 Thank you for the introduction, and I'd like to thank the organizers for giving me this chance to present on behalf of the TCGA lower-grade gliomas analysis working group on the integrative genomic characterization of this very interesting group of tumors that are shown here. So this is one example of a lower-grade glioma, and this is the pathology. And compared to the normal brain over here, you can see the infiltrative nature of this disease. The background is whiter, and that represents a lot of edema that's being generated by the infiltration of these tumor cells. This is a family of neoplasms rather than one specific type. And the WHO classification of the lower-grade gliomas is shown here. They include astrocytomas of grades 2 and 3, oligodendrogliomas of grades 2 and 3, and oligoastrocytomas of grades 2 and 3. These tumors are thought to be biologically related, but the data for that has not been complete. They're all uniformly fatal, but the time of survival varies considerably from about one year to 15 years. Many of the tumors with astrocytic lineage do progress to glioblastoma. That has been the subject of a previous TCGA investigation, and you'll hear some references to that in my talk. Now, unfortunately, the classification of gliomas is not as simple as this in practical terms. At the ends of the spectrum, the ends of the biologic spectrum, we have astrocytomas of grades 2 and 3, and these tumors are characterized by the infiltration of these tumor cells that are dark and large. They're hyperchromatic. They're irregular in shape. They're known from previous studies to be characterized by IDH mutations, P53 mutations, ATRX mutations, and under the microscope, neuropathologists can prognosticate these and grade them based on morphologic appearance. Grades 2 have about a 60-month mean survival, whereas grades 3 have a 36-month survival. These are the tumors that will progress to glioblastoma and represent secondary glioblastomas when they progress along these lines. Oligodendrogliomas, on the other hand, are characterized as infiltrative tumors that have these round regular nuclei, perinuclear halos, the so-called fried egg or honeycomb appearance. They're characterized by, again, IDH mutations, which you'll hear a lot about today, and 1P19Q co-deletion is kind of their genomic signature, at least, of a subset of them. They also have CIC mutations, FUBP1 mutations, and mutations in the TURP promoter. They, in general, have a better prognosis, grade for grade, than astrocytomas, and it's important to recognize these because they are deemed to be more chemo-sensitive as a group. Now, those oligos and astros that I just shown, I think everybody would agree with their classification. The problem in classification, and therefore, biologic understanding, is that there's a large set of tumors that are called oligoastrocytomas that seem to have both these round regular oligodendroglioma cells and large hyperchromatic irregular astrocytoma cells. There's also a group that are just morphologically ambiguous and well-trained, well-intentioned neuropathologists will disagree on what to call these. So there is a real problem in our classification and biologic understanding of this family of neoplasms. Part of this has to do that we're still using the same histogenic classification of these neoplasms that we've used for almost a century now at the time of Harvey Cushing and Percival Bailey's classification based on presumed histogenesis of these tumors. You'll see that they used terms like astrocytoma and oligodendroglioma and glioblastoma nearly a century ago. Interestingly, there is no mention of oligoastrocytoma in this paradigm. Chances are we'll be returning to this in the future based on some of our studies. So even back a century ago, the authors of these original classifiers said it's impossible to classify these tumors as oligo versus astro under the microscope. They appear just like one another. And a little secret of our neuropathology community is that we really only have 60 to 70% concordance among neuropathologists. So there's a lot of inter-observer variability and in fact there's a lot of intra-observer variability. So we have a problem in our classification and understanding and it's clinically meaningful. That is, there are neuro-oncologists around the country and internationally that have really criticized our histologic classification and are just dying for a better molecular classifier that is more reproducible and clinically meaningful because these tumors that we're going to be discussing are treated in a different way by watchful waiting, chemotherapy, different doses of radiation, etc. So I was more than thrilled when the TCGA decided to launch a study on the lower grade gliomas and many of you are from the institutions listed here. You're well aware of the pipeline that goes from tissue sample to analyte extraction, sequencing, etc. And then finally the integrative data analysis that I'm going to be telling you about. The goal is the analysis of 500 lower grade gliomas. Surprise to some people is that we've already exceeded that goal of collecting more than 500 lower grade gliomas and they've been analyzed on the platforms listed here. Again, many of you in the audience are from these institutions and have done a wonderful job leading these centers. At the time of the data freeze, which was a slow freeze with some thaws, we had about 293 cases with overlapping data on 254 for the major platforms involved and these are the platforms and the numbers of samples for each of those platforms. So we are in the process of finishing up our first biomarker manuscript and hopefully what we submitted within a couple of weeks. So a very important slide is shown right here and I'm going to go over it in some detail. It is the analysis of significantly mutated genes in the lower grade gliomas and it's been separated into specific mutational classes. So in summary, I think you need to know that IDH mutations occur in 80% of lower grade gliomas. We're going to call those the IDH mutant lower grade gliomas and you can see those are the blue bars, IDH1 and IDH2, isocitrate dehydrogenase right here. These are IDH wild type. Now within the IDH mutant group, there seems to be two classes of mutations. One shown here, dominated by P53 mutations and ATRX mutations, those are mostly astrocytomas and oligoastrocytomas. The other subset does not have P53 mutations but has these CIC mutations, FUBP1 mutations and notch 1 mutations also pick 3CA mutations. These are mostly oligodendrogliomas and they are also characterized by a loss of 1P and 19Q chromosomal arms that will go into. So the IDH wild type tumors are an interesting subgroup. This represents 20% of the tumors and they don't have these characteristic mutations in ATRX P53 for the most part. They don't have these mutations either in CIC, FUBP1 or notch. Instead, they have mutations in genes like EGFR, NF1, P10, PIC3CA as well. Those are mutations that are more typical of glioblastoma and IDH wild type glioblastoma, the highest grade of diffuse glioma. So it's apparent that there's three mutational classes of these lower grade gliomas based on significantly mutated genes. When we look at copy number alterations, we can look at it in a couple of different ways. If we look at it by histology by oligodendroglioma, oligoastrocytoma and astrocytoma, we'll see that there are characteristic 1P and 19Q losses in oligodendroglioma but they're also seen in other histology types. We also see that there's gains on chromosome 7 seen in all histology types, losses on chromosome 10 in all histology types with some clustering. However, if we look at unsupervised clustering shown here, you have really tight clustering of specific gains and losses in specific subsets. So this is, again, just unsupervised clustering and we have gains of 7 and losses on 10 in one specific group. Those happen to be that IDH wild type group of lower grade gliomas that have the mutations like glioblastoma. 1P and 19Q losses correspond to this molecular cluster which is the group that has the CIC FUBP1 and is mostly oligodendrogliomas. And then this group is mostly astrocytomas and oligoastrocytomas that have the P53 mutations and the ATRX mutations. And they're characterized by a frequent 8Q24 amplification event that occurs in the region of MIC and includes MIC. So based on copy number and mutational findings, we did another unsupervised clustering. This is an algorithm called Ancosine developed by Giovanni at Memorial Sloan Kettering and it takes into account both mutations and copy number alterations and focal deletions and amplifications as well. And by employing this algorithm, again, there were three subgroups of lower grade gliomas that came out called OSC1, OSC2, and OSC3. And what we found is that OSC1 was IDH wild type gliomas that had findings of glioblastoma, CDI, and DKN2A loss, EGFR amplification of mutation, P10 mutation, and NF1 mutation. OSC2 was the IDH mutant 1P19Q intact group with the P53 and ATRX mutations, always in the setting of IDH mutations. And OSC3 was the IDH mutant 1P19Q co-deleted that also had the FUBP1, NOTCH1, and CIC mutations. So three distinct subgroups based on unsupervised clustering. Some of the other groups that we, some of the other platforms also yielded interesting results. This is the DNA methylation status. And you'll see that there's five distinct subgroups based on methylation status. This group that has relatively lower levels of methylation is IDH wild type. This is in stark contrast to the GBMs, the DeNovo GBMs that are almost entirely all relatively under-methylated in IDH wild type. And then the other stats include IDH mutant, which have that G-SIMP or hyper-methylator type. But these are divided into P53 and ATRX mutant forms and IDH mutant 1P19Q co-deleted. There's also a very small subset that we find interesting that was relatively under-methylated compared to others. That would probably be IDH wild type as well, but they are not clustered as well in terms of recurrent genomic alterations. Gene expression analysis yielded four strong subsets. Again, these gene expression subsets seem to correlate with IDH status in 1P19Q. So the purple group right here, which is a tight cluster on gene expression analysis, includes mostly the IDH mutant 1P19Q co-deletions. This expression group includes mostly the IDH mutant P53 mutant group. And this group represents mostly IDH wild type. There's a fourth group that was very mixed in its histologies and molecular classes, but it clearly had a defining signature as being more pronural or neuroblastic using multiple types of gene signatures. So based on these individual platforms of copy number, methylation, gene expression, and I didn't talk about microRNA expression, but we came up with four classes using that platform as well, there were between three and five molecular classes, three based on copy number that was very strong and tight, five based on methylation, four based on gene expression, and four based on microRNA. So it's possible to take each of these clusters for each case and do a cluster of cluster analysis to see what the overlaps of these individual clusters on individual platforms would be. And this was a very nice analysis done by Mia Gifford and Sophie Salama, which again reiterates that biologically there are three very tight clusters that are robust and clinically meaningful. So if we look at all the clusters located here on the left hand axis, you can see microRNA, RNA expression, methylation, copy number, and then this group right here is a clear tight cluster, and those are the IDH wild type lower gray gliomas that have the genomic alterations like GBM. These are the IDH mutant co-deletion tumors that are rich in oligodendrogliomas, have IDH mutations and 1P19Q deletions, and these are the IDH mutant non-co-deleted tumors that have P53 mutations, ATRX mutations. So there's three very robust non-overlapping lower gray glioma tumor types. You'll also recognize from this that we have kind of an astrocytoma signature and an oligodendroglioma signature if you will. The molecular data does not support a mixed histology group, so although there may be mixed histology or ambiguous morphology under the microscope, the molecular findings are going to cut these in two. Is this clinically relevant? Well, we think so. You can see the survival curves and progression free survival or event free survival really separate these tumor types based on IDH and 1P19Q status. This is the IDH mutant 1P19Q co-deleted 1P19Q intact and IDH wild type gliomas, and we separate them very nicely based on overall survival and progression free survival. We then wanted to look at the IDH wild type group because they seemed to have the molecular signatures of glioblastoma, the highest grade form of astrocytoma, and in fact we saw that the lower grade gliomas that were IDH wild type had almost the exact frequency and types of mutations as IDH wild type glioblastoma, including P53, P10, EGFR amplification and CDK and 2A amplification, but also on the less frequently mutated genes, they were very similar in nature. Similarly, we found that specific oncogenic gene fusions noted in the IDH wild type lower grade gliomas were very similar to GBMs. They included the FGFR3 translocations and EGFR gene fusions that are believed to be oncogenic in similar frequency to the GBMs. So in fact, the lower grade gliomas that are IDH wild type really have the molecular findings and the outcomes of glioblastoma rather than the other lower grade gliomas that are IDH mutant. This is a very important step in the field of neuro-oncology to be sure. What does this mean for therapy? Well, the RPPA analysis did shed some light on that. There were certain proteins that were upregulated in this IDH wild type group that might be targets of therapy, and certainly tyrosine kinases like EGFR, CMAT, HER2 might be targets of therapy. So in summary, six histologic diagnoses can be distilled into three robust clinically relevant molecular classes. IDH mutant 1P19 Q-coated tumors have CIC, FUBP promoter notch, and PIC3A mutations. IDH mutant non-coated tumors are characterized by P53, ATRX, and AQ24 amplifications, and the IDH wild type lower grade gliomas really have molecular alterations and clinical behavior similar to GBM. So thank you very much for listening, and I'd like to thank the analysis working group and the leaders of it. This has really been an incredible project for me to help lead. Time for one question. One question. The question is really, what's the major difference between the IDH wild type and GBM? You know they're very similar, right? Because they're histologically a little bit different. Is it any molecular differences? Because you're saying the mutation profile is the same, everything is the same, but there must be something epigenetically different or something? Yeah. At the genomic level, copy number and mutations are virtually the same. In terms of survival, the IDH wild type LGGs may have a slightly improved survival. And then if you look at gene expression analysis, they're not the same biologically per se. I think the introduction of necrosis and hypoxia into GBMs does lead them to segregate themselves based on gene expression profiling. So, slightly different biologically but genomically identical. So I think this could be epigenomically different? Epigenomically, they look very similar. So, as TCGA generates huge amount of data, the fundamental question is how to use this genome data to guide patient care. So next week is Dr. Andrew Mado from British Columbia and cancer agents here to talk about how to use TCGA data to inform procedure medicine in late stage cancer settings.