When one thing goes wrong, do several? In infant KMT2A-rearranged leukaemia the answer is no, and that cuts against the intuition
also called single-hit versus multi-hit · the quiet genome problem · what is the second hit here
The usual expectation is that cancer needs several things to go wrong at once. This leukaemia is the exception: its genome is one of the quietest ever sequenced, with barely more than one extra mutation on average. That is good news for finding a single target and bad news for finding a second one.
The same thing, for a scientist
The PI's stated intuition was that when one thing goes wrong, several do. For infant KMT2A-rearranged ALL the data contradict it, and the contradiction is worth stating plainly rather than softening. The dominant clone carries a mean of 1.3 non-silent mutations, one of the lowest burdens of any sequenced human cancer; kinase-PI3K-RAS lesions are present in 47% of cases but are often subclonal and frequently lost at relapse; and non-infant KMT2A-rearranged leukaemia carries a mean of 6.5 with frequent epigenetic-regulator mutations that are nearly absent in infants [E1]. Twin concordance approaching unity with a latency of months points the same way (`epi-twin-concordance-infant-kmt2a-leukemia`, `epi-latency-in-utero-to-diagnosis-kmt2a`). **And yet the fusion is demonstrably not sufficient in the one human experimental system available**: MLL-AF4 in cord-blood CD34+ cells enhances engraftment and clonogenicity but does not transform, and adding FLT3 activation does not rescue it [E2]. So the question is not whether a second hit exists but what kind of thing it is, and the leading candidate is not a mutation at all but the fetal cell context itself (`hyp-second-hit-is-developmental-window-not-mutation`).
The tension, laid out without resolving it prematurely1 deeper
Evidence that one thing went wrong:
- 1.3 non-silent mutations in the dominant clone, 0.6 expressed 1
- Cooperating RAS lesions subclonal and lost at relapse 1
- No second recurrent driver from whole-genome sequencing 1
- Near-total monozygotic twin concordance with months-long latency
- No recurrent structural second event identified
Evidence that something else is needed:
- MLL-AF4 does not transform human cord-blood CD34+ cells 2
- Adding the most obvious candidate cooperating lesion, FLT3 activation, also fails 2
- Older children with the same fusion need five times as many mutations to get the same disease 1
- A documented fusion-positive infant clone cleared spontaneously rather than progressing (
mechanism-preleukemic-clone-spontaneous-clearance)
Both lists are solid. The resolution that fits both is that the second requirement is real and is not genetic.
What a non-genetic second hit could be
Four candidates, in rough order of how much support exists in this graph:
- The fetal target cell itself. A first- or second-trimester fetal liver progenitor has a chromatin and transcriptional state that a postnatal cord-blood CD34+ cell does not, which would explain 2 as a wrong-cell result rather than an insufficiency result.
- The epigenetic ground state. Bivalent, primed chromatin at HOXA in the fetal progenitor (
cell-state-primed-bivalent-chromatin,mechanism-bivalent-primed-chromatin-fusion-targets) may be what the fusion requires and adults must acquire by mutation. - Loss of the reciprocal. The reciprocal derivative is variably retained, and its loss is a structural event that bulk mutation counts do not capture (
mechanism-reciprocal-kmt2a-fusion-chromatin-opening). - A non-coding driver. Every published burden figure is a coding count; a recurrent regulatory mutation would be invisible to all of them.
Consequences for this project's strategy
A near-sole driver is the best possible setting for an interface-directed drug: there is one thing to hit, it is truncal, and it is retained at relapse (mechanism-fusion-clonal-stability-diagnosis-to-relapse). It is simultaneously the worst possible setting for combination therapy, because the genome hands over no synthetic-lethal partner and no co-target. The combinations available are therefore all against the same node by different mechanisms — menin interface plus degrader, menin interface plus RNA agent, menin interface plus DOT1L — rather than against two independent nodes. That is a much weaker form of combination and it is the honest framing the therapeutics beat should inherit from this one.
- A near-sole driver means no synthetic-lethal partner is handed over, so every combination this project can build attacks the same node twice rather than two nodes once, and correlated failure is the expected mode.
- It also means the ~50% relapse rate cannot be explained by mutation, so relapse must be explained by cell state or plasticity, which are far harder to drug than a mutation would have been.
- If the real second requirement is a developmental window that has closed by the time of diagnosis, then the disease at diagnosis may depend on something that is no longer present, and the therapeutic logic has to be rebuilt around what maintains rather than what initiated.
- Has anyone sequenced infant KMT2A-r whole genomes at adequate power for recurrent NON-CODING drivers? Every burden figure quoted anywhere in this graph is a coding count, and a regulatory second hit would be invisible to all of them. This is the most likely place a missed co-driver is hiding.
- Is the reciprocal derivative chromosome lost in the cases that progress and retained in those that do not? That is a structural second event that mutation counts do not capture and that is measurable in existing banked material.
- Would introducing the epigenetic-regulator lesions found in older children (CHD4, KDM6A, NSD1) into an MLL-AF4-expressing cord-blood CD34+ cell permit the transformation that FLT3 could not [E2]? That experiment would directly test whether older children's mutations substitute for infants' cell context, and nobody has run it.
- What is the minimal fetal cell state that permits transformation? If it can be specified, it can be engineered in an iPSC-derived progenitor, which would give the field the human model it has never had.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Infant KMT2A-rearranged ALL has one of the lowest somatic mutation frequencies of any sequenced cancer, with a mean of 1.3 non-silent mutations in the dominant clone, subclonal RAS-pathway lesions frequently lost at relapse, and a mean of 6.5 in older children with the same fusion.
The landscape of somatic mutations in infant MLL-rearranged acute lymphoblastic leukemias · Andersson AK, Ma J, Wang J, Chen X, Gedman AL, Dang J, et al. · Nature Genetics 2015What this evidence cannot show
Coding-focused; whole-genome coverage was applied to a subset, so recurrent non-coding drivers would be underpowered to detect. The infant versus older-child comparison differs in lineage as well as age.
Evidence that argues against this (1)
- A later, deeper cohort finds the genome is not quiet in the subgroup that matters most. Multi-omics on 84 infants with KMT2A-rearranged leukaemia - RNA sequencing, methylation array, whole exome and TARGETED DEEP sequencing - resolved five integrative clusters and showed that the number of RAS-pathway mutations predicts prognosis, with the most refractory cluster (IRX-type undifferentiated, IC2) carrying RAS-pathway mutations at 100% frequency and the heaviest burden. A mean of 1.3 non-silent mutations in the dominant clone is an average over a heterogeneous disease, and deep sequencing rather than standard exome coverage is what makes the RAS burden visible - so 'lowest mutation frequency of any sequenced cancer' is partly a statement about sequencing depth and about averaging across clusters. Isobe T et al. 2022
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
MLL-AF4 expression in human cord-blood CD34+ cells enhances engraftment and clonogenicity but does not immortalise or transform, and combining it with ligand-independent FLT3 activation still does not, implying additional (epi)genetic requirements.
Ligand-independent FLT3 activation does not cooperate with MLL-AF4 to immortalize/transform cord blood CD34+ cells · Montes R, Ayllon V, Prieto C, Bursen A, Prelle C, Romero-Moya D, et al. · Leukemia 2014What this evidence cannot show
Lentiviral overexpression in postnatal cord-blood cells with a 16-week xenograft follow-up. A negative transformation result is compatible with the wrong cell, the wrong developmental window, the wrong expression level or insufficient follow-up, and the authors' own framing is that cooperating lesions are needed rather than that the model is wrong.
Evidence that argues against this (1)
- A competing interpretation the experiment cannot exclude, with positive evidence behind it. Targeting Mll-AF4 to the earliest definitive haematopoietic cells of the mouse EMBRYO showed enhanced B-lymphoid potential, repopulation and self-renewal confined to a restricted window between embryonic days 12 and 14, strongest in the lymphoid-primed multipotent progenitor population. If the fusion's pre-leukaemic effect is window-dependent in this way, a negative transformation result in postnatal human cord-blood CD34+ cells is what that experiment would be expected to produce whether or not additional (epi)genetic requirements exist. The same study does report that this pre-leukaemic state alone is insufficient for disease, so it cuts both ways - but it locates the insufficiency in a system where the window was hit, which the cord-blood experiment cannot claim. Barrett NA et al. 2016
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.review
Near-total twin concordance with months-long latency independently constrains how much additional postnatal work can be required.
Leukemia in twins: lessons in natural history · Greaves MF, Maia AT, Wiemels JL, Ford AM · Blood 2003What this evidence cannot show
Review of case series with strong publication bias toward concordant pairs; the first documented discordant infant KMT2A-r pair appeared only in 2009.
Evidence that argues against this (1)
- A discordant monozygotic infant pair with spontaneous clearance of the fusion-positive clone shows concordance is not universal. Chuk MK et al. 2009
the quiet-genome data this entry interprets
a non-genetic second requirement is the reading that fits both halves of the tension
- related to →The reciprocal fusion (AFF1::KMT2A and friends): a second oncoprotein that opens chromatin for the first
loss of the reciprocal is a structural second event invisible to mutation counts
bivalent fetal chromatin is the leading candidate for the non-genetic requirement
stability of the driver and silence of the genome are the same observation from two angles
- related to →Experiment: does MLL-AF4 transform a staged fetal progenitor when it cannot transform cord blood?
Counter-evidence (Isobe 2022, PMID 36042201): targeted deep sequencing of 84 KMT2A-rearranged infants found RAS-pathway mutations at 100% frequency and heaviest burden in the most refractory cluster, with mutation number prognostic -- the quiet-genome average hides a cluster in which a mutational second hit is universal
Counter-evidence (Barrett 2016, PMID 27396339): Mll-AF4 targeted to the mouse embryo confers self-renewal and B-lymphoid potential only in a restricted E12-E14 window, so a negative transformation result in postnatal cord blood is what that system would give either way