A pocketless fusion already had its protein-protein-interface drug taken to phase I/II - and TK216 turned out to be a microtubule destabiliser, not an EWS::FLI1 antagonist. This is the closest thing the field has to a controlled failure of the interface hypothesis.
also called TK216 · YK-4-279 · ONCT-216 · the Ewing interface programme · EWS-FLI1 / RHA disruptor
Ewing sarcoma is driven by a fusion protein with no drug-shaped pocket, exactly like the leukaemia fusion this project studies, and the field's answer was the same: find a small molecule that breaks the fusion's grip on a partner protein. That molecule was taken all the way into patients over fifteen years - and then a genetic screen showed it actually kills cells by poisoning microtubules, and the trial showed almost no activity.
The same thing, for a scientist
EWS::FLI1 and KMT2A::AFF1 pose the same problem - a transcriptional neomorph with no enzyme active site and no ligand pocket - and the Ewing field reached the same solution this project is pursuing: **disrupt an obligate protein-protein interaction the fusion needs.** Erkizan et al. screened by surface plasmon resonance for compounds binding EWS-FLI1, and YK-4-279 blocked the EWS-FLI1 / RNA helicase A interaction, induced apoptosis in Ewing cells and reduced orthotopic xenograft growth; the paper is explicitly framed as **proof of principle that inhibiting a fusion's interaction with a normal cellular partner is a general strategy for tumour-specific agents** [E1]. Its clinical derivative TK216 entered a first-in-human phase I/II trial. **Two things then happened.** First, Povedano et al. ran a forward-genetics screen using engineered hypermutation in Ewing lines and recovered recurrent **TUBA1B (alpha-tubulin) mutations sufficient to confer TK216 resistance**, then showed by reconstituted microtubule polymerisation and cell-based probe competition that TK216 is a **microtubule-destabilising agent** - which also explains its synergy with vincristine and its activity in EWSR1-FLI1-negative models, and the authors called for re-examination of the ongoing trials [E2]. Second, the trial read out: 85 patients, RP2D 200 mg/m2 once daily as a 14-day infusion, **two complete responses, one partial response and 14 stable disease in cohorts 9 and 10, 6-month PFS 11.9%, and no responses at all among the eight patients in cohort 11** [E3]. **The lesson is not that interface disruption is impossible.** It is that the Ewing programme had a plausible binding assay, a plausible cellular phenotype, in vivo efficacy and fifteen years of momentum, and never had the one control that would have caught the error: a demonstration that the drug's phenotype is abolished by an interface point mutation and by nothing else.
The mapping, and the place where the analogy protects the menin programme rather than indicting it
| EWS::FLI1 / RHA | KMT2A fusion / menin |
|---|---|
| pocketless transcriptional neomorph | pocketless transcriptional neomorph |
| the fusion binds a wild-type partner (RNA helicase A) that its oncogenic function requires | the fusion binds a wild-type partner (menin) that its chromatin residence requires |
| lead found by biophysical binding screen against the fusion 1 | lead found by structure-guided design against a defined pocket on the partner, with co-crystal structures |
| cellular phenotype: apoptosis, xenograft growth inhibition 1 | cellular phenotype: differentiation, HOXA9/MEIS1 loss, xenograft efficacy |
| genetic validation of the drug-target link: absent until a resistance screen was run, and it returned tubulin 2 | genetic validation present: patients relapsing on revumenib acquire MEN1 missense mutations at the drug-binding surface, and the same residues are the drug contact |
| clinical result: 6-month PFS 11.9%, no responses in cohort 11 3 | clinical result: 30% CR/CRh in relapsed/refractory disease |
Read the fifth row carefully, because it is the whole point. The menin programme has the control the Ewing programme lacked, and it has it in the strongest possible form - an acquired human resistance mutation at the drug's own binding site. That is a genetic proof that the clinical effect runs through the intended interface, and TK216 never had anything like it. The analogy is therefore a warning about method, not a prediction about revumenib.
What it does indict is the rest of this graph's interface-nomination pipeline: ana-comparative-interface-druggability-scan scores pocket character and chemical matter, and hyp-obligate-partner-interface-generalises already records that seven partner-side candidates nominated inside this project were struck on re-examination. TK216 is what it looks like when a nomination is not struck and proceeds anyway.
The specific control that would have caught it, stated as a rule
Povedano et al. did not find the error by chemistry. They found it by genetics: engineered hypermutation, selection under drug, and sequencing of what came back 2. What came back was alpha-tubulin, which has nothing to do with EWS::FLI1.
The transferable requirement is: a drug that claims to work by disrupting interface X must have its phenotype abolished by a mutation at X and by nothing else. Two directions, both needed.
- Forward (resistance). Mutagenise and select under drug. If the recovered mutations are in a protein that is not either partner, the mechanism is not the interface. This is cheap, unbiased and was decisive here.
- Reverse (phenocopy). An interface point mutant, not a knockout, must reproduce the drug's phenotype.
hyp-obligate-partner-interface-generalisesalready names this as the minimum sufficient test, citing the WDR5 WIN-site case where occupancy worked chemically and the phenotype ran through ribosome depletion and p53 instead.
The Ewing case adds a third observation that is easy to miss: TK216 was active in models that do not express EWSR1-FLI1 2, and that fact was visible before the trial. An on-mechanism fusion-interface drug should be inactive where the fusion is absent, and activity in fusion-negative models is a mechanism alarm, not a bonus indication.
What the trial adds that the mechanism paper does not
It would be possible to read Povedano et al. as a mechanistic curiosity that leaves the clinical question open. The trial closes it. Across 85 patients with a median of three prior systemic regimens, at the selected RP2D, the yield was two complete responses, one partial response and 14 stable disease across cohorts 9 and 10, with 6-month progression-free survival of 11.9%, and zero responses among the eight patients in cohort 11 3. The authors' own conclusion is 'limited activity at the RP2D'.
Two details matter for this project. First, vincristine was permitted from cycle 2 onward at investigator discretion 3 - and vincristine is a microtubule agent, which is precisely the mechanism Povedano et al. assign to TK216 2. A combination of two drugs with the same mechanism, described as synergy in the preclinical package, is a coherent story only under the tubulin model. Second, the responses that did occur are not evidence for the interface model; under the tubulin model they are what a microtubule agent does in a sarcoma.
The honest summary is that the field spent fifteen years and 85 patients on a mechanism that a resistance screen would have refuted in months.
What this graph already knew, and the one thing it did not
fusion-ewsr1-fli1 already records the TK216 programme and its negative record, sourced from registry status: NCT02657005 terminated, NCT05046314 completed without an approval, NCT03752138 withdrawn, with the accurate comment that this is the single most direct test in oncology of drugging a disordered fusion's protein interaction and that it produced no medicine. That entry is right and this one does not correct it.
What it does not contain is why, and the why is the entire transferable lesson. A terminated trial is compatible with many explanations - wrong dose, wrong population, wrong schedule, a real but insufficient effect. The deconvolution 2 removes all of them and leaves one: the compound was never acting on the interface. Neither Povedano et al. nor Erkizan et al. is cited anywhere in this corpus, and a search for TUBA1B across every node file returns nothing outside this entry.
The practical difference is what a reader takes away. From a registry status, the lesson is 'interface drugs are hard'. From the deconvolution, the lesson is a specific, runnable control - mutagenise, select, sequence, and require the answer to be one of your two partners - and a specific alarm - activity where the fusion is absent. Only the second form changes what this project does next.
| domain | engineering |
|---|---|
| where the analogy breaks | The menin programme possesses the control the Ewing programme lacked: acquired MEN1 missense mutations at the drug-contact surface in patients relapsing on revumenib, which is a human genetic proof that the clinical effect runs through the intended interface. TK216 never had a resistance mutation in either partner - the one it did have was in tubulin. The analogy therefore does NOT predict that menin inhibitors will fail; it predicts that the NEXT interface this project nominates will fail in this way unless the genetic controls are run first. |
| what it predicts here | Any new partner-interface candidate emerging from this project's interface scans must clear two tests before a chemistry programme: a forward resistance screen whose recovered mutations lie in one of the two partners, and an interface point mutant that phenocopies the drug. Activity in fusion-negative models should be treated as a mechanism alarm rather than as an expanded indication. |
| modality | small-molecule protein-protein interface disruptor, pocketless fusion |
| clinical trial | NCT02657005 (TK216, first-in-human phase I/II, Ewing sarcoma) as identified in the source publication's trial description |
| relationship to existing entries | CORRECTED DURING WRITING. fusion-ewsr1-fli1 ALREADY records the TK216 programme and its negative clinical record, sourced from ClinicalTrials.gov registry status: NCT02657005 terminated, NCT05046314 completed without approval, NCT03752138 withdrawn, with the comment that this is 'the single most direct test of drugging the disordered fusion's protein interaction in oncology, and it has not produced a medicine'. What that entry does NOT have is WHY, and why is the whole lesson: neither of the two primary papers is cited anywhere in this corpus. This entry adds (a) the target deconvolution - TUBA1B resistance mutations and direct microtubule destabilisation [E2], (b) the published efficacy numbers rather than a registry status [E3], and (c) the methodological rule and its contrast with the MEN1 resistance-mutation control [E4]. A registry status says a programme stopped; it does not say the mechanism was never the mechanism. |
| corpus absence checked | A first grep for TK216 was run with -E and a BACKSLASH-escaped pipe, which matches a literal pipe character and returned a false zero; the entry originally asserted on that basis that the corpus had no TK216 content. Re-run correctly on 2026-09-23, 'grep -rlEi "YK-4-279|TK216|TK-216" data/graph/nodes' returns fusion-ewsr1-fli1. TUBA1B returns nothing but this entry, and Povedano and Erkizan are cited nowhere else, so the deconvolution and the two primary papers are genuinely new; the programme's existence and failure are not. |
mapping
[
{
"bio": "EWS::FLI1 is a pocketless neomorph that requires a wild-type partner (RNA helicase A) for oncogenic function",
"analog": "the KMT2A fusion is a pocketless neomorph that requires a wild-type partner (menin) for chromatin residence"
},
{
"bio": "YK-4-279 found by biophysical binding screen; blocks the EWS-FLI1/RHA interaction; apoptosis and xenograft efficacy",
"analog": "revumenib/ziftomenib found by structure-guided design against the menin pocket; differentiation and xenograft efficacy"
},
{
"bio": "no genetic validation that the phenotype runs through the interface; a resistance screen returned TUBA1B",
"analog": "GENETIC VALIDATION PRESENT - acquired MEN1 missense mutations at the drug-binding surface in relapsing patients"
},
{
"bio": "active in EWSR1-FLI1-negative models, read at the time as broad activity rather than as a mechanism alarm",
"analog": "menin inhibitors are active in NPM1-mutant AML without a KMT2A rearrangement - but there the shared dependency on the menin-MLL1 interface is itself established, so it is not the same signal"
},
{
"bio": "clinical outcome: 85 patients, 6-month PFS 11.9%, no responses in cohort 11",
"analog": "clinical outcome: 30% CR/CRh in relapsed/refractory KMT2A-rearranged or NPM1-mutant leukaemia"
}
]generated hypotheses
[]
- This is the strongest published counter-example to the lab's central hypothesis as a GENERAL programme: a pocketless fusion, an obligate partner, a rationally identified interface disruptor, fifteen years, and the compound was a tubulin poison. Any grant or paper arguing that interface disruption is the answer for pocketless fusions has to cite this case or it is not arguing in good faith.
- The failure mode is silent and flattering. Binding assays worked, the phenotype was real, the xenografts responded, and none of it was the interface. Nothing in the preclinical package looked wrong.
- Activity in models lacking the fusion was visible for years and read as good news. A programme that treats breadth of activity as an asset has disabled its own best mechanism control.
- The combination that looked synergistic - TK216 plus vincristine - is two microtubule agents. A synergy that is explained by the wrong mechanism is worse than no synergy, because it makes the wrong mechanism look predictive.
- Delivery: TK216 required a 14-day continuous intravenous infusion at RP2D. For an infant leukaemia population, a mechanism that only works under continuous infusion is a delivery problem before it is a biology problem, and this project should ask what exposure its own interface candidates would require.
- The base-editor screen behind the MEN1 resistance mutations [E4] scanned MEN1 and found MEN1 - which is unbiased within the gene but cannot surface an off-target mechanism. Has a GENOME-WIDE mutagenesis or engineered-hypermutation screen of the kind that unmasked TK216 [E2] ever been run on revumenib? That is the one experiment whose result is not already known, and it is the only one that could return a tubulin.
- Are menin inhibitors active in cell lines with no KMT2A rearrangement and no NPM1 mutation, at the concentrations used? Under the TK216 lesson that is the single cheapest mechanism alarm available, and a clean answer either strengthens the interface claim or opens the same question TK216 failed.
- Does any candidate on this project's interface-druggability scan currently rest on binding plus phenotype with no genetic link between them? That is the TK216 signature, and the scan as designed scores pocket character rather than causal validation.
- What would a phenocopy test look like for the menin interface specifically - a MEN1 interface point mutant knocked in to an otherwise wild-type KMT2A-rearranged line, with the drug required to add nothing on top? This is the reverse control and it does not appear to have been published even though the mutations are known from patients.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
YK-4-279, identified by surface plasmon resonance screening for compounds binding the disordered EWS-FLI1 fusion, blocks RNA helicase A binding to EWS-FLI1, induces apoptosis in Ewing sarcoma family tumour cells and reduces growth of orthotopic Ewing xenografts; the authors present this as proof of principle that inhibiting the interaction of a mutant cancer-specific transcription factor with the normal cellular partner required for its oncogenic activity is a promising general strategy.
A small molecule blocking oncogenic protein EWS-FLI1 interaction with RNA helicase A inhibits growth of Ewing's sarcoma. · Erkizan HV, Kong Y, Merchant M, Schlottmann S, Barber-Rotenberg JS, Yuan L, Abaan OD, Chou TH, Dakshanamurthy S, Brown ML, Uren A, Toretsky JA · Nat Med 2009What this evidence cannot show
Abstract read in full from the PubMed record on 2026-09-23. The proof-of-principle framing is the authors' own and is the exact framing this project uses for the menin interface, which is why it is quoted. The paper establishes binding and phenotype; it does not establish that the phenotype is caused by the binding, and that gap is what E2 later fills in the wrong direction.
Evidence that argues against this (1)
- Povedano et al. show the cytotoxicity of the clinical derivative is mediated by microtubule destabilisation and that TK216 is active against cell lines and xenografts that do not express EWSR1-FLI1, which means the cellular and in vivo phenotypes reported for this chemical series cannot be attributed to the EWS-FLI1/RHA interaction without a genetic control that was never run. Povedano JM et al. 2022
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
A forward-genetics screen using engineered hypermutation in Ewing sarcoma cell lines identified recurrent mutations in TUBA1B, encoding alpha-tubulin, that are sufficient to drive resistance to TK216; reconstituted microtubule polymerisation in vitro and cell-based chemical probe competition assays show TK216 acts as a microtubule destabilising agent; this explains TK216's activity against cancer models that do not express EWSR1-FLI1 and its synergy with vincristine, and the authors call for re-examination of ongoing clinical trials with TK216.
TK216 targets microtubules in Ewing sarcoma cells. · Povedano JM, Li V, Lake KE, Bai X, Rallabandi R, Kim J, Xie Y, De Brabander JK, McFadden DG · Cell Chem Biol 2022What this evidence cannot show
Abstract read in full from the PubMed record on 2026-09-23. The experiments are in Ewing cell lines; the paper establishes that microtubule destabilisation is sufficient to account for cytotoxicity and for resistance, and it does not formally exclude that TK216 also engages EWS::FLI1 at some concentration. The practical consequence is the same either way: the cytotoxic mechanism is not the interface, so the interface cannot be credited with the phenotype.
Evidence that argues against this (1)
- The original series was selected and characterised on direct binding to EWS-FLI1 by surface plasmon resonance and on a specific block of the RHA interaction, and those biophysical results are not retracted. A defender of the interface model would argue that TK216 is a dual-mechanism compound whose microtubule activity dominates at cytotoxic concentrations - which, if true, still means the clinical programme was testing the wrong mechanism. Erkizan HV et al. 2009
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
In the open-label multicentre phase I/II first-in-human trial of TK216 in relapsed/refractory Ewing sarcoma, 85 patients (median age 27, range 11-77, median three prior systemic regimens) were enrolled across 11 cohorts; the maximum tolerated dose of the 14-day continuous infusion was 200 mg/m2 once daily, selected as the recommended phase II dose; in cohorts 9 and 10 there were two complete responses, one partial response and 14 stable disease, with 6-month progression-free survival of 11.9%; there were no responses among the eight patients in cohort 11; vincristine could be added from cycle 2 at investigator choice; and the authors conclude TK216 showed limited activity at the RP2D.
Open-Label, Multicenter, Phase I/II, First-in-Human Trial of TK216: A First-Generation EWS::FLI1 Fusion Protein Antagonist in Ewing Sarcoma. · Meyers PA, Federman N, Daw N, Anderson PM, Davis LE, Kim A, Macy ME, Pietrofeso A, Ratan R, Riedel RF, Trucco M, Breitmeyer JB, Toretsky JA, Ludwig JA · J Clin Oncol 2024What this evidence cannot show
Abstract read in full from the PubMed record on 2026-09-23. This is a heavily pretreated relapsed/refractory population in which any agent would be expected to perform poorly, so the low activity is not by itself proof that the mechanism was wrong - it is the mechanism paper [E2] that supplies that. The permitted vincristine co-administration means some observed activity cannot be attributed to TK216 alone.
Evidence that argues against this (1)
- Two complete responses in relapsed/refractory Ewing sarcoma is not nothing, and a defender of the programme could argue the trial was underpowered for a real but modest effect, or that the RP2D was set by infusion logistics rather than by target engagement. The counter to that counter is that a microtubule agent would produce exactly this pattern in a sarcoma, so the responses do not discriminate between the two mechanisms. Povedano JM et al. 2022
- E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Somatic MEN1 mutations at the revumenib-menin interface were identified in patients with acquired resistance to menin inhibition; the same inhibitor-menin interface mutations recur as a resistance mechanism in xenograft models and in an unbiased base-editor screen; and the mutants attenuate drug-target binding through structural perturbations that affect small-molecule binding but NOT the interaction with the natural ligand MLL1, preventing inhibitor-induced eviction of menin and MLL1 from chromatin.
MEN1 mutations mediate clinical resistance to menin inhibition. · Perner F, Stein EM, Wenge DV, Singh S, Kim J, Apazidis A, Rahnamoun H, Anand D, Marinaccio C, Hatton C, Wen Y, Stone RM, Schaller D, Mowla S, Xiao W, Gamlen HA, Stonestrom AJ, Persaud S, Ener E, Cutler JA, Doench JG, McGeehan GM, Volkamer A, Chodera JD, Nowak RP, Fischer ES, Levine RL, Armstrong SA, Cai SF · Nature 2023What this evidence cannot show
Abstract read in full from the PubMed record on 2026-09-23. This is the control TK216 lacked, and it is cited here for that comparative purpose rather than as a claim about resistance management. The base-editor screen is a saturating scan of MEN1 and is therefore unbiased WITHIN the gene; it is not a genome-wide screen, so it could not have surfaced an off-target mechanism of the kind that unmasked TK216. That distinction is the residual gap this entry names.
Evidence that argues against this (1)
- Because the base-editor screen was scoped to MEN1, it cannot exclude an additional off-target contribution to revumenib's cellular activity in the way a genome-wide mutagenesis screen could. The patient mutations are strong evidence that the clinical effect runs at least partly through the interface; they do not bound how much of the phenotype is on-mechanism, which is the exact quantity the Ewing case shows can be zero. Povedano JM et al. 2022
- related to →EWSR1::FLI1 (EWS-FLI1), t(11;22)(q24;q12)
the family whose interface programme reached the clinic and was deconvoluted to tubulin
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
the index fusion whose interface programme the analogy bears on
the worked clinical failure of the generalisation, and the reason its minimum-sufficient-test open question is not optional
- related to →Analysis (planned) — comparative druggability scan of every obligate fusion-partner interface in the atlas
supplies the causal-validation column the scan does not score
the partner whose patient-derived interface mutations are the control TK216 never had
the panel should include a forward resistance screen arm; TK216 is why
sibling entry on the same family - that one on its interface programme, this one on its addresses