Oncofusion

start / graph / mechanism-rag-off-target-recombination

Mechanismstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q1-fusion-genesis

RAG off-target activity: the dominant mutator in ETV6::RUNX1 ALL, and probably not the cause of KMT2A fusions

also called RAG-mediated recombination · illegitimate V(D)J recombination · cryptic RSS

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.

The same thing, for a scientist

RAG1/RAG2 introduce programmed DSBs at recombination signal sequences during V(D)J recombination. Off-target RAG cleavage at cryptic RSS motifs is the predominant mutational process in ETV6::RUNX1 ALL: exome and low-coverage WGS of ETV6-RUNX1 cases showed RAG-mediated deletions dominating, with RSS motifs near breakpoints, non-templated sequence at junctions, ~30-fold enrichment at promoters and enhancers of genes actively transcribed in B-cell development, and an unexpectedly high recurrent-to-non-recurrent SV ratio. By contrast, sequencing of reciprocal breakpoints in 14 primary t(4;11) leukaemias found duplications, deletions and inversions of up to a few hundred nucleotides, the signature of error-prone DSB repair, and the authors explicitly concluded the events were NOT produced by immune-system recombinases. This distinction matters operationally: it means the ETV6-RUNX1 prevention/detection logic does not transfer to KMT2A-r infant ALL. Two caveats belong with that conclusion: the KMT2A negative rests on a single n=14 Sanger-era study that has never been repeated at cohort scale, and no published study has ever argued the positive case at KMT2A - so the exclusion is supported by an absence of contrary work rather than by a contest the RAG hypothesis lost.

Go deeper
What RAG off-targeting looks like when it happens1 deeper

In ETV6-RUNX1 ALL the secondary lesions carry an unmistakable fingerprint: RSS-like motifs at the breakpoints, non-templated nucleotide addition at junctions (the work of TdT, another V(D)J enzyme), and roughly 30-fold enrichment at promoters and enhancers of genes actively transcribed during B-cell development 1. The process runs throughout leukaemic evolution and even shows localised clustering and reiterated deletions. Two new tumour suppressors, ATF7IP and MGA, were identified this way 1.

Why an active promoter is a RAG target

RAG is recruited to open, H3K4me3-marked chromatin. Active promoters in the B-lineage programme are therefore exactly where a cryptic RSS gets found. This is the same general principle, transcription equals accessibility equals fragility, that also underlies the KMT2A loop-anchor model, but with a different nuclease 3.

The KMT2A junctions say something different2 deeper

Marschalek's group sequenced reciprocal chromosomal breakpoints from primary biopsy material of fourteen t(4;11) patients. Every case showed duplications, deletions and inversions of fewer than a few hundred nucleotides, indicative of malfunctioning DNA repair, and the conclusion was explicit: these translocations were initiated by several DNA strand breaks on both chromosomes followed by error-prone repair, and not by the action of immune-system recombinases 2. There is no RSS enrichment and no TdT-style N-region signature at KMT2A junctions.

The timing argument reinforces it

KMT2A::AFF1 infant ALL arises in a fetal progenitor with a pro-B or even CD10-negative CD19+ CD34+ phenotype, in the first two trimesters. RAG activity is a feature of committed B-lineage progenitors undergoing receptor assembly. The cell of origin and the timing are not a natural fit for a RAG-driven mechanism.

What genuinely argues the other way

This entry states a negative, so honest counter-evidence means published reasons to keep the door ajar. Three exist, and none of them is strong.

  1. RSS-like motifs at an MLL breakpoint region have been reported once. The AFX intron that carries the t(X;11) breakpoint contains several immunoglobulin heptamer-like sequences (GATAGTG) throughout its 3.7 kb [E2 counter, Borkhardt 1997]. That is a sequence observation in a rare partner, not evidence of RAG cleavage, and short heptamer-like motifs are common by chance.
  2. The negative is small and old. Fourteen patients, Sanger sequencing, one laboratory, 1999, and never repeated at cohort scale 2. Absence of an RSS signature in 14 junctions does not exclude a minority mechanism, and breakpoint distribution at this locus is known to be age-heterogeneous [E2 counter, Reichel 2001].
  3. Recombination machinery has been linked to infant B-ALL by a different route. Rare PRDM9 alleles are enriched in parents of children with B-ALL and, on replication, in aneuploid and infant B-ALL cases [E1 counter, Hussin 2013]. PRDM9 is meiotic, not RAG, so this is not the same claim - but it is a live proposal that recombination-machinery variation contributes to the rearrangements of infant leukaemia.

What does not exist, and this is itself a finding: after a targeted literature search (PubMed title/abstract and Europe PMC full text, September 2026) I could find no published study claiming that RAG causes KMT2A fusions. The RAG hypothesis for KMT2A is not a contested position in the literature; it is an untested one that the field has quietly dropped. The entry's verdict is therefore safe but under-evidenced rather than well-supported.

RAG can still cleave non-immune loci in principle

It would be wrong to say RAG only ever touches antigen-receptor genes. RAG can nick the BCL2 minor breakpoint cluster region in a nonamer-independent, CCACCTCT-motif-dependent manner, and that region can synapse with a canonical RSS in a RAG-dependent way in cells 4. So RAG-mediated fragility at a non-immune locus is a real phenomenon. It has simply not been demonstrated at KMT2A, and the junction evidence argues against it.

The RAG-negative rests on two studies, not one - and both are from the same group

This entry says the KMT2A RAG-negative "rests on a single n=14 Sanger-era study". There are two: PMID 10467413 and PMID 9881706 (Reichel et al. 1998), which sequenced both derivative chromosomes in 3 cell lines and 2 patients and concluded there was "no evidence for the involvement of specific target sequences and recombinases of the immune system".

Two studies is better than one, and they are both from the Marschalek group - the group this graph already leans on hardest, supplying 33 of its 1,731 cited papers. Under the source-independence measure, two papers from one laboratory are one group's result, so the correction upgrades the evidence less than the count suggests.

What makes the negative interesting has changed anyway. PMID 38650597 shows V(D)J recombination is demonstrably running in KMT2A::AFF1 blasts - proximally biased IGHV usage, IGHV6-1 significantly elevated across 48 adult, 36 paediatric and 37 infant cases, and more abundant de novo DJH recombination than in other BCP-ALL subtypes. So the comfortable reading of the negative - that the cell of origin is simply pre-RAG - is not available. RAG is active in these cells and the breakpoints still carry no RSS signature.

Facts and figures
mechanism classprogrammed recombinase off-target cleavage
verdict for kmt2aDisfavoured on junction-sequence evidence (Gillert 1999, n=14): KMT2A junctions show error-prone-repair indels, not RAG signatures. The evidence base is one small Sanger-era study; no paper has argued the positive case, so this is an untested rather than a settled question.
unopposed noteTargeted search (PubMed tiab and Europe PMC full text, 2026-09-19) found no published claim that RAG causes KMT2A fusions. The closest published positives are RSS-like heptamers reported in the AFX intron at an MLL translocation breakpoint (Borkhardt 1997, PMID 9010221) and rare PRDM9 alleles enriched in infant B-ALL (Hussin 2013, PMID 23222848).
effectors
[
  "RAG1",
  "RAG2",
  "DNTT (TdT)"
]
signature features
[
  "cryptic RSS heptamer/nonamer motifs",
  "non-templated N nucleotides at junctions",
  "enrichment at active promoters/enhancers"
]
applies to
[
  "fusion-etv6-runx1 secondary lesions",
  "some IGH/TCR translocations",
  "BCL2 mcr in follicular lymphoma"
]
does not apply to
[
  "fusion-kmt2a-aff1"
]
How this could bite us back
  • If RAG is not the initiator, then RAG inhibition, which is the only lymphocyte-specific nuclease intervention anyone has proposed, is useless for KMT2A-r prevention. This closes off the most lineage-selective preventive option.
Still open
  1. Has anyone re-tested the RAG hypothesis at KMT2A with modern breakpoint sequencing across the 3401-patient Meyer/Marschalek cohort, searching systematically for cryptic RSS motifs? A negative result at that scale would close the question permanently.
  2. In the rare KMT2A-r cases with a pro-B phenotype and productive IGH rearrangement, do junctions differ from those in CD10-negative fetal-type cases?
  3. PRDM9 does not appear anywhere in this graph, yet rare PRDM9 alleles have been reported as enriched in infant and aneuploid B-ALL (Hussin 2013). Is PRDM9 allelic status known for any KMT2A-r infant cohort? This would be a cheap germline lookup in existing trio data and nobody appears to have done it for KMT2A-r specifically.
Every source behind this entry (5)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    RAG-mediated recombination is the predominant mutational process generating secondary lesions in ETV6-RUNX1 ALL, with RSS motifs at breakpoints, non-templated junctional sequence and ~30-fold enrichment at active B-development promoters and enhancers.

    PMID 24413735DOI
    What this evidence cannot show

    Applies to the secondary copy-number lesions in ETV6-RUNX1 ALL, not to formation of the ETV6::RUNX1 fusion itself, which arises in utero and is not RAG-signature. Exome plus low-coverage WGS limits detection of balanced events.

    Evidence that argues against this (1)
    • A germline determinant of RECOMBINATION machinery has been associated with infant B-ALL, which the entry's error-prone-repair-only account does not accommodate. Hussin et al. found a substantial and significant excess of rare allelic forms of PRDM9 - the meiotic recombination hotspot-specifying methyltransferase - among parents of children with B-ALL (44 parental exomes), replicated the excess in an independent cohort of 50 children with B-ALL, and reported that the excess in the replication cohort was found in ANEUPLOID AND INFANT B-ALL patients specifically. PRDM9 is not RAG and this is an association study with small numbers and no breakpoint-level mechanism, so it does not resurrect the RAG hypothesis. It does show that recombination-machinery variation is on the table as a contributor to infant leukaemia genomic rearrangement, and this graph contains no entry for it. Hussin J et al. 2013
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Reciprocal t(4;11) breakpoints in 14 primary patient samples show duplications, deletions and inversions of under a few hundred nucleotides consistent with error-prone DNA repair and explicitly not with immune-system recombinase activity.

    PMID 10467413DOI
    What this evidence cannot show

    n=14, Sanger-era sequencing, and absence of an RSS signature is an argument from absence. It remains the best direct junction-level test of the RAG hypothesis at KMT2A and has not been contradicted in the 25 years since.

    Evidence that argues against this (2)
    • RSS-like motifs HAVE been reported at an MLL-translocation breakpoint region, on the partner side. Characterising the AFX gene that fuses to MLL in t(X;11)(q13;q23), Borkhardt et al. sequenced the single 3706-nucleotide AFX intron that carries the breakpoint and reported that it contains a chi-like octamer and SEVERAL IMMUNOGLOBULIN HEPTAMER-LIKE SEQUENCES (GATAGTG) distributed throughout the intron, with the KARPAS 45 breakpoint falling at nucleotides 2913/2914 of that intron. This is a sequence observation in one rare partner and one cell line, not functional evidence of RAG cleavage, and heptamer-like 7-mers occur often by chance in any few-kilobase intron. But it is a published report of RSS-like motifs at an MLL translocation breakpoint region, and the entry's claim that there is 'no RSS enrichment' should be read as 'no systematic RSS enrichment has been demonstrated', which is a weaker statement. Borkhardt A et al. 1997
    • The argument from absence rests on one study, one laboratory, one assay generation and n=14, and has never been re-tested at cohort scale. Gillert et al. is Sanger-era junction sequencing of 14 t(4;11) patients; a negative for cryptic RSS motifs at that sample size cannot exclude a minority mechanism, and the same group's subsequent work on the same cohort lineage (Reichel et al., 40 t(4;11) patients) did not revisit the recombinase question but did find that breakpoint distribution differs significantly between infants and older children, implying more than one age-linked mechanism operates at this locus. A mechanism that is heterogeneous by age is exactly the situation in which a 14-patient negative is least informative. Reichel M et al. 2001
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.review

    Programmed DSBs in lymphocytes are targeted by chromatin accessibility and transcription, and 3D genome organisation both directs and suppresses their translocation.

    Mechanisms of programmed DNA lesions and genomic instability in the immune system · Alt FW, Zhang Y, Meng FL, Guo C, Schwer B · Cell 2013
    PMID 23374339DOI
    What this evidence cannot show

    Review; predominantly mouse data.

    Evidence that argues against this (1)
    • Accessibility and transcription set where programmed breaks occur; they do not set where those breaks go, and the measured ordering is the reverse of the emphasis here. In G1-arrested mouse pro-B cells mapped by Hi-C and genome-wide translocation sequencing, RAG-cleaved antigen-receptor loci were the dominant partners for target double-strand breaks REGARDLESS of genomic position — a frequency effect, not a proximity effect — and proximity-dependent enrichment only emerged once breaks were equalised genome-wide by ionising radiation. '3D organisation both directs and suppresses translocation' is true but subordinate: at a locus that RAG cuts often, break frequency dominates partner choice. Zhang Y et al. 2012
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    RAG can nick a non-immune locus (the BCL2 minor breakpoint cluster region) via a nonamer-independent, CCACCTCT-motif-dependent mechanism and synapse it with a canonical RSS in cells.

    PMID 22275374DOI
    What this evidence cannot show

    In vitro RAG nicking plus a cell-based recombination assay; demonstrates possibility at BCL2, not at KMT2A. Requires Mn2+ in addition to Mg2+ for the triple-nick behaviour, which is not physiological.

    Evidence that argues against this (1)
    • Demonstrating nonamer-independent RAG nicking at BCL2 shows the enzyme CAN work off-target; the sequence evidence at prenatal fusion breakpoints argues it did not do so there. Long-distance inverse PCR of genomic fusion junctions from nine patients with the other paradigm in-utero-initiated fusion found breakpoints scattered across 14 kb of intron with NHEJ signatures, and DNA sequences near the breakpoints revealed 'no consistent characteristic signal sequences of the V(D)J recombinase, topoisomerase II consensus sites, or other sequence motifs associated with recombination'. If a RAG-mediated mechanism operated at these prenatal B-lineage fusions, the cryptic-signal footprint that BCL2 supplies is absent from the junctions where it would matter most. Wiemels JL et al. 1999
  5. E-second-studymoderate confidence 3 of 5: One strong study, or several smaller consistent ones.journal-article

    The KMT2A RAG-negative rests on two sequencing studies rather than one - PMIDs 10467413 and 9881706 - both from the same research group, which under a group-level independence measure counts as one source.

    PMID 9881706
    What this evidence cannot show

    Read at the level of title and reported conclusion. Both studies are Sanger-era and small; 9881706 covers 3 cell lines and 2 patients. Neither used the high-throughput junction mapping that would be required to exclude a minority RSS-driven mechanism.

    Evidence that argues against this (1)
    • Counting this as a strengthening would be a mistake of exactly the kind the lab-clustering analysis was built to catch: **two papers from one laboratory are one group's opinion**, and this graph's largest single group is that laboratory. The negative is better supported than 'a single study' implies and less well supported than 'two independent studies' would. Oncofusion knowledge graph (internal analysis) 2026
Connected to