Cancer Technologies
Agent-based Modeling of Cell Interactions to Predict Metastasis in Pancreatic Cancer
Christian Palacios-Gomez (he/him/his)
Student
Berea College
Huntsville, Alabama, United States
Matthew Lazzara
Principal Investigator
University of Virginia, United States
Shayn M. Peirce
Professor and Chair of Biomedical Engineering
University of Virginia, United States
Epithelial-to-mesenchymal transition (EMT) occurs in cancers of epithelial origin, such as pancreatic ductal adenocarcinoma (PDAC), and is the process whereby epithelial cells decrease E-cadherin expression, lose their adhesive bonds with neighboring cells, and transdifferentiate into mesenchymal cells that express higher levels of vimentin. The dogma in the field is that mesenchymal cells become migratory and leave the primary tumor, invade surrounding or distal tissues, and become the seeds that create new tumors, or metastases (1). However, our group has identified two novel behaviors, which we have termed “scission” and “capture”, that are exhibited by mesenchymal cells and may be responsible for creating and shedding multi-cell clusters of metastatic cells from the primary tumor. Since multi-cell clusters are known to have an even more robust capacity to form metastases, it is important to understand the cell behaviors that give rise to their formation in the primary tumor. Our objective was to develop a computational, multi-cell agent-based model (ABM) to study how scission, the breaking of adhesive bonds between neighboring epithelial cells by migrating mesenchymal cells, and capture, the adhesion-based pushing or pulling of mesenchymal cells by neighboring epithelial cells, leads to the formation of multi-cell clusters that are shed from the primary tumor. ABMs simulate individual cells as agents, assign specific rules to dictate their behaviors, and predict how cell-cell interactions over time produce changes at the population-level. Modeling scission and capture events in populations of cells undergoing EMT may inform new strategies to therapeutically restrict the shedding of multi-cell clusters.
We developed our ABM using NetLogo 3D software. Our model simulates 200 randomly dispersed epithelial and mesenchymal cells programmed to aggregate into a spherical mass, representing a primary tumor. The simulated cells exhibited the expected adhesive and migratory behaviors (repulsion and adhesion) and unexpected behaviors (scission and capture) that we have previously observed in 2-dimensional co-cultures of epithelial-like and mesenchymal-like PDAC cells (Figure 1). Cell adhesion and migration in epithelial and mesenchymal cells are regulated by relative E-cadherin and vimentin expression levels. Therefore, simulated epithelial cells were encoded with random relative levels of E-cadherin expression, and mesenchymal cells with random levels of vimentin expression, ranging from zero to ten relative units (r.u.). We then used BehaviorSpace, a tool embedded within NetLogo that allows the user to run multiple simulations, to conduct three different in silico experiments. The first experiment represented the baseline EMT condition in early PDAC whereby E-cadherin expression levels in epithelial-like cells decreased by 0.01 r.u. and vimentin expression levels in mesenchymal-like cells increased by 0.01 r.u. The second experiment simulated the inhibition of EMT by a hypothetical drug that limited protein expression changes to +/- 0.001 r.u. The third experiment simulated late-stage PDAC by instituting a complete protein expression change of +/- 0.1. We repeated each in silico experiment twenty times and recorded the total number of cells shed from the primary tumor. Results across experiments were compared using one-way ANOVA with Tukey’s Honest Significant Difference for post hoc comparisons.
The results from the three in silico experiments are presented in Figures 2 and 3. In the baseline experiment, scission caused approximately 43 cells to shed, and capture events shed about seven cells. The shedding events created small gaps between cells within the tumor mass and led to the escape of small (3-6 cells) and moderately-sized (6-10 cells) multi-cell clusters. Introducing a hypothetical drug in the second experiment decreased the number of shed cells due to scission by three-fold and the amount due to capture by four-fold. Additionally, the tumor remained more intact, as there was limited shedding of cells, demonstrating that the inhibition of EMT may help prevent cell-cell interactions in the primary tumor that lead to metastasis. In the third experiment, the number of shed cells due to scission more than doubled compared to the baseline experiment, while the number of shed cells due to capture increased even more (p< 0.00001). Furthermore, the composition of the primary tumor was fragmented, as scission created large ( >10 cells) multi-cell clusters that were shed, and capture events formed several additional smaller multi-cell clusters. The predictions from the third experiment showcase how increased EMT could lead to more robust PDAC metastasis.
While the ABM predicts that scission and capture behaviors exhibited by cells undergoing EMT cause cell shedding and determine the size of multi-cell clusters that are shed from a primary tumor, it makes many simplifying assumptions. Our ABM only simulated two hundred cells, while tumors in the body may be comprised of thousands or millions of cells. There is also frequent cell proliferation and apoptosis in cancer, which our model fails to include. Future extensions of our ABM will represent tumor biology and the tumor microenvironment with more fidelity, including hypoxic conditions, desmoplasia, and other growth factors that drive EMT and impact the progression of PDAC. Ultimately, experimental validation of the model predictions can confirm how EMT in individual tumor cells impacts scission and capture behaviors in the primary tumor and the resulting shedding of multi-cell clusters.
Tiwari, N., Gheldof, A., Tatari, M., & Christofori, G. (2012). EMT as the ultimate survival mechanism of cancer cells, Seminars in Cancer Biology, 22(3), 194-207. https://www.sciencedirect.com/science/article/pii/S1044579X12000491