STRC h01 Phase 6c Full 6-Target Panel 2026-04-24

Extends STRC h01 Phase 6c COX Selectivity Panel 2026-04-24 from 2/7 → 6/6 dockable targets after literature curation of bound-ligand PDB IDs (none guessed, all verified bound HETATM resnames). Same 10-ligand probe panel, same ensemble pipeline. 60 docks delivered in 4 min 42 s.

Receptor curation — Stage A revision

TargetNew PDBBound ligandClassSource
TRPM48RD9A1H0Canthranilic-class pore blockerEkundayo 2025 Nat Commun 16:833
Cx50 (via Cx36 paralog)8QOJYMZ (mefloquine)quinoline antimalarial / pan-connexin blockerDing 2024 Cell Discov 10:68
KCNQ47BYMFBX (retigabine)KCNQ-family activatorLi T 2021 Mol Cell 81:25
TMEM16A7ZK3JRF (1PBC)aromatic pore blockerLam, Rutz & Dutzler 2022 Nat Commun 13:2798
COX-11Q4GBFL (bromoflufenamic)fenamate(existing)
COX-25IKRID8fenamic-class(existing)

Old IDs replaced: TRPM4 7MF0 (404), Cx50 7JN0 (apo), KCNQ4 7BYL (was KCNQ1), TMEM16A 5OYB (apo). BK channel skipped — no NSAID/blocker-bound PDB exists in RCSB; only NS1619 activator-bound is available, which is not a fenamate-class proxy. Cross-paralog warning logged for Cx36→Cx50.

Boxes derived from cif HETATM centroid (TRPM4 obabel pdb conversion drops HETATMs, falls back to cif parser); 22 Å cube (channel pockets are wider than h01 18 Å).

Stage B — full 6-target dock results

Per-ligand selectivity matrix (Kd in µM)

Ligandh01 mut Kd (5e)COX-1COX-2TMEM16ATRPM4Cx50 (via Cx36)KCNQ4
niflumic-acid41.10.420.271.6744.6218.7410.7
flufenamic-acid32.70.510.370.7724.156.1278.5
diflunisal104.20.721.043.0943.2238.4675.9
meclofenamic-acid94.00.913.815.4575.9270.6882.6
tafamidis(untested)1.521.544.7255.1127.2461.5
benzox-3CN-COOH (4h #1)(untested)1.041.742.8884.1167.2441.6
benzox-3OMe-COOH (4h #2)(untested)1.000.992.6082.3166.1559.4
iododiflunisal(untested)9.422.954.2851.1247.9614.9
aspirin (neg ctrl)40.322.136.1666.81066.42736.6
DMSO (proxy neg)176001410727396699472936063420

Bold = sub-5 µM Kd → off-target liability at therapeutic 10 µM dose.

Pipeline calibration (vs published IC₅₀)

TargetParentPredicted KdPublished IC₅₀Within 1 OOM?
COX-2niflumic0.270.1–0.3
COX-2flufenamic0.370.5–3
TRPM4flufenamic24.12.8⚠ 9× weaker (anthranilic-class A1H0C site is more permissive than flufenamic site)
TMEM16Aflufenamic0.7712⚠ 16× tighter (1PBC pore-blocker site is the JRF aromatic anchor, fenamates fit)
Cx50 (Cx36 proxy)flufenamic56.13 (Cx50, not Cx36)⚠ 19× weaker (cross-paralog underestimate, expected)
KCNQ4flufenamic278.556 (KCNQ-family)⚠ 5× weaker (retigabine site is activator-targeting, fenamates likely block elsewhere)

Two calibration take-aways:

  1. COX scoring is robust (within 2× of published) — interpretable Kd-level differences.
  2. Channel scoring shows site-specific deviations of 5-19× — interpret rank order, not absolute Kd. Use relative selectivity (ligand_X / ligand_Y at same target) as the primary metric, not absolute Kd.

Real findings

Finding 1 — TMEM16A is a SECOND off-target wall, not just COX

Phase 6c COX-only proof note flagged the COX selectivity wall. Full panel reveals TMEM16A is just as bad:

  • Tafamidis TMEM16A Kd 4.7 µM, COX-2 Kd 1.5 µM (3× ratio, both sub-10)
  • Benzoxazole-3CN TMEM16A 2.9 µM, COX-2 1.7 µM
  • Benzoxazole-3OMe TMEM16A 2.6 µM, COX-2 0.99 µM

Both COX (Arg120 carboxylate salt bridge) and TMEM16A (pore-blocker JRF aromatic anchor) accept the bicyclic-COOH pharmacophore at sub-5-µM. The tafamidis-playbook Tier-1/Tier-2 set is a double-off-target liability, not single.

For Misha’s pediatric target this matters because TMEM16A is highly expressed in stria vascularis (key cochlear epithelium for endocochlear potential maintenance) — TMEM16A inhibition at 1-5 µM would risk ototoxicity beyond just COX-driven blood-flow effects.

Finding 2 — KCNQ4 cleanly de-selected by all bioisosteres

CompoundKCNQ4 KdTherapeutic 10 µM bound fraction
tafamidis461 µM2%
benzoxazole-3CN-COOH442 µM2%
benzoxazole-3OMe-COOH559 µM2%
niflumic411 µM2%
flufenamic278 µM4%

The retigabine (FBX) pocket on KCNQ4 (at S5/S6 interface) is dimensionally and chemically incompatible with the carboxylate-anchored fenamate/benzoxazole pharmacophore. KCNQ4 = DFNA2 paralog — this is the most safety-critical channel for a hearing-loss pediatric drug, and Phase 6c says we’re clean on it.

Finding 3 — TRPM4 mildly de-selected by bicyclic bioisosteres

CompoundTRPM4 KdImprovement vs flufenamic (24 µM)
flufenamic24.11× (baseline)
diflunisal43.21.8×
niflumic44.61.8×
iododiflunisal51.12.1×
tafamidis55.12.3×
meclofenamic75.93.1×
benzoxazole-3OMe82.33.4×
benzoxazole-3CN84.13.5×

Modest 3-4× selectivity gain from anthranilic→benzoxazole pivot. At 10 µM therapeutic, tafamidis TRPM4 bound-fraction = 15% — borderline, would still register on patch-clamp as ~20% inhibition. Better than parents but not transformative.

Finding 4 — Cx50 (via Cx36 proxy): bicyclic bioisosteres mildly preferred

Cross-paralog caveat applies (the 8QOJ Cx36-mefloquine pocket is a different ligand class than fenamic-class Cx50 binding). Pattern: flufenamic 56 µM (best, but Cx36 not Cx50), tafamidis/benzoxazoles 127-167 µM, diflunisal/iododiflunisal/meclofenamic 238-271 µM. Bicyclic bioisosteres are 2-4× tighter than non-fenamic NSAIDs but 2-3× weaker than flufenamic. Insufficient resolution for actionable selectivity claim — Phase 8 wet-lab dye-transfer assay is the authoritative gate.

Selectivity heatmap — what dies, what survives

For each compound at therapeutic 10 µM dose, classify off-target as:

  • KILL if bound-fraction ≥ 50% (sub-10 µM Kd) → would register on activity assay
  • WARN if bound-fraction 20-50% (10-40 µM Kd) → may register on sensitive assay
  • CLEAN if bound-fraction < 20% (Kd > 40 µM) → safe at therapeutic
CompoundCOX-1COX-2TMEM16ATRPM4Cx50KCNQ4Verdict
tafamidisKILLKILLKILLWARNCLEANCLEAN3 kills — not viable as-is
benzox-3CN-COOHKILLKILLKILLCLEANCLEANCLEAN3 kills — not viable
benzox-3OMe-COOHKILLKILLKILLCLEANCLEANCLEAN3 kills — not viable
diflunisalKILLKILLKILLWARNCLEANCLEAN3 kills
iododiflunisalKILLKILLKILLWARNCLEANCLEAN3 kills
meclofenamicKILLKILLKILLWARNCLEANCLEAN3 kills
niflumicKILLKILLKILLWARNCLEANCLEAN3 kills
flufenamicKILLKILLKILLWARNKILLCLEAN4 kills

Every Phase 4h Tier-1/Tier-2 candidate has ≥ 3 sub-10 µM off-targets. None are wet-lab viable as currently designed.

Implication for v4 in-flight (Stage 1 700/1572 as of writeup)

v4 fragment-grow library uses the same locked cores (3-amino-benzofuran-2-COOH + benzoxazole + benzothiazole) and same -COOH / -CONHOH / -CONHSO₂Me / tetrazole acid bioisosteres as the Phase 4h playbook. Top-N v4 GREENs/YELLOWs will inherit the COX + TMEM16A liabilities unless they break the carboxylate engagement entirely.

Decision for v4 post-verdict:

  • If v4 returns GREEN/YELLOW → cross-dock top-30 against this Phase 6c 6-target panel (5 min wall) BEFORE any Phase 8 ordering. Real shortlist = compounds that pass both K1141 affinity AND COX/TMEM16A selectivity. Likely 0/30 will pass currently — drives next-step toward v4b carboxylate-deselect sub-strategy.
  • If v4 returns RED → COX/TMEM16A liability is moot, focus on Phase 3c v5 RFdiffusion de novo pocket design.

v4b proposed sub-strategy (queue, do not launch yet):

  • Lock cores 3-amino-benzofuran-2-COOH + benzoxazole-5-COOH (v3b/v4 winners)
  • Replace -COOH with non-anionic K1141-engaging bioisosteres: amidoxime, N-hydroxyacetamide, 1,2,4-oxadiazol-5-ol, tetrazole (already partial), N-acylcyanamide
  • Each candidate must (a) keep K1141 ε-NH₃⁺ H-bond geometry, (b) drop the COX Arg120 salt bridge, (c) drop the TMEM16A pore-blocker aromatic-anchor charge interaction
  • ~500-1000 ligand library, focused enumeration, ~30-45 min wall

Action items

  • Cross-dock v4 top-30 against Phase 6c panel as soon as v4 lands (~50 min from this writeup)
  • Phase 4h playbook revision (already started in-place); add TMEM16A as second deselect target
  • Phase 8 SOP: promote both COX and TMEM16A activity assays to G2.5 kill gate (alongside MST), not deferred to G3
  • Decide on v4b carboxylate-deselect launch after v4 verdict (likely yes)
  • Replace BK channel skip with patch-clamp Phase 8 G3 gate (no NSAID-bound PDB exists; cannot dock-screen)
  • Cx50 native PDB needed (Cx36 paralog uncertainty) — wait for new connexin literature or run Cx50 AF3 + dock against pore-lining residues

Ranking delta

Tier A held. mech 3 / deliv 4 / misha_fit 4 all held.

  • Phase 4h “tafamidis-class bioisostere kills off-targets” thesis falsified for two off-targets (COX + TMEM16A), not just COX. KCNQ4 (DFNA2 paralog, most safety-critical) cleanly de-selected — partial validation.
  • Pipeline now has 6-target self-consistent off-target Kd matrix usable for medchem triage.
  • Material constraint: no current Tier-1/Tier-2 Phase 4h compound passes the 3-of-3 (h01 Kd + COX + TMEM16A) selectivity gate. Any wet-lab order of these compounds is a calibration purchase, not a lead-development purchase.
  • next_step unchanged: v4 still primary; v4 verdict triggers either cross-dock-then-Phase-8 (if YELLOW with selective compounds) or v4b carboxylate-deselect launch (if v4 chemistry inherits same off-target pattern).
  • lit_audit: fixed held; this is calibrated dock against verified bound-ligand PDBs.

Connections