(D-125) Nuclear image analysis to predict cancer cell migration in confinement

Introduction:: During cancer metastasis, tumor cells migrate through spatially confined microenvironments. In vivo, individual cancer cells travel along blood vessels, muscle fibers, and extracellular matrix components [1]; these confining microenvironments provide mechanical cues that alter cell morphology and behaviors. Moreover, the nucleus of the migrating cell has been revealed to actively serve as a mechano-sensor under biophysical cues. Previous work has shown that the deformation of the cell nucleus increases cell body contractility and migration speed [2]. Additionally, changes in nuclear chromatin structure correlate with tumor cell progression [3]. However, it remains largely unknown whether and how the nuclear pre-confinement morphologic and epigenetic states regulate the initiation and completion of confined migration. Specifically, are 2D nuclear morphology and chromatin organization sufficient to predict a successful migration event through a confined space? To investigate this question, we built a single-cell, machine learning-based image analysis pipeline of cell migration events in confining microchannels, and implemented this pipeline to classify successful cell migration events.


Materials and Methods:: The input dataset consisted of single-cell live tracking images of confined cell migration events; to recreate the in vivo confining environment, we fabricated a PDMS-based microchannel platform. Channels were 3 μm wide and 150 μm long, and chips were coated with a type 1 collagen solution for cell attachment. MDA-MB-231 cells tagged with NES-GFP and H2B-BFP to visualize the cell body and nucleus were then seeded onto the chips. Stable Fucci HeLa cell line was used to track the cell cycle during live imaging. Lastly, cell migration through the channels was recorded for 24 hours at 37ºC using the ZEISS Celldiscoverer7 live microscope.

An image processing pipeline was built to pre-process the input dataset with ImageJ. The dataset was then processed by the ResNet 34 classification model. Global nuclear morphology, boundary features, and cell speed before entering the channel were extracted using ImageJ and the TrackMate plugin.


Results, Conclusions, and Discussions:: Τhe aspect ratio of the nucleus was higher in cells that did not successfully migrate (p=0.007), and cell circularity was higher in cells that successfully migrated (p=0.0015). This suggests that global nuclear morphological features play a role in predicting cell escape from confinement. Surprisingly, pre-confinement migration speed did not change between successful and unsuccessful migration events, which indicates a distinct mechanism governing cells’ ability and decision making during 2D and confined migration. Seeing these differences, we introduced the Fucci-Hela cell line to the microenvironment to investigate the influence of cell cycle in the migratory capacity of cancer cells under confinement. Preliminary data suggested a preferred bias towards cells under G1 phase.

To gain further insights into the potential correlation between nuclear biomarkers and migration ability, we implemented the ResNet 34 convolutional neural network to classify nuclei live images. The model was able to classify cell migration events with a 40% error rate; however, when analyzing the top losses of the model, no clear visual similarity was observed between the model’s incorrect predictions. Ongoing experimentation – increasing the size of the dataset and testing different image classification models – will likely reduce the error rate and allow chromatin architecture and nuclear morphological features to be more clearly extracted.

Understanding the factors that influence a cell’s decision to undergo confined migration is relevant to predicting and treating cancer metastasis. Nuclear features of tumor cells could be a potential diagnostic biomarker for cancer malignancy, and future study might indicate pathways to target to decrease the likelihood of migration in a malignant tumor. Further characterizing the classification might also provide insight on how the nucleus is implicated in tumor cells’ epithelial-to-mesenchymal transition and other cell fate decisions.


Acknowledgements (Optional): : The authors would like to acknowledge the support of the Amgen Scholars Program and the National University of Singapore.

References (Optional): : [1] Paul, C., Mistriotis, P. & Konstantopoulos, K. Cancer cell motility: lessons from migration in confined spaces. Nat Rev Cancer 17, 131–140 (2017). https://doi.org/10.1038/nrc.2016.123

[2] A. J. Lomakin et al. The nucleus acts as a ruler tailoring cell responses to spatial constraints. Science 370 , eaba2894 (2020). DOI:10.1126/science.aba2894

[3] Venkatachalapathy, S., Jokhun, D.S., Andhari, M. et al. Single cell imaging-based chromatin biomarkers for tumor progression. Sci Rep 11, 23041 (2021). https://doi.org/10.1038/s41598-021-02441-6