 So, I'd like to thank the organizers for inviting me to come out and give this talk. I think we have some really exciting data tying P53 mutation to heterozygous loss of 3P chromosome lumbarum. So, really briefly, head and neck cancer represents about a half a million new cases a year, but its incidence is increasing due to risk factors such as HPV infection and alcohol and tobacco use. So, we're starting to understand a little bit of the molecular basis of the disease. Outcomes still aren't very good with limited targeted therapies and only about 40 percent five-year survival. So, this study kind of started with that aim to really gain a better understanding of the disease. It's spurred on by our clinical collaborators. And from the methods side, we were really looking for ways to develop new methods to integrate data across platforms and ultimately to try and isolate genetic interactions from a cancer cohort. So, I'm going to go pretty quickly through this to try and stay in time, but feel free to find me at my poster tomorrow. I can go into more of the details. We've tried to make this analysis as transparent as possible, so the whole reproducible pipeline that will be available online, everything from downloading the data to a generation of figures and statistics as soon as this paper reaches publication. So, this head and neck cohort is very diverse and to try and homogenize it for the kind of analysis we're doing. We limited our patients to those under 85, and for the majority of the analysis, we're only looking at patients that don't have presence of HPV. So, the study design, again, I'm going to go pretty quickly, but feel free to find me at my poster. We started out kind of top down on by screen, and we wanted to try and isolate, you know, one of the core sort of candidate biomarkers that we can find in this dataset. And so, this included frequently mutated genes, differentially expressed mRNA and microRNA, and kind of recurrently altered copy number events, as well as clinical variables. So, after identifying these events, we just ran a basic screen to try and find markers that correlated with patient outcomes, and we identified about 60 of these. And then kind of, we went another step to finally look for associations among pairs of these prognostic biomarkers. And so, as you can see here, P53 is mutated in about 200 of these patients. We have heterozygous loss of the 3-P arm and another in 200 patients as well. And if you look at the overlap of these two patients, this occurs in about 70 percent of our patient cohort, and this is much higher than you would expect by the marginal probabilities. And so, at this point, we decided that this was kind of a good interaction to go after, because we know that had in the cancer, P53 has well been understood as a kind of prognostic biomarker, as well as an indicator of, you know, aggressive disease in this tissue. But what hasn't been known is that this 3-P deletion is very commonly co-occurring with us, and this happens in 70 percent of this patient cohort. So, one of the really nice things about TCGA is, while we were undergoing this study, another, you know, another set of patients were sequenced. And so, we were able to, you know, train our, find our associations in this discovery cohort, but then validate them in 111, and now I think it's up to 170 new recent TCGA patients, and indeed we see actually even higher effect size within this validation cohort of this co-occurrence. So, we implicated each of these genes based on their association with prognosis. So, the next question we asked is, you know, what's going on? Is the prognostic effect additive between the two, or do we see some sort of interaction? And indeed, while there is a small effect of each individually, the majority of the prognostic effect is mediated by the patients that have both of these events together, such that P53 mutant patients that don't have the deletion have actually pretty decent outcomes with respect to this cohort. And so, this was an interesting finding. Unfortunately, our TCGA cohort doesn't have updated clinical follow-up for these recent samples, but we were able, we were able to collaborate with Jen Grandis and Aguilof at University of Pittsburgh. And this is kind of one of the early exome sequencing studies that was published in 2011, and since then they've kept really good clinical follow-up for these patients, and we were able to recapitulate this finding that within these P53 mutations, the addition of the 3P loss does confer a worse prognosis. So I'm sure a lot of you are thinking right now that really what we're seeing is this link between P53 mutation and chromosome listability that's been described very often, and this is what we thought originally too, but when we test for the co-occurrence, we see that the co-occurrence of P53 with the 3P loss is much more frequent than any other chromosomal alterations, including alterations that are more highly deleted or amplified in the cohort. In addition, we kind of did this experiment to construct a very simple multivariate model. And so what we can see here, these are the hazard ratios in a multivariate model with these two, with the 3P loss in a general chromosomal instability variable. And what we can see is that when you take 3P into account, there is no additional effect of chromosomal instability. This result recapitulates in P53 mutated patients, but when we look at the wild type patients, we see kind of a reversal of this trend. So the kind of prognostic trend of 3P is actually a general trend with chromosomal stability, so there's no additional effect of 3P in these patients. And this kind of gives us further evidence for an interaction effect when we look at these events with respect to prognosis. So far, we've only been looking at these HPV positive patients, but we know that HPV negative patients, I'm sorry, but we know that part of the mechanism of action of HPV is that it integrates into the host genome, and the E6 viral protein binds P53, leading to its degradation. So then we tested what is there an effect of 3P in these HPV positive patients, which we know have P53 inactivated. And indeed, we do see a relatively large prognostic effect in our cohort of HPV positive patients. So another question we asked is we've done this analysis and had a neck cancer so far, but we know that a lot of patients have P53 mutation, and a lot of patients have this loss to this, this heterozygous loss to the chromosome alarm. So could we generalize these results? And for this, we did a pan-cancer survival analysis. And off the bat, you can see that the survival occurs for these different tissues are very heterogeneous. So to try and homogenize a little bit, we only looked at non-matostatic patients. We didn't look at patients such as glial blastoma, which had very poor outcomes for thyroid cancer, prostate cancer, which had very good outcomes. And indeed, when we plug it into this patient, this cohort of about 4,000 patients, we do see a clear effect of the combination of these events versus each individually. And this holds, even in a multivariate model, including tissue types, stage and age. So we think that there might be a broader, this could be a broader signal, at least within subsets of other tissues. So we've kind of established this combination of events and having that cancer. And so the next question we ask is, given this clear subtyping, can we further stratify the patient cohort? So for this, we repeated our screen. So we, from feature construction to this prognostic screen. And we see that amongst this poorly performing group, these patients with the combination of events, the presence of this microRNA, MIR548K, leads to even worse prognosis within those patients. So this microRNA is very interesting. So it's not expressed in normal tissue, and it is induced in at least a fraction of tumors. And this is how it got characterized as sort of a candidate biomarker. It's located at a focal copy number amplification and the expression trends with the copy number gains. But when we look at the prognostic effect, it segregates only with expression. So patients that have the amplification but don't have expression. So that would be these patients down here. They actually are in the well-performing group. And so while we were working on this, we got really excited because another group working in esophageal squamous cancer recently implicated this gene from a similar unbiased screen. And they did some really nice functional work in cell line showing that this does indeed have oncogenic potential. This paper was published in Nature I think two weeks ago and it's a really good read. So we think that this microRNA could have broader implications for at least a variety of squamous cell carcinomas. So finally, we've talked a lot about these patients with this combination of events, but there are about 72 patients that don't have these events going on that have relatively good outcomes. So can we look into those patients and see what's going on with them? And so for this, we conducted a secondary association screen. We didn't really have enough data to look at prognosis, but we can look for events that are specific to these 72 patients. And indeed, we do find that these patients are rich for caspase-8 mutations, which we know have a role in apoptosis, as well as RAS signaling pathway mutations, including a number of atrascana function mutations. And again, these results validate in the recent TCGA patient cohorts with even a larger effect size. So in conclusion, we've shown that P53 mutation is frequently co-occurring with 3P loss in the head and neck cohort. This leads to worse prognosis. And we show that in a couple of different cohorts, as well as a pancancer cohort of about 4,000 patients. Amongst these patients, we find that the expression of MyR548K leads to worse prognosis. While in the absence of this kind of common driver event, we see other pathways that may be important drivers in this tissue. So with that, I'd like to acknowledge this was really a team effort. We had a lot of clinical collaborators that, you know, a lot of people contributed good insight, including Scott, Litman, Quinn, Wen, Ezra, Cone, and Ryan Orozco. Jen, Grannis, and Ed Nugiloff who contributed the validation cohort, Neil Hayes really helped us get through this TCGA data and understand the inner workings. This is the whole Idaker lab and specifically the cancer team each contributed to this project. So, Sal Shari, JP Chen, Matan, Hoffry, Hanna Carter, and Trey. So with that, I'll take questions. We have time for questions. Rachel? I'm wondering if you looked at the difference between disruptive mutation and P53 and non-disruptive. Yeah, so we actually, we did, and what we find is that while disruptive mutations did have worse prognosis, the non-disruptive was significantly different from wild type as well. And so we weren't able to get down to that granularity into this pipeline. And we really did try and make this an unbiased screen. So injecting that kind of prior information that you have for P53 kind of gives it an unfair advantage against all the other it's significantly mutated genes. You have some parity in your one of the slides. Can you elaborate on, do you see this in lung squamous cancer or? So there's not a lot of data for lung squamous. We do see it in lung adeno. We think that there's an effect. And in the pancancer analysis, we see small effects in a large number of cancers. So it's really hard to pull out exactly where we see the effect in the pancancer context. I had a question to follow up on Rachel's actually really quick. How did you define when you did the check on P53 damaging versus non-damaging? How do you define those classes? So we ran a few analyses on this. We looked at different functional classification. So we looked at truncating mutations to the binding domains. And there's been some published work that have built sort of a classifier as disruptive mutations. So these would be mutations to the binding domains that kind of have polarity change or something like that, or truncating mutations. And so we did that as well. Well, that does add a small amount of information that would induce too many false negatives into the analysis. I had one other question. Did you guys do the reverse take 3P loss and look for other mutants that also interact with 3P loss events as far as their prediction of survival? Besides the HPV cases. Oh yeah. So by far the strongest signal in this cohort is this interaction of P53 and 3P. And generally anything else, any other correlations we can find can be explained away by this interaction. So it's kind of such a big, it's kind of the elephant in the room. And so anything else? Samples to check other. Yeah, yeah, to really get any further, it's pretty hard to overcome this relatively large signal. Did you try to go after specific genes on 3P? I mean, put in a different way, do you have an idea of what are the genes individually participating in this cooperation? Yeah, so this is, we would love to do this, but in general we see kind of, most patients have a full or the majority of the arm lost. And we don't really see peaks. There is a slightly higher fraction that have lost at a fragile site at 3P14. But we can't really break it down any further than that. Great, let's thank Andrew one more time and all the speakers for the session that rounds us up. Matt, Matt had an announcement I think before we adjourn, so let's wait for Matt. All right, so in addition to thanking the speakers, I just want to thank the session chairs, Han Liang this morning and Josh for this afternoon for a great leadership of the sessions. And then two announcements. First, the workshops begin at four o'clock. They're all in the schedule packet and they're running twice, both at four o'clock and five o'clock. And the second announcement, there are several cancer type analysis working groups that will be this evening. If you've registered for one of those working groups, Margie Sheth was saying that everyone should meet in the cafeteria at 545 and she will direct you to the rooms for each of those. And that's all, yeah, we're adjourned, yeah.