Scientists have developed a machine-learning based approach that allows them to predict how tumours will evolve in response to anti-cancer drugs and therapeutics, laying the groundwork for personalised medicine for cancer patients.
Like all living things, cancer cells evolve in response to selective pressure and changing environments. Cells with advantageous traits are more likely to survive than those without, and so become more prevalent in tumour tissues. At a microscopic level, it is a world where the strong dominate and the fittest survive. Recently, scientists have begun to wonder what it really means to be the “fittest” in the competitive world of tumour tissues. Cancer cells that can thrive under conditions saturated with chemotherapy drugs are expected to be “fitter” than those that are drug-sensitive. But are they?
In a study conducted by scientists from Memorial Sloan Kettering, who collaborated with the University of British Columbia/BC Cancer in Canada, researchers have developed a novel machine-learning approach that can accurately predict how tumours, particularly breast cancer tumours, evolve over time by leveraging the principles of population genetics. By measuring tumour evolution, this inventive tool has allowed scientists to uncover new findings on tumour growth and development. It is also expected to help clinicians predict drug responsiveness of tumours and identify cells that are most likely to be responsible for relapses, paving the way for more personalised treatments and better clinical outcomes for cancer patients.
Led by MSK computational biologist Sohrab Shah and BC Cancer breast cancer researcher Samuel Aparicio, this machine-learning project was developed using a combination of three innovations: patient xenografts, single-cell sequencing technology, and a machine-learning tool called FitClone.
Patient xenografts are realistic cancer models constructed by removing human cancers from patients and transplanting them to mice. The researchers analysed these mice tumour models repeatedly over extended timeframes of up to three years, taking snapshots of these models as they went along. These snapshots allowed the scientists to obtain a clearer picture of the progression of cancer.
The team also applied single-cell sequencing technology to simultaneously document the genetic makeup of thousands of individual cancer cells in the tumour in an efficient and automated manner.
The third component, FitClone, was built together with UBC statistics professor Alexandre Bouchard-Côté and applies the mathematical principles of population genetics to cancer cells in the tumour. The equations involved demonstrate how a population comprised of individuals with different starting frequencies and varying levels of fitness will evolve.
Combined, these three innovations have enabled scientists to model how individual tumour cells and their clones will behave, and predict how cancers will evolve. “The beauty of this model is it can be run forwards to predict which clones are likely to expand and which clones are likely to get outcompeted,” commented Dr. Shah.
To demonstrate their model’s predictive powers, the team used the model to track the effects of platinum-based chemotherapy treatments in mouse models. Their findings revealed that treatment with platinum chemotherapy led to the emergence of drug-resistant tumour cells with distinct copy number variants. Copy number changes refer to the differences in the number of specific DNA segments in cancer cells, which can significantly impact fitness.
Having observed these changes, the team became interested in how stopping platinum-based treatments would affect the tumour. Upon halting the treatment, they discovered that the drug-resistant cancer cells that initially dominated the tumour when subjected to platinum chemotherapy began to decline or disappear. In place of these cells, drug-sensitive cells grew in number and outmatched the drug-resistant ones, suggesting that drug resistance has an evolutionary cost. In other words, in an environment devoid of chemotherapy, drug-resistant traits may not be optimal to support cell survival, and so are replaced by the original drug-sensitive cells.
“This study is an important conceptual advance,” commented Dr. Shah. “It demonstrates that the fitness trajectories of cancer cells are predictable and reproducible.”
The team hopes to expand the applications of their approach on blood samples to help identify particular clones in a person’s tumour. In doing so, they can predict how the tumour will evolve and respond to drugs and therapies, and better personalise therapeutic approaches accordingly. [APBN]
Source: Salehi et al. (2021). Clonal fitness inferred from time-series modelling of single-cell cancer genomes. Nature, 1-6.