Entry for:The Bioinformatics Peer Prize
Tumors consist of multiple populations of cells that differ in their behavior and the size of these populations varies greatly. The response to treatment differs between populations, therefore we need to know the make up of a tumor, which can be achieved by identifying the mechanism that causes intratumor heterogeneity. A common explanation for this intratumor heterogeneity is the presence of cancer stem cells (CSCs) that have an unlimited renewal potential. Alternatively, clonal dominance may be the result in differences in division potential between the cells.
In this work we try to identify the source of intratumoral heterogeneity in, previously published, in vitro experiments of iterated growth and passage of cancer cells (Porter et al, Genome Biol., 2014). In these experiments genetic tags, barcodes, were added to cancer cells in order to track subpopulations over time. The barcoded cells were than grown for 3 days after which 10% of the population was passed to the next generation. Repeating this growth and passage 30 times resulted in a drop of the number of subpopulations and a strong increase of clonal dominance: a few subpopulations are large while all other subpopulations are small.
To test which hypothesis could explain these observations, we build a computational model, using Gillespie's stochastic simulation algorithm, that accurately simulates the iterated growth and passage experiments. The CSC hypothesis is simulated by incorporating two cell types: CSCs and differentiated cells (DCs). The division of a CSC may produce either 2 CSCs, a CSC and a DC, or 2 DCs, and DCs may only divide a limited number of times, only producing DCs. The alternative hypothesis is simulated by assigning a different division rate, taken from a normal distribution, to each subpopulation, which is inherited upon division.
Before testing either hypothesis we tested the effects of random growth and passage by assuming that all cells divide at the same rate. This model does not result in increased clonal dominance, but does closely match the loss of subpopulations observed in vitro.
Next, we compared the effects of each hypothesis. Simulations based on the CSC hypothesis did not result in clonal dominance but instead caused a pronounced drop in the number of subpopulations. This happened because any subpopulation initialized with little or no CSCs eventually disappears. Alteration of the model parameters can reduce the massive loss of subpoplatuions, but did not affect the development of clonal dominance.
In contrast, simulations based on variability in division rate, resulted in a close match to the experimental data. Since this model is only based on one parameter, the division rate standard deviation, we could easily obtain an optimal fit for the number and size of the subpopulations.
Simulations based on the CSC hypothesis could not match the experimental observations, while simulations based on the assumptions that all cells divide with a variable, inheritable, division rate did match those observations. Hence, our findings suggest that tumor cells exhibit a heritable variation in the division rates of individual cells.
5. Future ideas/collaborators needed to further research?
In the future, the model could be further extended to improve its power, especially for comparison with in vivo data. For this, the model should be extended with an explicit representation of space and physical interactions between cells. With such a model it becomes possible to explore the consequences of division rate variability while comparing with intra-vital images studies.