Profiling co-occurrent morphological phenotypes and their degree of expression severity in vacuolated cells by holo-tomographic flow cytometry and fractal analysis

Abstract

Cells are complex systems characterized by large phenotype heterogeneity. Conventional single cell classification approaches usually separate cells expressing a certain phenotype (e.g. associated to a disease condition) from the healthy control. However, multiple phenotypes typically coexist within the same cell as a result of complex intracellular interactions, machineries, functioning and external stimuli. Here we use label-free optical microscopy, powered by AI, to investigate how morphological phenotypes co-occur within vacuolated cells. Cytoplasmic vacuoles are important hallmarks of several pathological states (e.g. lysosomal storage diseases, viral infections, cancer). We rely on Holo-Tomographic Flow Cytometry (HTFC) to obtain 3D refractive index tomograms of vacuolated cells in continuous flow. Then, we propose a strategy to reduce the dimensionality of the tomogram using cross-sectioning and minimum intensity projection maps. We extract a set of morphological, refractive index-based, and fractal parameters demonstrating that the complex heterogeneity of vacuole patterns can be captured and can foster classification based on interpretable features. For training the AI, biologist domain-experts provided annotation of the different morphological phenotypes expressed and ranked them in terms of expression severity from the tomographic observations. Thus, we introduce a pipeline for morphometric phenotype profiling, in which each cell is associated with a 7-digits classification code representing the combination of coexisting phenotypes it expresses and their expression severity levels.