STRC Engineered Homodimer Phase 1 Results
Phase 1 AF3 test of 4 single-point mutations (R1581F, R1581Y, S1579F, S1579W) on the STRC Engineered Homodimer Avidity weak homodimer interface. Result: ALL 4 FAIL. Mutations destabilize the homodimer further (mean ipTM 0.14-0.19 vs WT 0.28-0.30 baseline) and 3/4 also destroy TMEM145 binding. The R-R repulsion hypothesis is falsified: R1581F removes the arginine pair but the homodimer stays weak.
Method
- 4 mutants × 5 AF3 models each × 2 seeds (original = 42, restart = 1625243150) = 10 models per mutant for R1581F/R1581Y (restart missing → 5 models) and 10 models per mutant for S1579F/S1579W.
- 4-chain complex per job: A,B = Ultra-Mini mutant (701 aa × 2), C,D = TMEM145 full (493 aa × 2). Total 2388 aa per job.
- Gates (MANIFEST): (1) homodimer A↔B ipTM ≥ 0.50; (2) each Ultra-Mini × TMEM145 pair (A↔C, B↔D) ipTM ≥ 0.40.
- WT_ref job not returned; baselines taken from prior runs: homodimer 0.28-0.30 (STRC Homodimer Interface From CIF), Ultra-Mini × TMEM145 monomer 0.43 (STRC Ultra-Mini Full-Length TMEM145 AF3).
Results (combined, per mutant)
| Mutant | n | Homodimer A↔B (mean / max) | Ultra-Mini × TMEM pair 1 (mean / max) | Ultra-Mini × TMEM pair 2 (mean / max) | Verdict |
|---|---|---|---|---|---|
| R1581F | 5 | 0.176 / 0.190 | 0.462 / 0.480 | 0.434 / 0.440 | FAIL (homodimer) |
| R1581Y | 5 | 0.162 / 0.180 | 0.180 / 0.200 | 0.180 / 0.200 | FAIL (both) |
| S1579F | 10 | 0.156 / 0.170 | 0.180 / 0.210 | 0.178 / 0.200 | FAIL (both) |
| S1579W | 10 | 0.148 / 0.160 | 0.307 / 0.450 | 0.292 / 0.420 | FAIL (both, seed-sensitive) |
Gate never met by any single model, for any mutant. Multi-seed SD on homodimer ipTM is 0.004-0.015 — the signal is tight; not noise.
Full output: ~/STRC/hypotheses/h26-engineered-homodimer/artifacts/phase1_af3_2026-04-23/af3_jobs_2026-04-23d_engineered_homodimer_analysis.json.
Interpretation
The R-R repulsion hypothesis is falsified. R1581F removes both arginines; if R-R repulsion was the destabilizer, F-F stacking should have rescued the dimer. It didn’t — homodimer ipTM actually dropped from 0.28-0.30 (WT) to 0.17-0.19 (R1581F). Every other mutation was worse.
Two plausible readings of the result:
- Real biology. The interface genuinely prefers the S-G-R sequence; something beyond simple charge repulsion (e.g. context-specific H-bonding, helix-edge electrostatics across the ARM repeat) is load-bearing. Single-point substitution at 1579-1581 cannot improve it.
- AF3 MSA/template bias. The wild-type sequence has more paralog/ortholog templates; mutation drops template coverage → ipTM drops by default. This would be a systematic artifact, not biology. Phase 1b (double-mutant) or wet-lab SEC-AUC on WT Ultra-Mini would disambiguate.
TMEM145 binding: R1581F preserves monomer binding (0.43-0.48, matches WT baseline 0.43) — this is a useful control showing the mutation didn’t globally break fold. The three other mutations destroy TMEM145 binding (drops to 0.18-0.20), suggesting 1579-1581 are more central to the TMEM145 interface than predicted. This is new information about the interface topology.
S1579W restart seed anomaly: S1579W original seed destroys TMEM145 (mean 0.17); restart seed with different model seed preserves it (mean 0.44). Same sequence, same structure templates, different random init → 0.27 gap. This is an AF3-convergence concern for all Phase 1 results, but does not rescue the homodimer gate (both seeds fail on A↔B).
Decision
Single-point engineering at 1579-1581 is dead as a path. Two possible continuations:
- Phase 1b: double-mutant AF3. Try paired substitutions (e.g., S1579W + L1582Y, or L1580C + L1580C for disulfide) that couple geometric repacking. 4 AF3 jobs, budget-compatible. Cheap test, but prior is low now that single-point data shows the interface resists R/S substitutions.
- Phase 1c: structural review. Revisit STRC Homodimer Interface From CIF with broader contact tolerance (3.5 Å → 5 Å cutoff) to identify secondary contact residues outside the 1579-1581 triplet. If new contact cluster found (e.g., aa 1640-1650 in the adjacent ARM repeat), design new mutants there.
- Kill h26 entirely. Accept that avidity-boost via engineered homodimer is not achievable with the current Ultra-Mini scaffold. Upstream consequence: STRC Mini-STRC Single-Vector Hypothesis #3 loses its planned Kd-upgrade path but retains the monomer baseline (ipTM 0.43 already sufficient for Phase 2 wet-lab commitment).
Recommended: Phase 1c structural review first (no new AF3 cost), then Phase 1b conditionally. Do not kill outright — the avidity principle is independently sound; it’s the specific 1579-1581 targeting that failed.
Upstream impact
Downgrade of h26 does not weaken #3 Mini-STRC — the #3 S-tier case rests on monomer ipTM 0.43, GOLD-zone pruning 0.68, and CpG-depleted CDS, none of which depended on the homodimer working. h26 was a bonus improvement layer. Removing it restores the original #3 plan.
Ranking delta
#26 Engineered Homodimer Avidity: A → B.
- Mech 4→3 (R-R repulsion mechanism falsified; avidity principle still valid but no viable engineering path identified)
- Deliv 5→5 (unchanged; still drop-in to h03 AAV if any future mutant passes)
- Misha-fit 4→4 (unchanged)
- min = 3 → B-tier
Next step: Phase 1c structural review of STRC Homodimer Interface From CIF at 5 Å contact cutoff. If new cluster identified → Phase 1b double-mutant AF3. If no new cluster → demote to C-tier.
Connections
[part-of]index[part-of]STRC Engineered Homodimer Avidity — parent hypothesis; this note is its Phase 1 result- STRC Engineered Homodimer Avidity §“Why the dimer is weak: R-R repulsion” — single-point R→F removes arginines but homodimer still weak
[applies]STRC Homodimer Interface From CIF as the weak-dimer baseline[applies]STRC Ultra-Mini Full-Length TMEM145 AF3 as the monomer × TMEM145 baseline- STRC Mini-STRC Single-Vector Hypothesis — removes the planned Kd-upgrade path; core #3 unchanged
[see-also]STRC Hypothesis Ranking[see-also]STRC Hypothesis Ranking Log[about]Misha