Elucidation and pharmacological targeting of tumour checkpoint dependencies
Dr Andrea Califano - Colombia University Medical Centre, New York, USA
It’s remarkable how cancer goes through phases and there are phases of tremendous excitement and then all of a sudden we are brought back to the harsh reality of the complexity of the disease. We started with metabolism a long time ago, it’s probably one of the very first targets for cancer - everybody thought we could defeat cancer through targeting metabolism, it didn’t work that way. Genetics has played tremendous excitement in the community in terms of identifying through a mechanism called oncogene addiction that we would be able to just target genes, oncogenes, that are mutated and therefore kill cancer cells. But the latest results, large scale trials, have shown that maybe between 5-10% of the patients that are treated with these targeted inhibitors based on mutational data actually respond and the rest have no long-term benefit. Finally there is immunotherapy and, again, tremendous excitement and I am myself extraordinarily excited but we have to remember that actually still the majority of cancers are not immunogenic, meaning that they still evade the mechanism of detection because they don’t present antigens.
So we’re left back to going back to the drawing board and trying to understand really how cancer works. This is where we come in just trying to target the fundamental mechanism of the behaviour of the cancer biology by reconstructing the entire assembly manual of the cancer cell and then finding the universal entry points that we’ve called tumour checkpoints or master regulators that can be targeted with drugs to induce cancer death.
I’m trying to pitch this as a complementary approach, not just the approach that will replace them all and will be the approach. But I want to make sure that what we understand is that right now there is a fundamental paradox in cancer and that’s what we’re trying to address. So if you take, for instance, triple negative breast cancer, from the state of the cancer cell, which is represented by something called the transcriptome, these cancers are virtually identical, they look exactly the same. If you, however, look at the mutation of these cancers they’re all over the place. There is literally almost no two patients that have the same exact driving mutations. So you’ve got to ask yourself the fundamental question – how is possible that you have all these different mutations and yet in some way they generate the same exact tumour state? So what we’ve discovered is that not just in triple negative breast cancer but in a variety of different tumours, I think we’ve published probably about a dozen of them, and now we’ve done it for every single patient in TCJ, so about 12,000 patients that are [?? 2:49] large enough for the [?? 2:50] analysis, that in fact there is a fundamental mechanism we call the tumour checkpoint that integrates all the genetic alterations, essentially think of it as a colander with all the mutations being the holes and what people are trying to do is just block the individual holes with their hands, which is very hard. What we are discovering is there an actual funnel underneath the colander and just plugging one single whole can essentially capture the entire result of all these different mutations. So that’s what we’ve now discovered – there is essentially a very small number of these tumour checkpoints, according to our analyses 93 different tumour subtypes and only 28 tumour checkpoints and each one of them explains one of the characteristics of cancer and each one of them is eminently druggable. So, in fact, we have just launched a Columbia study called the N-of-1 where we automatically match the tumour dependencies that are initiated based on the activity of these tumour checkpoints with specific drugs or drug combinations drawn from the repertoire of FDA approved and experimental compounds, late stage experimental compounds, so we can bring them directly to the clinic, so phase II and phase III compounds. And every single tumour we’ve looked at we do find certain combinations that seem to completely collapse the tumour checkpoint and cause tumour loss in vivo. There has been quite a striking set of responses that we get from this analysis, typically the top six drugs that we have prioritised using this method, anywhere from two to five of the top six have actually worked in vivo extremely well.
Acting before cancer manifests?
Before and after because unfortunately what happens is that we think of cancer as a collection of heterogeneous cells because of the mutation but it’s actually not just that, it’s more than that, it’s a collection of heterogeneous states. So one of the things that we’re showing, and actually other labs have discovered these as well, is that within the same tumour you don’t just have different mutational states but representing roughly the same tumour type, you really have almost completely orthogonal tumour states and that’s why certain drugs that are very effective, for instance to cure, say, a luminal breast cancer, all they do is they actually shift the cell to a more basal triple negative state which is now resistant to that particular set of drugs. So there’s something called tumour plasticity which allows cells without changing the genetics to re-programme themselves into states that are fundamentally different.
How are you applying this approach?
Yes, it does actually have quite a number of applications. First of all, if you build a complete genome-wide model of how the cell works, how the cancer cell works, you can use that model to do anything from understanding how drugs work, so something called elucidation of the drug mechanism of action. That’s a very important problem for the pharmaceutical industry because the mechanism of action of a drug not only explains how it works but also what could be potential toxicity that you get downstream. It also explains why certain drugs that have been built to do almost exactly the same thing, for instance, to, say, MEK inhibitors may or may not work in exactly the same patients. This is because no drug is actually hitting a single target with the same affinity, there’s an entire distributional repertoire of targets within the cell that they hit.
You can use this approach to discover what are the optimal biomarkers for some tumour progression or tumour prognosis or response to drugs and that’s very important because the mechanisms you discover in this way are the ones that are really in charge of regulating the cell rather than the ones that are simply statistically associated with the endpoint. Statistical association is a great concept but does not imply causality and so sometimes you find things that are just there and you think that they are meaningful but they’re really not. In this case what you find is actually directly implicated into the presentation of the phenotype and therefore if you use that as a biomarker it’s what you would call a mechanistic biomarker. It’s important because it’s helping us to define a completely novel class of targets, of therapeutic targets.
So far what we’ve been trying to target are proteins that are oncogenes, so these are involved in the initiation of disease. It turns out that once you initiate a tumour cell state, the tumour cell state becomes somewhat independent of these initiating events. A very interesting example which is now a clinical trial is the idea that in HER2 positive breast cancer once you initiate the tumour with the amplification of ErbB2, which is the surface receptor, you now get an autocrine loop which goes to a protein called STAT3, IL-6, IL-6 goes out of the cells, binds the receptor, activates a JAK kinase and then finally activates STAT3. So now you have this loop going - you can turn off the oncogenic signals and the loop will keep going. You’ve essentially, literally, switched the cell into a new state and that state, like most cellular states, is incredibly stable. So you’ve got to figure out how to destabilise this new state that the cell discovered and we do that through a combination of two drugs, one hitting HER2, the other one hitting the JAK kinase which is called ruxolitinib. So that’s a clinical trial that is now enrolling patients with some very exciting results. So I think that the idea that cancer is a state and rather than figuring out what initiated it we ought to figure out what actually keeps it in that state is certainly a novel idea that needs to be further explored.
First of all if what we’re saying is reasonable, or at least in some of the cases, it turns out that the proteins that we actually need to target are different from the proteins that we’ve been targeting so far. These proteins are the ones that are at the core of maintaining the tumour state. So that means that we may want to start thinking about developing novel classes of therapeutic agents and reusing the ones that we have to achieve different endpoints, not the endpoints, for instance, of viability but the endpoints of really abrogating the activity of these tumour checkpoints.
The other thing is that we need to start thinking about cancer at the individual level, not just the individual patient but the individual cell. What we find by looking at tumour checkpoints is because they act as the fundamental integrator of the funnel underneath the colander, if you want, they actually are much more universal than the individual mutations. Think of it this way – there’s 20,000 genes, there’s 220,000 possible states of these genes you could have if they were just on or off or mutated or non-mutated; there’s only two to the 400 atoms in the universe, you do the math, it’s essentially impossible to numerate that. Now if you go within one disease, let’s say cancer or obesity or whatever it is you’re interested in, maybe there’s a hundred to a thousand genes that are relevant that could be mutated. So now you only have two to the hundred to two to the thousand possible states of the cell, that’s still an extraordinary number. So unless we have some kind of simplification logic that allows us to draw some more universal conclusions we simply cannot explore these patterns one at a time. Most of them won’t even exist because there aren’t enough cells on this planet to implement them all and the one that exists will probably be in a single patient. So we need to figure out how to understand cancer using a mechanism of greater universality which we think these tumour checkpoints provide because that’s where whatever pattern of mutation you have you’ll collapse into aberrant activity of one of these tumour checkpoints.
So the idea that we can study things at the individual patient or individual cell level means that we can actually keep finding the same thing over and over again but can figure out in which patients or in which cells we have a certain dependency versus other dependencies. That’s going to be a crucial key answer to defeating this disease once and for all. As Joe Biden says, defeating this cancer once and for all.
Are you working together with any other organisations?
Yes, so first of all we’re working with many organisations, we have a study open right now on endocrine tumours, we’re collaborating with Dana-Farber, with Sloan-Kettering, Columbia and actually Heidelberg now. We have several studies where we are recruiting patients, it’s called the N-of-1, we’re recruiting 260 patients across 14 different malignancies. So this is, I would say, one of the first true pan-cancer studies. It is completely not based on genetics, we don’t even do a genomic analysis, we do an RNA-seq analysis and from the RNA-seq we can predict what are the tumour checkpoints that we need to target. This has actually had already some pretty striking results in the actual clinical setting and certainly in the preclinical setting we keep seeing very, very exciting results. So we are expanding this, there are a few organisations that are really excited about working with us and we are excited about working with them, it’s a little bit too early to declare names but what I can say is these are some of the institutions that have already run very large genetic trials and have seen that the signal is actually weak. It’s there, it’s doing something viable, like 5-10% of patients do get a long-term clinical benefit but most of the patients still fail to respond. So they know that they have to complement these genetic studies with something else and they’re very excited about using the type of technology we developed by using cancer system biology approaches to understand cancer.
Do you have a take home message?
I think so. If you want to take a step back, one thing I would I would like to mention is that almost every science, whether it is chemistry or physics or whatever science you want to think of, meteorology, economics, has had a golden age where the ability to understand things empirically by experimentation has been met by the ability to actually model the underlying reality. So, for instance, when Galileo started to model or Newton started to model how a ball would fall on an incline you don’t even have to do the experiment, you can simply model the experiment and then you can draw some sort of conclusion that would lead you to make predictions that can then be validated experimentally. In biology we’re still doing a tremendous amount of work using associations, that is statistical relationships between the data which are not relying on an underlying model. So one of the things that distinguishes the kinds of things that we do is that we are, for the first time, trying to really use underlying models of behaviour to model how the cancer cell works and to predict very effectively what are the actual therapeutic approaches that can target the dependencies that are created because of the model. So it’s not because of one mutation or another mutation it’s because of the concerted activity of all the mutations that make up the cancer cell and all the epigenetic signals and all the extra-cellular signals that the cancer cell is in that state. If we can model it and understand it we have a much better chance to really target vital core entry points of the cellular machinery than if you actually simply study is it raining then people have umbrellas open which is what we’re doing right now with association [?? 14:11] approach. So I think modelling is changing the way we think about biology and this is just the onset of the change in the discipline that is going to have an increasingly large contribution.