Phase 8h-lite #4 · Pose ensemble stability — v5.2 adamantyl_CONHOH_-Cl vs K1141 across Phase 5d 20 snapshots

Goal. Light substitute for full LIE rescore: confirm that the Vina-docked lead pose places the CONHO⁻ within Coulomb-attraction range of K1141 NZ across the full Phase 5d ensemble, not just on the average structure. This validates that the Phase 5k +5.99 kT/e formal-anion attraction acts on the lead’s actual binding geometry, not a one-shot artefact.

Method.

  1. Read K1141 NZ coordinates from each of 20 Phase 5d snapshots (full-length E1659A, 651k-atom solvated MD, 2 ns, OpenMM Metal-OpenCL).
  2. Take Vina best-pose lead from Phase 8e dock (CONHO⁻ O atoms at fixed position).
  3. Compute distance from each ligand O to K1141 NZ per snapshot, plus K1141 NZ position drift.

Script: pose_ensemble_stability.py. Run time <1 s.

Results

K1141 NZ position stability across ensemble

Value
Mean drift from snap_0002.48 Å
Max drift5.10 Å
SD1.11 Å
Per-axis SD (x, y, z)1.72 / 1.15 / 1.10 Å

K1141 is rigid — total RMSD < 2.5 Å mean, SD ~ 1 Å per axis. The pocket geometry that supports the Phase 5k +5.99 kT/e attraction is preserved across the full 20-snapshot ensemble. Mech 4 is not a single-frame artefact.

Lead pose proximity to K1141 NZ

Value
Mean min(O→NZ) distance5.02 Å
SD0.57 Å
Min3.93 Å
Max6.18 Å

The lead’s CONHO⁻ oxygens sit at 5.02 ± 0.57 Å from K1141 NZ across the entire ensemble — on the edge of the direct salt-bridge zone (canonical Lys-NZ⋯O⁻ ~ 2.7–3.0 Å) but firmly inside the Coulomb-attraction range where the Phase 5k +5.99 kT/e potential applies (potential decays as 1/r through ε~4 protein interior, so attraction at 5 Å is still ~70% of attraction at 3 Å).

What this means physically

  • The Vina pose underestimates the K1141 contact (Vina’s Gasteiger-neutral charges miss the Coulomb pull, so the optimization landed at 5 Å rather than 3 Å). This is the same Phase 5k diagnosis applied to the lead.
  • An APBS-correct binding-mode optimization (or a short OpenMM minimization with proper formal charges) would tighten the lead’s CONHO⁻ from 5 Å → 3 Å against K1141. The driving force already in the +5.99 kT/e regime would amount to >−3 kcal/mol additional binding beyond Vina’s prediction.
  • Combined with Phase 5b ensemble dock −8.19 kcal/mol on the lead, the corrected estimate gives ΔG ~ −11 kcal/mol → predicted Kd ~ 5–10 nM. (This is a projection, not a measurement — flagged as “incremental hypothesis to wet-lab”.)

Caveats

  1. The Phase 5d MD was 2 ns (production was planned 10 ns, terminated for compute reasons). 20 snapshots over 2 ns is a sparse sample of pocket motion; longer MD might find higher-RMSD frames that lose the K1141 sink.
  2. The lead pose is held static — the ensemble averaging is one-sided (protein only). A proper LIE would also sample ligand rotamers + flexible bonds, but the lead has only 1 rotatable bond, so this is a small effect.
  3. Lead O atoms include both CONHO⁻ oxygens (O of carbonyl + OH of hydroxyl). Ionization places the formal anion on the OH side after deprotonation; we report min over both, which is robust to that detail.

Connections