By Aaron Krol
We know that cancer is a personal disease, with each case displaying a unique genetic makeup that demands uniquely targeted therapy. Yet even as private companies like Foundation Medicine rush to capitalize on personalized cancer treatment, the basic research that supports these efforts still relies on flawed models: animal cancers, or human cancer cell lines, which may not map appropriately to live human tumors at all, let alone to a specific patient’s disease. This week, Bio-IT World is taking a look at three groups who are trying to update cancer models for modern drug discovery with a personal touch.
December 8, 2013 | All Chris Sander’s lab was trying to do was a simple drug perturbation study.
Sander is the chair of the Computational Biology Program at the Sloan-Kettering Institute, a division of New York’s Memorial Sloan-Kettering Cancer Center devoted to foundational cancer research. His department does some truly innovative work, including building and maintaining the cBioPortal
, an open online tool for the access and analysis of cancer genomics data. But they also do the same basic research being painstakingly performed in smaller labs around the country, and on this particular occasion, a grad student wanted to test a pair of drugs on a cancer cell line to see if the cocktail might be effective in fighting high-grade serous ovarian carcinoma (HGSOC).
Cancer cell lines are a miscellaneous bunch. Some types of cancers only have two or three cell lines in regular circulation to represent their biology in laboratory tests. HGSOC, however, has around fifty, according to the widely-respected Cancer Cell Line Encyclopedia (CCLE) produced by the Broad Institute. The CCLE has been busily genotyping the hundreds of cell lines used in cancer research – some of them decades old – and making that information available online. Their work has made clear just how many options labs have to choose between. To start their drug perturbation trial, Sander’s group had some winnowing down to do.
“We wanted to pick an ovarian cancer cell line,” remembers Nikki Schultz, a member of the Computational Biology Program laboratory. “And instead of just doing what everybody else does, going next door and asking – ‘Hey, do you have an ovarian cancer cell line? Okay, thank you.’ – we wanted to ask, if we have a choice, we know that they’re now fully characterized, which of the fifty is the best match?”
It’s a strange quirk of laboratory culture that this question rarely gets asked. In the case of HGSOC, just two cell lines account for fully 60% of all published drug trials, despite no obvious reason being given to prefer them. Popularity breeds popularity, and in the absence of other guidelines, laboratories have assumed prior research papers represented best practices. Even when practical considerations come into play in choosing a cell line, genetic fidelity to the real cancer being modeled is not generally the major criterion. “A lot of it has to do with availability and access to cell lines,” Schultz told Bio-IT World, “and also the ease with which they grow, the speed at which they grow. Some cell lines are easier to maintain than others, more robust than others.”
These qualities make an experiment easier to conduct, but they don’t help validate the results.
“The good news is that they’re all cancer cell lines of some sort,” says Schultz, “so findings that are derived from any of these experiments are still relevant for the field of cancer research in its entirety. But whether they’re of benefit to high-grade serous ovarian carcinoma research is a different question.” To make sure their own trial was indicative of drug effects on HGSOC specifically, Sander’s lab decided to dig into the cell lines’ particular genetics. The CCLE records sequences for around two thousand genes for each cell line in its database, including levels of expression and copy number mutations. By taking that information, and cross-referencing it with whole exomes of 300 real HGSOC tumors from The Cancer Genome Atlas (TCGA) produced by the National Cancer Institute, the lab could chart the degree of relatedness between live ovarian cancer and the cell lines used to model it.
A Motley Crowd
The results were starker than the team had imagined. Some of the cell lines typically used in HGSOC trials weren’t HGSOC at all. “We had hypermutated cell lines that looked more like endometrial cancer cell lines,” says Schultz, “so possibly mislabeled and therefore misused. And there were some others that were typically used as models of high-grade serous ovarian carcinoma, but didn’t really resemble the genetics of the tumors.” Worse yet, the two most popular cell lines – representing a majority of published papers for this cancer type – both fell in this latter category, of cancers that better resembled HGSOC than anything else, but were highly atypical of the real disease.
This is important, because cancer is a very individualized disease. Drugs rarely work on a broad spectrum of cancers; instead, they narrowly target one or a few specific mutations that may occur only in a small minority of patients. Drug trials therefore need to be well-targeted to have a reasonable chance of translating into clinical success. But the Sloan-Kettering lab’s casual background checks, intended just to choose a cell line for a drug trial, indicated that, at least for this one type of cancer, most research in the field was targeted to peculiar mutants unlikely to ever be found in the real world.
A cancer cell line growing in culture
“We realized, wait a minute, this is actually something worth reporting,” says Schultz. “We didn’t even think this would yield a manuscript.” But the team’s findings had implications extending well beyond their own drug trial, and they decided to investigate further. To dig into the root causes of these cells’ genetic discrepancies, they dug into the medical literature for the initial publications that cited particular cancer cell lines. “Usually the original citation is pretty specific about where the cell line was derived from,” says Schultz. “But we found in some examples that those descriptions changed with time. That paper was cited by someone else, but then the annotation was changed… by mistake probably, and then other people now cited that new paper, and with that they cited that mistake.” In this way, even cell lines free of strange mutations could drift from their original purposes, and become adopted as widely-used models for the wrong types of cancer.
The lab’s work was published in Nature Communications this July, and included some good news. Twelve cancer cell lines from the CCLE were found to be exceptionally good models for HGSOC, meaning researchers wouldn’t have to scramble for brand new resources to perform quality laboratory studies on this cancer type.
Unfortunately, these twelve lines together accounted for less than 1% of the relevant literature.
Saving Cell Lines
Some researchers are trying to move past cell lines as cancer models, suggesting genetically engineered animals (see: A Fly of One’s Own
), or more lifelike cellular systems (see: The Tumor Organoid Biobank
), as more powerful alternatives. But Sander’s team would like to rescue cancer cell lines, which are easy for any lab to grow and maintain, easily transmissible from one lab to another, and deeply imbedded in universally understood laboratory protocols. The trouble with cell lines may not be that they’re inherently poor cancer models, but that they require some oversight to make sure they’re well-matched to specific cancers of interest.
What’s needed is a resource labs can visit to see which cell lines are best indicated for their studies. With the cBioPortal, Sloan-Kettering already has a convenient one running – if they could just get the data. “TCGA will soon have characterized twenty different cancer types,” says Schultz, “and the CCLE has cell lines from all different tissues, so we should really do this systematically for all tumor types.”
That’s easier said than done. The comparison of CCLE and TCGA records that Sander’s team published in July was finely tuned to the peculiarities of the HGSOC genome. HGSOC is an unusually homogeneous cancer, distinguished by its reliance almost entirely on mutations to the TP53 gene. Typical HGSOC tumors will exhibit very few mutations to genes other than TP53, BRCA 1 and BRCA 2. On the other hand, the tumor genomes tend to have unusually high levels of copy number variation. These distinctive genetic qualities could be weighted and scored to rank cell lines.
Other cancers will need their own eccentricities defined and translated into a scoring system before the lab can match cell lines to them. Moreover, most cancers have an assortment of possible base mutations instead of just one, meaning those cancers will have to be further subdivided to yield accurate models. Extrapolated to twenty cancer types, this is time-consuming, stepwise work.
“Now we’re in the process of developing an algorithm that does this systematically across tumor types,” says Schultz. The plan is to use the template of the lab’s work with HGSOC to write rules for describing which features tumors of the same type have in common. That way, a computer can automatically generate the typical features of a given cancer genome, then map cell lines’ genotypes to that model. This will standardize the analytical process, and also greatly speed the pace of cell line validations.
At that point, the lab intends to make publicly available both its full database of cell line recommendations, and the algorithm itself. Experts in a specific strain of cancer would be able to examine the algorithm and decide whether they agree with its evaluation of what is “typical” for their specialty’s genetics. And if they disagree, it will be easy to modify the algorithm to reflect their specialized knowledge, and rerun the cell line rankings accordingly.
Cell lines will never be ideal models for cancer, because each patient’s tumor responds to drugs differently. But for foundational research, they are still a crucial resource, and need to reflect the real spectrum of cancer genetics as closely as possible. “I think you should use cell lines,” says Schultz, “but you should also be aware of the genetic makeup of the disease you’re studying.”
Sloan-Kettering’s research in this area could help labs around the world rethink their standards for choosing a cell line as they embark on new drug trials. It could also make clear which cancer types lack good cell line models, and help prioritize the creation of new ones. Key to making a real impact on future research, however, will be spreading the message about which cancer cell lines are representative of their parent tumors, and which are genetic oddballs.
Considering how much crucial cancer research gets its start in the petri dish, it’s a message well worth sharing.