Why the fusion causes the disease
Two separate questions live here. Why does a chromosome break and rejoin in the first place, and once it has, why does the resulting protein lock a blood cell in an immature, dividing state?
Three ways to cut the same gene
The corpus holds three mechanisms that can cut the KMT2A breakpoint cluster region, all at moderate confidence and none established. They are drawn at equal weight because the evidence gives them equal weight — and the one observation that separates them is positional, not mechanistic: de novo breaks favour the centromeric half of the cluster, therapy-related breaks the TOP2-rich telomeric half.
Two properties are left off the chart on purpose. Every one of these entries carries published counter-evidence, and 464 of 465 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. 461 of 465 additionally carry a bite_back note, which is required only where an entry touches an intervention.
Fragile DNA, the enzymes that cut it, and the repair pathways that glue the wrong ends together.
Apoptotic nucleases cut KMT2A, and cells that survive the attempt can carry a translocation
When a cell starts to kill itself, it chops up its own DNA, and one of the first places it cuts is right inside the KMT2A gene. If the cell then changes its mind and survives, the repair crew can stitch that cut to a cut on another chromosome, leaving a leukaemia-causing fusion behind in a living cell.
At the KMT2A anchor itself, KDM3B loss does not raise H3K9me1 AND H3K9me2 - it raises me1 3.9-fold while me2 FALLS to 0.44x, an inversion of the mono-to-di ratio that the paper's abstract describes as a joint increase
The paper says switching off one enzyme raises two closely related tags on the KMT2A gene. At the exact spot that matters, only one goes up - the other goes down. The tags are not moving together, they are swapping.
Both arrows on KDM3B depletion are broad: H3K9me1 rises genome-wide AND CTCF falls at 84.6% of 12,443 anchors - the KMT2A anchor's 0.40x is the 3.6th percentile among strength-matched anchors, not a unique event
Knocking down one enzyme raises a silencing tag all over the genome, and it also loosens the anchor protein all over the genome - at most anchors measured, not just at KMT2A. The KMT2A anchor is among the harder-hit ones, but it is not picked out.
Class switch recombination: a B cell deliberately mutates and cuts its own antibody locus, and the same enzyme causes MYC-IgH translocations
To change an antibody from one class to another, a B cell chemically damages its own DNA on purpose and then repairs it badly, deleting the piece in between. The enzyme that starts this is also the enzyme that causes one of the classic cancer-causing chromosome swaps.
Gray 2023's own 'exon 11 CTCF site' in RPE cells, measured: centroid GRCh38 chr11:118,488,662-118,488,761, inside canonical exon 11 and 34 bp from the independently measured summit - the graph's CTCF adjudication is confirmed, not overturned
A 2023 paper said a protein called CTCF sits on 'exon 11' of the KMT2A gene and protects it from breaking, but never gave a coordinate - and 'exon 11' means two different stretches of DNA 3.8 kb apart depending on whose numbering you use. Reading the paper's own raw data settles it: the paper means the same stretch this project had already measured, not the other one.
H3K9me1 - the only H3K9 state whose level at the KMT2A breakpoint cluster region has ever been measured, and it is LOW there until KDM3B is removed
H3K9me1 is the first, single methyl tag on histone H3 at lysine 9 - the faintest version of the silencing mark. In healthy cells the stretch of KMT2A where this leukaemia's break happens carries unusually little of it, and when the enzyme that keeps it clear is removed the tag floods in and the gene starts breaking.
In-utero origin, proven: clonotypic KMT2A::AFF1 junctions found in the patient's own newborn blood spot
Every baby in most countries has a few drops of blood dried onto a card in the first days of life. Researchers went back to those cards for children who later developed leukaemia and found the child's exact leukaemia fusion sequence already there at birth, so the first event happened before the baby was born.
Meiotic recombination: hundreds of deliberate DNA breaks per cell, made by a topoisomerase relative, at sites chosen by a single zinc-finger protein
Every egg and sperm cell is made by deliberately snapping its own chromosomes in hundreds of places so the parents' chromosomes can swap pieces. Where the snaps happen is decided by one protein, and when the swap lines up the wrong repeats, the child is born with a deleted or duplicated chunk of genome.
RAG off-target activity: the dominant mutator in ETV6::RUNX1 ALL, and probably not the cause of KMT2A fusions
Developing B cells own a pair of molecular scissors called RAG whose legitimate job is to cut and reshuffle antibody genes. In some childhood leukaemias those scissors slip and cut the wrong genes. For KMT2A fusions, though, the junction sequences say RAG was not involved.
Read-through (cis-SAGe) chimeras: most 'fusions' called from RNA-seq in childhood ALL are normal conjoined genes
When RNA copying machinery runs past the end of one gene into the next-door gene, it makes a hybrid message from two completely normal, unbroken genes. This happens all the time in healthy tissue, and in childhood leukaemia most of the hybrid messages a computer finds in RNA data turn out to be exactly this rather than real fusions.
Replication stress and interstrand-crosslink repair as a cell state - an ATR dependency measured in mouse MLL-rearranged AML, and the same stress implicated in breaking the MLL locus in the first place
Copying DNA is not smooth: forks stall, and a repair crew has to restart them before the chromosome breaks. This entry records the evidence that leukaemia cells carrying an MLL/KMT2A rearrangement live under that kind of stress, and that the same stress is one of the ways the rearrangement is thought to be created.
Replication stress, AID, base-excision repair and endonuclease G cut the KMT2A locus
When a cell's DNA-copying machinery stalls, a normally mitochondrial nuclease called endonuclease G moves into the nucleus and cuts the KMT2A gene at a specific hotspot. Blocking one of the repair enzymes upstream of it (APE1) protected human cord-blood stem cells from that cut, which makes this the most concrete preventive target in the whole fusion-genesis story.
TDP2 unglues trapped topoisomerase II from the KMT2A locus - it suppresses translocations overall and manufactures one rare junction class
When topoisomerase II gets stuck on DNA, an enzyme called TDP2 snips it off so the two clean ends can be glued straight back together. Losing TDP2 makes cells break KMT2A more and swap chromosomes more - but one unusual kind of scar found in therapy-related leukaemia only appears when TDP2 is working.
The apoptotic priming state — KMT2A-rearranged infant ALL is the one childhood ALL subgroup that is BCL-2-dependent rather than BCL-XL-co-dependent, and BH3 profiling is how that was measured
Every cell sits somewhere on a scale between ready to die and hard to kill, and a test called BH3 profiling reads that position directly by dripping death signals onto isolated mitochondria. Infant KMT2A-rearranged leukaemia sits in an unusual place on that scale: it leans on one survival protein, BCL-2, where most childhood leukaemias lean on two.
The CTCF site in KMT2A exon 11 - a conserved insulator element sitting exactly on the functional cut point of the breakpoint cluster region, whose occupancy decides whether the locus rearranges
Inside the fragile stretch of the KMT2A gene there is one spot where the clamp protein CTCF sits, in every cell type anyone has looked at. That spot happens to be exactly the boundary that decides whether a break produces a cancer-causing fusion protein or a harmless one.
The H3K9me1/2-CTCF brake: a KDM3B-G9a balance keeps the KMT2A breakpoint cluster region methyl-poor and CTCF-bound, and losing either end makes the locus amplify and rearrange - in epithelium, in primary cord-blood progenitors and in mice
The stretch of the KMT2A gene where this leukaemia's break happens is normally kept clear of a silencing chemical tag, which lets a clamp protein called CTCF sit on it and hold it shut. Remove the enzyme that keeps the tag off - which is what the chemotherapy drug doxorubicin does - and the tag floods in, the clamp falls off, and the gene starts copying and breaking itself.
The KMT2A breakpoint cluster region in base positions: three incompatible definitions reconciled to one GRCh38 table, with the DCAL frame's three bp identities re-derived and the GRCh37 lift measured per interval
Three entries in this project describe the stretch of KMT2A where the gene breaks, and all three give different exon numbers for what they claim is the same piece of DNA. This converts every version into plain chromosome coordinates so nobody has to guess again.
The KMT2A exon-11 CTCF anchor, placed in its population: baseline strength 4.92x (91.7th percentile of 12,443 RPE anchors), KDM3B-depletion ratio 0.396x, rank 58 of 1,625 strength-matched anchors - a strong, strongly-responding anchor at the edge of a continuum, not an outlier
This is the record of where the loop-anchor inside the KMT2A gene sits among all twelve thousand similar sites in the same cells - how strong it is to begin with, and how much protein it loses when the enzyme is removed. It is stronger than 92 out of 100 of them and loses more than 96 out of 100 of the ones its own strength, which puts it at the edge of the crowd rather than outside it.
The KMT2A exon-11 CTCF/cohesin summit, measured: GRCh38 chr11:118,488,694 (CTCF) and chr11:118,488,677 (RAD21), 17 bp apart, both inside canonical exon 11
This project had an interval for the CTCF landing site inside KMT2A but never a single base. Reading the published raw signal gives one: the peak of CTCF binding is at base 118,488,694 of chromosome 11, and the cohesin ring peaks 17 bases away, both inside the short piece of gene the whole argument is about.
The KMT2A locus sits inside a 179-318 kb H3K9me2-domain-free block in human CD34+ bone-marrow progenitors - the 7th, 11th and 37th longest such block on chr11 in three ChIPs - and the same block is present in granulocytes, K562 and most primary AML
In real human blood stem cells, the KMT2A gene sits in one of the largest stretches of chromosome 11 that carries none of the silencing tag at all - a gap of two to three hundred thousand bases. That is the same baseline scarcity the founding paper reported in eye-derived cells, now seen in the right kind of cell. But the same gap is there in mature white cells and in leukaemia too, so it is a feature of the place, not of the cell.
Topoisomerase II cleavage and TOP2 poisons as a source of KMT2A breaks
Topoisomerase II is an enzyme that deliberately cuts both strands of DNA to untangle it, then sews them back up. Drugs like etoposide jam it mid-cut, leaving a real double-strand break, and the KMT2A gene happens to sit at a favoured cutting site, which is why children treated with these drugs can later develop KMT2A-rearranged leukaemia.
Trans-splicing: healthy human cells make cancer-associated fusion transcripts without any chromosome rearrangement
Some fusion proteins that we think of as cancer-causing are also made in perfectly healthy tissue, not by joining chromosomes but by joining two different RNA messages together after they are copied. The DNA is entirely normal; only the message is hybrid.
Transcription–replication conflict as the candidate reason KMT2A-rearranged cells need the Fanconi pathway — a chain with two measured links and three inferred ones
Cells copy their DNA and read it at the same time, and where those two machines meet on the same stretch of DNA they collide. The KMT2A fusion makes cells read certain very long genes much harder than normal, so the idea is that collisions happen more often there, and the Fanconi repair crew is what clears them up — which would explain why these cells need that crew. Most of this chain has not actually been measured in KMT2A leukaemia cells.
V(D)J recombination: the immune system deliberately cuts its own chromosomes, using a domesticated transposon
Every developing lymphocyte deliberately cuts its own DNA in several places and glues the pieces back in a new order, which is how the body builds millions of different antibodies from a handful of gene segments. The scissors that do this are a tamed jumping gene that a vertebrate ancestor captured and put to work.
Where RPA actually is: nucleoplasmic and diffuse until there is single-stranded DNA, then punctate prereplication foci at the G1/S border, damage-induced foci with RAD51, and a PML-body pool - and the subunits differ in how well any of this is measured
The RPA clamp has no fixed address in the nucleus. It floats diffusely until single-stranded DNA appears, and then gathers into bright spots - at sites where DNA copying is about to start, and at damage. It is also found in PML bodies, a distinct nuclear compartment.
Which repair pathway glues the fusion together: classical NHEJ vs alternative end joining
Two different repair crews can join the wrong chromosome ends. One works fast and blunt; the other looks for a few matching letters on each side and trims everything in between, leaving a tell-tale scar. Reading those scars in patient DNA tells us which crew made the fusion, and in human cells the answer is mostly the fast, blunt one.
Why KMT2A picks AFF1: spatial proximity, shared transcription hubs, and the unsettled proximity-versus-damage question
Chromosomes 11 and 4 are separate, but inside the nucleus the KMT2A gene and the AFF1 gene can end up in the same small neighbourhood because both are switched on and busy at the same time. Ends that are already next to each other are the ones that get glued together.
What the hybrid protein does that neither parent could.
'AF4 bodies' - a named subnuclear compartment that is non-nucleolar and coincident with no known nuclear structure, and the one place a KMT2A fusion has been shown to move a partner
In 2004 a group found that the two commonest partners of the leukemia fusion, AF4 and AF9, meet each other in small nuclear spots they called 'AF4 bodies' - spots that did not line up with any nuclear structure they tested. When the MLL-AF4 fusion was present, AF9 moved.
"Dispensable for haematopoiesis" is a summary of a real phenotype: halved white counts, perturbed HSC and progenitor numbers, and failing serial replating
Mice that lose LEDGF in their blood system live normally, which is the main reason people think a LEDGF drug would be safe. They also have about half the normal white blood cell count and blood stem cells that stop making colonies after repeated challenge.
2-hydroxyglutarate competitively inhibits the alpha-KG dioxygenases - and it hits the TET demethylases AND the KDM histone demethylases, which is why its net effect on a CXXC reader is not predictable
Mutant IDH enzymes make a molecule that looks enough like the fuel used by a whole family of enzymes to jam them. Two of the jammed families matter here: the ones that remove methyl marks from DNA, and the ones that remove them from histones - and the KMT2A fusion's competitors sit in the second family.
A 2026 preprint puts XPO1 upstream of the menin interface itself: XPO1 reported to stabilise menin on chromatin and menin's binding to KMT2A and KMT2A::MLLT3
A 2026 preprint reports that blocking the export protein XPO1 loosens menin's grip on DNA and on the KMT2A fusion - the very grip that menin inhibitors are designed to break. Combining the two drugs killed KMT2A-rearranged leukaemia cells better than either alone, in cells, in patient cells and in mice.
A long non-coding RNA binds DOT1L's catalytic domain and turns its activity up - and blocking it makes the fusion protein itself get eaten
A long RNA molecule called LAMP5-AS1 sticks directly to DOT1L's business end and makes it work harder, which raises the H3K79 tag and keeps leukaemia stem-cell genes on. Patients with more of this RNA do worse, and removing it makes the leukaemia's own fusion protein get broken down by the cell's recycling system.
A menin inhibitor does not only take something away - it causes PRC1.1 to put something on, and the drug needs the cell's own silencing machinery to work
The standard story is that a menin inhibitor works by prising the leukaemia's fusion protein off the DNA, after which the genes it was switching on simply fall silent. The newer picture is that silence is not automatic: the cell has to actively bolt the genes shut using a machine called PRC1.1, and if that machine is broken the drug does much less.
A phosphosite on a menin target predicts who responds to a menin inhibitor
MEF2C is a protein switched on by the KMT2A-menin machinery. Phosphorylating it at one spot marks the leukaemias that respond to menin-blocking drugs, and adding a drug that stops that phosphorylation makes the menin drug work better.
A second brake, unconnected to JunD: menin is a TEAD1 corepressor with VGLL4 in the beta cell
Menin does not only work through JunD. In insulin-making cells it also teams up with a protein called VGLL4 to switch off a growth programme run by TEAD1. That is a second, separate way menin holds the cell still.
A second disease finds the same DOT1L-to-non-canonical-PRC1 axis in the same year - and the mark moves the opposite way
The 2026 leukaemia paper that this project's whole downstream story rests on says that blocking DOT1L makes a silencing mark go up everywhere. A second 2026 paper, in lymphoma, finds the same two machines wired together - and reports the mark going down. Both cannot be a general rule.
A ten-residue export signal is the 5' partner's whole contribution - and the field disagrees about which direction it pushes
The only part of PICALM that the leukaemia actually needs is a short tag telling the cell to pump the protein out of the nucleus. Swapping that tag for the same tag taken from unrelated proteins works just as well - but one influential paper argues the opposite, that keeping the fusion IN the nucleus is what matters.
A two-compartment switch for a SEC cargo: nucleolar retention versus AFF4-induced nucleoplasmic droplets, set by one methylated arginine
HIV's Tat protein can either be parked in the nucleolus or pulled into liquid droplets built by the elongation machinery in the rest of the nucleus, and a single chemical mark on one amino acid decides which. It is the clearest worked example of a protein's job being decided by which nuclear compartment holds it.
Acetylation is a cycle, not a mark - and this graph holds one complete cycle, on a residue the fusion throws away
An acetyl tag is not a permanent label, it is a balance: one enzyme adds it, another reads it, a third removes it, continuously. This graph spent many cycles collecting the enzymes that add tags and never collected a single one that removes them - and the one place where it now has the whole loop, the tag sits on a piece of protein the cancer version has deleted.
AF10/MLLT10 fusions drive aberrant self-renewal of non-stem progenitors — found only by opening the paper
Fusions involving the AF10 gene make ordinary blood progenitor cells behave like stem cells that renew themselves forever. The paper showing this has a title that mentions none of that, so a keyword search misses it.
AF9's nuclear residence is chaperone-dependent, the Hsp90 site is the SAME C-terminal region the fusion keeps, and UniProt records no cytoplasmic pool at all
AF9 is usually described as a nuclear protein, but it needs a chaperone - a protein that escorts other proteins - to stay in the nucleus. Block that chaperone and AF9 moves out into the cytoplasm and comes off the gene it was controlling.
AFDN (AF6): nothing is discarded from the fusion, yet this is the clearest subtraction in the set — the fusion steals wild-type AF6 out of the cytoplasm
The AF6 fusion throws away almost nothing — the break happens before the first working part of the protein. But the fusion drags the cell's normal AF6 into the nucleus, and because normal AF6 was holding down a growth switch called RAS out in the cytoplasm, that switch gets stuck on.
AFF1 as the subtraction control: the refutation is WITHDRAWN - the degron is lost, not retained
For the fusion this project cares most about, the fusion partner brings something new rather than losing something old. We tested the best candidate for a hidden loss — a self-destruct tag that could have been cut off — and found the tag is still there.
After lineage switch, KMT2A-rearranged AML does not merely lose CD19 - it acquires a monocytic myeloid-derived suppressor cell phenotype and actively suppresses T cells
When this leukaemia escapes CD19-targeted therapy by turning into a myeloid leukaemia, it does not just drop the target. The cells it becomes are the kind that shut T cells down, which is a second reason the immune therapy stops working.
AK_7_3, a KMT2A::AFF1-derived fusion circRNA that holds down mitochondrial respiration and ROS in t(4;11) B-ALL - the only mitochondrial mechanism in this beat measured in the index fusion
The t(4;11) translocation does not only make a fusion protein; it also makes a small circular RNA that joins the two genes back to front. Removing that circle from leukaemia cells makes their mitochondria work harder, produce more damaging oxygen radicals, become misshapen, and the cells start dying - so the circle is normally keeping the mitochondria quiet and the cells alive.
AKAP95 binds the Pol II CTD directly and nascent RNA gates the condensate - a feed-forward loop where transcription recruits the factor that sustains transcription
AKAP95 sticks directly to the tail of RNA polymerase, and the RNA the polymerase is making feeds back to make the droplet bigger. More transcription makes more RNA, which recruits more AKAP95, which supports more transcription.
AKAP95 directly enhances H3K4 mono-, di- and tri-methylation - by the KMT2B complex, and ONLY by the KMT2B complex; the paper's own discussion says the MLL1/KMT2A case was never tested
AKAP95 makes one particular H3K4-methylating machine work better - the one built around KMT2B. People often quote this as if it applied to KMT2A, the gene that matters in this leukaemia, but that experiment was never done and the authors said so.
AKAP95's subnuclear addresses - nuclear matrix, nucleolus, and dynamic foci - and the finding that the famous one is the one its MLL-linked function does not need
AKAP95 sits in three places inside the nucleus: on the nuclear scaffold, in the nucleolus where ribosomes are built, and in small mobile droplets. The scaffold address is the one it is famous for, and it is the one it can do without when it helps switch genes on.
AKT and ATM push ELL between two complexes: K355 acetylation sends it to SEC, ATM phosphorylation sends it to LEC
ELL belongs to two different machines - one that reads out growth genes and one that reads out small housekeeping RNAs. Which machine it joins is decided by a chemical tag, and two different kinases push the tag in opposite directions: a growth signal sends ELL to the growth machine, and a DNA-damage signal sends it to the other one.
Alu repeats and nuclear-matrix anchors in the KMT2A breakpoint region — the genome-feature side of the break
The stretch of DNA where KMT2A snaps is unusually full of short repeated sequences called Alu elements, and it is also physically tied to the nucleus's internal scaffold. Both features change how easily that piece of DNA can break and be mis-joined, though the evidence on which parts break most is genuinely contradictory.
Are the fusions temporary? No: the KMT2A fusion is the one thing that does not change between diagnosis and relapse
Almost everything about this leukaemia can change when it comes back after treatment: which cells it looks like, which markers it carries, which extra mutations it has. The fusion itself does not change, which is why it is the right thing to aim a drug at.
Arginine methylation gates KMT2-complex access to chromatin at two residues, in both directions, independently of the menin tether AND of the WIN motif - a parallel control point, not a rival to menin dependence
Two arginines on the histone tails act as switches: depending on which enzyme methylates them and how, the KMT2 machine can either grip the chromatin or not. Neither switch touches the menin handle or the WDR5 handle this project is built around, so it is a second, separate lock on the same door.
At low drug doses the leukaemia dies through FLT3, not through HOXA9 - a second H3K79me2 pathway the dose-response hides
Everyone assumed a DOT1L drug works by switching off HOXA9 and MEIS1. But at doses far too low to do that, some KMT2A leukaemias still die - and the reason is that the H3K79 tag is also holding up the growth receptor FLT3. That means the drug has two separate dose ranges with two different mechanisms.
At the Hox cluster, menin is required for most of the H3K4 methylation and MLL1 is required for only a subset - the cofactor has a bigger chromatin footprint than the enzyme
You would expect the enzyme to matter more than the helper protein it docks onto. At the Hox genes, the opposite is measured: removing the helper, menin, wipes out nearly all of the chemical mark, while removing the enzyme MLL1 only affects some of it.
ATM phosphorylates ELL at T551, S561 and S589 - inside the OCEL domain this graph already calls essential for MLL-ELL immortalisation
When DNA is damaged, the alarm kinase ATM tags the elongation factor ELL at three spots; the tagged ELL grabs its partner EAF1 harder, clumps with itself, and lets go of the machine that drives genes to be read out. All three spots sit inside the exact piece of ELL that the leukaemia fusion keeps and needs.
ATM phosphorylates ENL at conserved SQ sites and flips it from elongation to Polycomb repression - on sequence every KMT2A::MLLT1 fusion keeps
ENL normally helps genes get read out, but when DNA nearby breaks the alarm kinase ATM tags it and the tagged ENL recruits a silencing complex instead - the same protein switching from 'on' to 'off'. Because KMT2A::MLLT1 fusions keep almost all of ENL, the fusion should carry this switch too.
ATR phosphorylates KMT2A at S516 to break the SCF(Skp2) contact - a phospho-gated degradation switch sitting inside the retained fragment
When DNA is damaged during copying, a sensor enzyme called ATR puts a phosphate on KMT2A at one specific spot. That phosphate stops a disposal machine from grabbing KMT2A, so the protein survives and slows DNA copying down. The spot is on the piece that leukaemia fusions keep.
BAF does not work where the fusion works: most direct MLL-AF9 targets do not respond to Brg1 loss, and the SWI/SNF address is a distal MYC enhancer cluster 1.7 Mb away
You might expect the SWI/SNF machine to be helping the MLL fusion protein at the genes the fusion controls. When somebody checked, most of those genes did not care whether SWI/SNF was there - the machine was working somewhere else entirely, on a cluster of switches more than a million bases away from the MYC gene.
BCOR, not CBX8, is the AHD contact MLL-AF9 cannot lose - and the same two partners bind ENL far more tightly than AF9
Four different proteins compete for the one small patch of AF9 that the cancer protein keeps, and they cannot bind at the same time. Break just one of those contacts - the one with BCOR - and the leukaemia never forms in mice, even though the cells grow faster in a dish.
Bivalent and primed chromatin at fusion target loci — and why the H3K4/H3K27 story is weaker than usually told
Some genes carry both an 'on' and an 'off' chemical tag at once, poised to go either way. A subset of the fusion's target genes are like this — but the evidence that this poised state is the main story is weaker than the textbooks suggest.
Blocking where DOT1L binds instead of what it does - the modality that removes the mark only at fusion targets
Instead of switching DOT1L off everywhere, you can block the handshake that brings it to the leukaemia's genes. One such compound left the rest of the cell's H3K79 tagging untouched and removed it only from the genes the fusion controls - which is the selectivity a whole-cell enzyme inhibitor can never have.
BMI1, CDKN2A and CDK6 are one circuit, not three genes - and in KMT2A-rearranged disease it has a loop nobody has measured
One gene, CDKN2A, makes the brake on cell division; BMI1's job is to hold that brake off and CDK6 is what the brake grips, so the three are a single loop rather than three separate targets. In KMT2A-fusion leukemia the brake is usually still physically present and merely held down, which is the situation where the loop matters most and where nobody has measured it.
Bortezomib is called an NF-kB inhibitor, and in MLL leukaemia it does two different things in two lineages — neither of which is reliably NF-kB in the lymphoid one
Bortezomib is routinely described as a way to block NF-kB, because it stops the brake protein from being destroyed. In myeloid MLL leukaemia it does look like it works that way. In the B-cell form it works by letting the fusion protein itself build up, which triggers a hidden self-destruct programme — a completely different mechanism with the same drug.
Breakpoint anatomy, and an audit of the prognosis claim
This is the source audit behind the breakpoint question: exactly where KMT2A breaks, how the popular version of the age story got reversed in retelling, and why the claim that one break predicts worse survival rests on only thirty patients.
CALM-AF10 is predominantly cytoplasmic in transfected non-haematopoietic cells - and NUCLEAR in leukaemia cells (title corrected 2026-09-21; the five systems are five instances of one design)
A protein that switches genes on should live in the nucleus. Imaging with tagged constructs in five non-blood cell types puts the CALM-AF10 fusion mostly in the cytoplasm - but in mouse leukaemia cells the full-length fusion was found in the nucleus, so which compartment the fusion occupies appears to depend on the cell it is in.
CALM-AF10 needs BMI1 - and so does MLL-AF9, so this is a shared polycomb dependency and not a point of difference (title corrected 2026-09-21)
CALM-AF10 leukaemia cells depend on a protein called BMI1 that keeps stem-cell genes switched off and stem-cell behaviour switched on. Removing it, or blocking it with a drug, stops the leukaemia in mice.
CALM-AF10 reaches DOT1L through exactly the same OM-LZ interface as KMT2A::MLLT10 - the recruitment transfers, the addressing does not
Both the KMT2A::MLLT10 and the CALM::AF10 leukaemia fusions keep the same small piece of the AF10 protein, and that piece grabs the DOT1L enzyme in the same way. What differs is how each fusion decides WHERE on the genome to take DOT1L.
CALM-AF10 strips H3K79 methylation from the rest of the genome while adding it at HOXA - one fusion, two opposite effects on the same mark
The CALM-AF10 fusion grabs the DOT1L enzyme and takes it away from most of the genome, so the chemical tag DOT1L writes almost disappears everywhere except at a few growth genes. Cells with less of that tag break their chromosomes more easily.
Canonical versus non-canonical NF-kB in KMT2A-rearranged leukaemia — the two arms have opposite signs, and the two papers usually read as contradicting each other actually agree
NF-kB is two different circuits sharing a name. In MLL leukaemia the fast one keeps the disease going and the slow one holds it back, so saying 'target NF-kB' means nothing until you say which. The two famous papers on this look like they disagree and, read properly, they do not.
CCNT1 and CCNT2 have no histone substrate and reach chromatin on someone else's back: Brd4 recruits cyclin T1 and Cdk9 to acetylated chromatin and drives Pol II Ser2 phosphorylation - a borrowed address, which is a third annotation state the corpus has no word for
Cyclin T1 and T2 are the partner halves of the kinase that switches paused polymerase into productive elongation. They do not touch histones themselves - a BET protein reading acetylated histones brings them to the right place on the genome.
CD65s has no known synthesising enzyme, and the three fucosyltransferases nearest to it are explicitly excluded
CD65s is a sugar structure on the surface of these leukaemia cells, not something a gene makes directly, so the only way to connect it to the rest of this graph is to find the enzyme that builds it. The surface beat resolved two candidate enzymes and deliberately refused to name either, because no paper attributes the structure to them. Checking that properly, the refusal was right, and the nearest enzymes are not merely unproven - a 2018 paper tested three of them and reports that none of them makes it.
Chromatin loop anchors are DNA-break hotspots - and the two papers that established it disagree about why
The genome is folded into loops, and the places where a loop is pinned shut turn out to be where the DNA most often snaps. Two teams showed this within two years of each other and reached opposite conclusions about whether the snapping needs the gene to be switched on.
Chromocentres and pericentromeric satellite heterochromatin — no KMT2A evidence, but the AT-hook argument says there should be
Chromocentres are dense clumps where the AT-rich repetitive DNA near chromosome centres from different chromosomes bundles together. Other proteins with the same DNA-gripping hooks as KMT2A pile up there and hold the clumps together — but nobody has ever reported KMT2A or its fusions at a chromocentre, and that silence is itself informative.
CK2 phosphorylates KMT2A at T2724/S2726 to license taspase1 cleavage - and that site is in the half the fusion throws away
Healthy KMT2A gets cut in two by an enzyme called taspase1, and being cut makes it less stable. CK2 has to add phosphate right next to the cut site first. Blocking CK2 leaves KMT2A uncut and abundant, and the extra healthy protein pushes the leukaemia fusion off the DNA.
Clec12a is REQUIRED for NUP98::NSD1 AML pathogenesis - a marker turning out to be a dependency, in a different fusion family, and untested in KMT2A-r
CLEC12A was thought of only as a flag on the surface of leukaemia cells - something to aim a drug at, not something the cell needs. In a different kind of leukaemia (driven by the NUP98::NSD1 fusion), deleting it made the cells die and made the disease much less able to take hold in mice, which means it may be a genuine vulnerability and not just a label.
Competing model 1: the reciprocal product starts the leukaemia and then leaves
One model says the mirror-image protein is the match, not the fire. Switching it on for only two days permanently changed which genes a cell could reach - and then it was no longer needed.
Competing model 2: the reciprocal allele is lost as the leukaemia evolves, and losing it selects for something worse
A second model says every patient starts with both fusion genes and some leukaemias throw the mirror-image one away as they grow. The ones that throw it away behave worse.
Complex I inhibition triggers the mitochondrial integrated stress response through DELE1 and HRI, and that is what creates the MCL-1 dependency - a metabolic drug manufacturing an apoptotic vulnerability
Damage a cell's mitochondria in a specific way and they send out a distress signal that changes which survival proteins the cell is leaning on. In leukaemia this makes the cell suddenly depend on MCL-1, so a drug that blocks MCL-1 - which does very little on its own - becomes lethal when given with a mitochondrial drug.
Condensates in KMT2A-fusion leukemia — an honest verdict: not established for AFF1/AFF4/AF9, load-bearing only in ENL-mutant and NUP98 fusions
There is an exciting idea that fusion proteins work by forming liquid droplets in the nucleus. For KMT2A fusions this has essentially not been shown. The good droplet evidence comes from different proteins — mutant ENL and NUP98 fusions — and should not be borrowed.
CTCF keeps CALM-AF10 leukaemia undifferentiated through TGM2 - an architectural dependency, not a menin one
A protein that organises how DNA is folded, CTCF, is needed both to start and to keep CALM-AF10 leukaemia going in mice. It works by keeping one gene, TGM2, switched on, and blocking TGM2 makes the leukaemia cells mature into harmless macrophage-like cells.
CTCF/cohesin boundaries and the HOXA topological domain — the 3D scaffold the fusion program runs on
DNA is folded into loops and neighbourhoods, and the HOXA genes that KMT2A fusions switch on sit inside one such neighbourhood held together by proteins called CTCF and cohesin. In this leukemia one experiment shows the neighbourhood matters - but when scientists destroy these neighbourhoods everywhere at once, most genes carry on as before, so the fence is not automatically the thing holding the disease up.
CXCL12-CXCR4 retention — the axis everyone calls the niche dependency, and the CRISPR screen that says the ligand is not required
Bone-marrow support cells release a chemical called CXCL12, and blood cells that carry its receptor CXCR4 are drawn to it and held in place; block the receptor and the cells wash out into the bloodstream within hours. This is the mechanism behind the drug plerixafor, and the standard explanation for why leukaemia cells in the marrow are harder to kill than ones in the blood.
Cytoplasmic localisation in KMT2A fusions: the SEPTIN case has an image, the GPHN case has none, and neither partner wins the compartment
A few of MLL's fusion partners are proteins that normally work outside the nucleus - at cell membranes or on the scaffolding that holds nerve-cell receptors in place. When they are fused to MLL the chimera is pulled into the nucleus anyway, and for the gephyrin fusion nobody has ever taken the picture at all.
Decitabine degrades DNMT1 - and in ALL cells the same treatment lowers KMT2A protein, with the shared CXXC-family DNA-binding module offered as the reason
Decitabine gets built into DNA and causes the cell to destroy DNMT1, the enzyme that maintains methyl marks. A 2025 study found that in leukaemia cells the same treatment also lowers the amount of KMT2A protein - and only the cells that lost KMT2A actually responded to the drug.
DEK::NUP214 is the sixth lesion on the XPO1 address - and the only one where the drug response was measured in primary patient cells
A rare AML caused by fusing DEK to the same nuclear-pore protein found in SET::NUP214 turns out to sit on the FOXC1 and HOX genes and to need the export receptor XPO1 to stay there. Among 57 patients' cells screened against 500 drugs, the ones with this fusion responded selectively to XPO1 blockers.
Directed extraction of P-TEFb from the reservoir: HIV-1 Tat is the worked example, and the only cellular candidate rests on one unreplicated paper
A virus solved this problem long before anyone asked whether a leukaemia had. HIV's Tat protein grabs the off-switch RNA directly, pushes the braking protein off it, and walks away with the elongation enzyme. Only one human protein has ever been proposed to do the same, in a single unreplicated study.
DNA damage raises menin levels through ATM/ATR phosphorylation, and the residue has never been mapped
After DNA damage, menin protein levels go up, and the reason given is that the damage-sensing enzymes ATM and ATR put phosphate on it. Which spot they use has never been identified.
DNA-PKcs phosphorylates AF9 at S395 and releases it from the super elongation complex - the only AF9 regulatory element the fusion keeps
When a cell's DNA is damaged, a repair kinase puts a phosphate on one exact spot of AF9 and that makes AF9 let go of the machine that drives genes to be read out. That spot is inside the piece of AF9 the cancer fusion protein keeps, so the cancer protein probably has the same off-switch and nobody has checked.
Do different breaks give different biology? A verdict on partner identity vs breakpoint position
Yes for the partner gene: which gene KMT2A joins strongly predicts whether the child gets lymphoid or myeloid leukaemia, how old they are, and how they do. Much weaker for where inside KMT2A the break lands: that changes the gene-expression flavour and correlates with age, but the evidence that it changes survival rests on about thirty patients.
Docking onto pre-bound XPO1 is a general route to leukaemic chromatin, used by at least seven different lesions - not three coincidences
Several completely different leukaemia mutations end up switching on the same set of HOX genes. The thing they share is not a DNA-binding protein but a lift operator: the export carrier XPO1 is already parked on those genes, and each lesion carries a different handle that grips it.
Does the fusion 'lose restraint'? The pattern of KMT2A-side loss, and why the tidy reading does not survive contact with the ledger
There is an attractive story that the leukemic protein keeps KMT2A's ability to find genes and throws away everything that would make it stop. The tidy version is partly true, but most of the discarded parts have never been tested, and the best-evidenced loss is not a brake at all.
Does the fusion need the PWWP arm? An artificial CXXC + H3K36me2-reader + AF4-interaction protein substitutes for a whole MLL fusion - but LEDGF itself is not required for the fusion to reach chromatin
A stripped-down artificial protein built from a DNA-recognising piece, a histone-mark-reading piece and an elongation-recruiting piece works as well as a real leukaemia fusion. That says the histone-reading function matters. It does not say LEDGF has to be the protein doing it.
DOT1L and H3K79 in KMT2A::MLLT1: a direct partner contact, mutually exclusive with the elongation complex, and the only fusion with a DOT1L-inhibitor complete remission
Unlike the commonest KMT2A fusion, this one grips the enzyme DOT1L directly, so the chemical mark that enzyme writes is stronger on its target genes. When a DOT1L-blocking drug was tested in 51 adults, the only two people who went into complete remission both had this chromosome swap - and one of them was confirmed to carry this exact fusion.
DOT1L and H3K79 in KMT2A::MLLT3: recruitment is a graded dial that sets methylation and transformation - and the fusion it was discovered in produced no remission in the DOT1L trial
The amount of DOT1L this fusion pulls in is not on or off: weaken the grip a little and you get a little less chemical mark and a little less cancer. That graded behaviour is exactly the shape that makes a drug hard, and the one DOT1L drug tested in patients helped nobody with this particular fusion.
DOT1L does things that have nothing to do with methylation - which is why blocking its active site was never going to remove all of it
DOT1L's day job is adding a chemical tag to histones, but it also acts as a physical scaffold that holds other machinery in place, and that second job does not need the tag at all. A drug that jams the enzyme's active site leaves the scaffold standing, which is one good explanation for why such a potent drug did so little.
DOT1L leaves chromatin before menin does when the fusion is degraded, and the ordering survives a change of instrument that destroys the magnitudes - ChIP-qPCR at selected loci, target-restricted read counts and genome-wide peak presence all rank DOT1L ahead of menin while disagreeing about every number
When the fusion protein is destroyed one of its partners comes off DNA quickly and another lingers, and three completely different ways of measuring this disagree wildly about how much comes off while all three agree on which one goes first. That kind of agreement is worth more than any of the individual numbers.
DOT1L methylates H3K79 non-processively - so the three states are one distribution, not three signals
DOT1L does not hold onto a histone and add all three methyl groups in one go; it lets go between each one. That means the amount of mono-, di- and tri-methylation you see is just one number - how much enzyme activity happened there - split three ways, rather than three separate messages.
DOT1L's localization record is a mouse kidney paper wearing a human label - and the one image of it in a leukaemia cell is a supporting figure a title search cannot see
DOT1L is the enzyme every KMT2A-fusion leukaemia depends on, and the database entry saying it can be in the cytoplasm is not a human measurement at all: it is copied across from mouse kidney cells, where over-expressing a partner protein pushed the mouse enzyme out of the nucleus. The only picture of DOT1L inside a leukaemia cell is a supporting panel in a paper about something else.
DPY30 has a second job at the trans-Golgi network, and the authors who found it could not put ASH2L or RBBP5 there
DPY30 is known as a small piece of the machine that puts an activating mark on DNA packaging proteins in the nucleus. It also sits on a membrane compartment in the cytoplasm, where it helps ferry a receptor back from endosomes - and when it is removed, that traffic goes wrong. The same group looked for two of its nuclear partners there and could not find them.
ELL is in Cajal bodies - or nuclear speckles, depending which paper you read - and the MLL-ELL fusion empties them
ELL normally collects in small nuclear structures together with its partner EAF1, and that concentration depends on the cell actively transcribing. When ELL is fused to MLL in leukemia, those structures fall apart - which makes this one of the very few cases where a KMT2A fusion has been shown to move something in the nucleus rather than just switch genes on.
ELL: the discarded N-terminal EAF-binding site, and a second subtraction the graph had not recorded — the fusion delocalises endogenous EAF1
ELL normally works with two partner proteins, EAF1 and EAF2, docking them at two separate sites. The leukemia fusion keeps one docking site and loses the other, and on top of that it pulls the remaining partner away from where it normally sits in the nucleus.
Endogenous aldehydes and silenced ALDH2 — the rival explanation for an FA dependency in acute leukaemia, and the only one with an experiment behind it
Cells constantly make small amounts of formaldehyde and acetaldehyde as by-products, and these chemicals glue the two strands of DNA together where they should slide apart. Two systems defend against this: an enzyme that destroys the aldehyde, and the Fanconi crew that repairs the damage it causes. Some leukaemia cells switch the first defence off, which leaves them completely dependent on the second — and that, not anything to do with the KMT2A fusion, may be why this project's screen found the Fanconi genes.
Enhancer usage in KMT2A-rearranged leukemia — occupancy and amplification, not hijacking, and not super-enhancers
The fusion does not build new control switches for genes. It parks on switches the cell already has — including hidden ones inside genes — and turns their volume up by wiring them more tightly to the gene's start site.
Enhancer-promoter contacts at KMT2A-fusion target loci — what the fusion uses, what it does not, and what survives the drug
Genes are switched on by distant DNA switches that physically touch the gene's start. In KMT2A-fusion leukemia a specific chemical mark keeps some of those touches in place, and removing the mark breaks the contact and the gene goes quiet. But blocking other proteins stops the gene being read while leaving the contacts intact, and at other genes switching an enhancer on actually moves it further away - so touching and switching-on are not the same thing, in either direction.
ENL's nuclear annotation has no citation at all, and every picture of ENL in a nucleus is of a mutant or a fusion, never the wild-type protein
ENL is one of the commonest KMT2A fusion partners, and the standard protein database says it lives in the nucleus without citing a single experiment for it. Every published image of ENL forming structures inside a nucleus is of a mutated or fused version of the protein, deliberately over-produced - so what normal ENL looks like inside a normal nucleus is, as far as this search could establish, unrecorded.
EPS15 (AF1p): the EH domains may or may not be discarded, and either way nobody has shown it matters
EPS15 normally helps cells pull receptors in from their surface, using three gripper modules at one end. Reports disagree about whether the leukemia fusion keeps those grippers, and no study has TESTED whether losing them matters - though one 2000 paper did propose that this endocytic axis is relevant to MLL-fusion leukaemia, and the transforming part identified so far is a different, retained segment that makes the fusion clump together.
Erasing H2Bub: three deubiquitinases share two adaptors, and removing the enzyme the field names lowers the mark instead of raising it
The ubiquitin tag on H2B is taken off again by a small module inside the SAGA complex, and the level on any gene is the balance between putting it on and taking it off. Confusingly, deleting the enzyme most people name — USP22 — makes the mark go down, not up, because two other enzymes compete for the same helper proteins.
ETO2::GLIS2 rewires chromatin in stages and hijacks the osteogenic homeobox factor DLX3 as its first required step — a genome-edited human iPSC model that resolves leukaemia initiation into an ordered sequence
Researchers built the exact chromosome inversion that causes a rare, aggressive infant leukaemia into human stem cells, then watched what happened step by step as those cells tried to become blood. The fusion protein's very first move is to switch on DLX3, a gene that normally builds bone, and if DLX3 is deleted the leukaemia never starts.
EWSR1::FLI1 — the comparator that shows condensate formation has an optimum, and that forcing a fusion into the nucleolus switches it off
In Ewing sarcoma the fusion protein's sticky disordered tail recruits a chromatin-remodelling machine to repetitive GGAA stretches, and that tail's phase behaviour is required. But the same group that imaged it found that pushing the stickiness higher switches the target genes off — and that deliberately dragging the fusion into the nucleolus, a natural droplet, stops it driving cancer at all.
Five KMT2A-free NUP98 fusions bind MLL1 and the NSL complex — and one unbiased mass-spectrometry study failed to reproduce it
Five different NUP98 fusions, none containing any piece of MLL1, were shown to physically grab MLL1 and a partner complex called NSL, and deleting MLL1 delayed the resulting leukaemia in mice. A later study that searched for every protein binding these fusions, without deciding in advance what to look for, did not find MLL1 at all.
FLT3 is the second largest KMT2A-selective dependency and it is HEAVIER in B-lineage than in myeloid cells
FLT3 is usually discussed as a target in myeloid leukaemia. In this project's own measurement the leukaemias that need it most among KMT2A-rearranged lines are the B-cell ones.
FLT3-ITD signals from inside the cell, not from the surface - and whether it is the ER or the Golgi decides which pathway fires
A healthy FLT3 receptor sits on the cell surface and waits for a signal. The mutant form found in leukemia never gets there: it is held inside the cell, and from there it switches on a different set of growth signals than it would from the surface. Drugs that block the kinase let it escape to the surface again, which means the drug changes where the protein is as well as what it does.
Fusion-positive clones that never become leukaemia: persistence in remission, spontaneous clearance, and silent rearrangements in healthy people
Carrying the fusion is not the same as having leukaemia. There are people who carry a fusion-positive clone for years without being ill, one documented baby whose fusion-positive clone simply vanished, and evidence that broken-and-rejoined KMT2A DNA turns up in healthy adults without ever producing a working fusion protein.
GATA2 outlier activation in KMT2A::USP2 B-ALL - the only transcriptional output ever measured for this fusion
In a large set of childhood leukaemias, two patients had wildly abnormal levels of a stem-cell gene called GATA2, and both turned out to carry the same rare KMT2A::USP2 fusion - the only two in the whole cohort. No other KMT2A fusion did this, which is the single piece of evidence that this fusion switches on a distinctive gene.
GSK3 is required by KMT2A-rearranged leukaemia and nobody has shown it touches KMT2A
An enzyme called GSK3, normally thought of as a brake on cancer, turns out to be needed by KMT2A-rearranged leukaemia specifically. But the reason runs through a cell-cycle brake protein, not through KMT2A itself.
Guanine nucleotide supply sustains the LEDGF/menin/MLL-fusion complex - the one place in this graph where metabolism sits UPSTREAM of the menin interface
The leukaemia's central machine needs two scaffold proteins, menin and LEDGF, to hold it on the DNA. Starving the cell of the raw material for one of the four RNA letters makes it build fewer ribosomes, which makes it make less menin and less LEDGF, which makes the machine fall apart - so a metabolic drug and a menin drug attack the same thing from two ends.
H2A.Z - two genes of identical length, one KMT2A paper, and the measured quantity is a ratio nobody has repeated
H2A.Z is a swap-in version of a core histone that sits at the nucleosomes flanking a gene's start, where it makes those nucleosomes easier to move. In KMT2A-fusion leukaemia one study found that the fusion protein brings an enzyme to the Hoxa9 gene and changes how much of the H2A.Z there is acetylated - and nobody has repeated it, in any other fusion, in any human cell.
H2AK119ub - the mark every PRC1.1 claim in this graph rests on, named in fifteen entries and written in none until now
Polycomb complex 1 silences genes by hanging a small protein tag called ubiquitin on histone H2A at position 119, and that tag - not the better-known H3K27me3 - is the direct product of the machine this graph keeps calling PRC1.1. Fifteen entries here already argue about what happens to this tag when a menin or DOT1L inhibitor is given, and none of them had ever said what the tag is.
H2BK120ub (H2Bub1) — the ubiquitin tag on histone H2B that licenses H3K79 methylation, and the mark this graph was missing
A whole ubiquitin protein is clipped onto lysine 120 of histone H2B while a gene is being read, and that tag is what allows DOT1L to place the H3K79 mark that KMT2A-fusion leukaemia depends on. It is written by the RNF20/RNF40 enzyme as the gene is transcribed and stripped off again by deubiquitinases, so the level on any gene is a balance rather than a switch.
H2BK34ub — a second ubiquitination site on H2B that also stimulates DOT1L, written by MSL1/MSL2, and important at HOXA9 and MEIS1
There is a second place on histone H2B that gets a ubiquitin tag, put there by a different enzyme, and it turns on DOT1L just as effectively. It was shown to matter for switching on HOXA9 and MEIS1 — the two genes KMT2A-fusion leukaemia is defined by.
H3.3 - the graph's most-named epigenetic object with no entry, 325 mentions, and not one measurement in a KMT2A-rearranged cell
H3.3 is a spare version of a core histone that cells slot in wherever DNA is being actively read, without waiting for the cell to divide. This graph's whole argument about how fast a drug can clear the H3K79 mark depends on how quickly H3.3 is swapped in - and despite naming it 325 times, nobody has ever measured it in a KMT2A-fusion leukaemia cell.
H3K27ac - named in 40 entries with no entry of its own, and the strongest KMT2A::AFF1 measurement of it is invisible to every abstract search
H3K27ac is the chemical tag that marks a gene switch (an enhancer) as switched on, and it is the most-used readout in the whole enhancer field. In leukaemia driven by a KMT2A fusion it has been measured directly - in cell lines, and in four patients including one infant - and the result is that a DOT1L inhibitor strips the tag off the fusion's own enhancers while leaving the rest of the genome untouched.
H3K27me3 - the repressive mark this graph argued about for seventy-seven cycles without ever writing an entry for it
H3K27me3 is the chemical tag that means 'keep this gene switched off', and it is the direct opposite number to the H3K79 tag that this leukaemia is defined by. It is unusual among histone marks because the machine that writes it also reads it, so once a patch exists it spreads itself.
H3K36me2/3 - the mark that physically switches PRC2 off, that LEDGF reads to reach chromatin, and that a comparator fusion uses to keep EZH2 off HOXA
H3K36 methylation marks the body of a gene that is being read. It is also a physical off-switch for the Polycomb machine: where this mark is, the repressive machine cannot work. One leukaemia fusion exploits exactly that to keep its target genes switched on.
H3K4me2 - the LSD1 substrate, and the only histone mark ever compared head-to-head between two KMT2A fusion partners in human cells
H3K4me2 is a half-methylated tag near active genes, and it is the tag the enzyme LSD1 rubs off. LSD1 is one of the few drug targets in KMT2A-fusion leukaemia whose whole rationale is a histone mark - and in the one study that compared two different KMT2A fusions directly, the amount of this mark at a differentiation gene was the thing that differed between them.
H3K4me3 - the mark KMT2A's own SET domain writes, that the fusion cannot write, that an interface in this graph already binds, and that switches PRC2 off
H3K4me3 is the 'this gene is on' tag sitting at the start of active genes. The healthy KMT2A protein writes it - but the leukaemia fusion has lost the part that does the writing, which is the central paradox of this disease: the enzyme's own product is not what the fusion delivers.
H3K79 methylation is what keeps Polycomb off - and that, not gene silencing, is what a DOT1L inhibitor actually undoes
Genes that a leukaemia keeps switched on are protected from a silencing system called Polycomb, and the H3K79 tag is the protection. When a DOT1L drug strips the tag away, Polycomb moves in and shuts the genes down - and it does this gradually, as the tag is lost, rather than the moment the drug binds.
H3K79me1 - the first methyl group, the one H2B ubiquitination is not needed for, and the state HOX loci fall back to as blood cells mature
H3K79me1 is a single methyl group on lysine 79 of histone H3, the first of up to three that DOT1L can add. It is the state that normal blood cells settle into at their HOX genes as they mature, and unlike the higher states it does not need the ubiquitin tag on histone H2B to be made.
H3K79me2 - the di-methyl state the leukaemia actually runs on, and the one menin turns out to govern only half of
H3K79me2 is two methyl groups on the same histone position, and it is the specific state that sits in broad blocks over the genes a KMT2A fusion switches on. Measuring it is how the field decided DOT1L was a drug target, and a 2023 screen showed that only part of it is controlled by menin - the rest is put there by the cell's own machinery.
H3K79me2 builds up as B cells mature - so in B-ALL the drug's target is what the rescued compartment runs on
In normal mice, the H3K79 tag accumulates as B cells grow up, and without DOT1L the pre-B stage does not expand properly, the antibody repertoire narrows and a whole class of B cells fails to form. Since infant KMT2A leukaemia is usually a B-cell leukaemia, a DOT1L drug is attacking the same machinery the healthy B cells need.
H3K79me3 - the third methyl group, the one that needs the most enzyme flux and the one human DOT1L makes least of
H3K79me3 is the fully methylated version of the same histone position, and it is the hardest state for the enzyme to reach because each methyl group has to be added in a separate binding event. In human cells it is the rarest of the three states, which makes it the first thing an inhibitor removes and a poor thing to measure a drug against.
H3K9ac - the mark KMT2A fusions pile up on their own targets, the one the AF9 reader actually binds, and the one a DOT1L inhibitor strips off first
H3K9ac is an acetyl tag at the start of active genes. In KMT2A-fusion leukaemia it is piled abnormally high on exactly the genes the fusion switches on - HOXA and MEIS1 are the two biggest patches in the genome - and when a DOT1L inhibitor is given, the cell strips that tag off those genes faster than off any others. It is also the tag the fusion partner AF9 physically grips.
H3K9me2 - named sixty times in this graph with no entry of its own, the mark that defines the lamina-associated domains the fusion's targets sit beside, and the only H3K9 methylation state ever measured at a KMT2A fusion's own binding sites
H3K9me2 is the two-methyl silencing tag that covers huge blocks of the genome, especially the parts pinned to the edge of the nucleus. It is the tag a DOT1L drug pushes onto the leukaemia genes, the tag that floods the KMT2A gene's fragile stretch when its eraser is removed, and it has never once been measured in a patient's infant leukaemia cells.
H3K9me3 - measured in KMT2A::AFF1 cells, in a supplementary figure, and it rises GENERALLY where H3K27ac falls SPECIFICALLY in the same experiment
H3K9me3 is the tag that locks a stretch of DNA into silent, tightly packed chromatin. In KMT2A-fusion leukaemia it is the brake: the enzymes that write it push down the very genes the fusion needs, and leukaemia stem cells carry less of it than normal blood stem cells. When a DOT1L inhibitor is given it goes up - but everywhere, not just at the genes the drug is aimed at.
H3R2me2a - the asymmetric arginine mark two residues from H3K4 that LOCKS the WRAD module off the histone tail, and the first arginine mark node in this graph
H3R2me2a is a chemical tag sitting two amino acids away from the spot the KMT2A machine works on, and while it is there the machine cannot grip the histone tail at all. It is written by a different enzyme, PRMT6, and it is the reason the tag next door - the 'gene on' mark - cannot be made.
H3R2me2s - the symmetric mark on the SAME residue as H3R2me2a, which permits rather than enhances WRAD access, and whose 'enhancing' half this graph inherited from the paper that was later revised
H3R2me2s is the mirror-image version of the tag that blocks the KMT2A machine: same spot on the histone, same two methyl groups, arranged differently - and it does not block. Whether it actively pulls the machine in, or merely fails to keep it out, is the point on which two papers disagree.
H4K16ac - the one mark rule 80 actually rescued in this beat: elevated at MLL-AF9 targets in a supplementary figure, and the reason MLL1's own enzyme turns out to be dispensable
H4K16ac is a single acetyl group on histone H4 that physically stops chromatin folding up, so it is the one histone mark with a direct structural job rather than just a signalling one. In KMT2A-fusion leukaemia the enzyme that writes it is required for the disease, and the mark itself is piled up on exactly the genes the fusion turns on - a fact that sits in a supplementary figure of a paper about a different protein.
H4K20me1 - the transcription- and replication-linked H4 state, whose gene-body level tracks Pol II; it HAS been profiled genome-wide in human haematopoietic stem cells twice, and a query scoped to 'leukemia' cannot see either dataset
H4K20me1 is a single chemical tag on the packaging protein histone H4. Where it sits along the body of a gene tracks how hard that gene is being copied, and it also tells the cell where to start copying its DNA. This project needed an entry for it because a 2025 paper found a protein that reads it and had nowhere to point.
H4K20me2 - the abundant repair-linked H4 mark, the 53BP1 docking site, and the first H4K20 state in this graph; never once measured in a KMT2A-rearranged cell
H4K20me2 is a chemical tag on histone H4 that acts as the landing pad for 53BP1, the protein that marks broken DNA for repair. Until this entry the graph had eighteen histone-mark records and not one of them was any H4K20 state, and no published study has ever measured it in a KMT2A-rearranged leukaemia cell.
H4R3me2a - the activating arginine mark that a KMT2-family enzyme's OWN tandem PHD module reads, where the Kabuki-syndrome mutations sit, and which no KMT2A-fusion study has ever measured
H4R3me2a is an activating tag on histone H4, written mainly by PRMT1, and it is the tag a reading module inside one of the KMT2 enzymes grabs in order to work on chromatin at all. The mutations that cause Kabuki syndrome sit in that reading module and weaken exactly this grip.
H4R3me2s - the repressive symmetric arginine mark that recruits DNMT3A, blocks a KMT2 enzyme's own PHD module, and is the ONLY arginine mark ever measured in a KMT2A-fusion system
H4R3me2s is a silencing tag on histone H4 that physically pulls in the enzyme which adds methyl groups to DNA itself, locking a gene off. It is also the one arginine tag anybody has actually measured in a cell carrying a KMT2A fusion - it fell when the fusion's partner complex was removed.
HEXIM2 quantitatively compensates for HEXIM1 loss to hold the 7SK-bound P-TEFb pool constant - which makes single-gene knockdown of either a buffered experiment, and is the reason a HEXIM1-only measurement cannot be read as the brake's strength
HEXIM1 and HEXIM2 both lock the transcription accelerator into its parked state, and they stand in for each other. Remove HEXIM1 and HEXIM2 takes over so completely that the amount of parked accelerator does not change - so an experiment that removes only one of them will look like nothing happened.
Hitting the same axis at two points was predicted to be sub-additive; the one published test found synergy in all four KMT2A-rearranged lines
If a menin inhibitor and a CDK9 inhibitor both block the same chain of events, blocking it twice should add little. The one experiment that tested this in KMT2A-rearranged leukaemia cells found the opposite: the two drugs together did more than the sum of their parts in every cell line tried.
How a cytoskeletal partner reaches chromatin at all — the KMT2A N-terminus drags it in, and the traffic runs both ways
A septin or a junction protein has no business in the nucleus, so the obvious objection to these fusions is that the partner can never reach DNA. The answer is that it does not have to: the KMT2A half carries its own nuclear address labels and pulls the partner in — and in at least one case the fusion also drags the normal copy of the partner in with it, which starves the cytoplasm of it.
How a spliced fusion works: the break lands in front of MLLT1, the gene is never cut, and there is nothing left to make a reciprocal product from
In half of these patients the chromosome does not break inside the partner gene at all - it breaks just in front of it, so the partner gene arrives whole and the KMT2A promoter simply reads straight through into it. Because the partner gene was never cut, the other derivative chromosome has no partner-gene piece on it, and no mirror-image fusion can exist.
How BAF maintains KMT2A-rearranged leukaemia: it holds the blast undifferentiated, and it does so in other leukaemias too
Take the SWI/SNF motor away from an MLL-rearranged leukaemia cell and it stops dividing, turns into something resembling a normal white blood cell, and dies. The same happens in several other kinds of leukaemia, which is the part usually left out.
How general is subtraction across KMT2A fusions? Common enough to matter, too heterogeneous to be one thing, and mostly invisible to the way this project searches
Across seven KMT2A leukemia partners, losing something contributes in about half — but the losses are of three different kinds, and only one of them can be spotted by looking at the protein sequence. The project's way of hunting for drug targets can see that one kind and is blind to the other two.
How HOXA9 binds DNA: a hexapeptide into a PBX pocket, a 1.9 A structure that is mouse HoxA9 plus human Pbx1 homeodomains, and a second route that makes the motif redundant
HOXA9 on its own is a poor, unfussy DNA binder. It gets its grip by docking a short peptide loop into a pocket on a partner protein called PBX, and the pair then reads a longer stretch of DNA together. There is a crystal structure of that pair - but it is mouse HOXA9 with human PBX1, and only the small DNA-binding pieces of each, which matters for anyone designing against it.
How menin holds the beta cell still: JunD recruits it, with HDAC3, onto the Pbk promoter
In insulin-making cells, a protein called JunD drags menin and a histone-stripping enzyme onto the switch for a growth kinase called Pbk, and keeps that switch off. A menin-blocking drug breaks that grip and the growth kinase comes on.
How menin silences JunD: pocket binding plus uncoupling from MAP kinase signalling
Menin holds JunD in a groove and stops the cell's growth signals from switching it on, so the gene programmes JunD drives stay quiet. Two separate blocks are involved, one for each of the main growth-signal pathways.
How much HOXA9 and MEIS1 is actually required? Partial dependence, a Hox code, and why this caps what any single target can do
You might expect that removing HOXA9 would stop the leukemia, but mice without it still develop MLL-AF9 leukemia at the same rate, just with more immature cells. The program is buffered, so partial knockdowns only ever give partial answers.
How much sugar the cell has decides how much DOT1L there is - a metabolic dial on the leukaemia's mark
DOT1L carries a sugar tag at one spot near its tail, and that tag is added when glucose is plentiful. The tag stops a disposal enzyme from finding DOT1L, so well-fed leukaemia cells keep more DOT1L, make more of the H3K79 mark, and grow faster - and starving that pathway makes them more sensitive to the DOT1L drug.
How P-TEFb gets out of the reservoir: a dephosphorylation, a dissociation, and a reassembly on whichever scaffold is waiting
Cells keep a large stock of the elongation enzyme locked away and let it out when they need to switch genes on fast. Getting it out is not one step: the enzyme is chemically untagged, the particle falls apart, the two halves of the enzyme separate, and then they are rebuilt on whichever scaffold grabs them first.
HOXA-high versus HOXA-low KMT2A-rearranged leukemia — the split that breaks the one-mechanism story
Not every child with the same KMT2A fusion has the same disease. Some cases run the HOXA gene program hard; others barely express it at all, and those tend to be the youngest infants with a particular breakpoint.
HOXA9 has its own nuclear bodies, and in monocytic AML they move into the eIF4E body - a subnuclear switch measured in patient marrow and never genotyped
HOXA9, the gene KMT2A fusions switch on hardest, does not just float in the nucleus - it collects into small dots. In normal bone marrow those dots sit apart from another set of dots built around a protein called eIF4E, but in one subtype of adult AML the two merge, and blocking NF-kB pulls them apart again.
ICOSLG on infant t(4;11) blasts converts T cells into regulatory T cells - the one immune-evasion route measured in this exact disease, in this exact age group
Infant leukaemia cells with the t(4;11) rearrangement carry a surface protein that turns ordinary T cells into the kind that switch the immune response off. The more of it a child has at diagnosis, the sooner the leukaemia comes back.
IDH and TET2 mutations are mutually exclusive in AML - so 'IDH2 enriched, TET2 depleted' in one cohort is ONE observation, not two
Two different mutations, in IDH and in TET2, produce the same over-methylated DNA, and leukaemias almost never carry both - one is enough. So if a group of patients has more of one, it will automatically look like it has less of the other, whether or not anything interesting is going on.
If the target promoter set is no more conserved within a patient than between patients, then nothing in the oncoprotein complex remembers the first disease
The cancer protein does not seem to carry a memory of where it used to sit when the leukaemia changes type. Where it lands looks like it is decided fresh by the cell it finds itself in.
IGF2BP3 works in cytoplasmic RNA granules - the only compartment in this beat where a drug is already pointed, and nobody has looked at it in a leukaemia cell
IGF2BP3 is an RNA-binding protein switched on by KMT2A fusions, and it binds the fusion's own messenger RNA. Its job happens in small droplets of RNA and protein in the cytoplasm, including the ones cells build under stress. There is now a drug against it that works in a mouse MLL-Af4 leukemia - and no published picture of where the protein sits in any leukemic cell.
IKK/NF-kB retains MLL on chromatin — BOTH the fusion and wild-type MLL, at two promoters, in mouse myeloid AML, and it is occupancy loss rather than protein loss
Blocking the IKK kinases makes the leukaemia fusion protein fall off the two genes that matter most, even though the fusion protein itself is still present in the cell in normal amounts. The same treatment also knocks normal MLL off those genes and strips away the chemical marks both of them leave behind.
IL-1/IRAK1/4 — the receptor-proximal input the MLL screen actually pointed at, with an Il1r1 knockout in MLL-AF9 mice and an inhibitor that spares normal stem cells
The kinase screen that first linked NF-kB to MLL leukaemia did not find a receptor; it found IRAK3, a relay in the interleukin-1 and Toll-like receptor branch. Later work in AML stem cells shows that deleting the interleukin-1 receptor in MLL-AF9 mice reduces the leukaemia stem cell pool and extends survival, and that a drug against the IRAK kinases kills leukaemia stem cells while leaving normal ones alone.
In a patient, the menin-fusion association is demonstrable before the drug and unmeasurable after it - and the asymmetry is what this pair establishes, not a rate of menin loss
In a baby's leukaemia cells taken before treatment, menin is sitting on the fusion protein's genes and not on the normal gene's. In cells from the same child taken during treatment, nothing is sitting anywhere that can be measured. The useful finding is the first half; the second half is a limit of what can be seen, not a measurement of how much menin left.
In human MOLM13, DOT1L's own occupancy under a menin inhibitor runs 0.1364 / 0.3083 / 0.3555 / 0.3913 across MLL1-N eviction-depth quartiles at Spearman -1.000 - the writer is lost in proportion to eviction depth, and its mark follows with a 2.9-fold smaller span
The enzyme that puts the activating tag on is itself pulled off the DNA in proportion to how much fusion protein the drug removed. So the tag's pattern is most simply explained by the enzyme's: where the fusion goes, the enzyme goes, and the tag decays behind it.
In MOLM13 a menin inhibitor removes three quarters of DOT1L from the fusion's own sites and leaves H3K79me2 there completely intact for three days - the writer and its mark are separable on this timescale, and in RS4;11 they are not
The drug strips away the enzyme that puts a chemical tag on histones, but in one of the two cell lines the tag itself is still fully there three days later. Removing the machine does not immediately remove what the machine made.
In SEM, H3K4me3 responds to menin displacement the way a partially redundant mark should: selectively at menin-bound sites, by about a quarter, and not at all under a DOT1L inhibitor - which bounds how much of a menin inhibitor's lethality H3K4me3 can be carrying
Menin inhibitors kill this leukaemia. If they worked by wiping out a chemical mark at the genes the fusion controls, we would expect that mark to collapse. In the one dataset that measures it in an infant-derived cell line, it drops by roughly a quarter, only where menin was - a real, specific effect, but far too small to be the whole story of how the drug works.
In this fusion family, MHC class I is not silenced - it is elevated and load-bearing, holding NK cells off the leukaemia stem cell
Killer T cells need to see a cell's identity card; a different killer cell, the NK cell, attacks cells that are missing it. Leukaemia driven by these fusions keeps its identity cards up and its NK-activating flags down, which is a good way to hide from NK cells - and in mice, removing the identity card cripples the leukaemia.
IRX1 binds the MLL-AF4 complex non-competitively — menin, LEDGF and DOT1L are all unchanged
IRX1 shuts down the leukaemia protein's output not by knocking its partners off, but by sitting alongside them. Menin, the drug target, stays exactly where it was.
Is MYC the node? Two papers, the same MLL-AF9 model, opposite answers - and the second one failed to rescue
One study found that SWI/SNF keeps MLL-rearranged leukaemia alive by keeping the MYC gene switched on, and proved it by showing extra MYC rescues the cells. A second study, using the same kind of leukaemia, found no change in MYC at all and could not reproduce the rescue. Nobody has resolved it.
Is the drug target always in the retained half? An audit of the project's own framing
This project keeps saying the thing to drug is whatever survives the break. A seven-partner audit found that is true for the two commonest partners and false for at least one other, where the oncogenic event is what got thrown away.
Is TIFAB a direct RelB target? Two papers in the same MLL-AF9 system answer differently
Two studies of the same mouse leukaemia disagree about whether one protein switches another one off directly. A 2021 paper says RelB acts on TIFAB as a direct target; the earlier 2018 study from the same system tested which genes RelB physically sits on and reported TIFAB was not among them. Neither paper mentions the other on this point, and the disagreement matters because a whole chain of reasoning about how this pathway suppresses leukaemia hangs on that one link being direct.
ITGA4 (VLA-4) and NG2 — the niche-adhesion axis of KMT2A-rearranged B-ALL, and the one approved drug that hits it
Leukaemia cells stick to the bone marrow using a surface hook called integrin alpha-4, and in this genetic type of leukaemia that hook works together with a second molecule, NG2, that the fusion protein itself switches on. An antibody already licensed for multiple sclerosis blocks the hook, and in mice carrying children's leukaemia it slowed the disease and made standard chemotherapy work better.
KAT8/MOF is not only a chromatin enzyme: a measured mitochondrial pool binds mtDNA, and the chromatin address depends on which complex it is in
MOF is usually described as a histone enzyme in the nucleus, but a well-controlled study found a real pool of it inside mitochondria, sitting on mitochondrial DNA and controlling respiration - and deleting it wrecks the mouse heart. Where it sits on nuclear chromatin also depends on which of its two complexes it is riding in.
KAT8/MOF sets the quiescence set-point of the blood stem cell - and its catalytic activity is separately required for MLL-AF9 leukaemia, which DepMap also cannot see
The amount of a chemical tag called H4K16 acetylation decides whether a blood stem cell rests or wakes up, and the enzyme that puts the tag on is KAT8. The same enzyme's activity is needed for MLL-rearranged leukaemia to grow in mice. A large cancer screen detects neither, because KAT8 is needed by almost every cell line ever tested.
KMT2A and several of its fusion partners occupy the same spatial cluster in individual nuclei before any break occurs — single-cell diploid Hi-C, RUNX1-dependent, and 3D architecture does NOT predict which fusion type a cell will make
Before a chromosome break ever happens, the KMT2A gene and several of the genes it later fuses to are already sitting next to each other in the same spot inside the nucleus of a normal cell — and reading one nucleus at a time shows that more than one partner can be next to KMT2A in the same cell at the same time. Knocking down RUNX1, the protein that gathers them there, pulls them apart. But the arrangement does not decide which partner a cell will eventually fuse to, or whether the resulting leukaemia will be lymphoid or myeloid.
KMT2A is called an H3K4 methyltransferase, and it is required for the mark at under 5% of the promoters that carry it
KMT2A's job description is writing a chemical tag on DNA packaging at switched-on genes. When you delete it and look genome-wide, more than 95% of the genes carrying that tag still have it. It only matters at a small, specific set of genes - which is why the gene can be a cancer driver without being a general-purpose switch.
KMT2A K1130 acetylation - the clock-controlled, SIRT1-erased site that every KMT2A fusion retains, 17 residues from the CXXC and 208 from the breakpoint
A small chemical tag is added to one specific spot on the MLL1 protein and removed again on a daily rhythm, by an enzyme that senses how much energy the cell has. That spot sits in the part of the protein that every leukaemia fusion keeps.
KMT2A K1130 ubiquitination - the same retained lysine as the clock acetylation site, carrying a chemically exclusive second mark that no KMT2A paper has ever mentioned
The exact spot on MLL1 that carries a daily on/off chemical tag can also carry a completely different tag - one that usually marks proteins for destruction. The two cannot be there at once, so the spot is a switch rather than a dial. No paper about MLL1 has ever said this.
KMT2A K1133 acetylation - the better-evidenced half of the pair, and the one that appears in the CBP/p300 acetylome itself
The second of two neighbouring spots on MLL1 that carry the same daily chemical tag. This one turns up in an experiment designed specifically to find what the enzymes CBP and p300 tag, which is the best clue yet about what puts the tag on.
KMT2A K1235 acetylation - inside the SIRT1-interaction region, retained, and the one member of the Choudhary quartet the source paper named and then never assayed
A third chemical-tag spot on MLL1 that sits inside the same stretch the clock enzymes grab hold of. The paper that studied the other two named this one in the same sentence and then never touched it again.
KMT2A K1364 SUMO - the one modification site inside the breakpoint cluster, retained in some patients' fusion proteins and lost in others
One chemical-tag spot on MLL1 falls right inside the region where the chromosome usually breaks. Whether a patient's fusion protein has it depends on exactly where their break happened - so this single spot is different between patients with the same diagnosis.
KMT2A K2528 SUMO2 - the single best-supported modification site on the whole protein, in the N320 fragment, and LOST in every fusion
The most reliably detected chemical tag on the entire MLL1 protein is not one of the acetylation spots at all - it is a SUMO tag two-thirds of the way along. Every leukaemia fusion throws it away, even though the fusion is often described as behaving like the front half of MLL1.
KMT2A K636 acetylation - the forgotten member of the Choudhary quartet, retained, and outside the SIRT1-interaction region
One of four chemical tag spots found on MLL1 in a 2009 survey. The famous follow-up paper mentioned it once and then ignored it, because it sits outside the stretch the paper was interested in.
KMT2A partial tandem duplication: an addition that only works once the normal copy has been switched off — 'recessive gain-of-function'
This version of the disease has no partner gene at all — a chunk of KMT2A is simply copied and pasted back into itself. The surprise is that the copy is not enough: the cell also switches off its remaining normal copy of the gene, and forcing that normal copy back on kills the leukemia cells.
KMT2A::AFF1 has no mapped interface to DOT1L: it lacks both structurally defined recruitment modules, and the one complex-level route claimed for it is contested by reconstitution
There are two known protein hooks by which a leukaemia fusion can grab the enzyme that writes the H3K79 tag, and the infant fusion MLL-AF4 has neither. Some work says its elongation complex binds the enzyme anyway, other work says the opposite, and nobody has pinned the contact to a specific piece of the fusion.
KMT2A::AFF1 needs a developmental window and a lymphoid-committed cell — the cell type is a condition of transformation, not a consequence of it
The fusion at the centre of this project cannot turn just any blood cell into leukaemia. It works during a particular stretch of development, and its full cancer-causing power only appears once the cell has committed to becoming a B cell.
KMT2A::MAML2 keeps MAML2's activation domain and loses its Notch-binding domain, and suppresses Notch output
MAML2 normally helps the Notch signal switch genes on, using one end to dock onto Notch and the other end to do the switching. The leukaemia fusion throws away the docking end and keeps the switching end, and in the one experiment that has been done it blocked Notch's own target gene instead of helping it.
KMT2A::MLLT1 engages P-TEFb through one 66-residue surface: the ENL AHD, residues 494-559, and L550E breaks it
Everything this fusion does to speed up gene copying runs through a single short tail at the very end of the partner protein, just 66 amino acids long. Changing one amino acid in that tail stops the fusion from binding its partners and stops it causing leukaemia in cells.
LARP7's measured readout is P-TEFb release and Pol II CTD Ser2 phosphorylation, not a histone mark - and the one compound shown to act through it disrupts the particle rather than the protein
LARP7 clamps the 3' end of the 7SK RNA and keeps the transcription accelerator parked. When a drug knocks LARP7 off, the accelerator is released and starts marking polymerase for productive elongation - a change measured on the polymerase, not on the histones.
Last-exon indels that truncate the ZNF384 C-terminus phenocopy the fusion - so the fusion partner may not be the necessary element
Two children had leukaemia that looked and behaved exactly like the ZNF384 fusion subtype, but they had no fusion at all - just a small insertion-deletion near the end of the ZNF384 gene. Standard software missed both; a human had to go and look at the raw sequencing reads.
LEDGF as a survival factor rather than a tether: it is upregulated in chemoresistant AML blasts, protects against daunorubicin, and knocking it down sensitises MLLr lines
LEDGF also helps leukaemia cells survive chemotherapy. Blasts from patients whose disease came back have more of it, and removing it makes cells easier to kill - which suggests a LEDGF drug would work best alongside chemotherapy rather than on its own.
LEDGF does NOT hold the fusion on chromatin: knocking it down INCREASES fusion occupancy, and an IBD-binding-dead fusion still binds its targets
The standard picture is that LEDGF is a clamp holding the cancer protein onto DNA. When researchers removed LEDGF from leukaemia cells, the cancer protein stuck to DNA more, not less. Whatever LEDGF does for the fusion, holding it in place is not it.
LEDGF occupancy under revumenib: the measurement the graph called missing exists, in one 2025 preprint, and it goes the way the tenfold cooperativity predicts
Menin makes the leukaemia protein grip LEDGF about ten times more tightly, so blocking menin should loosen the LEDGF grip too. A 2025 preprint measured it: giving the menin drug revumenib does reduce how much LEDGF sits on the leukaemia's target genes.
LEDGF/p75 is required in the leukaemic cell and dispensable in ADULT MOUSE HAEMATOPOIESIS - not in the normal organism, which its germline knockout does not survive
A protein called LEDGF is needed for MLL-rearranged leukaemia to form, and adult mouse blood cells without the fusion get along fine without it. But mice born without the gene at all mostly die, so the gap is between the leukaemia and one tissue, not between the leukaemia and the whole healthy animal.
LIN28B has two routes - blocking let-7, and binding mRNAs directly - and in KMT2A::AFF1 every published mechanism is the direct-mRNA one (let-7 IS measured in mouse MLL-fusion AML: in-session correction)
LIN28B is famous for one job: holding down a growth-braking RNA called let-7. But it has a second, less famous job - sticking to ordinary messenger RNAs and making them last longer. Every published explanation that names a specific target inside an MLL/KMT2A fusion is the second job, not the famous one.
LIN28B localization - a nucleolar pool that sequesters pri-let-7, a predominantly cytoplasmic steady state with cell-cycle-dependent nuclear entry, and NO measurement in any haematopoietic cell
LIN28B is mostly found in the body of the cell, but a pool of it goes into the nucleolus - a dense blob inside the nucleus - and that is where it traps the growth-braking let-7 RNA before it can be made. Nobody has ever looked at where LIN28B sits inside a blood cell, let alone a leukaemia cell.
Lineage switch in KMT2A-rearranged leukemia — the same clone changes what kind of blood cell it is
A KMT2A-rearranged leukemia can stop looking like a B-cell leukemia and start looking like a myeloid (AML) leukemia, while the DNA proves it is still the very same cancer clone. The cell has not been replaced; it has changed career, and that is how it walks away from treatments aimed at B cells.
Losing the AF10 PZP reader is not a missing function - it is a removed brake, and putting it back cures the fusion
The normal AF10 protein has a module that reads the histone tail and keeps DOT1L in check. Every leukaemia version of AF10 has lost it - and if you engineer it back into the fusion, the fusion stops causing leukaemia.
Lost from the KMT2A half: both taspase1 cleavage sites (D2666 and D2718) - the fusion is never processed
Normal KMT2A is cut in two by an enzyme, and that cut is the signal for the protein to be cleared away on schedule. The fusion is truncated long before either cutting site, so it is never cut and never cleared - and in a mouse, forcing the healthy copy to stay uncut pushes the fusion off the genes it controls and makes the animals live longer.
Lost from the KMT2A half: PHD2 (aa 1479-1533), an E3 ubiquitin ligase that limits KMT2A's own activity
One of KMT2A's own finger-shaped modules is an enzyme that tags KMT2A itself for destruction, keeping its levels and its activity down. The fusion protein does not have it, so that self-limiting loop is gone.
Lost from the KMT2A half: PHD3 and the bromodomain are one cassette, not two modules (aa 1566-1765)
The part of KMT2A that reads an 'on' mark and the part next to it work as a single hinged unit, and a helper enzyme bends the hinge to flip the protein from switching genes on to switching them off. The fusion throws away the whole hinged unit.
Lost from the KMT2A half: the four-residue HCF-binding motif (DHNY, aa 1799-1802)
Normal KMT2A carries a four-letter tag that hooks it to a control protein called HCF-1. That tag sits just past the place where the gene always breaks, so every leukemic fusion protein loses it.
Lost from the KMT2A half: the FYRN-FYRC clamp (2018-2074 / 3666-3747) that holds wild-type KMT2A together
After normal KMT2A is cut in two, a pair of matching clips on the two halves snap them back together. Without its partner, the clip on the front half is not inert - it is the very thing the cell grabs to destroy the orphaned piece. The leukemic protein is cut off long before either clip and has neither, which is part of why it survives.
Lost from the KMT2A half: the MOF/KAT8 contact - required in the cell, but not shown to matter to the fusion
KMT2A recruits a second enzyme that adds a different chemical mark, and the fusion cannot recruit it. That second enzyme really is needed for this leukemia - but when it was finally tested inside a fusion-driven leukemia, it turned out not to be running the fusion's own gene programme, and to be just as necessary in leukemias with no KMT2A break at all.
Lost from the KMT2A half: the SET domain (aa 3829-3945) - the lost module whose loss demonstrably does not matter
The enzyme part of KMT2A - the bit that actually writes a chemical mark onto DNA packaging - is deleted from every leukemia fusion. Deleting it from mice does not stop the leukemia, so this is a loss that appears not to matter.
Lost from the KMT2A half: the transactivation domain (aa 2829-2883) and its CBP contact
KMT2A has its own 'switch genes on' patch that works by grabbing a shared coactivator called CBP. The fusion loses that patch and borrows an entirely different switching mechanism from its partner instead.
Lost from the KMT2A half: the WIN motif and the whole WRAD interface (aa 3762-3767, Arg3765)
KMT2A only works as an enzyme when four helper proteins clamp onto it, and they all attach through a single arginine near the far end of the protein. Every leukemic fusion is cut long before that point, so the fusion protein cannot assemble the enzyme at all.
Mechanism class 1 — chromatin and elongation hijack (KMT2A-r, NUP98-r)
Some fusion proteins keep the part that finds genes but swap the part that decides whether to switch them on for a machine that switches them on permanently. The cell is then stuck reading a set of stem-cell genes it should have finished with, so it never grows up.
Mechanism class 2 — transcription-factor corruption (RUNX1::RUNX1T1, PML::RARA, ETV6::RUNX1, CBFB::MYH11) — a balance, not a pure dominant negative
Here the fusion keeps the part of a master switch that binds DNA but replaces the part that turns genes on with a part that shuts them off, so development stalls. It is not simply that the normal switch is disabled — these leukaemias still need a working copy of it, and die without one.
Mechanism class 3 — constitutive kinase activation (BCR::ABL1, NPM1::ALK, ETV6::NTRK3)
In these fusions an enzyme that should switch on only on demand gets stuck on, because the partner glues copies of it together. Enzymes have a real chemical slot a pill can plug, which is why this class produces drugs - but the pill has to be given early and alongside chemotherapy, because on its own, late in the disease, it does very little.
Mechanism class 4 — condensate-driven transcriptional rewiring (NUP98-r, EWSR1::FLI1)
Some fusions bring along a floppy, sticky stretch of protein that makes many copies clump together into droplets on the DNA. Those droplets concentrate the machinery that switches genes on, creating brand-new super-switches that the normal cell never had.
Mechanism class 5 — enhancer hijacking (inv(3) GATA2/EVI1 — and DUX4-r, which only half belongs here)
Sometimes the rearrangement does not build a new fusion protein: it moves a powerful genetic on-switch next to a gene that should be silent. But the cleanest example of that is inv(3) AML — in DUX4-rearranged leukaemia the relocation also truncates the gene, and the truncation turns out to matter.
MEIS1's 'nucleus' is a profile prediction, not a measurement - and a homeodomain-less isoform is cytoplasmic and does the opposite job
The database says MEIS1 is a nuclear protein, but that statement comes from a computer recognising a homeodomain in its sequence, not from anyone looking. A 2026 study found a shorter MEIS1 that lacks that homeodomain, sits in the cytoplasm, and blocks the very differentiation the nuclear form drives.
Menin and LEDGF hold WILD-TYPE KMT2A on chromatin; for the fusion LEDGF is functionally required but positionally dispensable
The fusion cannot grip chromatin strongly on its own. It borrows two helper proteins — menin and LEDGF — that together clamp it in place, and menin is the handle that today's approved drugs grab.
Menin builds bone early and blocks it late: two opposite jobs in the same lineage
In bone-forming cells, menin is needed at the start to make a stem cell become a bone cell at all - but later, once the cell is committed, menin holds back the final maturation step. So blocking menin does opposite things depending on which stage the cell is at.
Menin has no subnuclear compartment, and nobody has looked - the whole published record is two biochemical fractionations, both DNA-damage papers, and no microscopy at all
Menin is the protein at the centre of this project's main drug idea, and nobody has ever taken a picture of where inside the nucleus it sits. The only evidence about its position comes from grinding cells up and seeing which fraction it lands in.
Menin holds KMT2A and KMT2B, and the graph has been written as if it held one
Menin is the anchor a whole class of leukaemia drugs is built around. It grips two proteins, not one - KMT2A and its close relative KMT2B - and every entry in this graph describes only the first grip.
Menin inhibition in the gut, not the islet: MI-463 raises GLP-1 from intestinal L cells
Menin is also abundant in the intestine, where it holds down production of the gut hormone GLP-1. Blocking menin there raises GLP-1 in the gut wall and in the blood, and improves how obese mice handle sugar.
Menin inhibition is a reshuffle, not a subtraction: four proteins are documented to GAIN chromatin or persist under the drug, and none of them is the fusion
The drug is usually described as taking things off DNA. In fact several things move onto DNA when it is given - a demethylase complex, a transcription factor, a repressive mark - and one partner protein simply stays behind when the fusion goes. The one protein nobody has checked for this is the fusion itself.
Menin inhibition makes adult human beta cells divide in a dish - the hardest thing in beta-cell biology
Adult human insulin-making cells almost never divide, which is why diabetes is hard to cure. A menin-blocking molecule, combined with a second drug, got them to replicate without losing their ability to make insulin.
Menin inhibition reported to induce CLEC12A and sensitise KMT2A-r and NPM1-mutant AML to CLEC12A-directed CAR cells (2026 preprint, not peer reviewed)
A 2026 preprint reports that menin inhibitors - the drugs at the centre of this project - turn on a surface protein called CLEC12A on leukaemia cells, which makes a CLEC12A-targeting cell therapy work much better. If it survives peer review, it turns a drug we already use into a way of creating a target that was not there before.
Menin is a brake the beta cell releases on purpose: pregnancy and glucose both lower it
Insulin-making cells keep menin around to stop themselves multiplying, but when the body needs more insulin - in pregnancy, or when blood sugar is high - the cell deliberately lowers its own menin so it can divide. Losing menin for a while is therefore something the healthy beta cell does to itself, not only something a disease does to it.
Menin is a corepressor of the LXR sterol sensors - one paper supplies three of this graph's twelve missing partners, and the function has never been written down here
Three of the partners this graph was missing - LXR-alpha, LXR-beta and SREBP-1 - are not three discoveries. They are one experiment in one paper showing that menin switches OFF the genes that make fat. That is a job for menin this project had never recorded.
Menin is the switch that lets TGF-beta turn parathyroid hormone down - and it works on secretion, not just growth
In parathyroid cells, a local signal called TGF-beta normally tells the gland both to stop growing and to make less parathyroid hormone. Menin is required for that message to be heard: knock menin down and the cells ignore it and keep secreting.
Menin represses GLP1R - the receptor a whole class of diabetes drugs exists to switch on
Menin keeps the gene for the GLP-1 receptor turned down. That receptor is the target of the biggest class of diabetes and weight drugs there is, so blocking menin turns up the very thing those drugs try to stimulate.
Menin represses NF-kB transactivation — through a mapped central region, a SIRT1 cofactor, and an RPA2 off-switch, in every tissue except the one this project cares about
Menin sits on the NF-kB proteins and holds down the genes they switch on, by bringing in an enzyme called SIRT1 that strips an activating mark off NF-kB. All of it was worked out in liver, gut, breast and endocrine cells, and none of it has ever been looked at in leukaemia.
Menin stimulates homology-directed repair through Chk1 and the region required is 1-228: a 2010 paper that is cited in no entry of this graph and in no reference list of the 2026 paper that reports the same conclusion
In 2010 a group showed that menin makes human cells better at the accurate kind of DNA break repair, that menin piles up at the break together with the checkpoint kinase Chk1, and that removing menin's first 228 amino acids destroys both. Nothing in this knowledge graph cites that paper, and neither does the 2026 paper whose title says menin is a key scaffold for the same process.
Menin's departure is partial where the fusion's loss is total, so the two are not stoichiometrically coupled
Three hours after the fusion protein is destroyed almost completely, menin has only half left the genes they shared. Half the menin sitting on those genes is being held there by something that is not the fusion.
Menin's other address list: the same complex that switches on HOXA in blood writes H3K4 methylation onto the CDK inhibitors p27 and p18 in endocrine cells
In blood, menin helps switch on genes that keep leukaemia cells immature. In hormone-producing tissue, the same menin machinery switches on two brake genes that stop cells dividing. Losing menin in an islet turns those brakes off, which is why losing menin causes endocrine tumours and not leukaemia.
MEPCE caps 7SK before assembly and then stops being an enzyme inside the particle - so its epigenetics annotation is a methyltransferase that methylates an RNA phosphate, which is exactly the case a mark-name regex mislabels in both directions
MEPCE puts a protective chemical cap on the 7SK RNA. Once the 7SK particle is fully assembled, MEPCE's catalytic site is blocked and it carries on working as a structural piece rather than as an enzyme.
MHC class I is a bivalent gene set, and a menin inhibitor derepresses it - shown in K-562, SCLC and neuroblastoma, and measured on HLA-A in primary AML, but never once in a KMT2A-rearranged cell
The genes that put a leukaemia cell's identity card on its surface are held half-on and half-off, and a menin inhibitor tips them towards on. That has been shown directly in three kinds of cancer cell and in leukaemia cells from patients, but never in a cell carrying the KMT2A rearrangement this whole project is about.
Mixed-phenotype acute leukemia and KMT2A rearrangement — plasticity caught in the act
Some leukemias refuse to pick a side: the same blasts carry both B-cell and myeloid markers at once, and doctors call this mixed-phenotype acute leukemia. KMT2A-rearranged leukemia is one of the classic causes, and it is probably the same underlying flexibility that later lets these cells switch lineage entirely to escape treatment.
MLL was reported at regions attached to the nuclear envelope in 2000, which is indirect evidence at the lamina that this graph currently records as absent
A 2000 paper looked at where the front end of MLL goes inside the nucleus and described it sitting in speckles that included patches stuck to the nuclear envelope and the rim of the nucleolus. That is a small but real observation at the nuclear edge, and this knowledge graph currently says no such observation exists.
MLL-AF9 leukaemia destroys the nerves that run its own niche — a mechanism class this graph had nothing on
Leukaemia does not simply move into the bone marrow and take up space: it damages the nerves that control the marrow's support cells, which converts the neighbourhood from one that suits healthy blood stem cells into one that suits the leukaemia. Because the damage is to nerves rather than to the cancer cells themselves, it is in principle repairable with drugs that act on nerve signalling.
MLL1 stays on condensed mitotic chromosomes and tethers menin there — a menin-dependent function absent from this graph until cycle 34
When a cell divides almost everything is stripped off the chromosomes, but MLL1 stays bound to the promoters of genes that were busy beforehand and holds menin there with it, so cells that lose this are slower to switch those genes back on. This project is entirely about the menin-MLL contact, and the 2009 paper describing this job it does was not cited anywhere here.
MLLT1 (ENL) refutes the YEATS parallel: the reader is retained in 84% of patients and is required for leukemogenesis
We expected the two commonest partners to work the same way — keep the useful bit, throw away the controlling reader. For ENL the opposite is true: the fusion keeps the reader in most patients, and deleting it makes the leukemia weaker, not stronger.
MLLT10 (AF10): the discarded PZP reader — the one partner where loss is proven to be the oncogenic event
Normal AF10 has a gripper that grabs the DOT1L enzyme and a separate sensor that decides where DOT1L is allowed to work. The leukemia fusion keeps the gripper and throws away the sensor, and when researchers put the sensor back, the cancer-causing activity disappears.
MLLT3 (AF9): the YEATS acetyl-lysine reader is discarded in ~92% of MLL-AF9 patients — same shape as PZP, but the causal test has never been run
Normal AF9 has a reader that recognises chemically tagged histones and steers the DOT1L enzyme to the right genes. In the great majority of MLL-AF9 leukemias the fusion keeps the part that grabs DOT1L and throws that reader away — the same pattern as the AF10 partner, except nobody has yet tested whether throwing it away is what makes the cell cancerous.
MLLT3 runs normal stem-cell self-renewal and is required in essentially no leukaemia line — the inverted case
AF9 is one of the commonest partners in this disease, and the gene it comes from turns out to be important for healthy blood stem cells rather than for the leukaemia. Deleting it from cancer cells in a large screen does almost nothing.
MOF/KAT8 is required by both KMT2A::MLLT3 and NUP98::HOXA9 — which is why it is the wrong kind of convergence
An enzyme called MOF is needed by two completely different kinds of fusion leukaemia, which sounds like a shared vulnerability worth attacking. It is not: MOF is needed by essentially every leukaemia cell we can measure, so the overlap tells us about what myeloid leukaemia needs in general, not about what these two fusions share.
Myeloid-primed blasts are already present at diagnosis of KMT2A-rearranged B-ALL and expand under immunotherapy — the single-cell demonstration this graph recorded as not existing
When a baby's B-cell leukaemia turns into a myeloid leukaemia after antibody or CAR-T treatment, the myeloid cells were not created by the treatment — reading single cells shows they were already sitting in the marrow on the day of diagnosis, in small numbers, and the treatment simply cleared away their competition. In the two patients where this was tracked before and after the switch, the myeloid-capable cells were found in the original diagnostic sample both times.
Niche-mediated chemoprotection in KMT2A-rearranged leukaemia — real for cytarabine, unsupported for menin inhibitors, and the one hypoxia experiment points the other way
Leukaemia cells sitting against bone-marrow support cells survive chemotherapy better than the same cells growing alone, and in this particular leukaemia the effect is stronger than in other childhood leukaemias. But 'the marrow protects the cells' is not a general licence: for the newest drugs in this disease, the menin inhibitors, there is no evidence of it, and the one experiment that tested low oxygen found it made the cells easier to kill, not harder.
No second driver means no second target - the quiet genome forecloses the combination logic that works in every other leukaemia
In most leukaemias, when one targeted drug stops working you add a second one aimed at another mutation the cancer carries. Infant KMT2A leukaemia barely has another mutation to aim at, so that strategy has nothing to attach to. Any combination in this disease has to be chosen from mechanism rather than from the patient's genome.
Nobody has convincingly found the H3K79 eraser - and that constraint sets the clock on every DOT1L inhibitor
Most chemical tags on histones have an enzyme that removes them; for H3K79 nobody has convincingly found one. If the tag can only be lost by being diluted out as cells divide, then a drug that blocks the enzyme has to be kept on board for several cell divisions before the tag disappears - which is exactly what the DOT1L trial did not do.
Nobody has ever looked at whether a KMT2A fusion is at PML nuclear bodies - a counted zero, with the one biochemical hint pointing yes
Cell nuclei contain small distinct compartments called PML bodies, and for one other leukaemia fusion protein their behaviour is the whole story of how the disease is treated. For this fusion, no one has ever put the two in the same microscope field - and the one biochemical experiment that exists says the fusion and PML do come out of a cell together.
Normal blood needs DOT1L too - lose it in an adult mouse and the marrow fails within weeks
DOT1L is not a leukaemia-only protein. Delete it in an adult mouse and blood production collapses: progenitors disappear within a week and the long-lived stem cells follow, so any drug that removes DOT1L is working against the tissue it is trying to save.
NPM1c and XPO1: one drug, one output, two incompatible mechanisms - is the mutant needed in the cytoplasm or on the chromatin?
The commonest mutation in adult AML creates a new export tag that hands the protein to XPO1. Blocking XPO1 switches the HOX genes off - but two well-regarded laboratories disagree completely about why, and their explanations are mirror images of each other.
Nuclear import of every KMT2A fusion is a property of retained KMT2A sequence, and the determinant was mapped in 1997 to a region containing the AT-hook motifs
Some of the proteins that MLL fuses to normally live outside the nucleus, yet the fusion always ends up inside it. A 1997 experiment found that the part of MLL responsible for pulling the fusion into the nucleus is the stretch that carries its DNA-gripping hooks, which every leukemia fusion keeps.
Nuclear matrix and mitotic chromosome scaffold — KMT2A sits with topoisomerase II, in the same place the gene breaks
KMT2A stays attached to the nucleus's internal skeleton, and during cell division it rides on the protein scaffold of the condensing chromosomes right next to topoisomerase II. That is striking, because topoisomerase II is the enzyme whose cutting at the KMT2A gene is thought to start infant leukemia in the first place.
Nuclear speckles — the compartment everyone assumes KMT2A::AFF1 uses, on evidence that skips AFF1 itself
Nuclear speckles are droplet-like hubs of RNA-splicing machinery, and genes that sit close to them get expressed more strongly. Three of the four AFF-family proteins have been shown to live in speckles — but not AFF1, the one that is actually fused to KMT2A in infant leukemia, and no one has ever co-stained a real KMT2A fusion with a speckle marker.
NUP98 FG condensates, audited assay by assay: no half-bleach has ever been done and no NUP98 fusion has ever been endogenously tagged
NUP98 fusions form bright dots in the nucleus, and four prominent papers call those dots liquid droplets. The two tests most of them rely on cannot actually tell a liquid droplet apart from a protein clustered onto a sticky stretch of DNA — and the one test that can has never been run on a NUP98 fusion.
Only a handful of loci lose fusion occupancy under a menin inhibitor, and they are the same few whether menin is drugged or deleted: MEIS1, MEF2C, JMJD1C - not HOXA, not MYB
When the drug is given, the fusion protein leaves only a few genes. Those few are named in the original paper and they do not include the HOX genes, which are the ones everybody talks about. Deleting menin genetically hits the same short list.
P-TEFb phosphorylates AFF4 and rearranges the complex the fusion hijacks
The super elongation complex is the machine KMT2A fusions hijack. Its own kinase, CDK9, phosphorylates the complex's scaffold protein, which changes how the parts fit together and where the complex goes on DNA.
PAF1 is annotated as punctate everywhere in the nucleus EXCEPT the nucleolus - and a 2024 proximity map puts it in the nucleolus
The standard database entry for PAF1 says it is scattered in dots through the nucleus but not in the nucleolus. A 2024 study that mapped what sits next to RNA polymerase II inside nucleoli found PAF1 there, and showed that removing it wrecks nucleolar organisation. Both statements cannot be right, and nobody has noticed the clash.
PAF1C couples transcription to H2B ubiquitination - but the RTF1 half of that coupling is yeast-specific and contested in human cells, and PAF1C is one of the few complexes that genuinely binds the retained KMT2A fragment
The ubiquitin tag only lands on genes that are actively being read, and the reason is a complex called PAF1C that rides along with the reading machinery and brings the tagging enzyme with it. That matters here because PAF1C is one of the few things that binds directly to the part of KMT2A kept in the fusion.
PAF1c is needed by the normal stem cell and by the leukaemia, and the genes it regulates in each are different
The PAF1 complex is required both by healthy blood stem cells and by leukaemia cells, so removing it is not selective. But the genes it controls in the two settings are not the same, which is where a selective idea would have to come from.
Panobinostat lowers H2B ubiquitination in MLL-rearranged ALL by suppressing RNF20/RNF40/WAC — the only clinical-stage agent that reaches this axis, and its one ALL trial was stopped for lack of efficacy
An approved HDAC-blocking drug turns out to work partly by shutting down the enzymes that put the ubiquitin tag on H2B, and it worked well against MLL-rearranged infant-type leukaemia in mice. The only trial of that drug in adult acute leukaemia was stopped early for lack of effect, which is the sobering part.
Paraspeckles and NEAT1 — no KMT2A evidence at all, but a fusion oncoprotein in another leukemia uses NEAT1 as a differentiation brake
Paraspeckles are nuclear droplets built around a single long RNA called NEAT1. In one leukemia the fusion protein switches NEAT1 off, and switching it back on is part of how the curative drug makes the cells grow up — but nobody has ever measured NEAT1 or looked at a paraspeckle in KMT2A-rearranged leukemia.
PBX proteins are not constitutively nuclear: a MEINOX partner keeps them there by blocking export - and nobody has ever looked at PBX3 in a KMT2A-rearranged cell
A transcription factor only works inside the nucleus, and PBX proteins carry signals both for going in and for coming out, so it is their partner protein binding them that keeps them in. If that holds for PBX3 and MEIS1 in leukemia cells then breaking the pair would move one of them out of the nucleus altogether - and nobody has ever checked.
PKA phosphorylates menin at Ser487 and pulls it off chromatin - a kinase input on the drug target itself
Menin is the protein that menin inhibitors aim at. A hormone signal acting through the enzyme PKA puts a phosphate on menin at one spot, and phosphorylated menin gets pulled away from DNA by the cell's skeleton proteins. Something other than a drug can already move menin off the genes it sits on.
PML K160 SUMOylation - the retained-portion PTM that became an approved therapy, and the benchmark every other retained site should be measured against
In one leukaemia fusion protein, a chemical tag on a spot the fusion keeps is the exact thing arsenic trioxide exploits to destroy it. Mutate that one lysine and arsenic stops working. That is what a retained modification site looks like when it has been taken all the way.
PML nuclear bodies and PML::RARA — the best-worked case of a fusion destroying a nuclear compartment, and what it actually predicts for KMT2A
In one leukemia, APL, the fusion protein shatters a normal nuclear structure called the PML body into hundreds of tiny specks, and the drug that cures the disease puts those structures back together. It is the clearest example in cancer of a fusion working by putting things in the wrong place — but the twist is that reassembling the bodies turned out not to be the cure; destroying the fusion protein was.
Polycomb bodies — indirect evidence via CBX8, undercut by the finding that CBX8-PRC1 condensates are dissolved rather than built by the rest of PRC1
Polycomb bodies are dense spots where the machinery that keeps developmental genes switched off collects. KMT2A fusions physically bind CBX8, a component of that machinery, which is the only reason to think the fusion goes near these spots — but the protein that actually builds the droplets is a different family member, and adding the rest of the complex makes CBX8's own droplets dissolve.
Polycomb in KMT2A-r leukemia — core PRC2 required, EZH2 and BMI1 dispensable, and CBX8 acting outside its repressive role
Polycomb proteins normally switch genes off. In KMT2A leukemia they behave strangely: one of them, CBX8, is actually needed to switch the fusion's target genes ON, and the enzyme everyone tries to drug (EZH2) turns out to be the dispensable part.
PSIP1 keeps R-loops off the genes it sits on, and removing it causes transcription-replication conflict, DNA damage and PARP-inhibitor sensitivity
When genes are switched on, the new RNA can stick back onto the DNA and jam the machinery. PSIP1 stops that happening. Take PSIP1 away and the jams cause DNA breaks, which is a reason to be careful about drugging it and also a possible way to exploit it.
Put both fusion proteins in the same cell and you get something neither produces alone
Cells carrying both fusion proteins resist death, keep dividing and switch on stem-cell genes in a way that cells with only one of them do not. The pair does something the singles do not.
Re-founding the address argument: the CXXC domain cannot specify a location at any affinity, and a tighter one would be worse
This project used to argue that the fusion's DNA grip is too weak to hold it in place. That was based on the weakest of nine measurements. The better argument is that the grip reads a two-letter pattern that appears tens of millions of times in the genome, so it cannot pick a location however strong it is - and if it were stronger it would be worse at picking.
Releasing JunD does not merely de-repress it: JunD without menin becomes a growth promoter
JunD normally slows cells down, but only while menin is attached. When menin is removed, the same protein starts speeding cells up instead - the effect flips sign rather than simply disappearing.
Removing the fusion is a transcription event first and a chromatin event later, and the gap is enormous: nascent transcription changes within 15 minutes, while genome-wide RNA Pol II peak calls are 99.6% retained at 3 hours and not one active histone mark loses a peak call by 24 hours - the fusion's immediate job is licensing elongation, not building a chromatin landscape
When the cancer fusion protein is destroyed the genes it controls stop being read within fifteen minutes, yet the reading machinery is still sitting on the DNA three hours later and the chemical marks on DNA packaging have not budged a day later. So the fusion's immediate job is not to assemble machinery or write marks - it is to let machinery that is already there actually move.
Resolving cycle 83's threat: the MLL1 channel moves under both drug classes and is therefore not diagnostic, but the MEN1 channel separates them completely - so 'menin displacement' remains a specific readout of menin-interface engagement and the corpus must stop treating the two channels as one
A worry had been hanging over this project for months: if a drug that does not touch menin moves the same signal as a drug that does, then the whole menin idea might be non-specific. It turns out the worry applies to one particular measurement, not to menin itself - menin's own signal separates the two drugs perfectly.
Reversibility and kinetics — does the program collapse when the fusion is removed, and how fast?
If you take the fusion protein away, the cancer program falls apart — the first changes happen within about fifteen minutes, and even an established, genetically messy leukemia regresses. This is the single best reason to think that blocking the fusion's partner could work.
RNA m6A (METTL3/YTHDF/YTHDC1) in KMT2A-rearranged leukemia — the empty cell that turned out not to be empty
Cells put small chemical tags on their RNA, and one of those tags turns out to be needed for leukemia cells to keep their DNA folded the way the cancer program requires. In KMT2A-rearranged leukemia the enzyme that writes the tag is itself switched on by KMT2A, and there is already a drug for it.
RNA-binding proteins set how much KMT2A::AFF1 protein a cell makes from its own fusion transcript
The amount of fusion protein a leukaemia cell makes is not fixed by the broken chromosome alone — proteins that grip the fusion's messenger RNA decide how much of it gets built. Two different laboratories have now found RNA-binding proteins sitting on the t(4;11) message and turning its output down.
RPA-coated single-stranded DNA is the structure that licenses ATR: ATRIP reads it, ATR is brought to the lesion, and CHK1 is activated - which is how this corpus's ATR and CHEK1 entries connect to the clamp that had no entry until now
A cell cannot see a broken chromosome directly. What it sees is a stretch of DNA that has become single-stranded and has been coated by the RPA clamp - and that coated stretch is the signal that switches on ATR, the kinase that halts the cell cycle. RPA is therefore not just a protective coating; it is the thing that makes damage visible.
RUNX1 K43 acetylation - a retained-portion site that AML1-ETO genuinely requires, in the 5' partner, with the 3' partner supplying the acetyltransferase in trans
In a different leukaemia fusion, p300 adds a chemical tag to a single lysine in the front half of the protein. Take that one lysine away and mice that would have died in five months survive past eleven. The back half of the fusion is what recruits the tagging enzyme - so the fusion builds the modification that it then depends on.
SAM supply sustains the KMT2A-rearranged hypermethylated state - the metabolic requirement that IS a consequence of the chromatin programme, and can be attacked as one
Chemical tags on DNA and histones do not stay put on their own; the cell has to keep re-applying them, and each one costs a molecule called SAM. KMT2A-rearranged leukaemia keeps far more of these tags than a normal cell, so it has a far larger SAM bill - and if you stop paying the bill, the genes the leukaemia depends on switch off.
SEC-L2 and SEC-L3 contain ENL and AF9 exactly as SEC does, so AFF2 and AFF3 sit in complexes built around the same acyl-lysine readers as the index fusion's partner - and both entries carry no epigenetics annotation at all
The protein fused to KMT2A in the commonest infant leukaemia, AFF1, has two little-studied relatives, AFF2 and AFF3. Those relatives build their own elongation machines - and those machines contain the very same two histone-mark-reading proteins, ENL and AF9, that make the fusion dangerous.
Serine/glycine dependence in MLL-ENL leukaemia, reached through a DIRECT fusion target gene - and the reason a PHGDH counted absence in this graph would have been wrong
A gene the MLL fusion switches on directly, PTBP1, controls how the leukaemia cell splits sugar for energy and for building blocks. Knock it down and the cells use less glucose, make less lactate, and lose a key form of a glycolytic enzyme - and the leukaemia turns out to need serine and glycine, two amino acids the body can normally make for itself, so much that mice fed a diet without them survived significantly longer.
SET::NUP214 has no address of its own: chromatin-bound CRM1 is already at HOX, and the fusion's FG repeats dock onto it
You would expect a leukaemia fusion to find the HOXA genes itself and drag its helpers there. For SET::NUP214 it is the other way round: the export carrier CRM1 is already sitting on those genes, and the fusion sticks to it.
SET::NUP214 is KMT2A-free in its sequence but not in its mechanism: it binds and condenses wild-type MLL1, which reopens the menin question this beat was sent to close
SET::NUP214 contains no piece of KMT2A, which is why it looked like a clean test of whether leukaemia can drive HOXA without the menin-KMT2A machinery. But the fusion grabs the ordinary, unbroken KMT2A protein and pulls it onto the same genes - and ordinary KMT2A is held by menin.
SET::NUP214 is the second KMT2A-free HOXA fusion, and it converges on the same two proteins CALM-AF10 uses: CRM1 and DOT1L - and so does NPM1c, which IS menin-dependent (corrected 2026-09-21)
A second T-cell leukaemia fusion, SET-NUP214, also switches on the HOXA genes without involving KMT2A. It sits on the HOXA genes together with the same two proteins CALM-AF10 depends on, which suggests they are a shared route to the same output.
SET::NUP214 recruits DOT1L by a different surface from CALM-AF10 and KMT2A::MLLT10: the founding paper tested for AF10 in the complex and did not find it
Three different leukaemia fusions all end up using the same enzyme, DOT1L, to switch on the HOXA genes. Two of them grab it with the same molecular handle; SET::NUP214 does not have that handle, so it must be grabbing the enzyme some other way.
SIAH1/SIAH2: the disposal route that destroys wild-type AF4 and that the reciprocal fusion escapes
Normally the cell keeps AF4 levels down by tagging it for destruction through a pair of enzymes called SIAH1 and SIAH2. The mirror-image fusion protein carries the tag-binding piece but somehow avoids being destroyed, so it accumulates.
SPEN is the only member of the real-gap floor whose epigenetic readout is direct and published: its SPOC domain recruits NCoR/SMRT, NuRD and the m6A machinery to active enhancers and promoters, and it disengages from chromatin the moment silencing succeeds
SPEN is a repressor that latches onto genes while they are still switched on, pulls in enzymes that strip acetyl groups off histones and remodel the packaging, and then lets go once the gene is quiet. It is the one protein in this beat's list of true gaps that has a clear, published chromatin job.
Step 1 — How the fusion finds its genes: AT-hooks and the CXXC domain reading unmethylated CpG
The fusion protein has two gripping parts kept from KMT2A: sticky AT-hooks that hold onto DNA generally, and a CXXC finger that only sticks to CpG letters that have no chemical tag on them. Because the genes a stem cell keeps switched on are exactly the ones with untagged CpGs, the fusion is automatically parked on the wrong set of genes.
Step 3 — The hijack: the fusion drags the super elongation complex to paused RNA polymerase II
Genes usually keep a copying machine parked just past the start line, waiting for a green light. The fusion protein permanently attaches the green-light machinery — the super elongation complex — to its favourite genes, so the copier never stops.
Step 4 — The histone mark: DOT1L and H3K79me2/3 at fusion target genes (and why global expression barely moves)
The fusion brings along an enzyme called DOT1L that stamps a chemical tag (H3K79 methylation) on the genes it sits on, and that tag keeps them switched on. Remarkably, removing the enzyme barely changes the rest of the cell's genes — only the fusion's own targets collapse.
Step 5 — The output program: HOXA9/HOXA10/MEIS1/PBX3 plus MEF2C, MYB, BCL2, FLT3 and PROM1/CD133
The genes the fusion keeps switched on are the ones a blood stem cell normally uses only briefly during development. Chief among them are HOXA9 and MEIS1, plus a survival gene (BCL2), a growth-signal receptor (FLT3) and a stem-cell surface marker (CD133).
Step 6 — The consequence: self-renewal held on, differentiation never fires
The end result is a cell stuck in a loop: it keeps making copies of itself and never grows up into a working blood cell. Drugs that switch the fusion off let those cells finally mature, which is why they work — and why they cause differentiation syndrome.
Subclonal cooperating lesions lost at relapse - what was actually measured, and why 'lost' is the wrong word
In most cancers the extra mutations that help the tumour grow are in every cell and are still there when the disease returns. In infant KMT2A leukaemia they are in a minority of cells at diagnosis and often missing when the disease comes back - the opposite pattern. That is strong evidence that the fusion, not the extra mutations, is the disease; but a more sensitive test has since shown that some of those mutations were never gone, only too rare to see.
Switching menin off reverses diabetes that is already there - in three different diabetic mouse models
In mice that already had diabetes, deleting the menin gene did not just prevent the disease - it reversed the high blood sugar that was already established, by making more insulin-making cells. It worked in three different kinds of diabetic mouse.
Taspase1 is what turns the reciprocal product into an oncoprotein - and it is a conditional oncogene because of it
The mirror-image fusion protein is made in one long piece that the cell quickly destroys. An enzyme called Taspase1 cuts it in two, and the two halves clip back together into a package that survives for a very long time. Without that cut, there is no oncoprotein.
The 7SK elongation brake is not an off-chromatin reservoir: 7SK occupies enhancers and super-enhancers genome-wide, recruits BAF and repositions nucleosomes - which means NOT APPLICABLE is the wrong epigenetics annotation for LARP7, MEPCE and HEXIM2, and BOUNDED ABSENCE is the right one
7SK is a small RNA best known as a parking space that holds the transcription accelerator in reserve, which made it look like something with no chromatin biology. In fact 7SK sits on enhancers all across the genome, pulls in a nucleosome-moving machine, and changes how DNA is packaged there.
The adult-type model answers PLK1 inhibition by making more PLK1 and AURKA
When the adult-type cells were given the PLK1 blocker, they responded by making more of the very machinery being blocked. Adding a second drug against one of those backups restored the effect.
The AF9 acetylation arm: PCAF at K339, HDAC5 gated by CaMKII - a kinase-controlled switch the fusion throws away
A second control system on AF9 works by acetylation rather than phosphorylation, but a kinase still runs it: a kinase kicks the eraser enzyme out of the nucleus so the mark can build up. The spot it acts on sits in the part of AF9 that the leukaemia fusion discards, so this whole system is missing from the cancer protein.
The age of the marrow, not only the age of the cell, decides whether MLL-rearranged leukaemia comes out myeloid or lymphoid
Put the same leukaemia into a newborn mouse and into an adult mouse and you get different diseases: in the newborn it acquires B-cell character, in the adult it stays myeloid. The difference is not in the leukaemia cell at all — it is in the surrounding marrow, and one of the molecules responsible has been identified.
The bite-back surface: LEDGF tethers CtIP to active chromatin and is required for DNA-end resection and homologous recombination
Away from leukaemia, LEDGF helps cells repair broken DNA by bringing a cutting enzyme to the break. A drug that removes LEDGF would remove that too, in every cell in the body, in children who are also being given DNA-damaging chemotherapy.
The brake pad is actively thinned: autophagy degrades HEXIM1 in MLL-fusion and NPM1c AML cells
In AML cells carrying an MLL fusion or a mutant NPM1, the cell's own recycling machinery chews up HEXIM1, the protein that holds the elongation enzyme in check. Blocking that recycling brings HEXIM1 back.
The case that the KMT2A-fusion condensate is an artefact — what 1,6-hexanediol, overexpressed tags and in-vitro droplets actually prove
Almost every claim that a leukemia protein forms liquid droplets rests on three experiments, and all three have known ways of giving the wrong answer: a chemical that dissolves droplets also poisons the enzyme being measured, glowing tagged proteins clump when overproduced, and purified proteins in a tube form droplets at concentrations cells never reach. This entry states the strongest version of the sceptical case so that the graph's positive condensate claims can be graded honestly rather than repeated.
The cell types that need MLL1 are the uncommitted ones - dependence ends exactly where lineage commitment begins
The gene that breaks in infant leukaemia is needed by the earliest, most flexible blood cells, and stops being needed once a cell has decided what it is going to become. That boundary is measured, not assumed.
The chromatin exchange at a menin-inhibited fusion site, measured in one experiment: H2AK119ub and H3K27me3 in, H3K4me3 out, in proportion to how much fusion left - and PCGF1 loss attenuates the in-fill without abolishing it
When the drug removes the fusion protein from a stretch of DNA, that stretch does not just go quiet by default - it is actively taken over by the cell's silencing machinery, and the 'active' chemical tag is removed at the same time. How much of each happens depends on how much fusion was removed.
The class I and class II arms of menin inhibition are different mechanisms - the class I effect was tested against viral mimicry and NF-kB and excluded both
Two teams found that a menin inhibitor makes leukaemia cells more visible to the immune system, and it is tempting to assume they found the same thing. They did not: one route works by waking up ancient viral DNA and triggering an alarm, and the other explicitly does not need that alarm at all.
The clearance clock has machinery, and the machinery says the clock does NOT stop in a non-dividing cell - a postmitotic heart replaces its histones with a half-life of about two weeks
This project argues that the chemical tag its main drug target writes can only disappear when the histone carrying it is swapped out, so a slowly dividing cell should hold on to the tag and resist the drug. The machinery that does the swapping turns out to run even in cells that never divide at all: in adult heart muscle, histones are replaced with a half-life of about two weeks.
The clustered PHD fingers (KMT2A aa 1416-1640) are a chromatin-targeting module and the canonical t(4;11) junction removes them by ten residues
KMT2A has a cluster of small zinc-binding fingers that steer it toward already-active genes marked with a particular chemical tag. The commonest infant-leukemia fusion cuts the protein just ten residues before that cluster starts, so the fusion keeps the parts that get it into the nucleus and onto CG-rich DNA and loses the part that reads the activity mark.
The CTCF licensing cascade: CTCF recruits HP1-gamma, KAP1, SUV39H1, SETDB1 and KDM5A to a break, that repressive chromatin licenses EARLY BRCA1/BARD1, and early BRCA1/BARD1 licenses EXO1/DNA2 long-range resection - measured entirely in U2OS and HEK-293T, with engineered breaks and one etoposide arm
When DNA breaks, the loop protein CTCF calls in the machinery that normally silences genes, and that silencing machinery is what lets the BRCA1 repair complex arrive and start trimming the broken ends. Everything known about this was measured in a bone-cancer cell line and a kidney cell line with artificially cut DNA, never in a blood cell.
The CXXC domain (aa 1147-1195, PDB 2KKF) is a chromatin address, not a compartment address — CpG mutants lose transformation but keep their subnuclear position
The CXXC domain is the part of KMT2A that reads unmethylated CG letters in DNA and so decides which genes the fusion lands on. The decisive experiment showed that breaking it stops the leukemia without moving the protein anywhere else in the nucleus — so it tells the fusion which genes to sit on, not which part of the nucleus to live in.
The DOT1L-AF10 octamer - the second route to H3K79 methylation, required in MLL-AF10 and absent from KMT2A::AFF1
DOT1L reaches the genome by two different handshakes. One is with AF9 or ENL; the other is with AF10, and that second one only works when four copies of the pair assemble into a larger unit. The commonest infant leukaemia fusion contains neither AF9 nor AF10, so claims about how DOT1L gets recruited do not transfer between fusion partners.
The endogenous MLL-AF4 localisation experiment was unattributable, not negative: an anti-AF4 polyclonal in a t(4;11) cell images mostly wild-type AF4
In 1998 a group raised antibodies against AF4 and looked at real t(4;11) leukemia cells rather than cells forced to make extra fusion protein. They found the fusion in the same speckled nuclear pattern as normal AF4, indistinguishable from leukemia cells without the translocation — and they found there is less fusion protein than normal AF4, not more.
The ENL YEATS acetyl-reader inside KMT2A::MLLT1: what it reads, what it does, and why both control arms carry it
The partner half of this fusion protein carries a small module that grips histones tagged with an acetyl mark, and it is the module that lets the fusion sit on the genes it switches on. Deleting it stops the fusion causing leukaemia in mice outright, and breaking only its grip on the acetyl mark cuts the number of leukaemia stem cells about three-fold.
The FA core complex does not move together, and that is the strongest argument against reading this as a Fanconi-pathway dependency at all
The Fanconi repair crew works as one machine: eight proteins clip together to do a single job, adding a chemical tag to a ninth protein. If a cancer cell needed that machine, losing any one of the eight should hurt it about equally. In these leukaemia cells four of the eight matter a lot, three barely matter, and the protein the machine actually acts on barely matters either — which means whatever these cells need, it is probably not the machine doing its known job.
The fludarabine prodrug chain - three molecules called by one name, and the one that is measured is neither the one given nor the one that works
The drug that goes into the drip, the drug measured in a blood test, and the drug that actually works are three different molecules, and all three are called fludarabine. A blood level tells you about the middle one only.
The focal half: H3K79 hypermethylation at the HOXA cluster and MEIS1 - but which cluster member differs between CALM-AF10 and MLL-AF10
At a handful of growth-control genes, the CALM-AF10 fusion piles on the DOT1L chemical tag instead of removing it, and those genes switch on. Switching them on is what stops the blood cells maturing.
The fusion begins at AFF1 residue 347 and the cyclin T1-binding sequence ends at 308, so KMT2A::AFF1 is the one AFF-family protein that cannot enter the reservoir
Normal AFF1 carries a short stretch near its front end that lets it dock onto the elongation enzyme, and that same stretch is what pulls AFF1 into the off-switch particle. The leukaemia fusion protein starts after that stretch, so it can never be switched off that way - it has to borrow the enzyme from a normal copy of AFF1 or from AFF4 instead.
The KMT2A-fusion causal chain: six links, one measured ordering, and two arms that are parallel rather than sequential
This is the step-by-step account of how a KMT2A fusion protein turns a blood cell into leukaemia, and - just as importantly - which of those steps has actually been measured and which is only assumed. Four of the arrows in the usual picture have never been timed, two of the steps happen side by side rather than one after the other, and part of what today's drug does is not fusion biology at all.
The KMT2A-r DNA methylation signature — focal hypermethylation, no global hypomethylation, and a hypomethylating agent that did not help
In infant leukemia with a KMT2A fusion, many gene switches get chemically locked shut by DNA methylation — and the more locked they are, the worse the outcome. But the genome as a whole is not under-methylated, and a drug that strips methylation off proved safe without improving survival.
The lost SET domain, and the verdict: the fusion is a gain-of-function neomorph, not a KMT2A loss-of-function
The fusion protein throws away KMT2A's own enzyme head, which sounds like it should matter — but that enzyme turns out to be dispensable. The disease is caused by the new abilities the fusion gains, not by the abilities it loses.
The menin inhibitor's effect on H3K79me2 is a REDISTRIBUTION within the fusion's own site set, not a loss from it: the share falls 41.4% at the quartile the fusion is most evicted from and RISES 18.7% at the quartile it is least evicted from, in murine MLL-AF9
The drug does not simply strip the activating tag off the fusion's landing sites. Where most of the fusion is pulled off, the tag drops by about 40 percent; where least of it is pulled off, the tag actually goes up by about 20 percent. The tag is moving within the fusion's own territory, not just disappearing from it.
The menin interface cannot be what decides where KMT2A::AFF1 sits: within one patient the fusion's target promoters were almost entirely replaced while the menin-binding motif was unchanged
The fusion protein grips the DNA partly through a protein called menin, and that grip is the same in every cell of a patient. Yet in one child the set of genes the fusion sat on almost completely changed between two stages of illness, so the grip cannot be what chooses the genes.
The menin interface is identical in all 27 of these patients and the amount of chromatin their oncoproteins occupy differs by up to two orders of magnitude - so the interface can be a permission for binding but cannot be what sets how much binding happens
Every one of these fusion proteins has the same menin-binding handle. The amount of DNA they actually sit on differs a hundredfold between children, so the handle cannot be what decides the amount - something else in each child's cells is.
The menin-FANCD2 bridge is dead: it was tested in 2008, the mutation spectra differ, and there is no genetic interaction
Because cells lacking menin and cells lacking Fanconi proteins are both killed easily by the same DNA-damaging drugs, people assumed the two work together. In 2008 a group tested that directly in fruit flies and found the two genes leave completely different kinds of mutation behind, and that changing menin up or down does nothing to how a Fanconi-deficient fly responds to the damage.
The MLL-side twin of the 1998 anti-AF4 result: in 1997 the same punctate nuclear pattern was seen with and without an 11q23 translocation, and the authors said so
In 1997 a group made antibodies to the normal MLL protein and looked at leukemia cells. On a gel the antibodies separated the fusion protein from the normal one by size; under the microscope the picture looked the same whether or not the cell carried the translocation, and the authors themselves wrote that this rules out using the microscope to diagnose the disease.
The MYC contradiction can be advanced but not closed: the enhancer is independently validated, the SWI/SNF route is not, and the two papers never measured the same thing
Two papers disagree about whether MYC is what keeps this leukaemia alive downstream of a chromatin machine. A third paper, which nobody had connected to the argument, deleted the exact piece of DNA at the centre of it and found the leukaemia could not survive without it. That does not settle the disagreement, but it moves it: the switch is real, and what is in dispute is who operates it.
The N320-C180 clamp is the only KMT2A module whose disruption has ever been shown to change subnuclear localisation - and the fusion is the disruption
Normal MLL is cut in two and the halves clip back together; a 2003 paper showed that this clipped-together pair is what gives the protein its position inside the nucleus, and that breaking the clip makes one half spread out evenly instead of sitting in spots. Every leukemia fusion loses the second half entirely, so it can never form that clip - but a 2019 paper looking at the same system saw no change in position at all.
The near-silent genome of infant KMT2A-r ALL — 1.3 non-silent mutations, and why that is bad news as well as good
Most cancers carry hundreds or thousands of broken genes, but infant leukemia driven by a KMT2A fusion carries on average just over one extra mutation — the fusion does almost all the work by itself. That makes the fusion an unusually clean target, but it also means there is no second broken gene to attack and no obvious explanation for why the disease comes back so often.
The NF-kB story in KMT2A-rearranged leukaemia is a myeloid story, and the lymphoid half is not a weaker version of it — it is a different subject
Almost everything known about NF-kB in MLL leukaemia was done in myeloid cells and mice. The infant disease this project exists for is a B-cell leukaemia, and the handful of B-cell papers are about something else entirely — how the cells avoid being killed by an inflammatory signal, not how they depend on one.
The non-chromatin half: CALM-AF10 breaks the cell's iron intake and the cells are chelator-sensitive in vitro - but iron restriction did not extend survival in vivo (corrected 2026-09-21)
Because the fusion takes over the protein that normally helps pull iron into the cell, CALM-AF10 cells cannot take up iron properly and are unusually easy to starve of it in culture. When that was tried in mice it shrank the spleen tumour but the mice did not live longer.
The nuclear lamina and lamina-associated domains - evidence grade for KMT2A: weak and indirect, not none
The inner surface of the nuclear envelope is a silencing shelf, and genes parked against it are switched off. One 2000 paper put the KMT2A amino terminus near that surface, but it never looked at the lamina itself, so the question is barely opened rather than untouched.
The nuclear pore and NUP98 fusions — a fusion that takes a pore protein's disordered repeats off the pore and builds nuclear droplets with them
NUP98 is normally part of the nuclear pore, the gate in the nuclear envelope, and the sticky repetitive stretch it uses to make the gate selective ends up in leukemia fusion proteins. In the fusion that stretch no longer sits at the pore: it builds droplets inside the nucleus that stick the fusion to DNA over broad regions and pull in the machinery that switches genes on, including KMT2A itself.
The nucleolus — the one compartment with direct published images of a KMT2A fusion inside it
Of all the sub-compartments inside the nucleus, the nucleolus — the factory that builds ribosomes — is the only one where somebody has actually photographed a KMT2A fusion protein. Two independent groups, twenty-two years apart, saw it there, and the second found the fusion switches on a nucleolar stress response.
The only cellular proposal for how the SEC gets P-TEFb out of the reservoir is a condensate, and the two assays it rests on cannot tell that model from its rival
One published idea is that the elongation scaffold clumps together into liquid droplets and sucks the enzyme out of the off-switch particle by sheer local concentration. It is a good idea and the experiments used to support it cannot distinguish it from a much simpler explanation, so this entry records both.
The other arm of the crosstalk: H2Bub licenses H3K4 methylation through ASH2L and relieves a built-in COMPASS autoinhibition — and MLL3 does not respond at all
The same ubiquitin tag that lets DOT1L work also switches on the enzymes that put the H3K4 mark, and it does so by releasing a brake built into the enzyme itself. Strikingly, it works on MLL1 and not on MLL3 — so the tag does not simply turn up every related enzyme.
The other half of the break: what is actually known about the AFF1 breakpoint region, and the negative-control gene nobody cites
A translocation needs two breaks, and almost everything written about this one describes only the KMT2A side. The AFF1 side has been mapped, and the study that did it also examined a near-identical gene that never translocates - which is the comparison the KMT2A literature has been missing for thirty years.
The paralogue that covers for LEDGF: HRP-2 carries the only other IBD in the human proteome, substitutes for LEDGF in HIV integration, and the double knockout is embryonic lethal
There is a second human protein with the same grip module as LEDGF. In viral infection it can take over when LEDGF is gone, and mice missing both die before birth while mice missing either one alone survive. A drug against LEDGF alone may simply be covered for.
The parathyroid does not need the rest of the body: deleting menin only there causes neoplasia and hypercalcaemia
Scientists deleted the menin gene only in the parathyroid glands of mice, leaving it intact everywhere else. Those mice grew parathyroid tumours and had high blood calcium, which shows the gland goes wrong on its own rather than because of something happening elsewhere.
The PBX3-MEIS1 dimer, not HOXA9, is the unit that reproduces the MLL-fusion transcriptome - and it turns the HOXA cluster on from underneath
Everyone writes that KMT2A-fusion leukemia runs on HOXA9 plus its helper MEIS1. When a group actually tested the combinations, the pair that caused leukemia fastest and looked most like the real disease was MEIS1 plus PBX3 - with no HOXA9 added at all. The helpers turned the HOXA genes on by themselves.
The phosphatase side: SET binds both wild-type KMT2A and the fusion, and inhibiting it changes four kinases at once
Phosphates get added by kinases and removed by phosphatases. A protein called SET blocks the main phosphatase PP2A, and SET physically sits on both the normal and the fused KMT2A protein. Blocking SET with a drug stops leukaemia cells growing and changes the activity of several kinases at once.
The pivotal menin-inhibitor paper reports that the drug does NOT globally displace MLL1 or the fusion from chromatin - menin goes, the fusion largely stays
Everyone says menin inhibitors knock the leukaemia's fusion protein off the DNA, but the paper that launched these drugs says something different in its own results: menin comes off everywhere, while the fusion stays on the DNA almost everywhere and only leaves a handful of genes. The clearest test was in a cell line with no normal copy of the protein left, so every signal measured must be the fusion - and there too it stayed.
The plasma membrane - the one compartment this graph targets with drugs and had never written down
Three of the things doctors actually aim at in this leukaemia - CD19, CLEC12A and NG2 - are proteins sitting on the outside of the cell, and a drug can only reach them there. This entry is the graph's record of that compartment, because until now all 46 localization entries described places inside the cell.
The quiet genome, audited - the famous 1.3 mutations is a count of point mutations in the dominant clone, and all three of those words do work
Infant KMT2A leukaemia is famous for having almost no mutations - an average of about 1.3 per case. That number is real, but it counts only spelling changes in the DNA, only in the cells that make up the bulk of the leukaemia, and only down to what the machine of the day could see. Checking each of those limits does not overturn the claim; it changes what the claim licenses.
The reciprocal fusion (AFF1::KMT2A and friends): a second oncoprotein that opens chromatin for the first
A translocation makes two hybrid genes, not one: the famous KMT2A::AFF1 and its mirror image AFF1::KMT2A. The mirror-image protein is usually ignored, but there is evidence it pries chromatin open so the famous one can reach roughly ten times more target genes.
The result that runs the wrong way: in Psip1-null cells Hoxa and Hoxd expression goes UP, because Psip1 also recruits Bmi1 and Ctbp1
LEDGF is usually described as helping switch Hox genes on. In one careful study, cells without LEDGF switched Hox genes on more, not less, because LEDGF also brings in the machinery that keeps them off.
The RNA layer and the menin interface are separable: IGF2BP3 depletion sensitises MLL-AF4 leukaemia to menin inhibition and outperformed it alone
Menin inhibitors work by prying the fusion protein off the genes it switches on. Removing an RNA-binding protein instead attacks the same leukaemia from a completely different side, and in mice doing both together worked better than either alone — with the RNA side, on its own, working better than the menin drug.
The soluble-nucleoplasm versus chromatin-bound axis - the one localisation readout that keeps a size dimension, and nobody has run it on a KMT2A fusion
You can ask where a protein sits in the nucleus by taking a picture, or by breaking the cell into layers and weighing what is in each layer. Only the second method can tell the leukemia fusion protein apart from the normal protein it shares an antibody with, and for KMT2A fusions nobody has ever done it.
The strongest published case that condensates are load-bearing in KMT2A-fusion leukemia — and it is AKAP95, not AFF1
The best evidence that liquid droplets matter in this leukemia does not come from the fusion protein at all — it comes from a partner, AKAP95, whose droplets a 2026 paper showed are needed for MLL-fusion leukemia but not for normal blood formation, and which a designed peptide can break. The graph's earlier verdict that the condensate case was 'not established' was right when written and is now out of date.
The third category: modules lost from the fusion protein but supplied in trans by the wild-type KMT2A allele
The graph has been sorting KMT2A's parts into two boxes, kept and thrown away. There is a third box: parts the fusion threw away but still needs, borrowed from the intact second copy of the gene that every patient keeps.
The third job: LEDGF sits on transcribed gene bodies with splicing factors and changes exon choice in half the genes that have alternative isoforms
Besides holding onto chromatin and helping switch genes on, LEDGF helps decide which pieces of each gene end up in the final message. Removing it changes the messages cells make, in about half the genes that have a choice.
The three AT-hooks (aa 169-180, 217-227, 301-309) as a localisation determinant — and the correction that MLL's hooks read DNA *structure*, not AT-richness
KMT2A has three short hook-shaped motifs near its front end that grip DNA and are kept in every leukemia fusion. The textbook says they grab AT-rich DNA through its narrow groove — but the one paper that actually tested the MLL hooks found they recognise a DNA *shape*, the four-way cruciform, and said so explicitly: structure rather than sequence.
The trans-histone crosstalk H2Bub → DOT1L → H3K79me: modulatory for me1, effectively required for me2/me3
The ubiquitin tag on histone H2B does not just sit next to DOT1L's target — it physically reshapes the nucleosome so that lysine 79 of histone H3 swings out into DOT1L's active site. Without the tag DOT1L can still add one methyl group, but it largely cannot add the second and third, and it is the second and third that KMT2A-fusion leukaemia runs on.
The two t(4;11) products split AFF1 exactly at the P-TEFb-binding module, and the elongation half goes to the reciprocal
The translocation cuts AFF1 in two. The front piece, which is the part that grips the elongation enzyme, ends up in the reciprocal product; the back piece, which grips the other scaffold proteins, ends up in the main fusion. That division explains an old observation nobody had an explanation for: only the reciprocal product actually drives elongation.
The ZNF384-fusion immunophenotype - weak CD10 with aberrant CD13/CD33 - and the one marker that separates it from KMT2A-rearranged disease runs the OPPOSITE way to expectation
Leukaemia cells carrying a ZNF384 fusion look half B-cell and half myeloid: they carry myeloid badges (CD13, CD33) that B-cells should not have, and they are faint for the usual B-cell badge CD10. That mixed look is what first made pathologists suspect the subtype existed before anyone had sequenced it.
Transcription factories and Pol II clusters - no KMT2A::AFF1 imaging, condensate evidence for SEC and MLL-ENL, and 3D-FISH that actively contradicts the factory story for t(4;11)
For thirty years people have described transcription as happening at a few thousand shared 'factories' in the nucleus, and it was proposed that two genes meeting in one factory is how a translocation starts. Live imaging later showed those factories are fleeting clusters lasting about five seconds, and when someone measured where the MLL, AF4 and ENL genes actually sit, they found the wrong pairing to explain why t(4;11) is the common translocation.
Trithorax against Polycomb at HOXA: the textbook antagonism, and the two ways this disease breaks it
Textbooks say the KMT2A family switches genes on and Polycomb switches them off, and that they fight over the same genes. In this leukaemia that picture fails twice over: Polycomb and the fusion's anchor protein actually work together to keep differentiation genes off, and blocking that anchor turns bivalent genes on by a completely different route.
Twelve weeks of a menin inhibitor in diabetic mice: more beta cells, lower fasting glucose, and nobody looked for a tumour
Diabetic mice given a menin-blocking drug every day for twelve weeks ended up with more insulin-making cells, lower fasting blood sugar and better sugar handling, with no change in weight or appetite. The study never examined their pancreases for tumours, so its silence on that point is not reassurance.
Two drugs, two spatial mechanisms on one mark: catalytic DOT1L inhibition de-concentrates H3K79me2 out of its peaks and into mid-occupancy chromatin, while menin-interface inhibition lowers the mark's share at the peaks AND at mid-occupancy, so its effect is relocation rather than a smaller version of the same change
Both drugs lower the H3K79 tag where the fusion protein sits, but blocking the writing enzyme spreads the tag out of its peaks into the middle of the genome while blocking the menin interface lowers it in the peaks and the middle alike. One drug's effect is therefore not a fraction of the other's - they are different kinds of change.
Two pools of one protein: the chromatin-bound XPO1 that matters here has never been separated biochemically from the soluble export machine
The same export protein appears in two places at once - shuttling cargo through the nuclear pore, and sitting still on specific genes. Everything interesting for leukaemia is about the sitting-still pool, and no experiment has ever purified it away from the other one.
Two published ways a fusion oncoprotein engages BAF - be a subunit, or recruit it with a disordered domain - and KMT2A has never been tested for either
In two other fusion-driven cancers, scientists worked out exactly how the fusion protein takes over the SWI/SNF machine: in one the fusion is part of the machine, in the other it grabs the machine with a sticky disordered tail. Nobody has asked whether MLL fusions do the second thing, even though their partners have exactly that kind of tail.
Two surfaces, two jobs: the AHD picks the scaffold and the YEATS pocket reads the histone — so a YEATS ligand cannot ask which of the three SEC-family complexes it is disabling
ENL and AF9 each have two separate grips: one holds onto the scaffold that decides which genes the machine visits, the other grips a chemically tagged histone. A drug that plugs the histone grip leaves the scaffold grip untouched — so it disables the reading function in all three versions of the machine at once, including the two that have nothing to do with the leukaemia.
UBTF tandem duplications INVENT an export signal, and that invented signal is what puts the protein on chromatin - the cleanest proof the XPO1 address is real
In a high-risk childhood leukaemia, a short stretch of the UBTF gene is duplicated, and the duplication accidentally creates a new export tag. That new tag grabs XPO1 - and instead of exporting the protein, it glues it onto the wrong genes. Delete the tag and the disease programme does not start.
WDR5 and KMT2A are at the centrosome and the midbody - a non-chromatin address for both, with a patient-cell phenotype
WDR5 and MLL are thought of as chromatin proteins, but two recent studies put both of them on the machinery that pulls chromosomes apart and then pinches the cell in two. Removing either one leaves cells with crooked spindles, misaligned chromosomes and two nuclei - and cells from patients with an MLL-haploinsufficiency syndrome show the same defect.
WDR5 maintains a differentiation block — in CEBPA p30 AML, not in KMT2A-rearranged disease, and a title check nearly buried it
WDR5 holds a particular kind of leukaemia cell in an immature state: take WDR5 away and the cells start maturing again. That result is in a different genetic subtype from the one this project studies, and it was invisible to a search that only read titles.
What is actually conserved across KMT2A partner classes — forced self-association, with two documented exceptions that break the rule
Partners as different as an elongation scaffold and a cytoskeletal GTPase both cause leukaemia when joined to KMT2A, and the usual explanation is that all of them simply glue copies of the KMT2A fragment together. That explanation is right often enough to be useful and wrong in at least two places that have been directly tested.
What is lost: the PHD3/CYP33 switch that lets wild-type KMT2A turn itself off
Normal KMT2A has an off-switch: a small finger domain that, when a partner enzyme twists a nearby bond, flips the protein from switching genes on to switching them off. Every leukemic fusion throws that finger away, so the protein can only ever say 'on'.
What LEDGF actually does for the fusion: it brings AF4, CDK9, BRD4, P-TEFb and Ser2P Pol II to the target gene, and H4K16ac with them
LEDGF is not the hook that holds the cancer protein onto DNA. It is what recruits the machinery that actually switches the gene on, so removing it silences the target genes without moving the cancer protein.
What the paralogues' disease biology says about AFF1: the AFF family shares one SIAH degron — and the index fusion DISCARDS it, because the whole degron lies in the AFF1 leader that the junction at residue 347 throws away
All four of these scaffold proteins carry the same short disposal tag, and when that tag breaks the result is disease — a mouse with ataxia and two human syndromes. The leukaemia fusion protein still contains AFF1's tag, so the disposal route is intact in the oncoprotein and nobody has ever tried to use it.
What the reciprocal product assembles: P-TEFb, an H3K4 methyltransferase, an H3K79 methyltransferase, an arginine methyltransferase and an acetyltransferase
When the mirror-image fusion protein is purified out of cells it comes attached to at least sixteen other proteins, including the kinase that releases stalled RNA polymerase and three different enzymes that write marks on histones. That is a bigger machine than the normal AF4 complex.
When MYC falls after a menin inhibitor, there are three mechanisms it could be, and no published experiment separates them
MYC goes down when leukaemia cells are treated with a menin inhibitor, and people would like to use that as a way of checking the drug is working. But there are at least three different reasons MYC could fall, and telling them apart matters: one of them means MYC is a real drug target, and another means MYC is just what happens when a cell grows up.
Where does the fusion go when menin inhibition evicts it? Almost nobody has looked - the published readouts are chromatin occupancy, the one protein whose fate was measured properly is menin, and the single human observation of fusion loss is confounded by a lineage switch
Menin inhibitors work by knocking the fusion protein off the DNA. Everyone measures that it has left the DNA; nobody has taken a picture to see where it went - whether it floats free in the nucleus, piles up somewhere else, or gets destroyed. In one patient treated with the drug the fusion became undetectable, which hints that it is destroyed, but that patient's leukaemia was changing type at the same time, so it does not settle it.
Where KMT2A and its fusions sit inside the nucleus — SNL1/SNL2, nuclear speckles and the condensate question
KMT2A does not float evenly through the nucleus — it collects into bright dots, and two short stretches of the protein (amino acids 388-432 and 1,051-1,089) are what put it there. Every leukemia fusion keeps both stretches, so the fusion protein inherits the same dotted address, and whether those dots are true liquid droplets is still an open question for KMT2A::AFF1 specifically.
Where KMT2A-rearranged leukemia physically sits — marrow niche, CNS, blood, liver/spleen and skin
KMT2A-rearranged leukemia is not just in the bone marrow: it spills into the blood in enormous numbers, reaches the fluid around the brain in roughly half of infants, swells the liver and spleen, and in newborns can show up as lumps in the skin. Where the cells sit changes both how they behave and how hard they are to kill, because the marrow and the brain each shelter them in different ways.
Where KMT2A::MLLT1 actually sits: chromatin occupancy is mapped, subnuclear localisation of the endogenous fusion has never been imaged
We know in fine detail which genes this fusion protein sits on, because that is measured by pulling the protein down with its DNA. We have never seen where it sits inside the nucleus of a real patient's leukaemia cell, because every picture that exists is of a tagged protein pushed into a cell that does not have the disease.
Where KMT2A::MLLT3 sits in the nucleus: two papers, both overexpression, both from before 2005 - and the fusion carries no AF9 import signal
Only two published studies have ever looked at where this fusion protein sits inside the nucleus, and both pushed extra copies of it into cells that do not have the disease. Both found it goes where the KMT2A half tells it to go.
Where LEDGF/p75 actually is: bound to chromatin through mitosis, in the nuclear interior with H3K36me3 - and never once imaged in a KMT2A-rearranged cell
LEDGF is the protein that helps hold the leukemia fusion's partner complex onto DNA, and unlike most chromatin proteins it stays stuck to the chromosomes even while the cell is dividing. The best pictures put it in the open interior of the nucleus next to a mark called H3K36me3 and away from the dense edge - but every one of those pictures was taken in HeLa cervical-cancer cells, never in a leukemia cell carrying a KMT2A fusion.
Where SET::NUP214 actually is: mobile intranuclear bodies that trap CRM1, dissolve in mitosis and under CRM1 inhibitor, and reassemble on washout
The fusion protein does not spread evenly through the nucleus. It gathers into bright spots that also hold the cell's main export carrier, and those spots vanish when the cell divides or when the export carrier is drugged - then come straight back when the drug is removed.
Where wild-type PICALM actually is: coated pits, the Golgi, coated vesicles - and a nuclear pool nobody has explained
The normal PICALM protein sits mostly at the cell surface where it helps build the cages that pull things into the cell, and also around the Golgi. A smaller amount goes into the nucleus, and nobody knows what it does there.
Which step is rate-limiting, and what compensates when you block it — the bite-back map for the causal chain
Walking through the fusion's six steps, only two are both necessary and selective enough to drug: the menin clamp and the histone mark. For each, we now know what the cell does to get around the drug — and in every case the escape route is already published.
Why a nucleoporin fusion needs menin-MLL1: the second menin-dependent fusion family, and what it borrows rather than carries
NUP98 fusions contain no piece of the MLL1 protein at all, yet blocking the menin-MLL1 handshake shuts them down. The reason is that they borrow the cell's own normal MLL1 and use it to hold open the genes that keep the leukaemia cell immature.
Why endocrine tumours do not carry KRAS mutations: menin blocks the proliferative arm of RAS
In hormone cells, the famous cancer gene KRAS actually slows growth instead of speeding it, because menin blocks the growth-promoting branch of its signal. Remove menin and the same signal flips to promoting growth - and a diabetes drug that would do nothing on its own then starts making the cells divide.
Wild-type AF10 is called nuclear on a construct and found in both fractions in tissue - and that disagreement decides whether CALM-AF10's cytoplasmic life is a gain or an inheritance
AF10 is annotated as a nuclear protein on the strength of a paper that found a nuclear-targeting signal in it. But the one study that separated real mouse tissue into nuclear and cytoplasmic parts found AF10 protein in both. That matters, because the leukemic CALM-AF10 fusion is mostly cytoplasmic, and nobody can say whether that is something the fusion does or something AF10 already did.
With no eraser, how fast the histone itself is replaced decides how much H3K79 methylation there is - and that sets the clock a DOT1L drug has to beat
Because nothing removes the H3K79 tag, the only way to lose it is to swap out the histone carrying it - either when the cell copies its DNA or as histones are routinely exchanged. That means how much tag a gene has depends on how fast its histones are replaced, and it means a drug blocking the enzyme has to stay on board through several histone turnovers before the tag disappears.
XPO1 E571K: a change-of-function hot spot in the NES-binding groove - and it does NOT confer resistance to the drugs that bind the same groove
One spot in XPO1 is mutated over and over in a few B-cell cancers. It sits in the pocket that grips cargo, and it changes which cargo gets gripped rather than switching the protein off. Reassuringly, the drugs that bind that same pocket still work on the mutant.
XPO1 mRNA goes UP when you inhibit the protein - the pharmacodynamic marker that made the paediatric trial interpretable, and what it cannot tell you
When a drug blocks XPO1, the cell notices and makes more XPO1 messenger RNA. That rise is easy to measure in a blood sample and proves the drug reached its target - which is why a small children's trial could say with confidence that the dose was doing something.
You can lower LEDGF without a LEDGF ligand: its protein half-life is about 4 hours against a proteome average of 20, so blocking guanine nucleotide or rRNA synthesis takes it out first
LEDGF is an unusually short-lived protein, so anything that slows down protein-making removes LEDGF before it removes most other proteins. Two existing drugs do exactly that, and in mice they lower LEDGF, break up the leukaemia complex and make the menin drug work better.
ZNF384 fusions make lineage-ambiguous leukaemia by acting in a haematopoietic stem cell - and the founding lesion is in the HSC itself, unlike Ph-like B-ALL
The ZNF384 fusion is not made in an already-committed B cell. It is made much earlier, in a blood stem cell, and from there it pushes the cell to make a bit of everything - which is why the resulting leukaemia cannot decide what lineage it is.
AKAP95 is phenotypically silent at steady state and required only when a cell is changing state - the same shape in ATRA differentiation and in MLL-AF9 transformation, thirteen years apart
Taking AKAP95 away from a resting cell does almost nothing. Take it away while the cell is trying to change into something else - differentiating, or being transformed by a leukaemia fusion - and the change fails. It is a transition protein, not a maintenance protein.
Bone marrow haematopoietic stem cell — the postnatal comparator, and why the same fusion behaves differently here
After birth, blood stem cells live in the bone marrow. Putting the KMT2A::AFF1 fusion into these postnatal stem cells is much harder to turn into leukaemia than putting it into fetal cells, which is a central clue about why infant disease is different.
Bone-marrow adipocyte — the fat cell that suppresses normal blood formation and feeds leukaemia a chemoresistance signal
Bone marrow contains fat cells, and the more fat a region of marrow holds, the less blood it makes — fatty replacement of marrow is what a pathologist sees after chemotherapy or in marrow failure. Recent work shows these fat cells also release leptin, a hormone that makes leukaemia cells harder to kill with chemotherapy.
Bone-marrow sinusoidal endothelial cell — the vessel wall that keeps stem cells alive, and posts vesicles that accelerate MLL-AF9 leukaemia
The bone marrow drains through wide, leaky, slow-flowing vessels called sinusoids, and the cells lining them do more than carry blood: they release the survival factors blood stem cells need and form the doorway through which new blood cells leave the marrow. In a mouse model of this leukaemia, these lining cells also send out tiny packages that speed the disease up.
CAR cell (CXCL12-abundant reticular) / LEPR+ perivascular stromal cell — the cell that makes the signal holding blood stem cells in the marrow
These are branching support cells wrapped around the blood vessels of the bone marrow, and they produce most of the chemical beacon (CXCL12) that keeps blood stem cells from drifting away. In leukaemia they are the cells that feed and shelter the cancer's own stem cells.
Cardiomyocyte - the non-haematopoietic cell that two of this graph's drug targets are required in, which makes it the graph's clearest bystander-organ liability
Heart muscle cells are not blood cells and have no part in leukaemia, but two proteins this project wants to drug turn out to be needed in them. That makes the heart the place where these drugs are most likely to do harm that has nothing to do with the disease.
Common lymphoid progenitor (CLP) — the CD10-positive lymphoid route that infant KMT2A leukaemia appears NOT to take
The common lymphoid progenitor is the bone-marrow cell that has given up making myeloid cells and will become a B cell, a T cell or a natural killer cell. There appear to be two separate routes into B-cell development in humans, and the evidence points to infant leukaemia coming from the other one.
Common myeloid progenitor (CMP) — the fork at which KMT2A leukaemia almost always turns one way
The common myeloid progenitor is the bone-marrow cell that still has a choice: it can go on to make neutrophils and monocytes, or red cells and platelets. KMT2A-rearranged leukaemia almost always ends up on the first branch, and that lopsidedness is itself a clue about what the fusion can and cannot do.
Conventional (myeloid-related) dendritic cell - the DC lineage RELB actually builds, and the thymic part of that phenotype is architecture rather than a cell-intrinsic requirement
Dendritic cells are the immune system's messengers, and they come in kinds that are built by different genes. This entry records the kind that needs the gene RELB - and it corrects a near-miss in this graph, where RELB had been attached to a different dendritic cell type because no node for this one existed.
Diagnosis versus relapse in the same child: relapse comes from a subclone that was already there and already quiescent, shown twice, seven years apart, by two unrelated methods
When infant KMT2A-rearranged leukaemia comes back, it does not usually come from cells that changed during treatment. It comes from a group of cells that was already present at diagnosis and was already sitting quietly, which is why treatment aimed at the dividing bulk misses it.
Early T-cell precursor (ETP) — the thymocyte that has not yet given up being myeloid, and where KMT2A-rearranged T-ALL sits
The early T-cell precursor is the newly arrived cell in the thymus that is starting to become a T cell but has not yet abandoned the option of becoming a myeloid cell instead. Leukaemias arising at this stage look like T-cell leukaemias but behave like myeloid ones, and they respond badly to standard T-cell treatment.
Erythroid progenitor - the committed red-cell precursor, distinct from the bipotent MEP that two entries had to borrow, and the compartment that pays for any haem-pathway or fetal-globin intervention
Erythroid progenitors are the marrow cells committed to becoming red blood cells; they are the stage after the shared red-cell/platelet ancestor. This graph needed a node for them because two of its genes have their only measured blood-cell requirement here, and because anything that attacks haem or globin production in a leukaemia will hit these cells too.
Fetal CD19+CD34+CD10-negative PreProB progenitor — a cell that exists before birth and almost not at all after
There is a B-cell progenitor that is common before birth and nearly absent in adults, and it lacks the CD10 marker that adult B-progenitors carry. Infant KMT2A leukaemia blasts are typically CD10-negative and share this cell's gene expression, which makes it a strong candidate for the cell where the disease starts.
Fetal liver haematopoietic stem and progenitor cell — the leading candidate cell of origin for t(4;11) infant ALL
Before birth, blood-forming stem cells live in the fetal liver, not the bone marrow. The gene activity of t(4;11) infant leukaemia looks most like these fetal liver cells, which is the strongest clue that the disease is born in the womb.
GMP-like state — the destination of therapy-driven lineage switch and, separately, the state most sensitive to menin inhibition
Leukaemia cells can take on the character of a particular normal bone-marrow cell — the one that makes neutrophils and monocytes — without actually being that cell. Two separate studies found that this is where cells go when therapy forces them to change identity, and that this is also where the newest class of drug for this disease works best.
Granulocyte-monocyte progenitor (GMP) — the committed cell an MLL fusion can turn back into a stem cell
The granulocyte-monocyte progenitor is a bone-marrow cell that has already decided to become one of two kinds of infection-fighting white cell and has given up the ability to renew itself indefinitely. The landmark finding in this disease is that an MLL fusion can put that ability back: a committed, short-lived cell becomes a self-renewing leukaemia stem cell without otherwise changing what it is.
H3K4me3 oscillates around the clock and MLL1 is what lets it - a cell state of KMT2A biology this project has never once mentioned
The chemical tag KMT2A writes on DNA packaging rises and falls on a 24-hour rhythm at clock-controlled genes, and without KMT2A the rhythm fails. Nothing in this corpus's 2,396 pre-existing entries had noticed that the gene it is named after is part of the body clock.
Haemogenic endothelium — the vessel-wall cell that gives birth to blood
Haemogenic endothelium is a rare cell that looks and behaves like part of a blood-vessel wall but can peel away and turn into a blood stem cell. It exists for only a few days in the embryo, mostly in the floor of the main artery, and it is where the entire blood system comes from.
iALL-HOXA9 — the HOXA-high state of infant KMT2A::AFF1 ALL, and the one everything is modelled on
The other half of these infant leukaemias runs the classic HOXA programme. It has the better outlook, and it is the half that nearly every laboratory model represents.
iALL-IRX1 — the HOXA-negative, IRX1-high state of infant KMT2A::AFF1 ALL
About half of babies with this leukaemia have cells in a different state: the HOXA genes are off and IRX1 is on instead. These children are diagnosed younger and do worse, and almost nothing in the lab models this half.
LIN28B-dependent fetal B- and NK-lymphopoiesis - the NORMAL state the leukaemia requirement is parasitic on, and the reason the therapeutic window is narrow by construction
Before birth, blood stem cells need LIN28B in order to keep multiplying, stay stem-like, and make B cells and NK cells at all. That is normal biology, not cancer biology - which is why removing LIN28B stops infant leukaemia starting and also stops the baby's own immune system being built.
Lymphoid-primed multipotent progenitor (LMPP) — the bi-potent cell that explains how one fusion makes two lineages
The LMPP is a blood progenitor that has already started down the lymphoid road but has not thrown away its myeloid options. If KMT2A::AFF1 lands here, the resulting leukaemia can present as B-cell, as myeloid, or as both.
Megakaryocyte — the platelet factory whose loss causes the bruising that brings these children in, and which also holds stem cells still
A megakaryocyte is a giant bone-marrow cell that sheds thousands of platelets, the fragments that plug damaged blood vessels. When leukaemia crowds the marrow these cells are lost first, which is why unexplained bruising and pinpoint skin bleeding are often the reason a child is brought to hospital.
Megakaryocyte-erythroid progenitor (MEP) — the branch KMT2A fusions almost never take, which is why it was a usable control
The megakaryocyte-erythroid progenitor is the bone-marrow cell that makes red blood cells and the giant cells that shed platelets. KMT2A-rearranged leukaemia essentially never comes from this branch, and that absence turned out to be useful: it let the project rule out one explanation for its iron-and-haem finding.
Menin is dispensable for blood at rest and required for blood under stress - and that is a cell state, not a hedge
Mice without menin make normal blood while nothing is going wrong. Put those cells in competition or make them rebuild a blood system and they fail. So the safety argument for menin-blocking drugs depends on which state the patient's marrow is in, not on an average.
Monoblast — the immature monocyte that names FAB M5a, the dominant morphology of congenital KMT2A leukaemia
A monoblast is the youngest recognisable step on the way to becoming a monocyte: a large cell with a big pale nucleus that has committed to the monocyte path but cannot yet do a monocyte's job. When a leukaemia is made mostly of these cells it is called monoblastic, and that is the commonest appearance of leukaemia in newborn babies, most often with a KMT2A break.
Monocyte — the graph's most-discussed undefined cell identity, and the confound sitting under its own DepMap haem signal
A monocyte is a large white blood cell that patrols the bloodstream for a few days and then settles into tissue as a macrophage, where one of its main jobs is to swallow worn-out red cells and recycle the iron locked inside them. That normal day job matters here because most of the KMT2A cell lines this project uses are monocytic, so an apparent iron-and-haem dependency of the fusion could simply be the physiology of the cell the fusion happens to be in.
Multipotent progenitor (MPP) — the stage where the undifferentiated KMT2A-r infant leukaemias map
The multipotent progenitor is the step just after a blood stem cell: it can still make every kind of blood cell but can no longer renew itself indefinitely. Some of the most treatment-resistant infant KMT2A leukaemias look transcriptionally like this very early stage.
Nestin+ mesenchymal stem cell — the niche cell that MLL-AF9 leukaemia wakes up and then loses
These are the marrow's own stem cells for bone and fat, sitting right next to blood stem cells and next to nerve fibres, and they release the signals that keep blood stem cells healthy. Leukaemia disturbs them: it drives them out of their resting state and pushes them towards becoming bone-forming cells instead of blood-supporting ones.
Neuron - the off-target cell type that defines what KANSL1 loss costs a human being
A neuron is the cell that carries electrical signals in the brain. It matters to a leukaemia graph because losing one working copy of KANSL1 - a protein in a complex this project tracks inside the fusion interactome - causes a neurodevelopmental syndrome in people, which is the clearest measurement anywhere of what turning KANSL1 down would cost.
NG2+ periarteriolar pericyte — the cell that holds blood stem cells asleep, and the one MLL-AF9 leukaemia destroys
Wrapped tightly around the small arteries of the bone marrow is a rare kind of support cell that keeps nearby blood stem cells in a deep resting state. When these cells are removed, the stem cells start dividing and the long-term stem-cell pool shrinks — and in a mouse model of this leukaemia, the disease destroys exactly these cells.
Osteoblast — the bone-building niche cell whose expansion suppresses CML and makes MLL-AF9 AML worse
Osteoblasts are the cells that build bone, and they line the inner surface of the marrow cavity where some blood stem cells sit. They were the first cells proposed to form the stem-cell niche — and, awkwardly for anyone hoping to treat leukaemia by manipulating them, increasing their number helps one kind of leukaemia and hurts another.
Plasma cell - the end of the B lineage, and the only cell type in which ELL2 has a documented job
A plasma cell is a B cell that has finished growing up and turned into a factory that pumps out antibody. It is here because ELL2, one of the elongation factors this graph tracks inside the KMT2A fusion complex, has its entire known day job in this one cell type and nowhere else.
Plasmacytoid dendritic cell — the interferon cell that KMT2A-rearranged leukaemia occasionally becomes, and is more often mistaken for
The plasmacytoid dendritic cell is a rare white cell whose whole job is to detect viruses and flood the body with interferon, the alarm signal that makes you feel ill during a viral infection. A KMT2A-rearranged leukaemia can occasionally take on this cell's appearance, and more often it imitates it closely enough to be misdiagnosed.
Pre-B cell — the stage KMT2A::AFF1 leukaemia never reaches, and why that boundary is informative
A pre-B cell is the next step after the pro-B cell: it has built a test version of its antibody receptor and uses it as a checkpoint. KMT2A::AFF1 leukaemia is blocked before this step, and which leukaemias sit before or after it is a real clinical distinction.
Pre-pro-B cell (murine Hardy Fraction A) - the stage before pro-B, with germline immunoglobulin genes, and the stage RYBP loss actually removes
Before a mouse B cell becomes a pro-B cell it passes through an earlier step called pre-pro-B, in which its antibody genes have not been cut and rejoined at all. This entry exists because a gene in this graph, RYBP, was shown to be needed for exactly that step, and the graph had no word for it.
Pro-B cell — the stage at which KMT2A::AFF1 leukaemia is arrested, and the phenotype the diagnosis is made on
The pro-B cell is an early step in making a B lymphocyte, before the cell has built a working antibody receptor. KMT2A::AFF1 leukaemia is stuck at about this step, which is why the blasts look like immature B cells that never finish growing up.
Quiescence (G0) — the state 15 entries invoke to explain measurable residual disease, defined here for the first time
A quiescent cell is alive but has stepped out of the cycle of growing and dividing, like a seed that stays viable in the soil without sprouting. Almost all leukaemia drugs work best on cells that are dividing, so the handful of leukaemia cells sitting in this resting state are the ones that survive treatment and can restart the disease months or years later.
Quiescence in the blood stem cell is actively held there by MLL1 - and losing MLL1 pushes the stem cell INTO cycle while pulling the progenitor OUT of it
Blood stem cells spend most of their life asleep, and the gene that breaks in infant leukaemia is one of the things keeping them asleep. Take it away and the stem cells wake up and burn out, but the slightly more mature cells one step down do the opposite and slow down.
Replication stress and interstrand-crosslink repair as a cell state - an ATR dependency measured in mouse MLL-rearranged AML, and the same stress implicated in breaking the MLL locus in the first place
Copying DNA is not smooth: forks stall, and a repair crew has to restart them before the chromosome breaks. This entry records the evidence that leukaemia cells carrying an MLL/KMT2A rearrangement live under that kind of stress, and that the same stress is one of the ways the rearrangement is thought to be created.
Retained fetal programme — the infant-versus-adult stem cell difference, and the reason KMT2A::AFF1 needs no second mutation
A blood stem cell in a fetus is not the same as one in an adult: it grows faster, renews more readily and runs a set of genes that are switched off after birth. Infant leukaemia blasts keep that fetal setting, and new work shows that one fetal gene in particular has to be switched on for the fusion to cause leukaemia at all.
SAM / methionine dependence - the one metabolic state in this graph demonstrated in KMT2A-rearranged lymphoblastic leukaemia itself, with a drug and a PDX behind it
Every chemical tag a cell puts on DNA or on histones is paid for with a molecule called SAM, which is made from the amino acid methionine in food. KMT2A-rearranged leukaemia carries an unusually heavy load of those tags, so it has an unusually large SAM bill - and taking methionine away, or blocking the enzyme that makes SAM, kills it.
SET::NUP214 adult T-ALL is gamma-delta-lineage-restricted and resistant to steroids and chemotherapy early - the same cell state CALM-AF10 occupies
Patients whose T-cell leukaemia carries SET::NUP214 almost all have the same immature, gamma-delta type of T cell, and their disease barely responds to steroids or the first round of chemotherapy - though with transplant their final outcomes end up similar to everyone else's.
Sympathetic nerve fibre in bone marrow — the wiring that sets the niche's daily rhythm, through beta-3 in health and beta-2 in leukaemia
Bone marrow is wired: nerve fibres from the fight-or-flight system run alongside its small arteries and release noradrenaline on a daily schedule, which turns down the signal that holds blood stem cells in place and releases a pulse of them into the blood each day. Leukaemia damages these nerves, and so does chemotherapy.
Th17 cell - the mature CD4 T-helper subset whose differentiation and maintenance require MENIN, which makes it the clearest published on-target normal-tissue liability of the menin interface
Th17 cells are a kind of helper T cell that produces the signal IL-17 and defends mucosal surfaces, and they also drive several inflammatory diseases. This graph needs them because menin - the protein the leading KMT2A drugs block - is required to make Th17 cells and to keep them made.
The apoptotic priming state — KMT2A-rearranged infant ALL is the one childhood ALL subgroup that is BCL-2-dependent rather than BCL-XL-co-dependent, and BH3 profiling is how that was measured
Every cell sits somewhere on a scale between ready to die and hard to kill, and a test called BH3 profiling reads that position directly by dripping death signals onto isolated mitochondria. Infant KMT2A-rearranged leukaemia sits in an unusual place on that scale: it leans on one survival protein, BCL-2, where most childhood leukaemias lean on two.
The CD24-positive pre-leukaemic state — what the index fusion makes before it makes leukaemia
When the KMT2A::AFF1 fusion is switched on at the right stage of development, the cells do not become leukaemia straight away: they first enter a distinct in-between state carrying a surface marker called CD24. That in-between state is a target nobody is aiming at.
The cell CALM-AF10 arrests: an early T-precursor in adults and a TCR-positive thymocyte in children - two states, split by age (corrected 2026-09-21)
When CALM-AF10 causes T-cell leukaemia, it is almost always in a particular early kind of T cell - the gamma-delta lineage or an even earlier precursor - and essentially never in the commoner alpha-beta lineage. Which of those two it is predicts whether the patient survives.
The CNS-adapted state — leukaemia-propagating cells reprogrammed by the central nervous system niche, and worse at restarting systemic disease afterwards
KMT2A-rearranged infant leukaemia often spreads to the fluid around the brain, and the cells that settle there change: they switch on programmes for handling fat and cholesterol, for low oxygen, and for inflammation. Unexpectedly, when those cells were put back into mice they were worse at causing leukaemia again, not better.
The CSF1R-positive fetal lymphoid-primed multipotent progenitor — a fetal-restricted subset of LMPPs that makes AML rather than ALL when KMT2A::MLLT3 is switched on, and exists in human embryos
Among the multipotent blood progenitors in a fetal liver there is a subset carrying a receptor called CSF1R, and these are the ones that turn into myeloid leukaemia rather than lymphoid leukaemia when the KMT2A fusion is switched on in a mouse. The same population exists in early human embryos, and blocking CSF1R kills a KMT2A-rearranged paediatric leukaemia cell line, which makes it a possible target.
The differentiation block — what it actually is, operationally, in KMT2A::AFF1 leukaemia
The blasts in this leukaemia are frozen at an immature stage: they keep dividing but never finish becoming working blood cells. The freeze is maintained actively, by the fusion protein rewriting which genes stay switched on, which is why releasing it is a treatment strategy.
The early lymphocyte precursor (ELP) state — what KMT2A-rearranged infant B-ALL scores as against a developing bone marrow reference, and why it is earlier than the pro-B arrest this graph also records
When you compare single leukaemia cells against a map of normal developing bone marrow, KMT2A-rearranged infant leukaemia lands at a very early lymphoid point — earlier than where its surface markers suggest — and carries myeloid features at the same time. Those non-physiological marker combinations are exactly what a targeted therapy needs to tell cancer from healthy tissue.
The endocrine cell's resting state is an actively maintained G1 restraint, and menin is what maintains it
Hormone-producing cells are not idle by default - they are actively held still, by brake proteins whose genes menin keeps switched on. Remove menin and the brakes come off; that is what a menin-blocking drug does, and in the pancreas it is being developed deliberately as a way to regrow insulin cells.
The G1 state in KMT2A-rearranged leukaemia — where CDK6 does a job that is not about the cell cycle at all
Before a cell commits to dividing it passes a checkpoint in a phase called G1, controlled by a pair of proteins called cyclin D and CDK6. In this leukaemia CDK6 turns out to matter for a different reason: blocking it does not mainly stop cells dividing, it lets them mature — which is the same thing the fusion prevents.
The HOXA-negative, IRX1-negative KMT2A-rearranged blast
Some KMT2A-driven leukaemias have the fusion protein on neither HOXA nor IRX1. That breaks the idea that these two are the only two options, and nobody knows what the fusion is doing in those cells instead.
The HSPC-like blast fraction — a minority of blasts that transcriptionally resemble haematopoietic stem and progenitor cells, found in the blood of the youngest KMT2A-rearranged patients and predicting drug response and outcome across leukaemia subtypes
Inside one patient's leukaemia, not every cancer cell is the same: a small group of them reads, gene for gene, like a blood stem cell rather than like the B-cell the rest of the leukaemia is pretending to be. In KMT2A-rearranged leukaemia this group is found in the circulating blood of the youngest babies, and across many kinds of leukaemia the size of this group predicts how well chemotherapy works and how long the patient lives.
The hypoxic, glycolytic state — the one state that appears independently in leukaemia-initiating cells, in measurable residual disease, and in KMT2A-rearranged B-ALL as a class
The bone marrow is a low-oxygen place, and leukaemia cells switch on a survival programme in response. Three separate studies, using different methods, found that same programme switched on in the cells that start the leukaemia, in the cells left behind after chemotherapy, and in KMT2A-rearranged disease as a whole — and blocking it made chemotherapy work better.
The innate-immune-rewired state — LAMP-5 is a direct fusion target that routes signalling away from interferon and toward NF-kB, and the cell depends on that routing
Leukaemia cells receive alarm signals through the same machinery as normal immune cells, but a protein called LAMP-5 — switched on directly by the fusion — diverts those signals down an inflammatory route and away from an antiviral one. Removing LAMP-5 flips the switch back and the cells stop growing.
The interferon state is not baseline in this disease — it is something a menin inhibitor creates, through endogenous retrovirus induction and MHC class II
Cells can be pushed into an antiviral alarm state by switching on ancient virus sequences buried in our own DNA, and that alarm makes them visible to the immune system. In KMT2A-rearranged leukaemia this alarm is not switched on by default — a menin inhibitor switches it on, and the cells then become much easier for donor T cells to kill after a transplant.
The lineage-switched AML state of infant KMT2A::AFF1, described by chromatin rather than by flow cytometry - fusion retained on PAX5 and MEIS1, lost from EBF1 and HOXA, myeloid promoters gained by DOT1L and ENL
After an infant's lymphoid leukaemia turned myeloid under treatment, the fusion protein stayed on some genes, left others, and the cells' switched-on genes changed. This is what that state looks like measured on the DNA rather than by surface markers.
The lipid-dependent state — ACSL4 and polyunsaturated fatty acid utilisation in KMT2A-rearranged AML, against ELOVL1 and saturated elongation in KMT2A::AFF1 infant ALL
Leukaemia cells need particular kinds of fat, not just energy. In the myeloid form of KMT2A-rearranged leukaemia the requirement is for an enzyme that activates long polyunsaturated fats; in the infant lymphoid form it is a different enzyme that lengthens saturated ones. Same founding lesion, two different fat requirements.
The MTAP-deleted, MTA-high state - a genotype-imposed metabolic state that this project's KMT2A-rearranged arm turns out NOT to have, and that its own screen could never have detected directly
When a tumour deletes the tumour-suppressor gene CDKN2A it usually deletes a neighbouring housekeeping gene, MTAP, by accident. Losing MTAP lets a waste metabolite build up that partly jams an enzyme called PRMT5, which leaves those cells unusually easy to finish off by blocking that enzyme - and it also makes drug-sensitivity numbers in mixed cell-line panels hard to interpret.
The neonatal window — the state in which an MLL fusion can transform, bounded at both ends
There is a period around birth when blood cells can be turned into leukaemia by an MLL fusion much more easily than before or after. The window is a property of the cell's condition and of its surroundings, not of the fusion.
The NK cell - the effector arm of haploidentical transplant, the one this disease is documented to evade, and the cell CLEC12A was named on
NK cells kill cells that look wrong without needing to be trained first, and they are the main reason a mismatched bone-marrow transplant can cure leukaemia. This disease has three documented ways of hiding from them, and almost nothing in this graph was about them until now.
The non-synchronisable state: five KMT2A-rearranged lines refuse mimosine, double thymidine and nocodazole, and the fusion protein does not fluctuate through the cycle
Researchers normally line up cells at the same point in their division cycle using drugs, so that everyone in the dish is doing the same thing at the same time. Five of the standard KMT2A-rearranged leukaemia cell lines simply will not line up, and the fusion protein inside them stays at a constant level throughout division while the normal version rises and falls twice.
The OXPHOS-dependent state — established in KMT2A-rearranged AML, and inverted in KMT2A-rearranged B-ALL where OXPHOS inhibition makes leukaemia-initiating cells appear
Cells make energy two ways: burning fuel with oxygen in the mitochondria, or a faster, cruder route that does not need oxygen. Myeloid KMT2A-rearranged leukaemia leans hard on the mitochondrial route and can be attacked there, but in the lymphoid form of the same disease blocking that route backfires — it shrinks the bulk tumour while increasing the cells that can restart it.
The primed chromatin state — what is actually established, and why the word 'bivalent' should be used carefully here
The leukaemia cells keep two sets of instructions accessible at once: the B-cell program they are running and the myeloid program they are not. That dual readiness is what lets them change lineage later, and it is written in the chemical marks on their DNA and histones.
The pro-B arrest state — what a KMT2A::AFF1 blast is actually doing while it sits there
The leukaemia cells are held at an early B-cell step where they are still receiving growth signals but have not built a working antibody receptor. They are not dormant: they are actively proliferating, resistant to steroids, and still able to switch lineage if pushed.
The relapse-initiating persister state — present at diagnosis, identifiable before a single dose is given, and defined by basal glucocorticoid-response activation plus small size plus quiescence
In infant KMT2A-rearranged leukaemia, the cells that will cause relapse are already present at diagnosis and can be told apart from the rest by reading single cells: they are smaller, they sit still rather than dividing, and their steroid-response genes are already switched on before any steroid is given. Counting them at diagnosis predicted relapse better than the test currently used.
The state under menin inhibition — differentiation in AML, rapid death in ALL, and a GMP-like exit in the cells that switch lineage instead
Menin inhibitors are usually described as drugs that let frozen leukaemia cells finish growing up. That is what happens in the myeloid form of the disease. In the lymphoid form — which is most infant cases — the same drug kills the cells quickly instead, and describing it as a differentiation therapy there is simply wrong.
The stem-like self-renewal state — a limited stem-cell module grafted onto a committed cell
Normal blood cells lose the ability to copy themselves as they mature. KMT2A fusions switch a small part of that self-copying program back on without making the cell a stem cell again, which is what lets an already-committed cell keep the leukaemia going forever.