Tcga - Integrated genomic characterization of endometrial carcinoma - Douglas levine

Douglas Levine: Thank you. Good morning. It's a pleasure to be here to discuss the findings from the Endometrial Disease Working Group. I always love coming to Washington. Last time I was here, I walked past the Government Accountability Office. Talk about a oxymoron. This time I was walking out of the subway and I walked up the Metro, I saw the VRE Trains. So all the doctors in the audience will know that I took an immediate right turn and went the opposite direction. So endometrial cancer begins here on the lining of the uterus. It invades into the muscle of the uterus, and then will spread to lymph nodes or to the omentum within the abdomen. There's two main types of endometrial cancer. There is low-grade glandular-forming endometrioid cancer, which has a low nuclear-to-cytoplasmic ratio and preserve architecture, and there's a high-grade serous type, which is more solid and is much more atypical nuclei, and these are essentially two different diseases. Now the complicating part is that it's very difficult for a pathologist to differentiate these two types of tumors when they are high-grade. And an article in press here where three different pathologists from three different academic centers tried to look at intra-/inter-observer variability, and, in fact, more than a third of the cases was a major disagreement when trying to classify high-grade endometrial carcinoma. Type I cancers are endometrioid. They have a favorable outcome, and often they're treated with radiation therapy, when appropriate. They Type II serous cancers, which are more aggressive and often metastatic, are treated with chemotherapy. So these two types have different treatment paradigms, which is somewhat schizophrenic among the professional community as far as, you know, who gets which type of treatment. But classification is the first important step to determining what treatment a patient receives. There's known mutations in endometrial cancer. PTEN mutations, P53 mutations, and the ones you see in this slide are previously known and have varying frequencies between Type I and Type II cancers. And the outcomes are different, of course, like any other cancer. Early-stage tumors do well. And for a disease that's spread beyond the uterus, the recurrence at about one year, 50 percent of patients will recur at about one year. But interesting to note is in the advanced recurrent cases, about 25 percent are serous or high-grade, and in the early diagnostic cases, at the time of diagnosis, only about 15 percent of cases are serous or high-grade. So when we started the TCGA project, we didn't want to collect all the tumors that were available because the ones that are commonly available are low-grade, good-prognostic tumors that do not result in death. And so that would not be the best cohort to study. And so we tried to skew the accrual towards the more aggressive cases that would lead to death and try to accrue cases based on what we see in the recurrent disease setting, as shown here on the slide. And we did pretty good. We've collected 373 cases for the data freeze. The average age is about what you'd expect for endometrial cancer. We have 20 percent of patients that have recurred. And here you can see we have about 100 cases of each, including the Grade 3, high-grade endometrioid cases. And we have 50 of the serous cases, and we wanted to make sure we got to 50 cases, and so we reached our accrual goal here before doing the analysis. And as with most endometrial cancers, there's only a few deaths. Many patients will die of other causes. And so we're powering most of our outcomes on recurrence rather than death, which doesn't occur as frequently in endometrial cancer; endometrial cancer being the fourth most common cancer among women, at least 8,000 deaths per year in the United States. To date, we have 400 -- I'm sorry, 248 exome pairs, just over 100 low pass whole genomes. And you can see the other numbers here for the other platforms that have been used in this analysis, including almost 300 samples on the reverse phase protein arrays. So the first thing we'll look at is a copy number. And this is just organized based on histologic subtype, and you can see that as the histology of it becomes more aggressive, the copy number profiles become more complex. Now if you do unsupervised clustering, as Andy Cherniack did, you can find four copy number groups here, and interestingly, the first copy number group is essentially diploid with no somatic copy number alterations. There's two predominantly endometrioid groups here in the middle that have few copy number alterations, and then there's very complex group four here. And if you'll look at the second bar below the figure, the blue here are all the serous subtype cases. But mixed in here are a few other lighter colors, and, in fact, one-quarter of the high-grade endometrioid cases cluster with the serous tumors. And we refer to these as serous-like cases. If you look at progression free survival, as you would expect, the serous group does much worse. But, in fact, copy number 3, which is essentially defined by this 1q broad amplification, and it is endometrioid in histology, also does significantly worse than the other two groups up here. If you look at the focal GISTIC peaks, again, you can see copy number 1 right here. Copy number 2 has very few alterations. Copy number 3, again, cluster 3 has this broad 1q amplicon, and then cluster 4 has extensive copy number alterations like you would see in ovarian serous carcinoma, probably lung squamous, and in a basal-like breast cancer. If you want to move to mutations, just the common mutations, just to sort of lay the landscape here, PTEN mutations are most common in low-grade endometrioid tumors. P53 mutations are obviously common in the high-grade serous and in a portion of the high-grade endometrioids. And just for reference, the PIK3CA mutations are essentially equally distributed across these different histologic subtypes. So now we get more into the exome data. And this part of the figure, I'll take a few minutes to go through it. On the top here, you're seeing mutation rate. And there's a group of samples here that have a very high mutation rate, probably 100 times higher than most solid tumors, on the order of 100 per megabase or more. This group is defined here by microsatellite instability and also has a high mutation rate. And these two groups over here have low mutation rates and are split up specifically based on copy number alterations, which you see in this row over here. This is, again, the serous cases are all over here, as I'll show you later, and the serous-like cases are here. But this is how we initially divided up -- divided up the cohort, again, based on mutation rate, microsatellite instability, and copy number alteration. Now, moving on a little further, if we highlight this subgroup over here, these 17 cases, which have very high mutation rates, are all defined by universal mutations in polymerase E, which plays a role in transcription-coupled repair. And the very interesting thing we discovered in this study is that 75 percent, or 13 of the 17 cases here, actually have one of two hotspot mutations in POLE. And this results in a different mutation spectrum with a significantly greater frequency of transversions rather than transition mutations. And so this was quite exciting when this was discovered. And similar findings have been seen in colorectal; in particular, in the marker paper, they had a small group that had these ultramutator cases, and since that time they've also identified these hotspot mutations in colorectal. And this afternoon, David Wheeler will talk about this. If we take these four mutation spectrum groups and look at their outcome, we see two interesting things. As expected, again, the serous cases and the high-grade frequent copy number cases have a worse outcome. The two middle groups, which includes the microsatellite stable and instable groups, have about the same progression for survival, keeping in mind that the follow-up data in this particular study is somewhat limited compared to other studies in literature and clinical trials. It's controversial in endometrial cancer whether patients who have microsatellite instability do better or worse. In colon cancer, it's almost uniformly agreed that those patients do better. And, again, endometrial cancer, we see here that they do the same. And then this subgroup of POLE mutations, although it's only 17 patients, it may be hard to draw conclusions, but so far, in some of these, despite the access right now being in days, some of these sensorings do get out to three to five years, there's not a single event in the POLE group, so it's possible this is a good prognostic subgroup and further data -- further study will be required, and this is underway by both David Wheeler and collaborators. So now if we add a few other rows to this figure, we can add in just PTEN and P53 mutations. Again, you see all of the P53 mutations are here. Virtually all the PTEN mutations lie in the other groups. And now we add on histology and grade, and, again, these purple bars are serous cases. The dark grey here are the high-grade cases. And so, again, these are all the serous cases, plus about 25 percent of the high-grade endometrioid cases, which still are behaving like serous tumors and called "serous-like." Now, you always find interesting cases here. And this is really where the molecular data can certainly be value-added onto what you otherwise would get without ever doing this. And here, before we get into that, we'll look at some of the significantly mutated genes. There's about 50 SMGs that are different, have differential frequencies between these groups. Some of the more attractive ones are highlighted here. Again, you can see PIK3CA mutations are fairly evenly distributed. Certainly, the ultramutator group has more mutations in every gene. ARD1A is not found in serous cases. KRAS is not found in serous cases. β-catenin is interesting here because there's a very high frequency peak among the cases that have low mutation rate. And, in fact, the hypermutator microsatellite cases do not have a higher frequency, as you'll see in every other gene, suggesting that β-catenin is really playing a role here in the endometrioid low-grade, low-mutation rate cohort, and a few other genes are here. Now, getting to an interesting case here, this is zoomed in on one part of the figure, and you'll see at the bottom, this is a serous case, but, in fact, it does not have a P53 mutation, there's no copy number alterations, and it has a very high mutation rate. This seemed odd, so we went into the patient portal, which you heard about yesterday from the cBio group at MSK. And if we pull up this patient's profile, you can see, in fact, there's a KRAS mutation -- again, not what you'd expect in a serous case. And there's an ARAD1A mutation, which is also less common in the serous cases. So then we went back to the histologic section, which now, gratefully, is fully available for all the TCGA cases. And we had one of our specialty pathologists look at this case, and he said definitively this is an endometrioid case, not a serous case, but, in fact, it has some micropapillae architecture which can be confusing to differentiate this from serous and endometrioid. And he looked at the morphology, and based on some other studies he had done, suggested this case may actually have an MSH6 mutation. I did tell him that it had a high mutation rate and seriac [spelled phonetically] -- seriac, we know seriac [unintelligible] from Wash U -- just last night, went back to look at the traces and there may be a MSH6 insertion in a homopolymer track, so it's not clear if that's a true mutation or not. But nonetheless, this is what would be a misclassified serous case that really only could be identified from doing these types of studies. We've done -- the Vancouver group has done microRNA sequencing and clustering. We found six microRNA subgroups. In addition, methylation was done both in this project on the 27 and 450K arrays. They've developed some approaches to integrate that data. Again, like many other tumor types that's for methylation subgroups, this group here is a hypermethylator group that has a CIMP-like phenotype that's seen in some of the other diseases. And again, here are all the serous cases that have very low levels of DNA promoter methylation. The MD Anderson Wei Zhang and Yuexin Liu have done expression clustering, identifying three gene expression clusters, giving them names of mitotic, hormonal, and immunoresponsive, based on the components and members of these clusters. The hormonal group has increased expression both at the gene and protein level of the hormonal receptors, and looking at the progression free survival, according to gene expression clusters, again, the mitotic group, which contains all the serous and serous-like cases, again, does worse, and the other two groups do about the same. Now -- and this -- and again, you can see that confirmed by all the P53 mutations over here, and the PTEN mutations over here, which really is the hallmark for differentiating these two subtypes. Now, I think for the first time at TCGA, we actually took the RPPA data and used it in a supervised manner to see if we could understand and validate the biology from the gene expression clusters here. And doing this on this heat map with 36 proteins, in fact, you can see that the mitotic cluster has increased expression of DNA and proliferative genes. Again, the hormonal cluster at the protein level, we see the hormone receptors being active here, and in the immunoreactive cluster, we see STAT3 as well as LKB1, again, confirming what we think is proper labeling of this cluster of immunoreactive. And now looking at the full RPPA data, courtesy of Gordon Mills, that was a supervised RPPA analysis. This is the unsupervised RPPA clustering. Here there are five RPPA clusters. I'll give you one second to look over that slide. The first cluster in pink, here, has basically signaling on. The second cluster contains most of the serous cases. The third cluster is basically signaling pathways that are off. The fourth cluster has reactive proteins as well as MAP kinase pathway. And the small fifth cluster, I believe, is a stromal signature here. The paradigm folks did the paradigm analysis, looking at the expression and copy number data. And it also identified five clusters, which you can see at the bottom, but two of them are quite small. The third cluster in the middle contains all the serous and serous-like cases, which has MYC activation. An interestingly here, we see the P53 pathway is suppressed due to the P53 in activating mutations. Cluster 5 has MYC activity and hormonal activity, which is consistent with -- of certainly this disease process. And here's an interest in cluster 1, we see low MYC signaling but very high WNT signaling, which, again, one confirms the high-frequency β-catenin mutations, and I'll also show you some of the specific mutations on the next few slides. This is -- speaking of the next slides -- this is from the cBio group. I think names are on the next slide. This is the RAS β-catenin pathway, which is the most significant module that came up in their mutually exclusive mutation analysis. And what we have here in these three different boxes, throughout the whole figure, we have the hypermutator microsatellite unstable cases. We have the microsatellite stable endometrioid cases and the serous-like cases here. The serous-like cases have amplification of ERBB2, which may or may not be associated with sensitivity to herceptin; a trial previously done by the Gynecological Oncology group was negative using mostly immunohistochemistry. The hypermutator samples have frequent KRAS mutations. And then, again, we see the frequent β-catenin mutations here. Now this is a different mechanism of activation in which KRAS can stabilize β-catenin as opposed to APC-associated degradation, which is seen in colon cancer. And the reason that this comes up is because the β-catenin and the KRAS here are mutually exclusive, that you see in multiple of the subgroups, also identify where SOX17 mutations with two hotspots shown on the bottom part of the figure. Looking back to one of our favorite pathways, PI3K/AKT, this is certainly the disease where we see the most activity in the pathway. Again, the mutations are evenly distributed across the various subtypes. PTEN mutations are more common, of course, in the endometrioids. And here we see mutual exclusivity between PIK3R1 and PIK3CA. This is not a new finding. It was reported about two years ago by Gordon Mills and has also been reported by others. But what's interesting here is if you look at the PIK3CA mutations, we see the common exon 9 and exon 20 hotspots, but we also see frequent mutations here in exon 2, which is a P85 binding domain. And then if we look at P85 or PIK3R1, we see many mutations with one hotspot here in the SH2 domain, which is the domain that binds over here to the P85 domain of PIKC3A. And so, in fact, these mutations here are likely functionally related to interactions over here with these mutations. And for that reason, you see almost perfect exclusivity here between the two genes. And I would imagine these cases are just due to the hypermutation status of these samples, and, of course, a few of these samples in this group do have a high mutation rates but do not have microsatellite instability. Another interesting finding is that there's a very high frequency in PTEN of the codon130 hotspot mutations. More than a third of the endometrial cases and in other diseases that have PTEN mutations, this particular hotspot has mutated at a much lower rate. Looking at PIK3CA, again, we went over this just in the past couple of slides. But again, this is also different than you're seeing in other diseases, particularly breast cancer, which has a lot of PIK3CA mutations and has targeted trials going on, but the spectrum is very different than we see here in endometrial cancer, which has clear implications. SuperClusters were done, which you've seen before. Again, this is sort of summarizing all the various platforms. There were four different clusters identified. Again, the cluster that contained all the serous cases does worse; a common theme, of course. And the rest of the clusters show no difference in outcome, based on this analysis. Finally, the question is whether uterine serous cases share similarities with ovarian serous and basal like breast. In the breast cancer paper, they showed similarities with ovarian serous as far as cyclin E amplifications and MYC amplification, and BRCA mutations, as well as correlation of expression profiles between the ovarian serous and the basal-like breast subgroups. So we ask the question, well, does uterine serous also share the features, because histologically and clinically, they share many features. And so here is a multiplatform approach, looking at copy number. Again, expression correlations between the uterine serous, ovarian serous, and basal group here. And, again, now the methylation plots are here with the uterine serous, ovary serous and basal-like breast, looking very similar. And from the paradigm group, you can see the consensus clustering here with the uterine serous, basal-like breast and all of the ovarian cases, based on the silhouette what's here. Unfortunately -- not unfortunately, but when we look at the mutations, we do find some different -- the differences here. So, they're very similar, looking at multiple platforms. But when you do add in the various mutations, you see quite a few differences. Certainly, ovarian serous and basal-like breast have generally a lower mutation rate, particularly, PIK3CA, ERBB2 amplifications, PIK3R1, PTEN, and some of the other genes here are highly mutated. In uterine serous cancer, which is, again, not a hypermutator subtype, but there are frequent mutations here, which you do not see in ovarian cancer, which generally has very few mutations overall, and you don't see them as commonly in basal-like breast. So, there's many similarities. They share many genomic similarities, probably related to the shared high-frequency of P53 mutations. And some of the GYN specialists and biologists ask the question, well, do all these cases come from the same place? Are these all tumors that begin in the fallopian tube and some of them fall down onto the endometrium and some of them fall out onto the ovary, giving you endometrial cancer and ovarian cancer? The mutation data would suggest that's not the case, but you easily could make the argument that they come from the same original site, but the micro environment then induces them to become differentiated tumors. So to summarize, again, we've identified recurrent POLE mutations that are associated with the altered mutation spectrum and very high mutation rate. Very active PI3K/AKT pathway, which certainly has ramifications for targeted inhibition. But, again, one of the main points is that this genomic stratification can really complement or supplant histologic subtyping, particularly where there's poor interobserver concordance. And this may have very important effects in whether a patient receives a completely different modality of treatment -- radiotherapy versus chemotherapy -- after having a hysterectomy. And so, in this era of precision medicine, these types of findings will help to design clinical trials and bring the targeted agents to the clinic in a rational manner. Just a few announcements before I finish. You know, TCGA is doing a lot to combat cancer. I think we can do better with cigarette smoking. This is back to the Government Accountability Office, where they're helping you figure out where to smoke. And in the elevators of this building, you can have a $25 fine if you actually smoke in the elevators. I think the fine should be a little higher for smoking in the elevator. [laughter] The endometrial group will meet tonight at 5:00 in Salon II to discuss the manuscript and go over our punch list for the manuscript. What else do I have? There's many people to thank. I tried to include the key players on this slide. If I left you off, I'm very sorry. Elaine is the co-chair of the Analysis Working Group, with me, and has been wonderful to work with. JJ Gao is our data wrangler, and Niki Schultz has been coordinating all the figures for the manuscript. And so we're grateful to all those people. And I have 35 seconds. And so my last slide -- whoa, where'd my last slide go? My last slide went away. Can I have my slides back? The last slide will show you that the AACR is having a special conference on ovarian cancer next May. And the reason I put this up is not to advertise the conference, because people are studying ovarian cancer here, but they really came and said because of TCGA Ovarian Project, which was finished about a year or two ago, we now think it's time to have a special conference. And so now the work that's being done here is really, as we all know, being spread to the broader community. The subtitle should be "From TCGA to Clinic," but it's "From Concept to Clinic." So if you're interested in this disease, come next September to sunny Miami. And thank you for your attention. [applause] Charles Perou: Time for a quick question. Actually, I got a quick question. So I noticed the high degree of overlap in PTEN loss and PI3K/AKT mutation. PTEN loss -- or PTEN mutation and PI3K/AKT mutation co-occur quite frequently. Douglas Levine: Yes. So almost all of the PTEN mutations have either PIK3CA or PIK3R1 mutations in this cancer but not others. And when you look at the protein, there is some correlations as to which dual mutation you have and what that functional effect is. I think basically in all these clinical trials, which you're involved with and I'm involved with, we have to basically look at the outcomes of the responses in light of these types of commutations and see what the significance is. Charles Perou: So does the RPPA data show activation of markers down the string of PI3K/AKT in all these samples? Douglas Levine: Gordon has previously published that there's more functional activity if you have a co-occurrence of loss of PTEN protein with one of these as opposed to just the mutation loss of protein. Charles Perou: All right. Thank you, Doug.