 Okay. Well, I'd like to thank the organizers for the opportunity to present on behalf of our UC Santa Cruz and Buck Institute G-DAC, and I'm also presenting on behalf of the first author of this study, a scientist in my group by the name of Christina Yao, who couldn't attend. Clinical breast cancer physicians actually have been long subsetting breast cancer phenotyping it by virtue of the receptors, the estrogen and progesterone receptor denoted as HR here or in tumors that are HR negative either as triple negative or HER2 positive. And we do this for predictive reasons because we have targeted therapeutics that treat patients that are HR positive or HER2 positive, but we only have empiric chemotherapy to treat those that are HR negative. Well, since the seminal work of Sorley, Peru, and colleagues in 2001 who took the transcriptome of breast cancers and came up with intrinsic subtypes as you see the five different subtypes here, we see the importance of this is that it's prognostic that we have two estrogen receptor positive groups that are in the luminal A and luminal B categories that have very different outcomes. Luminal A has a very good outcome as shown by the top Kaplan-Meier curve, luminal B the poor outcome, HER2 positive and basal poor outcome as well, but clinicians do not use this because it still requires fresh or fresh frozen tissues to do a microarray or RNA-seq type studies to do this intrinsic classification. So biological and clinical heterogeneia breast cancer is most evident by these transcriptome analyses and the intrinsic subtypes and with a 50 gene classifier we can now do this known as a PAM 50 and in fact the TCGA data is now annotated both for the clinical subtypes, the receptors as well as the PAM 50 calls of these five intrinsic types. With the exception of the HER2 subtype however, the pathways and signaling properties drive the other three major subtypes are really still unknown. These are the baseloid, the luminal A and B. Of course we know luminal A and B have estrogen receptor but we don't know why the outcome is so very different between those two. So we wanted to search for pathway differences that might be able to discriminate between these three intrinsic subtypes and we employed the pathway inference tool that you've heard about this morning and yesterday known as Paradigm. This obviously integrates DNA copy number and transcriptome data onto over 500, in fact 508 TCGA breast cancer samples that I'm going to describe here. Now Josh is giving you a very good overview of Paradigm and also the novel applications of this. We're using the most recent version that's actually now present in Firehose. It's a super pathway, curated 1,300 different pathways with about 16,000 different features. Features represent either proteins, protein complexes or cell outcomes such as G2M transition or ribosome biogenesis. And you get then a heat map as Josh pointed out before where the samples are on the bottom. The pathway features are on the vertical and you can see the clustering of pathway activities into potential networks. So our workflow is shown here using the copy number changes in the expression data and we then use the inferred pathway activities derived from Paradigm. We use some minimum activity filters to generate from the 16,000 down to 12,000 varying activities. We then used consensus clustering to identify inherent clusters within the path inferred pathway activities and we compare these to the intrinsic subtype calls from the PAM 50. We use both parametric and non-parametric analyses in the ANOVA or Crusco Wallace to look for differences in the pairwise comparisons. And then in the differentially expressed pathway features we use both pathway enrichment and subnetwork analysis of the super pathway where we look for ten or more edges. What you see here is the cluster analysis of the four different feature pathways, inferred pathway activities. There's actually three major ones. There's one very small one here which I'm going to ignore because of its size but did come out in every significance test. So pathway one however you see this light blue one and you see now the clinical annotations for HER2 estrogen receptor status and the PAM 50 calls. What's obviously in the HER, in the estrogen receptor where its negative is dark purple, you see all the ER negative tumors are over on the left side in this dark blue and you see that the HER2 positive tumors are scattered throughout and you see as well that the PAM 50 calls are enriched so that this cluster three down here is largely basal. Cluster two here is a largely luminal B and cluster one there is largely luminal A. So the enrichment is shown here in these diagrams. When I show you the entire heat map, again with those enriched activities present, we can annotate the heat map and show genes that are CMIC repressed as well as those that are CMIC activated of interest. The CMIC repressed and activated ones indicate that CMIC levels are very low in this cluster one which is the luminal A and in fact two examples of the CMIC repressed genes that are elevated in the luminal A represent the two microRNAs 146 and let seven which actually serve to down regulate invasion and growth pathways and breast cancer in these have been well validated. Fox A1 and estrogen receptor activities are seen to be strong in the two luminals and absent in the basal or the cluster three and you see the HIF1 alpha and integrins are strong in the basal or cluster three as well as co-associating with MIC and Fox M1 or Aurora kinase B and polokinase III genes you've heard about before but these are actual pathways associated or hubs associated with these pathways that are higher in the cluster three or basal like. So there's the enrichment shown in those three clusters. When we did pathway enrichment by ease score we actually found common differences that were shared by the luminal A and B that differentiated them from the basal like breast cancer and whereas the heat map had indicated Fox A1 and estrogen receptor and HIF1 alpha as being higher in the basals and different from both luminal subtypes the pathway enrichment score also identified gleamated hedgehog signaling and HDAC signaling 1 which are differentially represented between these luminal and basal subtypes. Now when we specifically look at a luminal versus a basal I'm going to show you luminal A versus basal but the virtually identical with some minor changes between when I look at luminal B versus basal you see a strong enrichment and in other words blue activity is higher in luminal A red activity is lower in luminal A or higher in basal you see that the HIF1 alpha or aryl hydrocarbon nuclear translocator hub is very high in basal and you see the estrogen receptor as you would expect and it's complex in interaction with Fox A1 is very high in the luminal. When we look at luminal A versus luminal B a very interesting division because we know that the outcomes are very different at least historically but we don't really understand the mechanistic differences we see that the luminal B's are very rich in activities with mcmax complex as well as the MIB complex in addition if we look at the luminal A versus luminal B we also see Fox M1 and pololite kinase 1 activities are enriched in the luminal B. So what was interesting from the TCGA data that confirmed historical data and probably is the only evidence of survival differences that we've seen so far in the 508 breast cancers from the TCGA is the significant difference between luminal A and luminal B in the outcome now these cases would be uniformly treated with a hormonal therapy so I and that's the most effective therapy in these two subtypes so I think these are real significant differences. What we look to see is whether or not the activity hubs that we found differentiating luminal A and B could also be used to distinguish overall survival and in fact they could as you see here the high mcmax complex is associated when you lump luminal A's and B's together with worse outcome high Fox M1 is also associated with worse outcome when you lump them together and you look at the composition of the green and red survival curves they're actually mixed with regard to luminal A and B status we didn't have enough cases in order to do a multivariate analysis that showed any significance for luminal status versus mcmax versus Fox M1 but it looks like the Fox M1 and the mcmax actually represent reasonable surrogates to differentiate survival between the luminal subsets. So in conclusion we've used unsupervised consensus clustering based on paradigm features inferred pathway activities showing significant associations and enrichments with the intrinsic breast cancer subtypes. In pairwise comparisons between the intrinsic breast cancer subtypes identified Fox A1 estrogen receptor and lower HIF1 alpha pathway activities as being shared in the luminal versus the basal or you could say the other way around the basals have low Fox A1 ER and high HIF1 alpha ARNT relative to the luminals this pathway activity also identified mcmax Fox M1 MIB and polo like kinase one is network hubs with elevated activities in the luminal B versus the luminal A to these show comparable prognostic value and survival associations in luminal as we see with luminal status and this super pathway analysis may therefore help to identify some pathways and signaling features between clinical intrinsic breast cancer subtypes and ultimately point to subtype specific therapeutic strategies I would point out that there are clinical studies in progress with polo like kinase one inhibitors as well there are newly identified chemical entities that are quite specific for Fox M1 so these would be strategies that one might pursue to try to get a handle on the treatment of this poor outcome luminal B type subset as well we have therapies that can attack the angiogenic hypoxic and oxidative stress associated network that seems to be coming up in the basal like breast cancers so I don't thank you for your attention and acknowledge the the UC Santa Cruz team as well as my buck co-investigators our breast analysis working group led by Chuck Peru and the entire organization here for everything that's allowed us to do this very comprehensive analysis thank you questions for dr. Benz very nice presentation question the luminal B's are they all hormone receptor positive and her to positive or are there subsets within there that might have something to do with the progesterone receptor and or proliferation probably not with regard to progesterone receptor but probably I mean definitely with regard to proliferation in fact the pathologic poor surrogate for a luminal B intrinsic subtyping is Ki67 and we and clinically the clinicians will use a Ki67 value of over 11 or 12 percent to designate a potential luminal B or poor outcome patient but what I'd like to point out is that based on this analysis I don't think proliferation is the only thing that's really being driven in this pathway you know so I I think that we need more clues as to to get handles on on how to therapeutically attack this subset of tumors I had one so there's been some discussion in the literature about autophagy and its linkage to poor prognosis in breast cancer I was wondering whether in your analysis you saw any evidence of autophagy genes well it didn't come out as a feature but I will say that our analysis was filtered so that we we looked for features that had a 10 or more interconnections so some of the things in the super pathway that are out perhaps not as well annotated pathways you know may still reveal these kinds of things and obviously as you heard this morning paradigms a sort of a work in progress and at this present time with you know has about 1300 different curated pathways involved in it but that's growing so hopefully we'll get more answers like that okay good thanks very much please join me in thanking the organizers and the speakers for the morning session uh coffee break I think