Retooling Cancer Model

Cancer is an ever-present scourge in modern society. More than 1.6 million Americans were diagnosed with cancer in 2016, and it is estimated to cost American citizens more than $156 billion annually by 2020.

Understanding how a tumor changes through time and recurs after surgery or treatment, as well as what types of drugs best kill the tumor, are essential for improving human cancer therapies.

Frequently, mouse models of cancer are used to study the disease and evaluate possible therapeutics, but a recent study from the Broad Institute Cancer Program demonstrates that mouse models do not represent human tumor evolution as well as thought, and these models may yield false-positive drug responses.

Their work highlights the continuing need for new model-building in biomedical sciences, particularly for the preclinical evaluation of cancer therapeutics.

Studying cancer in animals, like mice, maintains tumor complexity and tumor/microenvironmental interactions that are absent in cell cultures. The toxicity of a drug can be better evaluated when it is delivered to and processed by a complex set of organs, as opposed to simply adding it to liquid on cells.

Mice are used to study tumors in three different ways:

1) Mice can be exposed to a mutagen, like radiation, to cause tumors

2) The DNA of mice can be engineered so that tumors arise as the mice age, which produces tumors of murine origin.

3) Immunocompromised mice can be injected with human cells that will give rise to tumors of human origin.

When mice are injected with cells that were derived from a patient’s tumor, they are called a patient-derived xenograft, or PDX.

PDXs are thought to be an ideal way to model cancer because they maintain tissue architecture, interactions of the tumor with its environment (minus immune cell interactions), and the genomic alterations associated with the human tumor.

In practice however, PDX might not faithfully model patient tumors as well as originally thought.

Researchers from the Broad Institute Cancer Program recently published a paper in Nature Genetics that shows that PDX models evolve differently in mice than tumors in humans.

The researchers compared how the number of chromosomes in a PDX changed compared to the tumor from which they were derived, as well as through PDX in vivo passages.

An in vivo passage occurs when a fully formed PDX is surgically removed from a mouse, dissociated into single cells, and re-implanted into a new generation of mice.

In cancer, erroneously repeated or deleted regions of chromosomes, called copy number alterations (CNAs), and additional copies or loss of entire chromosomes, called aneuploidy, give the cell some of its tumorigenic properties.

The authors examined DNA sequencing information from 133 PDXs to determine CNA status. They inferred CNAs from RNA sequencing data from another 977 tumors, which allowed them to evaluate the evolution of 1100 PDXs from 24 cancer types.

When compared to CNAs of human tumors from The Cancer Genome Atlas (TCGA), PDXs generally represented the tumors from which they were derived.

When single PDX models were studied through in vivo passages, however, the PDXs rapidly acquired new CNAs and lost some of the CNAs they started with, becoming less like the original tumor.

Some PDX models were more susceptible to changing CNAs than others. Gastric cancer PDXs are the least stable, while brain tumor PDXs are the most stable.

Not all cells in human tumors are the same. Populations of tumor-forming cells, called subclones, contain different sets of genetic mutations that affect that subclone’s ability to grow, divide, and spread.

Subclones do not compete in PDXs the same way they do in human tumors. In fact, mutations that are considered causal in the development of human tumors, and are retained upon recurrence, were found to disappear in some PDX models, especially after they had been passaged several times.

Gains and losses of chromosomes, and therefore the genes they carry, are known to affect therapeutic responses to drugs.

The authors demonstrated that CNAs that were uniquely acquired in colon, pancreatic, and breast cancer PDXs (and are absent in the human TCGA data) made the PDXs sensitive to targeted therapies.

This phenomena may help explain why there exists such a high number of false-positive therapeutic responses to drugs in mouse models of cancer.

Clearly, more work is needed to create effective forms of model-building that will yield reliable and clinically-actionable conclusions.