High-throughput cell motility assay for systemic analysis of the cancer cell invasion

Introduction: Tumor metastasis represents a major obstacle for achieving long-term remission in patients suffering from non-hematopoietic malignancy and is the leading cause of cancer-associated death. Ill-regulated cancer cell motility is a fundamental characteristic that drives cancer metastasis. Many components within the tumor microenvironment (TME), cell signaling pathways, and genomic mutations affect cancer cell motility. A better understanding of how components jointly regulate cancer invasion could lead to the identification of molecular targets and therapeutic strategies to inhibit metastasis. However, in vitro methods to assess the invasive potential motility of cancer cells are built upon 2D platforms which have limitations in terms of cost, reproducibility, and throughput. A key limitation to the throughput of motility assays is a lack of automated analysis tools. Manual analysis is time-consuming, making it difficult to study multiple conditions within a reasonable timeframe.

To overcome these limitations, we established a liquid-handling workflow that allows imaging real-time cancer cell motility in a 96 well-plate format. Importantly, we established a machine-vision and deep-learning analysis method to automatically detect and track the migration process of individual cells from time-lapse videos without any additional markers. By conducting a systemic study of cytokine effects and screening a panel of FDA-approved compounds, we demonstrate that our assay can examine cancer cell motility behaviors from ~100 different conditions within a day as opposed to months using conventional methods. Through these studies, we can determine potential therapeutic targets and better understand the joint regulatory effect of cytokines within the TME.

Materials and

Methods: To study the effects of various compounds (i.e.: cytokines, cancer drugs, etc.) on 2D cell motility, cancer cells of different degrees of metastatic potentials were used. ECM components, such as 50µg/mL type I collagen, were used to coat the wells of a 96-well plate. Cells were seeded at 1000-2000 cells/well in 96-well plates and incubated at cell culture conditions (37°C and 5% CO2). Following incubation, treatment conditions were added to each well accordingly. The plates were then placed on a microscope (Nikon TI) with an on-stage incubator to maintain cell culture conditions. Timelapse brightfield images were acquired at 10-minute intervals over a minimum of 16 hours (4 positions per well; 384 positions per plate). The live nuclei in the brightfield images were detected using a custom DL model with DeepLabv3+ network architecture, and the trajectories of individual live nuclei were tracked using the previously established tracking methods. Our developed automated analysis system identifies live/dead cells using a deep-learning model trained using brightfield and corresponding fluorescent images (H33342 for live cell detection and PI for dead cell detection). It then uses the identified cells to track motility from the live-cell microscopy images. To quantify motility, the trajectory of individual cells that migrate across the well are recorded. This method is fully automated and allows for the analysis of 96 conditions in parallel.

Results, Conclusions, and Discussions: Our developed deep-learning model can detect cells within brightfield images with high overall accuracy for various cell lines and densities. Displacement/motility results found using our automated cell tracking algorithm were comparable to results from conventional manual tracking from the same image sets. A panel of breast cancer cell lines (MDA-MB-231, MCF-7, SUM149, SUM159, HCC1954) were used to test the effects of various cytokines found within the TME. Using our developed assay and deep-learning model, we determined the effect of cytokines (EGF, TNF-α, IL-6, and HGF) on these breast cancer cell lines. These results agree with previous studies which found that EGF, TNF-α, IL-6, and HGF increase breast cancer cell motility. Additionally, we identified synergistic and non-linear effects on the cancer cell lines when various TME factors were examined in combination (collagen coating, EGF, TNF-α, FBS concentration).

Initial findings using our deep-learning model and our liquid-handling workflow show that our automated cell tracking platform agrees with conventional methods and previous studies. This suggests that this developed assay is valid for 2D applications and can be used to identify complex relationships between combinations of factors on cell motility. As differences in motility can be observed between individual cytokine conditions, this assay has a high potential for applications using high-throughput CRISPR screening and for high-throughput drug screening. For the future development of this assay, we will push to develop and optimize a 3D assay model that can track cells embedded within ECM matrices.

Acknowledgements (Optional):