Oncofusion

start / graph / exp-surface-proteomics-of-an-isogenic-human-kmt2a-aff1-model

Experimentmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q138-kmt2a-aff1-surfaceome

Cell-surface capture proteomics on an isogenic human KMT2A::AFF1 model against its matched normal progenitor - the experiment an mRNA screen cannot replace

also called the isogenic surface proteomics experiment

Take human blood stem cells, put the fusion gene into half of them, and then chemically label and identify only the proteins sitting on the outside of each cell. Comparing the two lists gives the surface proteins the fusion itself creates, measured as protein rather than inferred from gene activity.

The same thing, for a scientist

This beat's screen, and the published surfaceome screen before it, both infer surface localisation from annotation and abundance from mRNA. Neither measures protein on a membrane. The experiment that does is **cell-surface biotinylation or enzymatic cell-surface capture followed by mass spectrometry**, run on an isogenic pair: human cord-blood CD34+ HSPCs transduced with KMT2A::AFF1 versus empty-vector controls from the same donor. **The design template already exists and worked for another fusion.** CBFA2T3::GLIS2 expressed in human cord-blood HSPCs on endothelial coculture recapitulated the transcriptome, morphology and immunophenotype of the human disease, and a fusion-specific gene library from that model yielded FOLR1 and a CAR-T against it [E1]. The specific lesson to carry over is that **the endothelial coculture was necessary to recapitulate immunophenotype**, so a KMT2A::AFF1 surface screen run without a niche would produce a list that is partly an artefact of culture.

Go deeper
What must be measured alongside, or the result repeats this beat's problem

Every hit must be carried immediately to a normal-tissue protein panel, not an mRNA one - immunohistochemistry on a paediatric normal-tissue array including brain, kidney, lung and vasculature. Both published KMT2A-r antigens failed a normal-tissue mRNA filter that their discovery papers did not apply 23, and a third antigen discovered without that filter would repeat the cycle.

And because this beat's own matched-permutation test showed that fusion-driven antigens are no cleaner and no dirtier than expression-matched controls 4, the expected yield of leukaemia-restricted surface proteins is low. The experiment is worth running anyway, because it measures a quantity nothing else measures - surface protein - and because the alternative is continuing to argue about it from annotation.

Facts and figures
designIsogenic pair: human cord-blood CD34+ HSPCs from a single donor, transduced with KMT2A::AFF1 versus empty vector, maintained on endothelial coculture (the niche condition shown necessary to recapitulate immunophenotype in the CBFA2T3::GLIS2 template). Cell-surface capture by hydrazide-based glycoprotein labelling or surface biotinylation, followed by LC-MS/MS. At least 3 donors.
readoutProteins enriched on the fusion-expressing surface proteome at >= 2-fold with FDR < 0.05, ranked by absolute surface abundance.
controlsEmpty-vector same-donor HSPCs (the isogenic control); an unrelated fusion (KMT2A::MLLT3) to separate KMT2A-rearrangement effects from AFF1-specific ones; a no-niche arm to quantify how much of the antigen list is a culture artefact; CD19 and NG2 as expected-positive surface proteins that must be recovered.
cost tierhigh
timeline12-18 months including model establishment
must be paired withImmediate immunohistochemistry of every hit on a PAEDIATRIC normal-tissue array including brain, kidney, lung and vasculature - protein, not mRNA. Both published KMT2A-r antigens failed a normal-tissue filter their discovery papers did not apply.
expected yieldLow. The matched-permutation test predicts fusion-driven antigens are ordinary members of a population in which 98.2% of leukaemia-expressed surface proteins fail a normal-tissue filter.
How this could bite us back
  • A surface-proteomic screen will find surface proteins, and the base rate says almost all of them will be present in some normal tissue. Running it without the paired normal-tissue protein panel produces a third antigen with the same problem as the first two.
  • The isogenic comparison is against a NORMAL progenitor, which answers 'what does the fusion add' and not 'what distinguishes this leukaemia from other leukaemias' - the second is the question a therapy needs answered and needs a different control arm.
  • An experiment justified by a template that produced FOLR1 should note that FOLR1's own normal-tissue profile is among the worst in this beat's entire screen.
Still open
  1. Does a KMT2A::AFF1 human HSPC model require an endothelial niche as the CBFA2T3::GLIS2 model did, and if so does the niche contribute NG2-positive cells that would contaminate a surface-capture preparation?
  2. Should the comparison be against the matched normal progenitor, or against a KMT2A::MLLT3 model? The first gives fusion-versus-normal, the second gives AFF1-versus-KMT2A-rearranged, and only the second addresses the partner question this project keeps hitting.
  3. Is there an existing deposited surface-proteomic dataset on any KMT2A-rearranged line? None was located on this beat and a targeted search of PRIDE was not run.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Expressing CBFA2T3::GLIS2 in human cord-blood HSPCs on endothelial coculture recapitulated the transcriptome, morphology and immunophenotype of the human disease and produced aggressive leukaemia in xenografts; interrogating the model's transcriptome against primary disease produced a fusion-specific gene library, from which FOLR1 was developed as a CAR-T target.

    PMID 36136600
    What this evidence cannot show

    Transcriptome, not surface proteome - so even this template did not measure surface protein directly. The niche requirement means the antigen list depends on the culture system.

    Evidence that argues against this (1)
    • FOLR1 itself is present at 2,860.9 nTPM in choroid plexus and 142.6 in kidney, so the template produced an antigen that fails the filter this experiment is designed to apply early. Human Protein Atlas 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    NG2/CSPG4, a direct MLL-AF4 target gene, reaches 124.5 nTPM in normal blood vessel and 246.21 TPM in GTEx tibial artery.

    PMID 39093982DOI
    What this evidence cannot show

    Fusion-target mechanism from the abstract; normal-tissue values from bulk adult mRNA in two atlases.

    Evidence that argues against this (1)
    • Pericytes are abluminal and may be inaccessible to a circulating therapeutic, so the mRNA objection may not translate into toxicity. Human Protein Atlas 2026
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    LAMP5, the output of the published MLL-rearranged surfaceome screen and a direct MLL-fusion target, reaches 297.8 nTPM in normal basal ganglia.

    PMID 33910331DOI
    What this evidence cannot show

    Partner-agnostic; abstract read in full.

    Evidence that argues against this (1)
    • LAMP5 is substantially an intracellular-vesicle protein by UniProt annotation, so its neural mRNA may not correspond to surface epitope - which is exactly the uncertainty this experiment's method resolves. Haematologica 2022
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    An expression-matched permutation test found fusion-enriched surface antigens no more normal-tissue-exposed than matched controls (ratio 0.911, p = 0.789), so the expected yield of leukaemia-restricted surface proteins from any such screen is low.

    What this evidence cannot show

    n = 14, mRNA-based. A null at this sample size excludes a 2-fold effect, not a small one.

    Evidence that argues against this (1)
    • A surface-proteomic screen measures a different quantity from an mRNA screen, so an mRNA-derived base rate may not predict its yield at all. Oncofusion Knowledge Graph 2026
Connected to