Computational Design of Ten Engineered Capsicum Cultivars Spanning 0.87-2.96M SHU: Vanillylamine Precursor Supply, Not Pun1/AT3, Controls the Scoville Ceiling

Summary We report the computational design of ten engineered Capsicum chinense cultivars targeting 866,611–2,956,310 Scoville Heat Units (SHU) via a three-tier intervention stack: peroxidase knockout (Tier A), plus PAL and COMT overexpression (Tier B), plus Pun1/AT3 triple mutant (Tier C). Products are designated SS-0087 through SS-0096. Full cultivar specification papers and design packages for each product are deposited as companion records (hasPart links below) and available for free download at https://www.scovillesplice.farm. The central finding is that vanillylamine precursor supply — not the terminal condensation enzyme Pun1/AT3 — controls capsaicinoid output. Pun1/AT3 carries 0% flux control coefficient; halving its catalytic capacity leaves pathway output unchanged. The true bottleneck is COMT, whose expression crashes 69% from 24 DPA (535 TPM) to 34 DPA (166 TPM), creating temporal substrate starvation at the peak accumulation window. Three independent lines of evidence converge on this conclusion: (1) expression-weighted FBA (+20.5% flux with sustained COMT), (2) ODE kinetics (+78% vanillylamine deficit at 34 DPA), and (3) tier stack equivalence (Tier C = Tier B in flux output). These cultivars were designed using the Capsaicin Design Platform (https://platform.scovillesplice.farm), a computational biology platform integrating RNA-seq expression data, flux-balance analysis, ODE kinetics, ESM2 protein language model variant analysis, and CRISPR guide design. Greenhouse validation is ongoing at Scoville Splice LLC wet laboratories in Montecito, CA, USA and Geneva, Switzerland. Methods RNA-seq: BioProject PRJNA789050, C. chinense SL08 placental tissue, 5 time points (4, 24, 34, 44, 54 DPA), n=5 per point. Quantification: Salmon 2.8.0. DE: PyDESeq2 0.5.4, BH correction (alpha=0.05). Result: 4,365 significant genes of 28,074 tested. FBA: COBRApy 0.32.1, GLPK solver, 37 metabolites, 44 reactions, expression-weighted flux bounds. ODE kinetics: 7-variable Michaelis-Menten model; COMT V_max modulated by time-series expression data. ESM2: 650M parameter model. Three Pun1/AT3 mutations: S39L (LLR 2.828), L345G (LLR 2.695), C175S (LLR 2.030). AlphaFold2: 440 residues, mean pLDDT 90.2. CRISPR: 24 SpCas9 guides, 4 targets, 79% low off-target risk. Constructs: 25 constructs, 31,911 bp total, GC 37.4–41.8%. 19/25 require codon wobbling. Data Availability Construct sequences (QB169135–QB169174, GenBank SUB16548149; embargo until 25 October 2027). Raw RNA-seq: NCBI BioProject PRJNA789050. USPTO Provisional: SS-CDP-2026-001P (filed 9 October 2026). Free design packages for all ten cultivars: https://www.scovillesplice.farm. Individual cultivar papers deposited as companion records (hasPart links).