How can a systems biology approach aid precision cancer medicine?

Dr Andrea Califano, Columbia University Department of Systems Biology, New York, USA

What was the topic of your keynote lecture here at the WIN Symposium 2015?

The topic was the fact that we are increasingly working on studying how tumours respond to drugs using the genetics of the tumour. The foundation for that has been this theory of oncogene addiction, that is if you have a mutation in an oncogene then shutting down that oncogene will actually cause the tumour to die. Unfortunately this has really failed a little bit to deliver on promises; in some cases it works exceptionally well but in a very small number of patients.

So I was reflecting on what are the reasons why this has failed and what can we do to address that and, in particular, what causes tumour initiation and progression is actually not the individual mutation, it’s the specific pattern of aberrant activity of the proteins, a relatively small number of proteins. You can actually activate a protein by mutating it but that’s just one way to do it, you can do it in a quadrillion, quintillions of other ways – by mutating upstream regulators, interactors, all sorts of other things. So what we decided to do is instead of just measuring the mutation in the proteins or in the corresponding gene we should perhaps think of having better methods to understand why the proteins are aberrantly activated. We do that using the targets of the protein as the best reporter. Basically if a protein regulates some targets, if those targets are differentially expressed then this protein is active but if the targets are not differentially expressed then it’s not active. That has turned out to be a phenomenal predictor of drug activity and response, clinical response, in patients and has also generated a larger number of new discoveries that we’ve published over the years, including the ability to identify proteins that don’t work individually but in pairs so that one or the other doesn’t do anything but the combination really induces very strong tumorigenesis – silencing both proteins induced remission.

Could you give an example of one of these protein reporter systems?

This is, for instance, a clinical trial that we have just initiated in breast cancer. These are patients that have HER2 positive breast cancer that have failed therapy with trastuzumab which is the inhibitor for HER2. What we discovered using this approach, by studying all the regulator proteins that are non-muted in the cell that has HER2 amplifications, we discovered that a particular protein called Stat3 was actually aberrantly activated but never mutated. Downstream of these proteins there is secretion of a protein called Interleukin-6 that goes outside of the cell, binds the receptor and activates a pathway called the JAK-STAT cascade which ends up on Stat3. So you start a loop that is a really vicious loop because now you can turn off HER2 and the loop keeps going.

So by using another drug that is an inhibitor for the JAK kinase, which is called ruxolitinib, we were able to induce very strong synergistic response in patients. Now four patients have been enrolled in the trial, two have actually responded and one died for a completely independent reason and one has not responded was actually predicted not to respond because he had ER positive state and we predicted that this would only work in the ER negative state. So it is a very interesting way of doing it and there are many other examples.

Can you tell us more about the 260-patient trial now enrolling at Columbia University looking at nine rare or untreatable malignancies?

That’s a different trial. The 260 patients is a really unique trial which we call an N of 1 trial where every patient is actually treated with an independent strategy because we’re using that trial to actually understand patient response. So for every patient that is enrolled in that study, which covers nine rare or untreatable malignancies, we are generating a complete map of the protein or proteins that the patient’s tumour is dependent on and then we figure out how to match them to very specific drugs or drug combinations that can be tested in a mouse where the patient tumour has been transplanted. Then we can advise the physician, treating physician, about the drug or drug combinations that worked.

What other research you have conducted in this area?

Some of the really interesting research that has now profound clinical implication is in the identification of biomarkers that can predict, for instance, whether patients with a prostate cancer will actually die of their disease or not. In prostate cancer overtreatment is one of the major problems and, in fact, most prostate cancer patients received either therapy or, most likely, surgery that they may not have needed. So what we did in that case is a collaboration with Michael Shen at Columbia. We showed that by using this approach, by looking at the target of the protein, we were able to discover two proteins, one called FOXM1 and one called CMPF that in isolation do absolutely nothing but when they are co-active in a tumour they are a harbinger of the worst possible response because they actually regulate the entire set of programmes that are associated with metastatic progression etc. So if you shut down one or the other nothing happens in this tumour; you shut down both of them this tumour dies. More importantly, at diagnosis even in a cohort of 900 patients that were followed up for twenty years you can see that the patients that had the double negative, that had no activity of either protein, by it’s called immunohistochemistry, where essentially none of them died, literally about maybe three patients died. In the cohort of the double positive, which had basically both proteins active at diagnosis, that represented 90% of the burden of the tumour with an extra 10% represented by the single positive of one of them called FOXM1.

So this basically tells you that, and in fact now it has been developed into a clinical biomarker, it has been licensed, and the idea is that these essentially represent one way in which at diagnosis you can predict over a far window of opportunity of many years whether the patients will have a very aggressive type of tumour or not.

Can you explain the title of your lecture, what is systems biology and how can it be used in relation to precision cancer medicine?

The foundation of this entire approach that I told you is that we can actually measure the activity of the protein by measuring the expression of its targets. That is like saying that I can figure out who is the person that committed a crime but is sitting in the shadow by knowing who was at the crime scene and knowing who are the friends of that person. Now, the big problem that people have not yet been able to solve is to figure out in each cancer context what are the targets of each protein. It’s called systems biology because it means that you have to build a regulatory network that has millions of interactions where every regulator protein out of 6,000 proteins that are either transcriptional regulators or signal transaction regulators is associated with somewhere between fifty and a thousand targets. So that analysis is done with algorithms that now have been extremely experimentally validated in the lab and that typically have 70-80% accuracy in predicting true physical targets of the proteins. Only when you have these targets resolved in each tissue type, each tissue corresponding to a tumour, you can now apply this approach. Without that all the different assays that we have to measure protein activity are not really very good because measuring the RNA is many steps removed from the protein activity. Even measuring protein phosphorylation is very noisy and actually has very large variability, even within a sample of the patient. Because if you leave a sample out for five minutes versus ten minutes the phosphorylation state on the protein will change. So the actual expression of the targets is one of the most stable and most reproducible predictors of the aberrant activity of proteins.

On a practical level, how do you use a systems biology approach?

So what we do is the following. First of all we build a model; we have to build this model of about a million regulatory interactions. For that we use a large cohort of patients, typically 200, 300, 1,000 patients have been made available, say by TCGA. We now don’t need to do that anymore because we’ve built so many of these models that we can actually take a tumour that has never been… that is very rare, for instance, and there’s not enough patients, and we can use the combined knowledge in all these models to predict what the targets are. Then we take the sample of that particular patient and we run it through the model and we ask if this is the differential expression of the genes in this particular patient and these are all the interactions then what are the proteins that are in use in that differential expression. That gives us a very short list, typically about twenty activated and twenty inactivated proteins that we call the master regulators. We can now match those to the response of drugs, so we take a large repertoire of FDA approved developmental drugs and we ask the opposite question, we ask which drug takes the specific targets of the protein that we want to shut down and completely flips their status. So if they were overexpressed now they’re under-expressed, they were under-expressed, now they’re overexpressed, because those drugs will actually inhibit the activity of that protein. So we can match a drug to the activity of all of these proteins; we can match drug combinations to the activity of all of these proteins and what is important is that when you actually hit the right targets the entire master regulator signature collapses. So you have an effect where you don’t have to target a single protein but you’re actually targeting an entire module of proteins that all work together to support the activity of the tumour.

So this has actually shown that you can with pinpoint precision identify drugs by studying only the human tumour that you can then put in a mouse xenograft that is in a patient transplant to an immunocompromised mouse and will actually put the mouse in remission without ever having been screened in cell lines or in mouse models.

Do you have a final take-home message from your presentation?

The take-home message is that we have to abandon what is easy to do and right now discovering mutations is very easy to do because sequencing technology has become extremely cheap and really buy more into more integrated approaches that bring both the mutational status, the status of proteins, the status of RNA etc, and use that as a compendium of information to give us a full perspective on the vulnerabilities of tumours. This actually doesn’t just work in tumours, we just published the same exact methodology in Parkinson’s disease, in ALS, in Alzheimer’s, in alcohol addiction and in stem cell pluripotency. So this is a very general paradigm, it’s not just restricted to cancer biology. So if you use this integrated approach you tend to be able to extract more information but some of them, unfortunately, are complex, they’re not easy and that’s why we have to use a systems approach to bring this to the patient.