Unsupervised Machine Learning-Based Clustering of Nanosized Fluorescent Extracellular Vesicles

Area of Science:

• Biotechnology
• Nanomedicine
• Cell Biology

Background:

• Extracellular vesicles (EVs) are crucial for intercellular communication and hold diagnostic potential.
• Current methods lack robust single-particle quantification of EV biomarkers, hindering clinical application.
• EV heterogeneity necessitates single-vesicle analysis for accurate biomarker profiling.

Purpose of the Study:

• To develop and validate a single-particle method for multiplexed biomarker detection on EVs.
• To assess EV purity and inflammatory status using multicolor single-molecule burst analysis.
• To demonstrate the utility of the method for identifying disease-specific EV subgroups and contaminants.

Main Methods:

• Multicolor single-molecule burst analysis microscopy for detecting multiple biomarkers on single EVs.
• Machine learning (t-distributed stochastic neighbor embedding) for classifying and clustering multidimensional single-particle data.
• Application to assess purity, inflammatory status, and identify specific EV subgroups in cell culture and plasma samples.

Main Results:

• The method successfully quantifies multiple biomarkers on individual EVs.
• Demonstrated assessment of EV purity and inflammatory status.
• Identified intercellular adhesion molecule-1 specific EV subgroups from inflamed endothelial cells.
• Confirmed apolipoprotein-a1 as a marker for lipoprotein contamination in plasma EVs.

Conclusions:

• The developed single-particle methodology enables robust multiplexed biomarker quantification of EVs.
• This approach facilitates standardized quality assessment of EV preparations and diagnostic profiling.
• The technique is applicable to standard confocal microscopes, promoting widespread clinical and research use.