Multi-parameter phenotyping of platelets and characterisation of the effects of agonists using machine learning

Abstract

Platelet function is driven by the expression of specialised surface markers. The concept of distinct circulating sub-populations of platelets has emerged in recent years, but their exact nature remains debatable.

Objective To design a spectral flow cytometry-based phenotyping workflow to provide a more comprehensive characterisation, at a global and individual level, of surface markers in resting and activated healthy platelets. Secondly, to apply this workflow to investigate how responses differ according to platelet age.

Methods A 14-marker flow cytometry panel was developed and applied to vehicle- or agonist-stimulated platelet-rich plasma and whole blood samples obtained from healthy volunteers, or to platelets sorted according to SYTO-13 staining intensity as an indicator of platelet age. Data were analysed using both user-led and independent approaches incorporating novel machine learning-based algorithms.

Results The assay detected differences in marker expression in healthy platelets, at rest and on agonist activation, in both platelet rich plasma and whole blood samples, that are consistent with the literature. Machine learning identified stimulated populations of platelets with high accuracy (>80%). Similarly, machine learning differentiation between young and old platelet populations achieved 76% accuracy, primarily weighted by FSC-A, CD41, SSC-A, GPVI, CD61, and CD42b expression patterns.

Conclusions Our approach provides a powerful phenotypic assay coupled with robust bioinformatic and machine learning workflows for deep analysis of platelet sub-populations. Cleave-able receptors, GPVI and CD42b, contribute to defining shared and unique sub-populations. This adoptable, low-volume approach will be valuable in deep characterisation of platelets in disease.