 So it's a pleasure to be here to present some highlights from the thyroid project. I'm going to start with a very simple model of thyroid cancer. Let's see. So we start with the normal follicular cell, and we progress to well differentiated tumors, either papillary or follicular tumors, follicular adenoma and follicular carcinoma. Most tumors stop here, but rarely they evolve into poorly differentiated or anoplastic carcinoma. And the point I want to make here is that there's a progressive loss of differentiation. Differentiation is very important for thyroid cancer, and it's actually the foundation of our classification. We talk about differentiated carcinoma and undifferentiated or anoplastic carcinoma with poorly differentiated in the middle. So it's really a foundational thing for thyroid cancer. So when TCJ started this, with 85% of the cases being papillary, it really became clear that we needed to focus on papillary. So our project is restricted to papillary thyroid carcinoma. So there's three main types of papillary carcinoma. There is a classical type that gets its name because it has these well-developed papillary structures. There's actually a follicular variant that is recapitulating normal thyroid architecture. And then there's a tall cell variant that has abundant cytoplasm and tall cells. And those are the three main types. There's other types, but we wouldn't get enough to really power an analysis of the rare type. So we kept it simple. And there's a very strong genotype-phenotype correlation as shown here. So before the TCJ, this was sort of the view of what genes were known to be mutated in papillary carcinoma, mostly in BRAF and RAS, but also rearrangements of RET and NTRK1 with infrequent PI3 kinase mutations. So that's sort of the foundation from which we started. So this is our cohort. We had up to 496 tumors with 391 on all platforms. And we had 49 whole genomes that we targeted at tumors that didn't have obvious driver mutations. So I really like this slide. I think it explains a lot. It shows that amongst all these tumor types, that thyroid carcinoma, papillary thyroid carcinoma, has relatively low mutation density, and actually the lowest, if you restrict your view of this, to solid tumors like especially carcinomas. So the next one that starts in is prostate here in breast. And we think this is partly the reason why it's such an indolent carcinoma. Whoops. A little jumpy. So this is our overview of our somatic alterations, and I'll go through just a few points here. So we have mutation rate and clinical information and significantly mutated genes. The first thing to notice is there's not much up there, right? There's a lot of white spaces here. So it's a relatively quiet genome. We can see that BRAF alterations are common in about 60% of cases, and then we have mutually exclusive RAS mutations. We have a new mutation here, EIF1AX, which I'll talk a little bit more about, which is mutually exclusive with BRAF and BRAF. And then we also discovered BPM1D and CHEC2 as significantly mutated. And then we jump to the fusions, and you can see we have a diverse collection of fusions that make up about 15% of the cohort, some expected RET, but also diverse fusions of BRAF. So that's one of our interesting findings that we have, these BRAF fusions. And you can see that these are mutually exclusive with each other and also with the point mutations. So that's a very nice story. And then we jump down to the copy number changes, and then many, many of the BRAF mutated tumors don't have a lot going on in terms of copy number change. And then you can see very interestingly we have a concentration of arm level copy number changes that start up when the point mutations and the fusions go away. So we're speculating that there are drivers embedded in here. We obviously can't pinpoint the genes, but we think this is an interesting result and very provocative. And so if you allow us to count those as potential drivers, we end up with only about 14 cases without a driving event out of 400. And this is relevant because, you know, in using existing genes, if you genotype 100 cancers, you'd only find a driver in about 75%. So we've expanded the universe of driving events, and that will have profound influences on molecular diagnostics. So this is EIF-1AX, and here's our mutations, and they're sort of landing in the same region as those reported in the Cosmic Database, and in this paper here on UVO melanoma. So we haven't functionally proven this. Jim Fagan's actually working on this at Memorial, but we haven't proven this as a driver, but we think it will turn out to be a real event. So here's our fusions. We have some new red partners. We have, and we think they're real because they retain the kinase domain and express the kinase domain. We have diverse BRAF fusions. We've discovered, and a few other people have discovered, that ALK fusions are present in about 1% of the cohort, and then ETV6, NTRK3, which also came up in a screen of radiation-induced papillary carcinomas recently. So we think this is a big part of our story, these new fusions. Oops, it's a little jumpy. So if you look now at the 14 cases that are officially dark, they're actually not that dark. We do have some interesting mutations, ATM, APC. We have some hits on some potential fusions. So if you allow us to count maybe some of these as potential drivers, we're really down to about five cases that don't have any explained cancer mutation. So there was a report that suggested BRAF was subclonal. So we specifically looked at the five drivers and using absolute to make a statement about their clonality. And we're making a strong statement in the paper that, you know, these drivers are in fact clonal, which has implications for targeted therapy. So what were some of the challenges that we faced when we did this project? One, it's restricted to papillary carcinoma, which is a very indolent, well-differentiated tumor that's cured 95% of patients. And if you're going to study the outcome of this disease, you really need 20-year data, and our cases don't have anything close to that. So that was a challenge. And then we have this relatively low mutation density. So that presented us really with two choices. We could sort of write up the biomarker paper that, you know, had some new point mutations, talk about the fusions, do the clustering and call it a day, or we could push. And Gatti and I may be synergized and we pushed. So we focused on the fact that these drivers were mutually exclusive, that we had a relatively quiet background genome, in which we could explore the role of these drivers. We had all this multi-dimensional data, and we had a very, you know, great and imaginative AWG. So the first thing we did is we said, well, let's explore this difference between BRAF and RAS. And we developed a signature, a 71 gene expression signature, that separated these tumors out, and then we converted that into a score, a BRAF v600D RAS score that we call BRS. And we scaled that from minus one to one, and then displayed the tumors. And what you can see here is that displays the tumors along this gradient, and it's not black and white. There's a transition that goes on here in the middle. And then we can use this score to explore how the other mutations, you know, where they fall on this gradient. And what's really fascinating, you can see that there's some BRAF mutations that are not v600D that are actually RAS-like. And this one's actually the BRAF K601E mutation, and that's consistent with the literature, because those are thought to be, you know, the follicular variant, which tend to be more RAS-like tumors. And then you can see some of the other fusions like the Pax-8 P-pargamma or weakly RAS-like, and that makes sense. So this was a nice way to explore how these other mutations might signal. And then if you bring in all the other data into this figure, you can really see that the biology of these RAS-like tumors is very different across all platforms from the BRAF-like tumors. And that's one of our overarching conclusions that these are fundamentally very different tumors. Then we turn to thyroid differentiation, and this was known that when you pick up a BRAF mutation that you have a loss of differentiation and particularly silencing of the iodine metabolism machinery. And so we wanted to explore this in our cohort. And so, oops. So this is a complicated slide, but basically these genes are those responsible for the iodine metabolism machinery. And we displayed them in context with genotype. Here's a BRAF V600E. Here's RAS. Here are the fusions. And you can immediately see that tumors on this side, which are the follicular variant tumors, are more differentiated than the BRAF tumors, that's not a shock. But what's really interesting is within the BRAF V600E cohort, we see a range of differentiation. And say, well, why is that, why do we care? We care because there's probably at least 100 papers that have looked at BRAF as a biomarker in isolation. In other words, the field is treating BRAF V600E papilla carcinoma as a homogeneous tumor group. And our data suggests that that's maybe not appropriate. So then we got interested in, well, what are the drivers of that? What are the potential drivers? What's correlated to that? And we found out some interesting genes, like trefoil factor 3 and 2-oncomeres, mirror 21 and 146, and a potential tumor suppressor mirror were correlated to the TDS score that we derived from these genes. And so we have some interesting possibilities and keep those in mind because they'll resurface later. So we used different kinds of data. We used the messenger RNA and the RPPA data, the group at Memorial spent a lot of time in figuring out the signaling consequences of these two drivers. The take-home point is that the BRAF-like tumors signal pretty much exclusively through MAP kinase and that the RAS tumors are, have a much more complicated, a little bit more PI3 kinase, but also some MAP kinase. So we explore this and really show that there's fundamental differences here. And then onto the clustering. We, I don't really have time to go through each platform, so I'll show the supercluster. Basically, all the cluster, all platforms, like I showed in the earlier figure, show that there's a striking difference between the RAS-driven tumors and the BRAF-V600-like-driven tumors. And there's histologic differences, et cetera. So that's not shocking. Again, it confirms our big conclusion. What's interesting though is that there's a cluster here that is very robust and it matches up between the different platforms, methylation, messenger RNA, microRNA, and that it's maybe hard to see, but these are enriched for the tall-cell tumors. So we think that there's a distinct cluster of tall-cell tumors and distinct expression profiling. And so that becomes more important in the context of this, where we really focused on the mirrors as part of the story. And I'll spend a little time going through this. So basically, mirror cluster one are RAS-like tumors, and then we have five classes of the BRAF-like tumors. And you can start to see that there's some interesting molecules that are preferentially expressed in some of these mirror classes. In this one, mirror 21, known alchemyra, is here. And why do we think that's relevant? Well, we actually took the scores that we developed in the middle part of the paper, the BRAF-RAS score and the thyroid differentiation score, and used them throughout the clustering section to make it more rich and informative. And you can see here this cluster has three-fourths of the tall-cell tumors. It's in a BRAF background. There's not a lot of other mutations going on. It has a higher risk. It has, I graded all the tumors, and it has a higher grade. It has clearly different messenger RNA profiles. But most importantly, it's the most BRAF-like, because it has the lowest BRAF scores, and it's the least differentiated. It has the lowest thyroid differentiation scores. And so why do we care about this? You know, because I admitted that the tall-cell is, you know, a recognized variant. Well, we care about it, because as come through in all pathology talks, we tend to disagree. And so what I call a tall-cell, another person may not call a tall-cell. So if we can actually uncover a molecular marker of the tall-cell, that would be adapted widely by the thyroid community. So we're very excited about this. And then there's a similar story here for cluster five and mirror 146, which I remind you were the mirrors uncovered back when we were looking for correlates of the TDS. So we think the different parts of the paper fit together nicely and support each other. And we really worked hard to tell this integrated clustering story. So our overarching conclusions were that RAS-driven PTCs and BRAF V600 PTCs are basically fundamentally different. And so it begs the question, do we reclassify thyroid cancer to sort of separate them? And there's some data that suggests we should. There was a paper in the New England Journal from Jim Fagan's group where they're looking at a MEK inhibitor that had differential responses for recovering susceptibility to radioactive iodine. And it depended on what your underlying genotype was. So I think the days where you could lump papillary carcinoma and run a clinical trial without knowing what the underlying drivers are and what the phenotype is, is just coming to an end. Likewise, we identified clinically-relevant subgroups of BRAF-driven PTCs, and we have a potential role for mirror. So we think things are going to come from that. So we're actually very excited about how this whole project played out. I know somebody sent me an email like, well, Tom, you know, a year ago, this was kind of looking kind of maybe not dull, but you know, and now it's really turned into this kind of interesting story. So we're very excited about this. And if you don't believe me, then here we are, like Gordon especially. So we do think this will be a landmark study for the thyroid field. And so far, it's already having impact. As I mentioned, Jim Fagan's working on the biology of EIF1AX. In part, catalyzed by this, Yuri Nikiforov is starting a working group of pathologists to explore the follicular variant and can we come up with the proper way to diagnosis? There's a lot of argument and disagreement amongst pathologists. So he's actually looking for support from the NCI to study this. And then we had a collaboration, Hopkins, Mayo, Michigan, and Cornell, looking at BRAF and isolation, like many other people. And then when the mirrors came out, I convinced the group to say, well, let's expand this study. So now we have hundreds of cases. And the whole goal is can we use these mirrors in combination with or without BRAF to predict central compartment lymph node positivity and sort of guide surgery. So I do think our paper will be very impactful. Of course, there's many, many people to thank, but I do need to give a special thanks to Chip, who was the analysis coordinator and really cranked through massive amounts of analyses. And of course, Gaddy was amazing to work with. So I wanted to just take a moment to thank TCJ leadership for giving me the opportunity to work on this project. What's nice about this is that Gary Hammer, who's here from Michigan, and I convinced Kenna to start a project on adrenal cancer. So we use the opportunity of being involved in thyroid to span adrenal cortical project and then the pheo project, which I'll also play a role. So this has really been a very exciting time and a fun time for me. And I would just like to thank everyone and acknowledge the AWG. I'll be glad to take some questions. Hopefully I did it in time. Tom, very nice talk. So the question I have is for those genomically very silent ones, where do they fall? Do they fall to the rest like or do they fall to the B ref? The remaining dark matter tumors? Yeah. They tend to be enriched in the follicular variant. And so the truth is those are the cases that pathologists argue about. Right, because it's really complicated. So basically those are different disease, basically. It happens to many kind of the kidney that we see that. It could be. I mean, so we can talk afterwards. It's a long answer. Yeah, Matthew. Hey, Tom. So I was a lot of incredible stuff in there. But I was really struck by the mutations in the thyroglobulin gene. And I was curious if those are loss of function mutations. Yeah, so we put them in the paper, Matthew, but we're not exactly sure how meaningful they are. They're in there. We went back and forth on that. Maybe chips here. Not sure. That's why we put it in there, but we didn't make a big deal about it. The guys at MD Anderson, Steve Sherman, were excited about that. But the story didn't pan out as strongly as we thought. So not sure how meaningful they are. So it's just a striking feature of different cancer types that I don't understand. There's lots of loss of function mutations of albumin in hepatocyte or gastronoma. There's loss of function mutations of collagen in chondrosarcoma. And this might be part of that same pattern. It might be. I mean, mutic did not pick up on them because it is a very large gene. So I mean, I think there's still more work to do on those individual mutations. So what Matthew was saying is that we look specifically at thyroid receptors and thyroid globulin. And it's sort of a controversial point. There are a lot of mouse models on receptors, but that didn't pan out in our dataset. We had a low incidence, maybe a couple percent of thyroid globulin mutations. We're not really sure how significant they are. They've not picked up on mutic. So, you know, mutic has different versions. I think some of the earlier versions picked up thyroid globulin, but not the later ones. Okay. Great. Thank you very much. Thank you, Tom. That was a wonderful story. So now the next presentation will be by Ali Amin Mansour on somatic alterations in clinically relevant cancer genes among 12 TCGA tumor types. Ali.