TL;DR

We generated 52 AAV9 VP1 capsid variants with redesigned receptor-binding loops VR-V (490-506) and VR-VIII (581-597) using RFdiffusion (backbone) and ProteinMPNN (sequence). This is one leg of the three-component Simon cure protocol for SMA: ABE base editing + motor-neuron-tropic AAV delivery + Fasudil/LIMK2 inhibitor for surviving motor neurons.

Method

• Pipeline: RFdiffusion (backbone diffusion) + ProteinMPNN (sequence design)
• Target: AAV9 VP1 capsid for improved motor neuron tropism
• Contig: A219-489/10-25/A507-580/10-25/A598-736 — fix scaffold residues 219-489, 507-580, 598-736; hallucinate 10-25 residues replacing each variable region
• Variable regions redesigned: VR-V (residues 490-506) and VR-VIII (residues 581-597) — the two loops that determine tropism in known engineered serotypes (AAV-PHP.eB, AAV9.45)
• N designs: 52 backbones, 1 ProteinMPNN sequence per backbone
• GPU: A100 PCIe 80 GB (Sweden), instance 34565416
• Wall time: ~2.5 h

Why VR-V and VR-VIII

These are the two surface loops at the receptor-binding interface. Every published motor-neuron-tropic AAV variant (PHP.eB, AAV9.45, AAVMYO, MyoAAV) modifies one or both of these regions. They are the only region where rational redesign is known to successfully shift tropism without breaking capsid assembly.

Role in the cure protocol

   1. ABE base editing       -> restores SMN2 -> SMN1 (Liu lab platform, we extend)
   2. Motor-neuron AAV9      -> delivers ABE specifically to motor neurons (THIS CAMPAIGN)
   3. Fasudil / LIMK2 inh.   -> reverses actin pathology in surviving motor neurons (Track 2B)

Without (2), the editor reaches the wrong cells; without (3), edited motor neurons still inherit accumulated cytoskeletal damage.

Next steps

1. ESMfold validation of every redesigned loop (pLDDT > 70 cutoff)
2. AAVR (KIAA0319L) docking to filter out designs that lose primary AAV9 receptor binding
3. Top-10 selection for downstream cell-line transduction testing at a partner lab
4. Cryo-EM-quality renders of top 3 for figures

Where the data lives

• GitHub (metadata + top 5 PDBs): https://github.com/Bryzant-Labs/sma-research/tree/main/campaigns/AAV_capsid_design/2026-04-10_rfdiffusion
• Dropbox open_data (full 52 PDBs + ProteinMPNN sequences): Dropbox/SMA/open_data/aav_capsid_2026-04-10/

What this is NOT

• Not a validated tropism shift (cell-line testing pending)
• Not a replacement for clinically established AAV9 — these are computational designs only
• Not affiliated with any published engineered serotype

CC-BY-4.0. Compute cost: ~$5 (A100 PCIe Sweden). All artifacts open-source.