KMT2B, AFF2, AFF3, AFF4 and ENL carry most of the contact residues of the two interfaces the fusion keeps (menin-KMT2A and AF9-AFF1), so a substitute is structurally possible and selectivity cannot be read from either surface; at the menin side KMT2B differs at W7
also called paralog substitution at retained interfaces · KMT2B MBM conservation · AFF family AHD contact conservation
If a drug blocks the contact between menin and KMT2A, or between AF9 and AFF1, a close relative of one of the proteins could in principle step in, and the drug might hit those relatives too. We looked at the exact amino acids that touch in solved 3D structures and asked whether the relatives have the same ones. They mostly do: the menin-grabbing stretch of KMT2B shares about 11 of 18, and the surface AFF1 uses to grab AF9 is almost identical in AFF2, AFF3 and AFF4, while ENL matches AF9 at 17 of 18 spots. So these two contacts are family-wide, not special to the fusion. This is about whether a substitute could exist, not whether it binds, is made in the cell, or is used.
The same thing, for a scientist
Contact residues were taken from PDB 3U88 (menin, KMT2A 75-mer, LEDGF; contacts present in both crystal copies, 4.5 A) and PDB 2LM0 (single-chain AF9-AFF1 NMR ensemble, contacts in at least half of 10 models). Conservation of those residues in paralogs was scored with a local alignment and compared with the same aligner on shuffled paralog sequences. KMT2B is identical at 11 of the 18 KMT2A residues that contact menin in MBM1 and MBM2 (null mean 5.6; unrelated proteins 2 of 18), keeping F9, P10, R12, P13, R6 and R8 but carrying glycine where KMT2A has W7. AFF2, AFF3 and AFF4 are identical to AFF1 at 10, 11 and 10 of its 12 AF9-contact residues (null about 2.2); ENL is identical to AF9 at 17 of 18 AFF1-contact residues. A third menin-contacting stretch (KMT2A 103-126) in 3U88 lies in an engineered chimera with about 68 native residues deleted and is not conserved beyond chance in KMT2B (5 of 19); it is excluded. Sequence identity at contact residues is not binding, expression or use.
Method and the controls that could have failed
Contacts from the deposited coordinates, not from literature descriptions 12. The KMT2A chain in 3U88 has two copies in the asymmetric unit; each pairs with exactly one menin copy (40 and 38 contacts, zero for the crossed pairs), so the contacts are not crystal packing. The RWRFP stretch had to appear among the contacts (it does, at 6-10) and unrelated proteins had to score near the null (they do: 2 of 18 on the menin side; 2 to 4 of 12 and 2 to 5 of 18 on the AF9 side) 12. The null is the same aligner on shuffled copies of the paralog; the smallest possible empirical p with 300 shuffles is 0.0033, so the p-values reported at that value are a floor, not a precise number.
What each result means
Menin side: the two motifs that the corpus already treats as the interface (MBM1 4-15, MBM2 23-40) both contact menin in both crystal copies, and the paper describing a menin-MLL inhibitor structure calls MLL binding bivalent through two N-terminal fragments 3. KMT2B keeps the core of MBM1 (F9, P10, R12, P13) but not the tryptophan at 7, which is the one notable difference and a possible handle for KMT2A-selective chemistry. AF9 side: AFF2/3/4 keep 10-11 of 12 contact residues of AFF1 and ENL keeps 17 of 18 of AF9's; the AF9 structure paper states that the core aliphatic residues are conserved in each of AF9's binding partners 2, which agrees. The consequence is that neither interface is specific to the fusion. A blocker is expected to affect the wild-type counterparts, and a substitute is structurally possible if one component is lost.
What not to conclude
This does not show that KMT2B binds menin as well as KMT2A (the menin-KMT2B affinity is unmeasured; the corpus holds that as an experiment), that AFF2 or AFF3 bind AF9 at these residues, that any paralog is expressed in KMT2A-rearranged cells or rises under drug, or that it would be used. The 2LM0 construct is an engineered single chain. The third menin-contacting stretch in 3U88 comes from an engineered chimera and is not evidence about native KMT2A.
| script | analyses/q204-failure-mode-atlas/paralog_interfaces.py |
|---|---|
| output | analyses/q204-failure-mode-atlas/paralog_interfaces.json |
| kmt2b identical at mbm1 mbm2 contacts | 11 of 18 |
| kmt2b null mean | 5.6 |
| kmt2b w7 | glycine |
| aff2 aff3 aff4 identical at aff1 contacts | 10, 11, 10 of 12 |
| enl identical at af9 contacts | 17 of 18 |
| segment 103 126 | construct-dependent, KMT2B 5 of 19, excluded |
| min empirical p with 300 shuffles | 0.0033 |
PDB
[ "3U88", "2LM0" ]
chain pair contacts 3U88 — the numbers behind the chart
{
"M-A": 40,
"M-B": 0,
"N-A": 0,
"N-B": 38
}- A drug aimed at either interface will not be fusion-selective on interface grounds alone; for AF9-AFF1 it would also reach the AFF4 super-elongation arm of normal cells.
- If selectivity is bought through W7, a KMT2B-driven bypass is the failure the next-generation drug would then have to meet, and KMT2B mutations would be the expected escape.
- Treating contact-residue identity as evidence of function would over-call substitution; this entry is a structural possibility, and the measurements listed above decide it.
- What is the menin-KMT2B affinity, and how does each clinical inhibitor's IC50 compare (the corpus experiment exp-menin-kmt2b-mbm-affinity-and-inhibitor-ic50)?
- Does the W7G difference in KMT2B change the groove geometry enough to give KMT2A-selective chemistry?
- Do AFF2 and AFF3 bind AF9 AHD at these residues, and are any of AFF2/3/4 or KMT2B up-regulated in KMT2A-rearranged cells under drug? Public expression deposits have not been re-analysed here.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
In PDB 3U88, 18 KMT2A residues in MBM1 and MBM2 contact menin in both crystal copies; KMT2B is identical at 11 of them against a shuffled-target null mean of 5.6, versus 2 of 18 for unrelated proteins; KMT2B has glycine where KMT2A has W7.
The same pocket in menin binds both MLL and JUND but has opposite effects on transcription (PDB 3U88), contacts computed by analyses/q204-failure-mode-atlas/paralog_interfaces.py on 2026-10-14 · Huang J, Gurung B, Wan B, et al. · Nature 2012What this evidence cannot show
Sequence identity at contact residues is not binding. Contacts are a 4.5 A heavy-atom definition on a 3.0 A structure. The crystallised KMT2A chain is a 75-mer with an internal deletion; only MBM1 and MBM2 contacts (residues 6-35) are used. KMT2B affinity for menin is not measured here.
Evidence that argues against this (1)
- Conserved contact residues do not guarantee binding: neighbouring residues, the loss of W7, and local structure all affect affinity. The corpus holds the menin-KMT2B affinity as an unmeasured experiment.
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
AFF2, AFF3 and AFF4 are identical to AFF1 at 10, 11 and 10 of the 12 residues that contact AF9 in PDB 2LM0, and ENL is identical to AF9 at 17 of 18 residues that contact AFF1, against shuffled-target null means of about 2; the AF9 structure paper reports that the core aliphatic residues are conserved in each of AF9's binding partners.
Leukemia fusion target AF9 is an intrinsically disordered transcriptional regulator that recruits multiple partners via coupled folding and binding (PDB 2LM0) · Leach BI, Kuntimaddi A, Schmidt CR, et al. · Structure 2012What this evidence cannot show
Abstract read for the paper's claim (2026-10-14); coordinates read for the contacts. The NMR construct is a single-chain AF9-AFF1 chimera. Whether AFF2 or AFF3 bind AF9 at these residues is not shown by this analysis. Empirical p-values are at the floor for 300 shuffles.
Evidence that argues against this (1)
- The corpus holds the ENL-AHD to AFF1 contact as structurally unsolved; identity at AF9's contact residues makes the AF9 model a reasonable proxy but does not replace a structure.
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
MLL interacts with menin in a bivalent mode involving two N-terminal fragments of MLL, and a small-molecule inhibitor structure shows the compound binds the MLL pocket and mimics key interactions.
Structural insights into inhibition of the bivalent menin-MLL interaction by small molecules in leukemia · Shi A, Murai MJ, He S, et al. · Blood 2012What this evidence cannot show
Abstract read only (2026-10-14).
Evidence that argues against this (1)
- The corpus records MBM2 as about 26-fold weaker than MBM1, so how much the second motif contributes to resistance or to drug response in a cell is untested; the 2012 inhibitor was a tool compound, not a current clinical one.
contact residues present in both crystal copies, and their conservation in KMT2B
MBM2 residues contact menin in both copies; KMT2B identity across MBM1 and MBM2
KMT2B keeps the MBM1 core at the contact residues
structural possibility only; affinity unmeasured
- related to →Measure the menin-KMT2B affinity nobody has measured, and the four clinical inhibitors' IC50 against it
the measurement that decides the structural possibility
ENL identical to AF9 at 17 of 18 AFF1-contact residues
AFF2/3/4 carry the AFF1 contact residues of the AF9 AHD surface
the same retained AF9 surface, with partner-conserved aliphatic core
atlas modes FM-020 and FM-021, tested