AI-based analysis of label-free live cell imaging of T-cell mediated tumor killing assay enables competitive and robust hit calling

Abstract

For the discovery and optimization of personalized cancer treatments using immune cell therapeutics, such as T-cell receptor (TCR-T) therapy and bispecific antibodies (BsAbs), robust functional activity of candidates must be confirmed in immune-mediated killing assays. In these assays, co-cultures of several cell lines and patient-derived primary cancer cells often are imaged live using automated microscopy. Conventionally, such assays use fluorescent dyes or specifically expressed nuclear proteins for labeling, followed by classical image analysis reliant on cell segmentation. They are therefore subject to artifacts like phototoxicity and bleaching, inaccurate segmentation due to the typical variations in visual phenotype with time as well as requiring the constant adaptation of analysis parameters for experiments across different human tissue types or donors.

Here we present a new approach utilizing brightfield images in combination with a hands-free, scalable artificial intelligence (AI)-based analysis workflow, requiring no fluorescent markers at all. We have applied this new workflow to a T-cell mediated killing assay and benchmarked it against current semi-manual, cell segmentation-based analysis of fluorescent images. We found that the new workflow performs well on phenotypically diverse cancer cells, with greater efficiency though elimination of manual adjustment steps, and produces results of equivalent consistency.

We conclude that this AI-based analysis workflow has the potential to substantially simplify T-cell mediated live cell killing assays, eliminating the need for labeling, and allows their efficient analysis, operating on brightfield images and thus avoiding time-consuming and difficult analysis of labeled images using classical segmentation-based analysis.