Linker rules for EGFR-VHL PROTACs, read off the MD runs
The system is fixed: a gefitinib warhead (3-Cl-4-F-anilinoquinazoline, propyl-O,
piperazine) and VH032 (tert-leucine, 4-hydroxyproline, 4-methylthiazolyl-
benzylamine). Only the linker between the piperazine and the tert-leucine
varies. The EGFR-VHL SMILES bundled in protac_md.py are built differently: in
egfrvhl_rank4 and _rank5 the linker acylates the anilino nitrogen and the
piperazine NH is left free. Everything below comes from explicit-solvent MD on generated linkers,
run alongside the validated degrader MS39 and a PEG4 negative control.
Two gates
- Permeability. The linker has to let the molecule fold compact and bury its polar surface in the membrane: low 3D PSA, compact Rg, intramolecular H-bonds. Measured on the chloroform ensemble.
- Ternary geometry. It still has to reach an extended warhead-to-E3 span of 15-20 A at a cost the binding free energy can pay, below the 3.7 kcal/mol conformational penalty of MS39. Measured on the water ensemble.
Passing both means behaving like a molecular chameleon: mostly compact, with an accessible extended tail.
Runs
Same warhead and E3, linker only. Neither the trajectories nor the linker SMILES
are shipped here, so the table records what those runs showed rather than
something the scripts regenerate. 3D PSA and Rg come from psa_ensemble.py and
rg_basins.py; the clash-free fraction and the penalty come from a rigid
ternary-geometry check that is not part of this folder.
| linker | core | 3D PSA (A^2) | Rg in chloroform (A) | chameleon | clash-free fraction / penalty (kcal/mol) | verdict |
|---|---|---|---|---|---|---|
| dm2 | 2-(n-propyl)malonamide, short and branched | 140 | 5.96 | yes | 2.5% / 2.2 | passes both |
| perm4 | 2-(allyl)malonamide, short and branched | 141 | 5.92 | yes | 4.3% / 1.9 | passes both |
| MS39 (reference) | C10 linear diacid, sebacoyl | 178 | 6.51 | mild | 0.2% / 3.7 | validated degrader |
| perm2 | cyclopentane-1,2-diyl, rigid ring | 172 | 6.40 | no | 0.0% / >5 | fails ternary |
| perm3 | C6 linear diacid | 194 | 7.39 | no, folds in water and opens in the membrane | 1.3% / 2.6 | fails permeability |
| PEG4 | PEG4 diacid | 193 | 7.21 | no | 0.1% / 4.1 | fails permeability |
2D descriptors do not see this. TPSA is 191 for every linker except PEG4 at 228, so it cannot tell dm2 at 3D PSA 140 from perm3 at 194; the difference is folding in three dimensions. Rotatable bonds: 18 for the two winners, 20 for perm3, 24 for MS39, 27 for PEG4. cLogP: 6.5-6.7 for the winners, 7.3-7.9 for perm2 and perm3, 8.8 for MS39, 5.4 for PEG4.
Rules, cheap enough to pre-screen before MD
- Short, branched aliphatic core. A 2-substituted malonamide or succinamide with a small hydrophobic branch such as propyl or allyl folds compact while the warhead's propyl-O-piperazine extension still covers the ternary span.
- Avoid three failure chemistries. PEG and poly-ether raise TPSA and HBA and stay floppy, which kills permeability (PEG4); reject high ether-O counts, TPSA above ~200, HBA above ~14. A rigid ring across the linker locks in a non-spanning geometry, which kills ternary formation (perm2). Long linear alkyl or diacid chains fold poorly (perm3) or push cLogP and PSA up (MS39).
- Keep it stiff enough. Rotatable bonds around 18, not 24-27.
- Moderate cLogP, roughly 6-7.
Hard experimental constraint
Published EGFR-VHL SAR: linkers whose fully extended warhead-to-E3 span is under about 13 A, fewer than roughly 12 chain atoms end to end, show no degradation. Count the full chain from the piperazine anchor to the tert-leucine, not just the core, and keep it at 12 atoms or more (Cheng, Yu et al., J. Med. Chem. 2020, the MS39 paper).
What MD should then confirm
- 3D PSA of the folded state at or below ~150 A^2, a compact chloroform Rg basin at or below ~6.0 A, and at least 95% of frames under 7 A.
- Chameleonic behavior: chloroform Rg below water Rg, intramolecular H-bonds at 1.5 or more.
- Ternary reach: a span of 15 A or more in 1-2% of the water ensemble, a penalty at or below ~3 kcal/mol.
Caveats
Propose in 2D, then confirm the top few by MD.
The ternary axis is a rigid free-solution proxy, not a validated degradation score. The 12-atom rule still applies on top of it.
Cores worth imitating
- dm2, 2-propyl malonamide:
CCC[C@@H](C(=O)N-VH032)C(=O)N-piperazine - perm4, the same with an allyl branch:
C=CC