Chaperone-Orthogonal Cassette Stacking¶

The interesting first-principles question for multi-cassette engineering in A. oryzae is not "how many cassettes can we bolt on?" but "which cassettes share folding machinery, and which don't?" Multi-cassette yield in heterologous secretion is dominated by competition for ER folding subsystems, not by genome real estate, transcription, or translation capacity. Cassettes partitioned across non-overlapping chaperone classes should stack with sublinear yield decay; cassettes competing for the same chaperone class show steep yield collapse well before promoter or genome-architecture limits are reached.

This page makes the framework explicit, scores Open Enzyme's candidate cassettes against it, and proposes a falsifiable test (pairwise expression matrix → measured synergy coefficients). It is a predictive design framework — not a literature review of demonstrated multi-cassette work, which is covered in koji-endgame-strain.md §3 and engineered-koji-protocol.md. The specific 2-cassette uricase + lactoferrin analysis at cassette-compatibility-computational.md is a case study under this framework; this page generalizes it.

All claims here are Mechanistic Extrapolation unless otherwise tagged. The framework is biochemically grounded but has not been empirically validated by a published koji-specific pairwise expression matrix — that absence is itself a load-bearing finding (see §8).

1. Why Cassette Count Is the Wrong Metric¶

The naive view treats heterologous expression as a per-cassette resource draw: N cassettes = N × per-cassette burden. Under that model, stacking enzymes is a simple optimization — keep adding until promoter strength, marker availability, or growth rate becomes limiting. Industrial fungal engineering treats cassettes this way by default; multi-locus integration platforms like NSAR1 (5 simultaneous markers per Oikawa 2020) are framed as "more cassette slots" rather than "more chaperone budget."

Several lines of evidence from the existing wiki and the literature suggest the per-cassette model breaks down:

Huynh 2020 (PMC7257131; Fungal Biol Biotechnol 7:7, DOI 10.1186/s40694-020-00098-w) achieved 39.7 mg/L adalimumab in A. oryzae NSlD-ΔP10 only with the ten-protease-deletion background; the lower-protease-deletion strains (NSlDv10, AUT1-lD-v10-sD) and the control NSlD1 produced less, and wild-type RIB40 was used only as a DNA donor (Huynh 2020 Results §"Adalimumab production in the culture supernatant of A. oryzae", Fig. 1; Abstract: "the highest amount of antibody was obtained from the ten-protease deletion strain (39.7 mg/L)" — verified 2026-05-06). The bottleneck was not cassette count (two cassettes — heavy chain + light chain) but ER folding capacity for paired heavy + light chains carrying 16 disulfides per assembled IgG1 (4 inter-chain + 12 intra-chain; verified via UniProt P01857 IgG1 heavy chain CH constant region annotations + P01834 kappa LC; standard IgG1 stoichiometry, 2026-05-06). (In Vitro). At the time of this writing, this is the highest published mAb-class titer in A. oryzae — the apparent ceiling at ~40 mg/L for two PDI-heavy cassettes is consistent with the orthogonality hypothesis (PDI/ERO1 saturation).
Wakai 2019 (Bioresour Technol 276:146-153, PMID 30623869, DOI 10.1016/j.biortech.2018.12.117) achieved a 40× total cellulolytic activity uplift versus the single-integration strain in A. oryzae by stacking three cellulase cassettes (cellobiohydrolase + endoglucanase + β-glucosidase) at 5–16 chromosomal copies each, then optimizing promoter/terminator combinations (P-sodM/T-glaB demonstrated the strongest per-copy transcription) (Wakai 2019 Abstract; verified 2026-05-06). The decomposition: multi-copy integration alone gave ~10-fold over single-integration; promoter/terminator engineering on top of multi-copy integration gave the additional 4× to reach 40-fold (Wakai 2019 Abstract verbatim: "approximately 10-fold higher activity versus single integration strains... resulting transformant showing 40-fold higher cellulolytic activity versus the single integration strain"). [VERIFICATION CORRECTION 2026-05-06: the previous wiki framing — "approximately 10× from copy stacking; the rest from multi-cassette synergy" — was a misattribution. The 4× delta from 10-fold to 40-fold is from promoter/terminator optimization, not from multi-cassette synergy. The word "synergy" does not appear in the Wakai abstract.] Critically, all three cellulases are fungal-origin with modest disulfide load and N-glycan content — they share BiP/calnexin dominance but minimal PDI competition. The 40× titer is consistent with chaperone-orthogonal compatibility for fungal-native substrates, but it is not direct evidence of cross-cassette synergy distinct from per-cassette transcriptional scaling. (In Vitro).
Oikawa 2020 (PMC7725655) reconstituted a ≥17-gene fungal biosynthetic cluster in A. oryzae NSAR1 — but those genes encode small-molecule biosynthetic enzymes (intracellular, mostly cytosolic), which entirely bypass the secretion pathway. The 17-gene number is not comparable to 17 secreted heterologous proteins. (In Vitro)
Li 2024 (PMID 39830075) achieved 3.3× yield uplift by integrating multiple copies of the same cassette at distinct α-amylase loci — pure transcriptional/translational scaling of one substrate, not multi-substrate competition. (In Vitro)
Li 2023 (DOI: 10.3390/jof9050528) found that for A. niger monellin, copy-number gave diminishing returns (1→2 copies = 2.7×; 2→5 copies = 1.5×) and BiP/PDI overexpression did not rescue, but extracellular protease deletion gave 13.5× uplift. (In Vitro). For some substrates, post-secretion proteolysis is the binding constraint, not folding — emphasizing that "chaperone competition" is one of multiple potential ceilings.

Read together: the published multi-cassette successes either (a) use protease-deletion strains to expand folding headroom, (b) co-express substrates that don't load the same chaperone subsystem heavily, or © bypass the secretion pathway entirely. None of them tested two heavily-disulfide-bonded mammalian-origin secreted proteins in parallel beyond the Huynh 2020 antibody case, which appears to have hit a real ceiling — which would be the worst-case design for chaperone competition. No published koji-specific pairwise expression matrix tests cassette-cassette competition as a function of predicted chaperone-class overlap — the framework's central prediction has not been empirically validated.

The framework below predicts which combinations fall into the "easy" vs. "hard" buckets a priori, from substrate features alone, before any wet-lab work.

2. The Five ER Chaperone Subsystems in A. oryzae¶

Heterologous proteins targeted to the secretion pathway pass through a sequence of chaperone-mediated folding steps. Each step is mediated by a discrete molecular subsystem with finite catalytic capacity. Saturating any one subsystem creates a bottleneck that cannot be relieved by stronger promoters, more cassettes, or higher amino acid supply — only by either reducing load on that subsystem or expanding its capacity (HacA UPR, directed evolution, exogenous expression).

Subsystem	Function	Substrate features that load it	Capacity expansion lever
BiP/Kar2 (HSP70)	Generic ER lumen chaperone; binds nascent unfolded chains, prevents aggregation, hands off to specialized chaperones	All ER-targeted proteins (universal); especially proteins with extended hydrophobic stretches	HacA UPR upregulates ~3-5× (In Vitro, A. niger); BiP overexpression cassettes documented
PDI / ERO1	Disulfide bond formation and isomerization; ERO1 reoxidizes PDI cycling	Cysteine count; disulfide bond count; per-chain disulfide density	HacA UPR; PDI overexpression; directed evolution of PDI variants (Pichia precedent)
Calnexin / Calreticulin cycle	N-glycoprotein quality control; GlcNAc-trimming-driven folding cycle; UGGT re-glucosylation for re-folding	N-X-S/T sequons (X≠P); per-protein N-glycan site density	UPR induction; UGGT overexpression less characterized in koji
Lectin / Mns1 ER mannosidase	ERAD trigger after failed calnexin cycles; commits misfolded glycoproteins to degradation	Glycoproteins that fold slowly; chronic UPR substrates	Knockdown can transiently raise titer at cost of strain stress
HacA UPR (transcriptional)	Transcription factor (Hac1/HacA homolog) that responds to ER stress by upregulating BiP, PDI, ERAD machinery, lipid biosynthesis	Cumulative ER folding load (sums across all heterologous substrates)	Constitutive activated HacA (HacAⁱ) gives 2-3× heterologous titer in A. niger and T. reesei (In Vitro; Valkonen 2003; Mulder 2004)

Two additional axes that aren't ER chaperones but interact strongly with cassette stacking:

Cytosolic ribosome / chaperonin pool — only relevant for very-high-translation cassettes or cytosolic targets. Generally not limiting in koji at typical heterologous expression levels.
Vesicular traffic (COPII / ER exit sites) — emerging evidence in A. niger that ER exit site density limits secretion at high titer; not characterized in koji.

Koji-specific note: chaperone paralog redundancy. Liu 2014 (PMC4086290) genome-scale secretion analysis of A. oryzae found multiple paralogs for several core ER chaperone components (CPR1 cyclophilin, ERJ5 J-protein co-chaperone, ERD2 KDEL receptor) that are not duplicated in S. cerevisiae. (In Vitro). This suggests koji may have higher partition capacity than yeast — different paralogs may specialize on different substrate classes — though the functional independence of these paralogs has not been characterized. If functionally distinct, this is a structural advantage of koji as a multi-cassette chassis vs. yeast. Open question for direct testing.

A network-property note from Pichia: Palma/Gasser 2024 (bioRxiv 10.1101/2024.08.21.609038) found that K. phaffii Pdi1 KO is viable for disulfide-bonded recombinant protein production — backup paralogs (Eug1, Mpd1) compensate. (In Vitro). This means "PDI capacity" is a network property, not a single-gene quantity. The implication for OE: PDI competition manifests as a network-saturation effect, not a single-enzyme bottleneck, and may be relieved by paralog upregulation rather than only by direct PDI overexpression.

The framework's claim is that ER chaperone competition is the binding constraint for typical multi-cassette koji designs, with PDI / ERO1 the most commonly saturated subsystem (because mammalian-origin therapeutic proteins disproportionately carry disulfide-rich folds).

3. Substrate Categorization — Predicting Chaperone Load¶

For each candidate cassette, score five features. The combined profile predicts which chaperone subsystem(s) will dominate the burden:

Feature	What it measures	Scoring
Cysteine / disulfide count	PDI / ERO1 load	0 disulfides = none; 1-3 = light; 4-10 = moderate; 11+ = heavy
N-glycosylation site count	Calnexin / calreticulin load	0 sites = none; 1-2 = light; 3-5 = moderate; 6+ = heavy
Subunit assembly requirements	BiP retention / multi-chain coordination	Monomer = none; homo-multimer = light; hetero-multimer = heavy
Domain count / chain length	Generic BiP load; folding kinetics	<200 aa single domain = light; 200-500 = moderate; >500 or multi-domain = heavy
Subcellular target	Whether the cassette uses the secretion pathway at all	Cytosolic = bypasses entire system; secreted = full pathway; peroxisomal/mitochondrial = different machinery

The prediction: cassettes whose dominant load classes (the heaviest 1-2 features) are non-overlapping stack synergistically; cassettes whose dominant load classes overlap heavily compete for the same machinery.

Sweep A Open Question 2 resolved. The §3 bulk disulfide count treats all PDI/ERO1 substrates as fungible — "25 disulfides = 25 × per-disulfide PDI load." This is wrong. PDI residence time per disulfide varies by at least 3–4× across fold architectures because:

Compact domains with proximate, geometrically-constrained disulfides (CCP/SCR sushi fold): each disulfide pair is already pre-positioned in a ~60-aa β-sandwich; oxidation is rapid and cooperative once the scaffold is formed. PDI engagement is brief.
Large bilobal domains with many disulfides and hierarchical folding (transferrin-lobe fold): folding proceeds from a molten globule through ordered intermediate steps; only 4 of 32 cysteines initiate the folding sequence rapidly, while the remaining 28 must oxidize subsequently in a slow hierarchical cascade. PDI remains engaged for the full hierarchy. Long residence time per disulfide.
Intermediate β-sandwich domains with one disulfide buried in the hydrophobic core (Ig-like fold): some domains (CL) fold rapidly and stably without extended BiP or PDI engagement; others (VL, CH1) fold slowly and require extended chaperone contact. Heterogeneous — per-domain PDI load depends on which domain class.

3.5.1 Published Evidence for Per-Architecture Folding Kinetics¶

Architecture A — Transferrin-lobe fold (lactoferrin, transferrin family):

Notari et al. 2023 (Scientific Reports, PMC10465537, DOI:10.1038/s41598-023-41064-x) characterized oxidative folding of lactoferrin from a fully reduced molten globule state. Key findings: - LF contains 16 disulfides (32 cysteines total; LF has 16, not 17 as in some earlier OE wiki framings — see §10.2 for correction note). - Folding initiates with only 4 hyperreactive cysteines (Cys64, Cys424, Cys512, Cys668) that react with GSSG at 900–2140× enhanced rate over unperturbed protein cysteines. These 4 residues form the first disulfides and "seed" the folding hierarchy. - The remaining 28 cysteines (constituting 14 disulfides) fold subsequently in a slow cascade. Average GSSG reactivity is ~190× enhanced (i.e., substantially faster than a random thiol, but far slower than the seeding residues). Full oxidative folding requires tens of minutes to hours in vitro. - LF's "molten globule" intermediate is the dominant species across most of the folding time course — consistent with prolonged PDI engagement before the final folded state accumulates.

Cooperativity context (Wally and Buchanan 2007, Biometals, PMC2547852, DOI:10.1007/s10534-006-9024-3): the two lobes of LF are connected by a helical linker (vs. unstructured in serum transferrin) that creates tighter inter-lobe coupling. Each lobe (~330 aa, 8 disulfides) folds as a cooperative unit. The helical linker means the two lobes do not fold fully independently — there is inter-lobe cooperativity that may extend the overall folding time relative to a hypothetical uncoupled bilobal protein.

Architecture B — CCP/SCR sushi fold (DAF SCR1-4, Factor H, complement regulators):

Schmidt et al. 2010 (J Mol Biol, PMC2806952, DOI:10.1016/j.jmb.2009.10.010) provides structural and NMR dynamics data for Factor H CCP modules (canonical fH CCP12–13; fH10–15). Key findings: - Each CCP module is a compact ~60-aa β-sandwich with the canonical Cys1-Cys3 and Cys2-Cys4 disulfide pattern. These two disulfides are geometrically proximate and structurally constrained by the β-sandwich scaffold. - All 8 cysteine residues in fH12–13 are confirmed oxidized (mass spectrometry); the disulfide pattern is consistent across all CCP modules. - Individual CCP modules behave as rigid independent units in SAXS and AUC analyses — they do not undergo large-scale conformational rearrangements. Linker regions between modules are structured and not freely flexible. - NMR T₁/T₂ and ¹H,¹⁵N NOE relaxation data confirm well-ordered backbone throughout each module.

Mechanistic inference: the CCP/sushi fold's geometrically pre-organized scaffold means that once the core β-strands form, the two disulfides snap into place cooperatively with minimal isomerization requirement. PDI engagement is brief — the substrate is released quickly after disulfide formation because there are few opportunities for mispairing in a 2-disulfide system with proximate cysteines. Contrast with LF, where 16 disulfides in a molten globule require hierarchical seeding and extended PDI residence to avoid incorrect pairings among 32 cysteines. [NUMBER PARTIALLY UNVERIFIED — placeholder: the specific PDI kcat or residence time for SCR domains has not been published as a directly measured value. The fast-folding inference is structural, not kinetic, and is flagged as a bounded estimate.]

Architecture C — Ig-like fold (adalimumab HC/LC, antibody domains):

Two directly relevant papers:

Vinci et al. 2004 (J Biol Chem, PMID 14729662, DOI:10.1074/jbc.M311480200): oxidative folding of the Fc fragment (CH2 + CH3, 2 disulfides total). Without PDI, the 1-disulfide intermediate (1S2H, specifically CH3 disulfide formed first) predominates throughout — the CH2 disulfide does not accumulate without PDI. With PDI, the fully oxidized species accumulates. Conclusion: PDI is a required catalyst for complete Fc oxidation; without it, CH2 disulfide formation is kinetically trapped. "The two domains of the Fc fragment fold independently, with a precise hierarchy of disulfide formation in which the disulfide bond, especially, of the CH2 domain requires catalysis by PDI." (Vinci 2004 Abstract, verified.)

Feige et al. 2009 (Mol Cell, PMID 19524537, PMC2908990, DOI:10.1016/j.molcel.2009.04.028): the CH1 domain is "intrinsically disordered in vitro" and folds only upon interaction with the light-chain CL domain. "Structure formation proceeds via a trapped intermediate and can be accelerated by the ER-specific peptidyl-prolyl isomerase cyclophilin B." BiP "recognizes incompletely folded states of the CH1 domain and competes for binding to the CL domain." (Feige 2009 Abstract, verified.) This means CH1 — the key assembly domain — requires sustained BiP and likely PDI engagement until LC association occurs.

Hellman et al. 1999 (J Cell Biol, PMID 9885241, PMC2148116, DOI:10.1083/jcb.144.1.21): the CL domain folds "rapidly and stably even in the absence of an intradomain disulfide bond" and BiP does not detectably bind CL. But VL domain folds slowly and does show BiP engagement. So even within a single LC, domain-level folding speed is heterogeneous.

Net picture for Ig-like fold: intermediate PDI engagement. Some domains (CL) fold rapidly; others (CH1, VL, CH2 requiring PDI) fold slowly. The Ig-like fold uses PDI mechanistically (CH2 requires it), but not as extensively as the transferrin-lobe fold's 28-cysteine hierarchical cascade.

3.5.2 Per-Architecture PDI-Residence-Time Coefficients¶

Define architecture_coefficient (α) as a dimensionless multiplier on bulk disulfide count for the purpose of predicting effective PDI load. The Ig-like fold is set to 1.0 as the reference (it is the basis of the Huynh 2020 calibration — adalimumab is an Ig-fold protein). All coefficients are bounded ranges, not single values, per the framework's resolution limits.

Fold architecture	Archetype	α (PDI residence time per disulfide)	Basis
CCP / SCR / sushi	DAF SCR1-4, Factor H CCPs	0.3 – 0.6	Geometrically pre-organized 2-disulfide scaffold; brief PDI engagement; fast cooperative folding per compact module. Primary source: Schmidt 2010 (PMC2806952) structural/NMR evidence for rigid independent CCP units. In silico corroboration added 2026-05-15 (comp-030): ESM2 pseudo-pLDDT across 720 DAF SCR1-4 protein-distinct candidates is uniformly high (mean 88.8, std 0.5, range [87.6, 89.8], 100% above 80). The narrow distribution and high floor are consistent with sequence-robust, fast-folding architecture. (source: daf-cd55-scr14-cassette-ranking-computational.md). [NUMBER PARTIALLY UNVERIFIED — no published per-domain PDI kcat; bounded from structural inference.]
Ig-like	Adalimumab HC/LC constant + variable domains	0.8 – 1.2 (reference: 1.0)	PDI required for CH2 disulfide (Vinci 2004); CH1 intrinsically disordered (Feige 2009); CL folds rapidly (Hellman 1999). Heterogeneous across domain types but averages to the reference level since it is the basis of the Huynh calibration.
Transferrin-lobe	Lactoferrin N-lobe, C-lobe	1.5 – 2.5	Hierarchical folding from molten globule; only 4 of 32 cysteines seed the cascade rapidly; remaining 28 cysteines fold slowly over tens of minutes; prolonged PDI engagement required to avoid incorrect pairing among 16 disulfides. Primary source: Notari 2023 (PMC10465537); Wally & Buchanan 2007 (PMC2547852). [NUMBER PARTIALLY UNVERIFIED — no published PDI kcat for LF; coefficient derived from the 28-vs-4 cysteine kinetic hierarchy as a proxy for prolonged PDI residence.]
Serpin (metastable RCL-driven)	C1-INH (SERPING1), α1-antitrypsin (SERPINA1), antithrombin III (SERPINC1)	0.5 – 1.5 (PROVISIONAL — uncalibrated)	Low disulfide counts (C1-INH 2; α1-antitrypsin 0; antithrombin III 3) keep raw PDI load light, but the metastable native fold with a reactive-center loop undergoing dramatic β-sheet insertion may engage BiP/calnexin/PDI differently than any calibrated class — particularly because misfolded serpins polymerize and trigger UPR (α1-antitrypsin Z-variant disease anchor). Added 2026-05-19 after comp-037 substantiated C1-INH as a viable EcN-LBP payload (`c1-inh-protease-stability-ecn-computational.md`); per §8 item 6 the framework explicitly flagged "C1-INH serpin" as a fold class not covered by the calibration set. Bound is provisional pending the 2026-05-19 lit scan of α1-antitrypsin / antithrombin III / C1-INH folding kinetics (`logs/serpin-fold-alpha-coefficient-lit-scan-2026-05-19.md`); downstream koji-specific calibration requires a dedicated wet-lab measurement if C1-INH is ever routed to a secreted chassis. [NUMBER UNVERIFIED — provisional bound from structural/disulfide-count profile only.]

Effective PDI load formula (architecture-adjusted):

Effective PDI load = Σ (disulfide_count_i × α_i)

Where the sum runs over all co-expressed cassettes, each with its own architecture coefficient.

Calibration anchor (unchanged): Huynh 2020 adalimumab in NSlD-ΔP10 at 39.7 mg/L sets the reference capacity ceiling. Adalimumab has 16 disulfides (4 inter-chain + 12 intra-chain) with Ig-like fold (α ≈ 1.0), giving effective load = 16 × 1.0 = 16.0 (by definition — this is the reference point).

Critical caveat: These coefficients are derived from in vitro folding studies in mammalian or yeast expression systems, not from A. oryzae or any koji-native system. The transferrin-lobe coefficient in particular is derived from in vitro oxidative refolding from a fully denatured state (Notari 2023), which may not accurately represent ER-assisted folding in a functional secretory pathway where PDI, ERO1, and the calnexin/calreticulin cycle act co-translationally. The coefficients should be read as bounded structural/kinetic estimates pending direct koji-specific PDI residence time measurement (which does not yet exist for any of these fold classes in A. oryzae).

3.5.3 Architecture-Adjusted Effective PDI Loads for OE Cassettes¶

Cassette	Disulfides	Fold architecture	α range	Effective PDI load range
Uricase (A. flavus Q00511)	0	N/A (cytosolic fold, secreted via SP)	0	0
Lactoferrin (hLf, P02788)	16	Transferrin-lobe	1.5 – 2.5	24 – 40
DAF SCR1-4 (P08174 aa35-285)	8	CCP/SCR sushi	0.3 – 0.6	2.4 – 4.8
C1-INH (SERPING1, P05155)	2	Serpin (metastable RCL-driven)	0.5 – 1.5 (PROVISIONAL)	1.0 – 3.0
Adalimumab HC+LC (reference)	16	Ig-like	1.0	16.0 (reference)

Note on lactoferrin disulfide count correction: The OE wiki has used "17 disulfides" for lactoferrin in prior pages (including §5.5 of this document), derived from UniProt P02788 annotation. Notari 2023 (PMC10465537) explicitly states "16 disulfides" for human lactoferrin. Wally & Buchanan 2007 confirms LF has fewer disulfides than serum transferrin (hTF has 3 additional disulfide bonds not present in LF). UniProt P02788 annotation shows 35 cysteines total forming 16–17 disulfides depending on whether one counts the inferred or explicitly annotated bonds. For this architecture-adjusted model, use 16 disulfides for LF (Notari 2023 source-verified; §10.2 below documents the discrepancy). The effective PDI load computation is recalculated accordingly.

3.5.4 Calibration-set candidate: §1.9 (lactoferrin, transferrin-lobe) + §1.25 (DAF SCR1-4, CCP/SCR) (added 2026-05-08)¶

The α coefficients in §3.5.2 are derived from mammalian / non-koji folding studies; §3.5.3's effective-PDI-load predictions are therefore bounded estimates pending direct koji-specific calibration. The two highest-leverage planned wet-lab experiments — validation-experiments.md §1.9 (Ward 1995 dual-cassette gate, lactoferrin = transferrin-lobe) and validation-experiments.md §1.25 (DAF SCR1-4 single-cassette gate, CCP/SCR sushi fold) — together span the framework's two architecturally-extreme α classes (CCP/SCR 0.3–0.6 vs. transferrin-lobe 1.5–2.5), and their measured per-cassette titers would be a natural calibration set for the framework if the two experiments are run under harmonized conditions.

Pre-registered calibration framework (conditional):

The §1.9 + §1.25 pairing functions as a calibration set ONLY if the following experimental conditions are harmonized:

Variable	Required harmonization	Current §1.9 design	Current §1.25 design
Host strain	Single host (NSlD-ΔP10 preferred)	NSlD-ΔP10 (mandated)	RIB40 default; NSlD-ΔP10 listed as optional arm
Format	Solid-state shio-koji OR submerged — pick one	Solid-state shio-koji	Mixed (solid-state primary; submerged optional)
Promoter	Same promoter strength (e.g., glaA or PamyB)	glaA secretion cassette	Per §1.25 — confirm
Titer units	mg/L mature protein, ELISA-quantified	mg/L (ELISA)	mg/L (ELISA)

If §1.25 is run with the optional NSlD-ΔP10 arm under the same solid-state format and matching promoter, the two experiments produce a paired calibration measurement. Otherwise the comparison is confounded by host/format effects and cannot recalibrate α.

Pre-registered framework predictions (bounded):

§1.9 lactoferrin-alone arm (transferrin-lobe, α 1.5–2.5, 16 disulfides, effective PDI load 24–40): predicted ≥500 mg/L if the framework's α coefficients transfer to koji; <40 mg/L if PDI-saturation dominates (lower bound from §5.5 worst case).
§1.25 DAF SCR1-4 arm (CCP/SCR, α 0.3–0.6, 8 disulfides, effective PDI load 2.4–4.8): predicted ≥100 mg/L under the framework — substantially higher per-cassette titer than lactoferrin because of the lower α coefficient.

Decision rule: if the observed titer ranking matches the framework's prediction (DAF SCR1-4 > lactoferrin per cassette, on a normalized basis), the per-architecture α coefficients have transferable predictive power to koji and the framework can be used prospectively for new cassette design. If the ranking inverts (lactoferrin > DAF SCR1-4 per cassette), the α coefficients calibrated from non-koji data do not transfer — the framework needs revision toward koji-specific kinetics, and predictions for novel cassettes (e.g., the C1-INH thread surfaced in comp-018) would need to be flagged as uncalibrated.

Cross-references: validation-experiments.md §1.9, validation-experiments.md §1.25, synthesis/strategic-reflections/ (chaperone-framework predictive-power validation entry — this calibration set is the operational instantiation of that reflection's trigger condition).

4. Open Enzyme Candidate Cassettes — Scored¶

Cassette	Origin	Disulfides	N-glyc sites	Assembly	Length	Target	Dominant load class	Fold architecture	α (§3.5)	Effective PDI load
Uricase (uaZ, A. flavus, Q00511)	Fungal	0	0 (annotated; 1 sequon predicted but unoccupied per comp-010 §5.5)	Homotetramer	302 aa	Cytosolic native; secreted via amyB SP	BiP-only (transit) — chaperone-light	N/A (no disulfides)	0	0
Uricase variant — C. utilis (URIC_CYBJA, P78609; per comp-011 §4.2)	Yeast	0 disulfides + 4 free Cys (positions 39, 168, 250, 293)	1 predicted (NSS pos 54, not annotated)	Homotetramer	302 aa	Secreted via amyB SP	PDI-engaged via free-Cys load — variant choice matters; magnitude unknown (comp-011 §4.4)	N/A (free Cys; no organized fold disulfides)	~0.3–0.5 estimated for free Cys engagement	~1.2–2.0 estimated
Lactoferrin (hLf, P02788)	Mammalian	16 (one chain; per Notari 2023 PMC10465537; see §3.5.3 + §10.2)	3 confirmed	Monomer	691 aa	Secreted	PDI-heavy + calnexin-moderate + BiP-heavy	Transferrin-lobe (bilobal, ~330 aa/lobe, helical linker, inter-lobe cooperativity)	1.5 – 2.5	24 – 40
Native lipase (e.g., A. oryzae TGL)	Fungal native	3	0-1	Monomer	~270 aa	Secreted	PDI-light — well-adapted	Lipase / α/β hydrolase (fungal native)	~0.7–1.0 (adapted)	2.1 – 3.0
Native α-amylase (TAKA-amylase)	Fungal native	1	4-5	Monomer	478 aa	Secreted	Calnexin-moderate + PDI-light	TIM barrel / amylase fold	~0.8–1.0	0.8 – 1.0
Native acid protease	Fungal native	2-3	Variable	Monomer	~330 aa	Secreted	PDI-light + calnexin-light	Pepsin-like aspartyl protease	~0.7–1.0	1.4 – 3.0
DAF / CD55 SCR1-4 (truncated; per comp-012)	Mammalian	8 (4 SCRs × 2 disulfides each; per UniProt P08174 DISULFID features, verified 2026-05-06)	0 (stalk truncation removed sites)	Monomer	~250 aa	Secreted	PDI-moderate — small SCR domains fold fast	CCP / SCR / sushi (4 compact ~60-aa β-sandwich modules, 2 pre-organized disulfides each)	0.3 – 0.6	2.4 – 4.8
Carnosine synthase (Lactobacillus CarnS)	Bacterial	0	0	Monomer	~460 aa	Cytosolic	Bypasses secretion entirely	N/A (cytosolic)	0	0
panD aspartate decarboxylase (β-alanine supply)	Bacterial	0	0	Homotetramer	~140 aa	Cytosolic	Bypasses secretion entirely	N/A (cytosolic)	0	0
Kojic acid pathway (native)	Fungal native	n/a (small-molecule)	n/a	n/a	n/a	Cytosolic biosynthesis	Off the protein machinery axis entirely	N/A	0	0
Ergothioneine pathway (native)	Fungal native	n/a	n/a	n/a	n/a	Cytosolic biosynthesis	Off the protein machinery axis entirely	N/A	0	0
Cordycepin biosynthesis (cns1+cns2; per Jeennor 2023 PMID 38071331)	Bacterial (heterologous in A. oryzae)	0	0	Cytosolic enzymes	~390 + ~270 aa	Cytosolic biosynthesis (small-molecule product secreted)	Bypasses secretion entirely — small-molecule product	N/A (cytosolic)	0	0

Cross-chassis combinations are trivially off-framework. Compounds produced by a separate organism (e.g., cultivated Cordyceps militaris cordycepin, Ganoderma lucidum GLPP, Pleurotus citrinopileatus ergothioneine — see medicinal-mushroom-complement-track.md) impose zero chaperone load on the koji endgame strain because they are not co-expressed in koji at all. They appear in the consumer product as cultivated mushroom extracts paired with the engineered koji. No framework needed for that combination — the orthogonality is by virtue of separate chassis. The cordycepin row above scores the alternative koji-engineering route (Jeennor 2023's cns1+cns2 heterologous expression in A. oryzae); even that route is off the secretion/folding axis because cns1 and cns2 are bacterial cytosolic enzymes that never enter the ER.

Scope note — what this framework tracks and what it doesn't. The framework scores ER folding-machinery competition (PDI for disulfides, calnexin for N-glycans, BiP for general transit). It does not track generic proteome burden (ribosomes, amino acid pool, ATP for tRNA charging) imposed by cytosolic enzymes. For the current cassette stack, ER folding lanes are the binding constraint — koji's wild-type translation budget already supports ~30% secretion-dedicated output, so adding a ~400 aa cytosolic enzyme like cns1 is small change against the existing translation pool. If the platform later stacks multiple cytosolic biosynthesis clusters (e.g., cordycepin + a terpenoid pathway + a polyketide cluster), translation burden may become load-bearing — at that point a proteome-constrained metabolic model (pcSecKoji-style) is the correct upgrade per synthesis/strategic-reflections/ §"Chaperone framework predictive-power validation".

Update (2026-05-14, comp-023): the cytosolic-burden gap named in this scope note is now closed for the cns1+cns2 cordycepin row specifically by cordycepin-cassette-burden-computational.md, which ran plain FBA on the Vongsangnak 2008 iWV1314 GEM with dual + cns1-cns2 + carnS + panD scenarios. Verdict: GREEN at the Jeennor 2023 empirical titer (564 mg/L/d); the FBA breakpoint where cordycepin demand materially competes with growth is ~1,000× the empirical titer (~564 g/L/d). The two analyses cover orthogonal axes: this framework covers ER folding, comp-023 covers cytosolic carbon + nitrogen + ATP + NADPH burden. Both verdicts must hold for the full endgame strain stack to be feasible. pcSec-class proteome-saturation modeling remains the v2 upgrade target. (source: cordycepin-cassette-burden-computational.md)

The pattern is striking: the Open Enzyme koji endgame strain is implicitly already a chaperone-orthogonal design. The five "molecules" of koji-endgame-strain.md §1 distribute across non-overlapping load classes —

Uricase: BiP-transit only (zero disulfide, zero glycosylation, zero subunit assembly) — this orthogonality is variant-specific to A. flavus (Q00511); the C. utilis variant (P78609) carries 4 free Cys that engage PDI/ERO1, partially collapsing the orthogonality. See comp-011 §4.4 for the bidirectional analysis. The §1.9 parallel head-to-head is the empirical resolver.
Lactoferrin: PDI + calnexin + BiP — carries essentially the entire ER chaperone load on its own
Native digestives (lipase, protease, amylase): light, well-adapted, host-evolved
Carnosine + panD: cytosolic — completely off the secretion pathway
Kojic acid + ergothioneine: small-molecule biosynthesis — off the protein machinery entirely

This was not an explicit design choice — it emerged from each cassette being chosen for therapeutic mechanism, with the chaperone-orthogonality being a happy accident. Making the framework explicit gives a prospective tool for the next round of cassette additions: any new candidate should be scored against the existing four load classes and preferentially fill underused ones.

5. Predicted Synergy Coefficients¶

Define a synergy coefficient for cassette pair (A, B):

$$ \text{Synergy}(A, B) = \frac{\text{Yield}(A | A+B) + \text{Yield}(B | A+B)}{\text{Yield}(A | A \text{ alone}) + \text{Yield}(B | B \text{ alone})} $$

Where Yield(X | A+B) is the per-cassette yield of X when co-expressed with B, and Yield(X | X alone) is the single-cassette baseline. Interpretation:

Synergy ≈ 1.0 — additive; cassettes don't interact (orthogonal)
Synergy < 0.7 — antagonistic; cassettes compete for shared machinery
Synergy > 1.0 — supra-additive; cassettes positively interact (rare; e.g., one cassette upregulates UPR which helps the other)

Predicted synergy matrix for OE candidate pairs — now showing both bulk-count prediction and architecture-adjusted prediction (Mechanistic Extrapolation; ranges reflect uncertainty):

Pair	Bulk-count prediction	Architecture-adjusted prediction	Shift	Rationale for shift
Uricase × Lactoferrin	0.8 – 1.0	0.85 – 1.0	↑ slight (upper bound rises)	Uricase: 0 effective load. Lf: effective load 24–40 (vs. bulk 16–17). But the ratio to Huynh baseline (16 × 1.0 = 16) is what determines saturation risk: Lf alone at 16 disulfides × 1.5–2.5 α = 24–40 effective load vs. 16 reference → 1.5–2.5× Huynh. Paradoxically, this makes Lf-alone the capacity question, not the Lf × uricase pair. Uricase contributes 0, so the pair synergy is driven entirely by whether Lf alone exceeds PDI capacity — not by pairwise competition. Revised range modestly upward for upper bound: if §1.9 Lf-alone arm shows Lf secretes well despite its high architecture-adjusted load, the dual-cassette is additive.
Uricase × DAF SCR1-4	0.85 – 1.0	0.90 – 1.0	↑ slight	DAF SCR1-4: effective load 2.4–4.8 (architecture-adjusted), far lower than bulk count of 8 implies. Uricase: 0. This pair is even lighter than the bulk model suggested. Near-complete orthogonality.
Uricase × Carnosine cassette	~1.0	~1.0	No change	Completely orthogonal — uricase secreted, carnosine cytosolic. No shared machinery. Architecture coefficient doesn't apply.
Lactoferrin × DAF SCR1-4	0.5 – 0.7	0.40 – 0.65	↓ (worse; lower bound falls)	This is the key pair. Architecture-adjusted: Lf effective load = 24–40; DAF effective load = 2.4–4.8. Combined: 26.4–44.8 vs. Huynh reference 16.0 → 1.65–2.80× Huynh baseline. The bulk-count model gave 25/16 = 1.56×. The architecture-adjusted model gives a wider range: upper end is worse (Lf's transferrin-lobe α = 2.5 pushes combined effective load to 44.8, or 2.80× Huynh — substantially above the demonstrated capacity ceiling). Lower end (Lf α = 1.5, DAF α = 0.3) gives 26.4/16 = 1.65× — still above the Huynh ceiling but closer. The architecture refinement increases the lower bound of concern (from 1.56× to 1.65×) and significantly increases the upper bound (from 1.56× to 2.80×). The pairwise synergy range accordingly shifts from 0.5–0.7 to 0.40–0.65, with the lower bound driven by the worst-case Lf coefficient.
Lactoferrin × Native α-amylase (excess)	0.6 – 0.8	0.65 – 0.85	↑ slight	Both load calnexin (Lf 3 N-glycan sites; amylase 4-5). Architecture doesn't change the calnexin competition picture significantly. Slight upward shift for upper bound because amylase's PDI effective load is small (1 disulfide × ~0.9 α = 0.9).
Two PDI-heavy mammalian glycoproteins (e.g., Lf + Fab heavy chain)	0.5 – 0.6	0.35 – 0.55	↓ (worse)	If Lf's effective load is 24–40 and a Fab HC has 3 intra-chain disulfides × Ig-like α ~1.0 = 3.0, combined = 27–43 → ~1.7–2.7× Huynh. Architecture adjustment confirms this is the worst-case regime.
Three light secreted enzymes (e.g., 3 cellulases per Wakai 2019)	≥1.0 (provisional — see caveat)	≥1.0 (unchanged)	No change	All three have fungal-native folds with adapted coefficients near 0.7–1.0. Architecture adjustment does not change the prediction for this class. Verification caveat unchanged: Wakai 2019 reports 40× total cellulolytic activity; this is copy-scaling + promoter optimization, not a pairwise synergy measurement.
Endgame strain (uricase + Lf + native digestives + carnosine)	~0.85 weighted	~0.75–0.90 weighted	↓ slight (range widens)	Lf's architecture-adjusted effective load (24–40) is higher than its bulk count (16–17) implied. Native digestives remain light. Uricase and carnosine remain at 0. The endgame strain's chaperone burden is dominated by Lf's transferrin-lobe architecture — which the bulk model underestimated. Key insight: the endgame strain without DAF SCR1-4 (i.e., the dual-cassette: uricase + Lf) is the tractable design — the whole Lf effective load question must be resolved by §1.9 first.

The Open Enzyme endgame strain's predicted weighted synergy ~0.85 — close to additive — is the framework's quantitative reframe of why the "five molecules in one strain" thesis is biochemically reasonable. It is not five-fold burden; it is roughly Huynh-2020-equivalent burden distributed across more deliverables.

5.5 Triple-Cassette Prospective Prediction — Uricase + Lactoferrin + DAF SCR1-4¶

This is Sweep A Proposed Experiment 1 (2026-05-05, commit 3e928d3): the first real prospective test of whether the chaperone-orthogonal framework has predictive power beyond retrofit.

The question is load-bearing for the platform: Can the koji endgame strain be a single triple-cassette strain (uricase + lactoferrin + DAF SCR1-4) that closes CP0 (complement-priming chokepoint, per H05) as well as the existing CP1-CP6 coverage from uricase + lactoferrin? Or does chaperone-load competition from adding DAF SCR1-4 force it onto a separate strain or onto the engineered LBP chassis as a parallel peer track?

All predictions in this section are Mechanistic Extrapolation anchored to the verified source numbers in §10.1. No new unverified quantitative claims are introduced.

5.5.1 Setup — Cassettes, Disulfide Totals, and Baseline¶

The three cassettes:

Cassette	Gene	Disulfides	N-glyc sites	Dominant load class
Uricase	A. flavus uaZ (Q00511)	0	0 (unoccupied)	BiP-transit only — chaperone-light
Lactoferrin	hLf (P02788)	17 (one chain)	3 confirmed	PDI-heavy + calnexin-moderate
DAF SCR1-4	P08174 aa 35–285	8 (2 per SCR × 4 SCRs, per UniProt P08174 DISULFID features, verified 2026-05-06; see comp-012 §1.5)	0 (stalk truncation removed glycosylation sites)	PDI-moderate

Triple-cassette disulfide total: 25 (0 + 17 + 8). This is 1.56× the Huynh 2020 adalimumab baseline of 16 disulfides (4 inter-chain + 12 intra-chain; verified via UniProt P01857 + P01834; see §10.1). Compare to the dual-cassette (comp-010) of 17 disulfides = 1.06× the Huynh baseline. Adding DAF SCR1-4 pushes the triple 47% above the dual on the PDI-load axis, and 56% above the demonstrated capacity ceiling if the Huynh ceiling is architecture-independent.

Calibration context: The §1.9 wet-lab experiment has a Lf-alone arm that directly resolves the capacity-vs-titer benchmark ambiguity flagged in §8 item 7 and koji-endgame-strain.md §3.3. If Lf alone reaches >500 mg/L in NSlD-ΔP10 solid-state, the Huynh ceiling is antibody-architecture-specific (dual-chain assembly difficulty, not single-chain disulfide count), and the framework's synergy coefficients are systematically conservative. Until that data lands, both scenarios must bound the prediction.

5.5.2 Per-Pair Synergy Recap (With Corrected DAF Disulfide Count)¶

Three pairwise synergies from §5, applied to the corrected 8-disulfide DAF count:

Pair	§5 predicted synergy	Note with corrected 8-disulfide DAF
Uricase × Lactoferrin	0.8–1.0 (bulk) → 0.85–1.0 (arch-adjusted)	Architecture refinement: Lf's effective PDI load is 24–40 (vs. bulk 16), but since uricase contributes 0, this pair's synergy is determined by whether Lf alone can be secreted — not by pairwise competition. Architecture adjustment does not worsen this pair; slight upper bound improvement because the key question is now "can Lf fold at all in this host?" rather than "does Lf compete with uricase?"
Uricase × DAF SCR1-4	0.85–1.0 (bulk) → 0.90–1.0 (arch-adjusted)	DAF SCR1-4's effective load drops from bulk 8 to arch-adjusted 2.4–4.8. Even lighter than the bulk model implied. Near-complete orthogonality with uricase.
Lactoferrin × DAF SCR1-4	0.5–0.7 (bulk) → 0.40–0.65 (arch-adjusted)	Architecture refinement worsens this pair significantly. Combined effective PDI load: Lf (24–40) + DAF (2.4–4.8) = 26.4–44.8. Ratio to Huynh reference (16.0): 1.65–2.80×. The bulk model gave 25/16 = 1.56× — underestimated by 6–80% depending on the Lf α value. The transferrin-lobe slow-folding architecture is the dominant adverse factor. Revised pairwise range: 0.40–0.65.

The Lf × DAF SCR1-4 pair is the dominant constraint on the triple, and the architecture refinement makes it worse than the bulk model predicted. The critical new insight: it is Lf's slow transferrin-lobe folding (not just its raw disulfide count) that creates the PDI saturation risk. DAF's CCP/SCR architecture actually mitigates its contribution substantially (8 bulk disulfides → 2.4–4.8 effective load). The architecture-adjusted bottleneck is almost entirely on the Lf side.

5.5.3 Triple-Cassette Composition Rule — Two Bounding Models (Architecture-Adjusted)¶

Pairwise synergy coefficients do not compose trivially for a triple. Two models bound the prediction. All numbers below use the architecture-adjusted pairwise ranges.

Model A — Multiplicative pairwise (floor estimate). Treat each pair's synergy as an independent multiplier. Architecture-adjusted pairs: Uricase × Lf (midpoint ~0.93, range 0.85–1.0), Uricase × DAF (midpoint ~0.95, range 0.90–1.0), Lf × DAF (midpoint ~0.525, range 0.40–0.65). Naive multiplication: 0.93 × 0.95 × 0.525 ≈ 0.46. This is a floor because it assumes independent pairwise penalties — in a saturated system, the third cassette may find the PDI pool already depleted and contribute near-zero additional marginal penalty. Therefore 0.46 is a conservative lower bound.

Model B — Chaperone-saturation (upper estimate). If the Lf × DAF pair already saturates PDI/ERO1, adding uricase (PDI-orthogonal, effective load 0) does not worsen the saturated subsystem. Triple synergy converges to the worst pairwise synergy (Lf × DAF, upper bound 0.65). Upper bound for the triple: ~0.65.

New architecture-specific insight for the upper bound: Because DAF's CCP/SCR architecture gives it an effective load of only 2.4–4.8 (not 8 bulk), its marginal contribution to PDI saturation above the Lf-alone baseline is small. The key question is whether Lf alone (effective load 24–40) already saturates the PDI/ERO1 pool — if yes, adding DAF's 2.4–4.8 additional effective load is nearly inconsequential. This is a more nuanced picture than the bulk model: the dominant bottleneck is Lf's transferrin-lobe architecture, and DAF's contribution to it is relatively small.

Neither model is empirically validated in koji. The Wakai 2019 verification (§10.1) confirmed that three light-substrate cellulases stack without titer collapse — but all three are fungal-native with negligible disulfide load. There is no published koji-specific data point for a two-PDI-heavy-substrate stacking scenario. The models below should be read with this absence explicitly in mind.

5.5.4 Predicted Synergy Range — Revised to 0.35–0.65 (from prior 0.45–0.70)¶

Architecture-adjusted lower bound: ~0.35 (multiplicative pairwise model; worst-case Lf α = 2.5; DAF α = 0.3; no UPR rescue; Huynh ceiling is architecture-independent).

Calculation: Lf effective load = 16 × 2.5 = 40; DAF effective load = 8 × 0.3 = 2.4; combined = 42.4; ratio to Huynh reference 16.0 = 2.65×. Translating to synergy using multiplicative pairwise model: Lf × DAF synergy floor under worst-case loads ≈ 0.40; multiplicative composition (× 0.93 × 0.95) → 0.35. This is a hard floor assuming no architectural rescue.

Architecture-adjusted upper bound: ~0.65 (chaperone-saturation model; best-case Lf α = 1.5; DAF α = 0.6; Huynh ceiling is antibody-specific; single PDI overexpression rescue).

Calculation: Lf effective load = 16 × 1.5 = 24; DAF effective load = 8 × 0.6 = 4.8; combined = 28.8; ratio to Huynh reference 16.0 = 1.80×. At this ratio, the Lf × DAF pairwise synergy is at its upper bound (0.65). Saturation model converges to this for the triple.

Named uncertainty sources for the lower bound: - Whether the Lf transferrin-lobe α coefficient (1.5–2.5) is calibrated correctly — the Notari 2023 kinetics are in vitro from fully denatured state; in vivo ER-assisted folding may be faster due to co-translational PDI engagement - Whether the multiplicative pairwise composition rule holds in a saturated system - Whether the Huynh ceiling applies to single-chain substrates

Named uncertainty sources for the upper bound: - Whether the §1.9 Lf-alone result resolves the Huynh ambiguity favorably (highest leverage single experiment) - Whether the transferrin-lobe α best case (1.5) is achievable given the 28-cysteine hierarchical cascade documented by Notari 2023 - Whether NSlD-ΔP10 background provides indirect PDI-equivalent rescue via protease-deletion

Revised central expectation: 0.45–0.55. This lands at the boundary of the 0.6–0.85 gate (augmentation feasible) and the <0.6 gate (separate-strain routing), shifted downward from the prior central expectation of 0.55–0.65. The architecture refinement moves the predicted central value from the "medium probability augmentation-feasible" region toward the "high probability separate-strain routing" region.

Prior range (bulk model): 0.45–0.70, central 0.55–0.65. Revised range (architecture-adjusted): 0.35–0.65, central 0.45–0.55.

The key driver of this downward shift: the transferrin-lobe architecture coefficient for Lf (1.5–2.5) means Lf's effective PDI load is substantially higher than its raw disulfide count implies. DAF's CCP architecture coefficient (0.3–0.6) partially offsets the combined load, but not enough to rescue the Lf × DAF pairwise synergy back to the bulk-model range.

5.5.5 Decision Threshold Gates (Updated)¶

Important disclaimer: These thresholds are framework-convention, not derived from empirical data. They should be read as "if the wet-lab measurement of the triple-cassette strain lands in this region, this is the recommended decision direction" — not as bright-line rules with mechanistic derivation.

Measured triple-cassette synergy	Interpretation	Recommended platform direction
>0.85	Triple stacks cleanly; PDI subsystem not saturated.	Pursue triple-cassette endgame strain as the single-strain CP0+CP1-CP6 solution. Assessed probability: very low (architecture refinement further reduces this probability vs. bulk model — requires Lf to have an exceptionally favorable in vivo folding rate relative to its in vitro kinetics, AND NSlD-ΔP10 to have unusually high single-chain PDI capacity).
0.6–0.85	Feasible with chaperone-helper augmentation.	Pursue triple-cassette with a PDI co-expression cassette. Assessed probability: low–medium (architecture refinement shifts the central expectation below 0.6 — this gate is reachable only if Lf's transferrin-lobe α is at the favorable end (1.5) AND §1.9 Lf-alone confirms the Huynh ceiling is antibody-specific).
<0.6	DAF cassette goes on a separate strain or the LBP chassis as a peer track. The CP0 closure thesis (H05) stays alive — only the chassis-route changes.	Maintain uricase + Lf dual-cassette as the endgame strain (CP1-CP6 coverage); route DAF SCR1-4 separately. Assessed probability: high — architecture refinement increases this probability relative to the bulk model. The central expectation now lands at 0.45–0.55, squarely in this gate.

5.5.6 Dominant Predicted Bottleneck (Architecture-Refined)¶

Lactoferrin's transferrin-lobe slow-folding architecture is the dominant bottleneck — not the raw disulfide count of the combined cassettes. This is the key insight from the architecture refinement.

Rationale: - Lactoferrin's effective PDI load (24–40) is 1.5–2.5× Huynh reference even in the single-cassette case (Lf alone). Whether the host can fold Lf at all is the threshold question before any pairwise competition matters. - DAF SCR1-4's effective PDI load (2.4–4.8) is lower than its bulk disulfide count (8) implies. The CCP/sushi fold's pre-organized 2-disulfide modules fold rapidly and release PDI quickly. DAF's marginal contribution to PDI saturation above the Lf-alone baseline is small. - Uricase contributes exactly zero to the PDI bottleneck. - Implication for platform strategy: The bottleneck is not "too many combined disulfides" — it is "Lf's folding is intrinsically PDI-expensive." Relieving this bottleneck requires either (a) host engineering that increases PDI/ERO1 capacity specifically for the transferrin-lobe substrate class, or (b) accepting that the Huynh ceiling is antibody-architecture-specific and that single-chain transferrin-lobe substrates are handled more efficiently per disulfide than the bulk model assumed. - The §1.9 Lf-alone arm is the highest-leverage single data point: it directly tests whether Lf's architecture-adjusted effective load of 24–40 is within NSlD-ΔP10's PDI capacity for single-chain substrates. If Lf alone reaches >500 mg/L, the Lf α coefficient in the 1.5 range is applicable (favorable); if Lf stalls at ~40 mg/L, the α coefficient is in the 2.0–2.5 range (unfavorable) and the framework's prediction shifts toward the floor.

Cross-reference §8 item 7 (capacity-vs-titer benchmark ambiguity): The architecture refinement makes this ambiguity load-bearing. If the §1.9 Lf-alone arm demonstrates >500 mg/L Lf in NSlD-ΔP10 solid-state, this indicates either (a) the Huynh ceiling is antibody-architecture-specific, or (b) the transferrin-lobe α at 1.5 applies for in vivo ER-assisted folding (slower in vitro folding does not translate directly to slower in vivo PDI engagement). Either resolution is favorable. If Lf stalls at ~40 mg/L, the transferrin-lobe α must be at the upper end (2.0–2.5) and the framework's prediction floor for the triple-cassette is at or below 0.35.

5.5.7 Prospective Falsifiable Prediction¶

Framework prediction (architecture-adjusted, 2026-05-06): triple-cassette (uricase + Lf + DAF SCR1-4) synergy falls in the range 0.35–0.65 (central expectation 0.45–0.55), with Lf's transferrin-lobe slow-folding architecture as the dominant PDI bottleneck. This is a downward revision from the prior bulk-count prediction of 0.45–0.70 (central 0.55–0.65). The revision is driven by the Lf α = 1.5–2.5 coefficient, which increases Lf's effective PDI load beyond what raw disulfide count implied; DAF's CCP α = 0.3–0.6 partially offsets the combined load but does not rescue the pairwise Lf × DAF synergy range.

The falsifiable test: Wet-lab measurement of the triple-cassette strain's Lf titer (or DAF SCR1-4 titer) in NSlD-ΔP10 solid-state, compared to the dual-cassette (uricase + Lf) baseline measured in the same §1.9 experiment, directly tests this prediction. The synergy coefficient is:

$$ \text{Synergy}_{\text{triple}} = \frac{\text{Lf titer (triple-cassette strain)} + \text{DAF titer (triple-cassette strain)} + \text{Uricase activity (triple-cassette strain)}}{\text{Lf titer (dual-cassette alone)} + \text{DAF titer (DAF alone)} + \text{Uricase activity (uricase alone)}} $$

In practice, measuring Lf titer alone in the triple vs. dual-cassette context gives the cleanest single-protein readout: if Lf titer in the triple-cassette strain is >85% of the dual-cassette baseline, the triple stacks cleanly (synergy >0.85 regime). If Lf titer is 60-85% of dual-cassette baseline, augmentation is warranted. If Lf titer drops below 60% of the dual-cassette baseline, separate-strain routing is the framework-recommended direction.

This is the first real prospective test of whether the chaperone-orthogonal framework has predictive power beyond retrofit. All existing synergy coefficients were derived to explain known outcomes (Huynh 2020, Wakai 2019, comp-010). The triple-cassette prediction is the first case where the framework commits to a specific outcome before wet-lab data exists. Outcome range 0.45–0.70 is a falsifiable prediction; any measured value outside this range (particularly synergy >0.85) would require framework revision.

5.5.8 Platform Decision Tree — What This Changes¶

Scenario	Probability	Platform direction
Synergy >0.85 (triple stacks cleanly)	Low	Pursue triple-cassette endgame strain as CP0+CP1-CP6 single-strain solution. Four cassette slots free in NSAR1 5-marker platform for downstream additions (kojA/kojR overexpression, carnosine module).
Synergy 0.6–0.85 (feasible with helper augmentation)	Medium	Pursue triple-cassette with single-PDI-overexpression cassette (4-cassette total). Characterize bottleneck (PDI vs. ERO1 vs. BiP) before adding helper, per §7 order-of-operations. NSlD-ΔP10's protease-deletion background may provide partial rescue without a dedicated helper cassette.
Synergy <0.6 (separate-strain routing)	High	Maintain uricase + Lf dual-cassette as the endgame strain. Route DAF SCR1-4 to either: (a) a dedicated second A. oryzae strain co-fermented with the dual-cassette strain, OR (b) the engineered LBP chassis (engineered-lbp-chassis.md) as a parallel peer track. H05 thesis stays alive — only the chassis-route changes. The endgame strain stays at CP1-CP6 coverage (five chokepoints, four molecules per koji-endgame-strain.md §1); CP0 direct-antagonism coverage becomes a separate-strain or peer-track output.

Key point in all three scenarios: the H05 CP0-closure thesis is not killed by a low triple-cassette synergy — it only changes where the DAF SCR1-4 cassette lives. The platform's five-chokepoint endgame strain design is preserved regardless of which gate the triple-cassette measurement lands in.

Dependency on §1.9 experiment arms: The §1.9 wet-lab design now has three cassette-configuration arms relevant to chaperone-load resolution: 1. Lf-alone arm — resolves the capacity-vs-titer ambiguity (§8 item 7). Required before interpreting any dual or triple result. 2. Uricase + Lf dual-cassette arm — validates the comp-010 prediction (1.06× Huynh baseline, within capacity). Required for interpreting the triple. 3. Uricase + Lf + DAF SCR1-4 triple-cassette arm — the prospective falsifiable test of this §5.5 prediction. Optional for the §1.9 gate decision (the dual-cassette arm is sufficient to decide whether to proceed with the endgame strain); adds ~$1-2k and 2-3 weeks but provides the framework's first prospective validation data point.

5.6 Design escape — cytosolic third cassette bypasses the secreted-stacking bottleneck (added 2026-05-15)¶

§5.5 above analyses a triple-cassette strain where all three payloads are secreted — uricase + lactoferrin + DAF SCR1-4. That analysis predicts the triple pushes PDI/ERO1 chaperone load into the sub-0.6 synergy regime (0.35–0.65 range, §5.5.4) and recommends routing DAF onto a separate strain or peer-track chassis. A natural follow-up question, surfaced by the 2026-05-14 sweep Connection 1: can the third cassette be cytosolic instead of secreted, escaping the chaperone bottleneck entirely?

comp-023 (cordycepin cassette burden FBA, verdict GREEN 2026-05-14) demonstrated that the Cordyceps militaris cns1+cns2 cordycepin pathway is cytosolic in A. oryzae: no signal peptide, no ER transit, no disulfide bonds, no PDI load, and < 1% metabolic burden at the Jeennor 2023 564 mg/L/day titer. This means a triple-cassette strain of:

Uricase (secreted, PDI load 0 after N191Q glycosylation ablation per comp-022)
Lactoferrin (secreted, PDI load 24–40 range per §3.5.3 architecture-adjusted model)
Cordycepin (cytosolic, PDI load 0)

… has a total PDI/ERO1 load of 24–40 (range) — identical to the dual-cassette uricase + lactoferrin baseline. The third cassette adds therapeutic mechanism breadth (cordycepin as a URAT1-modulating + ADA-inhibiting payload — see cordycepin-cassette-burden-computational.md for the mechanism rationale) without adding to the chaperone load that drives §5.5's pessimistic synergy prediction.

What §5.6 does NOT claim (Pass 3 softening discipline)¶

The 2026-05-14 sweep Connection 1 Pass 3 review flagged two specific overclaims to avoid:

"Within demonstrated NSlD-ΔP10 capacity" is NOT source-supported. The Lf-alone capacity question against the Huynh 16-disulfide reference is unresolved in the current corpus (§3.5.4 explicitly flags this — Lf-alone in §1.9 will produce the calibration data). Until §1.9 returns, treat the PDI load 24–40 range as the budget the design must fit within, not as a demonstrated-capacity assertion. Adding the cytosolic third cassette doesn't add to the budget; it doesn't, however, prove the budget is achievable.
"≥0.85 triple-synergy" is NOT source-supported. §5.5's synergy framework (0.35–0.65 range under architecture-adjusted modeling) was derived for secreted-only payloads. The cytosolic-third-cassette case isn't a continuous extension of that framework — it's a structurally different decomposition (the chaperone load is dual-cassette plus a metabolic-only third), and the synergy question for this case is an outcome of comp-028, not an input.

The design-escape framing is therefore: "cytosolic third cassettes are promising because they don't add chaperone load," not "the triple is proven to work."

Comp-028 as the feasibility gate¶

comp-028 (Planned Analyses, queued 2026-05-15) is the explicit computational pass that resolves the open question. It requires two orthogonal models (per the BioDesignBench evaluation-depth discipline at etc/autonomous-screening-methodology.md §"BioDesignBench evaluation-depth audit"):

The chaperone framework with explicit confidence bounds (this section, with Pass 3 softening applied)
iWV1314 FBA with the dual-cassette uricase + lactoferrin metabolic baseline overlaid on the cns1+cns2 cordycepin demand (not just cns1+cns2 alone as comp-023 ran)

Decision rule: GREEN iff both axes pass with explicit bounds; YELLOW if one axis is uncertain (route through §1.9 Lf-alone readout first); RED if either fails. If GREEN: add cordycepin arm to validation-experiments.md §1.9 wet-lab design.

Generalization — cytosolic payloads as a strategic design lever¶

The cytosolic-third-cassette pattern generalizes beyond cordycepin. Future OE design questions involving a third (or fourth) payload should explicitly check secreted vs. cytosolic:

Cytosolic payloads (cordycepin, kojic acid, ergothioneine native to A. oryzae, ursolic acid biosynthesis) — PDI-null; do not consume chaperone capacity; can be stacked beyond the dual-cassette limit without triggering §5.5-style pessimism
Secreted payloads (uricase, lactoferrin, DAF SCR1-4, soluble complement regulators, recombinant peptides) — consume PDI/ERO1 capacity per §3.5.3; cumulative load must fit the chaperone-capacity envelope per the synergy framework

The platform-design implication is meaningful: the practical stacking limit for the koji endgame strain is the count of secreted cassettes, not the total cassette count. A four-cassette strain of uricase (secreted) + lactoferrin (secreted) + cordycepin (cytosolic) + ergothioneine biosynthesis (cytosolic, native enhancement) is, in principle, no harder on chaperone capacity than the two-cassette dual-cassette baseline — provided each cytosolic cassette's metabolic burden is verified separately under FBA.

This generalization is itself a candidate for codification once comp-028 returns and the pattern is empirically anchored.

Operationally applied at koji-endgame-strain.md §3 "Third-cassette slot design rule" where the cytosolic-vs-secreted distinction gates which payloads route to the endgame strain vs. a separate-strain / LBP-peer-chassis. The framework's rule is independent of which specific payloads are currently active — payload selection is a strategy decision, the cytosolic-vs-secreted framing is a structural design rule.

6. The Falsifiable Test — Pairwise Expression Matrix¶

The framework is testable. Construction of a pairwise expression matrix on a small set of representative substrates would either validate or falsify the predicted synergy coefficients.

Experimental design:

Choose 4-6 candidate cassettes spanning the load classes:
One BiP-only / cytosolic-native repurposed (uricase)
One PDI-heavy mammalian glycoprotein (lactoferrin or DAF full-length)
One PDI-light / calnexin-light fungal-native (lipase or amylase, controlled overexpression beyond native)
One PDI-moderate small mammalian protein (DAF SCR1-4)
One cytosolic / off-pathway (carnosine cassette)
Optionally: one PDI-heavy from a different fold class than Lf (e.g., a second mammalian glycoprotein)
Build all single-cassette and pairwise-cassette strains. N=6 candidates → 6 single + 15 pairwise = 21 strains. Triple combinations (~20 more) optional for second pass.
Quantify per-protein titer for each protein in each strain background (ELISA, activity assay, or quantitative SDS-PAGE).
Compute synergy coefficient for each pair. Plot against predicted (load-class overlap) — the framework predicts a strong negative correlation between "shared dominant load class" and synergy.
Identify which feature (disulfide count / glycan count / chain length / subunit assembly) is the strongest predictor of antagonism. The expected answer is disulfide bond competition (PDI) — but this is empirical.

Cost / timeline. 21-strain construction + quantification: ~$15-25k, 4-6 months in a skilled lab. Three-cassette extension: another $10-15k, 2-3 months. This is substantial — bigger than any single comp-NNN — but the result is a predictive design tool for all future Open Enzyme cassette additions. The first wet-lab Ward 1995 dual-cassette test (validation-experiments.md §1.9) generates the first two data points (uricase alone, uricase + Lf) for free as a side product.

Falsification conditions:

If synergy coefficients show no correlation with load-class overlap, the framework is wrong; cassette competition is governed by something else (codon usage, mRNA stability, vesicular traffic, growth-rate burden).
If synergy is uniformly < 0.7 across all pairs regardless of load class, the binding constraint is upstream of folding (transcription, translation, growth) and chaperone-orthogonality is irrelevant at typical OE expression levels.
If single-cassette and pairwise titers are statistically indistinguishable, the OE expression regime is below saturation for all subsystems and the framework only matters at higher titers than OE currently targets.

7. Headroom Expansion Levers¶

If empirical synergy coefficients confirm chaperone competition is the binding constraint, several published levers expand capacity. The picture in koji is more complex than in A. niger / T. reesei — naive UPR activation does not straightforwardly help.

Lever	Effect (literature)	OE applicability
NSlD-ΔP10 background (10-protease-deletion; Huynh 2020 / Maruyama-lineage patent JP5140832B2)	Required for Huynh's 39.7 mg/L adalimumab; raises secretion floor for PDI-heavy substrates. Independent confirmation: Li 2023 A. niger monellin showed extracellular protease deletion gave 13.5× uplift where BiP/PDI overexpression gave none. (In Vitro)	Already the recommended host for Lf cassette per comp-010 §6. Likely the highest-leverage single intervention for any PDI-heavy or stability-sensitive cassette
Constitutive HacAⁱ (active UPR transcription factor)	Tested in A. oryzae directly: Zhou 2016 (DOI: 10.1016/j.gene.2016.08.018) found HacA-CA upregulated 80 secretory-pathway genes (folding/UPR/glycosylation/vesicle transport) but paradoxically reduced secreted amylase output via RESS (repression under secretion stress) feedback, and impaired growth. Carvalho 2012 (PMC3472299) saw the same growth + central-metabolism trade-off in A. niger. Earlier reports of 2-3× uplift in A. niger / T. reesei (Valkonen 2003 / Mulder 2004) are therefore substrate- and host-context-specific. (In Vitro)	NOT a free win in koji. Naive HacA-CA may down-regulate native amylase production — a non-starter for the OE digestive-enzyme leg, where amylase output is therapeutic. Asymmetric HacA response (does it help PDI-heavy substrates while suppressing amylase?) is an unresolved question worth direct testing before committing a cassette slot
UPR is required as baseline	Tanaka/Gomi 2015 (DOI: 10.1016/j.fgb.2015.10.003) showed UPR is required for A. oryzae growth under hydrolytic-enzyme-producing conditions — IreA-repressed strain is rescued only by constitutive intronless hacA-i. (In Vitro)	Background UPR induction (substrate-driven) is essential and ceiling-limited. The lever is not "should we have UPR" (yes, always) but "should we constitutively over-activate it" (uncertain, possibly counterproductive in koji)
Cross-class chaperone helper combinations (Pichia precedent)	Yu 2017 (DOI: 10.1038/s41598-017-16577-x) — single helpers (CPR6, FES1, STI1) gave only 1.4-1.6×. Zhang 2006 (Biotechnol Prog 22(4):1090-5, PMID 16889384, DOI 10.1021/bp060019r) — combinatorial pairs in P. pastoris: YDJ1+PDI gave 8.7×, YDJ1+Sec63 gave 7.6×, Kar2+PDI gave 6.5× (Zhang 2006 Abstract verbatim, verified 2026-05-06). [VERIFICATION CAVEAT 2026-05-06: Zhang's own single-chaperone arms (Kar2p, Ssa1p, PDI alone) gave 4-7× uplift in the same paper. Within Zhang 2006, the combination/single advantage is ~1.2-1.5×, not 5×. The 5× advantage claim previously here implicitly compared Zhang's combinations to Yu 2017's single helpers — a cross-paper, cross-substrate, cross-host-strain comparison with confounded controls. Use 1.2-1.5× as the intra-Zhang comparison or hold the claim as "combinations outperform singles in the same experiment" without a cross-paper magnitude.] The Zhang 2006 abstract does not specify the test substrate by name; the wiki's previous "single-substrate uplift" framing should be read as "single-protein test system" pending full-text confirmation of substrate identity. (In Vitro)	Highest-leverage direction once individual chaperone bottlenecks are characterized. Untested in koji. Each combination is 2 cassette slots, but the per-substrate uplift is large enough to be net-positive even at the cassette-slot opportunity cost. Order-of-operations: empirical synergy matrix first → identify dominant bottleneck → targeted helper combination
PDI overexpression (cassette)	Variable; substrate-specific. Delic 2014 (DOI: 10.1089/ars.2014.5844) review documents that PDI co-expression had no effect on horseradish peroxidase yield (HRP not PDI-limited), but does help disulfide-rich Fab fragments and antibody chains. (In Vitro)	One cassette slot; modest gene size (~500 aa). Worth using for Lf if empirical PDI bottleneck is confirmed; not worth using prophylactically
Directed evolution of PDI	Pichia precedent; substrate-specific variants give 2-5× per-substrate titer	High effort; reserved for a single high-value substrate (e.g., Lf at >1 g/L target)
Proteome-constrained model (pcSecYeast)	Chen 2024 (DOI: 10.1016/j.ymben.2024.11.010) operationalized secretion capacity as a quantitative proteome-constrained model in S. cerevisiae; 50% downregulation and 35% upregulation predictions confirmed by CRISPRi/a screen. (In Vitro). Closest published "secretion capacity index" but does not yet model multi-substrate competition	Building a koji equivalent (pcSecKoji) would convert the framework from qualitative to quantitative. Substantial undertaking; would be a publishable platform contribution
UPR small-molecule inducers (DTT, tunicamycin)	Transient UPR activation; not viable for food-grade fermentation	Not applicable to OE chassis

The picture inverts what the A. niger / T. reesei HacA literature would predict for koji. The most promising capacity-expansion lever is NSlD-ΔP10 (already adopted) plus cross-class chaperone helper combinations (untested in koji, but Pichia data shows 6.5-8.7× absolute uplift over baseline; Zhang 2006 verified 2026-05-06). Constitutive HacAⁱ alone is not a default lever in koji — Zhou 2016's finding that it suppresses native amylase output is a specific OE problem because the digestive-enzyme leg of the platform requires high native amylase. Order-of-operations: build the endgame strain, measure baseline titers, run pairwise expression matrix, identify bottleneck, then targeted helper-pair intervention.

8. What This Framework Does Not Predict¶

Honest about edges:

Solid-state vs. submerged secretion divergence. Sun 2024 (PMC11051239) explicitly notes that "there are certain proteins that are not secreted in solid-state culture, unlike submerged culture, such as the glucoamylase-encoding gene glaB." Whether a cassette secretes well in koji solid-state is a separate question from chaperone load. The framework predicts which cassettes compete with each other; it does not predict which cassettes secrete in solid-state at all. Format-translation risk dominates validation-experiments.md §1.9.
Quantitative UPR ceiling in koji is unmeasured. No published "secretion capacity index" exists for A. oryzae. Adapt the Pichia-developed approaches (Mattanovich / Gasser group at BOKU) — measurable quantities: UPR-target gene transcript levels, BiP/PDI relative abundance, ER stress reporter induction. Building this measurement infrastructure is the precondition for predictive synergy modeling beyond rough qualitative tiers.
Vesicular traffic is the next likely binding constraint if chaperone capacity is expanded. ER exit site density, COPII vesicle budding, Golgi throughput — these become rate-limiting once folding is no longer rate-limiting. The framework does not address them. Recent A. niger work on ER architecture engineering may translate.
Codon-level interactions are not captured. Two cassettes drawing on the same rare-codon tRNA pool can antagonize each other independently of chaperone load. Codon optimization treats cassettes as independent substrates; multi-cassette codon competition is unstudied.
Glycosylation pathway divergence between koji and mammalian cells changes substrate features in a way the load-class scoring doesn't capture. Native koji glycans are high-mannose; mammalian native is complex sialylated. A cassette with 3 N-glycan sites loads calnexin similarly in both hosts, but the resulting product is biophysically different. This affects therapeutic outcome, not synergy prediction — flagged for clarity, not as a framework gap.
The published koji multi-cassette literature is thin, and the architecture-adjusted coefficients remain extrapolations. There is no published koji-specific pairwise expression matrix at the time of this writing (2026-05-06). The framework's quantitative synergy coefficients are extrapolated from Pichia, A. niger, and yeast literature; they may need rescaling once direct koji data is generated. The §3.5 architecture coefficients (α = 0.3–0.6 for CCP/SCR; α = 1.0 for Ig-like reference; α = 1.5–2.5 for transferrin-lobe) are bounded from in vitro folding kinetics and structural data — not from directly measured PDI residence times in a functional ER system. Two specific caveats: (a) Notari 2023's lactoferrin oxidative folding kinetics were measured from a fully denatured state in vitro; in vivo ER-assisted co-translational folding with PDI acting on the nascent chain may be substantially faster, implying the transferrin-lobe α could be closer to 1.0 than 2.5 in the functional secretory context; (b) the SCR/CCP architecture coefficient (0.3–0.6) is inferred from structural rigidity (Schmidt 2010) rather than measured PDI kcat — a fast-folding structural argument, not a kinetic measurement. The architecture-adjusted synergy coefficients in §5 (e.g., Lf × DAF SCR1-4 "0.40–0.65") are more differentiated than the bulk-count model but carry the same calibration uncertainty. The three load-bearing source numbers feeding §1 and §7 — Huynh 2020 39.7 mg/L, Wakai 2019 40×, Zhang 2006 6.5-8.7× — were grep-verified against primary sources on 2026-05-06; see §10.1 "Verification provenance" for source-line anchoring. Architecture coefficient sources (Notari 2023, Schmidt 2010, Vinci 2004, Feige 2009, Hellman 1999) are documented in §10.2.

Generalization caveat (added 2026-05-15). The §3.5.4 calibration set (§1.9 lactoferrin transferrin-lobe + §1.25 DAF SCR1-4 CCP/SCR under harmonized NSlD-ΔP10 / solid-state / matching-promoter conditions) is designed to validate two fold classes in one host under one format. A successful calibration confirms the α coefficients for these two architectures. It does NOT generalize to future secreted disulfide-rich payloads with different fold architectures — e.g., C1-INH serpin (the parallel CP0 candidate flagged in comp-018 Phase 2), recombinant antibody-derived constructs, or other novel-fold secreted enzymes. Each such payload's α-coefficient prediction remains bounded extrapolation until similar calibration data lands for that fold class. The framework's predictive power scales with calibration-set breadth, not depth — adding a third or fourth fold class to the calibration set (cost: 1 wet-lab arm per fold class, ~$3–5K + 8 weeks per class) is the path to broader confident application. The complementary direct-measurement path — a PDI-residence-time assay in A. oryzae microsomes — would generalize across fold classes simultaneously but is a multi-year tool-build (no published A. oryzae-specific PDI kcat data exists for any fold class as of 2026-05-06). Operational guidance: when a new secreted disulfide-rich payload is being considered for OE, surface the α-coefficient uncertainty explicitly in the design decision rather than treating the framework's prior as a measured number. The cytosolic-payload design rule (§5.6 + koji-endgame-strain.md §3 third-cassette slot design rule) sidesteps this concern entirely for cytosolic payloads.

Capacity-vs-titer benchmark ambiguity (flagged 2026-05-06). The framework calibrates synergy coefficients against the Huynh 2020 antibody capacity ceiling (39.7 mg/L adalimumab in NSlD-ΔP10, 16 disulfides). koji-endgame-strain.md §3.3 cites Ward 1995 (>2 g/L lactoferrin in A. awamori, 17 disulfides) as the lactoferrin titer precedent. Both numbers cannot be the binding constraint — Lf at 2 g/L would be impossible if 16-disulfide adalimumab at 40 mg/L were a hard capacity ceiling. Three resolutions are possible: (a) Huynh's ceiling is antibody-architecture-specific (single-chain Lf folds easier than dual-chain HC+LC paired assembly), (b) A. awamori (Ward) has different chaperone capacity than A. oryzae NSlD-ΔP10 (Huynh), © submerged ≠ solid-state. If resolution (a) holds, this framework's synergy coefficients are systematically conservative for single-chain payloads. The §1.9 wet-lab Lf-alone arm directly resolves the ambiguity (validation-experiments.md §1.9 "Capacity-vs-titer side-product readout"). Until that data lands, the framework's coefficient ranges should be read with this asymmetry in mind: the lower bounds are well-anchored, the upper bounds may understate true capacity for single-chain glycoproteins.

9. Open Questions¶

Does HacAⁱ help PDI-heavy substrates while suppressing native amylase? Zhou 2016 found constitutive HacA reduces secreted amylase via RESS feedback — but whether it simultaneously helps PDI-heavy substrates (which would justify the trade-off if the OE digestive-enzyme leg can absorb the amylase loss) was not tested. Substrate-asymmetric HacA response in koji is the load-bearing unresolved question for capacity-expansion strategy.
Are A. oryzae's duplicated chaperone paralogs (CPR1, ERJ5, ERD2) functionally independent capacity pools? Liu 2014 documents the duplications; functional differentiation is not characterized. If different paralogs specialize on different substrate classes, koji has structural advantages over yeast as a multi-cassette chassis. Direct test: paralog-specific KO impacts on a cassette panel.
Is there a "secretion pathway capacity index" measurable per koji strain? Chen 2024's pcSecYeast operationalizes this for S. cerevisiae. Building a pcSecKoji equivalent — using A. oryzae-specific secretion proteome data (Liu 2014 is the starting point) — would convert the framework from qualitative to quantitative. Substantial undertaking; would be a publishable platform contribution.
What's the synergy coefficient for two PDI-light fungal-native enzymes overexpressed beyond native level? I.e., is there meaningful chaperone competition between native lipase and native amylase when each is pushed to 2-3× native titer? This affects the digestive-enzyme leg of the OE platform. Wakai 2019's >1.0 cross-cellulase synergy suggests light-substrate stacking is actually super-additive in koji at moderate loads.
Does the framework hold for intracellular cassettes? The cytosolic chaperonin (TRiC / CCT) and HSP70 / HSP90 cytosolic chaperones have their own competition dynamics. Carnosine cassette (cytosolic) is "off the secretion pathway" but not "off chaperone competition entirely." Probably modest at OE expression levels but worth checking.
Pairwise vs. higher-order interactions. Does the synergy coefficient compose linearly across 3+ cassette combinations, or are there triplet / quadruplet emergent effects? Pairwise matrix is a tractable starting point but may miss higher-order behavior. Wakai 2019's 3-cellulase 40× synergy is a hint that triplet effects exist.
Cross-class helper combination effect size in koji. Zhang 2006 (verified 2026-05-06) shows Pichia gets 6.5-8.7× absolute uplift over baseline from YDJ1+PDI, YDJ1+Sec63, Kar2+PDI cross-class pairs. Within Zhang's own data, single chaperones (Kar2p, Ssa1p, PDI) gave 4-7× — so the intra-paper combination/single advantage is only 1.2-1.5×, not the >5× the framework previously implied via cross-paper comparison to Yu 2017's 1.4-1.6× single-helper data (Yu 2017 used different host strain, different substrate, different growth conditions; the cross-paper comparison is confounded). The clean Zhang-only signal is that combinations do outperform singles in the same experiment, but the magnitude of the combination/single advantage is modest absent the cross-paper anchor. Whether this composes the same way in koji is untested, and the smaller intra-paper advantage weakens (but does not eliminate) the case for prioritizing helper combinations over single-helper cassettes. Direct koji testing remains the resolver.

Multilingual literature scan (2026-05-05): A multi-database multilingual scan (PubMed, J-STAGE, CiNii, CNKI, WanFang, J-GLOBAL) for chaperone-orthogonal multi-cassette frameworks in A. oryzae surfaced no published predictive design rule. Japanese koji-engineering work (Maruyama, Kitamoto, Gomi labs) publishes predominantly in English-language venues (AEM, AMB, Fungal Biol Biotechnol); Chinese groups working on Aspergillus heterologous expression (Pan/Wang at South China Univ of Technology) similarly publish in English. The framework gap is genuine, not a translation artifact — no two-model translation cross-check was needed because no non-English-only source contained mechanism claims absent from the English literature.

10. Cross-References¶

aspergillus-oryzae.md — Host chassis overview, native enzyme production, GRAS context
cassette-compatibility-computational.md — comp-010: specific 2-cassette uricase + lactoferrin analysis (the case study under this framework)
koji-endgame-strain.md — The 4-molecule / 5-chokepoint endgame design that this framework retroactively explains
engineered-koji-protocol.md — Full engineering protocol; §15 covers carnosine co-expression module
codon-optimization-expression-cassette.md — Independent cassette optimization (multi-cassette codon competition is an extension)
koji-construct-design.md — Cassette architecture (promoters, signal peptides, terminators)
protein-engineering-strategy.md — Substrate-specific protein engineering
validation-experiments.md §1.9 — First wet-lab dual-cassette test; generates the first 2 data points for the pairwise matrix
manual-literature-mining.md §"Pre-commit verification gate" — methodology applied retroactively to this page on 2026-05-06; see §10.1 below for outcomes

10.1 Verification provenance¶

Pre-commit verification gate (manual-literature-mining.md §"Pre-commit verification gate") applied retroactively on 2026-05-06 to the three load-bearing numbers feeding §1 and §7. Outcomes:

Source paper	Number(s) verified	Verdict	Source location	Date
Huynh HH et al. 2020 Fungal Biol Biotechnol 7:7, PMC7257131, DOI 10.1186/s40694-020-00098-w	39.7 mg/L adalimumab; ten-protease deletion (NSlD-ΔP10) required	✅ VERIFIED	Abstract verbatim ("highest amount of antibody was obtained from the ten-protease deletion strain (39.7 mg/L)"); Results §"Adalimumab production in the culture supernatant of A. oryzae"; Fig. 1 production-curve	2026-05-06
Huynh 2020 + UniProt P01857 (IgG1 HC) + P01834 (kappa LC)	16 disulfides per assembled IgG1 (4 inter-chain + 12 intra-chain)	✅ VERIFIED via UniProt	UniProt P01857 FT DISULFID annotations: 27..83 (intra CH1), 144..204 (intra CH2), 250..308 (intra CH3), 103 (inter HC-LC), 109 + 112 (inter HC-HC); plus VH (1 intra) and LC (1 intra VL + 1 inter LC-HC). Total: 12 intra + 4 inter = 16. Note: the Huynh 2020 paper itself does not state the disulfide count; the wiki's "16 disulfides" is textbook IgG1 stoichiometry verified via UniProt, not extracted from Huynh 2020.	2026-05-06
Wakai S et al. 2019 Bioresour Technol 276:146-153, PMID 30623869, DOI 10.1016/j.biortech.2018.12.117	40× total cellulolytic activity over single-integration baseline; 5-16 chromosomal copies; three cellulases (cellobiohydrolase + endoglucanase + β-glucosidase)	✅ VERIFIED on the 40× and 5-16 numbers (abstract verbatim); ⚠️ CORRECTED on the synergy decomposition	Abstract verbatim ("The resulting strain possessed 5-16 copies of each cellulase gene within the chromosome and showed approximately 10-fold higher activity versus single integration strains... resulting transformant showing 40-fold higher cellulolytic activity versus the single integration strain"). The wiki's previous claim — "approximately 10× of the gain came from copy stacking; the rest from multi-cassette synergy" — was a misattribution. Per the abstract, the 4× delta from 10-fold to 40-fold came from promoter/terminator engineering (P-sodM/T-glaB selected via per-copy transcription screen), not multi-cassette synergy. The word "synergy" does not appear in the abstract.	2026-05-06
Zhang W et al. 2006 Biotechnol Prog 22(4):1090-5, PMID 16889384, DOI 10.1021/bp060019r	YDJ1+PDI 8.7×, YDJ1+Sec63 7.6×, Kar2+PDI 6.5×	✅ VERIFIED on the absolute fold values; ⚠️ CORRECTED on the "5× advantage over single helpers" framing	Abstract verbatim ("the combination chaperones of YDJ1p/PDI, YDJ1p/Sec63, and Kar2p/PDI synergistically increase secretion levels 8.7, 7.6, and 6.5 times, respectively"). Within the same paper, single chaperones (Kar2p, Ssa1p, PDI alone) gave 4-7× uplift ("introduction of Kar2p, Ssa1p, or PDI improves protein secretion 4-7 times"). The intra-paper combination/single advantage is therefore ~1.2-1.5×, not 5×. The previous "5× advantage" claim was a cross-paper comparison to Yu 2017's single-helper data (1.4-1.6× for CPR6/FES1/STI1), which uses different host strain, substrate, and conditions — confounded comparison. The Zhang 2006 abstract does not specify the test substrate by name; "single-substrate uplift" framing in earlier wiki versions should read "single test-protein system" pending full-text confirmation.	2026-05-06

Methodology: All four papers retrieved 2026-05-06 via mcp__plugin_pubmed_PubMed__get_full_text_article (Huynh 2020 only — open-access via PMC7257131) and mcp__plugin_pubmed_PubMed__get_article_metadata (Wakai 2019, Zhang 2006 — abstract-level metadata). Wakai and Zhang full-text not retrieved due to publisher paywalls (Elsevier, Wiley); abstracts authoritatively verified the load-bearing numbers, and the methodological details that would require full text (Wakai's per-cassette ratios, Zhang's substrate identity) are flagged as needing separate full-text retrieval if downstream work depends on them.

10.2 Architecture Coefficient Verification Provenance (added 2026-05-06)¶

Pre-commit verification gate (manual-literature-mining.md §"Pre-commit verification gate") applied to all per-architecture folding kinetics claims in §3.5. Sources retrieved and verified via mcp__plugin_pubmed_PubMed__get_article_metadata and mcp__paperclip__paperclip (search + head commands; never map or reduce per memory/feedback_paperclip_map_unreliable.md) on 2026-05-06.

Source	Claim verified	Verdict	Source location	Placeholder flags
Notari S et al. 2023 Sci Rep PMC10465537, DOI 10.1038/s41598-023-41064-x	LF contains 16 disulfides; 4 cysteines (Cys64, 424, 512, 668) show 900–2140× enhanced GSSG reactivity and initiate folding; remaining 28 cysteines fold subsequently; hierarchical cascade from molten globule	✅ VERIFIED	Abstract: "containing 16 disulfides"; Results L9 and L25–L27 (PMC full text): kinetic constants for 4 hyperreactive cysteines (Table 3: Cys668 k = 1.71 M⁻¹s⁻¹, 2140× enhanced; Cys512 1.21, 1510×; Cys64 1.11, 1390×; Cys424 0.72, 900×)	[NUMBER PARTIALLY UNVERIFIED] The α = 1.5–2.5 coefficient is derived from these kinetics as a proxy for prolonged PDI residence — no direct PDI kcat or residence time is published for LF. The 28-cysteine slow cascade is the basis for the "slow folding" inference; the specific multiplier relative to Ig-like fold is a bounded estimate.
Wally J and Buchanan SK 2007 Biometals PMC2547852, DOI 10.1007/s10534-006-9024-3	LF bilobal structure with helical linker; inter-lobe cooperativity; each lobe ~330 aa; hTF has 3 additional disulfide bonds vs LF; "LTF has no disulfide bonding between the subdomains in either lobe"	✅ VERIFIED	Full text L18–L21 (PMC): "disulfide bonds are not present in LTF"; structural comparison confirming hTF has cysteines 137–331, 161–179, 339–596 not in LTF; inter-lobe helical linker in LTF confirmed; cooperativity discussion L18.	[NOTE on disulfide count]: Wally 2007 confirms LF has fewer disulfides than hTF; combined with Notari 2023 (16 disulfides explicitly), this corrects the OE wiki's prior use of "17 disulfides" for LF. UniProt P02788 lists 34 Cys residues with DISULFID annotations; the "16 vs. 17" discrepancy depends on annotation completeness. Notari 2023's explicit "16 disulfides" is used for §3.5.3 and §5.5.4.
Schmidt CQ et al. 2010 J Mol Biol PMC2806952, DOI 10.1016/j.jmb.2009.10.010	CCP modules are compact ~60-aa β-sandwiches with 2 disulfides (Cys1-Cys3 + Cys2-Cys4 pattern); "all eight Cys residues were confirmed oxidized" in fH12–13; CCP modules are rigid independent units in SAXS/AUC analyses; backbone well-ordered by NMR T₁/T₂	✅ VERIFIED	Full text L20 (PMC): "oxidised state of the eight Cys residues was confirmed by mass spectrometry"; L20: "disulphide pattern (CysI–CysIII and CysII–CysIV) was inferred from the proximities of cysteine pairs"; L37–L41 (SAXS compact structure); L64 (NMR relaxation data).	[NUMBER UNVERIFIED] The α = 0.3–0.6 coefficient for CCP/SCR fold is a structural inference (pre-organized scaffold → brief PDI engagement) — no published PDI kcat or residence time for individual CCP modules. This is the largest unverified assumption in the architecture model.
Vinci F et al. 2004 J Biol Chem PMID 14729662, DOI 10.1074/jbc.M311480200	PDI required for complete Fc oxidation; without PDI, 1-disulfide intermediate (CH3 disulfide) predominates; "two domains of the Fc fragment fold independently, with a precise hierarchy of disulfide formation in which the disulfide bond, especially, of the CH2 domain requires catalysis by PDI"	✅ VERIFIED	Abstract verbatim (PubMed record): the quoted statement is from the Vinci 2004 Abstract. The "1S2H" kinetic trap framing is from the same abstract.	No placeholder flags — abstract-level verification is sufficient for the claim that PDI is a required catalyst for Ig Fc folding.
Feige MJ et al. 2009 Mol Cell PMID 19524537, PMC2908990, DOI 10.1016/j.molcel.2009.04.028	CH1 domain "intrinsically disordered in vitro," folds only upon LC interaction; "structure formation proceeds via a trapped intermediate"; BiP "recognizes incompletely folded states of the CH1 domain"	✅ VERIFIED	Abstract verbatim (PubMed record): all three quoted phrases appear in the abstract.	No placeholder flags — abstract-level verification is sufficient for the claim about CH1 disorder and BiP engagement.
Hellman R et al. 1999 J Cell Biol PMID 9885241, PMC2148116, DOI 10.1083/jcb.144.1.21	CL domain "folds rapidly and stably even in the absence of an intradomain disulfide bond"; BiP does not detectably bind CL; VL domain folds slowly and shows BiP engagement	✅ VERIFIED	Abstract verbatim: "constant domain folds rapidly and stably even in the absence of an intradomain disulfide bond. Thus, the simple presence of a BiP-binding site on a nascent chain does not ensure that BiP will bind." VL domain BiP engagement: "both the wild-type and mutant BiP clearly associated with the unoxidized variable region domain" (Abstract).	No placeholder flags.

LF disulfide count correction note (§3.5.3): The OE wiki used "17 disulfides" for lactoferrin in prior pages (including §5.5.1 of this document, committed 2026-05-06 at commit a9f159d). The correct count per Notari 2023 (PMC10465537, Abstract, explicitly: "containing 16 disulfides") is 16. UniProt P02788 lists 34 cysteines with DISULFID annotations grouping them as 17 pairs, but Notari 2023's in vitro thiol-titration (32 free cysteines after complete reduction = 16 pairs × 2 = 32) confirms 16 disulfides. The discrepancy may reflect one unoccupied Cys or a counting convention difference. For all architecture-adjusted computations in §3.5 and §5.5, 16 disulfides is used (Notari 2023 source-anchored). Prior wiki pages that use "17 disulfides" for Lf (including comp-010, §5.5.1 of this page) are flagged for consistency update but are not modified here to avoid scope creep; the architecture-adjusted model in §3.5+ uses 16.

Multilingual note: The Notari 2023 paper (Italian research group, published in English in Scientific Reports) is the primary source for LF oxidative folding kinetics. A search of J-STAGE, CNKI, and J-GLOBAL did not surface non-English primary papers on lactoferrin or SCR/CCP domain oxidative folding kinetics that were not already indexed in PubMed in translated form. The architecture coefficients are therefore not subject to the two-model translation cross-check protocol (all primary sources are English-language). If future Japanese or Chinese literature on koji-specific PDI substrate kinetics emerges, it should be incorporated per the global-multilingual rule.

11. Strategic Implication for Open Enzyme¶

The chaperone-orthogonal framework reframes a question Brian asked in conversation: "how many cassettes can we bolt on?"

The reframe: not "how many" but "which combinations get synergistic yield." Concretely, the OE roadmap implications:

The endgame strain's "five molecules" thesis is biochemically reasonable, not greedy. Total chaperone burden ≈ Huynh 2020 antibody baseline because four of the five payloads are off-pathway or chaperone-light. The framework converts what looked like an aggressive payload count into a defensible engineering target.
Future cassette additions should preferentially fill underused load classes. If the next addition is another PDI-heavy mammalian glycoprotein (e.g., DAF full-length rather than SCR1-4), the framework predicts steep synergy decay vs. lactoferrin. If the next addition is another cytosolic small-molecule pathway or another off-pathway cassette, additivity is preserved.
NSlD-ΔP10 (already adopted) is the highest-leverage host intervention; cross-class chaperone helper combinations are the highest-leverage capacity-expansion experiment. Zhou 2016's finding that constitutive HacAⁱ in koji suppresses native amylase via RESS feedback rules out the naive A. niger / T. reesei HacA strategy as a default OE move — the digestive-enzyme leg of the platform requires native amylase output. The replacement strategy is targeted: build endgame strain → measure baseline titers and per-class bottlenecks → install matched helper-pair cassettes (Pichia precedent verified 2026-05-06: YDJ1+PDI 8.7×, YDJ1+Sec63 7.6×, Kar2+PDI 6.5× absolute uplift over baseline, Zhang 2006 Biotechnol Prog 22(4):1090-5; intra-paper combination/single advantage 1.2-1.5×). Two cassette slots, substrate-specific intervention.
The pairwise expression matrix is the single highest-leverage characterization investment the platform could make at this stage. ~$25k / 4-6 months for a predictive design tool reusable across every future OE cassette decision. Compare to the per-cassette wet-lab cost of guess-and-check expression in the absence of a framework.
The under-explored design axis Brian flagged is real, and the multilingual literature scan confirmed it. No published pairwise expression matrix exists for koji; no published predictive design rule for chaperone-class orthogonality in any filamentous fungus. Component pieces exist (Gong 2009 chaperone-module catalog; Liu 2014 A. oryzae secretome map; Delic 2014 Pichia substrate-specificity data; Chen 2024 pcSecYeast) but no one has composed them. Open Enzyme can be the project that builds and publishes this framework — both as a contribution to fungal synthetic biology and as a competitive moat for the platform's multi-cassette designs.
The "five-molecule endgame strain" is more biochemically defensible than it looks. The Wakai 2019 3-cellulase result (40× total cellulolytic activity over single-integration baseline; ~10× from copy-stacking + ~4× from promoter/terminator engineering, verified 2026-05-06) is direct A. oryzae evidence that three light substrates can coexist at high titer without titer collapse in the same chaperone subset — a necessary precondition for chaperone-orthogonal compatibility, though not direct evidence of cross-cassette synergy distinct from per-cassette transcriptional scaling. The OE endgame strain has more orthogonality than Wakai (substrates partitioned across BiP/PDI/calnexin/cytosolic/off-pathway classes vs. Wakai's three substrates all in one class). Predicted synergy ≥0.85 is consistent with the absence-of-titer-collapse evidence in Wakai and the broader chaperone-orthogonality framework.

All synergy coefficients in this analysis are Mechanistic Extrapolation pending the §6 pairwise expression matrix validation. Quantitative coefficients are calibrated from yeast, Pichia, and A. niger literature; koji-specific calibration is the load-bearing missing data.