Phase-2 dossier: the non-small-molecule modalities - degraders first, and the honest limits of each
also called degrader modality dossier · PROTAC readiness · non-menin modalities · beyond occupancy
Instead of jamming a protein's pocket, you can tag the whole protein for destruction, silence the gene that makes it, or teach the immune system to find the cell. Each of these is a real option here, and each fails for a different reason that we can name in advance.
The same thing, for a scientist
This dossier covers the modalities that are not orthosteric small molecules, because three of the targets in this beat fail specifically on occupancy-based logic. ENL is a scaffold as much as a reader, so ligand occupancy may not phenocopy loss. KMT2A's CXXC domain is a protein-DNA surface that orthosteric chemistry cannot address. The fusion junction has no pocket at all. Targeted degradation answers all three in principle. The phase-2 question is not whether degraders work in cell culture - they do - but whether the three constraints that actually govern them (ternary-complex formability, E3 ligase expression in the relevant cell, and exposure in infant marrow) hold for this disease. Each gets a named gate. The dossier also records why RNA-directed and gene-editing approaches are being deferred rather than pursued, so the deferral is a decision rather than an omission.
Degraders: the three constraints that decide everything
1. Ternary-complex formability. A degrader must hold target and E3 ligase in a productive geometry. This is the one constraint that is now cheaply predictable: menin has a well-folded 84-pLDDT AlphaFold model and a 1.96 A inhibitor co-crystal, and ENL has a 2.10 A inhibitor co-crystal, so both have the warhead-bound structures a ternary-complex model needs. Boltz-2 can be used to triage linker geometries, with the explicit caveat that its own authors state the affinity module does not handle multimeric binding partners or cofactors 3.
2. E3 ligase expression in the target cell. CRBN and VHL are the workhorses. Their abundance in infant KMT2A::AFF1 blasts specifically - not in AML cell lines - is the datum nobody has checked, and it is obtainable from open expression data. This is a G1 item.
3. Exposure. Degraders are large, often beyond rule-of-five, and marrow penetration in an infant is the failure mode that cell culture cannot see. This is why the degrader gate in the go/no-go register is written on primary blasts ex vivo rather than on a line.
Where degradation beats occupancy, target by target
| Target | Why occupancy may fail | Degrader case |
|---|---|---|
| ENL/MLLT1 | YEATS reader also acts as a SEC scaffold; a pocket-blocker may leave the scaffold in place | Strong. An ENL degrader is reported in the literature and recorded in gene-mllt1; the head-to-head with the SGC-iMLLT-class probe is the cheapest decisive experiment in this beat |
| KMT2A (CXXC) | Protein-DNA surface, no orthosteric pocket; family-wide selectivity problem | Conditional. Degrading KMT2A removes CXXC function as a by-product, but wild-type KMT2A is LOEUF 0.070 and haploinsufficiency causes a developmental syndrome, so a non-selective KMT2A degrader is unacceptable in a child |
| Menin | Occupancy already works; an approved drug exists | Weak as a first-line case, but a degrader would be insensitive to the MEN1 interface mutations that confer clinical resistance only if it binds outside the mutated surface - and the reported mutations cluster at the drug-binding interface, so most current warheads would lose binding too |
| LEDGF/PSIP1 | Multivalent chromatin anchoring may defeat single-site blockade | Strong in principle, but there is no LEDGF warhead worth converting (3 ChEMBL rows, best 470 nM) |
The menin-degrader trap, stated precisely
A degrader is often assumed to beat resistance mutations. Here it probably does not, and the reason is specific. The MEN1 substitutions observed in patients cluster at four residues - S160, M327, G331 and T349 - at the inhibitor-menin interface 1. A degrader built from a current menin warhead binds that same interface, so the mutations that abolish inhibitor binding would abolish degrader binding as well. A degrader only escapes this if its warhead engages a different menin surface, and no such warhead exists.
This is the kind of statement a readiness review exists to make before someone spends a year making a menin PROTAC in order to defeat resistance it would not defeat.
RNA-directed and gene-editing approaches: deferred, with reasons
Fusion-transcript-directed ASO or siRNA. Conceptually ideal: the junction is unique at the RNA level too, and RNA does not need a pocket. Deferred because delivery to marrow blasts in an infant is unsolved, and because the junction sequence is patient-specific, making this a bespoke product. The same junction-class coverage calculation that gates the immunological route gates this one, and it is free to run.
Gene editing of the fusion. Deferred: excising or disrupting the fusion allele in vivo in every blast is beyond current delivery, and partial editing selects for unedited cells.
Immunotherapy. Not deferred and not covered here - the graph already holds substantial entries on CD19 CAR-T, CD22-directed therapy and myeloid-directed antigens, including the finding that myeloid-directed antigens are the only axis that survives lineage switch. Phase-2 readiness for that axis belongs with those entries, not here.
| phase2 verdict | START the ENL degradation-versus-occupancy comparison, which is cheap and decisive. DO NOT start a menin degrader intended to defeat resistance - the resistance mutations sit at the warhead's own binding surface. Defer RNA-directed and editing approaches, with the junction-class coverage calculation as the free prerequisite for reconsidering them. |
|---|---|
| compute backend | Boltz credits for ternary-complex triage, used as a filter and never as a decision; Explorer Slurm GPU only if a linker enumeration campaign is approved. |
modalities covered
[ "targeted protein degradation (PROTAC, molecular glue)", "fusion-transcript ASO/siRNA (deferred)", "gene editing of the fusion allele (deferred)" ]
modalities elsewhere
[ "CD19 CAR-T, CD22-directed therapy and myeloid-directed antigens are covered in the therapeutics entries, not here" ]
degrader constraints
{
"ternary_complex": "menin and ENL both have warhead-bound co-crystals (7UJ4 at 1.96 A, 6HT1 at 2.10 A) suitable as modelling starting points",
"e3_ligase_expression": "CRBN and VHL abundance in primary infant KMT2A::AFF1 blasts is unchecked; obtainable from open expression data; G1 item",
"exposure": "marrow penetration of beyond-rule-of-five molecules in an infant is the failure mode cell culture cannot detect; the gate is therefore written on primary blasts ex vivo"
}menin degrader trap
{
"observed_patient_resistance_residues": [
"S160",
"M327",
"G331",
"T349"
],
"observed_substitutions": [
"S160T",
"M327I",
"M327V",
"G331D",
"G331R",
"T349M"
],
"implication": "these cluster at the inhibitor-menin interface, so a degrader built from a current menin warhead would lose binding to the same mutants the inhibitor loses. A degrader escapes resistance only with a warhead at a different surface, and no such warhead exists."
}go no go
{
"G1_feasibility": {
"question": "Is the machinery present and is a ternary complex plausible?",
"GO": "CRBN or VHL detectably expressed in primary infant KMT2A::AFF1 blasts, and at least one linker geometry produces a self-consistent ternary model",
"STOP": "Neither E3 ligase is expressed above background in primary infant blasts"
},
"G2_biology": {
"question": "Does degradation do what occupancy does not?",
"GO": "For ENL: degrader reduces HOXA9 by at least 50% where a matched occupancy probe achieves less than 20%, at comparable target engagement",
"STOP": "Degrader and probe give the same result - occupancy was sufficient, and the degrader adds only complexity"
},
"G3_window": {
"question": "Exposure and normal-cell cost in the relevant setting",
"GO": "At least 90% target protein loss for at least 48 h in primary infant KMT2A::AFF1 blasts ex vivo, with under 30% healthy-donor CD34+ CFU loss",
"STOP": "Under 50% target loss in primary blasts at any tolerated concentration"
}
}- Degrading KMT2A removes wild-type KMT2A too, and wild-type haploinsufficiency causes a developmental syndrome; in an infant this is a serious, foreseeable harm.
- Degraders select for resistance through loss of the E3 ligase itself, a route that no amount of warhead optimisation addresses and that has been seen repeatedly with CRBN-based agents.
- A menin degrader marketed as resistance-proof would fail against the observed patient mutations, because those mutations sit at the warhead's own binding site. Believing otherwise would waste a programme.
- Boltz-2's authors state its affinity module does not handle multimeric partners; using it to rank ternary complexes is using a tool outside its declared domain, and a confident-looking score there is a trap.
- Are CRBN and VHL actually expressed in primary infant KMT2A::AFF1 blasts, and at what level relative to the AML lines where degraders are usually demonstrated?
- Does any menin binder engage a surface away from residues 160, 327, 331 and 349 - and if one were found, would it be potent enough to serve as a degrader warhead?
- Is there a ligandable surface on wild-type KMT2A that is occluded in the fusion, or exposed only in the fusion, which would permit fusion-selective degradation?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Clinical resistance to menin inhibition is mediated by somatic MEN1 mutations at the drug-menin interface; in a cohort of 31 patients treated for more than two cycles with material available, 12 (38.7%) carried one or more MEN1 mutations, with M327V/M327I, T349M, G331R and S160T reported.
MEN1 mutations mediate clinical resistance to menin inhibition · Perner F, Stein EM, Wenge DV, et al. · Nature 2023What this evidence cannot show
31 sequenced patients from one programme, all revumenib; the fraction is an estimate from a selected subset with material available and is likely biased upward relative to all treated patients.
Evidence that argues against this (1)
- A later cell-free DNA series adds G331D and shows these mutations detectable at very low variant allele fractions, implying the true prevalence and the diversity of escape substitutions are both under-counted by tissue sequencing. Blood Cancer Journal 2026
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
Structural starting points for ternary-complex modelling exist for both degrader-relevant targets: menin with SNDX-5613 at 1.96 A (7UJ4) and the ENL YEATS domain with the SGC-iMLLT probe at 2.10 A (6HT1).
RCSB PDB data API entry records for 7UJ4 and 6HT1 · RCSB PDB 2026What this evidence cannot show
A warhead-bound structure is necessary but not sufficient for ternary-complex design; the exit vector and its tolerance to linker attachment are not read off these structures.
Evidence that argues against this (1)
- Ternary-complex modelling from binary structures is the weakest step in degrader design, because the productive ternary geometry is not predictable from the two binary complexes and is usually found empirically. Having starting points is not the same as having a model. RCSB PDB 2026
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.preprint
Boltz-2, the current public release (v2.2.1), approaches free-energy-perturbation accuracy on a small benchmark (average Pearson 0.66 on a four-target FEP subset) while running roughly 1,000 times faster, but its authors state explicitly that the affinity module does not handle cofactors, ions, water or multimeric binding partners - which is exactly the regime a ternary degrader complex occupies.
Boltz-2: Towards Accurate and Efficient Binding Affinity Prediction · Passaro S, Corso G, Wohlwend J, et al. · bioRxiv 2025What this evidence cannot show
Preprint; the repository states that updated evaluation code for Boltz-2 is not yet released, so the reported accuracy cannot currently be reproduced independently. The 0.66 correlation is on four targets, none of which resemble menin or a YEATS domain.
Evidence that argues against this (1)
- The authors also report that performance varies strongly between assays and that the cause of that variance is unresolved, so a single headline correlation should not be used to justify triaging a real chemistry programme. source 2025
sibling modality: RNA agents need only a sequence, degraders need a binder to a largely disordered protein
the G1 gate is a degradation-versus-occupancy comparison
recommended alternative route for this candidate
registers the decision gates for the degrader modality
- related to →MEN1 drug-interface mutations — the first proof that a chromatin drug can drive Darwinian escape in patients
the observed substitutions sit at the warhead binding surface, which is why a menin degrader does not escape them
ENL is the target where degradation may beat occupancy; the head-to-head is the cheapest decisive experiment
- related to →Myeloid-directed antigens (CD33, CD123) — the only immunotherapy axis that survives lineage switch
immunotherapy readiness lives with the therapeutics entries, not in this dossier
the degrader dossier predates the HR scaffold and does not price this cost