2026-04-20-am-clonal-hematopoiesis-blood-cancer-solid-tumors

Date: 2026-04-20 Primary topic: single-cell-spatial-omics Cross-referenced from: single-cell-spatial-omics Word count: ~13,955

Auto-generated captions

This transcript is Vimeo auto-generated English. Expect misheard names and technical terms. Use the ripgrep/search approach for exact quotes.

Welcome everyone, to those in the room and those joining us online, to this major symposium, Clonal Hematopoiesis: Aging and Drivers of Both Blood Cancer and Solid Tumors. My name is Jyoti Nangalia. I'll be chairing the session. I'm a hemato-oncologist based at Cambridge University Hospital, but also the Wellcome Sanger Institute, where I do my research. We have three speakers today. I'm one of them. We have Siddharth Jaiswal, who I'll introduce first. He's a current associate professor of pathology at Stanford? Associate. Associate at Stanford. One of the first people, together with Ben, to describe the concept of clonal hematopoiesis. So we're really fortunate to have amazing speakers with us today. His lab focuses on understanding the biology of aging, of the hematopoietic system in particular, and how these mutations associated with clonal hematopoiesis contribute not only to cancer, but also to age-related phenotypes. So first up, we'll have Dr. Jaiswal, then myself, and after that, Ben Ebert. All right. Over to you, Sid. Thank you, Dr. Nangalia, and it's my great pleasure to be back at AACR presenting our latest work. These are my disclosures. So clonal hematopoiesis is a premalignant expansion of hematopoietic stem cells, typically due to acquired somatic mutations in a handful of driver genes. And we know from first principles, clonal hematopoiesis should be very common with aging. Thanks to great work from the Sanger Institute, we can estimate that there are about 50,000 to 200,000 hematopoietic stem cells in adult humans. And we also know that these HSCs acquire about one protein coding mutation every 10 years. Which doesn't sound like a big number, but if you multiply these two figures together, you can calculate that by the age of 70, there will be about 1.4 million protein coding mutations within the HSC pool of each of us. And if just one of these mutations is sufficient to confer a fitness advantage to a hematopoietic stem cell, then these clonal expansions should be common with aging. And indeed, that is what we see. And so I thought it'd be useful to spend one minute just talking about the terminology of clonal hematopoiesis, because it is a bit confusing. So clonal hematopoiesis, as I mentioned, is any clonal expansion due to a fitness-increasing mutation in a hematopoietic stem cell. And we know that clonal hematopoiesis, by this definition, is ubiquitous in all humans. Clonal hematopoiesis can be further subdivided into CH with known cancer-associated driver mutations. So that's how we first identified this state, is by looking in genes that were recurrently mutated in heme malignancies, and then going back to healthy individuals to see if they had those mutations. However, there is also clonal hematopoiesis with unknown driver mutations, and we don't really understand why these mutations expand. But we know that this is almost as common as clonal hematopoiesis with known drivers. And then within the known driver clonal hematopoiesis space, there's what we have defined as CHIP, clonal hematopoiesis of indeterminate potential. That's based on a variant allele fraction of greater than 2%. And then clones that are smaller than this are sometimes called micro CH or microCHIP. These are very common, because only a subfraction of these small clones go on to develop into large clones. And then within the CHIP pool, there's CHIP due to single nucleotide variants or indels, so think mutations in TET2 and DNMT3A, and then there's also mosaic chromosomal alterations. So surprisingly, there haven't been that many studies that have reported the prevalence of these different kinds of clonal hematopoiesis all in a single population, mainly because you need whole genome sequencing data to identify clonal hematopoiesis with unknown drivers. It's typically identified because there's an accumulation of passenger mutations, which doesn't occur, at least is not detectable from whole blood sequencing, unless there is an expansion event. But we've done this in a few cohorts now, and this is the prevalence of the different kinds of clonal hematopoiesis as a function of age. So you can see that in those under the age of 60, it's relatively uncommon, about 10% in total. And this rises to about 80% of all individuals who are age 90 or older have a large detectable clone. And you can see that clonal hematopoiesis with indels or single nucleotide variants is almost as common as clonal hematopoiesis with unknown drivers. And one of the bigger surprises in the initial surveys of clonal hematopoiesis was the spectrum of genes that we typically see mutated. And if you think about typical oncogenes, things like RAS signaling pathway or tumor suppressors, like TP53 loss of function mutations. We do see these in clonal hematopoiesis, but they're relatively uncommon. Only about 10% of the total. Instead, the other 90% seems to be genes that are involved in epigenetic regulation, like DNMT3A, TET2, ASXL1, or spliceosome factor mutations.And so why are these particular genes that don't seem to have an obvious link to oncogenes or tumor suppressors, why are they positively selected? And we hypothesized that we could make headway into this question by studying it directly in humans. And specifically, if we could identify a way to calculate the clonal expansion rate in a large number of people, we could do large-scale association studies to try to figure out why these particular genes are mutated. And so we developed a method that we call passenger approximated clonal expansion rate. So this is a method that uses the total number of passenger mutations in a clone to infer how old that clone is. So basically, these passenger mutations are accumulated linearly with respect to time, and so it's like a timestamp for when that clone was born. And once we know how old that clone is, we can infer how quickly it's been growing using a simple linear regression model, as shown here. And what we hypothesized was that the residual, so if you plot the passenger mutations versus the input variables for each individual, which is shown by a dot here, the residual, which is the difference on the y-axis from the line of best fit to their number of passenger mutations, that should be a measure of the fitness of that clone in that individual. And so we went on to validate this method in several ways. And one of the things that we found was that the specific driver mutation itself is a pretty large determinant of clonal fitness. So these are our PACER fitness estimates for the most common drivers seen in CHIP. These are all shown relative to DNMT3A mutations. And what you'll note is that most of the drivers have a higher fitness than DNMT3A. So DNMT3A, somewhat paradoxically, despite being the most commonly mutated gene, also has one of the lowest fitness scores. The highest fitness scores were seen in the spicy som genes, U2AF1, SRSF2, and SF3B1. JAK2 was also quite high. And then in the middle, you see a lot of genes that are common, TET2, PPM1D, ASXL1, et cetera. So we know that driver mutation is a major determinant of growth rate, but we wanted to know, using our method, what environmental and genetic factors may also drive positive selection of these clones. So we turned first to environmental factors. And this is data taken from the UK Biobank, as well as some other cohorts. So we first measured biomarkers or demographic factors that were available in this data set and asked, did any of these associate with clonal expansion rate? And you can see that of these commonly assessed demographic variables, including sex, smoking, diabetes, BMI, inflammation, and air pollution, only smoking was associated positively with clonal expansion rate. This was a little surprising given the amount of animal model data supporting things like obesity or inflammation as being drivers of clonal expansion, at least in mice. However, we did not see these associations in humans. The reasons for that difference are not entirely clear. We've also looked at this in other ways. This is a collaborative project with Alex Bick's group at Vanderbilt. And what his group did was to create polygenic risk scores for different protein levels that you might have in your body, and then asked if you have a higher genetic risk of having certain proteins circulating in your body, does that associate with clonal expansion rate? And what we found generally was that genetically predicted cytokine levels were not very predictive for clonal expansion rate. The one exception being there was a modest association between predicted IL6 levels and the rate of growth of DNMT3A clones. And then there were a handful of other TNF-like factors that were also modestly associated with DNMT3A clonal expansion. So overall, what our findings have seemed to indicate thus far is that of the measured environmental factors that we can assess, it's really only smoking that seems to have a strong correlation to growth rate. And so we then hypothesized that perhaps we could do a genome-wide association study or a GWAS to look for germline variants of clonal expansion rate to reveal genes or pathways that may be associated with fitness advantage. And this is the result of our first survey. So this was done in the TOPMed cohort. This finding has been replicated now in additional cohorts. And what we found was that there was a single locus that reached genome-wide significance in the survey. The lead polymorphism in that locus is called RS2887399. It's a single nucleotide polymorphism in the promoter of a gene called TCL1A. The reference allele is a guanine. The alternative allele is a thymine. And what we observed is that the alt allele was associated with slower clonal expansion in several non-DNMT3A driver genes, including TET2, ASXL1, and some splicing factors.TCL1A is an interesting gene because it has been studied in the context of heme malignancies before. It was originally cloned by Carlo Croce's lab back in the 1990s because it was found in a translocation in a rare malignancy called T-prolymphocytic leukemia. That's a leukemia of mature T cells. And this translocation causes the TCR alpha locus to come into proximity with TCL1A to drive aberrant expression of TCL1A in the mature T cells, and it's thought to be the causal driver of that malignancy. However, there was no known link between this gene and myeloid malignancies or HSC biology. TCL1A is a bit of an enigmatic protein in the sense that it's not really very clear what it's doing. It's a tiny protein. It's 14 kilodaltons. It forms an obligate homodimer with itself, and early studies show that it was able to bind to the PH domain of AKT and potentially activate AKT via an unknown mechanism. And so what we went on to show in our published work, through a variety of experiments, was that we think TCL1A is a major driver of clonal expansion in hematopoietic stem cells. What we showed was that the TCL1A locus is normally closed or lowly expressed in hematopoietic stem cells. And in fact, the only cell types in adult humans where TCL1A is expressed are B cells, plasmacytoid dendritic cells, and progenitors of sperm cells. And so in this sense, TCL1A might actually be a very good therapeutic target because you might predict that you'd be able to block its function without really affecting normal tissues very much. However, if you now acquire a mutation in one of these driver mutations that we showed was associated with the polymorphism, we believe that this de-represses the TCL1A locus. So now the gene goes from off to on. This promotes TCL1A protein expression and we think is a driver of clonal advantage for these mutations. And the reason we were able to find this serendipitously was that there happens to be a subset of the population that has a thymine at this TCL1A promoter at this SNP. And we think that having this thymine variant prevents the chromatin from opening and abrogates the clonal advantage, and therefore reduces prevalence of CHIP in these individuals. So what is the full spectrum of driver mutations in CHIP that might be associated with TCL1A? And so what we can do now that we have this polymorphism that we know is strongly linked to TCL1A expression, is to systematically ask what driver mutations are enriched or de-enriched if you have the T allele at this locus. And if the prevalence of the driver mutation is below the dashed line, we would predict that these driver mutations are somehow dependent on TCL1A for their clonal expansion. And so we did this in the UK Biobank dataset, and we replicated our previous findings, which were that ASXL1, TET2, SF3B1, and SRSF2 mutations were strongly depleted in individuals who carried this T allele. But we found a number of other surprising things, such as NRAS mutations, IDH2 mutations, and JAK2 mutations may also be depleted in carriers of the T allele. And furthermore, we found that clonal hematopoiesis with unknown drivers, so this large class of clonal expansions where we don't really know why it's happening, it turns out the strongest genetic hit, so if you do a GWAS for clonal hematopoiesis with unknown drivers, the strongest hit is in this TCL1A polymorphism, suggesting that many of these mysterious clonal expansions may also depend on TCL1A expression. So we think that TCL1A is a critical factor for the development of clonal hematopoiesis, and it was found using GWAS. And many of you in this room may note there have been many polymorphisms, many genes that have been linked to clonal hematopoiesis through GWAS. But what I would say is distinct about TCL1A is that it is the only gene of all the GWAS hits that goes from off to on in hematopoietic stem cells when a somatic mutation in one of these driver genes is acquired. And this suggests that this is not just a modulator of clonal expansion, but it may be a major causal driver, so the reason that these mutations are actually selected. And if this model is true, there are two predictions. So one is that ectopic expression of TCL1A should be sufficient to cause clonal expansion phenotypes in vivo, in a model system, for example, by altering key HSC biology. And a second prediction is that TCL1A expression should promote the risk of at least some CHIP-related phenotypes, such as hematologic malignancy. And so we decided we would like to study this in an animal system. And as many of you in this room know, there are many models for these common CHIP driver mutations, including loss of function, TET2 alleles, TET2 catalytic domain mutants, truncating mutations at the C-terminus of ASXL1, and then point mutations in SF3B1 and SRSF2. So these have all been built in mice.And they variably show a clonal expansion phenotype at the HSC level. The majority of these mice show things like anemia, but only a couple of these models actually develop malignancy. And when they do develop malignancy, it's typically at a very late stage. And we looked in all of these models, and in none of these models could we see that TCL1A became expressed in the HSCs. So unlike what we see in humans, if you introduce these mutations into human HSCs, TCL1A goes from off to on. It doesn't seem to happen in the mouse. One reason for that is that if you look at the promoter region of TCL1A, this is where the polymorphisms that I described lie. You can see that there is generally poor conservation between humans and other species at this locus. And so we think that maybe because the promoter is not really conserved at this region, mice do not have the same effect. So even if you put the mutation in the mice, it doesn't lead to the de-repression at this locus. And so to get around this, what we decided to do was to build a transgenic mouse model where we could enforce expression of TCL1A in a tissue-specific manner. And so we introduced the TCL1A human open reading frame into the H-11 locus. It's a safe harbor locus in mice, and the promoter is separated from the open reading frame with box stop lux cassettes. And so what we can do is cross these mice to various Cre-expressing strains to get tissue-specific expression of TCL1A in the mice. And so we have done this using Bav1 Cre mice as well as Tau1 Cre mice. And what we find is that these mice, for most of their life, look largely normal. So we followed their CBCs up to a year, and you can see that there's really no difference in the white blood cell count, in the red blood cell parameters, in the platelet counts, or in the differential up to nine months. We did note that by midlife, there was a marked HSPC expansion in the bone marrow of these mice, and this was particularly evident in the long-term hematopoietic stem cell fraction of the bone marrow and less evident in the myeloid progenitors. So we think that TCL1A, as perhaps is expected, does lead to some sort of fitness effect on the hematopoietic stem cells and causes some sort of expansion in this model. We've also done competitive transplants, where TCL1A was forced to be expressed in some HSCs and then competed against wild-type HSCs, and you can see that over time, the TCL1A-expressing HSCs take over. So there is this competitive advantage as well. In terms of the mechanism of why TCL1A may cause clonal expansion of these HSCs, that's still being worked out. But we suspect that there is increased proliferation, so increased cell cycling of these HSCs and also resistance to apoptosis. So, we thought that these mice would probably not have a very exciting phenotype beyond what I just showed. But much to our surprise, when we kept the mice alive for up to a year, we found that there was 100% mortality in the TCL1A transgenic mice. So you can see that by one year in both Cre crosses, all of the mice have died. And we observed there's a pronounced splenomegaly and a pronounced hepatomegaly, as well as an expansion of HSPCs in the moribund mice. The phenotypes at death range from pancytopenic, that's the majority of the mice, to marked leukocytosis. And we further characterize what are these mice dying of, and it looks like the majority of the mice are developing a myeloid leukemia phenotype, characterized by sheets of abnormal myeloid cells in the bone marrow that express an abnormal phenotype. They're Kitlow CD11b positive, so they're myeloid, as well as CD34 positive and negative for lymphoid markers. So is this relevant potentially in human myeloid malignancies as well? And so we've asked if individuals who have this T allele, the one that prevents TCL1A expression, do they have any alteration in their risk of developing myeloid malignancy? And we can see that indeed that is the case, that carriers of the T allele are less likely to develop any myeloid malignancy, and an especially strong protective effect for CMML, chronic myelomonocytic leukemia. And finally, we've asked, is TCL1A still expressed in the hematopoietic stem cells from patients with myeloid malignancies? And we've looked at this in a couple of different ways. First, using publicly available single-cell RNA-seq data from healthy or malignant hematopoiesis. We can see that healthy HSCs, as expected, there's essentially no TCL1A expression. However, if we look in the HSCs from patients with CMML or MDS, so each dot represents a single patient, we now see TCL1A is expressed in these HSCs, confirming that it may actually play a role in the malignancies as well as clonal hematopoiesis.And we've also looked at this at the protein level using flow cytometry. So this is gated on a HSC fraction. You can see that in healthy donors, there's, as expected, essentially no TACE1A expressed, whereas we do see robust expression in these high-risk MDS cases. So to summarize, clonal hematopoiesis becomes ubiquitous with aging, but the factors leading to positive selection of these drivers is incompletely understood. Of the environmental factors we've studied so far, smoking seems to be the strongest promoter of clonal expansion. And genetically, we think that expression of TACE1A seems to be a major driver of clonal expansion of clones of both known and unknown drivers. And we hypothesize that TACE1A may be a therapeutic target for myeloid malignancies arising from antecedent clonal hematopoiesis. And I'd like to acknowledge all the people that did the work. The majority of the studies that I've showed today were done by a very talented postdoc, Isaac Tingestdal, and an MD-PhD student, JK Gopalkumar. Thank you for your attention. Beautiful talks. Thank you. The talk's now open for questions. Can I ask people to come to the microphones because this is being live-streamed, and for those joining us online to use the QR code and ask questions via the online chat so we can ask the speaker. Hi, Tim Ley from Wash U. Sid, beautiful work. I must have been shocked to see those tumors arise in those mice late. The long latency suggests that there's either cooperating mutations. I wonder if you know what they are or whether there's some kind of transcriptional earthquake associated with escape. Yeah, Tim, great questions. So we are trying to sequence the bone marrow from these mice now to determine if there are any recurrent cooperating mutations that to be determined. The thing that is a little bit surprising is the very narrow window in which these mice all seem to die. So they seem to be pretty healthy up to about nine months, and then by 12 to 13 months, they're basically 100% dead. So I don't quite understand at this point why that window is very short, and it would be a little bit surprising if they all acquired the same cooperating mutations within that time frame. So I'm starting to suspect it's some sort of epigenetic state or something else that goes wrong around that time point. Are they transplantable? Yes, they are transplantable. Yes. Very interesting presentation. I have a couple questions, actually. Now that you can measure fitness, if we measure the fitness of the non-clonally involved hematopoietic stem cells in a patient who has clonal hematopoiesis, are they as fit as normal or less fit than normal? Yes, a good question. So the method that I talked about, we can only assess the fitness of the expanding clone. So I think you're asking, can we assess the fitness of the other HSCs that are not part of that clone? I'm by no means an expert on this, but I was always wondering, is the clonal hematopoiesis the result of the background failing from aging? Yeah. So now you can see the clones because they are the only ones left contributing. Yeah. Or is it really that they are more effective, that's why they are succeeding more and becoming more prominent? The other question was, when we say they are more fit, is it because they are more constantly dividing or more actively dividing, meaning is it the amplitude or the time? Can we distinguish these two? Because we say it's fit, but is it fit as a result of the constant firing, or is it fit because it does a lot when it fires? Yeah. So for your first question, we don't know for sure what the reasons are. Sorry, could you repeat the first question? We've been always told in oncology, this is an activated mutation. Yeah. Okay? I'm now trying to ask, is it more potent, the mutant, or is it more constantly on, and as a result, you end up having more function? So here, the same thing, the fitness, is it because of more proliferation per unit of time, or is it proliferating more than normal, so it ends up being represented more than normal? Can we distinguish these two scenarios? So we could try to do that using the mouse models. So far, we have observed an increased proliferation rate using things like BrdU over short time points. But I guess we don't know, is that also true in humans or is it perhaps some sort of inciting event in the environment that promotes the proliferation, and when that signal goes away, perhaps the cells aren't proliferating as much? We don't know that for sure. One of the things that was surprising is that in the mouse HSC literature, it is believed that causing HSCs to cycle reduces their overall stem cell potential and fitness. So if you force HSCs into cycle, they kind of exhaust over time. But at least when you acquire this TACE1A gene expression, that doesn't happen. One quick last question, if I may. Sure. Hi. Lovely talk. Just a quick question about the SNP that you found. How common is it in, say, the UK Biobank, and does it modulate the interaction between CHIP and other diseases like lung cancer and other malignancies? Yeah, great question. So the allele fraction for the T allele is about 30%. So that means that about 25% of people are heterozygotes and 5% are homozygotes for the T allele. We know that this SNP doesn't cause or associate with heme malignancy risk. So far, we haven't seen a lot of associations with any of the other sort of chip associations, like lung cancer, for example. I'm afraid we're going to have to move on in the interests of time, but there might be some time at the end for a general panel discussion. Thank you very much. I'm up next. So over the next 20, 25 minutes, I'm going to give you all a summary of the sort of work that we do in our lab, and it's really about looking back in time to understand the history of somatic evolution in blood. These are my disclosures. So this audience will now need no introduction really to the concept that every cell in our body, right from where the zygote dividing, is accumulating somatic mutations such that within our tissue, over here on the left-hand side, every cell is a little bit unique compared to its neighbor. And depending on, as beautifully illustrated earlier, the germline factors or the environmental factors, some cells will be winners, some cells will be losers, and there's beautiful substrate for Darwinian evolution within our tissues. Now in our lab, we've developed approaches to study this and look back in time using phylogenetic approaches to really try and time and understand how this somatic evolution is occurring within our tissues right from the start of life. And just to give you an example, the way we do this is we can take any tissue, but we use blood or bone marrow. We grow them into single-cell-derived colonies, and by doing so, what you're really doing is amplifying the DNA of a single cell without much in the way of artifact creation, so that you have enough DNA to undergo whole genome sequencing that is reflective of that single cell within that tissue. And then when you do this several times, hundreds of times within the same tissue, within the same individual, you can then use the pattern of somatic mutations across the whole genome with an individual cell to reconstruct a phylogenetic tree. And this phylogenetic tree essentially allows you to go back in time right to the fertilized egg to track somatic evolution within any tissue. And it's been a really fun time over the last five years because we've been using these approaches to understand the ancestral origins of clones, mutation timing, speed of clonal outgrowth, but also, much to our surprise, we can infer population size, stem cell dynamics, just from the genomics approaches. So at the top is some of our studies in case you're interested in looking at them. But we collaborate broadly, as shown in gray. And so far, we have about 13,000 whole genomes, perhaps one of the largest whole genome dataset in blood that's publicly available, including all the tools that we've developed to analyze these phylogenies. And just to give you a bit of a flavor of the kind of things that you could learn and some of the insights that we've had, I'm going to first orientate you as to how to read a phylogenetic tree. So these little red dots at the bottom are all the cells that we sequence with whole genome sequencing. And the top of the tree is the start of life. And what you do is these lines, the vertical axis of the somatic mutation says you walk up the tree and you go sideways into these shared branches, right to the root of the tree. Those are all the mutations acquired by that history, that ancestral lineage, right down to the sample that you have in your dish today. And you can see already that at the top of the tree, there are all these little tiny branches. And what that's telling you is that early in life, there is rapid expansion of cells, rapid genetic divergence of cells, including the stem cell pool. And then beyond that point, the stem cells don't really talk to one another. They're independently making their family of progeny that we then sample today. On the right-hand side are the cancer cells in this individual that has a chronic myeloid neoplasm, such as essential thrombocythemia. It's often JAK2 mutation driven, nothing else. And this occurred on this shared branch here before 1.3 years of age. And as you can see, the pattern of expansion of that single stem cell into a clone is trackable. And you can see that this pattern of expansion is slower than early in life because the clone has grown more slowly than the embryos growing early in life. So you can really recapitulate pace of clonal expansions within cells. And then you can sort of estimate that fitness, much like Pacer is doing. And for me, fitness is really an interest rate on your bank account, constant interest rate yields compound returns. And when you model the interest rate really that a JAK2 is conferring on a stem cell, we find that it's quite variable between individual patients. But what's remarkable is that growth rate perfectly correlates with when that individual will present with disease. So fast-growing clones, 70% per year, take about 20 years to present to clinic. Slow-growing clones take about 50 years, which has therapeutic implications. You may not need to eradicate clones, you may just need to slow them down. And it also has implications for early detection if your window of detection is of the order of decades. In the same clinic as MPNs, these myeloid disorders, I also treat chronic myeloid leukemia. Now, this holds a special place in our hearts because it was really the first ever cancer to be associated with a acquired genetic change, the Philadelphia chromosome.Here, discovered in the 1960s and 70s and here we see something very different. Again, the wild type hematopoiesis looks much like the previous picture, but this is the clone driven by the fusion protein BCR-ABL, and this actually looks much like the start of life. This clone is growing so fast, 30,000% growth per year, at least at the start, because you could not maintain that. And the latency to diagnosis is about four years. So if you detect a BCR-ABL at the DNA level in your blood, don't wait. It's going to grow quickly, you treat it. And interesting that this is one of the most treatable cancers and it's the most fast-growing cancer. So this is quite exciting. You can get growth rates, you can get timing, but you can also get clonal dynamics. So, Dr. Joyswell's nicely recapitulated how all of us have clonal hematopoiesis as we're getting older, and often in multiple clones, not just a single driver mutation. We'll have driverless clones as well. And we can simulate how this is happening with these sort of younger individuals having perhaps some driver mutations, but they haven't had a chance to grow out. And then as you simulate and get to 80 to 100 years old, you can beautifully recapitulate what's happening to hematopoiesis and this collapse in polyclonal due to the clonal expansions, with a simulator we've created called RSimpPop. And in that, what we find is that with a very simple model of 100,000 stem cells dividing around once per year and click go and let that play out for about 80 years. If you have around 200 coding mutations that are entering your stem cell pool every year, and if the interest that those mutations confer is around 15% per cell and you let that run, you can nicely recapitulate clonal hematopoiesis over lifespan. Now, we thought that we would need to do other things. We thought that we would need to kill off some stem cells, induce population bottlenecks, but a very simple model was sufficient. So then we started to think, well, what happens when you play around with these parameters? What happens if you have bone marrow failure and fewer stem cells? What happens if your stem cell rate is higher, you're dividing more frequently? And we're playing around with some of those experiments at the moment. But in the first instance, I spoke to Peggy Goodell at Baylor College of Medicine, who works on stem cell biology in the mouse. And we thought, well, what happens in an old mouse? It's used for studies of clonal hematopoiesis all the time. It's used for aging. It's used for stem cell biology. Does an old mouse that's two to three years old look like a 100-year-old, 80-year-old human, or does it look like a two to three-year-old human? Does this process scale with lifespan or does it scale with time? So we did this study in the mouse with a talented PhD student called Chirag, and we found a completely different pattern in the mouse to the humans. So the old mouse, 30 months, quite hard to get to 30 months as a mouse, really old, no clonal expansions, no collapse in clonality. And when you use very deep, sensitive, targeted sequencing, you can find all the clonal hematopoiesis clones, but they're tiny. Super, super, super tiny. And then we tried to understand why. Well, we found that the fitness, that interest rate from a DNMT3A mutation or a BCOR mutation is actually similar between a human stem cell and a mouse stem cell. Now, this wasn't done functionally. This was done using the phylogenetic trees and population dynamics. We found that the HSC pool size wasn't too different, about 70,000 stem cells in a mouse. I have no idea why a little mouse needs 70,000 stem cells, but perhaps it's legacy. Maybe it's phylogenetic legacy, and the mutation rate's only three times higher. It doesn't scale with lifespan. Mutation rates to really recapitulate human, the human system would need to be maybe 30 times higher because lifespan of a mouse is 30 times shorter, but it wasn't. And so overall, we concluded that the HSC division rate in the mouse, which was around nine to 10 divisions per year, was just too low for meaningful clonal expansions, over a two to three-year period. So what this says is that somatic evolution, as we all know, is not just about the environmental factors, not just about the cell intrinsic driver mutations, but it's also about your baseline tissue dynamics. How fast is the clock ticking? And in the mouse, I'd say a two to three-year-old mouse is more like a 10 to 20-year-old human, in terms of its pace of somatic evolution. So this was quite interesting, and we have other studies now going on where we have fewer stem cells, or perhaps increased cycling of stem cells to see how does that influence clonal hematopoiesis. But I want to, for the next 10 minutes of my talk, move on to some really exciting unpublished data that we've been working on for many years. It's been a bit of a labor of love, and we sort of asked ourselves, okay, we do all these phylogenies, we have all these genomes now. We're sort of really able to track the history of every somatic mutation in our tissue. But what about other forms of heritable genetic information? So we have 28 million CpG sites in our genome, and each site has a cytosine on the forward strand and a cytosine on the reverse strand. These little dyads, as shown here. And methylation states are associated with density of CpGs, with CpG islands being unmethylated, and then these introgenic regions that are perhaps more sparse in CpGs being more methylated. And we know that in blood, we have DNMT3 mutations, we have Tet2 mutations. They're affecting the methylome. And we know that there are things like epigenetic clocks that measure our age. So it seemed like a really good candidate. And so we had this dream, can we build methylation trees?And it ended up being longer than it took longer than we would have liked, but we're nearly getting there. So we first had to collaborate with NEB to get methylomes from these individual colonies. Now we had two nanograms of DNA, sometimes 20 nanograms of DNA from a clonic crypt or a hematopoietic colony, and we were trying to get a whole genome PCR-free. But we're also now trying to get a whole methylome. So we first tried bisulfite sequencing, and that just vaporized our DNA. That didn't work. So then we worked with NEB to lower the input of DNA required by adjusting the enzyme concentrations. And then we finally managed to get to a situation where we now have an automated pipeline at the Sanger, where we can put in a tiny amount of DNA, get a whole genome and a whole methylome. And the beauty of this is that for every CpG in the whole genome, full genome coverage, we have a binary readout. So for every cell, you have two alleles. So each CpG will either be fully unmethylated on both alleles or fully methylated. And very occasionally at imprinted sites, you might have half methylation. So this is the picture of one colony. All the sites are sort of unmethylated or methylated, and this little hump in the middle are imprinting control regions. So we have now a binary readout. No longer do we have to go 20% methylation goes to 30% methylation. We have zero, we have one, we have two. And then what was really tough is to do the informatics. We have 28 million CpG sites per colony. We have thousands of colonies. Every time you try and interrogate it, it runs out of memory. So we needed to build an informatics pipeline for this, and which we now have. And so for every cell, effectively, we now have 50 million CpG sites, duplicate rates less than 10% for those who look at genomic technologies, that's really good, and global methylation rates that are completely biologically plausible. So this is our study now. We have about seven healthy donors from 17 weeks post-conception to 77 years and six hematological cancer donors that include some of the patients that I have just shown you from the somatic mutation analysis. And we sequenced about 1,200 colonies with both whole genomes and methylomes. The idea is that we use the somatic mutation tree as the gold truth of the ancestry, and then we assign CpG methylation changes to the phylogeny. So for every CpG in the genome, we can say has it changed? Which cell is it changing in? When in life did it change? And just to give you a sense, most of the CpGs were in open seas, and some of them were in islands, shelves and shores, as you'd expect. And we treated each CpG as a dyad, combining the read counts from this with the read counts for this into one dyad just to increase our power. And we didn't find there was much discordance within a dyad, actually. Dyads changed together, which was good. And as expected, we had fully methylation predominantly in the open seas, mostly unmethylation in the CpG islands. And when you look in the women, these two donors, blue and purple are women, we're finding a lot of half methylation in the islands and the shores because this is X inactivation in the females. What's really interesting is X inactivation in the females is not X chromosome wide. It is predominantly at the islands, the promoter regions, and much less in the open seas, which was quite interesting for us to note because I just assumed as a hematologist, you see these bowel bodies on your blood films and you sort of think the whole chromosome's just compacted down. So we did some further checks now. This is a phylogenetic tree now on its side, and those in my lab will immediately be looking at that and going, "What's the age of that patient?" Because we're quite good at aging someone based on their phylogenetic tree now. So this individual's probably in their 50s. And each row here is a consecutive CpG, and red is methylated, blue is unmethylated, white is methylated but only on one allele. And every row here, every dot here is the CpG for all the different colonies. So this is a block of CpGs, and we can see that it's unmethylated at the promoter of CSF3R, which we'd expect. This is a line element. It's fully methylated across all of our colonies. And this is an imprinting control region which is nicely, heterozygosely methylated. So these sites haven't changed. And now we have the challenging task of assigning methylation states to the trees. So what we have here now is we take the reads that are methylated for individual tips, we take the full sequencing depth, and then what we're essentially asking is we convert this into a methylation state, methylated, unmethylated, heterozygous methylation, and we account for in vitro changes. In vitro, there might be some demethylation, there might be some methylation, and we correct for that. And then we're basically saying, what is the probable sequence of methylation states that's occurred from the tips all the way to the root so we can see what the most likely changes are using a hidden Markov approach. And what we found was really quite staggering. So I'm just going to talk you through this one example. This is a tree from an MPN patient, and the y-axis is the number of mutations, and they have about 800 mutations. They're maybe about, I don't know, 40, 50 years old. And there are two clonal expansions here with drivers, which I'll talk about later. And I've expanded the top of the tree, the first 50 mutations that you acquire before you're born. So we can see the top of the tree really clearly, otherwise it's really squidged. And you can see that this block of CpGs, these 33 consecutive CpGs in a promoter of PLD6, the fact that it's PLD6 doesn't matter. Well, it might matter, but there are loads of sites like this. This has changed from unmethylated to methylated one, two, three, four, and maybe a few more times, and each of those has happened very early in life.Around 10 mutations, which is basically around gastrulation. So during gastrulation and very early in life, each cell is going, am I going to be heterozygously methylated? Am I going to be unmethylated? Am I going to be fully methylated? And it's choosing an epiallelic state, which we'd never appreciated before. And there are thousands of these regions that are changing early in life. So then we went back and said we can't consider each CpG change as an individual change because they're changing together in segments. So we went back and segmented the genome using various approaches, and only last week we decided exactly which approach we're going to be using for the study. And as expected, most segments that change together are in promoters, around two million segments. And in transcribed regions, about 75% of your promoters have these large blocks of CpG segments. And in the quiescent regions, they're mostly single CpGs, two CpGs that change together. And then we found so many more. So because I said that these are changing early in life, we went to our early samples. We went to our cord blood and our fetal samples to go, are they really changing early in life? So this is a cord blood phylogeny, about 50 mutations by the time you're born. And here is a site, again, around 30, 40 CpGs, unmethylated in all of your cells. It's changed and gained methylation on one allele, but not the other, and it's heterozygously methylated here and here, around 10 to 15 mutations in life. This has happened twice in the tree. This site has happened so many times, and sometimes it's allele A that's methylating, sometimes it's allele B. And this site I find particularly fascinating because every time it's acquired this segmental allelic state, it's the same allele. Now, what I don't know is this the paternal allele or the maternal allele because they're treated differently early in life, particularly with methylation, but this is all work in progress. So overall, we have about 4,000 or 5,000 of these segments of CpGs that are at least five CpGs long changing early in life, and thereafter, the rate is much slower. This is very different to somatic mutations that are quite linear over life. And in total, we have half a million CpGs changing early in life, and then slower. Now, you may notice this here, that's so much higher. This is in an in utero DNMT3A R802H mutation. Its acquisition is associated with 200,000 CpGs that have demethylated. So then we asked what's the rate of somatic methylation changes in life compared to SNVs, and late in life we have about 100 changes per SNV, but early in life we have about 10,000 changes. So this is just a regression model. So much, much higher than we ever appreciated. And when you look at the really simple question, do clonal hematopoiesis driver mutations impact your methylome? The answer is yes. So every dot here is colored by the donor, and these are the number of epimutations per branch of the tree. And you can see some are above the line here, and they're all associated with driver mutations, 1q+, DNMT3A Y66F. And for losses, it's the same, around 5,000 large segmental losses early in life, and the ones that are above the line are often our familiar driver mutation players. And this may be how some of these mutations are conveying their phenotypic effects on stem cells. We need to see what these sites are and perhaps what genes they may be associated with. But what's really more interesting to me is, are these epialleles themselves drivers? Do somatic mutations prefer certain epiallelic backgrounds? And can these epialleles act as drivers themselves? And the answer to both of them is yes, we think. So going back to that example I showed you earlier, this is actually a JAK2 mutation-driven clonal expansion. This is a DNMT3A mutation. When you do a differential methylation analysis, DNMT3A mutations versus wild type, what are the methylation changes? This is the first hit that comes up. But it's not due to DNMT3A. It predates DNMT3A. It's a background that DNMT3A prefers. So now we're trying to see whether there are certain embryonic epialleles that are preferred by certain driver mutations. But this is even more fascinating. Can these epialleles act as the first hit as a driver mutation themselves? And the answer is yes. So this project has involved many PhD students and this Laurie Kreeger, who's a joint PhD student between Peter Campbell and myself, with Lucy Yates. They had this idea to take individuals with breast cancer that have BRCA-like breast cancer but do not have a germline BRCA mutation. And some of you in the audience will know that there's this concept that some people with BRCA-like breast cancers have methylation of the promoter of BRCA, but weirdly, it's detectable in their blood. So why is it in the blood? People have thought, well, is it some sort of environmental hit? This is the phylogenetic tree of two individuals who developed BRCA-like breast cancer, and this is a segment within the promoter of BRCA. It's methylated on one allele, and it's normally unmethylated. So they have epigenetic silencing of one allele of BRCA in their blood, and here, too. But look at when this occurred. I've not expanded the top of the tree here, but it's right at the top of the tree in this tiny little branch you can barely see. It's right at the top of the tree in this tiny branch you can't see. These two individuals got epigenetic silencing of one promoter of their BRCA allele pre-gastrulation.They have epigenetic mosaicism in their blood. They have epigenetic mosaicism in their breast because it's pre-gastrulation, and the subsequent breast cancer has emerged from the lineage that has this silenced BRCA allele in the breast. So when we map, and we're not showing it here, when we map the breast cancer onto this, it emerges from this lineage. So this is an evidence of an epi driver that occurs early in life. So we're very excited by this field. This is what our lab's now doing across cancers, and we're hoping to put this together for submission soon. So key messages, phylogenetics are a beautifully powerful approach to understand cancer evolution. I've presented new data on methylation mutations being stably inherited for decades after occurrence. Most changes are heterozygous, unmethylated to heterozygous methylation or fully methylated to heterozygous, and they occur in segments, and they are stable, and they're vast changes early in life. And I believe that somatic methylation changes are both due to driver mutations, but most excitingly, can be drivers themselves. So just before I end, a little paper plug, because I think this paper's been released to time with this session, but I haven't had a chance to talk about it, but it's on the evolution of MPNs and how serial genomic monitoring in the clinic can be used to predict disease progression years in advance of clinical progression. And interestingly, we find that both azacitidine and hydroxycarbamide, which are two common agents used in MDS and MPN, are mutagenic, and we describe their signatures to normal tissues, and then we do some work on why skin cancers are emerging in the context of some of these therapies. So I encourage you to check it out. Most importantly, those shown in bold, including Xin Wang, who's sitting in the audience, they're the brains behind the work that I presented today. They're brilliant at the analysis that they do and the ideas that they have, and I'm so proud that they're in our team. Our funders, our collaborators, and the patients for donating their samples. Thank you very much. Should we take some questions? No? Me. Hi. Amazing talk as always. So my question is allelic-specific methylation and clonal selection. You show the several independent crates have the same methylation. So I'm not totally understood your explanation. But two explanation, one is the sum result of the positive selection of the particular allele. The other explanation might be the sum epigenetic memories through the... Because after paternalization. So once the old methylation erased but not completely erased. So this seeds of some methylation could prefer the particular allele rather than the original. Which is the more feasible plausible explanation- Yes. According to your data? So in terms of these sites that are either gaining methylation or losing methylation early in life, you're wondering whether it might be selection or something else. I'm not sure if it's selection because it's at a time... Well, it could be. It's at a time when the growth is so rapid that any minor selective advantage will sort of hitchhike with clonal expansion. So they could have a minor selective advantage. We don't know. We have to model that, and it's quite hard to model because the background growth rate is so high that literally even a 5% selective advantage would come up. It could also just be stochastic loss. Maybe when there's reprogramming going on because it's at a time when you are lineage priming your germ layers, that there are pockets that get missed. I don't know the mechanism at the moment, and I also don't know how the maternal and paternal chromosome alleles might be differentially treated. I think it's a really key question that what's driving this. For me, it's like autosomal imprinting. We're seeing this beyond the X chromosome. Imprinting? Well, it's like that. We always think of heterozygous methylation as only associated with the X chromosome, but that's essentially what we're seeing here across 5,000 sites in the genome. And I don't know what's causing it. Okay. Thank you. Okay. Nuria Lopez-Bigas from Barcelona. I think it's very powerful to have the mutations and also the methylation. If you would get non-regulated CpGs, in principle, you could also build trees, right? Have you tried that and do they correlate with the mutation trees? So the issue with building de novo methylation trees is you'd have to restrict it to certain methylation sites that are likely to only change once or twice. So the ones that I showed you change once early in life and then stay stable. So those would be very hard to use because when you do a de novo tree, it would try and place it as one event, and then it breaks the phylogeny. So it's a bit trickier, but you'd have to find CpGs that change once-- That have a constant rate of change over life and then use those to build a tree. We can resolve our trees with methylation. So where we have what we call polytomies at the top of the tree where we don't know what the ordering is because there's multiple branches happening very quickly, because the methylation rate is so much higher, we can beautifully resolve those with the methylation. Great. Thank you. Really great talk. Could you comment on the role of these somatic mutations, methylations, the ordering in other diseases outside of cancer? And I feel like you're sitting on a dataset where you can start to look at that.Outside of oncology? Yeah. Outside of oncology. Yeah. Yeah, really great question. So when we look at some of these genes associated with these segments that change early in life, just gene set enrichment pathway analysis, weirdly, they're kind of neurological genes. And it's almost like methylation barcoding and methylation diversity is for neurons as DNA VDJ is for immune cells, strangely enough. And when we look at EWAS, EWAS is like the methylation version of GWAS hits in Parkinson's, Alzheimer's, schizophrenia. Some of those hits, because they're done in blood, are these early changes that are happening. So whether that's just coincidence because these are stable changes over life or whether they have some impact on neurological disease is currently under investigation. Thank you. How are we doing for time? I think we should probably move on to the next talk. Okay. Let me introduce Ben. I mean, Ben really doesn't need an introduction, does he? It's a real honor to have Ben Ebert speak, CEO of Dana-Farber and President Professor of Medicine at Harvard, and again, together with Sid, Ben and Sid really described this field, and opened up this new area for research. His lab focuses on clonal hematopoiesis, and the molecular basis of blood cancers, but also protein degradation for cancer therapeutics. So delighted to have you here, Ben. Thanks, Jyoti, and it's easy to give a talk now because you guys did such an incredible introduction to this field. So these are my disclosures. So I'm going to jump right into some of the consequences of CHIP that occur in terminally differentiated blood cells. And this was a finding that Sid made in his original analysis of exome sequencing data, that while we clearly saw that CHIP was associated with the increased incidence of myeloid malignancies, there was also an increased mortality that was beyond what we could explain from just blood cancers, and that came from some of the common age-associated inflammatory diseases, and that has subsequently been validated in a number of studies, both using human genetics, some using more population science epidemiology methodologies as well as animal models. But just in the heart, we can see evidence of increased atherosclerosis, increased myocardial infarction, heart failure, as well as arrhythmias, including atrial fibrillation. There's quite a strong association with chronic liver disease as well, with chronic kidney disease, pulmonary disease, osteoporosis, blood clots. I didn't add in the lung cancer findings more recently, that are actually mediated apparently by similar biology with increased levels of interleukin-1 beta in many of these circumstances being associated with these inflammatory consequences. Sid, in his own lab, has made the striking observation that individuals with CHIP have a decreased incidence of Alzheimer's disease and other neurodegenerative disease. So there's a broad effect of these mutant blood cells on their function and their risk of a variety of diseases. Been very convenient that one of the mutations that is most powerfully associated with inflammatory disease is TET2, and that's been very consistent across many disease states. And the animal model of knockout of TET2 recapitulates many things, as was described in Sid's talk. It recapitulates the competitive advantage of TET2 mutant cells. It recapitulates the increased risk of blood cancers, and it also recapitulates the increased inflammatory phenotypes. Sid initially showed that in atherosclerosis and the exposure to a very high-fat diet in LDL receptor knockout recipient mice, that just changing the genotype of the blood cells to TET2 knockout resulted in increased atherosclerosis, shown here in the descending aorta. In mice with TET2 knockout blood cells exposed to cigarette smoke, you can see greater airspace disease and mimicking COPD. And in chronic liver disease in mice fed a diet that predisposes to first fatty liver and then NASH, and then cirrhosis and fibrosis, you can see that just changing the genotype of the blood cells increases the reticulin staining here, leading to end-stage liver disease in these mice. In many of these cases, we can interrupt either interleukin-1 beta or the NLRP3 inflammasome, which is in the same pathway, and reverse these phenotypes, implicating IL-1 beta as a critical mediator of many of these phenotypes. And actually, one other thing that I might just add to all of these studies is that over the years, we've recognized that different genotypes have different risk of inflammatory diseases. And really, it's the individuals with the largest clones that have the greatest risk with a varying allele fraction of around 10% or greater. So that becomes an enormous clone comprising quite a large percentage of all the terminally differentiated blood cells in the body, and those are the individuals who carry the greatest riskSo what is the mechanistic basis of inflammation Tet2 knockout cells? We like to use genetic screens in the lab, and that is a challenging thing to do in vivo. And we're talking about very complex phenotypes where there's a lot of interaction between hematopoietic stem cells or differentiated blood cells and their surrounding environment. So we wanted to do this type of screen in the native microenvironment. A challenge for doing these large-scale in vivo screens is that you transduce a very heterogeneous pool of cells. Unlike CRISPR screens in cell lines where every cell is pretty similar, in vivo, the cells are incredibly different. Even if you try to purify stem or stem and progenitor cells, they have very different proliferation rates. So we wanted to be able to quantify not only the size, the number of cells with every guide, but every individually transduced cells using a unique molecular barcode. And so every guide RNA in this library has a unique molecular barcode, so we could track both the number of clones and the total number of cells transduced with each guide and trace that clonal progeny from infected stem cells and progenitor cells. The reason for this is just illustrated here. You can imagine that in some cases, a cell that is transduced might produce a couple of cells, and even if it's a non-targeting guide that has no biological effect. And another non-targeting guide could happen to transduce a cell that gives rise to many progeny. And you would think based on just the sequencing data that this guide causes the cells to be less fit than this guide, but it's just the heterogeneity of the cells. And using this approach, we can track the number of clones as well as the total number of cells. So we actually managed to do a genome-wide in vivo screen for inflammation in Tet2 knockout cells. We started with Cas9 transgenic mice, with or without Tet2 inactivated in the blood cells. We used this genome-wide barcoded library as an inflammatory stimulus. And I should also note that we think that really with any of these inflammatory states, CHIP by itself doesn't immediately cause any inflammatory diseases. It increases the risk and there needs to be another stimulus. In the case of cardiovascular disease, that could be a high cholesterol exposure, or in the case of gout, it's uric acid crystals, and that inflammatory response is exaggerated in the case of the Tet2 knockout. So here we used a classic stimulator of myeloid inflammation, which is zymosan from a yeast cell wall, and then harvested cells from the peritoneum, which becomes very inflamed in this circumstance. And we actually looked at the bone marrow and the peritoneum, looking specifically not just for clonal expansion broadly, but for genes that mediate inflammation and accumulation of guides in the peritoneum. When we did this, we were struck that a very large percentage of the guides related to lipid metabolism, and particularly one of the key regulators of the lipid metabolic pathway is ACLY, as well as OGT, which I'll talk about in a little bit. It also reminded us of some of the very early studies that we did, that Sid did, that showed sheets of macrophages, foamy macrophages in mice with Tet2 knockout exposed to a high-fat diet. So if you look at some of the individual genes that hit in this screen, they include ACLY, which is a regulator of conversion of citrate to acetyl-CoA, a number of genes that are critical regulators of cholesterol and fatty acid metabolism. And each of these individual mutations takes the level of inflammation in the peritoneum from increased inflammation with Tet2 compared to wild type back to baseline. So the same effect of the inflammatory stimulus in the Tet2 knockout and the wild type. If you do lipid profiling of these mutant cells, both in human cells and in mouse cells, you see a very dramatic increase of accumulation of triacylglycerols, diacylglycerols, as well as cholesterol esters in neutrophils, macrophages in mice and in human macrophages that have been edited at the Tet2 locus. And if you stain these cells, these green droplets are lipid droplets that we see in the Tet2 knockout cells much more than in wild-type cells. And in humans as well, using a different stain, you can see these lipid droplets in Tet2-edited HSCs, HSPCs, compared to wild type. So there's a quite profound metabolic defect related to lipid metabolism in these Tet2 knockout cells. If you look by gene expression profiling in the Tet2 knockout cells exposed to this inflammatory stimulus, it's quite marked compared to wild-type cells. But if you knock out ACLY, one of the critical regulators, you largely return the gene expression profile to the wild-type state. I mentioned that OGT was one of the other hits. That struck us because OGT is one of the primary binding partners of Tet2, and OGT is a regulator, is actually a sensor of metabolic state in cells, and it adds a sugar moiety, glycan acylation, to many different proteins in the cell in response to metabolic stimuli.We tried to look at that interaction between TET2 and OGT and map it. We did a base editing screen, tiling all of TET2 and found some hits right where we predicted the binding interaction between TET2 and OGT to be. And then we developed a TR-FRET assay to look directly at the interaction of the TET2 with OGT right at this location, and you can see there's a nice interaction here with wild type TET2 interacting with OGT. And if we mutate the critical amino acids shown here, we lose the interaction between TET2 and OGT. So we think there's a very specific interaction that occurs at this locus between TET2 and OGT. If TET2 is gone, OGT is liberated from that interaction. So there's a lot more to be worked out there, but we think that TET2 mutations result in a major metabolic defect and it's related to OGT and its metabolic sensing. If you interrupt, so if you knock out TET2, OGT is liberated from its interaction with TET2, it results in glyco-N-acylation that's aberrant. It results in this lipid accumulation. If you block AOLY, you remove that lipid droplet formation. Lipid droplets are known to cause inflammation, and that, again, can be inhibited by various inhibitors of lipid metabolism. It also aligns with findings from our lab and a number of labs that have shown mitochondrial defects in TET2 knockout cells as well, which we think are probably related to the substrates going into the mitochondria. And then quickly, I just wanted to note prediction of outcome for CHIP, because this is critical for us to think about how we are going to intervene in individuals with CHIP who might have these various outcomes. We did this in the UK Biobank in 500,000 individuals. If you look at instant myeloid malignancy in individuals with CHIP, you see a huge variation in the risk of progression, DNMT3A heterozygous mutations being among the lowest, splicing factor mutations being among the highest. This aligns very well with the PACER analysis that Sid showed us, though there are exceptions like PPM1D, which had a higher rate of progression in the PACER analysis but has a very low risk of developing a myeloid malignancy. DNMT3A mutations, though they have a very low risk compared to the others, still, of course, increase the risk of progression to myeloid malignancy. It's just much smaller than the others, at least for the heterozygous DNMT3A mutations. By putting the various factors that we observed together, which are quite simple, what the genotype is that's driving, how many mutations are present, how big the clone is, RDW, which in all studies have been very powerfully associated with all the adverse outcomes, MCV, cytopenias, and age, get a very simple score that really subdivides the risk. High-risk individuals, which fortunately is only about 1% of individuals, have getting close to an inevitability of progressing to a myeloid malignancy, and there are already multiple clinical trials underway to intervene on risk of progression, specifically in this subgroup. Again, that's only 1% of individuals, which is good. 90% of individuals are in green in the very low-risk group. Only a threefold increased risk of developing a myeloid malignancy, which is a very low baseline, so the absolute risk is very low. And then the intermediate risk has an intermediate, about a 35-fold increased risk, but still a relatively small overall risk. And even though we built this model for progression to myeloid malignancy, it correlated very powerfully with overall survival. The high-risk group having about 50% overall survival, low risk and no CHIP having about a 95% survival. And this has been validated now in a number of cohorts, but illustrating that even in this cohort that was very different from the UK Biobank, in that it was older and less healthy, the absolute risk of heme malignancy is most affected by this risk group, but the total mortality is still driven by cardiovascular disease as the major cause of death, though solid organ malignancy also was significantly elevated, particularly in the high-risk group. So just wanted to note that, again, CHIP is extremely common, associated with overall mortality, increased risk of a variety of inflammatory diseases, and that TET2 inactivation exacerbates these inflammatory phenotypes and alters lipid metabolism, and intervening on that lipid metabolism removes the inflammatory phenotype. And risk of progression to malignancy or to overall mortality is affected by individual genetic alterations and can be predicted by this and other clonal hematopoiesis risk scores. And again, highlighting the seminal work that Sid did, as well as various members of the lab. The work I talked about today was led by Peter Kim and Chris Hergot here. Happy to take any questions. Yeah, go ahead. Beautiful talk. I was wondering, when you see infection, cancer, or anything other than autoimmune, you start to wonder about sex effects. Mm-hmm. And so I was wondering, do you have a component for mosaic loss of sex chromosomes in your risk score? And to what extent is there an overlap between the effect of any given clonal hematopoiesis-associated allele? It's a great question. Minus Y is included in there, but there is clearly an effect of sex on each of these individual mutations. The vast majority of mutations are much more common in males than females. DNMT3 mutations are slightly more common in females than males. And we've actually been able to recapitulate some of that in animal models by just modulating hormone exposures and estrogen-driving clonal advantage in DNMT3A. Very cool. Thank you so much. Hey, Tim. Ben, great talk. Fantastic work, as always. Clonal hematopoiesis of indeterminate potential. Since the algorithms are so good now, is it time to drop IP and just call it- Yeah ... clonal hematopoiesis? I suppose every human disease has some indeterminate potential. We were trying to align with MGUS, which also has almost an inevitable component to it. But it's still a smaller percentage that fully progress, which is the idea. It's a good question. My question's related to the ACLY component. Yeah. Do you see the same dependency in earlier, less mature cells being MPPs and HSCs, or is this lipid accumulation phenotype restricted to these different- Great question. We're doing that now, and it is not a trivial experiment because ACLY is essential for all cells, and so doing that genetically carefully is hard. There are some ACLY inhibitors, some of which have been moved forward clinically, but they are developed to have their primary activity in the liver, to modify lipid metabolism. And so, understanding the tet2 specificity is not trivial, but we think it's there. But we're still working our way through those experiments. Good question. Thanks. Question? Yeah. Wonderful talk. Just a quick question. Building on what you mentioned about the lipid metabolism, do you think there's a role to investigate CHIP on tumor microenvironments and how it would evolve, like response to immunotherapy and things of that line? Absolutely. I think the most elegant work has been done in lung cancer, where infiltrating tet2 mutant clones seem to increase the risk of lung cancer. That also aligns with the striking finding of the CANTO study, in which a large number of patients, about 10,000 patients in a randomized controlled clinical trial, were treated with an IL-1 beta inhibitor called canakinumab, and the risk of lung cancer was approximately halved in the individuals who got canakinumab. And when mature lung cancers were treated with canakinumab, it didn't have great efficacy. So it may be in those very earliest stages where there's an initial mutation that the inflammatory microenvironment that CHIP is part of, can lead to whether that clone becomes established, expands, and then ultimately progresses to lung cancer. So, yes, I think at least in the case of lung cancer, that seems to be true and may be true in other cases as well. Yeah. Thank you very much. Great work again. I just remember from the very first paper of DNMT3A that Dr. Lei published in the "New England Journal." Yeah. There were two mutants that were recurrent. Yeah. The rest were one hit each. Yeah. Now with more sequencing, maybe this is not completely true. So the question is, are they all relevant or only some that are recurring and have functional impact? Yeah, great question. It's a loss-of-function mutation, but as Tim showed and has held up both in CHIP and in mature MDS and AML, there are hotspot mutations, and those hotspot mutations that are dominant negative mutations. And so if you look at those mutations, specifically, the risk of progression is greater than a heterozygous loss-of-function mutation and associated with worse outcomes as well. So in the model, when they say DNMT mutation, any one counts or only the hotspots? Yeah. We have lumped them together, which is an apples and oranges. It's probably true for many mutations, P53 as well. They're hotspot mutations that act as dominant negatives. So, we do sometimes lump just because even with these huge data sets, you start to get into sample size problems when you get down to individual types of mutations. Thank you. Great talk. My name is Amulia Singh. We work on blood tests to detect lung cancer, and so almost everything you all said was fascinating in that light. For your COPD phenotype, do you see it sort of differential or kind of a differential diagnosis where you have individuals who have COPD but don't get lung cancer, and what does that look like relative to those who do? Yeah. It's not something that we've worked on ourselves, but that's a great question. We certainly haven't looked at that at all. But that's a good question because that is the at-risk population and whether there are other stimuli that lead somebody in a more oncogenic direction versus airspace disease or whether that's a stochastic event. So, great question. Thanks. Great. We can now open the floor for a panel discussion, for the last five minutes actually, because there are quite a lot of questions online, and if there are questions for any panel members, please come to the microphones. So the first one is for you, Sid. Is expression of TCL1A increasing with age in the healthy population? That's a great question and one that we're looking at right now. So I don't have any data yet, but that's something that we're going to want to look at. It could be complicated because, as I mentioned, clonal hematopoiesis with unknown drivers seems to be strongly associated with the TCL1A SNP. So if you start to look at older individuals, many of them will have these unknown expansions. Maybe could I ask Jodi a question? In your methylome analysis, have you seen any TCL1A promoter hypermethylation? Yeah. Well, we'll check when we get back. Now, I'm looking at Shin in the audience, who's probably going to be very keen to check. We haven't looked. Question for both of you, actually. Were there any protective environmental factors for CH expansion that either of you have come across? Pradeep Natarajan published a paper showing that a vegetarian diet actually had a lower risk of CHIP, so that's one that I can think of. But as Sid said, the exact relationship between CHIP and inflammation from a human genetic standpoint has been difficult to see. Yeah, I don't have anything additional beyond what I showed. Smoking seemed to be the strongest risk factor, so not smoking is protective. But yeah, I'm not sure. And perhaps, I think there are sex hormone differences that are driving clonal selection in ways that we don't understand yet, or hematopoiesis in ways, and we've always known that there's differences in blood count parameters between males and females, so there's some interaction there. Perhaps the most powerful selective pressure that we see on a routine basis in all of our cancer centers is chemotherapy and radiation, which has an extremely powerful selective pressure for p53 and p53 pathway mutations, which ends up being the basis of therapy-related malignancies, ultimately. And I suspect the protective factors will be very much gene dependent. So, with JAK2, for example, you can do these polygenic risk scores for all your different blood counts. Are you likely to have a high hemoglobin or a low hemoglobin in the population? And for example, if you're likely to have a low platelet count in the population and you acquire a JAK2 mutation, you're much less likely to develop an MPN, and you're more likely to have just clonal hematopoiesis due to JAK2. So your germline polygenic risk scores, as you've also shown, can sort of influence selection on specific drivers. There is an interesting question here online. In germline carriers of DNA repair defects, do you see more CHIP? We don't at the moment, but we haven't looked in detail. Any thoughts on that? So there are, for all of the really powerful germline predisposition to leukemia disorders, clonal hematopoiesis does develop early in life and is a predictor. So patients with RUNX1 mutations or Shwachman-Diamond syndrome, and the particular mutation that is inherited does, to a large degree, dictate what those initiating mutations are. So individuals who inherit one RUNX1 allele mutated will acquire a mutation in their second RUNX1 allele. Patients with Shwachman-Diamond syndrome will develop p53 mutations as one of their initial drivers. So the inherited allele does have a powerful effect on the initiating mutation. And sometimes what's growing out in that context can be maladaptive. But occasionally, what's growing out in that context can be germline rescue. Yeah. So it can be a really powerful tool for discovering what actually reverses the germline defect because those good clones will also grow out in addition to the maladaptive clones. It could be a powerful discovery engine. Some more questions on the floor? A quick question about GATA2. Very famous epiallele, right? Because when you methylate the wild-type allele and the mutant allele is expressed on the other one, it's associated with progression to monoMAC and AML. Did you look at GATA2 in your studies to see whether it was affected early on? Sorry, didn't quite catch the gene. Which one? GATA2. GATA2. It hasn't come out in our unsupervised screens. We will do a more supervised screen on known drivers. So far, what we're finding is we're finding some early signals in known tumor suppressor genes where there's promoter methylation early in life. What's tricky is knowing which segment in the promoter is critical. So there might be five segments in the promoter, one of which has shown this beautiful allelic methylation in association with the lineage. Is it changing gene expression? We don't yet know, but we will check out GATA2. We haven't looked at it directly. It's very susceptible to allele-specific methylation. It just might be a place to look because there's something very special about it. We should be able to. We have genome-wide coverage. Yeah. Thank you. My question, in general, is I guess related to clinically, we now sequence all of our cancer patients and find their CH incidentally, or their primary care found them. And obviously, we shouldn't be changing people's chemo exposure at this point without a randomized controlled trial to suggest that avoiding doxorubicin prevents therapy-related myeloid neoplasms. But I'd just like put it to you three experts on where you think we should be going and what types of questions we should be asking, and how do we incorporate this data in our care of cancer patients over the next five years? And I can start. So, we've looked at the effect of chemotherapies, lots of different chemotherapy agents, both for solid tumors and liquid tumors in normal blood cells. And sometimes you find quite interesting things. You can find that two agents within the same class that actually have quite similar efficacy, one is mutagenic to the blood cell and one isn't. So in some scenarios like that, clinicians may be quite happy to switch. Um, we found that long-term alkylator use can be mutagenic. So you can have two routes. You can have direct mutagenesis to cells, or you can just have clonal selection, such as on PPM1D and TP53. I think what will be quite interesting is to survey that before you start. So if you have someone who has a history of chemotherapy use and is about to get more treatment, surveying what proportion of TP53 selection there is already in the marrow may be quite useful. So not only might be useful in the future in terms of making our chemotherapy regimes a bit safer, but it could be useful for predicting how damaged wild-type hematopoiesis is following the first bout. I don't know if there's anything to add. Yeah, we see it a lot because it's in the sequencing reports. I think the vast majority of cases it wouldn't give it a thought in the midst of therapy, but as those patients become cancer survivors, it becomes perhaps a bigger issue or to identify the people who might be at greatest risk of a therapy-related malignancy or something. So, they end up in our survivorship clinics and thinking about it. And the other point is that the tumor-infiltrating clonal hematopoiesis may be a little bit different than just identifying the mutations in the blood, and the associations with lung cancer most powerfully with the individuals where the cells are at a higher percentage in the tumor. So, I think we're probably too early to know how to change therapy. And as you said, you would never change therapy without evidence from a randomized trial first. In your descriptive... So the question's around whether we've stratified by hormone therapy. We're looking at a cohort to look at just that question right now, but don't have any answers yet. Okay, probably the last question. Thanks. Oh, just a quick question, more of a technical question. Thank you for the great talk. For the epigenetic phylogenic trees, did you use short-read sequencing, and have you done it with long-read sequencing? You know, because that has better- So it's Illumina short-read sequencing. So with long-read sequencing, you need large DNA input, whereas we have two nanograms of DNA. Okay. So you can do long-read sequencing, but it's going to be on the whole sample. So it's useful for phasing the germline, for example. But we can phase about 15% of sites on short-read sequencing. Gotcha. Thank you. Great. Okay, well, with that, thank you all very much for attending in person and I'll close the session. Thank you to the speakers.