AI Assisted Analysis of 3D Cell Models Using Widefield Microscopy

Introduction

Drug discovery & development is a costly and time-consuming process with a high candidate failure rate¹. Two-dimensional cell cultures have been a common method for investigating drug candidates and have improved success over biochemical methods. However, while valuable, these cultures do not accurately represent the complex in vivo tissue environment and failure rates remain high². There has been an increased interest in three-dimensional models that better recapitulate the in vivo microenvironment. Recent advances indicate that these can become an important tool in drug discovery².
Viability staining is one of the most common phenotypic assays performed with 3D cell models and is an important aspect of cancer therapeutics and other drug discoveries. There are many ways to measure cell viability, most methods involving stains, imaging, or fluorescence measurements^3,4. The most common methods measure metabolic function of cells with the percentage of viable cells extrapolated. Examples include measuring ATP by using a luminogenic ATP assay, such as CellTiter-Glo, and the rapid colorimetric MTT assay, measuring metabolic activity through the quantity of water-soluble dye converted to in-soluble Formazan^4,5. Since these methods acquire measurements through cell solubilization, they do not permit further downstream analysis, limiting versatility. Similarly, Trypan Blue staining of

dead cells initially allows for cell retention, but the dye is toxic to viable cells and after 30 min all cells, viable or not, are stained⁶. Conversely, Thallium 201 imaging to detect myocardial viability retains cell structure but is much harder to quantify than other methods⁷.
Fluorescent live/dead imaging of 3D cell models has proven to be a robust and versatile process that permits further downstream analysis of biomarkers and phenotypic profiling^8,9. A consideration in adoption of this protocol is the type of microscope to use for image acquisition. Confocal microscopes provide high “Z” resolution and clearer images of cells in 3-dimensional structures. However, confocal systems are more expensive, tend to have longer acquisition times, and create larger data files than widefield counterparts. Widefield microscopes are more common, can acquire signal from a whole 3D structure in a single image, and produce smaller datasets. However, the cellular resolution is poorer than confocal and the ability to count Live and Dead cells is compromised. To overcome this limitation, we present a method for artificial intelligence (AI) assisted analysis of cell viability images taken with widefield microscopy. We show that similar data can be achieved compared to confocal microscopy, but at a more affordable and accessible cost point and streamlined workflow.

Fig 1. The workflow outlined shows forming and loading spheroids into a flowchip, inserting the flowchip into Pu·MA System, starting the automated assay, imaging of spheroids post assay, and AI assisted analysis of images.

Download