Morphological single-cell analysis of peripheral blood mononuclear cells from 390 healthy blood donors with Blood Cell Painting

Abstract

AbstractRecent molecular profiling and phenotyping methods combined with machine learning based analyses enable genotype-phenotype discovery at an unprecedented scale. The challenge now lies in unraveling the biological mechanisms underpinning these associations. High content imaging is a cost-effective approach for morphological and functional profiling of single cells that has provided insight into mechanisms of disease phenotypes, and consequences of genetic and drug perturbations. However, the morphological variability of healthy immune cells − instrumental to understanding disease-specific deviations from the healthy state − is still relatively uncharacterized. To elucidate this variability at scale, we generated high-resolution fluorescent confocal imaging data of peripheral blood mononuclear cell (PBMC) samples from 390 healthy blood donors with the Blood Cell Painting protocol. The protocol, developed here from the popular Cell Painting morphological profiling assay, optimizes for efficiency and throughput, and includes PBMC thawing, plating and fluorescence marker staining of non-adherent blood cells, followed by confocal and widefield imaging with a high content microscope. We assigned cell types based on cellular features with a classifier trained expert annotations, and observed monocytes to be five-fold more frequent in imaging data compared to flow cytometry baseline, with B and T cells being two-fold less frequent. We hypothesize this discrepancy is due to differential adherence between the cell types. We also evaluated three computational methods for correcting batch effects in imaging data, and found Harmony to perform the best, compatible with previous reports. Finally, we performed the Blood Cell Painting protocol on PBMCs in acute myeloid leukemia, and showed the protocol to be able to distinguish between AML FAB subtypes. Our study highlights the utility of high-content imaging with Cell Painting in characterizing and understanding health and disease phenotypes, opening avenues to further studies with integrated imaging and molecular profiling data.This manuscript is a work in progress, and we anticipate incorporating additional results into subsequent versions.