Thermodynamic trade‑off and expression bottleneck in Prime Editor evolution Protein engineering and random laboratory evolution performed to increase the catalytic activity of Prime Editors have expanded the potential for precise genomic correction. However, the evolved variants carried a fatal paradox at the molecular level. Amino‑acid substitutions introduced to improve editing performance instead compromised protein thermodynamic stability and collapsed soluble expression levels, creating a trade‑off. This structural instability caused a rapid decline in intracellular effective editor protein concentration after mRNA delivery, forming a critical technical bottleneck and blind spot that limited final editing efficiency in vivo.

ProteinMPNN neural‑network‑based inverse‑folding architecture: catalytic‑preserving multiplex mutagenesis In the study released this May, the team fully deployed the AI‑guided protein design network ProteinMPNN, which uses structural information, to overcome intrinsic collapse barriers of protein architecture. The three‑dimensional coordinates of the reverse transcriptase (RT) domain were reduced to AI‑driven inverse‑folding algorithm inputs. While preserving the essential catalytic pockets required for nucleic‑acid complex formation and activity, the backbone’s hydrophobic core and surface energy were optimized through computational parallel introduction of 30–163 extensive amino‑acid substitutions. This engineering redesign became the central engine that maximized intracellular folding stability of the editor complex.

Ex vivo primary‑cell soluble expression amplified by 200 % and in vivo PE6/PE7/PEmax efficiency barriers broken The next‑generation PE8 Prime Editor architecture, fully reconstructed with ProteinMPNN, increased intracellular soluble protein residuals up to twofold (200 %) upon mRNA delivery compared with the original. Multi‑center ex vivo clinical screening across diverse cell types and delivery modalities, including human primary cells, showed dramatically accelerated correction kinetics. In head‑to‑head in vivo mouse benchmarks, the PE8 platform demonstrated up to 2.9‑fold higher final target‑editing efficacy relative to the current state‑of‑the‑art backbones PE6, PE7, and PEmax, establishing a clear performance gap.

Establishment of in silico protein design architecture and standardization of next‑generation synthetic biology components The computational protein‑engineering and molecular‑genetics dataset delivers a uniquely impactful contribution to global biopharma R&D and programmable nucleic‑acid therapeutic businesses. By complementing the thermodynamic depletion barrier of laboratory evolution with an AI‑based inverse‑folding algorithm, the work establishes an “AI‑in‑silico evolutionary hybrid enzyme‑engineering standard framework.” The high‑stability RT backbone design metrics serve beyond Prime Editors, functioning as a filtration engine that pre‑calculates immunotoxicity and efficacy thresholds for all synthetic‑biology devices requiring high soluble expression, such as base editors, self‑replicating RNA replicons, and biosensors. It perfectly filters out false‑positive genotoxicity noise caused by protein aggregation in preclinical stages, thereby acting as a master reference that can dramatically shorten IND approval timelines for regulatory agencies worldwide.