Investigators led by a group of scientists from Harvard Medical School have developed a high-throughput platform to profile 500 cell lines derived from diverse epithelial cancers to help them predict their sensitivity to 14 kinase inhibitors.
Although the researchers found that the majority of cell lines were largely refractory to each inhibitor, they also noticed that a small subgroup of very sensitive cell lines could be identified for most inhibitors.
In addition, although cells with sensitivity to EGFR, BRAF, MET, or HER2 were marked by activating mutations or amplification of the drug target, the screen revealed low-frequency drug-sensitizing genotypes in tumor types not previously associated with drug susceptibility. Furthermore, a comparison of kinase inhibitors thought to have the same target revealed striking differences predictive of clinical efficacy.
Writing in a recent issue of the Proceedings of the National Academy of Sciences, the researchers concluded that genetically defined cancer subsets, irrespective of tissue type, predict response to kinase inhibitors and provide an important preclinical model to guide early clinical applications of novel targeted inhibitors.
Jeffrey Settleman, a professor of medicine at Harvard Medical School,
director of the Center for Molecular Therapeutics at Massachusetts General Hospital Cancer Center, and the corresponding author on the paper, spoke with CBA News recently about the work, its applicability to drug discovery, and how it could save drug makers money.
Can you give me some background on your work?
As a starting point, we are interested in the concept of personalized cancer medicine. The idea is that cancer is a very heterogeneous disease, at the tissue level as well as at the molecular level.
The genetics of tumors, it is becoming very clear, are playing an important role in determining the response to drug treatment. This is something that is becoming more appreciated with time. It is particularly true with these new targeted therapies, such as Gleevec [Novartis’ tyrosine kinase inhibitor for chronic myelogenous leukemia, gastrointestinal stromal tumors, and a number of other malignancies] and Tarceva [a tyrosine kinase made by Genentech to treat non-small-cell lunch cancer and pancreatic cancer], as opposed to the conventional toxic chemotherapy drugs, like taxol and platinum.
We developed a platform to study this relationship between tumor genetics and drug responses by modeling those responses in cell culture, where we could collect a large number of tumor cell lines. The idea was that we would capture the genetic heterogeneity that exists within the cancer patient population.
By looking on a large scale at a large number of cell lines, and looking at their differential responses to some of these targeted drugs, we were able to demonstrate that there are, in fact, genotypes that seem to correlate closely to sensitivity to these drugs in culture. We can use that information to potentially optimize the use of these drugs and eventually match patients with the appropriate therapy based on their underlying genetics, and make those predictions about response prior to treatment by looking at the genetic features of an individual’s tumor.
This paper was essentially a proof of concept that cell lines are a good model for genetic heterogeneity in cancer patients that seems to predict therapeutic response. The real proof of concept came with the use of the drug Tarceva in our study. We were able to show that sensitivity was well-correlated in lung cancer cell lines that had EGF receptor mutations, and that is a relationship that has already been demonstrated clinically in patients. Those with EGF receptor mutations are much more likely to respond to Tarceva than those without those mutations.
We were able to show that relationship holds true in cell lines as well. That is important because a lot of people feel that cancer lines may not accurately represent human disease. I think one of the key take-home messages from our report is that cancer cell lines can, in fact, be good models for looking at the genotypes that are directly responsible for response to at least some targeted cancer drugs.
We went on to look at some of the other drugs that are being explored clinically and found similar relationships with specific genetic features of tumor cells. Applied broadly, we feel that this is a mechanism or a platform that all pharmaceutical companies should be using with their newly developed anticancer drugs before going into the clinic, to get a sense of how those agents are likely to behave clinically, and possibly to identify a patient population that should be treated with those drugs to optimize the likelihood of response.
How would this screen be applicable to drug discovery?
We have had a lot of interactions with pharmaceutical companies, by the way, whereby they hear about our platform and bring us their compounds. We have established relationships where they are interested in examining the profile of their drugs across a large panel of cancer cell lines. They see the value in that.
Specifically, what we can be bring to the table is information that goes beyond what drug companies typically do with their new compounds, which is to look at just a few cell lines and maybe a couple of mouse models. We feel that provides a very limited perspective on how the drug is likely to behave once it gets into people. We feel that looking at a large panel of cell lines adds another important dimension to the preclinical evaluation of new drug candidates that would be difficult to obtain through other mechanisms.
We cited the example in our paper of how a drug, in this case sorafenib [marketed as Nexavar by Bayer], a clinically approved drug for renal cancer, was initially believed to be an inhibitor of the BRAF kinase, which is implicated in melanomas in particular. It turned out that sorafenib was tested in the clinic in melanoma and failed. It also turns out that it failed because it is probably not a good BRAF inhibitor.
We showed in cell lines that, in fact, it is not a good inhibitor of BRAF, and it does not work against melanoma cells. Had we known that before the clinical trial was launched, we probably would not have pursued a clinical study of that drug in melanoma. That is an example of how this kind of analysis can help guide the clinical trial design, which is relevant no matter what stage of pre-approval testing you are talking about.
Ultimately, you have to figure out how to get the drug into the clinic, even if it looks good preclinically. Many drugs that look good preclinically fail when they get to the clinic. That is an important bottleneck, because clinical trials are very expensive and many of them fail. It is a cost-saving measure for companies to learn as much as possible about a drug before designing a clinical trial.
That is where I feel a lot of the value of this technology is to the drug companies. They can learn a lot more about where their drug is likely to have clinical activity.
Are you performing these assays for drug companies on a fee-for-service basis?
That is not our preferred mechanism. We have a lot of sponsored research agreements. We do, however, get into the issue of intellectual property and proprietary information.
When you are talking about preclinical agents from companies, there is often a lot of protected information that is unavailable for public disclosure. Unfortunately, that conflicts with our academic goal, which is publishing our findings. So we do occasionally run into conflicts where we just cannot work with a company and pursue these kinds of studies, because of the constraints imposed by intellectual property.
These are pharmaceutical companies and biotech companies?
Both. I cannot name them because nondisclosure is often a part of the fine print in these agreements. Basically, we have had some connections with every major company that has an oncology drug pipeline.
Could this screen be used to test compounds for other kinds of diseases?
Not really. Certainly, some diseases are like cancer in that drugs to treat them may work in some patients but not in others, due to underlying genetic differences. Cancer is a pretty unique case in terms of this kind of approach, so I think that we are talking about something that pretty specifically applies to cancer drugs.
What is the next step in this research?
We would like to expand the studies to include as many drugs as possible, because I think that with every new drug we test, we learn more and more about the differences in tumor cells that affect their sensitivity to drugs.
We are constantly grappling with the problem of acquired drug resistance. Even in cases where drugs work really well, they generally work only for a short time. That is the problem with cancer — it often responds initially, but then becomes resistant. We have to better understand the mechanisms by which tumors become resistant.
We can use cancer cell lines as an important tool for modeling resistance in culture. That is an important area for further study.
Do you have any future publications planned?
Yes. We have a paper that is in press now in Cancer Research. It relates to inhibitors of a kinase, ALK. It has been getting a lot of press lately, because it has been implicated in the treatment of a number of diseases including neuroblastoma, a type of lymphoma, and some types of lung cancer as well.
We have been profiling of tumor cell lines on a large scale with inhibitors of ALK. We have found also found some interesting relationships between tumor response and the underlying genetics.