Experiment (proposed) — cross-family interface point-mutant panel: is the obligate interface real in each fusion?
also called interface point-mutant panel · separation-of-function fusion panel
For each fusion protein we build versions with tiny, precise changes that break just one contact point, then ask whether the cancer program still works. Knock-outs tell you a protein matters; this tells you which handshake matters, which is what a drug has to break.
The same thing, for a scientist
A single coordinated wet-lab panel that replaces knockout with separation-of-function point mutagenesis across five fusion families. For each fusion, interface point mutants (not deletions, which confound folding) are introduced into a retroviral fusion cDNA and scored in a common readout set: serial colony replating in murine lineage-negative bone-marrow or human CD34+ cells, HOXA9/MEIS1 and program-specific qPCR or RNA-seq, immunoprecipitation for complex integrity, and transplantation for the mutants that matter. The design deliberately tests candidate interfaces nominated by `ana-comparative-interface-druggability-scan` so that computational triage is validated or discarded in one pass.
The panel1 deeper
| Fusion | Interface to mutate | Prediction if the interface is obligate |
|---|---|---|
| KMT2A::AFF1 | (a) MBM1 menin-docking motif (positive control, expected obligate); (b) candidate AFF1 self-association / AFF1-AFF4 heteromerisation residues | (a) loss of transformation, validating the readout; (b) loss of HOXA9/MEIS1 with menin docking intact — a new target |
| RUNX1::RUNX1T1 | ETO NHR2 coiled-coil residues | loss of high-molecular-weight complex, restored target-gene transcription, loss of replating 1 |
| CBFB::MYH11 | CBFB surface contacting RUNX1 | loss of RUNX1 sequestration and of transformation; AI-10-49 should phenocopy 2 |
| PML::RARA | PML RBCC coiled-coil dimerisation residues | loss of priming; forced-dimer constructs are the published complement of this test 3 |
| NUP98::NSD1 | catalytically dead NSD1 SET; separately, FG-repeat IDR swap | distinguishes enzyme-dependence from condensate-dependence 4 |
| EWSR1::FLI1 | PrLD tyrosine residues (negative control for pocket character) | loss of BAF retargeting but by a diffuse mechanism, i.e. biologically obligate yet not druggable 5 |
The cross-family design is the point. Doing any one of these rows alone tells you about one disease; doing all six in one readout set tells you whether 'obligate interface' is a general property of fusion oncoproteins, which is the question the lab's whole strategy rests on.
Controls and interpretation traps
Essential controls: expression-matched wild-type fusion; a folding control (circular dichroism or thermal shift on the isolated mutated domain) to prove the point mutant did not simply unfold; an unrelated surface mutant at the same domain as a specificity control; and, for each family, the published pharmacological phenocopy where one exists (revumenib for menin, AI-10-49 for CBFB-SMMHC).
Trap 1: a point mutant that abolishes transformation may have destabilised the whole protein. Hence the folding control, which is where most published interface claims are weak.
Trap 2: murine progenitor replating overstates KMT2A::AFF1 biology, which is notoriously hard to model; the human CD34+ arm is not optional for that row.
Trap 3: absence of a phenotype may mean redundancy rather than dispensability, so negative rows must be followed with double mutants before being reported as falsifying.
| design | Retroviral expression of wild-type versus interface point-mutant fusion cDNAs in murine lineage-negative bone marrow and in human CD34+ cord-blood cells, across six fusion families, with expression-matched, folding and specificity controls plus pharmacological phenocopies where an agent exists. |
|---|---|
| readout | serial colony replating; HOXA9/MEIS1 and program-specific qPCR plus RNA-seq; co-immunoprecipitation for complex integrity; flow-cytometric differentiation markers; transplantation for mutants with a colony phenotype |
| controls | expression-matched wild-type fusion; thermal-shift or circular-dichroism folding control on each mutated domain; unrelated same-domain surface mutant as specificity control; revumenib (KMT2A row) and AI-10-49 (CBFB row) as pharmacological phenocopies; empty vector |
| cost tier | med |
| timeline | 9-15 months, staged: cloning and folding QC 2 months, colony and expression readouts 4 months, transplantation for positives 6-9 months |
| go no go | If the KMT2A MBM1 positive control does not abolish transformation in the human CD34+ arm, stop and fix the model before interpreting any other row. |
| highest value single row | the AFF1 self-association mutants in KMT2A::AFF1, because a positive result nominates a target that is on the fusion rather than on a wild-type host protein |
tests hypotheses
[ "hyp-obligate-partner-interface-generalises", "hyp-oligomerisation-interfaces-as-mutation-resistant-targets", "hyp-nup98-partner-obligate-interface" ]
collaborators needed
[ "a group with validated human CD34+ KMT2A::AFF1 transformation capability", "a structural group to nominate interface residues and run folding QC" ]
design risks added cycle 33
[ "Acceptance criterion missing: the drug phenotype must phenocopy the interface point mutant, or the panel validates off-interface mechanisms (WDR5 WIN site, PMIDs 30865883/38682900).", "Partial disruption misread as redundancy: distributed interfaces such as PML's coiled coil lose only 74-84% of self-association per single point mutant (PMID 39587079), so a surviving program after one mutation is not evidence the interface is dispensable. Combinatorial mutants are needed per row.", "Protein-stability confound: an interface point mutant that destabilises the protein scores as loss of interface when it is loss of protein. Each row needs a matched abundance control." ]
design amendments 2026 09 19
[ "Add a degradation arm to every row: genetic interface mutant, occupancy inhibitor, and targeted degradation of the partner. The three can disagree, and at WDR5 they did (PMID 34586829).", "Replace binary disrupted/not-disrupted calls with a titration of interaction strength or occupancy against program output, because at least one interface in the panel (EWSR1 PrLD) has an optimum rather than a threshold (PMID 35483357).", "For the NUP98 row, read out condensate CONSTITUENTS as well as puncta, because condensate formation can persist while transformation is lost (PMID 35073946).", "Record the treatment history of every primary sample and include a matched early/late pair from the same patient where possible, because prior therapy alone shifts menin-inhibitor sensitivity with on-target activity preserved (PMID 41027490).", "For any peptide-based arm, measure free intracellular peptide in live unfixed cells; the classical cell-penetrating-peptide uptake readout has a fixation artefact (PMID 12678851)." ]
- Interface point mutants frequently destabilise the protein, so an apparently clean loss-of-function result can be a folding artefact; without the folding QC this panel would generate confident false positives.
- KMT2A::AFF1 is notoriously hard to model in mouse progenitors, so the family the lab cares most about is the row most likely to produce an uninterpretable result.
- A negative result at an interface does not mean the interface is undruggable, only that it is redundant in the assay used, and over-reading negatives could discard a good target.
- Which residues actually form the candidate AFF1 self-association interface, given that no structure of AFF1 in the fusion context exists?
- Can human CD34+ cord-blood transformation by KMT2A::AFF1 be made robust enough to serve as the primary readout, or must the panel rely on murine progenitors with all their known distortions?
- For rows that come back negative, how many double mutants are needed before redundancy can be excluded rather than assumed?
- For each row, how many simultaneous interface substitutions are needed to fall below the threshold where the oncogenic program collapses? PML suggests one is not enough, and the panel currently assumes it is.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
Interfering with the ETO NHR2 oligomerisation domain dissolves the RUNX1/ETO high-molecular-weight complex, restores target-gene transcription and reverts the differentiation block, establishing the readout this panel would use.
Targeting the oligomerization domain of ETO interferes with RUNX1/ETO oncogenic activity in t(8;21)-positive leukemic cells · Wichmann C, Chen L, Heinrich M, et al. · Cancer Research 2007What this evidence cannot show
Used a competing peptide rather than interface point mutants, which is precisely the gap this experiment closes.
Evidence that argues against this (1)
- If the panel adopts the NHR2-peptide readout, it inherits that readout's delivery problem. Apparent energy-independent entry of TAT and other arginine-rich cell-penetrating peptides was traced to redistribution of peptide and cargo during the fixation step used for fluorescence microscopy, and to strong ionic interaction with the cell surface that overestimates cell-associated signal; uptake is endocytic and energy-dependent. A panel arm that scores 'interface disrupted' from a peptide experiment needs an independent measure of free intracellular peptide - target engagement in live, unfixed cells - or a negative result is uninterpretable as a statement about the interface rather than about delivery. Vives E et al. 2003
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
A small molecule that disrupts the CBFB-SMMHC:RUNX1 interface restores RUNX1 transcriptional activity and selectively kills inv(16) blasts, providing a pharmacological phenocopy against which the corresponding point mutant can be benchmarked.
A small-molecule inhibitor of the aberrant transcription factor CBFbeta-SMMHC delays leukemia in mice · Illendula A, Pulikkan JA, Zong H, et al. · Science 2015What this evidence cannot show
AI-10-49 availability and potency limit how cleanly the pharmacological arm can be run today.
Evidence that argues against this (1)
- The panel benchmarks a point mutant against a pharmacological phenocopy, which is the right design -- and the WDR5 WIN-site case shows why it must be an explicit ACCEPTANCE CRITERION rather than a convenience. Picomolar WIN-site inhibitors displace WDR5 from chromatin and yet act in MLL-rearranged cells through ribosome depletion and p53/MDM4. If the panel accepts 'drug kills the cells' as interface validation, it will validate interfaces whose drugs work elsewhere. Howard GC et al. 2024
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Forced RARA homodimers alone prime mice for APL-like leukaemia, which is the gain-of-function complement to the loss-of-function dimerisation mutant proposed here.
Forced retinoic acid receptor alpha homodimers prime mice for APL-like leukemia · Sternsdorf T, Phan VT, Maunakea ML, et al. · Cancer Cell 2006What this evidence cannot show
Long latency in the published model means the loss-of-function arm will need large cohorts or a sensitised background.
Evidence that argues against this (1)
- Single interface point mutants may be insufficient for distributed interfaces, which is a design risk the panel does not price. In PML, self-association runs through a long coiled coil in which each CC helix of one RBCC dimer engages a helix of a neighbouring dimer, and single point mutants in this graph's cycle-32 reading lose only 74-84% of self-interaction. A panel that scores 'interface mutant retains the program' as evidence of redundancy will misread partial disruption as dispensability. Tan Y et al. 2024
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
FG-repeat deletion abolishes and an unrelated LLPS-competent IDR restores NUP98-fusion chromatin occupancy and transformation, validating the IDR-swap design used in the NUP98 row.
Phase separation drives aberrant chromatin looping and cancer development · Ahn JH, Davis ES, Daugird TA, et al. · Nature 2021What this evidence cannot show
Published with NUP98-HOXA9; extending the design to NUP98::NSD1 adds the untested assumption that the FG-repeat requirement is partner-independent.
Evidence that argues against this (1)
- The IDR-swap design validates occupancy and phase behaviour, not causality for transformation, and the panel's readout depends on which of those it is measuring. In NUP98-NSD1/FLT3-ITD immortalised primary murine HSPC, inhibiting the condensate constituent SMARCA5 abolished the transformed phenotype while leaving the NUP98-NSD1 condensates fully intact; the authors state that condensate formation is not sufficient for maintaining transformation. A panel arm that scores an IDR swap as restoring 'the interface function' will score condensate formation, which this experiment shows can be present without the disease state. The NUP98 row therefore needs a constituent-composition readout alongside the puncta readout. Jevtic Z et al. 2022
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
EWSR1 prion-like-domain tyrosine residues required for phase transition are required for BAF retargeting by EWS-FLI1, giving a published negative control for pocket-character claims.
Cancer-specific retargeting of BAF complexes by a prion-like domain · Boulay G, Sandoval GJ, Riggi N, et al. · Cell 2017What this evidence cannot show
Tyrosine mutants conflate phase behaviour with specific contacts, so this row is a control for diffuseness rather than a clean mechanistic test.
Evidence that argues against this (1)
- This negative control assumes monotonicity and the measurement says otherwise. In patient-derived Ewing cells, EWS::FLI1 requires a narrow optimum of low-complexity-domain interaction; increasing those interactions toward putative phase separation represses its GGAA-microsatellite target genes, and driving the fusion into the nucleolus inhibits transformation. A panel that assumes 'more disruption, less program' will score this arm wrongly in both directions, and the design fix is a titration - occupancy or interaction strength varied across a range, with program output measured at each point - rather than a binary disrupted/not-disrupted call. Chong S et al. 2022
- E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Menin-KMT2A disruption produces loss of the fusion program, differentiation and disease control in KMT2A-rearranged models, giving the panel a pharmacological positive control for the KMT2A row.
A Menin-MLL inhibitor induces specific chromatin changes and eradicates disease in models of MLL-rearranged leukemia · Krivtsov AV, Evans K, Gadrey JY, et al. · Cancer Cell 2019What this evidence cannot show
Pharmacological disruption and genetic interface mutation are not equivalent; the drug also affects wild-type KMT2A-menin complexes, which a fusion point mutant does not.
Evidence that argues against this (1)
- Choosing menin-KMT2A disruption as the pharmacological positive control makes the panel's calibration depend on which cells the control is run in. Serial samples from multiply relapsed KMT2A::AFF1 patients were markedly less sensitive to menin inhibition than earlier samples from the same individuals, with the resistance acquired in the complete absence of menin-inhibitor exposure, and with the drug's on-target repression of its canonical target genes fully preserved in the resistant cells. A positive control that is on-target and non-lethal in heavily pretreated material will set the panel's dynamic range wrongly for every other arm. The panel should state the treatment history of every sample it uses and, ideally, include a matched early/late pair. Mahdavi L et al. 2025
- E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
An interface-disruption panel must require that the pharmacological arm phenocopy the genetic interface mutant, because a liganded, chromatin-displacing partner interface (the WDR5 WIN site) produces a cellular phenotype driven by ribosome depletion and p53 rather than by the interface.
Ribosome subunit attrition and activation of the p53-MDM4 axis dominate the response of MLL-rearranged cancer cells to WDR5 WIN site inhibition. · Howard GC, Wang J, Rose KL, et al. · eLife 2024What this evidence cannot show
Abstract only; cell lines, one chemical series. This is a design requirement derived from a counterexample, not an experimental result about the panel itself.
Evidence that argues against this (1)
- The requirement stated here - that the pharmacological arm phenocopy the genetic interface mutant - is right, and the WDR5 case shows it needs a third arm. Occupancy-based WDR5 inhibitors lacked in vivo efficacy while a VHL-recruiting degrader of the same protein (MS67) outperformed them at suppressing WDR5-regulated transcription, at reducing chromatin-bound MLL complex components and c-MYC, and at suppressing MLL-rearranged AML patient cells in vivo. A genetic interface mutant removes the interface; an occupancy inhibitor blocks it; a degrader removes the whole protein. The three can disagree, and a panel with only the first two will conclude that an interface is not the mechanism when the real finding is that occupancy is not the modality. Adding a degradation arm - or at minimum a degron-tagged genetic control - is the design change this row implies. Yu X et al. 2021
Design risk added (Tan 2024, PMID 39587079): PML self-association runs through a coiled coil in which each dimer's CC helix engages a neighbouring dimer's, and single point mutants lose only 74-84% of self-interaction. A panel that reads 'program survives one interface mutation' as redundancy will misread partial disruption; combinatorial mutants and matched abundance controls are needed.
- related to →NUP98::NSD1, t(5;11)(q35;p15.5)
Puncta are the wrong readout for the NUP98 arm (Jevtic 2022, PMID 35073946): SMARCA5 inhibition destroys NUP98-NSD1 transformation while the condensates persist, so the panel needs a constituent-composition readout.
- related to →EWSR1::FLI1 (EWS-FLI1), t(11;22)(q24;q12)
Binary disruption calls will misread the Ewing arm (Chong 2022, PMID 35483357): the LCD-interaction/program relationship has an optimum, so the arm needs a titration rather than a disrupted/not-disrupted call.
The positive control depends on sample treatment history (Mahdavi 2025, PMID 41027490): pretreated KMT2A::AFF1 samples resist menin inhibition with on-target repression intact, which would set the panel's dynamic range wrongly.
- related to →WDR5 — the partner that is not a partner: it binds the half of KMT2A the fusion throws away
Add a degradation arm (Yu 2021, PMID 34586829): genetic interface mutant, occupancy inhibitor and degrader can disagree, and the WDR5 case is one where occupancy failed and degradation worked.
same cross-family interface question, different reagent
the panel should include a forward resistance screen arm; TK216 is why