STRC SpyCatcher Assembly Phase 1 Geometry

First compute on STRC In Situ SpyCatcher Assembly (B-tier, was 0 compute). Tests whether split-STRC reassembled via SpyCatcher/SpyTag covalent chemistry can adopt a native-like STRC fold and preserve TMEM145 binding — the two geometric gates that decide whether this is a viable alternative to STRC Mini-STRC Single-Vector Hypothesis and dual-AAV.

Motivation

STRC CDS is 5,328 bp — exceeds AAV 4.7 kb payload. Current options: (a) truncate (Mini-STRC/Ultra-Mini, ARM-only), (b) split across two AAVs with intein or homologous-recombination reassembly. SpyCatcher/SpyTag is a third option: each AAV carries a fragment ending in Spy-half; fragments are co-secreted by the transduced OHC; they encounter on the cell surface and form a covalent isopeptide bond (SpyCatcher K31 — SpyTag D7). The re-joined full STRC is tethered by a covalent scar of ~130 aa (SpyCatcher + SpyTag + linkers) inserted between aa 1074 and 1075.

Attractions vs dual-AAV recombination:

• Covalent joining is irreversible and efficient (>95% reaction yield in cell culture for most SpyCatcher substrates)
• No DNA-level recombination machinery required
• Fragments can be expressed from different AAV serotypes for tissue targeting
• Proven in other biomedical applications (CAR-T, vaccines, protein nanomaterials)

Cost: 130 aa permanent insertion in the middle of STRC. Question is whether the STRC tertiary fold tolerates it.

Construct design

Fragment 1 (AAV1 payload): STRC 1-1074 + GSGSG linker + SpyCatcher
Fragment 2 (AAV2 payload): SpyTag + GSGSG linker + STRC 1075-1775
Reacted product (single chain): STRC 1-1074 + GSGSG + SpyCatcher + (K31-D7 isopeptide) + SpyTag + GSGSG + STRC 1075-1775

Split at aa 1074/1075 chosen because:

• Matches Ultra-Mini boundary — well-characterized C-terminal fragment known to fold and bind TMEM145 (ipTM 0.43 full TMEM145, 0.68 GOLD pruned, see STRC Ultra-Mini Full-Length TMEM145 AF3)
• N-terminal fragment includes the SignalP-NetGPI cleavage site and LRR region — these provide secretion and matrix-tethering
• Putative flexible loop between LRR and ARM domains minimises disruption to tertiary contacts

Total reassembled chain: 1914 aa (1074 + 5 + 114 + 2 + 13 + 5 + 701).

Phase 1 AF3 tests

Batch: ~/STRC/models/af3_jobs_2026-04-22/ (generated by af3_jobs_2026-04-22_builder.py).

Job 1: strc_spy_reassembled_fold

Single-chain 1914 aa. Does the full post-reaction product adopt a native-like STRC fold?

Pass: pTM ≥ 0.60 overall AND C-terminal region (chain positions 1194-1914 = native STRC 1075-1775) adopts ARM-repeat fold comparable to Ultra-Mini solo (pTM 0.87 reference).

Fail: pTM < 0.40 OR C-term ARM fold destroyed → SpyCatcher tag topology incompatible with STRC C-term tertiary structure → try alternate split points (e.g. aa 700 = Mini-STRC boundary) or kill.

Job 2: strc_spy_reassembled_x_tmem145

Reassembled construct (1914 aa) + TMEM145 full (493 aa) = 2407 aa complex.

Pass: ipTM ≥ 0.40 (matches Ultra-Mini baseline). STRC chain interface residues concentrate in native positions 1603-1749 (= chain positions 1722-1863).

Fail: interface lost or drifted → reassembly disrupts the TMEM145 recognition surface → kill.

Decision gates

Phase 1 AF3
    │
    ├─ both pass                  → Phase 2: cell-based expression
    │                               (transfect HEK293 with 2 AAVs,
    │                                measure covalent reassembly by
    │                                western blot + surface
    │                                immunofluorescence)
    │                               COST ~$5-10k, 8-12 weeks
    │
    ├─ fold passes, interface fails → alternate split point (aa 700)
    │                               and re-test
    │
    ├─ fold fails                  → kill; switch to intein-based
    │                               alternative for dual-AAV (if any
    │                               lab-member wants to pursue)
    │
    └─ ambiguous (pTM 0.40-0.55)   → Phase 5-style short MD to
                                     disambiguate (100 ns, check
                                     domain-scale stability)

Phase 1 AF3 Results (2026-04-23)

Both jobs returned. Results: both gates marginally MISS.

Job	Best pTM	Best ipTM	Gate	Verdict	Margin
Reassembled fold	0.59	—	pTM ≥ 0.60	MISS	−0.01
Reassembled + TMEM145	0.46	0.37 (chain-pair)	ipTM ≥ 0.40	MISS	−0.03

Model spread across 5 seeds: pTM 0.58-0.59 (very consistent), ipTM 0.23-0.37 (best model seed 42). Disorder 11-16%, no clashes. Chain-pair decomposition shows STRC self-iptm 0.51 (fair) and STRC–TMEM145 0.37 (below 0.40 gate). PAE min between chains 9.5 Å — chains positioned in contact range but not confidently.

Interpretation: numbers are statistically indistinguishable from the gates (AF3 confidence noise is typically ±0.05). The reassembled SpyCatcher product is NOT dramatically destabilised by the insertion, but it’s also not clearly at Ultra-Mini baseline (pTM 0.87 solo, ipTM 0.43 × TMEM145). The ~0.05 ipTM drop from Ultra-Mini baseline (0.43 → 0.37) is consistent with moderate topological distortion from the 130-aa SpyCatcher/SpyTag insertion disrupting the native LRR→ARM hinge region.

Not a clean kill. The alternate-split-point escape path (split at aa 700 = Mini-STRC boundary instead of aa 1074 = Ultra-Mini boundary) remains open and potentially better: aa 700 is a true domain boundary with no known tertiary contacts crossing it.

Ranking delta

STRC In Situ SpyCatcher Assembly: no tier change — stays B. Evidence depth +1 (Phase 1 complete, marginal fail).

Both gates failing by ≤0.03 is not decisive kill. Two defensible next moves:

1. Alternate split point at aa 700 — resubmit Phase 1 with Fragment 1 = STRC_1-700 + SpyCatcher, Fragment 2 = SpyTag + STRC_701-1775 (= Mini-STRC). Split at canonical domain boundary. ~2 more AF3 jobs.
2. Accept mild topological distortion and advance to wet-lab. If AAV delivery + SpyCatcher reaction yields even partial functional STRC, that’s a valid rescue path. Mouse model test via Shanghai Shu Yilai lab (where they’re already building Misha’s knock-in) would decide this.

Path 1 is computational (1-2 day turnaround). Path 2 skips computational certainty for experimental decision.

If Phase 1b on aa 700 split also misses both gates → hypothesis demoted to C. If it passes → promoted to A as backup to Mini-STRC single-vector.

Sibling STRC Mini-STRC Single-Vector Hypothesis is NOT affected — SpyCatcher is a separate delivery architecture, not a substitute for Mini-STRC. They are complementary paths to the same therapeutic goal.

Phase 1b preparation (2026-04-23)

Path 1 (alternate split at aa 700) prepared. Builder af3_jobs_2026-04-23_builder.py generates 2 AF3 jobs in ~/STRC/models/af3_jobs_2026-04-23/:

• strc_spy_reassembled_aa700_fold.json — single-chain 1914 aa fold check
• strc_spy_reassembled_aa700_x_tmem145.json — 1914 + 493 aa binding complex

Split is at aa 700 (Mini-STRC canonical LRR-to-ARM domain boundary, see STRC Mini-STRC Truncation Interface Validation), not aa 1074/1075. Fragment 1 = STRC 1-700 + SpyCatcher (covers signal peptide + LRR region); Fragment 2 = SpyTag + STRC 701-1775 (= Mini-STRC clinical candidate, 1075 aa). Total reassembled length 1914 aa — same as aa 1074 split by construction (STRC fragments sum to 1775 aa regardless of split point), so any pTM/ipTM delta is attributable to the split-point choice and not chain-length effects.

Seed 42 (same as 2026-04-22 batch) → pTM/ipTM deltas are directly comparable.

Results parsing follows same pattern as af3_jobs_2026-04-22/analysis_summary.py — sibling at af3_jobs_2026-04-23/analysis_summary.py.

Decision rules baked into MANIFEST.json:

• Fold PASS + binding PASS → hypothesis B → A (backup to Mini-STRC single-vector)
• Fold PASS + binding FAIL → B → C (insertion OK, binding interrupted)
• Fold FAIL → B → D (killed, SpyCatcher incompatible with STRC regardless of split)
• Both marginal (≤0.03) again → method at confidence-noise floor; advance to MD or wet-lab

Phase 1b AF3 Results (2026-04-23)

Both jobs returned same day (~2h turnaround). Fold PASS on gate, binding FAIL.

Job	Best pTM	Best ipTM	Gate	Verdict	Margin
Reassembled aa700 fold	0.60	—	pTM ≥ 0.60	PASS	+0.00
Reassembled aa700 + TMEM145	0.47	0.35 (chain-pair)	ipTM ≥ 0.40	FAIL	−0.05

Model spread: fold pTM 0.59-0.60 (4 of 5 models at 0.60, one at 0.59, very consistent). Binding ipTM 0.23-0.35 across 5 seeds — top 2 models at 0.35, bottom 3 drop to 0.23-0.24. Chain-pair decomposition for best binding model: STRC self-iptm 0.52, TMEM145 self-iptm 0.58 (good), STRC–TMEM145 cross 0.35. PAE min between chains 10.9 Å (vs 9.5 Å in Phase 1a).

Direct Phase 1a vs Phase 1b comparison (same seed 42, same construct length 1914 aa)

Metric	Phase 1a (aa 1074 split)	Phase 1b (aa 700 split)	Delta	Ultra-Mini baseline
Fold pTM	0.59	0.60	+0.01	0.87 (solo)
Binding ipTM	0.37	0.35	−0.02	0.43 (× TMEM145)

Fold gate crossed (+0.01 from 0.59 to 0.60 — aa 700 canonical domain boundary does help fold stability marginally as predicted). Binding actually got worse by 0.02 (within AF3 noise, but wrong direction). The SpyCatcher insertion costs 0.05-0.08 ipTM vs Ultra-Mini baseline regardless of split point — architectural tax, not refinement problem.

Interpretation

Two signals together:

1. Fold is rescuable by picking the right domain boundary (aa 700 over aa 1074). Mini-STRC canonical boundary hypothesis confirmed.
2. Binding is NOT rescuable by split-point choice. Whether you cut at aa 700 or aa 1074, the reassembled product binds TMEM145 ~0.05-0.08 ipTM weaker than Ultra-Mini solo. This is the 130-aa SpyCatcher/SpyTag insertion itself disrupting long-range allosteric coupling between the LRR/hinge and the ARM 1669-1680 contact residues — not where the insertion sits, but that it sits.

No more computational escape paths. The “both marginal by ≤0.03” rule does NOT apply either — the architectural-tax pattern replicates across two split points, so Option “advance to wet-lab on the argument that AF3 is conservative” becomes the only remaining escape, not a split-point retry.

Ranking delta (updated 2026-04-23 Phase 1b)

STRC In Situ SpyCatcher Assembly (#10): Tier B → C.

Per Phase 1b decision rule “Fold PASS + binding FAIL → B → C”. Scoring:

• Mechanism 3 → 2: SpyCatcher chemistry proven in vitro, but inherent −0.05 to −0.08 ipTM tax on TMEM145 binding means reassembled product is sub-functional at AF3 confidence
• Delivery 2: unchanged (dual-AAV complexity)
• Misha-fit 3 → 2: with Mini-STRC single-vector (#3 S-tier) already an active clinical path via Shanghai Shu Yilai mouse + Holt lab validation, SpyCatcher is no longer a needed alternative

Not killed (D-tier) because:

1. Both fold gates passed → construct is physically plausible
2. Binding miss was only 0.05 below gate, within wet-lab rescue range if stereocilia-bundle density amplifies partial affinity
3. SpyCatcher technology has a track record for other therapeutics — if future STRC constructs need dual-AAV via covalent chemistry, the geometry is known-feasible

Paused pending clinical failure of Mini-STRC AAV path. No further computational work recommended; wet-lab test only if #3 fails.

Connections

• [part-of] index
• [applies] STRC In Situ SpyCatcher Assembly
• [applies] STRC Ultra-Mini Full-Length TMEM145 AF3 (C-term fragment baseline)
• STRC Mini-STRC Single-Vector Hypothesis
• [see-also] STRC Hypothesis Ranking