STRC Research

STRC AF3 Static Pocket Blindness to Loop Dynamics

6 min read

STRC AF3 Static Pocket Blindness to Loop Dynamics

TL;DR. Two WT-vs-E1659A static-structure docking tests (Phase 4c 2026-04-21, Phase 4c-v3b 2026-04-24) returned marginal WT preference for acidic ligands. This does NOT falsify the pharmacochaperone hypothesis — it falsifies the premise that static AF3 structures can test it. The claim is fundamentally dynamical: E1659A destabilises loop 1642-1651, and the compound rescues folding by damping that loop. Static snapshots — whether AF3 or even MD-minimised — erase the ps-ns mobility that is the actual substrate of the rescue. Honest tests: MD ensemble comparison (Phase 5d mutant — exists) and holo-MD residence-time analysis (Phase 5g — not yet run).

Problem

Two rounds of static docking have tested the WT-vs-E1659A preference of candidate pharmacochaperones against AF3-predicted structures:

roundrostern ligandsmean ΔΔG(WT−mut)fraction preferring mutant
STRC Pharmacochaperone Phase 4c WT Decoy (2026-04-21)5 Phase 4b leads + diflunisal + 3 negatives9-0.455 kcal/mol0/5 leads
Phase 4c-v3b (2026-04-24, this note)29 v3b YELLOW + 7 Phase 4h Tier-1 seeds36pending full runpending full run

The early signal from the in-progress Phase 4c-v3b smoke (3 ligands, full run live at PID 75335) matches the 2026-04-21 direction: v3b chemistry (tafamidis parent + benzoxazole-COOH + benzofuran-COOH cluster) also mildly prefers WT. Whatever improvement Phase 3c v3b bought in absolute affinity (Kd 4.57 µM vs 30 µM fenamic), the WT bias is unchanged.

The naive reviewer reads this as: “If the compound binds WT at least as well as mutant, it is not rescuing anything — the pharmacochaperone mechanism is falsified.”

That reading conflates two different claims.

Two distinct claims

  1. Static geometric claim. The E1659A pocket adopts a unique static conformation that a small molecule can selectively bind, with WT pocket binding substantially weaker.
  2. Dynamic folding claim. E1659A destabilises loop 1642-1651 on the ps-ns timescale, increasing the probability of misfolded states. A small molecule that binds the K1141 neighbourhood in any conformation reduces loop mobility and shifts the folding equilibrium back toward the native state. This is the mechanism that tafamidis uses on TTR (Bulawa 2012, Coelho 2013) — it does not prefer amyloidogenic monomer over native tetramer; it stabilises tetramer by raising the kinetic barrier to dissociation.

Static docking tests (1). The pharmacochaperone literature on TTR makes claim (2). These are different experiments.

Why AF3 cannot resolve the dynamic claim

AF3 — like AF2 — predicts an equilibrium structure from MSA + pairwise features. The output is a static model. Critically:

  • Single-residue surface substitutions rarely perturb the backbone template. E1659A removes a glutamate sidechain; Ala1659 fits the existing backbone geometry without forcing Cα displacement (Ala is geometrically permissive, smaller than Glu by 3 heavy atoms). AF3 produces a near-identical global fold.
  • The differentiator is mobility, not geometry. The Glu1659→Ala1659 substitution removes the E1659-K1141 salt bridge that tethers loop 1642-1651. In WT the loop is constrained by the ~2.8-Å contact; in mutant the loop is free to sample a wider conformational ensemble. This is a Boltzmann-weighted distribution effect, not a mean-structure effect. AF3 reports only the mean.
  • Vina sees what AF3 gives it. Vina scores a single static receptor pose. If the two poses (WT and mutant AF3) are geometrically near-identical in the K1141 pocket, the Vina ΔΔG will be near-zero regardless of whether the underlying dynamics differ drastically.

This is not a bug in AF3 or Vina — it is the design envelope. The pharmacochaperone claim lives outside that envelope.

What the compound actually does

The mechanism, stated precisely: acidic rescue ligands (COOH / tetrazole / CONHOH, including the v3b benzofuran-2-COOH / benzoxazole-COOH / tafamidis-class cluster) engage K1141 ε-NH3+ in a salt bridge. Once engaged, the K1141 sidechain is pinned, which:

  1. Constrains the K1141-D1140-D1173 acidic cluster (see STRC Pharmacochaperone K1141 Fragment Pocket) through local electrostatic coupling.
  2. Reduces the conformational entropy of loop 1642-1651 indirectly, because the K1141 sidechain-loop contact network is the principal anchor.
  3. Shifts the folding equilibrium toward the native-state ensemble that both WT and E1659A transiently visit, but from which E1659A escapes at higher frequency.

None of these effects are visible to single-structure Vina. All are visible — in principle — to MD ensemble comparison (RMSF profiles, residence-time distributions, mutual-information networks).

The only honest tests

teststatuswhat it probes
STRC h01 Phase 5 MD Ensemble Rescoring 2026-04-23 (Phase 5a/5b)delivered 2026-04-23 (WT Ultra-Mini)ensemble Vina on 20 MD snapshots of WT pocket
Phase 5d E1659A full-length MDdelivered 2026-04-24, 2 ns × 20 snapshotsfirst ensemble on the actual disease target
STRC h01 Phase 5e Mutant Re-Dock Delivery 2026-04-24 (Phase 5e)delivered 2026-04-2411 ligands × 20 mutant snapshots — validates WT-based dock as mutant proxy at the Kd-ratio level (max 2.09×)
Phase 5d extended (50-100 ns × 3 replicates)not startedconvergence + RMSF profile comparable to reviewer standards
Phase 5d-WT (matched full-length WT MD)not startedWT-vs-mutant loop RMSF contrast
Phase 5g-holonot startedligand residence time, K1141-COOH contact persistence, loop RMSF with ligand present
Phase 5i-network (DCCM/generalized correlation)not startedallosteric pathway K1141-ligand → loop 1642-1651 → E1659A site

Phase 5e is the closest thing we have to direct evidence that the K1141 pocket is accessible on the actual disease target — it showed v3b WT-dock scores are reliable proxies for mutant Kd (max kd_ratio 2.09×). That validates docking as a screening tool but does not test the chaperone claim.

The chaperone claim lives in Phase 5g/5i. Neither exists yet.

Design implication

  1. Phase 4c and Phase 4c-v3b results are not falsification evidence against the pharmacochaperone hypothesis. They are evidence against using single-structure Vina to test it. Historical Phase 4c was interpreted as a fundamental challenge to h01; that interpretation should be revised in the Phase 4c-v3b proof note.
  2. Phase 4d K1141A double-mutant decoy (in flight) does test a falsifiable sub-claim. If removing K1141 ε-NH3+ costs ≥ 2 kcal/mol for acidic ligands on the E1659A background, the salt-bridge anchor claim is validated on static pocket — a necessary but not sufficient condition for the full chaperone claim. If the loss is < 1 kcal/mol, K1141 is not the critical anchor and the scaffold direction needs revision regardless of dynamics.
  3. The chaperone claim blocks on Phase 5g-holo. Until we have 50-100 ns of E1659A + top-3 ligand holo MD with RMSF / residence-time / contact-map analysis, any Nobel-tier reviewer will flag h01 as having no direct mechanistic evidence. Vina scores — however sub-µM — are not mechanism.
  4. Static-structure comparability of WT and mutant pockets is a methodological constraint, not a biology finding. Do not publish “compound binds WT equally well” as a negative. Publish “static-structure docking is orthogonal to the dynamical hypothesis; see Phase 5g”.

Ranking delta

  • No change. This note reframes interpretation, not evidence weight. h01 tier A held; mech 3 held; next step unchanged (Phase 3c v4 → medchem triage on v4 YELLOW/GREEN).

Connections

Graph View

Backlinks