Surface Induced Unfolding Reveals Unique Structural Features and Enhances Machine Learning Classification Models

Native ion mobility-mass spectrometry combined with collision-induced unfolding (CIU) is a powerful analytical method for protein characterization, offering insights into structural stability, and enabling the differentiation of analytes of similar mass and mobility. A surface induced dissociation (SID) device was recently commercialized, enabling broader adoption of SID measurements and surface induced folding (SIU). This study evaluates SIU, benchmarking its reproducibility and performance against CIU on a Waters CyclicIMS ion mobility-mass spectrometer. Reproducibility studies were conducted on model proteins, including β-lactoglobulin (β-lac), bovine serum albumin (BSA), and immunoglobulin G1 kappa (IgG1κ). SIU and CIU exhibited comparable reproducibility, with root-mean-square deviation (RMSD) values averaging less than 4% across multiple charge states. Notably, SIU achieved unfolding transitions at lower activation thresholds, enhancing sensitivity to subtle structural differences and providing additional analytical information content such as unique high arrival time unfolding features and additional unfolding transitions. Furthermore, the differentiation of closely related protein subclasses, such as IgG1κ and IgG4κ, was improved with SIU, as evidenced by higher RMSD pair-wise comparisons and the identification of unique unfolding intermediates. These advancements translated into enhanced supervised machine learning models for IgG subclass classifications. SIU-trained models outperformed or matched CIU-trained models, achieving high cross-validation accuracies (>90%) and robust classifications of biotherapeutics Adalimumab and Nivolumab. This work establishes SIU as a complementary and efficient alternative to CIU, offering improved sensitivity, and analytical depth without loss in reproducibility. This work highlights the benefits of including SIU in protein characterization workflows, particularly in high-throughput and machine learning-guided applications.