Correcting the fusion with genome editing: verdict NOT FEASIBLE for infant leukaemia, though intron-targeted disruption is a real published strategy
also called CRISPR fusion junction · base editing the breakpoint · genome editing of fusion oncogenes
You cannot glue the chromosomes back the way they were. What you can do is cut the fused gene again so it stops working, and this has been shown to shrink tumours in mice. Getting that machinery into every leukaemia cell in a sick baby, without hitting healthy cells, is the part nobody can do.
The same thing, for a scientist
Two things get conflated under 'correction' and only one of them exists. **Reversal** — restoring the two parental chromosomes — is not a real option: it would require a precise, bi-allelic, inter-chromosomal reciprocal recombination in every leukaemic cell, and nothing in the editing toolkit approaches that. **Disruption** is real and published. Martinez-Lage et al. devised a non-patient-specific CRISPR/Cas9 strategy that targets **two introns**, one in each partner gene, so the cut pair is only present together in a rearranged cell; this robustly disrupts the fusion oncogene specifically in cancer cells and reduced tumour burden and mortality in in-vivo models for both transcription-factor and tyrosine-kinase fusions [E1]. The intron-targeting trick is elegant precisely because it removes the need to know the patient's exact breakpoint, which is the usual objection to junction-directed editing. **The verdict is nonetheless not feasible for infant KMT2A-rearranged leukaemia**, for reasons that are about delivery and safety rather than about the editing concept: no delivery vehicle reaches a disseminated leukaemic compartment at the near-complete efficiency an oncogene-disruption strategy requires; a double cut on two chromosomes in a haematopoietic cell is itself a translocation-generating event, which is the mechanism of the disease being treated; and the target population is infants, in whom a genotoxic intervention with decades of follow-up risk is the hardest possible risk-benefit case.
Why the required efficiency is the killing constraint1 deeper
Oncogene disruption is not like gene addition. In a monogenic disease corrected in HSCs, a 20% edited fraction can be curative because corrected cells have a selective advantage. Here the opposite holds: unedited leukaemic cells retain the fusion, retain the fitness advantage, and regrow. The strategy therefore needs to approach completeness in a compartment that is by definition disseminated through marrow, blood, CNS and often skin (concept-cns-involvement-at-diagnosis, concept-leukemia-cutis-infant). No current delivery modality — LNP, AAV, or ex vivo — achieves that in vivo across those compartments.
The self-referential safety problem
The intron-targeting design makes simultaneous double-strand breaks on two different chromosomes 1. That is, mechanistically, exactly the configuration that produces a translocation: two broken chromosomes in spatial proximity, handed to error-prone end joining (mechanism-nhej-vs-alt-ej-at-fusion-junctions, mechanism-3d-genome-proximity-partner-choice). In a cell type whose defining lesion is a translocation, and in a patient with a whole lifetime ahead, inducing paired DSBs genome-wide is a hard sell. The evolutionary precedent is sobering: vertebrates needed to suppress RAG's transposition activity by more than a thousand-fold to make a programmable nuclease tolerable in lymphocytes, and even so it drives recurrent oncogenic deletions (mechanism-vdj-recombination-programmed-breaks).
Where editing IS the right tool in this project
Not as a therapy, as a reagent. Every high-value experiment in this beat and the mechanism beat needs precise editing: knocking a degron into the endogenous KMT2A allele (exp-degron-washout-recovery-kmt2a-aff1), building an endogenous-locus KMT2A::AFF1 in a fetal-like human progenitor to resolve the transformation failure (epi-fusion-positive-newborns-versus-leukemia-risk), and separation-of-function alleles at the menin interface. The editing verdict here is 'not a medicine', not 'not useful'.
| modality | genome editing (CRISPR/Cas9 intron-targeted disruption; base or prime editing at the junction) |
|---|---|
| verdict | NOT FEASIBLE for infant KMT2A-rearranged leukaemia as a therapy |
| what works in principle | intron-targeted double-cut disruption of the fusion, published and effective in vivo in non-leukaemic fusion models [E1] |
| reversal is not an option | restoring the two parental chromosomes would require precise reciprocal inter-chromosomal recombination in every cell; no tool approaches this |
| correct use in this project | as a laboratory reagent for degron knock-in, endogenous-locus fusion models and separation-of-function alleles |
blocking failures
[ "requires near-complete editing of a disseminated leukaemic compartment, because unedited cells retain the fitness advantage", "paired double-strand breaks on two chromosomes are themselves translocation-generating, in a disease caused by a translocation", "no delivery vehicle reaches marrow, blood and CNS leukaemic cells at the required efficiency", "infant population, so decades of secondary-malignancy follow-up risk" ]
- Inducing paired chromosomal breaks to treat a disease caused by paired chromosomal breaks is the clearest example in this graph of biology coming back to bite an intervention.
- Partial editing selects for unedited cells, so a subtherapeutic edit is worse than none: it debulks without curing and leaves a clone that has already survived the intervention.
- Any germline or fetal application would be permanent and would be applied on the basis of a penetrance nobody has measured.
- Would a nickase or base-editor design that disables the fusion without a double-strand break remove the translocation-generating objection? Disrupting a splice site in the retained partner intron with a base editor is conceptually available and, as far as this beat could establish, has not been attempted for any fusion oncogene.
- Does the intron-targeting strategy [E1] work for KMT2A fusions specifically, where the breakpoint cluster region is an 8.3 kb intronic window and the partner is one of more than a hundred genes? The KMT2A side is unusually favourable because the intron is large, well mapped and shared across partners; nobody has tested it.
- What is the maximum residual unedited fraction compatible with cure in a leukaemia? This is the number that decides whether editing will ever be a leukaemia therapy, and it is answerable by a mixing experiment in a xenograft model.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
A non-patient-specific CRISPR/Cas9 strategy that targets one intron in each partner gene disrupts fusion oncogenes specifically in cancer cells and reduced tumour burden and mortality in in-vivo models.
In vivo CRISPR/Cas9 targeting of fusion oncogenes for selective elimination of cancer cells · Martinez-Lage M, Torres-Ruiz R, Puig-Serra P, Moreno-Gaona P, Martin MC, Moya FJ, et al. · Nature Communications 2020What this evidence cannot show
Proof-of-concept in cell lines and xenograft/nude-mouse models with localised or favourable delivery; the fusions tested were not KMT2A fusions; efficacy is measured as reduced tumour burden rather than eradication, which is the crux for a leukaemia. The authors do not claim clinical readiness.
Evidence that argues against this (1)
- The strongest objection to a two-cut fusion-disruption strategy is not delivery, it is that the on-target edit itself is mutagenic at chromosome scale. Leibowitz et al. used single-cell whole-genome sequencing to show that CRISPR-Cas9 editing generates micronuclei and chromosome bridges that initiate chromothripsis (extensive rearrangement restricted to one or a few chromosomes), and demonstrated it not only in model cells but by editing at a clinically relevant locus in clinically relevant cells. The Martinez-Lage design deliberately makes TWO cuts on TWO different chromosomes in the same cell, which is the configuration most likely to produce a bridge or a micronucleus. In a haematopoietic cell, in an infant, to treat a disease caused by a chromosomal translocation, that is the intervention biting back in the most literal available sense. This does not refute the in vivo efficacy in E1; it says the safety argument is now empirical rather than analogical, and it should outrank the RAG analogy at E3. Read: abstract only. Leibowitz ML et al. 2021
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Depleting the KMT2A::AFF1 fusion product is sufficient to impair clonogenicity and engraftment, confirming that a hypothetical complete-disruption strategy would have the intended biological effect if it could be delivered.
Targeting MLL-AF4 with short interfering RNAs inhibits clonogenicity and engraftment of t(4;11)-positive human leukemic cells · Thomas M, Gessner A, Vornlocher HP, Hadwiger P, Greil J, Heidenreich O · Blood 2005What this evidence cannot show
RNA-level depletion in cell lines and xenografts, not genome editing; establishes target validity, not editing feasibility.
Evidence that argues against this (1)
- Disrupting the gene does not promptly remove the protein, and this graph already holds the reason. Liu et al. showed that wild-type MLL is cleared in defined cell-cycle windows by SCF(Skp2) at S phase and APC(Cdc20) at late M phase, that degradation initiates within the N-terminal approximately 1400 amino acids (exactly the portion retained in every fusion), and that the prevalent MLL fusions tested are RESISTANT to that degradation, which the authors proposed as a universal mechanism underlying MLL leukaemias. An editing strategy that ablates transcription therefore leaves behind an abnormally long-lived fusion protein, so the interval between a successful edit and loss of the oncoprotein is set by a pathologically extended half-life. Any efficacy model that equates gene disruption with protein loss, or any preclinical readout timed to normal protein turnover, will overestimate the effect. Read: abstract only. Liu H et al. 2007
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
A programmable nuclease active in the lymphoid lineage produces recurrent oncogenic rearrangement even after evolution suppressed its transposition activity by more than a thousand-fold, which sets the safety bar for any engineered nuclease in haematopoiesis.
Transposon molecular domestication and the evolution of the RAG recombinase · Zhang Y, Cheng TC, Huang G, Xu A, Xiong Y, Schatz DG, et al. · Nature 2019What this evidence cannot show
An evolutionary and structural argument used here as an analogy for engineered-nuclease safety, not as direct evidence about Cas9.
Evidence that argues against this (1)
- The analogy names the wrong residual hazard. Next-generation sequencing of rearrangements in primary murine B cells showed that the RAG recombinase's in vivo genome-destabilising activity is not transposition at all: it releases cleaved gene fragments that reintegrate into breaks generated independently elsewhere in the genome, a pathway the authors state is distinct from DNA transposition and from trans-V(D)J recombination, and the same signature appears in human lymphoma and leukaemia. Only a few putative in vivo transposition events have ever been documented. So the safety lesson for an engineered nuclease in haematopoiesis is not 'transposition was suppressed a thousand-fold and residual transposition is the risk' - it is that cut ends migrate into whatever other breaks exist, which is a property of making a break at all and is not engineered away by a cleaner enzyme. Rommel PC et al. 2017
- related to →KMT2A / MLL1 — the gene that breaks
Counter-evidence (Leibowitz 2021, PMID 33846636): Cas9 double-strand breaks initiate chromothripsis via micronuclei and chromosome bridges, shown at a clinically relevant locus in clinically relevant cells -- a two-cut, two-chromosome fusion-disruption design is the configuration most likely to trigger it, in the exact cell compartment the disease arises in
Shared counter-evidence (Liu 2007, PMID 17908926): KMT2A fusions resist the SCF(Skp2)/APC(Cdc20) degradation that clears wild-type MLL, so both DNA-level disruption and transcript-level knockdown are rate-limited by an abnormally long protein half-life; fusion protein, not transcript, is the correct pharmacodynamic endpoint for either modality
Counter-evidence (Leibowitz 2021, PMID 33846636): the 'borrowing the machinery' argument no longer needs the RAG-domestication analogy -- Cas9 editing is directly shown to initiate chromothripsis via micronuclei and chromosome bridges, including at a clinically relevant locus in clinically relevant cells
- inhibits →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
intron-targeted double cut would disrupt the fusion specifically in rearranged cells
- related to →Which repair pathway glues the fusion together: classical NHEJ vs alternative end joining
paired induced breaks are handed to the same error-prone joining that made the disease
RAG's engineered-down transposition sets the safety bar for a therapeutic nuclease in haematopoiesis
editing's correct role in this project is as a laboratory reagent, e.g. degron knock-in
borrowing the machinery: what a fusion-directed nuclease would be, and why it is hard
correcting the junction rather than drugging it
Counter-evidence (Rommel 2017, PMID 28179379): the residual hazard of a lymphoid-active programmable nuclease is not transposition but migration of cut ends into whatever other breaks exist -- a property of making a break, not of enzyme cleanliness