Challenge. Stem cell derived 3D tissue models bear the unique potential to study human organ development and malfunction. The main challenge for properly analyzing these organoids lies within their heterogeneous cell-patterning and morphology. Hence, current studies mainly focus on single-cell transcriptome assessment and 2D histology, thereby sacrificing spatial information. Nonetheless, microfluidic cell culture technologies continue to evolve rapidly, in tow with the urgent need of high-throughput analytical methods suitable for the full exploitation of these samples. Importantly, tracking and phenotyping such 3D cultures at the single-cell level not only requires the development of image acquisition methods providing high time and spatial resolution, but also powerful computational tools capable to handle the increasing amount of data generated.
Solution. In a recently published study in Cell Methods Reports, teams from PioneerCampus and AI4Health addressed this challenge. The Pioneer team around Matthias Meier presents a label-free, live imaging approach enabling faster image acquisition of 3D stem cell cultures cultivated on chip. This microfluidic platform integrates a deep learning method called Bright2Nuc, specifically developed for this platform by Carsten Marr and his team from AI4Health, which allows to predict in silico nuclear staining from confocal bright-field images in 3D.
Results. Bright2Nuc reliably segmented individual nuclei from bright-field images, quantified their morphological properties, predicted stem cell differentiation state, and tracked cells over time. The morphological features from inferred nuclei and corresponding bright-field images facilitated determining different transcriptional cell states across their developmental trajectory. Bright2Nuc was also able to capture cell dynamics, for example epithelial-mesenchymal transitions. Importantly, the inferred cell states were obtained in this label-free setting with higher precision than by using a single cell marker in current state of the art fluorescence staining approaches. Within the constraints of the studies’ framework, Bright2Nuc allowed for the assessment of motion directionality, cell migration, and cell-cell contacts during the experiment.
Taken together,Bright2Nuc resolved real-time information from live nuclei in 3D cell cultures highlighting the so far untapped potential of information contained in bright-field images. As such, Bright2Nuc opens a diverse field of applications for dynamic high-content screening of organ-on-chips settings.