Employing deep mutational scanning in the Escherichia coli periplasm to decode the thermodynamic landscape for amyloid formation

• 成果类型: Article
• 署名作者: McKay, Conor E.; Deans, Miles; Connor, Jack; Saunders, Janet C.; Lloyd, Christopher; Radford, Sheena E.; Brockwell, David J.
• 署名单位: University of Leeds; University of Leeds; AstraZeneca
• 刊物名称: PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
• ISSN/ISSBN: 0027-12164
• DOI: 10.1073/pnas.2516165122
• 发表日期: 2025-09-23
• 关键词: beta-peptide; aggregation; oligomerization; STABILITY; fibrils

摘要

Deep mutational scanning (DMS) assays provide a powerful method to generate large-scale datasets essential for advancing AI-driven predictions in biology. The tripartite beta- lactamase assay (TPBLA), in which a protein of interest is inserted between two domains of beta- lactamase, has previously been reported as capable of detecting and quantitating the aggregation of proteins and biologics in the oxidizing periplasm of Escherichia coli and used as a platform for identifying small molecule inhibitors of aggregation. Here, we repurpose the TPBLA into a high-throughput DMS platform. We validate this format using a single-site saturation library of the intrinsically disordered peptide A beta 42, linked to Alzheimer's disease, demonstrating strong agreement between observed variant fitness scores and variant behavior using our previously reported low-throughput TPBLA. The results of DMS revealed variant fitness scores that correlate with known amyloid-promoting regions. An in silico approach using FoldX-derived per-residue thermodynamic stability confirmed that the TPBLA reports on amyloid fibril stability. In vitro experiments support this finding, showing a strong correlation between variant fitness scores and the critical concentration of amyloid formation. Machine learning using the DMS dataset identified beta-sheet propensity and polarity as primary drivers of variant fitness scores. The derived model is also able to predict thermodynamically stabilizing regions in other amyloid systems, underscoring its generalizability. Collectively, our results demonstrate the TPBLA as a versatile platform for generating robust datasets to advance predictive modeling and to inform the design of aggregation-resistant proteins.