 I thank the organizers for asking for an oral presentation of our abstract and I want to point out that this is the work of senior scientists at our institute, Dr. Christina Yau who was unable to attend. I do also want to point out that this afternoon you'll hear more from Josh Stewart who co-led the PENCAN analysis. It was the freezing of that data set on over 3,000 samples that enabled this kind of pathway analysis. So most people are familiar with the phosphatidinositol-3 kinases, a very large family, heterodimers between regulatory subunits, P85 alpha is most commonly known but there are other four others and the catalytic unit, the P110 alpha which is encoded by the PIC3CA gene which is very commonly mutated in human cancers as you heard. Modular domains of this PIC3CA gene are shown here. There's five of them. It's frequently mutated in multiple types of cancers I pointed out but there are hot spot mutations in both the kinase and helical domains that we'll talk about. These are thought to be early and possibly initiating events and malignancies like breast cancer. There's a notable absence of mutations in the RAS binding domain RBD and also within the core catalytic site of the kinase domain and that's seen across cancers. Now preclinical evidence which led to the rationale for doing this study indicates that PIC3CA mutations are activating and gain a function but domain-specific mutations may have different downstream effects in particular phosphorylation and activation of AKT and other downstream pathways including mTOR and also phenotypic consequences but these are preclinical studies in model systems. So the TCGA pan-cancer dataset is big enough to offer an opportunity to ask questions about possible pathway differences linked to domain-specific PIC3CA mutations in different tumor types. Are there common domain-specific pathway activities across the different tumors or are these cancer types specific? So in terms of pathways we're going to use an algorithm known as Paradigm that's been used in most of the marker papers of TCGA briefly. This integrates RNA gene expression data with DNA copy number data. Superimposes this on a super pathway to infer activities which refer to as inferred pathway links or IPLs and encompassing over 13,000 pathway features which can be displayed in a heat map or analyzed for differences between tumor subsets. Now the pan-can dataset characteristics that were used here and these were the frozen data sets employing the platforms including exon sequencing RNA expression and DNA copy number. We had over 3,500 samples that had the paradigm analysis in other words that we had IPL values. We had over 3,200 samples with exon sequencing data so we have an overlap set shown in the VEM diagram of 2637 with both IPL and PIC3CA mutation data. As you can see in the pan-can histograms here that the cancer type distribution among the valuable samples is very different. That's actually reflected a lot in in the overall TCGA cruel of these samples. So that's going to have to be taken into account. Now somatic PIC3CA mutations have an overall mutation frequency in this pan-can dataset of 22 percent that's 572 out of the 2637 cases but the frequency per tumor type is quite variable as you can see from 0 percent as seen in AMLs leukemia to greater than 50 percent as seen in uterine endometrial cancer with the intermediate values as shown there. Now in order to look at domain specific mutations we also filtered so that we would use only nonsense mutations because we're interested in coding gain of function abnormalities and also mutations that occurred in only one domain as you can see in nearly 100 of the cases there were multiple mutations across PIC3CA. So we ended up with a set of 447 cases that we could evaluate and you can see we eliminating the AML we have a very different again we have an uneven distribution across the tumor types that's going to have to be accounted for. The domain specific PIC3CA mutations are shown here as you can see there's a preference for the helical and kinase domains and you can see also the hotspots in particular specific mutations histidine to arginine at 1047 in the kinase domain glutamate to lysine at 545 and at 542 so the helical domain has two nearly adjacent hotspots and the kinase domain has a very common hotspot at one spot. So 83% of the cases had either helical or kinase domain mutations and 64% of these were at either of these three hotspots as shown. So the domain distribution of the mutations by cancer type was significantly different color coded here the helical domain is an orange and the kinase domain is in red and that's most common as shown and across the different tumor types and as an example again of this variation we see in breast cancer seems to prefer kinase domain mutations with 52% of those being in the kinase domain whereas in squamous cell of the head and neck and lung there's a preference for helical domain mutations 65% of those being in that domain and only about 20% in the kinase domain. So how did we look at pathway activation in cup N and control for or just for cancer type? Well we use the IPLs and identified significant IPLs for each domain versus all others by logistic regression and we adjusted for the cancer type with wall test P.05. The kinase domain ended up having 711 significantly different IPLs the helical domain over a thousand and amazingly these did not overlap much there were actually 1500 that were non-overlapping IPL features between the two kinase and helical domains. We then determined in pathway enrichment among the different IPLs with an FDR corrected e-score and we visualized by side escape. So as a kind of proof of consequence we first went through to the PI3 complex itself and here you see a color coding that indicates if it's very dark red this is a kinase domain positive association and a reciprocal helical domain negative relationship and if you have lighter shades of red increasing the pink this is again kinase domain positive blue is more helical domain positive or kinase domain negative for green but here what you see is in the PI3 kinase catalytic subunit itself in the center here where it's the red areas that the kinase domain mutations most strongly associate with the super pathway hub showing the PI3k catalytic subunit activation and this hub is negatively associated with the helical domain mutations and this is in fact consistent with the preclinical evidence which is why we looked look to that to begin with but then asking what was distinguishing the kinase domain mutations from the helical domain mutations in in other areas we first turned to cell cycle and proliferation activities and these were featured by the poli kinase 1 and FoxM1 hubs which were most strongly associated with kinase domain mutations you can see red and pink virtually everywhere and negatively associated with helical domain mutations using the same color coding as before and then if we go to helical domain mutations we see very different we see the ROGTPA's families and also gap junction degradation as hubs and pathway enrichments that were most strongly associated with the helical domain mutations and negatively associated with kinase domain mutations so although there's these are common across different cancer types in the TCGA pan cancer data set missense pic3ca mutations distribute very differently with respect to total mutation frequency and domain specificity examples being breast cancer with greater than 50 percent in the kinase mutants whereas squamous cell head and neck and lung less than 25 percent opposite sort of relationships with regard to the helical domain mutations the kinase domain mutations appear to be linked more strongly with pathway features like FoxM1 and poli kinase 1 that enable cell proliferation while the helical domain mutations appear to be linked more strongly with features enabling cell migration and dissemination such as the ROGTPA's now functional studies are clearly needed to confirm these findings including the suggestion that breast cancers preferentially mutate pic3ca to drive cell proliferation while lung and head and exquamous cancers prefer to helical domain mutations potentially to drive their malignant cell motility but additional comparisons are also needed here to identify potential tumor type specific or context dependent differences between the kinase and helical domain pathway preferences because as we've begun now to do the functional studies we then commit ourselves to a particular tumor type in fact working with the Hopkins group that has made isogenic lines in the mammary cell line MCF10A for these different mutants we need to know what context specific differences there are between the helical and kinase domain mutations as well as general differences across the entire pancan dataset as i've shown you today so i'd like to thank our UC Santa Cruz and Buck Institute GDAC team members again Christina Yao who did all this work our partners at 5.3 who've developed the paradigm and perform that for us and then also a point again to this afternoon session where i think you'll hear more about the pancancer 12 analysis working group and the subtype uh data that's uh in press thank you very much i'll take questions so we have time for a few questions yeah p85 alpha is that the regulatory uh i i guess i'm trying to understand it myself but i think the question is that do we did we see any mutations in the regulatory subunit which would be p85 alpha we did not look specifically for it but i do not believe many were reported if any at all maybe i think i could be corrected by that for someone in the audience that might know better so the uh the other interesting aspect okay that's the p85 that's a p85 we did not specifically look at that but that's an interesting aspect because these mutations that are outside the kinase domain are felt to alter the dependency on that regulatory subunit so when you look at the helico domain and kinase domain mutation and infer their functions for those low frequency mutations that could be bystanders or passenger mutation have you then go back to use those low frequency mutation in the kinase domain or in the helico domain and look at whether they still carry the same signature we didn't because most of the time it's highly the hotspot mutation you skew everything that you're seeing uh i mean the hotspot mutation of the kinase hotspot mutation of the helico domain they are so predominant and so the signature you're seeing could mainly from them i'm wondering in besides doing all the functional studies of the individual mutations can you take the individual mutation and go back to see whether they actually carry the same signature that can infer whether they actually have the functional uh we we could do that we didn't specifically do that no question so we usually see mutations some mutants are co-current pattern or mutually exclusive so have you checked that this co-current or mutually pattern for patient kinase different cross different cancer now we didn't we only looked for the no we didn't control for anything that was concurrent or mutually exclusive with regard to pic 3ca mutations we only looked at that question most broadly thank you so let us move on so now we move to the our last speaker um dr suyuan zheng from the university texas um md anderson cancer center he'll talk about comprehensive molecular profiling of um aderectal casemona um so this is the last talk hopefully serve as a appetizer for our lunch