Abstract
BACKGROUND/OBJECTIVES: Single-domain immunoglobulins are small protein modules with specific affinities. Among them, the variable domains of heavy chains of heavy-chain-only antibodies (VHH) as the antigen-binding fragment of heavy-chain-only antibodies (also termed nanobodies) have been widely investigated for their applicability, e.g., therapeutics and immunodiagnostics. However, despite their advantageous biochemical and biophysical characteristics, protein aggregation throughout recombinant synthesis is a serious drawback in the development of nanobodies with application perspectives. Therefore, we aimed to develop a computational method to predict the aggregation propensity of VHH antibodies for the selection of promising candidates in early discovery. METHODS: We employed a deep learning-based structure prediction for VHHs and derived from it likely biophysical and biochemical properties of the framework region 2 with relevance for aggregation. A total of 106 nanobody variants were produced by recombinant expression and characterized for their aggregation behavior using size exclusion chromatography (SEC). RESULTS: Quantitative characteristics of framework region 2 patches were combined into a function that defines an aggregation score (AS) predicting the aggregation propensities of VHH variants. AS was evaluated for its capability to forecast recombinant VHH aggregation by experimentally studying VHH Fc-fusion proteins for their aggregation. We observed a clear correlation between the calculated aggregation score and the actual aggregation propensities of biochemically characterized VHHs Fc-fusion proteins. Moreover, we implemented an easily accessible pipeline of software modules to design nanobodies with desired solubility properties. CONCLUSIONS: AI-based prediction of VHH structures, followed by analysis of framework region 2 properties, can be used to predict the aggregation propensities of VHHs, providing a convenient and efficient tool for selecting stable recombinant nanobodies.