the ligand a PHD finger reads
PIP4K2A is an enzyme that converts one rare signalling fat into another, and it sits at one of the strongest inherited risk locations for childhood leukaemia. The fat it consumes is the molecule that a chromatin-reading finger grips, so a lipid kinase turns out to sit one step from the histone-reading machinery this project is about.
PML — domain architecture and residue coordinates
PML builds tiny droplet-like bodies inside the cell nucleus that act as collection points for stress responses. The front third of the protein is the scaffolding that makes those bodies, and in acute promyelocytic leukaemia that scaffolding is welded onto a hormone receptor, which is why the leukaemia cells never mature.
PPIE (CyP33) — the off-switch the fusion throws away, and the reason PHD3 reinsertion kills transformation
CyP33 is a protein that docks onto a small finger of normal KMT2A and flips it from switched-on to switched-off. Every leukemia fusion deletes that finger, and putting it back makes the fusion stop transforming cells — so this is a brake the cancer removes rather than an accelerator it adds.
PRC1.1 — the complex this graph has been leaning on, now enumerated subunit by subunit, and the enumeration changes the conclusion
PRC1.1 is a small machine that finds unmethylated stretches of DNA at gene starts and puts a silencing tag on the surrounding histones. This graph has used it to explain why some KMT2A leukaemias resist menin inhibitors — but until now only two of its six parts had entries, and writing the other four in shows the complex is a brake on the leukaemia rather than a fuel line.
PRC2 - the complex, not the subunit, is the unit every experiment in this disease actually resolves
PRC2 is a five-part molecular machine that stamps a 'keep this gene off' mark on chromosomes and then reads its own stamp to spread it further. In KMT2A leukaemia, every experiment that has produced a clear answer removed the whole machine rather than one part - which is why this entry exists as a complex rather than as a list of genes.
PRMT1 — an MLL-fusion cofactor whose first-in-human inhibitor was terminated for lack of efficacy and toxicity
PRMT1 tags proteins with methyl groups and is an essential part of one particular MLL fusion's machinery. A drug against it reached patients, and the trial was stopped early because it did not work and caused too many problems.
PRMT3 - the mostly-cytoplasmic ribosome methyltransferase that also writes the ACTIVATING mark on the residue PRMT5 represses, has an acute-leukaemia degrader, and is invisible to this project's own mark-query rule
PRMT3 mostly sits outside the nucleus and tags a ribosome protein, which is why this project reached it through a tumour suppressor that switches it off. But it also puts an activating tag on the very same histone spot that PRMT5 puts a silencing tag on - and a drug that destroys PRMT3 already slows acute leukaemia cells.
PRMT5 - the type II arginine methyltransferase this graph has reached for sixteen times, whose mark on H3R2 RECRUITS WDR5 to chromatin
PRMT5 is an enzyme that adds a pair of methyl tags to arginine residues, on histones and on many other proteins. The tag it places on one spot of histone H3 pulls in WDR5, a part of the very machine this disease's fusion protein hijacks.
PRMT6 - the enzyme whose mark is MUTUALLY EXCLUSIVE with the MLL complex on the H3 tail, and which shuts HOX genes down by evicting WDR5
PRMT6 writes a single chemical tag on histone H3 that physically blocks the MLL machine from landing on the same stretch of protein. When there is more of that tag, HOX genes - the genes this leukaemia lives on - go quiet.
PRMT7 - the enzyme whose class is still disputed, which represses a KMT2 enzyme by writing H4R3me2s, which is required for leukaemia stem cells in CML, and whose only H3R2 claim is the one this beat was dispatched to check
PRMT7 adds methyl tags to arginines, and it is the odd one out in its family - biochemists still disagree about whether it can add one tag or two. The tag it puts on histone H4 blocks a KMT2 enzyme from working, and losing PRMT7 kills leukaemia stem cells in another leukaemia without harming normal blood.
PSIP1 / LEDGF p75 — the tether with the best therapeutic index in the whole interactome
LEDGF is the protein that reads a chemical mark on chromatin and uses it to park the menin-KMT2A complex on the right genes. Cells in general do not seem to need it at all, but KMT2A-rearranged leukemia cells do, which makes it the most promising second target after menin.
RARA — domain architecture and residue coordinates
RARA is a receptor that reads vitamin-A signals and, when it does, tells blood cells to finish maturing. In acute promyelocytic leukaemia its front end is chopped off and replaced by PML, which jams the receptor in the off position until a high dose of vitamin-A acid forces it back on.
RBBP5 — a WRAD subunit that binds the one part of KMT2A every fusion throws away
RBBP5 is part of the four-protein core that switches on KMT2A's enzyme activity. That enzyme sits in the SET domain, and the SET domain is lost in every KMT2A fusion — so RBBP5 is not a partner of the fusion, it is a partner of the normal protein the fusion left behind.
REL (c-Rel) — the third canonical subunit, measured once and found unchanged, and named in zero papers in this disease
c-Rel is the third member of the classical NF-kB family and the one nobody in this field has looked at. The single time it was measured in an MLL leukaemia cell its level did not change, and no paper has ever been written about it in this disease.
RELA (p65) — menin binds it and represses it, and it is required to keep MLL oncoproteins on chromatin
RELA is the business end of the classical NF-kB switch, and menin binds it and holds it down. Separately, RELA is required to keep the KMT2A fusion protein sitting on the genes it controls, so a drug that removes menin may be letting go of a brake on the very thing it is trying to switch off.
RELB — the NF-kB subunit that SUPPRESSES MLL-AF9 leukaemia, with a direct ChIP target list and a genetic rescue, which makes it the best-evidenced anti-leukaemic node in this beat
RelB is the transcription factor of the slower, second NF-kB circuit, and in mouse MLL leukaemia it does the opposite of its famous cousin RELA: switching it on shrinks the disease. It works by turning on a gene that restrains blood-stem-cell self-renewal and turning off one the leukaemia needs.
REST / NRSF - the transcription factor that delivers G9a to chromatin through MED19 and MED26, is a paediatric cancer gene, and is the worst search token in this corpus
REST is a switch that keeps nerve-cell genes turned off in every cell that is not a nerve cell, and it does it by hiring silencing enzymes - including the H3K9 methyltransferase G9a, which it reaches through two Mediator subunits this project has already written about. It is also a Wilms tumour gene, and its name is an ordinary English word, which makes it almost impossible to search for.
RING1 (RING1A) - the paralogue that is not RING1B, and the reason a substring search for 'RING1' in this graph returns the wrong gene
RING1 is the twin of the better-known Polycomb enzyme RING1B, and the two do the same chemical job so well that removing just one of them changes almost nothing. The trap this entry exists to close is a naming one: RING1B is not this gene - it is a gene called RNF2 - so a paper or a search that says 'RING1' may mean either, and telling them apart matters because the published leukaemia experiments removed both at once.
RNF2 (RING1B) — the one catalytic subunit shared by every PRC1 flavour, which is why the genetics point two ways at once
RNF2 is the enzyme that attaches a small tag (ubiquitin) to histone H2A, the chemical mark that Polycomb complexes use to keep genes quiet. Because the same enzyme sits inside every version of Polycomb complex 1, deleting it removes several different machines at once — which is why experiments that delete it and experiments that delete only its PRC1.1 partners give opposite-looking answers.
RNF20 (BRE1A) — the H2B ubiquitin ligase that MLL-rearranged leukaemia needs, and that DepMap says every other cancer line needs just as much
RNF20 is the enzyme that attaches the ubiquitin tag to histone H2B, the tag that DOT1L needs in order to mark KMT2A-fusion target genes. Switching it off does shut down those genes and stops the leukaemia growing — but the same enzyme turns out to be needed by almost every dividing cancer cell, so it is a real dependency without a therapeutic window.
RNF40 (BRE1B) — the obligate partner of RNF20, and the one of the pair with a documented link to differentiation and bivalency resolution
RNF40 is the second half of the two-protein enzyme that tags histone H2B with ubiquitin; neither half works alone. Its distinctive contribution is that cells need it to switch genes from a poised, undecided state to an active one during differentiation — which is the exact step KMT2A-fusion leukaemia cells fail to take.
RPRM (reprimo) - the other half of menin's Ewing 2007 pair, a protein with no experimental localisation at all, and a promoter methylated in 45% of paediatric AML
Reprimo is a very small protein thought to help p53 stop the cell cycle, and its gene is switched off by DNA methylation in nearly half of children with acute myeloid leukaemia. It was also pulled down once with menin in a large screen, and that single pull-down is the only physical link to this graph's central protein.
RTF1 — the PAF1C subunit the graph was missing, on a complex that does bind the retained KMT2A fragment
RTF1 is one of the parts of PAF1C, the complex that clamps onto RNA polymerase and that the KMT2A fusion grabs using a piece of KMT2A it still has. It completes a set of five parts, four of which were already in this graph.
RUNX1 — activated by KMT2A::AFF1 and then bound by the reciprocal fusion, in the one published two-oncoprotein circuit in t(4;11)
The KMT2A fusion switches the RUNX1 gene on, and the RUNX1 protein that results then sticks to the other, mirror-image fusion protein made by the same translocation. It is the only published case where the two halves of the t(4;11) break work on each other through a shared target.
RUNX1 — domain architecture and residue coordinates (with the RUNX1b/RUNX1c +27 offset)
RUNX1 is the master switch for making blood cells: one compact module grips DNA, and a partner protein clamps onto that same module to make the grip stronger. Two different official versions of the protein are in circulation and their residue numbers differ by twenty-seven, which is enough to put the DNA-gripping module in the wrong place on a diagram.
RUNX1T1 (ETO / MTG8) domain architecture
RUNX1T1 is the 3' partner of RUNX1::RUNX1T1, the t(8;21) fusion of core-binding-factor AML. This entry lists where each of its functional parts sits along the protein, measured as residue numbers, so the fusion diagrams can draw which parts survive the break and which are thrown away.
SAMHD1 - the dNTP hydrolase that is the real biomarker behind the IMPDH story, and the reason the KMT2A framing of that story does not hold
SAMHD1 is an enzyme that chews up the building blocks of DNA, including the chemotherapy drug cytarabine after the cell has activated it - so cells with a lot of SAMHD1 resist that drug. Two 2026 papers show that blocking a metabolic enzyme called IMPDH starves SAMHD1 of the signal that switches it on, which makes cytarabine work again; the same papers found this has nothing to do with whether the leukaemia carries a KMT2A fusion.
SEC — the AFF1/AFF4-scaffolded super elongation complex, the particle KMT2A::AFF1 is a permanent subunit of
SEC is the machine that tells a stalled gene-copying enzyme to go, and it is the one the infant-leukaemia fusion protein is bolted onto. It is built on the scaffolds AFF1 and AFF4 and carries the kinase P-TEFb, the ELL speed-up subunits and the two mark-reading proteins ENL and AF9.
SEC-L2 — the AFF2-scaffolded super-elongation-like complex: P-TEFb plus ENL and AF9, and NO ELL
SEC-L2 is one of the cell's several near-identical transcription-speeding machines. It is built on AFF2 instead of the AFF1 that the infant-leukaemia fusion uses, and it carries the same two mark-reading proteins, ENL and AF9 — but it is missing the ELL speed-up subunits the main machine has.
SEC-L3 — the AFF3-scaffolded complex, and the only SEC-family member with a known physiological job: genomic imprinting and X inactivation
SEC-L3 is a transcription-speeding machine built on AFF3. It carries the same two mark-reading proteins, ENL and AF9, as the machine the infant-leukaemia fusion hijacks. What makes it different is its day job: AFF3 sits on imprinted genes and on the switch that silences one X chromosome, so this is the copy of the machine that helps keep parent-of-origin gene silencing in place.
SEPTIN11 (septin-11) — KMT2A fusion partner at 4q21, adjacent to AFF1
Septin-11 is the closest relative of septin-6 and sits on chromosome 4 just next to AFF1, the commonest KMT2A partner of all. That accident of position means a KMT2A break joining it looks, under the microscope, exactly like the famous t(4;11).
SEPTIN2 (septin-2 / NEDD5) — KMT2A fusion partner in t(2;11)(q37;q23)
Septin-2 is the septin whose crystal structures taught the field how septins polymerise. It is also a rare KMT2A fusion partner, almost always in leukaemia that followed earlier chemotherapy.
SEPTIN5 (septin-5 / CDCrel-1) — KMT2A fusion partner in t(11;22)(q23;q11.2)
Septin-5, originally called CDCrel-1, is a cell-division septin that was the very first septin found fused to KMT2A, in infant twins with leukaemia. It is the shortest of the KMT2A-partner septins and has the shortest self-association helix.
SEPTIN6 (septin-6) — cytoskeletal GTPase, KMT2A fusion partner in t(X;11)(q24;q23)
Septin-6 is one of a family of proteins that bind the energy molecule GTP and snap together into filaments, forming part of the cell's internal scaffolding. It has no ability to read or write on DNA, which is why finding it fused to a gene-control protein in leukaemia was so surprising.
SEPTIN9 (septin-9 / MSF) — cytoskeletal GTPase, KMT2A fusion partner in t(11;17)(q23;q25)
Septin-9 is a scaffolding GTPase of the cell skeleton whose official protein name still records how it was found: 'MLL septin-like fusion protein'. It was discovered because it turned up fused to KMT2A in leukaemia, and only later studied as a normal protein.
SET / TAF-Ibeta domain map (Q01105, 290 aa) - dimerisation, earmuff, and the acidic tail that binds MLL
SET is a small protein that handles histones and also blocks a major cellular phosphatase. Its three parts are a clamp that makes it pair up, a curved 'earmuff' body that grips histones, and a negatively charged tail - and it is the tail that grabs MLL.
SETD1A — a leukemia dependency whose enzyme activity turns out to be dispensable
SETD1A is a separate H3K4-writing enzyme that shares the same four-protein core as KMT2A. Leukemia cells need it — but not for its enzyme activity, which can be deleted without hurting them, so the usual assumption about why it matters is wrong.
SETD1B - the COMPASS paralogue round 4 deferred for want of a leukaemia paper, which had already been published: its SET domain sets H3K4me3 breadth and MYC in MLL-rearranged AML
SETD1B is one of six human enzymes that put the activating H3K4 methyl mark on genes. In MLL-rearranged leukaemia, cutting its catalytic domain narrows that mark across the genome, switches MYC down and kills the cells - and adding MYC back rescues them.
SGF29 — a SAGA reader with a Tudor pocket, intermediate essentiality, and the best unexploited profile in this beat
SGF29 reads the same H3K4 mark that KMT2A writes and uses it to bring a large acetylation machine to those genes. It is needed by leukemia cells more than by most other cells, it has a real binding pocket, and almost nobody is working on it.
SIAH1 — the E3 ligase that degrades AFF1, and that the reciprocal AF4::KMT2A fusion escapes
SIAH1 is a disposal tag-writer that normally keeps AFF1 protein levels low by marking it for destruction. The t(4;11) translocation makes a second, reversed fusion protein that keeps AFF1's tag-binding end but is shielded from destruction, so it builds up in the cell.
SIAH2 — the second AFF1 E3 ligase, reported only alongside SIAH1 and never separated from it
SIAH2 is a close twin of SIAH1 and was reported in the same experiment as a second enzyme that marks AFF1 for destruction. Nobody has ever tested the two separately, so we do not know whether both matter or only one does.
SIRT1 - not a fusion partner, the least required gene in KMT2A-rearranged lines, and the one target here where the direction of the drug is genuinely disputed
SIRT1 is an enzyme that switches genes off, and it is not something a KMT2A fusion recruits. Whether the useful drug here turns SIRT1 on or off is genuinely disputed, and nobody has run the two directions side by side in KMT2A-rearranged leukaemia.
SKIC8 (formerly WDR61) — the last PAF1C subunit, and a gene-symbol trap worth recording
SKIC8 is the sixth and smallest part of PAF1C, the complex the KMT2A fusion grabs with a piece it still has. It used to be called WDR61, and looking it up under the old name returns nothing, which is how identifiers quietly go wrong.
SKP1 — a pan-essential housekeeping adaptor that is also a structural subunit of PRC1.1, which makes it the one PRC1.1 member that can never be a target
SKP1 is a small adaptor protein that every human cell uses constantly to help tag proteins for destruction. It is also, unexpectedly, a genuine structural part of the PRC1.1 gene-silencing machine — which means you can never drug it, because removing it kills every cell equally.
SMARCA2 (BRM) - the paralogue nobody measured, and the reason a DepMap rank cannot predict what a BRG1/BRM inhibitor does
SMARCA2 is the alternative motor that can slot into the same chromatin machine as SMARCA4. Knock out one and the other often covers for it - which is why the drugs that reached the clinic hit both, and why a genetic screen that tests them one at a time understates the target.
SMARCA4 (BRG1) - a tumour suppressor by mutation and a maintenance requirement by knockdown, and in KMT2A-rearranged lines the dependency data say it is not selective at all
SMARCA4 is the motor that pushes nucleosomes around inside the SWI/SNF machine. In many solid tumours losing it causes cancer, but leukaemia cells need it to stay alive and undifferentiated - and the honest reading of the current data is that leukaemia cells in general need it, not KMT2A-rearranged ones in particular.
SMARCB1 (BAF47/INI1/SNF5) - zero entries, zero mentions and, on the evidence, correctly so: it is lost in primary AML rather than required
SMARCB1 is the classic tumour-suppressor part of the SWI/SNF machine - losing it causes rhabdoid tumours in young children. In leukaemia the evidence points the same way: primary AML blasts have lost it, and KMT2A-rearranged cells do not appear to need it.
SMARCD1 (BAF60A) - rank 59 of 17,728, survives the lineage split, and had never been named in this graph or studied in this disease
SMARCD1 is the sister part to SMARCD2 in the same slot of the SWI/SNF machine. It has the strongest KMT2A-rearranged dependency signal of any canonical SWI/SNF subunit tested here - and essentially nobody has studied it in leukaemia.
SMARCD2 (BAF60B) - named in one KMT2A paper in the whole of PubMed, and the dependency it carries is myeloid identity, not the rearrangement
SMARCD2 is one of three interchangeable parts that sit in the same slot of the SWI/SNF machine. Leukaemia cells with an MLL rearrangement need it - but when the test is split by cell lineage, the need turns out to belong to myeloid cells generally, not to the rearrangement.
SPI1 / PU.1 — a myeloid dependency that a KMT2A panel mistakes for its own
PU.1 is one of the master switches that makes a blood cell myeloid. Cells that have taken that path need it. Our disease's cell lines are mostly myeloid, so PU.1 looked like a dependency of the disease when it is a dependency of the cell type.
SS18 - a SWI/SNF subunit that is itself a fusion partner in another cancer, and a lineage-independent dependency in KMT2A-rearranged lines
SS18 is a small subunit shared by two flavours of the SWI/SNF machine. It is also the gene that fuses to SSX in synovial sarcoma, producing a fusion protein that lives inside the machine itself - which is the cleanest example anywhere of a fusion oncoprotein hijacking chromatin remodelling, and a useful contrast with how KMT2A fusions work.
SSRP1 - the small FACT subunit, and the only protein in this beat that binds a confirmed KMT2A-fusion interactor directly (LEDGF/p75, via HMG box to PWWP)
SSRP1 is the small half of FACT, the machine that unpacks and repacks nucleosomes so RNA polymerase can read through them. Its interest for this project is one specific contact: SSRP1 grips LEDGF/p75, a protein that is itself a documented partner of the leukaemia fusion, which makes SSRP1 exactly two handshakes from the fusion complex even though no paper has ever put them in the same experiment.
SUPT16H (SPT16) - the large FACT subunit: pan-essential, an anti-dependency in KMT2A-rearranged lines, and drugged in MLL-r leukaemia by a compound that works through p53
SPT16 is the big half of FACT, a two-protein machine that takes nucleosomes apart in front of a moving RNA polymerase and puts them back behind it. Every cell needs it, KMT2A-rearranged leukaemia cells need it slightly less than other leukaemia cells, and the one drug that hits it in this disease kills cells by trapping it on DNA rather than by blocking the fusion.
SUPT4H1 (SPT4) — zero papers linking it to KMT2A, and the 138th most KMT2A-selective gene out of 17,728
SPT4 is the small half of DSIF, the two-protein clamp on RNA polymerase. Nobody has ever published anything connecting it to KMT2A leukaemia — yet in an unbiased scan of every human gene, KMT2A-rearranged leukaemia cells need it more than other leukaemia cells do, and it ranks in the top 1%.
SUPT5H (SPT5) — the pausing factor the KMT2A field has never studied, and the counted absence that undercuts 'the fusion hijacks pause release'
SPT5 is half of DSIF, the clamp that holds RNA polymerase still just after it starts a gene, and then rides along with it once the brake is released. If a KMT2A fusion works by releasing that brake, you would expect a large literature connecting the two — and there is essentially none.
TASP1 (taspase1) - the protease that cuts KMT2A, and the timer the fusion escapes
Taspase1 is the enzyme that cuts normal KMT2A in half. The leukemic fusion protein is missing the places where the cut happens, so it is never cut - and that turns out to be one reason it lasts so long in the cell.
TCF3 (E2A) — domain architecture and residue coordinates
TCF3 is a switch-on protein for developing B cells: one end shouts at the cell's transcription machinery, the other end grips DNA. In one childhood leukaemia the DNA-gripping end is cut off and replaced by a different protein's DNA-gripping end, so the shouting is aimed at entirely the wrong genes.
TET1 (LCX / CXXC6) - a CXXC reader that is also a KMT2A fusion partner, and whose leukaemic fusion throws its own CXXC domain away
TET1 reads the same unmethylated CG addresses that the KMT2A fusion reads, and it erases the chemical tags that block those addresses. It was discovered in the first place because it gets fused to KMT2A in a rare leukaemia - and in that fusion, TET1's own address-reading module is the part that is left behind.
TET2 is NOT a CXXC protein - its CXXC domain is a separate gene, and it is DEPLETED in KMT2A-rearranged AML rather than enriched
TET2 is the gene people reach for when they want to argue that a leukaemia's DNA-methylation landscape has been altered by a second mutation. But TET2 does not actually carry the CG-reading module that the KMT2A fusion uses - that module was split off onto a different gene long ago - and in two large adult studies TET2 mutations are LESS common in KMT2A-rearranged leukaemia than in other kinds.
TET3 - the CXXC reader whose domain prefers 5-carboxylcytosine, and the only CXXC-family gene ever found mutated in a KMT2A-rearranged cohort
TET3 carries the same kind of CG-reading module as the KMT2A fusion, but it does not read the same letter: its favourite target is a chemically modified cytosine, not a plain unmodified one. It is also the only member of this whole protein family that has ever turned up mutated in a study of KMT2A-rearranged leukaemia - and that was in older children, not infants.
The 7SK snRNP - a stable RNA core with a reversibly docked kinase, which is why the reservoir is a switch and not a sink
The 7SK snRNP is a small particle built around an RNA that holds the elongation enzyme P-TEFb switched off until the cell releases it. Its inner parts are permanently bolted together and the enzyme is the part that comes and goes, which is what makes it a usable switch rather than a dead end.
The ELL/EAF activation module (ELL2, EAF1, EAF2) — a general elongation stimulator, not a KMT2A-fusion dependency
ELL2 and its partners EAF1 and EAF2 form the part of the super elongation complex that actually makes RNA polymerase go faster, by gripping the polymerase in the same spot as a general start-up factor. But leukemia cells with a KMT2A fusion need these three proteins slightly less than other leukemia cells do, so this module looks like general machinery rather than a target.
The HUCA complex (HIRA/UBN1/CABIN1/ASF1a) - four proteins, one job: put H3.3 into chromatin without waiting for S phase
Four proteins work together as one machine that inserts a replacement histone into genes that are actively being read, at any point in the cell cycle. Each part does something different, and knowing which part does what tells you which one you could interfere with.
The IKK complex — the unit the eviction experiment actually resolves, because the chemistry that produced the result cannot see the subunits
The three IKK proteins work as one machine, and the drugs used to show that MLL leukaemia needs them block the machine rather than any one part. So when a paper says a subunit is required, check whether the experiment could have told the parts apart.
The KMT2A::MLLT1 chimeric protein: exactly what survives the join on each side, in residue numbers
The fusion protein is the front third of KMT2A welded onto most of ENL, and the weld point decides which working parts each parent contributes. KMT2A gives up every one of its own chromatin-writing and chromatin-reading modules and keeps only its targeting end; ENL gives up almost nothing.
The KMT2A::MLLT3 chimeric protein: what survives the join on each side, in residue numbers - and the four AF9 modules that do not
This cancer protein is the front third of KMT2A welded onto only the last fifth of AF9. Everything AF9 uses to find its way around a cell - its chromatin reader, its self-sticking region and even its nuclear address label - sits on the discarded side of the weld.
The little elongation complex (ELL + ICE1 + ICE2 + ZC3H8) — the other place ELL works, and why ELL's pan-essentiality is probably not about the fusion
ELL, one of the proteins in the machine the KMT2A fusion hijacks, also has a second job in a completely different complex that makes small nuclear RNAs. That second job is probably why every cell needs ELL, which means ELL's essentiality is not evidence that it matters to the fusion.
The NuRD complex - two separable modules, three paralogue choices that change what it does, and a graph that carried its ATPase and its shared core but never the scaffolds that make it NuRD
NuRD is the only chromatin machine that both slides nucleosomes and strips acetyl marks off histones, and it is built from two halves that work on their own. Which version of three swappable parts it contains decides whether it reads methylated DNA, which enzyme it clamps and which gene it silences - so 'NuRD' on its own does not name one object.
The replication protein A heterotrimer (RPA1/RPA2/RPA3) - the ssDNA clamp that every resection and replication-stress claim in this corpus silently depends on, and which had no entry until cycle 160
Whenever DNA is unzipped - to copy it, or to repair a break - the exposed single strand has to be covered immediately or it will fold, degrade or recombine wrongly. RPA is the three-part protein that does the covering, and it is also the flag that tells the cell's alarm kinase that exposed DNA exists.
TIFAB — a 161-residue NF-kB adaptor-layer protein with TWO direct KMT2A-fusion papers, which is more disease-specific evidence than any receptor or TRAF in this beat has
TIFAB is a small protein that dampens the NF-kB alarm signal, and in mouse MLL leukaemia it makes the disease worse rather than better. It is one of very few upstream signalling proteins with experiments done in this exact leukaemia, and it has the unusual distinction of being tested in two different KMT2A fusions.
TNF — the only NF-kB-activating ligand with real KMT2A::AFF1 B-ALL evidence, and the evidence says the blasts are RESISTANT to it
TNF is the classic alarm signal that switches on NF-kB, and it is also one way donor immune cells kill leukaemia after a transplant. Infant-type MLL-AF4 leukaemia cells are unusually resistant to being killed by it, which is a proposed reason transplants work less well in these children.
TOP1 domain architecture (UniProt P11387)
TOP1 is the enzyme partner in the NUP98::TOP1 fusion. Its business end is a single large catalytic region in the back half of the protein, and the fusion keeps all of it, including the exact amino acid that cuts DNA.
TRAF2 - a cytoplasmic NF-kB brake reported as an AFF4 two-hybrid hit, 97 partners in the screen that found it, and the live instance of the brief's own accession trap
TRAF2 is a cytoplasmic signalling protein that helps keep one branch of the NF-kB pathway switched off. It was listed as a partner of the nuclear elongation scaffold AFF4 on the strength of a single yeast screen, in which TRAF2 also bound ninety-six other things.
TRIM27 - the E3 ligase inside PRC1.1 that USP7 spends its time undoing, and the worst UniProt localisation record this graph has audited
PRC1.1 contains a pair of enzymes that work against each other: TRIM27 tags parts of the complex for destruction and USP7 removes those tags. Blocking USP7 lets TRIM27 win, and the complex falls apart - which is the only way anyone currently has to switch PRC1.1 off with a drug.
TRRAP — a 3,859-residue DotCom subunit that every cell needs and no KMT2A cell needs more
TRRAP is an enormous scaffolding protein found inside the H3K79 methylation complex, and also inside several other complexes that acetylate histones. Every cancer cell line tested dies without it, which makes it a very poor candidate for a leukemia-selective drug.
UBE2A (RAD6A) — the E2 for H2B ubiquitination in one primary account, contradicted by another, and with no KMT2A literature at all
Before histone H2B can be tagged, a small carrier enzyme has to hand the ubiquitin over; UBE2A is the carrier one major paper names, and a different major paper names a different enzyme instead. Nobody has ever tested either of them in KMT2A-rearranged leukaemia, and the cell-line dependency data say neither is needed for cells to survive.
UBE2T — E2; rank 33 of 8,269 for KMT2A-selective dependency, and no KMT2A literature at all
UBE2T is part of a cellular repair crew that fixes DNA when the copying machinery gets stuck. A screen run by this project found that leukaemia cells carrying the KMT2A fusion need it more than other leukaemia cells do — and nobody has ever published a paper on this gene in KMT2A leukaemia, so the finding is genuinely new and genuinely unconfirmed.
USP2 (ubiquitin carboxyl-terminal hydrolase 2) - the ninth commonest KMT2A partner and the only one that is a deubiquitinase
USP2 is an enzyme that removes ubiquitin, the small tag cells use to mark proteins for destruction, so USP2 makes its targets last longer. It is the ninth most common gene fused to KMT2A in leukaemia, and it is not the same gene as USP22, which this graph covers separately.
USP22 — a KMT2A-rearranged dependency that works through MEIS1 protein stability, and is not an interactor of the fusion
USP22 is an enzyme that removes the disposal tag from MEIS1, one of the main proteins the KMT2A fusion switches on, so MEIS1 survives instead of being destroyed. Blocking USP22 lets the cell throw MEIS1 away, and leukemia cells with a KMT2A fusion stop growing — but USP22 never touches the fusion itself.
USP7 - the enzyme that holds PRC1.1 together, and the only pharmacological handle on this complex that exists today
PRC1.1 is the silencing machine at the centre of this beat and there is no drug that hits it directly. USP7 is a deubiquitinating enzyme that keeps PRC1.1 assembled, so blocking USP7 makes the complex fall apart - which is, at present, the only chemical way anyone has to turn PRC1.1 down, and it comes with a large warning attached.
USP8 (UBPY) - the second deubiquitinase KMT2A partner, three patients, same UCH logic
USP8 is another enzyme that strips the destruction tag off proteins, and like USP2 it turns up fused to KMT2A in a handful of children with leukaemia. It matters because it is the second, independent example of the same pattern: a rare fusion that keeps KMT2A's chromatin fingers and adds a tag-removing enzyme.
Vimentin domain coordinates (P08670) - so the VIM::KMT2A breakpoint can actually be drawn
This is just the map of vimentin's parts by residue number, fetched from the protein database. It exists so that the picture of the VIM-KMT2A sarcoma fusion can be drawn with real coordinates instead of words.
WAC — the adaptor that hands RNF20/RNF40 to RNA polymerase II, and the only member of this axis whose knockdown alone killed MLL-rearranged ALL cells
WAC is the adaptor that grabs RNA polymerase with one end and the H2B-tagging enzyme with the other, so the ubiquitin tag only goes on genes that are actually being read. In MLL-rearranged infant leukaemia cells, removing WAC alone stripped the tag off and killed the cells.
WDR5 — the partner that is not a partner: it binds the half of KMT2A the fusion throws away
WDR5 is always drawn in pictures of the KMT2A complex, and there are sixty structures of it. But it grips a piece of KMT2A near the far end of the protein — a piece that every fusion deletes — and KMT2A-rearranged leukemia cells actually need WDR5 less than other leukemia cells do.
ZEB2 — an EMT transcription factor that turns out to be a leukemia dependency and a guardian of lineage fidelity
ZEB2 is best known for helping solid-tumour cells change shape and spread, but in blood it does something else: it stops immature blood cells from switching on the programs of the mature cell types they have not become yet. A scan run by this project found KMT2A-fusion leukemia cells need it, and independent laboratories had already shown that leukemia cells of several kinds do.
ZFP64 — rank 4 on selectivity, and there is not even a ChEMBL entry for it
ZFP64 is a zinc-finger protein that sits on the KMT2A gene's own promoter and keeps KMT2A switched on. Cells in general do not need it, KMT2A-rearranged leukemia cells do, and no drug programme anywhere is working on it.
6-Mercaptopurine - two years of a nightly tablet whose lethal dose is a genotype, given as a liquid to a baby who cannot swallow it and must not have it with milk
Mercaptopurine is the nightly tablet that runs for most of the two years of treatment. A small fraction of children carry a gene variant that makes the normal dose overwhelmingly toxic, and the practical problems in a baby are as much about how you get an accurate small dose into her as about the drug itself.
6-Thioguanine - the drug a 1,498-child randomised trial removed from maintenance and left in intensification, which is exactly where Interfant-06 uses it
Thioguanine is mercaptopurine's more potent sibling. A large randomised trial found it caused serious, sometimes permanent liver damage in about one child in nine when used long-term, so it was dropped from the two-year maintenance phase - but kept for the short intensive blocks, and that is how the infant protocol uses it.
Anti-thymocyte globulin - a rabbit antibody against human T cells, given to infants at a fixed mg/kg dose against a T-cell mass that nobody measures
This is an antibody made in rabbits that destroys human T cells. It is given in the three days before a transplant from anyone other than a matched brother or sister, to stop the child rejecting the graft and to reduce the donor cells attacking the child. The dose is set by the child's weight, but what it has to destroy is the child's T cells - and those do not scale with weight.
Azacitidine - the one hypomethylating agent actually given to infants with KMT2A-rearranged ALL: safe, demonstrably demethylating, and it did not change survival
Azacitidine strips methyl marks off DNA and is used in adult blood cancers. It is the only drug of its kind that has been given to infants with KMT2A-rearranged leukaemia in a trial: it was safe and it did strip the marks, and it made no difference to how long the babies lived.
Bleximenib (JNJ-75276617) — menin inhibitor with a cleaner cardiac profile, and the only one with a 2-year-old floor
Bleximenib is an experimental menin-blocking pill from Johnson & Johnson. It is interesting for two reasons: in early trials it did not prolong the heart's QT interval the way the approved drug does, and its first-in-human trial has a paediatric cohort open from age two — though that cohort sits in the phase 1 part, while both phase 3 studies that could support approval enrol adults only, so children will again get the drug after adults do.
Blinatumomab in infant KMT2A-rearranged ALL — the largest single outcome gain in thirty years
Blinatumomab is an antibody with two grabbing ends: one holds a leukemia cell by its CD19 marker, the other drags a T cell over to kill it. Adding one course of it to standard infant chemotherapy roughly doubled four-year survival, which is the biggest improvement this disease has seen in decades.
Busulfan - the alkylating agent that replaces radiation in infant conditioning, and whose labelled paediatric target range is below the range the modern trial data support
Busulfan is the chemotherapy drug used instead of whole-body radiation to empty a baby's bone marrow before a transplant, because radiation damages a growing brain and body. It is very hard to dose in a baby: too little and the leukaemia comes back, too much and the liver is injured, so the blood level has to be measured during treatment and the dose corrected.
Chidamide (tucidinostat) - the HDAC inhibitor in both of this graph's menin-inhibitor combination papers, and the only one paired with a menin inhibitor in the lymphoid index disease
Chidamide is an HDAC-blocking drug approved in China and Japan. It is the one drug in this area that shows up twice alongside menin inhibitors - the class this whole project is built around - once in myeloid leukaemia and once in the infant t(4;11) leukaemia that is the project's central disease. That makes it the most obvious partner drug in the field, and almost nobody outside China has noticed.
Cyclophosphamide - where the KMT2A literature reports two different activated oxazaphosphorines and only one of them is this drug's metabolite
Cyclophosphamide is an alkylating drug: it sticks DNA strands together so the cell cannot copy them. It needs the liver to switch it on, so laboratory studies use a pre-activated version - and two different pre-activated cousins appear in this disease's literature, only one of which is really this drug.
Cytarabine - the one conventional drug KMT2A-rearranged infant ALL is MORE sensitive to, and the only case in this beat where a mechanism changed a protocol
Cytarabine is the one ordinary chemotherapy drug that infant leukaemia cells are unusually good at taking up and being killed by - the opposite of the steroid story. That laboratory finding is the reason cytarabine appears so heavily in the protocols these babies are actually given.
Daunorubicin - the anthracycline an infant gets before her first birthday, with a measured null in vitro and sixty years of heart downstream
Daunorubicin is the red drug given on days 8 and 9 of induction and again later. Infant leukaemia cells are no more sensitive to it than any other child's, and every milligram is a permanent withdrawal from a heart that has to last sixty years.
Decitabine - degrades DNMT1, and in ALL cells it degrades KMT2A too: the hypomethylating agent with two mechanisms and a menin-inhibitor combination already in the clinic
Decitabine is an old, approved leukaemia drug that gets built into DNA and traps the enzyme that copies methyl marks, destroying it. A 2025 paper found it also destroys KMT2A protein itself in leukaemia cells - so the drug that strips methylation may be hitting the fusion's own parent gene product, and it works together with menin inhibitors.
Dexamethasone - the steroid an infant is actually on for most of the protocol, and the one with a high-certainty neurodevelopmental signal in babies
Dexamethasone is the steroid that takes over from prednisone on day 8 and runs for most of induction and again in the reinduction block. In premature babies given it for lung disease, randomised trials show it raises the risk of cerebral palsy - which is a different situation, but it is the only high-quality evidence about what this drug does to a developing infant brain.
DOT1L808 - a 5 nM DOT1L degrader that cleared an orthotopic leukaemia and still works after menin inhibitors fail
This is a molecule that does not block DOT1L but destroys it, tagging it for the cell's own disposal machinery. In mice with leukaemia it cleared the disease completely, and its sister compound still killed cells that had already become resistant to menin inhibitors.
Doxorubicin - named in twenty entries of this graph, cuts the KMT2A breakpoint where etoposide does, and appears nowhere in Interfant-06
Doxorubicin is the anthracycline everyone means when they say 'anthracycline', and it is the yardstick every other one is measured against for heart damage. It is not in the infant leukaemia protocol at all - but it cuts DNA at exactly the spot where this disease's translocation happens.
Ebselen: the only small molecule reported to bind LEDGF itself - and it does it by making selenium-sulphur bonds with whatever thiols it meets
Only one small molecule has ever been reported to stick to LEDGF directly. It works by forming bonds with sulphur atoms in the protein, which it will also do to many other proteins, so it is a starting clue rather than a drug.
Eltanexor (KPT-8602) - the second-generation SINE whose whole point is that you can give it more often
Eltanexor blocks the same nuclear export protein as selinexor but crosses much less into the brain, so it can be given far more frequently. That turns out to matter: in leukaemia models the difference between a weekly dose and a near-daily one is the difference between a temporary effect and a permanent one.
Enasidenib - an approved mutant-IDH2 inhibitor sitting next to the one methylome co-mutation that IS enriched in KMT2A-rearranged AML, and never tested there
Enasidenib blocks the mutant form of IDH2, the enzyme whose faulty product drives DNA over-methylation, and it has been approved for adults with IDH2-mutant leukaemia since 2017. IDH2 mutations turn up more often in KMT2A-rearranged AML than in other AML - and nobody has ever reported giving this drug to that group specifically.
Enzomenib (DSP-5336) — menin inhibitor with the highest reported KMT2A-r response rate and little differentiation syndrome
Enzomenib is an experimental menin-blocking pill from Sumitomo Pharma. In early trials it produced the highest response rate so far in KMT2A-rearranged leukaemia while causing much less of the dangerous differentiation reaction — but the numbers come from small, early-phase cohorts, not a full trial.
Etoposide (VP-16) - the drug that makes this disease and cures several others, and the only entry in this graph whose main finding is that you must not stop using it
Etoposide jams the enzyme that untangles DNA while it is mid-cut, so the cut never gets sewn up. That is how it kills cancer cells, and it is also the best-understood reason a child cured of one cancer can develop a KMT2A-rearranged leukaemia years later - which is why the answer is careful dosing and not avoidance.
FHD-286 - the dual BRG1/BRM inhibitor that carries every SWI/SNF therapeutic claim in this disease, with a minimum enrolment age of 16
FHD-286 is an oral drug that blocks both motors of the SWI/SNF machine. In the laboratory it kills MLL-rearranged leukaemia cells, works together with menin inhibitors, and is in a phase 1 trial in adults with relapsed leukaemia - but no trial of it admits children under sixteen.
Fludarabine - the immunosuppressive half of modern conditioning, given as a phosphate prodrug, cleared by the kidney, and dosed by a body-surface-area rule that measurably fails in children
Fludarabine is not there to kill the leukaemia; it is there to suppress the child's own immune system so the donor's cells are not rejected. The dose is worked out from body size, and in children that rule is measurably wrong often enough that dosing by measured blood levels instead cuts the graft-rejection rate from 8.5% to 1.8%.
I3IN-002 — the first small-molecule IGF2BP3 inhibitor with in vivo activity in an MLL-Af4 model
I3IN-002 is the first drug-like molecule made to block IGF2BP3 from gripping its target RNAs. It slowed leukaemia in mice carrying the MLL-Af4 fusion and was tolerated, which makes it a starting point rather than a medicine.
Icovamenib (BMF-219) — the covalent menin inhibitor that left oncology for diabetes
Icovamenib is a menin blocker that latches on permanently (covalently) rather than reversibly. Its leukaemia trials have been stopped, and the company has redirected the drug to type 1 and type 2 diabetes — because blocking menin makes insulin-producing cells grow, which is exactly the kind of side effect that should worry anyone planning years of menin blockade in a child.
Ivosidenib - the approved IDH1 inhibitor, recorded here so that nobody reaches for it: IDH1 is the paralogue that is NOT enriched in KMT2A-rearranged AML
Ivosidenib blocks mutant IDH1 and produces durable remissions in adults whose leukaemia carries that mutation. It is in this graph mainly as a warning sign: it is the wrong twin. IDH2 is the one that is more common in KMT2A-rearranged leukaemia, and ivosidenib does not touch IDH2.
JD-PI95 - a genetically encoded bridging peptide that drags AKAP95 to HSP70 and dissolves its condensates, with in vivo activity in THP-1
Rather than blocking AKAP95 with a drug, the authors built a short protein that grabs AKAP95 with one end and the cell's own protein-disassembly machine with the other. Forced together, the machine takes AKAP95's droplets apart, and the leukaemia cells grow more slowly.
JDI-12 - published as a JMJD1C inhibitor that preferentially kills KMT2A-rearranged leukaemia and spares cord blood, then used as 'KDM3i' to CREATE KMT2A rearrangements in cord-blood progenitors, by two papers that do not cite each other
JDI-12 is a laboratory compound found by computer screening against a DNA-tag-removing enzyme. One paper says it kills KMT2A-fusion leukaemia cells while leaving healthy cord blood stem cells alone; another paper uses the same compound to make those same cord blood stem cells start copying and breaking the KMT2A gene.
M-808 — a covalent menin-MLL inhibitor with a co-crystal, five years before the graph noticed it
M-808 is a laboratory compound that sticks permanently to menin rather than binding and letting go. It shrank tumours in mice and was never developed.
Melphalan - the third agent in Interfant-06 conditioning, dosed per square metre in babies who are dosed per kilogram for everything else in the same week
Melphalan is the last drug given before the donor cells go in, on the day before transplant. Its main injury is to the lining of the mouth and gut, and there is evidence that giving too much of it makes graft-versus-host disease more likely.
Methotrexate - 5 g/m2 through kidneys that are still maturing, and the one backbone drug the canonical infant sensitivity study never tested
Methotrexate blocks the folate chemistry cells need to make DNA. Infants get it in three completely different ways - a huge intravenous dose with a rescue drug, small weekly tablets for two years, and injections into the spinal fluid - and the huge dose leaves the body through kidneys that are only just finishing their development.
Mitoxantrone - the Interfant-06 protocol says it is less cardiotoxic than daunorubicin, and the largest survivor cohort ever assembled says it is about seventeen times worse per milligram
Mitoxantrone is a close cousin of the red anthracyclines, used in the experimental arm of the infant protocol and in relapse. The protocol tells doctors it is gentler on the heart than daunorubicin; the biggest long-term survivor study ever done found the opposite, by a very large margin.
MS2133 - the first molecule to degrade DOT1L rather than inhibit it
MS2133 is the first reported molecule that hands DOT1L to the cell's protein-shredder instead of blocking its active site. It is a laboratory tool rather than a medicine, but it is the reagent that lets anyone ask whether removing DOT1L entirely does more than switching its chemistry off.
Native E. coli asparaginase - the second-largest in-vitro resistance signal in infant ALL, and a different product from the pegylated one four doses later
Asparaginase is a bacterial enzyme, not a chemical, that destroys an amino acid leukaemia cells cannot make for themselves. Infant leukaemia cells are twelve times more resistant to it than older children's - the second-biggest difference of any drug measured.
Novartis 'compound 10' - a DOT1L methyltransferase inhibitor from Perner et al. (Blood 2020), active at 100 nM, whose chemical structure is not given in the GEO deposit and whose target engagement at that dose IS demonstrated in the same series by a 5.6-fold spike-in-normalised H3K79me2 collapse by day 3
One of two new experimental drugs that block the DOT1L enzyme, made by Novartis and tested in this study. It works at a much lower dose than the older drug of the same kind.
Novartis 'compound 11' - a DOT1L methyltransferase inhibitor from Perner et al. (Blood 2020), active at 100 nM, whose chemical structure is not given in the GEO deposit and whose target engagement is not measured anywhere in the series
One of two new experimental drugs that block the DOT1L enzyme, made by Novartis and tested in this study. It works at a much lower dose than the older drug of the same kind.
Panobinostat - the HDAC inhibitor this graph has reasoned about for cycles without having an entry for, with strong KMT2A-rearranged xenograft activity and a terminated adult trial
Panobinostat is an approved drug that blocks histone-deacetylase enzymes very broadly. In mice carrying human MLL-rearranged leukaemia it works well on its own, and it reaches a chemical pathway no other clinical drug touches - but the one trial that asked whether it works in adults with relapsed acute leukaemia was stopped early because it did not.
Pegaspargase - the same enzyme wearing a polymer, dosed in units that are not the other product's units, and now dosed to a level in every group except infants
Pegaspargase is the same bacterial enzyme as plain asparaginase with a polymer coat that makes it last far longer, so one dose replaces many. It is a different product with different dosing numbers, and older children are now dosed by blood-level monitoring while infants are not.
Pinometostat (EPZ-5676) — the DOT1L inhibitor, and the most instructive failure in this field
Pinometostat blocks DOT1L, an enzyme the KMT2A fusion hijacks to keep cancer genes switched on. It worked beautifully in the laboratory and almost not at all in patients, and understanding why is the single best warning we have about menin inhibitors.
Plerixafor (AMD3100, Mozobil) - a licensed CXCR4 antagonist that has never been given to an infant with leukaemia, and whose only paediatric efficacy study failed
Plerixafor blocks the receptor leukaemia cells use to stay anchored in the bone marrow, and it is already an approved drug for pushing stem cells into the blood before a transplant. In mice carrying infant MLL-rearranged leukaemia it made a targeted drug work much better, but it has never been tested in an infant with leukaemia and its one paediatric efficacy trial did not work.
Prednisolone - the active steroid, the drug every resistance experiment in this disease actually used, and the one given into the spinal fluid
Prednisolone is the form of the steroid that actually works - prednisone has to be turned into it by the liver first. Almost everything we know about why infant leukaemia cells survive steroids was measured on prednisolone in a dish, and in the Interfant protocol prednisolone is also the steroid injected into the spinal fluid.
Prednisone - the prophase drug, the prognostic factor, and a prodrug that has never itself been tested on a KMT2A-rearranged blast
Prednisone is the steroid every infant with leukaemia gets first, alone, for seven days, and how fast the leukaemia melts in that week is one of the strongest predictors of whether the child will be cured. It is also a prodrug: the liver has to turn it into prednisolone before it does anything at all.
Revumenib (SNDX-5613, Revuforj) — first approved menin–KMT2A interface inhibitor
Revumenib is a pill that pries apart the handshake between the KMT2A fusion protein and a partner protein called menin, which the leukaemia needs in order to stay immature. It became the first drug of its kind approved by the FDA, in November 2024, for children over one year old and adults whose KMT2A-rearranged leukaemia has come back.
Romidepsin - the HDAC inhibitor with the best preclinical case in KMT2A-rearranged INFANT ALL specifically, and a mechanism that is not about histones at all
Romidepsin is an approved HDAC-blocking drug that came out top in an unbiased drug screen against cell lines made directly from infants with MLL-rearranged leukaemia. What it does there is unexpected: it makes an existing chemotherapy drug, cytarabine, work better, by switching off the enzyme that destroys cytarabine inside the cell.
Selinexor (XPO1 inhibition) — a real paediatric response rate attached to an unproven mechanism in KMT2A-rearranged disease
Selinexor jams the doorway cancer cells use to push tumour-suppressor proteins out of the nucleus, trapping them where they can work. In children with relapsed leukemia it produced remissions in nearly half of those treated, but at doses that caused reversible brain toxicity.
SGC-iMLLT - the MLLT1/MLLT3 YEATS chemical probe, and the 2026 paper that put it into KMT2A::AFF1 in vivo with a healthy-HSPC comparison
SGC-iMLLT is a freely available research compound that blocks a small reading head on two proteins, ENL and AF9, which help leukemia cells keep the wrong genes switched on. A 2026 study showed it slows the growth of the specific infant-leukemia fusion this project is built around, in animals, without damaging healthy blood stem cells.
Silmitasertib (CX-4945) - a CKII inhibitor that works on KMT2A-rearranged leukemia by strengthening the allele the fusion does not have
This drug blocks a kinase that has nothing to do with the leukemic fusion protein. It works - in mice - by making the healthy copy of KMT2A last longer, and the extra healthy protein crowds the leukemic one off the genes it was switching on.
Tazemetostat - an approved EZH2 inhibitor, and the drug this disease's own biology predicts will not work
Tazemetostat is an approved pill that blocks EZH2, the enzyme that writes the main 'switch off' tag on chromosomes. It works in a lymphoma where EZH2 is overactive - but in KMT2A leukaemia the back-up enzyme EZH1 takes over, and a head-to-head test in leukaemia cells showed a drug that blocks both worked better.
Thiotepa - the third agent that reaches the central nervous system completely, and whose alkylating metabolites are excreted through the skin into a nappy
Thiotepa is added to transplant conditioning because it gets into the fluid around the brain and spinal cord as completely as into the blood, which matters in a leukaemia that often reaches there. It and its active breakdown product also come out through sweat, so in a baby it collects under a nappy and burns the skin - which is why children on this drug are bathed several times a day.
Treosulfan - a European conditioning standard for three decades that the FDA first approved in January 2025, for a diagnosis and an age that exclude the infant with KMT2A-rearranged ALL
Treosulfan is a close chemical relative of busulfan that works the same way but is gentler on the liver, and European transplant units have used it in children for years. The American regulator only approved it in January 2025, and only for a different leukaemia and only from one year of age - so a six-month-old with this disease is outside its licence on both counts.
UNC0642 - the G9a/GLP chemical probe that abolished doxorubicin-induced Kmt2a copy gain in mice, and the one intervention in this beat that has been given to an animal
UNC0642 is a laboratory compound that switches off the enzyme which adds the silencing tag to the KMT2A gene, and it is good enough to give to a mouse. Mice given it before doxorubicin did not develop the extra copies of Kmt2a that doxorubicin otherwise produces.
Valemetostat - an approved dual EZH1/EZH2 inhibitor that has already worked in an MLL-AF4 xenograft, in a disease nobody has enrolled it in
Valemetostat blocks both EZH2 and its back-up EZH1, which is the combination this leukaemia's biology says is needed. It is already approved in Japan for a different blood cancer, and in the laboratory it shrank a mouse tumour carrying the exact fusion that causes infant leukaemia - but it has never been given to a patient with that disease.
Venetoclax in KMT2A-rearranged leukemia — the apoptotic partner every other drug needs
Venetoclax removes the brake that stops cancer cells from killing themselves. On its own it rarely cures anything, but added to almost any other drug in KMT2A-rearranged leukemia it converts a growth-arresting effect into actual cell death.
Vincristine - measured to be no different in KMT2A-rearranged disease, and measured to be UNDERDOSED in infants by the very rule meant to protect them
Vincristine stops cells dividing by wrecking the scaffolding that pulls chromosomes apart. Infant leukaemia cells are no more or less sensitive to it than any other child's - but infants are given a reduced dose on the assumption that they clear it slowly, and when someone finally measured, they did not.
Vorinostat (SAHA) - the only HDAC inhibitor given to newly diagnosed infants with ALL, in a 50-patient trial whose endpoint was tolerability and which reported four treatment-related deaths, all infectious
Vorinostat is an approved drug that blocks the enzymes which strip acetyl tags off histones. It has been given to fifty newly diagnosed infants with acute lymphoblastic leukaemia, alongside bortezomib and standard chemotherapy - the only HDAC blocker ever taken that far in this exact population. The trial was designed to ask whether infants could tolerate it, not whether it worked, and four children died of infection during treatment.
VTP-50469 - the preclinical menin-KMT2A interface inhibitor that almost every chromatin experiment in this field actually used, and which had no entry of its own in this graph
This is the laboratory version of the menin-blocking drug. Nearly every experiment people cite about what menin inhibitors do to chromatin was done with this compound, not with the approved medicine.
Ziftomenib (KO-539, Komzifti) — once-daily menin inhibitor approved in NPM1-mutant AML
Ziftomenib is a second menin-blocking pill, taken once a day, approved by the FDA in November 2025 for adults whose NPM1-mutated acute myeloid leukaemia has returned. Its KMT2A-rearranged arm had to be paused early because too many patients developed a dangerous reaction called differentiation syndrome.
Resistance and toxicity, recorded as first-class entries rather than footnotes.
A cell that cannot deposit H2AK119ub is still evicted normally by a menin inhibitor and silences its vacated sites less - so a fusion-occupancy pharmacodynamic marker reads FULL ENGAGEMENT in exactly the genotype that responds worst, and this is now the fourth independent reason this corpus distrusts its own candidate markers
If a leukaemia cell has lost part of its silencing machinery, the drug still does its first job - pulling the fusion off the DNA - but the second job, switching those places off, works less well. A test that only checks the first job would say the drug is working perfectly.
A gene-set permutation is anti-conservative for a protein complex, and the fix is to permute the other margin
Checking whether a group of genes stands out by comparing it against random groups is only fair if the genes are independent, and the proteins in a machine are not. The fix is to shuffle which cell lines you call the test group instead of which genes, and when that fix is applied here the result survives.
A genome-wide peak-call census is the wrong pharmacodynamic assay for a menin inhibitor, in two independent ways
If you measure whether a menin drug is working by counting how many menin spots are still detectable across the whole genome, you will conclude it is not working even when it is - twice over: most of the genome is irrelevant, and the counting method cannot see the size of change the drug produces.
A menin degrader removes six surfaces; a pocket inhibitor blocks one and touches a second - and in a child receiving TOP2 poisons that difference is a repair-competence difference nobody has measured
Menin does several separate jobs, and one of them is helping cells repair broken DNA accurately. A drug that blocks one pocket on menin leaves most of the other jobs alone; a drug that destroys the whole protein removes all of them, and children with this leukaemia are also given drugs that deliberately break DNA.
A PCGF1-null cell - a published resistance genotype - loses H3K79me2 at the most-evicted sites exactly as a control cell does (0.6123 against 0.5860), so the new stratified readout scores it as fully drug-engaged, and stratification does NOT close the resistance blind spot it was hoped to close
A cell engineered to carry a change that should make it resist the drug still loses the activating tag just as completely as a normal cell where the drug works. A test based on that tag would report the drug as working perfectly in a cell that is not responding.
A target map read off the diagnostic sample does not describe the same child's disease after a lineage switch
If doctors profile a child's leukaemia at diagnosis and keep that list of target genes for later, the list will be about half wrong once the disease changes type - no better than a list borrowed from a different child.
Adaptive, non-genetic resistance — the leukemia keeps the fusion and rewires how it is maintained
Some leukemias stop responding to a drug without changing their DNA at all. They simply find another way to keep the same cancer genes switched on, which means sequencing the tumour will show nothing wrong and the drug will still fail.
An NG2-directed therapy meets pericytes - and unlike B-cell aplasia there is no replacement therapy for a damaged vascular wall
NG2 sits on the cells that wrap around small blood vessels and hold them together. A therapy that kills every NG2-bearing cell would damage those, and there is no equivalent of the antibody infusions that make the loss of normal B cells survivable.
An occupancy readout returned background in every channel of the only drug-exposed patient sample it has been tried in - so a negative from such an assay cannot currently be told apart from an assay failure
If doctors ever want to check whether a menin-blocking drug is working by looking at where menin sits on a patient's DNA, they should know that the one time this was tried, the test came back blank for everything it measured - including the things that should have shown up.
Anthracycline cardiotoxicity: the heart an infant keeps for sixty years
The drugs that cure infant leukemia include ones that permanently damage heart muscle, and a baby's heart is more vulnerable than an older child's. The damage does not show up at the end of treatment; it shows up twenty and thirty years later, which is why a drug that avoids these agents would be worth a great deal even if it only matched current cure rates.
Bite-back: a DOT1L inhibitor in CALM-AF10 deepens a genome-wide H3K79 deficit that already causes chromosome breaks
CALM-AF10 cells already have far too little of the DOT1L chemical tag across most of their genome, and that shortage makes their chromosomes break. A drug that blocks DOT1L removes the small amount that is in the wrong place, and also removes what little is left everywhere else.
Bite-back: a target nominated from an imported mechanism inherits a fusion the index disease does not have
If a drug target is chosen because of how a mechanism works in MLL-AF9 leukaemia, and the infant fusion does not have the protein part that mechanism runs through, the target can be perfectly real and still be the wrong one for the disease this project is about.
Blocking IKK/NF-kB in an infant is removing the switch every immune cell runs on — and the class has already failed clinically on exactly that
NF-kB is the switch that immune cells use to respond to infection, and it is also what normal blood stem cells rely on under stress. Drugs that block it have repeatedly failed in people because of the side effects, and the patients here are babies whose immune systems are still being built while chemotherapy is already suppressing them.
Both approved menin inhibitors cause multi-organ hyperplasia in animals, and one of them in the pancreas irreversibly
In animal safety studies, both approved menin-blocking drugs caused tissues in several organs to overgrow. For revumenib this included the pancreas, and in that organ the overgrowth did not go away after the drug was stopped.
Cardiac troponin release on MCL-1 inhibition - reported for two structurally distinct compounds and described by one developer as a likely CLASS effect, which makes it an essential-function ceiling rather than a chemistry problem
Two different drugs that block MCL-1 both made a blood marker of heart-muscle damage go up in patients, even when nothing else about the heart looked wrong. Because the two drugs are chemically unrelated, the likeliest explanation is that heart muscle needs MCL-1 - which would mean the problem belongs to the target, not to the drug.
Decitabine's target is a pan-essential gene, and KMT2A-rearranged lines depend on it LESS than other leukaemia lines - rank 16,949 of 17,728
The enzyme decitabine destroys is needed by almost every cell that has been tested, so the drug's safety comes from dosing and timing rather than from hitting only cancer. And in the cell lines available, KMT2A-rearranged leukaemias need it slightly less than other leukaemias do, not more.
Defect report: analyses/does-this-deposit-agree-with-itself/run.py keys each GEO record on !Sample_geo_accession, which in a SOFT record comes AFTER !Sample_title - one line, and it inverted the finding
A specific bug in one file of this project, written down precisely enough that someone can fix it in under a minute, together with what it changed.
Eleven immortalised lines passaged for decades: the competing explanation this beat could not close
Every number behind the Fanconi finding comes from cancer cells that have been growing in plastic dishes for twenty to forty years. Cells in long-term culture accumulate their own DNA-copying problems, so a repair dependency measured in them might be a property of the dish rather than of the leukaemia. Nobody has measured this in a child's actual leukaemia cells, and that is the gap this entry exists to name.
ENL stays after the fusion leaves — an epigenetic lesion that outlives the protein that made it
In one patient on a menin inhibitor the drug worked perfectly: both the fusion protein and menin became undetectable. The leukaemia carried on anyway, because a helper protein called ENL was still parked on the same stretch of DNA, keeping the genes running without the protein that originally put it there.
Every niche-disrupting drug strips the patient's own stem cells of the same shelter — the bite-back that applies to the whole beat
The bone-marrow hiding place that shelters leukaemia cells is the same hiding place that keeps a child's healthy blood stem cells alive and in position. Any drug that breaks it breaks it for both, at exactly the time the child needs those healthy stem cells most.
Evicting the fusion is not enough: PRC1.1 has to arrive, and losing it causes menin-inhibitor resistance in vivo
Menin drugs push the cancer protein off the genes it is holding open, but that alone does not switch those genes off. A repressive complex has to move in afterwards — and leukaemia cells that have lost that complex keep the genes on and survive the drug.
Giving a G9a/GLP inhibitor alongside etoposide or doxorubicin removes a DNA-repair licence at the moment TOP2 breaks are being made - and the sibling compound UNC-0638 is measured to block BRCA1/BARD1 arrival at a break, in children on curative therapy
This graph proposes giving children a drug that strips a silencing tag off DNA, alongside chemotherapy, to stop a leukaemia-causing break from happening. Three separate papers now show that removing that same tag stops the cell's repair machinery from reaching a break - so the drug might prevent one break and leave every other break unrepaired.
Giving a translocation-promoting epigenetic drug in a translocation-defined B-lineage leukaemia, to infants
The drugs that block Polycomb work by stripping off a 'switch off' tag, which gets replaced by an 'switch on' tag and makes nearby genes more active. In B cells that are actively rearranging their antibody genes, that extra activity makes chromosome breaks join up wrongly more often - and this leukaemia is caused by exactly that kind of wrong join.
Graft-versus-host disease in a transplanted infant - the only antileukaemic benefit and the main cause of death are the same process at different grades
When donor immune cells attack the child's own tissues it is called graft-versus-host disease. A mild version is good, because the same cells also attack any leftover leukaemia; a severe version kills more children than it saves. Doctors cannot aim for the mild version, and in a baby the drugs used to control it leave an immune system that has never met anything before.
In a live mouse, losing menin in bone cells means less bone - and menin overexpression means more
When menin is deleted from bone-building cells in mice, the mice end up with thinner, weaker bones; when extra menin is put in, they end up with more bone. Deleting it only in the bone cells buried inside the bone also causes osteoporosis, by making the bone-dissolving cells more active.
Intact PTH rises in 34% of patients on revumenib, and hyperparathyroidism is on the label
The approved label for revumenib reports that the parathyroid hormone level went up in about a third of treated patients, and lists hyperparathyroidism as a side effect. That is the same hormone that goes wrong first in people born with a faulty menin gene.
Irreversible femoral growth-plate closure in juvenile rats: the label already tells paediatricians to watch growth
In young rats given revumenib for three months, the growth plates in the thigh bone closed early and did not recover. Because of this the approved label instructs doctors to monitor bone growth and development in children taking the drug.
JUND liberation — the drug's mechanism of action and one of its mechanisms of resistance are the same event
The leukaemia drug works by prising a protein called menin off the fusion protein, but menin holds a second protein, JunD, in exactly the same groove. Freeing JunD lets it switch on its own set of genes, and those genes help the leukaemia survive the drug.
KMT2A::USP2 is invisible to karyotype and often to FISH, so these patients are misclassified as fusion-negative
Because this fusion is made by a flip inside one chromosome rather than a swap between two, the standard chromosome test looks normal and the usual fluorescent probe test is often normal too. Children with a high-risk leukaemia are therefore filed as having no fusion at all, and treated accordingly.
LAMP5-directed therapy for KMT2A-rearranged leukaemia would meet the basal ganglia, in a disease treated in infants and routinely treated into the CNS
LAMP5 is one of the most brain-concentrated proteins in the body, at its highest in a deep brain region that is still developing in babies. Leukaemia therapy in this disease is deliberately directed into the nervous system, so a LAMP5-targeting treatment would go where the protein is.
Lineage switch as a menin-inhibitor escape route — documented in a patient on revumenib, and invisible to every resistance-mutation screen
Some leukaemias escape a drug not by changing their genes but by changing what kind of cell they are, swapping a B-cell identity for a myeloid one while keeping the same underlying fault. This has now been seen in a patient being treated with a menin inhibitor, and no DNA test would have found it.
Lineage switch as antigen escape from CD19-directed therapy in KMT2A-rearranged leukemia
Drugs that hunt the CD19 marker on B-cell leukemia work extremely well, but in KMT2A-rearranged leukemia the cancer can dodge them by turning into a myeloid leukemia that has no CD19 at all. It usually happens within weeks to a couple of months, it is often missed because the B-cell test comes back clean, and most patients do not survive it.
Lineage switch to myeloid leukaemia under CD19-directed therapy in ZNF384-rearranged ALL - and in the one cohort that caught both families, the ZNF384 switch looked mutationally NOTHING like the KMT2A ones
Leukaemias with a ZNF384 fusion can escape CD19-targeted drugs by turning into myeloid leukaemia, the same escape route already known for KMT2A. But in the one study that caught both kinds in the same group of children, the KMT2A escapes carried almost no new mutations while the ZNF384 escape carried a large set - so the two families may be doing the same thing by different means.
MEN1 drug-interface mutations — the first proof that a chromatin drug can drive Darwinian escape in patients
Under treatment with a menin inhibitor, some leukemias evolve a tiny change in the menin protein exactly where the drug sits. The drug can no longer bind, but menin can still hold on to the fusion protein, so the cancer program restarts while the drug does nothing.
Menin displacement is a poor pharmacodynamic proxy for disabling the fusion, and the degron shows both halves of why: removing the fusion barely moves menin genome-wide, while the menin inhibitor that moves menin leaves half to three-quarters of the fusion's own chromatin sites occupied
The natural way to check that a menin drug is working is to measure whether menin has come off DNA, and two results here say that is the wrong check. Destroying the cancer protein hardly moves menin at all, while the drug that does move menin leaves most of the cancer protein still attached - so the thing you can measure and the thing you care about come apart in both directions.
Menin inhibitors are prescribed on 'KMT2A-rearranged', a category that contains proteins as different as KMT2A::AFF1 and KMT2A::USP2
The approved drug for KMT2A leukaemia is given to anyone whose KMT2A gene is rearranged, whatever it is joined to. But some of those rearrangements make a protein that keeps parts the common ones throw away, and nobody has checked whether the drug works on them.
Menin is required in neural crest for the palate to close - and both menin inhibitors cause cleft palate in animals
Deleting menin in the cells that build the face kills mouse pups at birth with a cleft palate and skull defects. Both approved menin drugs cause cleft palate and skeletal malformations in animal pregnancies, at doses below the human dose.
Mixed donor chimerism under three years of age - the same drug exposure ablates less marrow in a younger child, and dosing cannot fix it
After a transplant, doctors check what fraction of the child's blood cells come from the donor. In children under three, nearly half end up with a mixture of their own and the donor's cells, while in older children almost none do - and this happens even when they received exactly the same amount of the conditioning drug. In a leukaemia, the cells that were not replaced are where a relapse comes from.
MYC reactivation through PRC1.1 loss — a named, druggable bypass out of menin dependence
Menin inhibitors work by shutting off a set of genes. Some leukemias delete a different piece of machinery, PRC1.1, which lets them switch a second, hidden set of genes back on, including MYC. The cancer then runs on that hidden set instead, and the drug becomes irrelevant.
Nine survivors in ten carry a long-term complication, and it is worse the younger they were
Of 67 children who survived infant leukaemia for at least five years, 90% had at least one lasting health problem and two thirds had several - problems with learning, weight, teeth, hearing, lungs and hearts. Babies diagnosed before six months, and babies whose leukaemia carried the KMT2A rearrangement, ended up with more of them.
Pan-essentiality: checked for all eight, and cleared — but a shallow dependency is its own problem
A gene that every cell needs is useless as a cancer drug target, because blocking it would harm healthy cells just as much. All eight Fanconi genes behind this finding were checked against every cell line ever screened, and none of them is in that category. The catch is the opposite one: the effect is small enough that there may not be much of a window either way.
PSIP1 is a tumour suppressor in the lymphoid lineage next door: losing Psip1 ACCELERATES T-ALL in mice, while the same protein is a dependency in KMT2A-rearranged myeloid leukaemia
In one kind of childhood leukaemia LEDGF is what the cancer needs. In the T-cell kind, mice that lose LEDGF get leukaemia faster, and patients' tumours carry loss-of-function mutations in it. A drug that removes LEDGF could help one disease and promote another.
Raising MHC class I is not a free good: the same molecule is the inhibitory ligand that holds NK cells off, and in this fusion family it is an in vivo leukaemia dependency
Turning up a leukaemia cell's identity cards makes it easier for one kind of immune killer to find, and harder for another kind. The second kind is the one that matters most after a mismatched transplant, which is exactly the setting where this drug is being proposed.
Risk: CLEC12A-directed therapy ablates normal granulocytes and disables an inhibitory brake on NET formation - the toxicity is mature-myeloid, not stem-cell
CLEC12A therapies are sold on the promise that they spare the blood-forming stem cells, and they mostly do - but CLEC12A is also on ordinary neutrophils, the white cells that fight infection. In the first trial of a CLEC12A CAR, every patient lost their blood counts and two died of infection.
Risk: the class's pharmacodynamic logic measures menin and reports it as the fusion, in a drug class dosed in infants
The tests used to show these drugs are working measure menin coming off DNA, or one gene switching off. Neither shows what happened to the leukaemia's fusion protein. In a trial in babies, where you can take very few samples, measuring the wrong thing is expensive.
Signalling co-mutations (RAS pathway, FLT3-ITD, WT1) under menin-inhibitor pressure
Even if a menin inhibitor successfully switches off the fusion's gene program, a leukemia cell that also carries a mutation in a growth-signalling gene like KRAS or FLT3 may keep growing anyway. These mutations are common in KMT2A-rearranged leukemia and they are the most obvious reason the drug can work and the patient still relapse.
Sinusoidal obstruction syndrome after infant conditioning - an age signal, a regulatory exposure threshold, and a prophylaxis that failed its randomised trial
Busulfan and the other conditioning drugs can block the small veins draining the liver, which makes the liver swell, the belly fill with fluid and the child turn yellow. It kills about one in five children who develop it. Being under a year old looks like a risk factor, and the drug that was supposed to prevent it failed when it was properly tested.
Switching menin inhibitors: four substitutions defeat all of them, three defeat only some, and the map now exists
When a leukaemia becomes resistant to one menin inhibitor, the obvious next move is to try a different one. A systematic screen across five of these drugs shows that some resistance changes block every drug in the class, while a few block only particular ones — so switching sometimes helps and usually will not.
Targeting LIN28B - the essential-function and wrong-sign risks are established by the SAME paper that makes it interesting, and the sign may reverse with lineage
The paper that shows infant leukaemia cannot start without LIN28B also shows that a baby's own B cells and NK cells cannot be made without it. And other papers show that in myeloid leukaemia, adding LIN28B makes the cancer grow up and die - so a drug that blocks it could make that kind worse.
The bite-back of striking a target: a wrong strike costs a cure and generates no data that it was wrong
Deciding not to pursue a drug target is a decision with consequences, but unlike a failed experiment it never produces a result telling you it was a mistake. The target just stops being looked at, and nobody finds out.
The entire experimental model space for SET::NUP214 is two cell lines, neither of them in DepMap, with no mouse model found - so every mechanism in this beat is n = 1 or n = 2
Almost everything known about this fusion comes from two laboratory cell lines, and one of them supplies nearly all of it. That is a very thin base for any conclusion, and it is also why several obvious experiments have never been done.
The GMP-like state is being used as a therapeutic anchor in two independent arguments, and its chromatin definition rests on one post-lineage-switch sample that also defined it
Two separate lines of reasoning treat a particular myeloid-like state as the place where a new drug class works best, and as the place leukaemias go when they change identity. The chromatin version of that state is defined from one patient, and outside that patient this deposit cannot see it.
The graph's universal 'every KMT2A fusion loses the SET domain' is a silent, propagating error class
A statement that is true about the famous half of a chromosome swap has been written down as if it were true about all of them. It has travelled into at least three entries and into this beat's own instructions, and nothing in the checking machinery can catch it, because every citation attached to it is correct.
The H3K9me3 null that appears to exonerate prophylactic G9a inhibition was measured in retinal pigment epithelium, at 1 kb resolution, with n = 2 - and the warning it appears to lift lives in a leukaemic blast
This work found that one silencing tag does not change at the KMT2A gene, which makes a proposed preventive drug look safer. But the measurement was in eye cells, and the safety worry is about blood cancer cells. The reassurance does not reach the place the worry lives.
The lineage confound, tested: the FA signal is in both halves, the B-lineage half is stronger, and it survives holding lineage constant
The cells used to find the Fanconi result were mostly of one blood-cell type, so the finding might have been about that cell type rather than about the KMT2A fusion. Splitting them up shows the opposite of what that worry predicted: the effect is stronger in the minority half, and it is still there when the comparison is made within a single cell type.
The menin-inhibitor-resistant state has its own address: SMARCA4, BRD4 and CREBBP, with no MEN1 mutation anywhere
Leukaemia cells forced to grow through repeated doses of a menin inhibitor end up resistant without any change in the drug's target. When researchers asked what those resistant cells now need to survive, the answer was a short list of chromatin machines — and drugs against two of them killed the resistant cells.
The MLL3/4-UTX switch fails to throw — menin inhibition kills by turning something ON, and cells that cannot turn it on survive
Menin inhibitors are usually described as switching cancer genes off. In fact part of how they work is switching a set of protective genes back ON, using a different machine called MLL3/4-UTX. Leukaemias that have broken that machine feel the first half of the drug and not the second, and they survive.
The MRD target a centre can run decides which infants are intensified, and the two targets disagree in the direction that matters
Which leftover-disease test a hospital is able to run changes which babies get the most intensive treatment, because the two available tests disagree about roughly one child in ten at exactly the level where the decision is made. The disagreement is not random: the test that needs a specialist laboratory catches relapses the routine one misses.
The off-target surface, measured rather than feared: KDM3B depletion moves CTCF occupancy at 84.6% of 12,443 RPE anchors and halves it at 1,436 - so an H3K9-axis drug given to prevent one rearrangement is remodelling loop architecture genome-wide, and preferentially at the strongest anchors
The idea behind this drug is to keep one gene's loop anchor in place so the gene does not break. But when the enzyme is removed, the anchor protein shifts at most loop anchors in the whole genome - about eight in ten - and drops by half at more than a thousand of them. Giving a healthy infant a drug that rearranges the genome's loop structure everywhere, to protect one gene, is a very different proposition from the published picture.
The other half of the islet story: losing menin also costs the beta cell its identity and raises glucagon
Deleting menin in insulin cells does not only make more of them. Over time the cells lose the markers that make them insulin cells and start making glucagon instead, and mice with one faulty copy end up with high fasting blood sugar driven by too much glucagon.
The predictable clinical error this beat exists to prevent: a menin inhibitor given for a KMT2A-rearranged sarcoma because of the gene name
A menin-blocking drug is approved for KMT2A-rearranged leukaemia, and a molecular report on a sarcoma can say 'KMT2A rearrangement' in the same words. In these sarcomas the part of KMT2A that menin holds on to is not in the fusion at all, so the drug has nothing to grip.
The reversibility question: months of partial occupancy against a syndrome built from decades of loss
MEN1 syndrome comes from losing menin permanently, in every cell, for decades. A menin inhibitor takes away part of one of menin's jobs for a few months and then stops, and whether that is enough to matter is genuinely unknown.
The same pocket suppresses JUND: menin inhibition's unmeasured endocrine liability in infants
Menin inhibitors work by plugging a pocket on the menin protein, but that same pocket is where menin does a second job that normally stops tumours forming in hormone glands. No trial has followed a treated infant long enough to know whether blocking it for months matters decades later.
The shared reader sits in a complex whose only well-defined normal job is parent-of-origin silencing — and the published safety package for YEATS inhibition is entirely haematopoietic
The protein a YEATS drug grabs also sits inside a machine whose normal job is keeping parent-of-origin gene silencing in place, and another one built on the gene that causes an inherited learning disability. The safety testing done so far only looked at blood stem cells, which cannot see either of those things, and the patients are babies.
The whole bone-mineral axis moves on revumenib: phosphate in 51%, alkaline phosphatase in 33%, calcium in 15%
Revumenib's most common laboratory abnormality is a rise in blood phosphate, seen in half of patients, alongside changes in calcium and a bone enzyme. Taken together these are the numbers a doctor would look at to check whether bone and mineral handling is being disturbed.
TP53 inactivation and the apoptotic threshold — the drug still switches the program off, and the cell still does not die
A menin inhibitor kills leukaemia cells only if those cells are still able to carry out the self-destruct program. Leukaemias that have lost the p53 gene keep the self-destruct machinery switched off, so the drug can do its job on the genes and the cell simply survives anyway.
Treating a subclonal lesion is selection pressure applied to the clone you did not aim at
If only a fraction of a leukaemia's cells carry the mutation a drug is aimed at, the drug kills that fraction and leaves the rest - and the rest is most of the disease. In infant KMT2A leukaemia the drug-targetable mutations are usually in a minority of cells, which makes aiming at them a way of clearing the field for the cells nobody targeted.
Treating the five menin inhibitors as one drug, when a 2026 title says they are not
This project reasons about five menin inhibitors as if they were interchangeable because they all block the same pocket. A 2026 paper from the Perth infant-leukaemia group states in its title that they show different sensitivity in infant leukaemia cells - and the paper itself cannot be read anywhere, so we know the claim exists and not what it says.
Two reasons XPO1 is a hard target: every cell needs it, and blocking it switches on a survival pathway
XPO1 is needed by essentially every human cell, so there is no built-in safety margin from the target itself. And when you block it, leukaemia cells turn on a survival signal that partly rescues them - a resistance mechanism that appears with the first dose rather than after months.