 So, thank you for the opportunity to talk today about uterine carcinoma sarcoma. This is the integrated molecular characterization of uterine carcinoma sarcoma or UCS for short. And it involves the work of many people in the TCGA research network. Myself and Doug Levine are co-chairs of this UCS working group. So here's a little bit of background. UCS contains an admixture of carcinoma and sarcoma. So there's a carcinomatous component and a sarcomatous component. It is a rare and aggressive tumor that is found in less than 5% of all uterine cancers with a five-year survival rate of about 35%. The median survival is only 24 months compared to 60 months median survival for endometrial cancer. So you can see the large difference between the two tumor types. The median age of the patient is 65 years, mostly occurring in postmenopausal women. And the symptoms include vaginal bleeding, abdominal pain, and polyps. So TCGA collected 57 samples of this rare tumor type for the UCS project. The carcinoma sarcoma has two histological classifications. So if the sarcomatous component originates from tissues that are normally present in the uterus, then the classification is homologous. And if the tissues are not from the uterus, then the classification is heterologous, commonly rhabdomyosarcomas. Heterologous tumors, they have worse prognosis than homologous tumors. So we started off this analysis by doing the usual work that we do in TCGA, which is clustering. But when we tried to cluster these 57 samples across many different data types, we found that they were extremely hard to cluster. In fact, in all of the disease working groups that I'm involved in, I found UCS to be one of the hardest cluster. We did not find any robust clusters on any of the platforms. Here I've given an example of the copy number variation cluster right there, and you can see that there's some pattern there, but not a whole lot. So bottom line is that we thought, okay, we cannot do the regular clustering and subtype analyses that we have done for so many TCGA disease working groups. We need a new approach. We need a new kind of analysis, which I'll get to in a moment. But first, here's the landscape of DNA aberrations that we found in UCS. So there are quite a few copy number variations, which you can see right here. And then there are mutations of many different genes. The top mutated gene is P53, followed by FBXW7, PIC3CA, and then so on. So P53 is mutated at a rate of 91% out of all the samples, whereas FBXW7 and PIC3CA, they are mutated at 39% and 35% respectively. So now that we got these aberrations, the next question that we asked was, which of the aberrations that we have identified can have potential therapeutic implications? So for that, we looked at the drugs that are currently on the market and drugs that are in clinical trials and said, okay, which of the genes that we have found aberrations in, they fall into the category of targets of those drugs. And this is the map that came out. So in each of the columns, you can see the samples. And then each of the rows, you have a single gene. And then you can see, depending on the coloring of the cell, gray means the gene is mutated in that sample, red means it's amplified, blue is homozygous deletion, and green means fusion. So for instance, you can see in this sample, PIC3CA is mutated. So on this side, we have grouped these genes by their pathways and then drugs that act on them. So sensitivity to PI3 kinase inhibitors, mTOR inhibitors, CDK inhibitors, anti-heratotherapy, and so on. Now, just because, this is a limitation of the approach, that just because a gene is targeted by a drug doesn't necessarily mean that we can use it for therapy. But this is the first step that these are the genes that can potentially be used for therapeutic opportunity. The next thing we tried was we compared UCS with endometrial and ovarian cancers. So on the left-hand side, you can see endometrial cancers and the four subtypes that were identified by TCGA. And then you have ovarian cancer, and then on the right-hand side, you have UCS. And the interesting thing that comes out is the following. We have a, well, most of the UCS tumors, they look like serous ovarian, serous endometrial, and also ovarian cancers. So you can see their molecular profiles, copy number aberrations, DNA methylation, expression, and microRNA. It resembles those of serous endometrial. And then to a great extent, the serous ovarian as well. Furthermore, you have mutations in P53 that are observed in UCS, in ovarian, and serous endometrial. On the other hand, you've got a few samples, not a whole lot, but a few samples in UCS that have P10 mutations similar to the endometrioids in endometrial cancer. And they are similar in profile to the endometrioids in endometrial cancer. So moving on, we looked at EMT scores, epithelial to mesenchymal transition scores. So these are scores that are based on a study by Lauren Byers, and she published a signature for EMT genes. And then using that signature, we computed, for each sample, we computed the EMT score across multiple disease types. So what you see before you is a box plot where each dot represents a single sample, and it's EMT score. Anything that's below this red line is more epithelial-like. Anything above the red line is more mesenchymal-like. So instantly, we can see that UCS, which is on the right-hand side, is an extremely large range of values. In fact, the largest among all of the diseases that we have seen in this study. So we can see that it pretty much spans all the way from the bottom to the top. Whereas we can see, for instance, that AML and ACC and GBM and LGG, those are more mesenchymal-like as expected. On the other hand, we can see that bladder, breast cancer, colorectal, et cetera, they're more epithelial-like as expected. So this was an interesting finding that UCS spans the whole range of EMT values. So we said, okay, let's go ahead and study this phenomenon even further. So what you see before you is, in each column, we have a single sample. So all the samples are organized from left to right in ascending order of the EMT score. So the samples over here are more epithelial-like. The samples here are more mesenchymal-like. When we order them this way and then throw in some of the DNA methylation, microRNA, mRNA, protein, and mutation data, some interesting results come out. What we see is that these are the promoters for the microRNA-200 family. And I'm sure you know that microRNA-200 family is well known to play a part in EMT. Here are the microRNA-200 family members. And you can see that those samples with high mesenchymal scores, they have hypermethylation of these promoters for the mid-200 family. And a corresponding decrease in expression level of the mid-200 family as well. Interestingly enough, these are some of the targets of mid-200 as well as other markers of EMT. And you can see things like Zep1 and Zep2 and others, they are higher on the mesenchymal side of the spectrum rather than the epithelial side. So we can track these differences going all the way to mRNA and then even to protein, where we see e-caterin and cloudin-7, they are epithelial markers, and they are depleted in these samples that have low, that have high EMT scores. On the other hand, they are elevated in those samples that have low EMT score. So we can see a nice story developing here which basically says that DNA methylation of mid-200 family members causes the expression of mid-200 family to go down, which in turn causes the expression of mRNAs, those genes, that are regulating EMT to go up. And then we have a corresponding decrease in epithelial markers in RPPA. Interestingly enough, we didn't find a whole lot of association with mutations. And what you can see on the last two rows is that those patients, those samples with more epithelial scores, they have a higher percentage of carcinomas rather than sarcomas. And then on the other side of the spectrum, those patients with high mesenchymal scores have a higher sarcomatous component than carcinoma, which is to be expected and it is a sanity check. The next step we looked at was, well, okay, so we know that UCS is a uterine cancer. We also know that it is a sarcoma, it has a sarcomatous component. So the question we asked was, is UCS more similar to gynecological cancers, or is it more similar to the non-epithelial cancers? So what you see here, let me explain to you a little bit. Without looking at UCS, what we did, we looked at the group of gynecological cancers, endometrial and ovarian. We looked at a group of non-epithelial cancers, namely low-grade glioma, GBM, adrenocortical carcinoma, and kidney papillary. So we threw them all in one group, threw the gynecological cancers in the other group, looked at the proteins that were differentially expressed between those two groups. Then after getting that list of proteins, we got about 40 proteins. After getting that list of proteins, we threw in the UCS and said, where does UCS fall? And here's a zoomed in view of UCS. And what you can see here is that UCS shares proteomic characteristics with both the gynecological cancers as well as the non-epithelial cancers. Interestingly enough, the DNA damage response members, so these are proteins that are involved in DNA damage response. They are similar to endometrial and ovarian. Remember that I showed you that the copy number profiles of endometrial and ovarian cancer in UCS, they look very similar to each other. So the fact that we picked out DNA damage response genes that are also proteins that are also common between the gynecological cancers is quite interesting. The other thing that is common with gynecological cancers is cell signaling. So we see a lot of PKC and a lot of phosphoproteins that are here, which show that UCS shares cell signaling profiles with endometrial and ovarian cancer. On the other hand, we have EMT proteins, and the EMT profile looks very similar to the non-epithelial cancers that are right here. So we can see that certain pathways and certain molecular features are shared with gynecological cancers, whereas others are shared with non-epithelial cancers. The next point I want to talk about is that there are three computing theories of development of UCS. The first theory is called the collision theory, which says that UCS occurs because of a collision of adjacent and independent carcinomas and sarcomas. They come together to form this disease. The second competing theory says that UCS is a combination of cellular masses that diverged early from a common stem cell precursor. The third theory is called the conversion theory, and it says that basically UCS has a monoclonal origin with late divergence of the carcinoma into sarcoma. So essentially, the difference between the second and third one is timing, once and of course the cell of origin. So in the second one, combination theory, the cell of origin is more stem-like and it diverges early. In the third theory, conversion theory, the cell diverges late from the carcinoma into the sarcoma. So the question that we asked was which of these theories is supported by evidence from TCGA? Now bear in mind that collision theory has fallen off as being popular, so it is no longer that popular. Combination theory and conversion theory are still around, but more and more evidence is gathering in the literature for the conversion theory, the last of those. So what we found in our results is as follows. Here we've got pan-cancer analysis that was done. UCS is on the right-hand side, and then you've got about a dozen tumors going this way. And we've got the blue dots here. Well, first of all, each column is a sample. The blue dots here represent the total mutation rate in that sample. The red dots represent the clonal mutation rate, and then you've got the significantly mutated genes that are mutations in those genes, followed by the clonal rate of mutation in the significantly mutated genes. So it takes a little while to grasp this figure, but what I want to get at is that we found that 73% of all mutations and 82% of the mutations that are in significantly mutated genes are clonal in UCS. This compares very favorably to other tumor types. So what we saw was that UCS was no more heterogeneous than any other average tumor type. Therefore, we think that the conversion theory, that the evidence supports the conversion theory because the logic is that if we had the collision theory that was correct, then we'd had two separate components coming together, then we wouldn't see such high degree of clonality in the mutations. Also, if there was early divergence, we would expect a separation of the two components. And again, if there was early divergence, we would see more clonality compared to the other tumors. But we don't see that, so we think that the conversion theory, it's the data suggests that the conversion theory might be the right one. So to summarize, UCS is a rare and aggressive tumor that is found in less than 5% of uterine cancers with a median survival of 24 months compared to 60 months for endometrial cancer. It is a very heterogeneous disease with poor clustering, some of the worst clustering that I've seen in TCGA, so we can't really classify them into subtypes that are robust, which is unfortunate because the number of samples that we get is very small to begin with, and then you can't even classify them into subtypes, so we're stuck there as far as subtypes are concerned. Nevertheless, there are similarities among the samples, namely that we see extensive copy number variations and recurrent mutations in TP53, FBXW7, PIC3CA, PPP2R1A, and P10, ranging all the way from 91% down to 19% and even lower. We have identified DNA aberrations with potential therapeutic relevance, and we saw that most UCS samples resemble serous endometrial and ovarian cancer samples with P53 mutations and high CNVs, but a few samples do resemble endometrioids with P10 mutations and low CNVs. UCS spans a large range of EMT scores compared to the other tumors, in fact, the largest range of EMT scores that we have seen in our pancancer study. We saw that the promoters of microRNA 200 family are methylated in samples with high EMT scores with correspondingly low microRNA 200 expression. We also saw that UCS shares DNA damage response and cell signaling proteomic features with gynecological tumors, but EMT features with non-epithelial tumors, therefore it shares features with both groups of tumors. We also saw that most of the mutations are clonal, supporting the conversion theory of UCS tumor development. I would like to acknowledge many people, the TCGA Research Network, but especially my co-chair, Doug Levine, and the UCS Writing Committee, Andy Churniak, Chip Stewart, Brad Murray, Rianne Bolby, Juan Valter, Heway Shen, and Julia Zhang from the program office for doing an excellent job of coordinating this whole thing. And then the review committee, the manuscript is currently under review by Raju Kuchola-Party and John and Weinstein, Gordon Mills, and of course a pathologist, Rosemary Zuna, Russell Brodas, Rob Soslow. And a special thanks to JC for all the work that he's done in TCGA and for helping us along and bringing TCGA to almost a completion. Thank you. Very interesting, Rianne. Thank you, John. I was interested in the slide on the clonality and mutation rates. It looked good, yeah. Despite the heterogeneity in other senses of the disease, it looked as though there was very little heterogeneity in the mutation rates, and the only other one that was in the same class in that sense was the chromophobes. Any regularity there or any reason that you can think of for it? So chromophob, frankly speaking, came as a surprise to us as well. We wouldn't have predicted that before doing the analysis. And it turns out that chromophobes just are very heterogeneous. We have not studied that in this group, in the UCS group, but I'm sure the chromophob group will find it very interesting. I don't know if Chad Creighton is here and if he's got any comments on that. But yes, that did come as a surprise to us.