Scientist Stories: Phil Jones, How cancer associated mutants colonize our tissues and what we can do

Welcome to the next one in our Sanger seminar series. My name is Phil Jones and I'm a senior group leader at the Wellcome Sanger Institute, and today I'd like to talk to you about what I've called inner evolution. As we get older, we inevitably get more and more mutations in our cells, and some of these mutants will spread through our normal tissues as large clones. Competing for space and potentially promoting diseases like cancer. This process of mutant competition has parallels with evolution, and we, and we may be able to intervene to deplete the mutants that are detrimental. So in this talk, I'm going to outline what we know about the competitive selection of somatic mutations in humans, focusing on one epithelial tissue, the esophagus. And then I'm going to talk about the mechanisms by which some mutations colonise the epithelium, talk about the good and the bad aspects of somatic mutation, and then how we can change the environment and potentially deplete mutations that are undesirable. So it's become very clear over the last few years that most of the epithelial tissues in our body carry progressively more mutations as we get older, and this is just some of the literature on this theme. Many of these mutations that spread through our tissues are linked to cancer and some to other benign diseases. Today, for reasons of time, I'm going to focus on one of these tissues, the oesophagus. So this is a view of the human oesophagus, which connects your mouth and your stomach, And we're going to be focusing on the pink lining of this tissue that you can see around the central lumen. A few years ago, we wanted to understand whether there were mutations in, in this pink lining epithelium, and the way that you can do this is to use DNA sequencing. So what you can do is to take a small piece of the epithelial lining—that's shown here as this pink square—and and let's imagine it contains two clones, a blue clone and a green clone, both mutated in the same gene. If we sequence a small piece of epithelium very deeply using targeted sequencing to get many-fold coverage of the individual genes we're interested in, and then align the reads, we would expect to see a small proportion of reads carrying carrying the mutation in the blue clone and a slightly larger proportion of reads carrying the green clone mutation. And because this is normal tissue and essentially diploid, we can extract the clone size from the variant allele frequencies of these different mutations. And how we use this in the oesophagus was we collected grids of small pieces, each about 2 millimetres squared, of the epithelium, and then we sequence very deeply for 74 or 300 genes, and these genes were the usual suspects that we find in cancers. Using that data, we could firstly construct spatial maps connecting clones that span different grid squares and essentially producing an array of a physical map of mutations, as you see here. And this data also gives us information on mutation prevalence and on genetic selection, which I'll come back to in a moment. The other information we get from these sort of studies are mutational signatures. So those are a hint as to the causes of the DNA mutations. For example, in the skin, we see ultraviolet light as the main cause, but in the esophagus, most mutations are due to cell-intrinsic processes linked with aging. And then finally, for some of the larger clones, we're able to study copy number alterations and understand what clones where there's been loss of both alleles of a particular gene. So this is a very brief summary of a lot of data. We looked at some young people and some older people, but I'm just going to show you two examples. So on the left, we have a man in, in his 20s, and on the right, a man in his 70s. Both are non-smokers, and the coloured circles here are a representation of the size of clones. And imagine you're looking top down on one square centimetre of epithelium. This is what it would look like if you could see the mutant clones. And you can see that by the time you get to 70, this man is heavily colonised by mutations, and several of these are common in cancers of the oesophagus. So there are lots of mutations, but do they matter? Do these mutations actually alter cell behaviour to drive clonal expansions? And this is the first, if you like, of our evolutionary parallels. So a widely used technique in molecular evolution is to look at the ratio of mutations that alter protein structure compared with those that are silent. And as you'll know, a silent mutation almost always does nothing to protein function, but missense and nonsense or other protein truncating mutations have a major impact on the function of a protein. And we can talk about positive selection as being when you have an excess of these protein-altering mutations over silent mutations. So for a gene that doesn't alter cell behaviour, in this context, this dN/dS ratio will be 1. However, when we look at the genes that are prevalent in the oesophagus, we can see very strong competitive selection. So for example, for nonsense mutations in NOTCH1 and TP53, the ratios are approaching 100. And alongside these very strongly selected genes, we have a tail of multiple examples of other more weakly selected mutants. For this talk, we're just going to focus on a couple of these mutations, and the first is TP53. This is, if you like, the genomic spell check responsible for genome integrity maintenance. But it's very striking that in squamous cancers, which are the kind of cancers that developed from the part of the oesophagus that we studied, it's almost universal for them to be mutated for P53. However, if you look at the normal tissue, you can see that until you get to your 70s, the prevalence of P53 mutants is really quite low, and then it rises as people get beyond 70. So this argues that P53 is a gatekeeper to squamous carcinoma, that the cancers are developing from the small population of P53 mutant cells that lurk within the normal epithelium. Epithelium, but that P53 is strongly enriched as you progress from normal to cancer. A counterexample is the gene NOTCH1, and this is the most effective mutant at colonising the oesophagus and pretty much any other tissue that we know of. NOTCH1 is really quite rare in squamous cancers of the oesophagus and is much more prevalent in normal oesophagus in people aged 40 and above. And this relative prevalence, that there's far more NOTCH1 in the normal than in the cancer, suggests, hints, that maybe NOTCH1 is protecting against the process of malignant transformation. More recently, we've got on to try and visualise some of the mutant clones within the epithelium. And here Emily Abbey in our lab has studied NOTCH1 mutations, and what she's done is to take tissue sections and stain them for NOTCH1, shown here in green, and NOTCH1 is expressed in the lower layers of the tissue. And what Emily found was that you got large runs of positive staining alternating with areas where there was no staining at all, where Notch1 protein had been lost. And then what she did was to dissect out the positive staining areas and the negative staining areas. In the areas that lack Notch protein, it becomes very clear that there are protein disrupting mutations affecting one or both alleles and/or there's copy number alterations, copy-neutral loss of heterozygosity. So where we see an absence of NOTCH1, both alleles have been disrupted. But even in the positive areas, the majority of samples had mutations in one NOTCH allele, and there were also examples of copy-neutral loss of heterozygosity. So NOTCH is majorly disturbed by mutation, NOTCH1, in normal epithelium. And really what's most striking about this image is that there's absolutely no visible difference between the tissue that lacks NOTCH1 and the tissue that has it. Despite NOTCH1 being a very good clonal coloniser, it doesn't disturb the structure of the tissue. So these studies in humans have given us the insight that normal tissues, including the oesophagus, can be very heavily colonised by mutations which are not neutral but are under strong genetic selection, and these appear to accumulate inevitably with the passage of time. So how is it that we end up here with this very high burden of mutations in what appears to be a completely normal tissue? That's what I'd like to move on to now. To understand this, we needed to move from studying humans to studying transgenic mice, where we have a elegant set of genetic tools that let us resolve how mutations alter cell behaviour. But before we get to that, I need to give you a crash course in the mouse oesophagus. And like the human oesophagus, the mouse oesophagus consists of sheets of a single cell type called the keratinocyte. It's a very simple tissue, very uniform structure, has no appendages or glands. It's just layers and layers of keratinocytes.. And the keratinocyte life cycle is, if you like, that they're born by cell division in the bottommost cell layer of the tissue. The divisions generate daughters, some of them go on to divide, some of them will decide to differentiate, and when they differentiate they exit the proliferative layer, migrate towards the cell surface, and then they're shed into the lumen of the oesophagus. And this process of shedding goes on throughout life, so there's a continual need for more cell divisions to replace the cells that are lost. And across the whole of life, this is beautifully balanced, so exactly the right number of cells are made to replace those which are lost. There's a critical balance not only in the cells that are shed and the cells that are made, but also in what's going on in this dividing cell layer at the bottom of the tissue. And if the tissue is going to be sustained, it has to be true that on average a division will produce one cell that goes on to divide, maintaining the dividing cell population, and one cell that will differentiate and leave the dividing cell layer. If that balance goes wrong and you make too many dividing cells, you end up with a tumor. But if it goes slightly the other way and you systematically produce too few, you'll end up with tissue failure and an ulcer. So we wanted to understand how this beautiful balance was achieved in the mouse oesophagus, and the way we've done this is to use a technique called lineage tracing, where we can use a set of transgenic tools to genetically label scattered single cells in the dividing cell layer of the mouse oesophagus, shown here as the yellow cell on the left. And then as this— as time goes by, this yellow cell might divide and it might produce a group of yellow-labeled cells. And by generating multiple labeled cells and then analysing oesophaguses at different time points after we've induced labelling, we can get a set of family trees, if you like, showing how single cells have behaved. So an example of real data is shown here. We use 3D imaging at single-cell resolution to reconstruct individual single-cell clones, and this clone was labeled— comes from a cell that was labeled about a month ago, and it's divided to produce 2 cells which remain in the dividing layer, arrowed red here, and 3 cells which are flattened like dinner plates with the white arrows, and these are differentiated and they're moving towards the surface of the tissue from which they'll be shed. We can also see some other examples like this, and this is a 3-cell clone where it's— where there are no more remaining cells in dividing layer, so this clone is inevitably going to move to the surface and will be lost from the tissue in a matter of weeks. To show you a bit more about this sort of experiment, here are top-down views of a sheet of oesophagus that we've removed from a mouse at different time points. And we start at 24— at the 24-hour time point, you can see a single cell labelled with this genetic label, in this case a yellow fluorescent protein. And by 10 days, we've sort of zoomed out now, you can see many, many small clones with multiple cells. And those clones get progressively larger as we go out to a year, but the number of clones that remain gets smaller. And if you look at the graph here, you can see that's really rather nicely that the proportion of labelled epithelium, which started at just over 1%, finishes a year later still at 1%. So the population we've labelled is a self-sustaining population and reflects the behaviour of all the dividing cells in the tissue. And we can use other transgenic tricks to get other insights into cell behaviour and then formulate this data together in a quantitative model. And the key insight that comes from this exercise is that all the dividing cells are essentially equivalent, they're all the same, there's one population of dividing cells which we could call progenitors. And when a progenitor cell divides, its daughters might either be another progenitor or might be a differentiated cell that's destined to be shed from the tissue. And a crucial insight that you need to understand how mutations spread through tissues is that the fate of a cell can have 3 options. So a dividing progenitor can produce 2 progenitors, 2 daughters that will differentiate, or 1 cell of each type. And at each division, the outcome is unpredictable, but across the population, the odds, the likelihood of producing these kind of divisions, is balanced. So we can think of it as a game of dice, as I've cartooned here, where if you throw a 1, you have 2 dividing daughters, and a 6, 2 daughters that differentiate, and a 2 to a 5, 1 cell of each type. Now, statistically, this kind of behaviour is named by statisticians Gambler's Ruin, or geneticists would call it neutral drift. And we can see how it works out in an example of two progenitor cells shown here. So these cells are now going to behave according to this model. At the first division, this cell is thrown a 2, and it has one daughter of each type, but then it throws a 6, and now all the, all the cells in this tiny little group of cells have differentiated. So if this cell was carrying a neutral mutation, that mutation would be lost from the tissue, and indeed most cells that develop a neutral mutation will be lost from the tissue within a few rounds of cell division. But that isn't always the case. So here's a cell that divides, but then it gets lucky—it has two progenitor daughters—and as time goes on, again it gets lucky again, and now For this clone to be lost in the next round of division, you'd have to throw 3 6s, an unlikely event. So by chance, most clones are lost, most neutral mutations are lost, but a few would expand and persist in the tissue. What happens if we get a mutation that rewrites the rules of the game? This is beautifully illustrated with a mutation that we studied a few years ago a mutant of a gene called Mammal1 that regulates the Notch signalling pathway. And on the top is the same example of the neutral label I showed you before, but the lower row shows you a similar fluorescent protein hooked up to a mutant of Mammal1. And as you can see here, the pattern changes dramatically. We get huge clones appearing and these get progressively larger and larger till they fuse, and by a year the entire tissue has been replaced by mutant cells. Despite this, and very strikingly, the mice remain completely well, and as in humans who have extensive NOTCH mutations, there's no gross difference in the tissue structure and there are no tumours formed. So when we went further into this, we were able to show that early on in this colonising mutant had tilted the odds in favour of producing proliferating daughters over non-proliferating daughters. It changed the rules of the dice game and loaded the probabilities in favour of proliferation, and that means that each round of division on average you'll have more and more dividing cells, a bit like compound interest. And the mutant population will expand and it will spread through the tissue. And this is true not only of the mastermind mutant but also some other mutants found in human esophagus and linked to tumours. So we've studied the PI3 kinase mutant that forms large clones in human esophagus and is found in cancers. P53 mutants behave in this way, and also Notch mutants, where we lose one or both alleles of Notch, they expand by altering the fate balance of the cells. It doesn't stay like this, however. If you remember that green oesophagus that had been completely replaced by mutations, once all the mutant cells are surrounded by themselves, by mutant cells, once the whole tissue has been replaced, the behaviour changes and the mutant cells revert towards balanced cell production similar to that seen by wild-type cells in homeostasis. And this reversion towards balanced behaviour stabilises the mutant cell population and the tissue remains normal without tumours. And again, this isn't just seen for mammal, it's seen for p53 and for Notch and for other mutations we think that colonise the oesophagus. So I've talked about some really simple models where we generate mutations in single cells and we let them spread through a wild-type tissue, but that isn't what happens in humans. In humans, we've got this very complicated landscape of multiple clones competing over a long period of time. So can we model this in the mouse, this more complex landscape, and see how it develops? And to do this, we took some wild-type mice and exposed them to a mutagen called nitrosamine that's found in cigarette smoke, and we aged those animals for a year. And then we removed the esophagus and did this gridded sequencing exercise that we've done on the humans. And you can see a couple of graphs here where we've plotted these dN/dS ratios with the human on the right and the mouse on the left. And what we can see is that we're getting the same genes, the same top 3 genes are being selected in mouse and humans. And actually, as we go down the list, there's also positive selection for some other genes. So remarkably, you can recapitulate many aspects of a human mutational landscape in a mouse model over just a year. So knowing this, we went on to combine this process of generating these complex mutational landscapes with the lineage tracing technique. So this time we'll use— we used a mouse called a confetti mouse where we can paint the cells one of four different colours. So we treat with the mutagen, make lots of mutations, and then switch on this label. We label single cells one of four colours and wait for 9 months or 18 months. Then we remove the oesophagus, we cut out individual clones and sequence them, and then we compare the distribution of clone sizes against the mutations they carry. And what was quite surprising is it's not the case that clone size simply depends on the mutations carried. That model does not explain the distribution of clone sizes in the data. However, what does explain it is if we consider this reversion towards normal that I described earlier, and we call this the neighbour constraint model.. So let's think about wild-type cells on the left. They're all equal in fitness and their fate is balanced. Now let's think of a pink clone and a blue clone expanding through the tissue, and let's say the pink clone is the fittest but the blue clone is still fitter than normal wild-type cells. And for the blue clone and the pink clone, their fate's imbalanced. The dice are loaded in favour of producing excess dividing daughters. Those clones will expand and inevitably collide. This is a normal tissue, so the dividing cell layers are finite size, and you'll get clones colliding as they grow. And then what will happen is that the blue clone will lose and be displaced by the pink clone, and then once the entire field is colonised by pink cells, it will— the behaviour of the cells reverts towards balance,, and we get a picture similar to normal with the balanced production of dividing and non-dividing daughters. And to show you this in action, this is a simulation where each colour pixel represents a cell carrying a different mutation, and we'll run this over time, and this very nicely recapitulates the distributions of clone sizes that we saw in these experiments. And what you can see is the clones are getting larger, the number of clones gets smaller, and we end up with a patchwork of mutations that have competed against each other and they've leveled up the competitive fitness across the tissue. So to summarize this part of the talk, mutations spread through normal tissues by transiently biasing normal progenitor sulfate towards producing excess dividing cells. But once they put the expanding clones that result collide with each other, there's spatial competition and eventually neighbour constraints where clones of equal fitness are abutting and the cells within them revert towards homeostatic behaviour. So is all this mutation simply bad, or are there good things about it? And it's inevitably perhaps a mixed picture— there are good and bad aspects. So let's consider a couple of examples. So again, this is work from Emily Abbey in the lab, and she studied how Notch mutants aggressively colonise the normal oesophagus. But once they've done that, what happens if we start inducing tumours? If you remember back to the humans, I said there's a lot more Notch in a normal human oesophagus than there is in cancers that derive from it. So Emily took Notch mutant mice where she was able to delete one or both copies of Notch in the oesophagus and generated tumours and waited for a few months. And what was striking is that the tumours where Notch1 is wild type were substantially larger than those lacking Notch1. So in the absence of Notch, tumours are smaller, and those large areas of human oesophagus might indeed be generating less aggressive tumours than those coming from the non-mutated areas of the oesophagus. But Emily went further and used an antibody that blocks Notch1, so she again generated tumours but this time treated the mice, and you can see these tumours stain red here, with the antibody for a matter of weeks. And quite strikingly, when you treat with the antibody, the tumours are smaller. So antibody treatment against Notch1 cuts tumour growth. So carrying all those mutations in Notch1 is probably a good thing and might protect us against cancer. Let's think about a bad mutation. So I've said that p53 is found in almost all squamous cancers of the oesophagus, though it's quite rare in normals. So this is just an example of a series of experiments done by Kasumi Mirai in the lab, and she generated a mosaic mouse oesophagus. And she wanted to study what was the effect of losing both copies of p53. So here we can see the brown areas have got a wild-type copy of p53, but the other one's missing. The other areas have got a mutant copy of p53, but the second allele is missing. She treated with a mutagen, aged the mice, and generated tumours, and then took tiny samples of these different areas and looked by sequencing for changes in copy number. And what's very striking as you transition from normal oesophagus to oesophageal cancers is that there's marked chromosomal instability. So in the areas where we've got one wild-type copy of p53, there's hardly any chromosomal instability, but once we lose both alleles of p53, in this case one by mutation, one by deletion, you can see that we're getting large chromosomally unstable clones appearing. And in the tumours that develop where we've got no p53, there is really marked copy number change, far more than in tumours generated from wild-type epithelium. So one of the things that p53 mutation is doing is paving the way to chromosomal instability. And I would say also the p53 mutant tumours are larger than those from P53 wild-type tumours. So it acts at multiple levels in transformation. One of the advantages of the sort of approach we have is looking at the whole tissue in the oesophagus in great detail, and this is very useful for studying early events in cancer formation. So this is an example I'm going to show you now of what we call a microtumour. So this is far too small to see by the naked eye, might have 50 or 100 cells in it, and it's stained red here. We image the entire organ and can find dozens or hundreds of these very small tumours after mutagen treatment. These early tumours don't look that bad on a histopathological section, the normal way of diagnosing a tumour, but if we 3D image them, and that's shown here, you can see that the tumour is stained red and now we're going to zoom down below the level of the epithelium into the underlying stroma and you can see this network of capillaries and already the tumour has become surrounded by new blood vessels in the process of angiogenesis and that's a hallmark of much more advanced malignancies. So we're getting tiny microtumours formed early on after mutagenesis. However, what's striking is most of them are lost within 3 months. However, a few persist, they progressively get larger, and by 18 months we have some very large lesions as shown here, which have features of a more advanced transformation, they've got dysplasia, and they're enriched for cancer driver mutations in the mouse. So we're seeming to get a picture where we get lots of early lesions and a few survivors that are progressively transforming in a more malignant direction. Why is it that we're losing all these tumours early on? And We looked in a series of mouse models and were able to show it's not that the immune system is detecting and eradicating these very small tumours, it's not programmed cell death, it's not that the tumours are running out of proliferating cells. So having come up with a load of negatives, we went on to ask what it might be and formulated this hypothesis that maybe it's that very competitive clones in the normal epithelium that are affecting the survival of the tumours. So for a tumour to grow in a very mutated epithelium, it has to be fitter than the surrounding clones, mutant clones, in the normal tissue. But, and if it was the same fitness, then that tumour would just be mutually competing with the tissue. But if the surrounding clones are fitter than the early tumour, it might get lost. So to test this, we did several experiments, but one's shown here. We generate some tumours and then we turn on this very, very competitive mastermind mutant that I've talked about before. We let the mastermind clones expand and see whether tumours are eliminated. And the crunch result is shown here, that as we dial up the proportion of coverage of this mastermind mutant, this is directly proportional to the elimination of tumours. And how this might work is perhaps suggested by this image here. You can see one of these microtumours and it's got a very thin footprint in the dividing cell layer here. It's surrounded by mutant cells and Just in this cartoon of the image here, you can imagine that the mutant cells might extrude out this small footprint and dividing cell layer, and so that lesion would be lost. So in summary, colonising mutants might either promote or inhibit carcinogenesis, and in the more complicated landscape that characterises humans, epithelial mutants might eliminate early lesions. So finally, can we do anything about undesirable mutations? And this is a nice example of work done a couple of years ago by David Fernandes-Antoran, who's now got his own group, and David was studying p53 and how it was affected by very low doses of radiation. So each of these doses is 2 or 3 CT scans worth. And he generated oesophagus-containing tiny p53 mutant clones and then exposed the mice to 5 lots of very low-dose radiation. And what this did was dramatically expand the p53 mutant clones. And what's happening here is that as you give the radiation, the wild-type dividing cells, shown yellow here, differentiate, so they leave the dividing cell layer, and that creates space for the green p53 mutant cells to expand. So they spread through the tissue displacing the wild-type cells due to this wave of differentiation. Why do the wild-type cells differentiate? Well, it's not to do with DNA damage, which is the usual phenomenon people think of with p53, but rather at this very low dose of radiation, it's actually oxidative stress. So David measured the levels of oxidation in mitochondria in wild-type and p53 mutant cells after a shot of low-dose radiation, radiation, and you can see in this plot on the yellow side of the, of the line, wild-type cells experience a strong pulse of mitochondrial oxidation, but the p53 mutant cells don't change at all. And that's because p53 mutant cells are preloaded with high levels of antioxidant genes, so are able to mop up the oxidative stress created by the radiation. So they remain in the dividing cell layer and can take over the space made by the differentiating wild-type cells. So we reasoned, was it possible to do anything about this? And the mice were treated with an antioxidant, sort of over-the-counter antioxidant you can get from Boots or whatever, and then given 5 doses of radiation as before. And the effect is dramatic. If we add the antioxidant the p53 clones are much smaller, and what's happened is that this time it's the mutant cells that differentiate and create space, and then the wild-type cells place them. So we've made wild-type cells fitter than the mutant by giving the antioxidant, and indeed what we've done is we— when we do this, we reduce the level of p53 mutations below that in untreated mice. So potentially, I'm not suggesting people going to Chernobyl or somewhere and get exposed to radiation in the slightest, or take antioxidants, but it might be possible to manipulate tissues to deplete oncogenic mutants like p53 by processes like redox changes. So I think that's an exciting point to end on. It's worth thinking about how evolutionary principles operate and cell competition processes operate in all of our aging tissues, and the oesophagus is a very nice and perhaps hopeful example of how we might be able to reshape these inner evolutionary processes to cut the risk of developing cancers. And finally, I'd like to thank my research group and all of those who fund us. Thank you very much.