 All right. Hi, everybody. My name is Julianne Shee. I'm part of the Broad Institute's GCC or Genome Characterization Center, and today I'll be talking about somatic copy number alterations and aneuploidy events in UVL melanoma. So this AWG actually started fairly recently in October 2014, and we're having our first face-to-face tomorrow, so we'll have a lot more integrative analysis after that. And so while today I will bring in some of the other platforms, this talk will mostly focus on copy number analysis. All right, so just a quick introduction to UVL melanomas. The UVA is a layer of your eye right underneath your sclera, which is the white part of your eye, and it consists of the iris, the ciliary body, and the choroid. So these tumors are rare, but they're also the most common intraocular cancer and occur at a rate of about 6 per million per year. So what's important to note is that they're clinically distinct from cutaneous melanoma, and over 70 percent of UVM patients actually develop liver metastasis as the first and only site, despite less than 2 percent having clinically detectable metastases at time of presentation. These are also very molecularly distinct from cutaneous melanomas. They don't have BRAF mutations, and instead I have GNAQ and GNA11 mutations, as well as about 80 percent of metastasizing tumors, specifically having VAP1 inactivating mutations as seen in papers in the past. Currently, the most common prognostic test in the clinic is a 15 gene classifier that developed by Castle Biosciences that predicts metastatic risk fairly accurately. So there's actually been quite a bit of work done in aneuploid and UVM melanomas in the past, and in particular, this one paper from 2008 finds that about 50 percent of tumors have this chromosome 3 loss, and you can see white denotes a loss, whereas black denotes a gain in this heat map. And so chromosome 3 loss of heterosegosity is a marker for metastasis as well as poor prognosis, and they also found over here that higher aneuploid tumors tended to have worse survival than those tumors with lower aneuploidy. And so compare that to the TCGA-UVM cohort. We have 79 tumors looking at copy number with epimetric SNP6 chips, and you can see that this overview looks fairly similar to what's been seen in the past. So you have about half of our tumors also have a loss in chromosome 3, and the patterns in chromosome 6 as well as 8 also look very similar. So before I get back into copy number, the copy number landscape of UVM, I'd like to show some significantly mutated genes in this tumor type. So there aren't a whole lot, and underneath this blue line right here, you start dropping off in terms of power, but above that we do see GNAQ and GNA11 mutations in 73 out of 79 total tumors, and then there's SF3B1, EIF1AX, and BAP1 mutations. And these have also been seen in previous whole exome studies done on UVL melanomas. So I did hierarchical unsupervised clustering using just the copy number amplifications as well as deletions, and as you might expect, they separate very nicely into two main groups. So you have cluster 1, which is on the top, which does not have a loss in chromosome 3, and then you have on the bottom cluster 2 or subtype 2, which is defined by this chromosome 3 loss. But within these two clusters, you can actually separate them further into 1A and 1B on the top, where 1A is very, very quiet in terms of somatic copy number alterations, and 1B is characterized by a gain in 6P as well as 8Q. On the bottom, 2B is characterized by a very high gain in chromosome 8Q, as well as just overall increased aneuploidy. So there's also a very striking somatic mutation pattern here, and so moving from the left to the right, you can see that, once again, 73 out of 79 of these tumors have either GNAQ or GNA11 mutations. These are fairly mutually exclusive, although not entirely mutually exclusive. And then you have EIF1AX mutations, 10 of which are found solely in cluster 1A. Sf3B1 mutations, of which there are 18, 13 are found in cluster 1B, and for an all 16 back mutants, mutations are actually found in either subtype 2A or 2B. And this is particularly interesting because BAP1 is located on chromosome 3, and so all of these mutations are actually homozygous mutations. So here are a couple clinical correlations to the copy number subtypes, epithelial tumors which tend to have poor prognosis and are found more commonly in subtype 2A and subtype 2B, whereas spindle tumors are found more commonly in 1A and 1B. Subtype 2 tumors are also found more commonly in the ciliary body as well as the coroid, whereas subtype 1 tumors are found mostly in the coroid. And most importantly, subtype 1A and 1B tumors have significantly better survival than 2A and 2B tumors. So the difference here between 2A and 2B is, as you can see, not significant, but there's at least a hint of a suggestion that 2B, which is again characterized by increased aneuploidy and a very high 8Q gain, may actually have worse overall survival than cluster 2A. So as you can see here, there's not much chromosomal instability in UVL melanomas compared to a lot of other solid tumor types. And it's been suggested that this is because they occur in your eye, which is obviously a very sensitive area, and so people tend to catch these tumors early. But also with 79 tumors, we don't have a whole lot of power to detect rare events. So these amplification peaks here are very large and actually encompassed in my arm level analysis. But there are a couple of interesting deletions, specifically on chromosome 3. And so the ones I've listed here are all exonic and found solely in subtype 2 tumors, which means that they actually cause a homozygous deletion in these tumors. And furthermore, they're all mutually exclusive, including one additional BAP1 homozygous deletion. But unfortunately, there's not a whole lot of literature out there about those three genes. Okay, so BAP1 mutations, as I've mentioned before, are found predominantly in metastasizing tumors and in our cohort are found only in subtype 2 tumors. So germline BAP1 mutations have also been previously found in families and individuals that are predisposed to cancers such as mesothelioma and UVL melanoma, but none of these are actually seen in our cohort. So 12 out of 16 of the BAP1 mutations that we see are either frameshift, nonsense, or splice site mutations. And so theoretically, we should see a drastic decrease in BAP1 expression in these tumors. So I started with a quick sanity check, looking just at the BAP1 expression between cluster 1 and cluster 2 UVL melanomas. And so you can see here that within clusters 2A and 2B, which again have BAP1 loss of pedozygosity, there's a significant decrease in BAP1 expression. But when we just zoom in on subtype 2 and look at the BAP1 expression by mutation status, looking at those that are wild type or have a missense mutation versus the ones that should theoretically have a truncating mutation, it's harder to know what's going on with these mutations. For one, not all of these truncated mutations have a significantly lower BAP1 expression. But even if we choose to look just at this bottom half of the truncated mutations, there's still some evidence that there are some BAP1 alterations or silencing that we're missing in the LOH tumors. So we ran absolute, which is an algorithm that infers tumor purity, poity, subclonality, and whole genome doubling status, as well as generates an absolute copy number profile for each homologous chromosome or allele. And so you can see here that the majority of UVMs are very high purity with a median of 0.95. And here we've plotted poity versus whole genome status. And so most of these tumors are near diploid with a median poity of 2.02. But there are five tumors that are whole genome doubled. And out of these, four of them are in subtype 2a and actually have a chromosome 3 isodysame. And so here's a quick figure showing you how this could arise. So starting with this diploid tumor on the left, for whatever reason it loses this one red allele leading to a loss of heterozygosity in chromosome 3. And when it whole genome doubles, you get an isodysame of chromosome 3 and a tetrasome in all these other chromosomes. And lastly I'd like to end on a sneak peek at what the overall landscape of the TCGA platforms looks like in comparison with the somatic copy number alteration clusters. And so this is still work in progress. But it seems right now that methylation and messenger RNA expression correlate extremely well with our clusters. And in addition, looking at just the mutations, EIF1AX mutations are found again solely in cluster 1a. And BAP1 mutations are found solely in clusters 2a and 2b. This is extremely striking because these clusters are determined solely on looking at somatic copy number alterations. And yet all of these other platforms correlate extremely well with this. Okay, so just a summary. So somatic copy number alteration clusters are consistent with methylation and messenger RNA clusters. And each of them is actually defined by particular somatic mutations and are very predictive of what mutation would be found in any given tumor. So 1a was the most quiet in terms of copy number alterations. They also have better survival and tend to have EIF1AX mutations. 1b was characterized by a gain both in 6p as well as a 2 gain. Had better survival as well as SF3b1 mutations. Cluster 2a was characterized by loss of heterozygosity in chromosome 3, worse survival and BAP1 mutations, specifically truncating mutations. And cluster 2b also had loss of heterozygosity in chromosome 3 but are additionally characterized by high 8q gain as well as increased aneuploidy. This seems to have the worst survival out of the 4 clusters and tend to be also BAP1 mutated. So what does this ultimately mean? So we know now that copy number subtype 1 and 2 look extremely different in multiple platforms including methylation and messenger RNA. And we also know that chromosome 3 loss of heterozygosity is highly predictive of metastasis and worse overall survival. And so perhaps in the clinic copy number subtype 1 and subtype 2 patients could be actually treated differently. Going forward this AWG will be integrating all of the different platforms into one cohesive story about UVL melanoma that hopefully will impact patient care in the future. So I'd like to thank Andy Churniak for working with me on copy number analysis. Julian Hess who's also at the Broad for mutation analysis and Bita and Syriac for leading the group as well as the rest of the UVM analysis working group. All right, thank you. Very nice talk. Two questions. So number one, the EIF1AX mutation seems sort of mutually exclusive to the SF3B1. And one is initiation, the involving translation, the initiation when it's splicing. Do you know any crosstalk? It seems like mutually exclusive to me in terms of the distribution. Right. I'm not actually familiar with the biology behind this, but yes, they are mutually exclusive with one exception. And they're all found in the non-monosome 3 tumors. Yeah, and the other thing is the 3 mutation, chromosome 3, like clear cell kidney cancer, there are multiple tumor suppressors lost in addition to BAP1, including BHL, PBI, M1, and ZD2. You didn't see those mutations in your cohort, right? Right. We didn't see them significantly mutated. It's unclear. I haven't checked if there are just one-off mutations, but we should take a look at that. Yeah, because when we saw a subset of clear cell patients that have ZD2 mutation, and some of them have SF3B mutation, they're sort of mutually exclusive. It's almost very similar to here. So I don't know whether there's any. I might have missed this, but can you explain a little bit more about the cohort? But were they all metastatic samples that you were analyzing? So, again, it's difficult to tell because within the clinical data, so as I mentioned, it's difficult to determine whether a tumor is eventually going to be metastatic at time of presentation. And so in the clinical annotation file, there are only about two or three metastatic patients. But you do see some of these patients eventually go on to die of metastatic uveal melanoma. So it's kind of unclear which of these patients are actually metastatic. So have you compared your classification versus the gene signature that's currently in the clinic, and does it do better? So I haven't checked that, but the mRNA group, I believe, has checked that, and it looks pretty similar to what they see. And so just looking at the third bar from the right, those overlap very well with the somatic copy number alteration clusters. And finally, you mentioned that the GNAQ and GNA11, there were a few co-occurring mutations. Did you notice anything with regards to the type of mutations? Were they the non-hot spot mutations, or did those patients do worse? Do you know? I don't know that off the top of my head. Thank you, Julian. I'm going to introduce our last speaker for this morning, Nicholas Straski. And from Blueprint Medicine, he's going to talk about the landscape of driver kindness fusions in cancer. So a couple of announcements. We are going to have lunch after this talk, and the poster viewing will start at 1.30. We will come back to this room at hopefully 2.25.