 I think I will assume. It is my pleasure to introduce our speaker today for the first scientific director, Lectures Theories on Cancer Research, Jeff Trent. Our speaker today is Todd Gallop of the Broad Institute. He is the director and founding core member of Broad Institute of MIT and Harvard. He's a world leader in using genomics tool to understand the basis of cancer. And he pioneered the development of new cell-based approach for drug discovery in cancer and other diseases. Third is Charles A. Diner investigator in human cancer genetics at the Diner Father Cancer Institute and professor of pediatrics of Harvard Medical School. He is the recipient of multiple awards, including the Outstanding Achievement Award from the American Association for Cancer Research, the Paul Max Prize for Cancer Research from Memorial Sloan Cancer Center, and the Dan Land Prize from the American Physiology Society. And in 2014, he was elected to the National Academy of Medicine. So please join me in welcoming Dr. Gallop to give a Jeff Trent lecture on cancer research. Thank you. Thank you so much for having me. Back to NHGRI, it's great to be back. And I have to say, especially, an honor to give the Jeff Trent lecture. Jeff Trent is one of those real pioneers of what we now call cancer genomics as a field. But Jeff really should be credited with some of those earliest ideas for how genetics and genomics might impact not only the cancer research community, but impact patient suffering from cancer. So it's really a pleasure to be here. First, a few disclosures. I've had the privilege of helping to advise a number of biotech companies. And we have some sponsored research funding from a couple of companies. So this is, I think, for most people in the cancer field, take a slide like this for granted now. It's kind of a rubric for how we think about cancer precision medicine. And that is that a patient's journey would start with a sample of a tumor that would then be genomically characterized. And by tumor genome here, I mean not just the sequence of the tumor, but other molecular characteristics expression, proteomics, what have you. And based on that omic characterization, one would develop insights that would then lead to the selection of appropriate therapy, which targeted therapy, which would then lead hopefully not just to an initial transient response, but to durable responses that really improve quality and duration of life for those patients. Does everybody with cancer in 2023 benefit from this paradigm? Absolutely not. It's still the minority, I think it's fair to say. But I don't think it's a wrong idea simply because everyone today doesn't benefit from it. It's the right idea. I think there's no going back. It just means there's a lot of work to do. And I'll tell you a little bit about what our lab and others at the Broad are thinking about to try to make this a reality for more and more patients. I want to start just for fun with where I started in this field as a pediatric oncology fellow, as a postdoc, where I actually studied one of my own first patients as a pediatric oncologist, a kid with acute lymphoblastic leukemia. And based on the genomic methods of the time, discovered a new chromosome translocation, including a new gene now known as ETv6, which this was pre-genome sequencing era. And so you could discover genes back then. Discovered this gene fusion as a result of a cryptic chromosome translocation, a 1221 translocation. This fusion turns out in follow-up work to remain the most common gene fusion or chromosome translocation in childhood ALL. And it also turns out that the prognosis of those patients, those kids given standard chemotherapy, is extraordinary. Well north of 90% cures given chemotherapy. And that's allowed clinicians to now scale back the intensity of therapy for those patients if you're positive for this particular gene fusion. And that's important for kids where giving cytotoxic chemotherapy has quite substantial long-term side effects. So I think that's all good. And I'm proud of that initial discovery, which was now almost 30 years ago. I will note that just the year prior to this, another important leukemia discovery was made by Paul Yu and Francis Collins discovering a translocation involving another component of this so-called core binding factor complex, which made this even more interesting because it was clear that there was selection for rearrangements of this gene family. So then both these discoveries about 30 years ago. And so I think we have to ask the hard question of, is this a victory for precision medicine? On the one hand, it is because there's dose escalation for these kids with this particular gene fusion. But on the other hand, we're 30 years in. And as far as I'm aware, there are zero drugs in development, serious development in the industry to go after this fusion. And I think that's largely, it's not just because it's relatively uncommon childhood cancer where the commercial opportunity is large, but it's also because this involves the fusion of two transcription factors and transcription factors being largely relegated to the so-called undruggable class of proteins unlike enzymes, for example, that are more straightforward to target with small molecules. I think if we're really serious about precision medicine, not just for oncology, but in general, we need to get more serious about making it possible to develop therapeutics that target whatever the genome is providing. And I'll say a bit more about this towards the end of my talk. So what are some of the approaches that we can take to approach cancer discovery from a cancer precision medicine perspective? The two obvious ways are to learn from patients and to learn from experiments in the lab. And I would say that people tend to have bordering on religious views on which of these approaches are more important. The argument in favor of learning from patients is, of course, they're patients. And there's nothing more physiologically relevant than a patient suffering from cancer. That's good. But the challenge there, of course, is that there's probably no limit to the types of observations you can do in patients or in materials obtained from patients. But there is a limit to what you can do in terms of perturbations. That is clinical trials, which for any one patient will be limited in number. And then following patients in the routine clinical care. On the other side of the coin, we can learn from experiments in the lab. For example, cell lines growing in Petri dishes. There you can do an infinite number of perturbational experiments, which is how we learn best in general, but they suffer from their cell lines growing in plastic and they're not real tumors and real people. So I think rather than sort of arguing about these two, the goal really needs to be to take both approaches and I'll give and integrate them together and I'll give some examples of how we're trying to do that. First, starting and learning from patients. And of course, it's fun to give this talk at the NHGRI because possibly the single, the project that wound up having the single biggest impact on the field of cancer research in recent memory was the Cancer Genome Atlas project, which is a collaboration between NHGRI and NCI. This was a big bet for NCI, a really big bet for NHGRI relative to the overall budget. But this just changed the way, I think the whole field thinks about the role of the genome in cancer in a way that there's just knowing back that had a follow up international cancer genome project, the ICGC. And now, it's fair to say that cancer genome testing has become commonplace and standard of care throughout this country and many other countries, of course, largely happening in commercial testing settings in companies that are doing panel testing of 300 or 400 genes, increasing number of companies sequencing the entirety of the exome in tumors. So such that in the US alone, there are now hundreds of thousands of patients that are having their tumor sequenced in companies. And that's great and can be useful for those patients. But I wanna flag that I think the real challenge here is that we're not learning as much as we can and should from that patient experience. Most of the data that are generated by commercial testing companies, where that testing is being done at very high quality, goes into a black hole and is not available to research community to learn from. It's not that they're opposed to making the data available, it's not their mission to support research. And I think this is, I'll come back to this again as something I think we need to worry about not just in cancer, but beyond cancer is increasingly genetic testing becomes routine and becomes the work of the commercial world. How do we nevertheless learn from it among the research community? I wanna say a little bit about direct to patient efforts because in addition to the large academic medical center ways of focused mechanisms for doing cancer genome studies or the commercial testing I just mentioned, what about direct engagement of patients? And this is particularly important in oncology because as you may know, the majority of patients with cancer in the US are not treated at major research medical centers. They're treated by community oncologists, about 85% of those patients are treated in the community and only a tiny sliver of patients actually enrolled in clinical trials, mostly at major medical centers where there's a mechanism to ask patients whether they'd be interested in being study participants with informed consent and so on. And so most patients, cancer patients are very interested and eager to participate as partners in research but most of them weren't asked because they're treated in the community by busy practicing oncologists in private practice. And so the question was, could we connect with those patients directly using social media, not having to go through a major medical center, a research center to obtain informed consent and obtain samples from patients. And so the goal of this initiative which we've called Count Me In or CMI is to make it possible for any patient anywhere to contribute their genomic and clinical information as partners in research. And to again use that and to do that by using modern social media mechanisms to reach patients and obtain informed consent online. So I'd like to give a couple of examples and this is the work of Nick Wagle who spearheaded the Count Me In initiative initially focused on metastatic breast cancer. As you may know, breast cancer was a big component of the TCGA, initial cancer genome analysis project but that study focused on newly diagnosed untreated patients which was really important but there were really no large scale studies of metastatic breast cancer and that's what unfortunately patients who die of breast cancer die of metastatic disease. And so the question here was could we reach out into the community to obtain samples, do sequence analysis and with patient provided clinical follow up information for metastatic breast cancer and the answer was yes. And so very quickly the project was able to accumulate several thousand patients with metastatic breast cancer from all 50 states and parts of Canada as shown on the left. And interestingly as shown on the right with the size of these dots representing the number of patients that came from different centers where a center may be a practicing oncologist in the community, most of these patients came from different treatment facilities, most with an N of one patient from that one practice over 1600 different practices represented in this study. So I think that's really interesting and encouraging. I will say that with respect to this morning's discussion around the importance of diversity in genetic studies, this seems like it should be a mechanism to better engage patients as partners in research across the diversity spectrum. That doesn't and has not happened automatically simply by capturing all 50 states since there's an active component of the count me in project that is specifically trying to engage underrepresented patient populations. So metastatic breast cancer is unfortunately a common type of cancer. What about using the direct to patient approach to study rare cancers where this might be even more useful because even the major academic medical centers don't see enough of rare tumor types to have a large experience. So we initiated another project. This one led by Corey Painter for a tumor called angiosarcoma, a rare cancer with only about 300 cases per year. Corey herself as a survivor of angiosarcoma. So she was a passionate champion of this project. And what was remarkable here is that despite this being super rare because the patient communities were already connected is very easy to connect to them through social media. And we had about 300 patients within the first few months of this study did genetic analysis of those tumors together with the patient provided clinical information. And a couple of findings came out of that initial study which is published in Nature Medicine. Again, led by Corey and Nick Wagley. One was that particularly for some of these patients the number of mutations somatic mutations was quite high the so-called tumor mutation burden was quite high in these angiosarcoma patients. And second, while the numbers were small there were a couple of anecdotal responses of patients who had responded to immune checkpoint blockade therapy being given off label. Enough of a signal there while it didn't reach statistical significance to say, oh, that kind of goes together, high mutational burden response to checkpoint blockade. Somebody should do a clinical trial in angiosarcoma testing immune checkpoint blockade. That study was done, we had nothing to do with it, was reported at ASCO just this summer showing quite substantial responses to immune checkpoint blockade. So I think this is exciting because it suggests that particularly for rare cancers engaging patients wherever they are may be the most effective way to do this type of cancer genomics research. And beyond that, I think we're excited to be extending this count-me-in approach beyond rare cancers to rare non-cancers because it's the, to rare non-cancers because those rare disease of any kind, no, even the large academic medical centers just don't see enough to have enough experience but if we could reach them anywhere in the country and eventually anywhere in the world that could we believe accelerate rare disease research quite dramatically. And I should mention, of course, that as we do for all projects at the Broad, as these genomic data are generated, they're made available for the research community to benefit from. So I think these kind of genomic studies are really powerful for a snapshot of analyzing the genomes of cancer but you'd like to get more than a snapshot. And so this is connecting to the prior session a bit because the revolution of blood biopsy methods, essentially the same kind of approaches as for NIPT are becoming very exciting in the world of oncology. And that is the ability to use, to look for circulating tumor DNA, free DNA, in the peripheral blood of patients, either for screening or for following minimal residual disease or MRD response to therapy and then capturing a patient's recurrence before the tumor burden is so high that it's hard to get on top of it with salvage chemotherapy. This works far better than I thought it ever would and it's not just in our hands but in multiple groups showing that this works. The sensitivity though is still a real problem. In the screening setting, sensitivity particularly to detect early stage tumors is quite low. And in the recurrence setting, in the MRD setting, the ability to identify recurrences as early as you'd like is also limited. And so the question is why is that? Is it a failure of the tests themselves, of the molecular biology of the test or is it something else? And we believe that it's actually something else that's as simple as there aren't enough molecules of the tumor-derived DNA in the tube of blood. If you estimate that there are, let's say a billion cancer cells in a stage one cancer on average, plus or minus. And if you believe, as most people do, that most of the cell-free DNA and blood is coming from only the proportion of those cells that are dying, that's still about 10 to the seventh cells. If it's about, you say, a 1% dead or dying cells, that's still about 10 to the seventh cells. And yet the estimates of the number of cell-free, or of tumor equivalents in a tube of blood that goes into blood biopsy tests, it's on the order of a cell, or sometimes less than a cell, in those MRD kind of settings. And so the challenge is that if there aren't tumor-derived DNA molecules in the tube of blood, it doesn't matter how much you amplify the signal that's in the tube and whatever your test is. If there's no substrate there to begin with, you're out of luck. And so the question is, why does that happen? It is likely the case that there's some tumor types that shed less of their tumor DNA from those dying cells into the blood. But we know that a lot of the reason there's so little circulating tumor DNA in the blood is because it's so rapidly cleared, for example, by cup for cells, the macrophage-like cells in the liver. And so if that's the case, this rapid clearance, is there anything you could do about this? And so I wanted to share the work from the Gershner Center for Cancer Diagnostics at the Broad Institute in two projects led by Victor Adelsteinsen, one together with Sangeeta Bhatia from MIT, and one from Chris Love also at MIT, trying to address this question of, could you transiently pause this clearance of cell-free DNA by one of two strategies? Either transiently occupy these cup for cells so they cannot as effectively engulf the cell-free DNA, or give individuals a monoclonal antibody drug directed against DNA, thereby blocking it from this cup for cell mediated uptake. And so this is, of course, just in animal studies, but in mouse models bearing xenografted human tumors, they're able to show that, in fact, you can boost by 10 to 100 fold with either of these two strategies, the amount of cell-free DNA that you can collect from the animals that's a transient response. So the idea would be that, and we're interested in bringing this forward to patients in the future, could you give patients prior to their blood biopsy test an injection of an agent like this, and then 30 or 60 minutes later, collect the regular tube of blood and do whatever standard downstream assay you'd like to do? So we'll see whether that will work in patients, but I think it looks encouraging because the power of being able to detect cancer genomes in blood is pretty profound, and if we could optimize that even further, that would be a very good thing. Now I wanna turn to from how we can learn from patients to how we can learn from experiments in the lab, and of course, I mean, a lot of things you can learn from laboratory studies, but here I mean, primarily systematic genome scale types of studies that you're all very familiar with. So I wanna say a little bit about a project that I'm particularly proud of at the Broad that we call the Cancer Dependency Map, or DETMAP for short, and the concept here is simply to take a large number of patient-derived, human-patient-derived cancer models, cell lines, and we now have more than 1,000 of those that have been completely characterized, and by that I mean whole genome sequencing, whole exome sequencing, RNA-seq, various degrees of epigenetic profiling, proteomic analysis, metabolite profiling, and then for each of those 1,000 cell lines, subject them to genome-wide CRISPR, Cas9, or in some cases, Cas12, loss-of-function studies. So from this matrix of 1,000 cell lines, completely genomically characterized and systematically perturbed, knocking out every gene in the genome, you get a data matrix that allows you to create a map of cancer dependencies or vulnerabilities, where obviously if you knock out a gene and none of the models care with respect to their survival, that's not very interesting. If you find something that kills everything, that's probably not interesting either, because that will be something that's likely to just be toxic because it's a common essential gene, but if you find knockouts that produce a selective vulnerability, that's the most interesting with respect to identifying a future therapeutic target. So again, the idea is a matrix of now over 100,000 molecular features for each of the 1,000 cell lines, a dependency profile based on either CRISPR knockout studies or systematic small molecule treatment as another form of perturbation. We've done a limited number of CRISPR activation studies to look at gain of function across the genome, but not at depth map kind of scale, and then use what are pretty standard machine learning approaches to find predictors of particular vulnerabilities. All of these data are made publicly available and accessible through a quite useful portal called the depth map portal, and I'm pleased that the usage of this is now become pretty ubiquitous in cancer labs, particularly in industry, where now I think when people are considering a therapeutic target, first thing they'll do is to say, well, okay, what is its pattern of vulnerability look like in the depth map? And if it shows no pattern of vulnerability, you should probably think about other targets. I also just mentioned, but not going to detail, that there are a number of follow on studies as part of the next generation of the depth map project, which is led, important to say, by Paquita Vasquez, also with Bill Hahn and Bill Sellers and previously Jesse Bohm. New cancer models, patient derived models, particularly focusing on the models of cancers that are not well represented across the compendium of existing cell lines, and this is being done as part of the human cancer models initiatives funded by NCI. Parallog screens, these are screens largely led by Bill Sellers at the Broad, knocking out not just single genes, but pairs of genes, particularly paralogs, that often have redundant function. And so you could imagine, for example, that a small molecule might do a good job of identifying two paralogs, related paralogs, whereas individual genetic knockout of either one may not reveal a phenotype. And so those studies are ongoing. There are a number of in vivo screens that are being conducted to identify, are there vulnerabilities, for example, that look amazing in vitro? But if you could do those studies in vivo systematically, could you eliminate false positives and, or could you actually discover vulnerabilities that are uniquely found in vivo by our mist in vitro? And it looks like there may be some of those as well. And of course, there are always more kinds of omic characterizations and a growing number of cell lines that one could study. I wanna share one vignette that my lab is focused on coming out of the DETMAP project and that's been focusing on ovarian cancer, which as you probably know, has a real dearth of effective treatments or at least treatments that give sustained benefit to patients as opposed to transient responses, for example, that occur with cisplatin therapy, which is very effective in inducing remissions, but unfortunately, often as accompanied by rapid recurrence. And so here, Daniel Bondeson as a postdoc in the lab was interested in asking the question, are there selective dependencies for ovarian cancers selectively compared to other lineages, the measure of skew of the data towards ovarian cancer in that way without describing this plot in great detail, he was able to show that in fact, there was one protein in particular called XPR1 and a related protein called Kiddens 220, previously not particularly well annotated protein that were outliers with respect to being ovarian cancer dependencies, but not dependencies elsewhere. So this was interesting. It turns out that XPR1 isn't the single phosphate exporter encoded by mammalian genomes. And we're able to show in experiments that I don't have time to describe today that Kiddens 220, this previously unannotated protein, is part of an XPR1 complex. And if you have Kiddens 220 loss of function, you lose this XPR1 phosphate export activity as well. So XPR1, that's an interesting and phosphate homeostasis, is that an interesting therapeutic target and are there predictive biomarkers beyond the ovarian cancer lineage that might identify XPR1 dependent tumors? And so here we just looked this up on the debt map and said, what's the top predictive biomarker of XPR1 dependency? And it turns out that the most predictive molecular feature of XPR1 dependency is a phosphate importer known as SLC34A2. So the dependency is a phosphate exporter. It's predictive biomarker is a phosphate importer. That's unlikely to be seen by chance. It's interesting also that it had been previously established that SLC34A2, the importer, is under the transcriptional regulation of the lineage transcription factor Pax-8, which is known to be essential for the survival of ovarian cancer cells and their fallopian tube progenitors so that with this, you can't dial down Pax-8 transcription factor because it's work itself is required for survival. That drives expression of the importer SLC34A2. So you can see on the right that for those cancer cell lines that do not have high levels of expression of the importer SLC34A2, they don't care whether you knock out XPR1 with respect to survival, whereas in those models where the biomarker is high, you have a high degree of lethality associated with knockout of XPR1, the phosphate exporter. We went on to show that the, this predictive biomarker, the importer, SLC34A2 is highly expressed, shown by immunistic chemistry on the top panel. And interestingly also, XPR1 itself is the target of frequent genome amplification in ovarian cancer suggested that it must, for reasons that we still don't understand, be under some kind of positive selective pressure such that you see this amplification. And so the model that we're working under is that under normal cells, there's a normal homeostatic balance between the phosphate importer SLC34A2 and the phosphate exporter XPR1. In the case of ovarian cancer, because of this Pax-8 driven high level of expression of the importer, the cell needs to have a commensurate high level of XPR1 expression in order to compensate for this high phosphate influx to keep the cell in homeostasis. And so when you then genetically ablate XPR1, the cell collapses because of unopposed phosphate imports and we're able to show, Daniel was able to show that, in fact, you do see accumulation of phosphate within the cell leading to that cell death. But here again, we come to another challenging to target, therapeutic target. So the biology and the genetics is saying XPR1, that's the thing you should go after for ovarian cancer, but it's a large transmembrane protein. You might say, oh, it's easy just make an antibody, but these are actually tough antibodies to make. And so how could you go about trying to target this with some small molecule strategy? And so here, Daniel Bondeson again, turned to genetics and functional genomics to answer this question and used systematic base editing to tile across the coding sequence of XPR1 and ask which of those mutations were able to rescue XPR1 deficient cells from this cell death phenotype and which of those mutations were deficient in their rescuing ability. And what he was able to show is shown on the left hand side that there are two particular regions denoted A and B that are part of the XPR1 protein previously described as the so-called SPX domain. And it turns out that these two regions of mutations that resulted in loss of function directly oppose each other on the predicted, the alpha-fold structure of XPR1 as shown on the right. It also turns out that it had been previously recognized that this XPX domain is involved in binding of anositol pyrophosphates which are the downstream product of a couple of enzymes in particular an enzyme called IP6K. And the binding of these anositol pyrophosphates to the SPX domain results in derepression of XPR1 phosphate export function. So this raises the question that if you had an IP6K inhibitor, might it phenocopy XPR1 genetic loss of function that we discovered in the DETMAP? It turns out that there's a company, Schoia Pharmaceuticals in Japan, that actually made a quite good IP6K inhibitor. So we synthesized that compound and then tested it in a handful of ovarian cancer cell lines, some which have the high SOC34A2 biomarker and some that do not and showed that in this limited number of cell lines at least that this small molecule SC919 was able to phenocopy XPR1 loss. At least a few biomarker high and a few biomarker low cell lines. So that's encouraging, but from a genomicist's perspective, having a few of this and a few of that isn't particularly satisfying. And we'd really like to be able to do is to ask whether that really holds up across the scale of, for example, the DETMAP. That is, does a small molecule phenocopying the genetic loss hold up across the genetic diversity encoded in the DETMAP. And so to do this, we turned to a method developed by a postdoc in my lab called Channing U, a method that we call Prism. And this is a very simple idea, which is to say what if we introduce into different cancer cell lines, now we've barcoded over a thousand of them, a unique molecular barcode randomly integrated into the genome that then would allow you to pool hundreds of cell lines together and then treat with some perturbation of interest, genetic perturbation, the small molecule perturbation and simply count the barcodes. So that if you, a particular barcode was depleted, for example, treating with a small molecule, you could infer that the cell line that was tagged with that molecular barcode must have been killed. Simple concept, but means that you can do like 500 experiments at once in an internally controlled way. So again, we call this method Prism. And so here, what we could do is say, all right, let's take the Prism profile and we did this with just with 250 human cancer cell lines, treat with this IP6K inhibitor SC919 and then simply correlate it to all the existing depth map data and ask which of the genetic perturbations, the CRISPR knockout across the entirety of the genome are best correlated with the growth inhibitory activity of this small molecule. And you get this very dramatic result. So this looks like a line, it's actually a lot of green dots all lined up. And then you get this hockey stick showing that number one correlate, CRISPR knockout correlate of this small molecule is XPR1 and number two is Kitten's 220, which as you may recall, we showed was an obligate binding partner to XPR1. So this tells you a lot. This tells you that initially encouraging pattern that we saw that a small molecule now phenocopies and recapitulates the genetic ablation of XPR1 holds up across at least 250 cell lines. So that's good to validate the hypothesis. And it also tells you that this is a very good small molecule to the extent that this is a pretty good test of off target activity of a compound. Because if you have off target activity, it often manifests as a growth inhibitory signal. And we don't see that here because we only see it correlating with genetic knockout of its intended target. So I think we're increasingly using prism in this context to validate targets, intercredential the selectivity of small molecule inhibitors. I wanna just spend one minute, I added this slide in after hearing this morning stimulating discussion on Parkinson's disease and Francis' question about alpha-synuclein. And this really isn't ready for presentation, but I put it in any way, it's a preliminary result, but just to stimulate you with how we're thinking about could we use prism outside of oncology? And that is, could we understand the growth inhibitory effect of alpha-synuclein by looking at the genetic diversity of these human cancer cell lines? And so we did the following kind of crazy experiment, which is on a lentivirus to take an alpha-synuclein transgene that encodes a human variant, the A53T mutation that is prone to aggregation, introduce that into the panel of 250 cell lines and ask is there differential growth inhibition that is seen across the panel of cell lines? And if so, could we use that to identify models that might be used to study what are the mechanisms by which alpha-synuclein induces growth inhibition or cell death, which of course is what occurs in the substantia nigra in patients with Parkinson's disease? And so here's just a preliminary result looking at the effect at five days or eight days following transduction of alpha-synuclein across the first 250 cell lines. And what you see is very reproducible results and a great diversity of the growth inhibitory effect where the majority of the cell lines don't care actually whether you overexpress alpha-synuclein but then there are these outliers towards the bottom which reproducibly appear to have a growth inhibitory effect of alpha-synuclein expression. And then we could look back at the debt map data and ask what are the predictive biomarkers of having a growth inhibitory response to alpha-synuclein and I'll just point to one encouraging initial result which is an outlier is expression of endogenous alpha-synuclein itself, which I think tends to say that yes, these are cancer cell lines, they're not primary dopaminergic neurons from the substantia nigra, but there's something about the biology that we're capturing here. And so as a way to do for example genome-wide, initial genome-wide modifier screens that would then take a limited number of hypotheses and put them into more physiologically relevant dopaminergic neuron systems that that's a promising approach to take. That's all I have to say about that. So just a couple of final words about precision medicine beyond cancer. It's just early days, which is kind of exciting to think about where our field is, where the initial bottleneck was, oh, how could you technically accomplish sequencing of a genome or then many genomes or then it was technically feasible, but it was too expensive. Now it's neither of those things and the challenge is now like, how do we make good use of those data? And I think maybe people here will disagree, but it seems inevitable to me that whole genome sequencing, it's already at about 300 bucks, it's continuing to fall. Whole genome sequencing for everybody is going to be in our future. Maybe in the newborn setting, maybe in another setting, that's the world we're gonna live in and we need to figure out how to make the most of that information and how to do this responsively and to make sure we're learning from doing all that sequencing. Because what a disaster it would be if somehow routine whole genome sequencing as part of the medical record became a thing and we weren't maximally learning from that experience and so the Broad is committed to trying to help make sure we're learning maximally from that experience. And so I'll just say here that it seems likely that genetics in the clinic, whole genome sequencing or whatever, I think that will be what it is as opposed to panels of looking for mutant genes or whole exomes, will be a thing and it will mostly be a thing that is driven by commercial entities, by companies, that's good. They'll do a good job at this, but we're not poised just as I said on the cancer genome side to be learning from those data and I think we have a collective responsibility to make sure that we do. Everyone including the companies will benefit from sharing those data with the research community because the more we learn about the benefit of that sequencing, the more valuable their products will become, but we're not on that path now and I think that's something we need to figure out. I think there's an important role of the federal government probably in there somewhere to incentivize the sharing of that information and perhaps even be paying for the sharing of that information until and unless we've learned what there is to be learned from large scale information and we need to be thinking about rather than this sort of firewall between there's research and there's clinical practice getting more to the point where we're learning from routine care of patients because you think about how much information is generated in the clinic on a daily basis, it's enormous and when you compare that to what we learn from the clinical experience, that's tiny and we need to fix that. So I think my last slide on delivering on the promise of the genome, I think the genome has delivered. It's pretty spectacular. I gave one tiny early example of this fusion that piqued my interest in the genome, but we now have a mountain of genetic variants strongly associated with disease risk. That's no longer our problem discovering whether those things exist and whether they can be clues to the biological basis of disease. They exist and they're giving strong clues as to that basis. But that just exposes the next challenge that we need to deal with. And that I think you can sum up by saying we need to figure out how to target in a pharmacologic sense, whatever the genome is delivering. And I think there are two components to that. One is we need to figure out how to go from this mountain of variants to a molehill of therapeutic hypotheses, a small number of hypotheses. If there are a thousand genetic variants associated with a particular disease, that's not going to be a thousand mechanisms. It's gonna be a limited number of mechanisms that are consolidated from those large number of genetic variants. And I wanna make the distinction here from a variant to function kind of effort. That's important too to understand what is the biochemical function or the gene regulatory function of a given variant. Here I'm really focusing on what is the biological story that a collection of variants is telling that could be exploited therapeutically. I don't think we know how to do that now. We need to take the same kind of systematic functional genomic approaches and figure that out and get that to the point that it's as regularized as sequencing genomes in the first place. And then having identified whatever those key therapeutic programs are or therapeutic opportunities are, we need to get better at figuring out how to make therapeutics that are directed against whatever those data are telling us we need to target. And we can't actually believe that there's such thing as an undruggable target. Whether that's gonna mean figuring out clever strategies with small molecules that we hadn't figured out or it's gonna be programmable nucleic acid-based therapeutics. We need to figure out how to be able to be systematic with therapeutic discovery just like we're systematic with genetic and genomic discovery. So I'd like to conclude by just acknowledging the people that have contributed to these projects. Their names are on the screen and I've tried to mention them as I go along. Thank you all very much for your attention. And I think we have a few minutes for questions. Dr. Ortimi, can I start? No, Charles, is it okay if I ask? Okay, I don't wanna step on your toes and not have you call on me. Oh, okay, Todd, that was terrific. First of all, thanks for coming. Thanks for spending the whole day with us. I know you've met with various people. I think you had lunch with trainees and so forth. So it's wonderful to have you head of the Brode here visiting for all day. I have to admit I had no idea you were gonna be here first thing this morning so I had no idea that you were gonna see me dressed the way I was. Which leads me to my first of two questions. What is the most embarrassing costume ever worn by the head of the Brode Institute? Now, this could either be you or your predecessor, but what's the most embarrassing costume? I think it's yet to be worn. And you told me, Eric, that it was. If you wanna, for half price, I will sail you the entire Kengarb and you could impress the staff tomorrow when you go back. Excellent, I'll take you up on that. Okay, there you go. All right, serious question though, because it relates to both the previous session with the patients and also a question that came up and then you even touched on it. One of the questions that I think a lot of folks ask when they hear a presentation that we heard earlier or hear cancer genomicists talk about liquid biopsies why is it not routine? Why is it not everybody getting blah, blah, blah? And I think the more we see it in a non-invasive prenatal setting where it's clear this can be done for prenatal screening, why is it not done for cancer? So you gave one argument, which I know is legit, about sensitivity issues. But in your position, in your expertise, do you think sort of looking ahead five, 10 years? I mean, you sort of are very confident about we're all gonna get our genome sequence, if not at birth at some time. Do you think liquid biopsies screening for cancer is gonna become routine and if so, you sort of see it in the five or 10 year window or longer? Yeah, so I don't know is the honest answer. I think that, so I alluded to the technical challenge of sensitivity, but let's just imagine that somehow the sensitivity challenge is addressed through some technical solution. It's still not obvious to me that actually screening healthy populations for cancer is the right thing to do. I suspect that if you made a super sensitive test, you'd find a lot of cancer out there, much of which may not actually contribute to the demise of individuals. So I think at a population level, particularly patients that aren't otherwise at risk, either genetically or because of exposures, I'm not sure that kind of blood biopsy-based screening for cancer is what we should be doing. I don't know, studies still need to be done and we have this technical challenge of they're now pretty good at detecting advanced cancers, but not so good at earlier cancers. So it's a bit moot, but I'm still a little skeptical. Francis. Todd, that was a great talk and thank you for being here for the Trent lecture. I wanna ask a bit of a technical question about prism. You had to barcode all of those cell lines in order to be able to determine what happens in terms of their survival after some kind of intervention, a treatment, a drug. Steve McCarroll has this alternative of a cell village where you don't have to barcode your cell lines, you do have to know their genotypes and then basically at the end of your treatment, you simply look to see with very low pass sequencing who's still there, because you can deduce that with appropriate informatic approach. If one was trying to do this experiment and didn't wanna bother with the barcodes, it would seem like that was an alternative strategy, although maybe it's a little more complicated and a little more expensive. Just do you have any comment about the difference between a prism approach and the cell village approach? Yeah, so we've shown that actually you can do the detection just with endogenous genetic variations. You don't need the barcodes. Having a short barcode that you can PCR amplify and do a trivial amount of sequencing, the cost does matter, particularly if you're gonna sequence across thousands of small molecules in many doses and so on. And their copy number variation in cancer cell lines can make it a little more complicated. But it may be that if sequencing costs continue to drop, you may not need the barcodes at all. And it does introduce an additional step and maybe some selection when you're putting them through drug selection to insert the barcode. So yeah, it's possible. Got it. You gave us a very, very wonderful presentation and telling us, do you know what makes us happy in us to understand the root causes of disease? My question is, yes, but not everybody that carried these alterations will develop cancer. So how do we bring the other aspect of the things that determine human accessibility to disease within this framework of genomics? And my second question is, you said the ferragome may have a role to play in incentivizing businesses. How do we do that? Do you have some examples that could be useful? Yeah. So first, with respect to cancer development risk, most of what I focused on was focused on the somatic genome, not on the germline, because it's certainly the case that not everyone with germline risk variants will develop cancer. Those are interesting populations. However, for example, if one were to do noninvasive blood biopsy type monitoring for patients who are known to have genetic risk, that's an interesting patient population to focus on and to study environmental exposure in that context, or people that have an environmental risk, like smokers, as an example. Then with respect to incentives, this is outside of my area of expertise to be sure, but I'd like to see, for example, what if a condition of reimbursement to companies for their tests, obviously you can't require that patients share their data, but you could require of companies that they ask patients if they wish to share their data, and then patients should decide whether they wish to share. I've spoken to many companies as a second point that are saying we don't object to sharing data and making it freely all of this data if there's informed consent, with the community, but it doesn't make sense for it to be our financial obligation to do that. That's not the business we're in. So if there was a mechanism for federal dollars to support the ingestion of those data from the commercial world, where there's informed consent, and to store it and make it available to the research community, that would create the feasibility of data sharing, where there's not like a philosophical objection to it today. Barry? Hi, thanks for coming, an excellent talk. You made some very important points toward the end of your talk about, as you just did now, sharing the data and having an institutional and genomic learning health system. I'm wondering, we're seeing a lot of the barriers to doing this actually within institutions and concerns about HIPAA and other things, and even greater barriers about sharing across institutions. Do you have any experience with trying to do this in one of a man's greatest hospital and that in your system, and how have you overcome those barriers? Yeah, it remains a challenge, but it's getting better. I think as the scale increases, I think institutions are seeing that actually the number of patients that they have in their private biobanks is tiny compared to all of us or the UK Biobank, for example. And so the notion that by hoarding your samples and data, you're gonna somehow have a competitive advantage. I think that's going away. There are still legal concerns around ensuring that appropriate consent to share and so on. Is there technical challenges of how do you make the data actually freely available? But I think we're moving out of the era of institutions wanting to hoard the data because they see it as a competitive advantage. Having the patients is a competitive advantage, but the data I think shouldn't be monetized, shouldn't be hoarded. And I think the availability of NLP methods and other types of methods making it possible to extract value out of clinical data in whatever form it happens to reside. I think that's making that better as well. And then lastly, I think there was some hiding behind HIPAA as if HIPAA was there in a way to prevent hospitals from sharing data when in fact it was there to guarantee the ability of patients to be able to take their data with them. I think that's going away too. So I think that problem's gonna get better even if it's not solved today. Great, thanks. So we take one more question. Thank you for your talk. It was very interesting and very educational for me. I was just wondering if you envision integration of AI or machine learning in cancer genomics and do you think this will accelerate treatment in general? I mean, it's hard to think of areas, new areas of science that are more exciting than the intersection of AI and biology. I think it's one of the new areas of focus at the Broad. I will say that I think a big challenge there is understanding like what are the right questions really to be asking of these powerful methods because I think most biologists, whether they're cancer biologists or whatever biologists, don't understand enough about these methods to really use them in sensible ways or even to invent, make them better. And most computer scientists that aren't trained in biology don't really know what the right most exciting and important and tractable questions are. So there's a bit of a language barrier at the moment. And so I think where we need to focus on is those disciplines coming together and jointly defining, okay, in 2023, what are the questions that are just out of range today but not just fantasy thinking and what are the data that we need to answer those questions? And if we don't own the data today, let's generate them. I think that's the path that we need to go on. I think either the biologists using off-the-shelf methods or computer scientists expecting that the biology will come automatically. I don't think either one alone is gonna get there. Okay, thanks everyone very much.