NEGATIVE RESULT: Fasudil Scaffold Hopping Fails to Achieve LIMK2 Selectivity (0/20)
Summary (published as negative result with same rigor as positive ones)
We systematically modified Fasudil (approved ROCK inhibitor; DILI 0.28, BBB 0.93) across 9 chemical-modification categories to achieve LIMK2 selectivity for SMA. 115 variants generated, 65 passed ADMET-AI (56.5%), top 20 docked via DiffDock against LIMK2/LIMK1/ROCK1/ROCK2.
Result: 0/20 achieved the LIMK2-selective margin > 0.3 threshold.
Best variants
| Variant | Modification | LIMK2 | LIMK1 | ROCK1 | ROCK2 | Margin |
|---|---|---|---|---|---|---|
| B_7Cl | 7-chloro isoquinoline | +0.050 | −0.140 | −0.460 | −0.110 | +0.160 |
| B_7F | 7-fluoro isoquinoline | −0.090 | −0.090 | −0.020 | −0.160 | −0.070 |
| B_7CFFF | 7-trifluoromethyl | −0.460 | −0.360 | −1.560 | −1.180 | −0.100 |
| Fasudil_ref | Unmodified | −0.850 | −0.040 | −0.150 | +0.350 | −1.200 |
Best margin achieved: +0.160 (below the 0.30 selectivity threshold).
Categories tested
- Isoquinoline ring substitutions (positions 5, 6, 7, 8): F, Cl, CH₃, OCH₃, CF₃, NH₂
- Amine head replacements: morpholine, piperidine, pyrrolidine, homopiperazine
- Combined modifications
- Hybrid designs (LIMKi3/BMS-5 pharmacophores)
- Linker modifications (SO₂NH → CONH, CH₂NH)
- N-linked heterocycle heads
- Constrained analogs
- Fluorine scan
A positive finding inside the negative result
Scaffold hopping from Fasudil produced dramatically better ADMET profiles than de novo generation:
- Fasudil variants: 56.5% pass all ADMET gates (65/115)
- PocketXMol random generation: 0% pass all ADMET gates (0/1934)
This confirms Fasudil's isoquinoline core is inherently drug-safe — but it is also fundamentally ROCK-preferring.
Root cause
Fasudil's isoquinoline-sulfonamide scaffold has geometric complementarity with the ROCK kinase ATP pocket deeply encoded in its 3D shape. Key structural differences vs LIMK2:
- LIMK2 gatekeeper K383 (large, charged) vs ROCK2 M172 (small, hydrophobic)
- Position-7 modifications (pointing at the gatekeeper) moved LIMK2 confidence from −0.850 to +0.050, but could not overcome the scaffold's ROCK bias
- Sulfonamide linker geometry locks the molecule in a ROCK-favorable orientation
Conclusion and recommendations
- Do NOT pursue further Fasudil modifications for LIMK2 selectivity — scaffold is ROCK-intrinsic
- Fasudil remains the validated pan-ROCK/LIMK2 dual inhibitor for Track 1 → Simon
- LIMK2-selective compounds require a different scaffold — our PocketXMol DFG-out campaign already produced pyrazolo-pyridine hits (1219_0, margin +0.43), confirming novel scaffolds are necessary
- B_7Cl (margin +0.160) could be a starting point for ROCK2-selective optimization if that becomes useful
Methods
- Variant generation: RDKit SMILES enumeration
- ADMET: ADMET-AI v2.0.1 GNN (41 TDC endpoints, local GPU)
- Docking: DiffDock v1.1, 10 poses per compound, 4 targets
- Selectivity threshold: LIMK2_conf − max(LIMK1, ROCK1, ROCK2) > 0.3
Why publish this? Negative results with the same rigor as positive results keep the field honest and stop other labs from wasting resources on the same dead end. CC-BY-4.0. Full finding: /findings/2026-04-10/FINDING_2026-04-10_Fasudil_scaffold_hop_NEGATIVE.md.