 Anyhow, I'd like to thank everyone for inviting me and the organizers and also to let you guys know that I gave this talk about 2009, almost seven, eight years ago, among the same group of people. But we didn't call it the symposium. We call it the TCJ jamboree. I'm sure a lot of you guys probably remember that. Those are some good times. And today I'm going to talk about the efforts that was, no, that's not it, it's a PDF. Can you just go and view? Great. All right. So, yeah. So the title is Panglioma Integrative Analysis, the TCJ project since 2006 and one of three co-chairs, Roverhawk and Antonio Yavoroni, are leading this project and this analysis I represent the TCJ, LGG, GBM analysis working group. So brain cancer, as we know, is divided among grades in histology, as it was shown by Layla's presentation earlier today. Glioblastoma multiforme is a representative of high grade, grade four, typically of astrocytomates like cell types, generally have poor survival, whereas the lower grade gliomas are diffused, generally grade two and grade three with mixed histology, astrocytomas, oligodendrogliomas, oligoastrocytomas with an overall better survival. So generally we consider histological grading inversely correlated with outcome. And since 2006, as you all know, TCJ started with three primary pilot tumor studies, GBM being one of them, and GBM actually was one of the first marker papers that came out of our group, and I'm calling this here version 1.0 because subsequently in 2013, Cameron Brennan and Roverhawk and others and myself, we published the second marker paper version 2.0 with an expanded TCJ set. And in between that, we published two companion papers looking both at gene expression profiling for these GBMs as well as denomethylation profiling where we identified a very specific unique subtype, which we term G-SIMP. So we're very happy to note that TCJ produced four outstanding papers with high impacts and high citations, and the first or this version 2.0 marker paper in 2013 was actually presented by Roverhawk within the same meeting, then the second TCJ symposium, so I put up a link up here so you can view the YouTube if you like, but I just want to highlight three main points from that paper because it relates to what I'm going to present to you later today. So the first thing that we've identified was we identified novel mutations and rearrangements in EGFR using our expanded data set. Promoter mutation correlates with expression, suggesting a role in telomerase reactivation, and the G-SIMP, the champion there, was confirmed with best overall survival. IDH1, IDH2 mutation has been well-studied now in gliomas. We, and among others, reported a high correlation with IDH mutation with an impact on denomethylation as represented here. G-SIMPs having the lowest, I'm sorry, the highest methylation pattern among all high-grade GBMs generally defined as being younger, the age of onset, better survival, and as I said, associated with IDH1 mutation. And we've also shown that this is highly represented among the lower-grade gliomas among other people. So what we did here is kind of give you a broad-dive view of the epigenome landscape across all TCJ tumors. This is, we pulled down 10,000 plus tumor types across 34 different tumor types, calculated the average mean methylation level for each sample and plotted it here on a box plot and then sorted it by the overall mean distribution per tumor type. And what you can see is that the LGGs has the highest overall methylation level across all these different tumor types, whereas the testicular cancer, kind of the positive control there, has the lowest methylation level as expected since some synomers are mostly unmethalated. IDH mutations correlate with favorable outcome in GBM and astrocytomas. And the main marker paper that came out, or that's coming out in the coming weeks, was also presented last year by Daniel Bratt. And again, I'm not going to go through any of the topics that he mentioned. I'll just give you guys a link up here if you guys want to watch this video. But the main things that we are reporting in this paper is that we're identifying three major clusters. Layla kind of represented that earlier today. IDH mutant codels. These are samples that have the co-deletions of the 1P19 co-arm. IDH mutants with non-codels. And then there's a subset of IDH wild type. And as Layla has represented, these are generally of GBM-like. So we have IDH wild type. Regardless of histology, had genomic aberrations and clinical behavior similar to primary glioblastomas. Nearly all IDH mutant LGGs without the co-deletion had mutations in P53, ATRX, and so forth. And LGGs that were IDH mutant and co-del had the most favorable clinical outcomes and were associated with mutations among the SIC, FOPP1, NUCH1, and TURP promoters. So these two efforts by TCJ expanded our insight into the glioma field, particularly looking at the molecular associated with the clinical phenotypes. And given this, we then formed an analysis working group that started about a one-and-a-half years ago. We're currently in the writing stage. Again, the idea behind this is to then look and understand the characteristics molecular features between low-grade gliomas and GBM. And also, given this larger expanded data set, can we then understand the relationship between those GBM-like LGGs as well as the G-SIMP phenotypes that are very similar to the LGG phenotype. Now that we have this expanded set, can we now learn more about that? So the clinical molecular characteristics of the glioma data set that we have through TCJ includes over 1,122 samples. Majority of them are GBMs, but there's an almost even mix between GBMs and LGGs. Since the new paper that's coming out, New England Journal Medicine, which profiled 289 LGGs, we've expanded that now with an additional 290. And IDH status are known for over 87% of the samples. So molecular profiling of the largest glioma data set to date, we have a diverse set of platforms that we've used. Again, this is going historical here from 2006 beyond, so you can imagine that technology has advanced considerably during this time. So there's a lot of different platforms that are being represented per experiment. But the major experiments that we've profiled and have for this particular data set includes gene expression, DNA copy number, DNA methylation, somatic mutations. So the overall genomic landscape that we've found, looking at this entire pan glioma data sets is that GISTIC have identified 57 disjoint amplifications with 105 deletion-deleted regions. They're very large megabase regions across 1,000 glioma patients. MUTEC identified 100 candidate genes. This is profiled across the three different centers, 30 of which were previously reported, both from the first marker paper in GBM and the new marker paper that's coming out. And again, the usual suspects, IDH1, P53, ADRX, EGFR, P10, PIK3CA, PIK3R1, and NF1 have all been represented with an additional set of 70 genes that are now being reported. And here's just a snapshot of some of these genes, what we are now able to get from this larger sample set in power is that we can get lower frequent numbers of mutant events. So, and these are candidates for oncogenes and tumor suppressor genes, such as CETD2, ERD2, K-RAS, DNMT3A, and N-RAS and so forth that have not ever been reported in either GBM or LGGs, and so we're now able to now report this in this new manuscript. So, looking at the overall genomic landscape, Flores said what he done was organize the patients according to those three major clusters that we are looking at, the IDH mutants, IDH mutant non-codels, and IDH wild type. And then here, selecting specific genes associated with particular pathways, such as cell cycle, the RAS pathway, apoptosis, chromatin modifiers, mTOR, notch signaling, and cohesion. And what we can see from here from this global landscape is that the vast majority of the IDH wild type are represented by cell cycle and RAS pathways, as well as the mTOR pathway, whereas the IDH mutant non-codels are dominated by apoptosis related genes as well as chromatin modifiers, and the IDH mutant codels are dominated by the notch signaling. What we also did was look at the relationship between ATRX and TURP promoter mutations. We found that they're mutually exclusive in our cohort, as was reported in several recent articles in the past couple years. One thing that we did do with this dataset was use the normal tumor blood samples and estimated the tumor length, a telomere length for each of these tumors, and then associated that with whether they are ATRX mutants or TURP promoter mutants. And what we found is that the TURP promoter mutant subgroups have a lower telomere length overall compared to their normal partners, whereas the ATRX mutants have a longer telomere length. The RNA sequencing data, we have over 667 gliomas represented here, defines four major clusters, three of them dominated by IDH mutants and one of them dominated by the IDH wild type. We've also looked at dynamethylation, identified two macro clusters dominated by IDH mutants and IDH wild type, and within each of these groups we have three subclusters, LGM1, so we're calling all the clusters from here on out, LG is lower grade glioma and GBM, and then M would represent the platform. So in that case for the expression we'll call it LGR. And so here we're calling LGM1, 2, and 3 dominated by IDH mutants, and LG4, 5, and 6 are the wild types and there's a split between the GBMs and the LGGs. In collaboration with Sophie, she was able to, and her team were able to, merge the two expression and methylation data to generate a tumor map, and then overlaying that with sample types defined by the four different clusters by expression and the six different clusters by methylation. And you can see they nicely separate the IDH mutants and IDH wild types. What we also looked at, and what's quite interesting is that the IDH mutants are dominated by three different clusters, and if you look at the genomic, epigenomic profile, genome-wide is looking at over 20,000 probes per sample, there's clearly a difference between LGM1, 2, and 3 by methylation, even though they all are represented by an IDH1 mutation status. And even among the IDH wild type, there are some differences between them. So this is not looking at the methylation just from that heat map, but it's using that heat map as a discovery and then now looking at the entire genomic profiling and then seeing that there is true stratification genome-wide. Survival now shows that there's this strong stratification by these LGM methylation clusters. LGM1, 2, and 3, which as I said are IDH mutants, they have a higher methylation profile, have the overall best survival, whereas there is a subset of this IDH mutant, which returns LGM1 that have a lower survival, and this is quite significant even after adjustment by tumor type and age. IDH wild type also shows a striking difference in survival, whereas the LGM6 here highlighted in blue shows a tendency for an improved survival. Now I'm not going to talk about the IDH wild type in this talk, we don't have a lot of time, but I'm going to go ahead and mention to you some of the findings that we've come up with within the LGM1 cohort. So what we did touchy here, my postdoc in my lab, she's in collaboration with McKaylee, took the DNA methylation data as well as the gene expression data, and I didn't tell you this earlier, but the LGM1s are dominated by the non-codels. And so to eradicate any influences by the codels, we just focus only within the non-codel sub-cohort, and then looked at the differences between this LGM1 and LGM2, since LGM2 is dominated by the non-codels. And what we found is that there is a lot of genes that have lost methylation compared to the counterparts, these are all associated within, is all referenced in relation to LGM1, and we also find that genes that are up-regulated and genes that are down-regulated as they lose their methylation. So interestingly, when we take these 179 probes that we've identified as being different between these LGM1 cohort and the LGM2 cohort, what you find is that the dominant difference is not just the entire cohort of the LGM1 non-codels. We remind you again, the LGM1s were defined by a discovery panel that was designed to look at two more specific probes. Now, when we do a more supervised analysis, given that discovery analysis, we now discover a substructure within the LGM1, this hypo class within the LGM1, and it's represented even in a different dataset, like gene expression dataset, which have a pronounced expression pattern elevated within this cohort. You can see the second track there, the black and the white, the blacks are the GBMs and whites are the LGGs. So it's not just high-grade GBMs that are dominated within this feature. There's a mix of some LGGs as well in there. So she then looked at the overall mean methylation level, and lo and behold, those guys are genome-wide, are much lower methylation than the reciprocal LGM1 hypergroup, as well as the LGM2 hypergroup. And then when we looked at survival based on this new stratification, you see now that the dominant trend for poor survival within the LGM1 IDH mutant codel, or non-codel subtype, dominated by this hypo phenotype within the LGM1, whereas the LGM1 hypergroup is an elevated survival similar to its counterparts in the LGM2 and LGM3. So we set out to then investigate genome-wide, given that we have a 450K platform that many of you probably worked with and know very, very extensively, that these not only covers canonical promoters, but also covers intergenic regions, such as enhancers, candidate enhancers and silencers. And so what we set out to do is look at all these features that we have, given the differences between these two platforms and see whether we can find genomic features that are unique. And what we find is that, is that there, I don't know if I have a pointer, yeah, 67% of our set of probes that we find are enriched in the open seas or these intergenic regions. And when we do a motif analysis to define a genomic signature, we find that SOX2 and the SOX family motif factor is prominent and enriched within these cohorts of probes. And it's, we also found that the expression of SOX to be elevated, 12 out of the, I'm sorry, 10 out of the 12 SOX families are elevated in this cohort compared to this LGM2 cohort. We then set out to validate this data. We pulled in a DNA methylation data that was recently reported in nature. These are all IDH mutant cohorts. And using the same probe set that we have here, given this new data set, non TCGA data, mind you, we can now then see that there are three samples of very low methylation level here. And if you look at our data set here, we have 25 that we're identifying as LGM1 hypo. This represents a roughly a frequency rate of about five and a half percent of our data set. And in their data set, three out of 49 represents 6%. So it's kind of nice that the frequency is represented here. So one last thing, we also looked and tried to identify epigenetically silenced biomarkers or prognostic biomarkers that defines these differences in survival. What Thais did, student of my lab, she then defined five different regulatory groups. We're calling them e-regs, epigenetically regulated group one, two, three, four, five. These are groups of genes that are hypermethylated within a group of patients that we now know are different in survival and clinical output compared to all the other ones. And here are some non TCGA tumor brain samples here. And then we took the RNA sequencing data. So for example, these cohorts of genes here are hypermethylated and their gene expression for those same genes are down-regulated and up-regulated in all the other ones. And again, here, this is represented of the IDH mutant as a whole, down-regulated, up-regulated and so forth. You can see that there are sets of genes here that are well defining each of these different survival groups. So obviously, this is not as important unless you can validate this. And so we did validate this. We pulled in four different publications that was recently published within the last two years. The Edelberg Stern paper here looking at 136 GBMs among 59 pediatrics and some adults. 61 Pylocytic isocytomas. These are tumor types of low-grade grade one types. Lower-grade gliomas here mix of non-codels and codels as well as 46 oligodendroglioma tumors. And you can see very clearly using these same sets of genes here that we've identified using the TCGA cohort, you can see that they're very well reproduced within these sample sets. And then we've also been able to classify these data according to our subtypes of LGM1, 2, 3, 4, 5 and 6. The survival, very similar to what we're producing. Obviously, we have the advantage of a larger data set. So this LGM1 cohort is represented of 55 samples, whereas in our validation set we were able to identify seven in their cohort. Yet there's still a nice trend for the similar profile. OK, so I've got 30 seconds and I'm just going to end here with a summary, four major points. One, we identified several novel genes that like to contribute to glioma genesis. We show that mutations in Turton ATRX have an impact on telomere length. We identified molecularly tumor subtypes that defy traditional histology and we identified epigenetically regulated genes that can predict patient outcome. And again, I just want to acknowledge some of the key people that have worked on this project. Again, this is a project since 2006. So every member here has contributed significantly to this work. But leading this project right now is myself, Rue Verhoc and Antonia Yavororoni, as well as Flores, a student rules lab, Micheli, a collaborative artist from Italy, and two of my students here, Tatyani Malti and Taiisa Bidachi, who have been doing lots of work on the epigenetics. And again, lots of people involved in this project, as well as all the analysis working group from GBM to LGG are being represented here. So thank you very much. IDH is a metabolic gene. Do you expect to see differences and have you looked for differences in the metabolic profiles of the tumors? No, we did not look at that. But yeah, we would expect to see something if we have. Maybe the proteomic data can help us with that. But again, you have to understand that GBMs were started in 2006. Proteomic stuff hasn't started until later on. So we're underpowered there. I know that TCGA was not really designed to have clinical impact. But this is such a terrible disease. And I know brain cancers were selected to be an important TCGA tumor type because prognosis is so bad. Do you think that there's anything that you've done here that will have an impact on patient care? And of course, you can't count the discovery of IDH1, which you know came from another group. Yeah, right. And definitely, the clinical impact. No disrespect intended. The clinical impact of this is, I think it was pronounced, is very strong because there are subsets of patients within the lower grade gliomas, for example, that are wild type, that have differences clinically as well as molecular differences. And this is going to help, at least for treatment protocols, to define whether they should be exhibiting a particular treatment protocol or not. So thank you. It's a difficult question, but I'm not a clinician. It's OK. Go ahead. I'm thinking of a big clinical impact because I'm just looking at Antonio Gavarone, who's near me, a discovery of FGFR3 translocations. I'm not sure that was strictly speaking a TCGA thing, but he's part of the TCGA group. But my question actually was about a different kinase that you mentioned, which was KDR. And you talked about KDR as a new candidate oncogene. But it's part of the, you know, that's really driven by the fact that it's part of the same amplicon as PDGFRA and KET. And I'm wondering, what's the evidence for KDR being a separate actual target of the amplification or driver oncogene? Just wondering if you can comment on that. Yeah, I really can't answer that question effectively. And I'm just going to have to say that there's just a lower number of samples. We're looking at about five samples or 10 samples that exhibit this particular mutation. But yeah, I'm going to have to defer to Antonio for that. So Huda, I was just thinking it might be important to look at those five mutations and see where they land. OK. Yeah. I hope Ruel is listening. OK. Thank you very much, Hutan. We are going to move on to the next talk. So Angeliki Pantasi from Harvard Medical School will talk about somatic structure rearrangement in RAS pathway.