As a scientist, I want to find a treatment for cancer. To do this I need to perform many experiments testing my new therapy on cancer. I can’t just use human patients, they’re too annoying. How about cancer cells from a tumour and grown in flasks? Sounds good, but through my literature search, I find that cell lines undergo substantial alterations within their genome. That surely won’t do, my therapy targets a specific gene. I need a model which does not undergo such genetic alterations. I probably should look for something else.
Perhaps it’s because the environment of a flask is too different from a human? How about a mouse?
Mice are mammals, like humans. Sounds good.
If I can place the cancer cells from a human tumour into a mouse, since it is a similar environment, the cancer should not change as excessively. So, I take a sample of cells from a cancerous tumour and surgically graft it into a mouse. Great, now I have an accurate model of cancer. One that I can use to perform experiments on that does not involve those annoying humans. One that will better represent the original tumour than if grown in flasks. One that does not change its genetic make-up over time.
Humans and mice seem more similar than humans and flasks. But still wouldn’t the change from a human environment to a mouse also cause changes?
I mean, I like cheese, but not as much as a mouse. Surely this would have an effect.
It was assumed that Patient-derived Xenografts (PDX’s), where human tumours are surgically grafted into mice, accurately represent the genetics of the human primary tumour.
Well… until recently.
Uri, et al (2017) was not satisfied with this assumption. They set out to test if Patient-derived xenografts accurately represent the genetic landscape of a primary tumour and to compare with two established cancer models; cancer cell lines and cell line derived xenografts (CLDX’s).
First, we may need to establish some characteristics of cancer.
We need to view a cancerous tumour not as an individual entity, but as a population of individual cancer cells. The cancer cell population is subjected to evolutionary forces the same as any other population of individuals. Over passages (generations) of cells, certain genes will be lost or gained depending on evolutionary forces.
As you can imagine with a population of animals, if placed in a totally alien environment, the animals would undergo divergent evolution over generations in the new environment. Creating genetically different animals when compared to the original population. A population of cancer cells, in this aspect, is no different. When cancer cells are taken from a human tumour and grown in flasks (Cell Lines) they too will undergo evolution over passages – Gains/losses of DNA via mutations within the genome of cancer cells. This leads to cancer cells with altered genetic landscapes in comparison to the original tumour cells.
This is a current problem for cancer research.
Cancer research relies on models that accurately represent the genetic landscape of the primary tumour. If not, the observed responses may not be applicable to the original cancer in question, otherwise known as – useless.
Currently, cell lines and CLDX’s are known to undergo model specific mutations leading them to be genetically different from the primary tumour over time. It is therefore important for future research to develop a model which accurately represents the primary tumour.
As genetic changes in PDX models over passages had not been directly studied Uri, et al (2017) had to create a catalogue of large genetic alterations – gains or losses of genetic information – between early and late passages of PDX models from other papers. Each study had to include DNA-based genomic measurements from primary tumours and at least one PDX passage. However, this lead to a small data set, too small for a comprehensive analysis.
That won’t do…
To overcome the quantity issue they applied computational inference algorithms on studies with gene expression profiles. In other words, they predicted changes in the genomic landscape by inferring the expression of genes – somewhat concerning. It is unreliable to predict specific losses or gains of DNA based on gene expression. But until a direct study between PDX models and primary tumours is made this will have to do.
Overall, a final data set of 1110 PDX’s representing 24 different cancer types were used (Figure 1). This is a bit misleading as only fractions of this data set were suitable for many of their analyses leading to a much smaller sample size per analysis. This leads to an inability to make specific conclusions, however, general conclusions can be.
By comparing the passaged models genetic landscapes to the primary tumour’s genetic landscape allowed Uri, et al (2017) to determine if each model still retains – and which best represents – the genetic landscape of the primary tumour. Alterations from the primary tumour’s genetic landscape are considered model-acquired alterations. The greater the amount of model-acquired genetic alterations leads to a poor representation of the primary tumour by that model.
Using the above reasoning the assumption of PDX models retaining the same genetic variation as their primary tumours was shown to be false. After a tumour is implanted within a mouse there is a dramatic shift in the genetic landscape of the PDX tumours with a median of 12.3% (range, 0-58.8%) within 4 passages (Figure 2). This is a reduction of total genetic alterations in PDX models as compared with cell lines.
However, given the labelling “early, medium, and late” of cell line passaging, the actual amount of passaging would be much greater than 4 even for “early”. This could account for some of the increase of genetic alterations when compared with PDX models.
Next, when PDX’s were compared to CLDX’s, it is shown that PDX has a much higher change in the genetic landscape. This does not necessarily mean that CLDX’s better represent the primary tumour, unfortunately. Certain aspects of the CLDX model, such as they are xenografted from well-established cell lines. Thus, before being xenografted each CLDX would have already undertaken large changes within their genetic landscapes, causing less observed changes when passaged within mice.
Figure 2 – Mosaic of Model-acquired genetic alterations per model type. Credit: Uri, et al (2017)
Overall this does not look good for PDX’s – It does not look good for any of the three models.
The change in environment from a human to a mouse does not seem to be spared from specific selection pressures causing genomic alterations. Thus, PDX models diverge from their primary tumours after being xenografted and do not accurately represent the primary tumour. PDX’s also do not necessarily represent the primary tumour better than cell lines of CLDX’s.
What is even more concerning for current cancer research is their next finding.
The PDX models can lose signature chromosomal aberrations through passaging – large gains or losses of chromosomes that are believed to cause the specific cancer type. We see an example of this is in figure 3, the hallmark chromosomal gains of breast cancer chromosomes 1q and 8q are lost in PDX models.
This could have a huge therapeutic impact.
Most therapies target aspects like these signature alterations. If the signature alterations are no longer present in the models, it is highly likely that the targets of therapies are altered too. Potentially leading to many therapies having been discarded due to the failure of the model to accurately represent the primary tumour and not due to the therapy being ineffective.
Think of all the wasted time and research grants.
Most scientists already knew the flaws of working with cancer cells line and CLDX models but assumed the same flaws would not be present with tumours grown in mice. This assumption clearly was incorrect. Uri et al, 2017 has created a clear divide within cancer research where, after the publication, PDX’s (once thought infallible) are now subject to the same criticism subjected to cell lines and cell line derived xenografts. The search for cancers next model top model needs to continue.
I’m sorry to say.
Patient-derived Xenografts you are no longer in the running to be cancers next top model.
Topic Paper; Uri et al. (2017) Patient-derived xenografts undergo mouse-specific tumor evolution. Nature Genetics, 49 (11) 1567-1575. doi:10.1038/ng.3967
Cell line evolution’s effects on Drug Response: Uri, et al. (2018)Genetic and transcriptional evolution alters cancer cell line drug response, Nature, 560, 325-330. doi.org/10.1038/s41586-018-0409-3
Organ on a chip – new human drug model: Zhang, et al. (2018) Advances in organ-on-a-chip engineering. Nature Reviews Materials, 3(8), 257. https://doi.org/10.1038/
Further PDX Info: Tentler, et al. (2012) Patient-derived tumour xenografts as models for oncology drug development. Nat. Rev. Clin. Oncol. 9, 338-350. http://www.nature.com/doifinder/10.1038/nrclinonc.2012.61
Clinical relevance of Cancer Cell Lines: Gillet et al (2013). The clinical relevance of cancer cell lines. J. Natl. Cancer Inst. 105, 452–458. doi-org.ezproxy.massey.ac.nz/10.1093/jnci/djt007