Single-molecule FRET imaging and deep learning reveal concentration dependence of aggregation pathways during Aβ aggregation

Protein aggregation into amyloid fibrils is the hallmark of several devasting neurodegenerative diseases. Gaining an understanding of disease etiology hinges on our ability to understand the molecular mechanics of how soluble monomers assemble to form insoluble fibrils consisting of thousands of constituent monomers. Although amyloid fibril formation is a highly specific self-assembly process, growth patterns and resultant fibril morphologies are highly dependent on solution conditions. Using fluorescence lifetime imaging and deep learning, we have recently shown that amyloid assembly occurs via heterogeneous aggregation pathways resulting in a mixture of co-present mature fibrils with unique morphologies and physicochemical properties (Meng et al., PNAS_2022_e2116736119). Bulk biophysical methods are unable to fully characterize these mixtures of fibril polymorphs. Here, we further develop and use Förster resonance energy transfer (FRET) imaging to monitor the entire aggregation pathway of the Alzheimer's Disease related peptide amyloid β 42 (Aβ42) at the single fibril level in real time. We incubated a mixture of donor-labeled, acceptor-labeled, and unlabeled Aβ42 monomers, which resulted in the formation of fibrils with diverse FRET efficiency values, indicating structural heterogeneity. Single-fibril images reveal that increasing monomer concentration promotes the formation of more homogeneous fibrils. Fibrils formed at lower concentrations show assemble via highly heterogeneous pathways. Deep learning methods (https://github.com/hoisunglab/FNet) enable segmentation of single fibrils within images of highly overlapping fibrils, allowing for quantitative analysis of the aggregation process in terms of fibril growth rate and photon density for each fibril over the course of fibril assembly.