WIN 2016
Characterisation of somatic mutations using patient RNASeq profiles
Prof Andrea Califano - Columbia University Medical Center, New York, USA
What we presented was a series of results that have to do with how we can complement the existing paradigm of oncogene addiction and immunotherapy by using data from RNA sequencing from cancer patients that can be useful in predicting the specific drugs that these patients will be responsive to.
What did you find?
So we started from a very simple observation which is that really what determines the state of a cancer cell is not the mutations but actually the activity of the proteins. The presence of mutation in the corresponding genes is one possible way in which the activity of these proteins may be dysregulated but it’s not the only way. There are all sorts of things, for instance you may be exposed to oestrogen through normal hyper-production of the hormones or maybe through the pill for so you don’t have an actual mutation but you get aberrant signals that contribute to the activation of the corresponding protein downstream. So based on that observation we decided that wouldn’t it be nice if we had a way to actually directly access protein activity instead of having to infer it potentially from the presence of mutations. We developed an algorithm that has just been published in Nature Genetics, which has also been licensed exclusively to the Darwin Health Company, which allows us to essentially very accurately estimate the activity of a protein and in particular the differential activity, whether there has been an increase or a decrease in protein activity, from the expression of the protein targets. We have generated what you would call a multiplex gene reporter assay for each different tumour type, more than thirty different tumour types, and so we can now very accurately estimate the aberrant activity of about 6,000 proteins in more than thirty tumour types.
When will this become available?
You can bring it to your clinic and put it to use right now. We are working very hard to have within the next 60-90 days a New York CLIA certified version of the test. We have two tests right now that are generated from this approach, the first one is called OncoTarget and what OncoTarget does is essentially almost like a straightforward complement to what you would normally get by a mutational analysis. That is, in a mutational analysis you would get a list of the genes that are mutated and that are therefore potential candidates for therapy using the corresponding inhibitors. In this case you get a list of proteins that are aberrantly activated among the proteins that we already know how to inhibit with an existing drug. So, for instance it would tell you that HER2 is activated in this patient whether or not there is a corresponding mutation in that particular gene. The value there is that we see a number of patients that have absolutely no mutation and yet have totally aberrant activity of the corresponding protein. So, for instance, about 15% of lung cancers have activity of EGFR that are comparable to the patients that have the worst activating mutation in exome 20 and exome 19 etc. and yet have completely wild-type EGFR. So we would expect that these patients would actually be very good candidates for therapy using Tarceva or afatinib, another EGFR inhibitor. So that’s where this is helpful because while we normally find really actionable mutations only in about 25% of all cancer patients that present in the clinic using typical mutational landscape assays, using VIPER we typically find somewhere between five and twenty actionable proteins in 100% of cancer patients.
Will this improve personalised medicine?
Yes, actually the real issue is not that much one of restricting things but actually of having a slightly broader repertoire of choices because, for instance, 75% of the patients that we see in the clinic literally have no actionable mutation. They may have a KRAS mutation which is considered partially actionable because you can use a MEK inhibitor but MEK inhibitors don’t really work very well in KRAS mutant patients. You may have a p53 mutation and you give and again that’s not a drug that cures from the cancer, it may delay the ultimate untimely death. So the idea here is that having a broader repertoire of potentially actionable proteins that can be also associated with an inhibitor gives the oncologist a little bit more choice. Then, of course, these potential actionable targets need to be then vetted to a rigorous clinical study process. So we are already designing in collaboration with some of our major partners at MD Anderson or etc. where we’re already designing potential clinical trials where we would actually do the equivalent of what is called a basket study, that is identify what are these targets that come recurrently as being very highly activated in patients with a particular type of cancer and then basically designing a study that will test the corresponding inhibitors across that repertoire of targets.
So that’s the first technology, OncoTarget. The more exciting technology is the one called OncoTreat and what OncoTreat does is it starts from a very simple question. It says how is it possible that when you look at tumours that have wildly different genetics they actually have almost identical transcriptomic states. So, for instance, if you take triple negative breast cancers mutations are all over the place, almost no two patients have the same set of mutations, and yet when you look at their transcriptional profiles their spread is not very different from the spread of the normal tissue. So that’s surprising, it means that the cell, the cancer cell, is actually very tightly regulated and so it is legitimate to ask if this is this incredibly large landscape of tumour mutations that we see in triple negative and yet the state of the protein is so tightly regulated then there’s got to be a bottleneck, there’s got to be some proteins that are responsible for maintaining that state. We call those proteins master regulators and we call the little modules that they form, literally anywhere from two to about twenty proteins, we call them tumour checkpoints. We have shown that these tumour checkpoints really exist across every one of the tumours that we see in TCGA and that collapsing this tumour checkpoint either genetically or pharmacologically induces dramatic tumour regression in vivo. So what we’re now doing systematically, we can take an individual patient sample, or even a single cell from a tumour, and we can identify what are these proteins that maintain that tumour state, the master regulators. A very big surprise is that most often these are not the oncogene, the oncogene are actually further down in the list of the most activated proteins. The ones that are the most activated proteins are the ones that maintain tumour state. Then what we do is we basically test the entire repertoire of FDA approved and investigational compounds, about 450 drugs in oncology, and we ask is there any one that can completely invert the pattern of regulation that maintains tumour state, either alone or in combination. So far we’ve seen that basically in every tumour we’ve looked at there are somewhere between two to six of these compounds that we already have in the clinic that can actually do that. When we use them, for instance in a transplant of the patient or in some cases even in the patient themselves, we’ve seen dramatic results with the full abrogation of tumour viability in vivo.
Like trying to find a key that fits?
Yes, that’s exactly a good point. What we’ve been trying to do is to use one of those old-fashioned keys that were essentially flat, they had a single tooth and it was flat toothed, you turned the lock and it opens. So it turns out that cancer is a much more complex lock and requires a lot of different teeth, not just one. So the ability that we have to actually kill cancer cells by targeting only one protein, even if it’s the right protein, is limited and it has certainly only worked in tumours that are almost completely 100% driven by the aberrant activity of the protein like CML for instance. In most tumours targeting the oncogene only gives you a finite reprieve from the disease and eventually it will come back and will relapse and it will no longer be sensitive to the inhibitor. We are hoping that by targeting these more complex locks where you have maybe the ability to target anywhere from three to four to five to maybe twenty proteins at the same time, you actually now put the tumour in an irreversible state switch from which it can no longer recover. You’re targeting mechanisms that are downstream from the mutation so you are not offering the tumour the opportunity of escaping the therapy by basically selecting cells that have alternative mutations or alternative bypass mutations. So of course this is all theoretical, it’s very practical in the sense we’ve seen it functioning both in transplants from human tumours into immunocompromised mice and also directly in patients, but we need to still run very rigorous clinical trials to determine that this actually works in practice in the long run.
Do you have a summary message?
Yes, right now a couple of reflections that I have is that one of the things that came out from this research is that we have to spend a lot more time in understanding at a more precise molecular level the effect that drugs have in different cancers. We tend to, if you want, simplify intuitively the role of drugs by saying this is a MEK inhibitor or this is a HDAC1 inhibitor and unfortunately drugs tend to have extraordinarily tumour specific patterns of activity, we call it the mechanism of action of the drug. Really understanding more precisely how a drug participates, if you want, in modulating the dynamic of the cancer cell and also of normal cells has to become a really more fundamental area of investigation. Because we find very frequently through these studies with complete experimental validation that drugs that are thought to be exactly identical essentially, for instance two HDAC inhibitors, or an HDAC1 inhibitor or two MEK inhibitors, have profoundly different behaviour in the tumour because in fact when you look at their complete repertoire of targets and effectors it actually maybe overlaps what you consider to be the key target, maybe MEK or HDAC, but actually substantially different on the rest of effectors and targets. This is really important because we may be able to prioritise among a number of potential MEK inhibitors the one that really works in a particular patient with a KRAS mutation while the other MEK inhibitor may not work at all. This is because of this field effect that these inhibitors have, essentially they are not clean, they are very dirty small molecules, they hit a lot of different things in the cell and by the way they hit things depending on the context in which you put them. So the combination of that suggests that this new field which we call systems pharmacology really requires some very fundamental deep thinking and this is an area, for instance, in which Darwin Health is very, very much a player because we have developed assays to very precisely characterise compounds and are working with a number of companies to actually help them characterise their compounds’ pipelines.
