 So, I'm grateful to have this opportunity to present on behalf of the many members of our TCGA kidney analysis working group. Genetic changes underlying clear cell renal cell carcinoma, or CCRCC, include alterations in genes controlling cellular oxygen sensing, such as VHL gene, and the maintenance of chromatin states, such as PPRM1 gene. For 446 CCRCC cases, TCGA evaluated clinical and pathological features, genomic alterations, DNA methylation profiles, and RNA and proteomic signatures. Shown here is a data summary for our project. In all, 446 tumors were profiled on most of the platforms shown here, while 372 tumors were profiled for all of the RNA and DNA data types. And so some of the integrative analyses focused on this core set of 372, but then other analyses made use of all data available. We surveyed these tumors by whole exome sequencing and found 39 significantly mutated genes, the most frequently mutated subset of which is shown here. And the somatic mutations were initially called by three sequencing centers, Baylor, College of Medicine, Broad, and UCSC. So each center analyzed the BAM files using their respective mutation calling algorithms. And shown here are the numbers of variant calls for each center for the top 50 or so most significant genes in our data. And so we can see the intersection between any two centers as well as calls that were made by one center, but not by the other two centers. And so for our analyses, for the integrative and pathway analyses, we focused on the calls that were made by two or more centers as being of a higher confidence. But then recently we carried out validation of these mutation data using an orthogonal platform, using the ion torrent platform in this case. And so here are shown the corresponding validation rates for our top genes. And as we might expect, we see that those calls that were made by any, at least two centers, had very high validation rates on the order of the high 80s to 90 percent validation. Whereas those calls that were made by only one center, we find that the validation rate is much lower. So our significantly mutated genes as a group were validated by an orthogonal method, and that's a nice result to have that strengthens our confidence in our results, and there's more on the sequence analysis and the validation in this poster, 114. Now back to our significantly mutated genes. It is well known that in CCRCC, it's closely associated with VHL gene mutations and alterations that lead to stabilization of hip 1 alpha and hip 2 alpha. And as expected, most all of our cancers were impacted for VHL gene with well over half showing both alleles inactivated by some combination of copy loss, somatic mutation, or DNA promoter methylation. Several genes involved with chromatin regulation were frequently mutated, including PBRM1, set D2, BAP1, and R1a. And this would implicate a major role for epigenetic reprogramming in this disease. And these chromatin modifier genes have also recently shown up in the other cancers under study by TCJ, and so this appears to be a very important process in many cancer subtypes. We also found a number of mutations in PI3 kinase pathway regulators as well as alterations at the copy level, so the PI3 kinase pathway appears to be heavily targeted. Now in CCRCC, the copy number landscape consisted primarily of chromosomal arm level gains or losses. So for example, 3P is a well-known broad region of loss that would encompass many genes, but a few key genes are also in this region, including VHL and some of the aforementioned chromatin modifier genes. Now there were fewer focal regions of copy alterations in this disease as compared to what we have observed in other cancers. And so here a focal region would represent a smaller region of gain or loss as observed in a subset of the samples, and a focal region would be indicative of genomic targeting of a single gene or a handful of genes that gives us a smaller region in which to focus. And for the focal deletions, which are shown on the left here in terms of their overall significance, we find a handful of genes that would appear to be a period of tumor suppressors for some of these cancers, such as VHL, F-Head, C2K2AMP10, and on the right we have our focal amplifications, which include some oncogenes such as MDM4, MIC, and then we have this region 5Q35.3, which has two genes which have been found to be able to drive the PI3 kinase pathway, GMB2L1 and SQSTM1. Now based on RNA-seq analysis, our group identified a number of recurrent fusion RNA transcripts as shown here, and using targeted methods, 11 of these 13 recurrent events could be validated. We found that mutations in specific genes, particularly those involved with chromatin regulation, could be associated with widespread molecular alterations. So for example, here we see widespread DNA hypomethalation being associated with mutation of the H3K36 methyl transferase, set D2. So this finding would be consistent with the emerging view that H3K36 trimethylation may be involved in the maintenance of a heterochromatic state, whereby DNA methyl transferase 3A normally binds trimethylated H3K36 and then methylates nearby DNA. And so if we take out set D2 function, then apparently we lose this mechanism of methylation in the cells. We also found that the total extent of DNA hypomethalation in cells increases with advancing tumor stage and grade. And this would suggest that the epigenetic state of a more aggressive kidney cancer, as denoted by a higher stage or grade, differs from that of a less aggressive kidney cancer. And from paradigm and hotnet pathway analysis, we found that mutations involving the SWE SNF chromatin remodeling complex show potential far-reaching effects on other pathways. So SWE SNF complex includes P-Barm1, R1A, and SMAR-K4. So shown here, the circles represent genes, and so for each gene we have various levels of data being represented, including mutation, expression, and inferred pathway activity by paradigm analysis. These data are shown across all the tumors examined, and then the data are rolled up into a circle for compact data presentation. And then a line between any two genes would denote some type of inferred functional relationship. So based on databases approaching and approaching interaction and post-transcription regulation, we find that chromatin modification pathway intersects a wide variety of processes involving genes such as SAC1, NFCAPA-B1, hip-1-alpha, June, FOS, beta-continue, and TGF-beta receptors. Now a number of analyses carried out by our group are set out to define molecular signatures and pathways of aggressive CCRCC. Now previously, as part of the TCJ paper in a varying cancer, we defined a gene expression as signature of survival using a training data set and then apply the signature to multiple test data sets in predicting higher risk versus lower risk patients. So this is based on mRNA data. And in multiple validation data sets, both internal and external to TCJ, we found that the signature could distinguish higher risk from lower risk patients. So having these patterns be reproducible across multiple data sets is nice to see. And so for the Clear Cell Project, we wanted to apply this approach, but also to extend it beyond just mRNA data, but also here considering three other data types, micRNA, DNA methylation and protein, in addition to mRNA. And so we divided our sample profiles into discovery and validation subsets. And using the discovery set, we defined the top survival correlates for each data type examined and then applied the respective prognostic signatures to the validation subset to demonstrate reproducibility of patterns. And for each data type examined, their respective prognostic signatures could clearly distinguish higher risk from lower risk patients with very profound differences in terms of overall survival. Now this would indicate that the molecular alterations involved with more aggressive kidney cancer involve multiple levels of data, not just mRNA data, which has been the most studied until recently, but also we see micRNA features, protein features, and DNA methylation features that can also be associated with more aggressive versus less aggressive cancers. Now some may look at a result like this and think of the possibility of driving a molecular based assay for use in the clinical setting for providing prognosis information for patients. And that may be a possibility, but I think there's another application for these results in that we can ask the biological question of what are the pathways or processes involved with making some kidney cancers more aggressive than others? Are there clues that can be found in these catalog of survival correlates that we have gleaned from the TCGA data? For example, when we drill down into our protein data, we found that survival correlates include AMP-activated kinase, or AMPK, and acetylquacoboxylase. These are proteins that are anti-correlated with each other, where low levels of AMPK and high levels of acetylquacoboxylase are associated with a worse outcome. These proteins are particularly interest due to their key roles in metabolism, specifically we know that down-regulation of AMPK together with up-regulation of acetylquacoboxylase together contribute to this metabolic shift of the cell from relying on oxidative phosphorylation to relying upon aerobic glycolysis. This is also known as a type of Warburg effect, and it can provide a growth advantage to the cancer cell. So given that initial finding, we went back and looked at our TCGA data in the context of these core medical pathways that have been well studied over the past 50 years or more and include glycolysis as well as oxidative phosphorylation that occurs in the mitochondria. And so I actually had flashbacks in my undergraduate biochemistry as I was putting this figure together. So these are very well-established pathways. But on top of this, we've overlaid our TCGA survival correlates data, where the boxes represent mRNA features and the diamonds represent protein features. And the color for a given feature denotes a correlation with patient survival, with red denoting a correlation with a worse outcome, and blue denoting a correlation with a better outcome. Now when we put everything together, we find that these survival correlates do indeed underlie a glycolytic shift. Specifically, we find that there is up-regulation of genes involved in the pentose phosphate pathways being associated with a worse outcome, as well as down-regulation of AMBK at the mRNA and protein levels, and up-regulation of genes involved with fatty acid synthesis being associated with a worse outcome. And then in the mitochondria, we see a systematic down-regulation of genes involved with oxidative phosphorylation and the Krebs cycle being associated with a worse outcome. So here we have an example of taking a catalog of molecular features gleaned from genomics analysis and putting these in a context of known pathways, and thereby we hope to achieve a more pathway-level view or a higher-level view of what is happening in the cancer cell. And so these changes in gene expression that are observed here being associated with metabolism are intriguing, but then we can ask the question of just what might be driving these events, what might be driving these changes in metabolism and the more aggressive kidney cancers. Likely there are many factors involved here, but one pathway that appears to be involved is the PI3 kinase pathway. A PI3 kinase pathway has multiple roles, one of which is to help regulate glycolysis and metabolism. And so if the PI3 kinase pathway is altered, then that likely results in changes in metabolism in the cell. And we do find that this pathway is highly targeted at the genetic and genomic levels. So here is an analysis from the MIMO group showing that different genes may be altered in different samples, but that if we take the sum total of these alterations, we find that there is a large percentage of these kidney cancers that have at least one alteration in the key gene in the pathway. Now these key genes include GMB2L1 and SQSTM1, which are located on the 5Q35.3 region. These genes have been perhaps understudied in the context of the PI3 kinase pathway, but could perhaps receive more study in light of these findings. But in addition to the genetic and genomic levels, we find that the PI3 kinase pathway also appears to be a target at the epigenetic level. Specifically, we find that promoter methylation of Mir21 and Greb10 contributes to PI3 kinase pathway deregulation. So shown here is a schematic of the PI3 kinase pathway where boxes represent RNA features, diamonds represent protein features, and in this case ovals represent DNA promoter methylation features. Red denotes a correlation with a worse outcome, blue with better outcome. Now on the one hand, in the more aggressive kidney cancers, we find that there is a significant decrease in promoter methylation of the mycariname Mir21, and this decrease in methylation corresponds with an increase in expression of the gene. Now Mir21 has a well-known and established target, which is P10, and we also find in our data as Mir21 increases, P10 expression decreases at both the mRNA and the protein levels. So these three events are happening together, and they tend to happen in the more aggressive kidney cancers. On the other hand, we find that the more aggressive kidney cancers show an increase in promoter methylation of the Greb10 gene, and that corresponds to a decrease in expression. And normally, Greb10 serves to inhibit the PI3 kinase pathway. It provides negative feedback for the pathway. And so if Greb10 expression is decreased, we can infer that that could only serve to further increase PI3 kinase activity. So here we have evidence for multiple mechanisms targeting the PI3 kinase pathway, and that these changes we see with respect to PI3 kinase correspond with the changes that we see in metabolism, and also are associated with a worse patient outcome. And so this is really a story that comes about from having all these different data types on the same set of cancers. We have microRNAs contributing to this story, as well as DNA methylation and mRNA and protein and genetic and genomic data. All are helping to fill in this picture of what pathways are at work here. So in conclusion, we find that integrative analyses highlights the importance of both VHL HIF pathway and chromatin remodeling histone methylation pathway in CCRCC. We also find frequent targeting of the PI3 kinase pathway at the genetic, genomic and epigenetic levels. And finally, we see evidence for a metabolic shift to aerobic glycolysis that appears involved with more aggressive disease. And this brings us to the end, and I thank you for your time. A few minutes for questions. Maybe I'll start off asking Chad to comment. This disease has become something of a poster child for tumor heterogeneity studies, and maybe you can comment on the extent to which we cannot inform anything at all about the tumor heterogeneity story given the nature of the data collection and the sample collection. Sure, yeah. I think any cancer is going to have an amount of noise. I think this type of TCGA data is going to have not just technical noise, but I think there is biological noise. And one of these is going to be the tumor heterogeneity. I think some diseases may be more heterogeneous than others. I think that's yet to be more fully fleshed out in the clear cell cancers. But I think what we're seeing here is maybe more of a, I think we characterize more of the dominant clones. And I think our analyses would favor looking across inter-sample heterogeneity, what's different across samples. I think drilling down, and within a given sample, I think that's another type of study, and it's certainly well worth while. I think we need to look at both avenues. Thanks. There's a question over here. So if essentially all these different prognostic factors obtain on different types of data actually captures alteration of the same pathways, then could you also exploit like clinical data and multivariate analysis to actually see if they are dependant or independent and understand if they actually cover the same alteration? So the different prognostic signatures, maybe put them in a multivariate analysis, see which one is. We've tried that, and I think it depends on the question. If one wants to provide more, one wants to ask which is providing more information, it might be a little bit of a wash. I think in the multivariate model, maybe the DNA methylation signature was not as significant as the others, but it's a little hard to interpret. I guess it's this question of if you want to divide another molecular assay for use in the clinical setting, and I think that's something that could be looked into a bit more. I think what we showed here is more asking the biological question, what are the pathways? And I think a multivariate analysis is trying to get at this, well, can we come up with a better clinical biomarker? And maybe we can. I think they maybe do provide a little more information than the clinical biomarkers, but there could be some tweaking here in terms of getting that more refined signature. Hi, Chad. Let's talk. I just wonder, using the training site and the relations that you show, different molecular features can predict survival data. I just wonder, given that clinical data, like tumor state, other clinical data, why do those molecular data provide additional prediction power? Oh, so if the molecular signatures provide more information than the clinical data? Yeah. Well, so if you try to make a multivariate model for each signature, I think we did it separately for each signature, and I think three out of the four signatures could provide a little more information. They still provide significant information if you factor in grade and stage and age and the other variables. Thanks. Hi. This James just tried to make a comment that we talked about this, and it's really tried to address the first question brought up by the attendees. So the idea is basically for the kidney cancer, we look at 400-plus cases, about 100 of them actually went down to receive systemic treatment. So we assemble a so-called clinical TCGA, basically have many physicians involved, and we are actually actively collecting clinical data that related to the treatment response. And I think that's the first step, at least, that we can take to understand the genomic basis underlying the treatment response. And that will actually help a lot in terms of how to prioritize and how to treat patients. And I think that would be the best use of the TCGA data. No, I agree. I think that's a worthwhile avenue of research. And I think that would speak to a lot of these TCGA projects where maybe the clinical data are still being collected. So as we collect more data, I think we can get even more out of these data than what we've got from the initial analysis by going back and maybe fleshing out the clinical data and asking more questions. Thank you. Let's take one more. Yes, for the MIR-21 and GRM-10, can you quantify GRB-10? Can you quantify the frequency of that in the PIC-3CA, the methylation, roughly? And secondly, what orthogonal platform did you... Were you able to confirm that in, say, methylation-specific PCR or something like that? What platform to confirm the methylation patterns? I don't know that... Well, I don't think we've... No, I don't think we don't have... We don't tend to validate the methylation data. I think we try to see if it syncs up with the other data types, but that's something that I think could be looked at maybe in independent cohorts or in the TCGA data. I think in terms of a percent, it's more of a continuous value. So I think I'd have to maybe ask Pierre Laird's lab what actual percentage, but I think just looking at a heat map and maybe a sizable percentage, maybe... I think when we bin the samples in terms of good, bad, intermediate prognosis, we did that on thirds, but that's a little bit arbitrary. I think you can look at maybe, on the order of 25%, could have lower or decreased methylation levels. Thanks. Thanks again.