 Thank you for the introduction, and thank you to the selection committee for giving me this opportunity to present a progress report on the FFPE pilot study that we have done through TCGA. For the talk today, I will cover three main things. First will be the co-isolation of nucleic acids from FFPE, some of the preliminary results that we've obtained through the genomic and epigenomic characterization of those analytes, and then wrap up with some conclusions and future plans. In addition to serving primarily as the technical director of the molecular group at the BCR, it's also my great pleasure to serve as a co-chair for the analysis working group leading this effort, and so I wanted to start with this slide to give credit for all of the individuals who are contributing to this. It is truly a classic TCGA approach to new concept discovery. Before I begin, I basically just wanted to frame the context of this pilot and emphasize, or maybe just reiterate as we all know, that massively parallel sequencing has resulted in major advances in our understanding of tumor biology. Many of the presentations that we've seen this morning in fact speak to that, where new drivers of cancer tumor progression are being discovered, new therapeutic targets are identified, and even molecular taxonomies for cancer classification are being identified. I think it is important to note that many of these seminal studies have drawn from frozen biospecimens and the platforms that were used to generate those results were optimized for those frozen tissues. So it is the application of what is being learned through these research projects to the clinical environment that is commonly known as precision medicine, or as Dr. Mongol referred to it, personalized medicine. So one of the challenges in making that translation is that diagnostic specimens that are typically available are preserved as formal and fixed paraffin embedded samples, and it is fairly well known that that preservation method introduces many molecular artifacts. So the goals for this FFP pilot were to first identify and optimize the best practices for the extraction, characterization, and analysis of FFP samples, and then our goal is to define the patterns of artifact introduced by that fixation process, and then hopefully in the end contribute to bridging the gap to this diagnostic material and facilitate the application of the emerging cancer taxonomy to clinical testing environment. So from there I will begin with our results from co-isolation of DNA and RNA from FFP. So this pilot study actually started quite a while ago when the TCGA program office asked the BCR to identify a method of extracting DNA and RNA from FFP tissue, and they asked us to find an optimal method. And so from the BCR's point of view, an optimal method is one that maximizes yield and integrity with the hope that it provides consistent molecular characterization of the platforms outside of our domain. So integrity is measured at the BCR by DNA gel electrophoresis, and what we would expect to see in an intact sample is a high molecular genomic band from DNA and from RNA high molecular weight discrete peaks for the ribosomal subunits on a RNA bioanalyzer trace. So in fact this is what we see from control tissues that we used in this pilot study where we procured fresh tissue, split it in half, immediately froze one portion, and optimally paraffin embedded, formal and fixed and paraffin embedded the other half. It was that FFPE half that we then used to survey different commercially available FFPE extraction methods. So what we very quickly found was that there was no single off-the-shelf method that gave optimal DNA and RNA integrity. So as you can see here from these FFPE control tissues we obtained nice DNA integrity, but the RNA was very much degraded. Conversely we found other methods that didn't perform quite as well from a DNA integrity perspective, but the RNA integrity was much better. Still not up to the standard of frozen, but from FFPE this was perceived to be a pretty strong signal. So from there we undertook a task of customizing the methods that were available commercially to try and improve the integrity of the DNA in the context of improved RNA integrity. And what we ended up with in the end is this TCGA optimized method, where as I mentioned DNA is maximally, the molecular DNA is maximized and the molecular weight distribution of the RNA is also shifted to the trend of less degraded RNA. So here basically just briefly want to lay out what this method looks like, not dwell very much on it other than to say this technique utilizes scrolls, not cores, and we normalize the input surface area for each extraction, not necessarily the number of scrolls that go in. And to point out that after the key point here is that the actual lysis step results in a supernatant that both the DNA and the RNA are further extracted from. So with this extraction method identified then we turned our attention towards patient samples and there were basically three criteria for inclusion in this pilot study. First was the, there needed to be FFPE material available from a TCGA patient and that TCGA patient needed to previously qualified and shipped. And then lastly there needed to be enough residual analyte from those frozen extractions from the previous shipment so that we could send out a set of samples, a cohort that was frozen and FFPE matched patient samples. So after some, after a selection of patients we ended up with 38 patients in the study distributed across six different tumor types and there's quite a bit of information on the slide but I think the only thing I really want to emphasize two things is that the average age of these tissue blocks was fairly uniform just shy of three years old at the time of extraction and that consistent with what we see from frozen extractions the DNA yield from FFPE is also less than RNA and use this point to illustrate the fact that it typically required more than one extraction to meet the yield requirements for distribution and characterization. So all of the data that we have is from multiple pooled analytes prior to shipment. So then if we talk about the distribution of this I'm summarizing here the five platforms that paired tumor and FFPE analytes were sent to and when we distributed these it was DNA and RNA from FFPE tumor DNA and RNA from frozen tumor as well as a germline normal and we sent them to as many places as we had yield to send and there were some that were rate limiting and so that is the explanation for the gap. So at this point then we can start to turn to the results. The first platform that I will show you today is the results from the SNP6 array and we were not very surprised I would say to find that the SNP6 array did not perform very well with FFPE or FFPE did not perform very well on the SNP6 array and in fact zero of the SNP6 arrays passed the QC pipeline in place at the Broad and this was in large part due to the highly over segmented copy number profile and it's illustrated in this figure here where the fresh frozen samples are very tight, excuse me tight low segmentation but the FFPE was hyper segmented and spread out. When put into a cluster analysis FFPE was still able to validate copy number alterations in frozen but if you did not have the frozen reference to map to you found that a there was a high rate of false discovery I guess when you when you mapped it to the frozen you found that FFPE had far more copy number alterations suggesting there's a high rate of false discovery so the conclusion from this was that segmentation artifacts compromise the standalone utility for determining copy number alterations of FFPE through SNP6 array. So the next platform then would be exome sequencing and so I am showing you today results from one center and one of the studies and shown here is a Lego plot that represents the rate of different types of single nucleotide transitions found in a tumor sample relative to the germline control and so essentially what you see on the Z-axis is the rate of transitions per megabase and this is essentially should be viewed as a profile of what the Luad tumor type looks like in this view and from our analysts we heard that this picture is consistent with what was previously found through the marker paper for the Luad study. Now the interesting thing is when we when we compare this picture from the frozen to the FFPE we see a profound enrichment in CDT transitions in the FFPE sample and that's a little hard to see but the scale here is actually doubled and so the overall rate or distribution of other transitions is similar between frozen and FFPE it is that we have a significant enrichment in the CDT. So to gain a little more information in about those CDT transitions we bend the transitions by allele frequency and we saw interestingly that the CDT artifact or CDT signature was really confined to the low allele fraction compartment. So when we look at a low allele compartment above 10% we see now the profile of frozen to FFPE is very similar between the two is a comparable signature and then in this low allele fraction is where the CDT artifact segregates. So these results support the use of FFPE for exome sequencing however the additional tools are needed to compensate for this artifact. So the next platform then that I'll present is the mRNA sequencing results and the one of the methods that we use to evaluate the comparability of RNA sequencing between frozen and FFPE samples was through this pairwise Pearson correlation and essentially the takeaway message is there's fairly high correlation between transcript abundance in the frozen and FFPE samples. The lowest average was a 0.85 and the point of reference here is set of technical replicates run on the same platform at UNC. So overall comparable when we do another method of looking at this unsupervised clustering we see that the dominant clustering factor in the RNA-seq data is the tissue type and then within that we see that patients for some of the studies segregate out together suggesting that the effect of FFPE is very low in that in those studies so bladder and endometrial. However in other studies such as Luad and colon rectal we see that the effect of preservation method is much more profound suggesting that there may be some differences between the studies in the influence of FFPE fixation. So if we wanted we wanted to look a little bit more closely at that so the next the way that we did that was to isolate out only the significantly differentially regulated transcripts and so this is a cluster analysis comparing the differences between fresh frozen and FFPE and importantly what we saw was in this cluster analysis there was no dominant effect of tissue type in this we see a reasonable distribution of tissues across but we do see this this fairly profound and somewhat scary if we are approaching a pilot to use FFPE for precision medicine there are some robust differences between between these transcripts but I think the encouraging thing from this is when we drill down and look at the individual patients and then plot the relative transcript abundance of fresh frozen against its paired FFPE we see a fairly linear relationship between the transcription levels or the transcription abundance suggesting that there's a uniform over or under detection within these specimens and it is not a transcript specific effect. So before I get run off the stage then I'll go through the micro RNA sequencing results very quickly and and save with within this platform we used a principal component analysis to evaluate the the effect of FFPE and what we've found in this principal component analysis was that 70% of the variance that could be detected in these samples is restricted to principal component one and that the effect of FFPE through this analysis does not become apparent until principal component five where you can begin to separate out the frozen versus FFPE samples so this suggests in the micro RNA context there's very weak contribution of preservation method on the outcome and just briefly one of the interesting things that we've we've come across so far is that the the number of micro RNA that are significantly regulated in FFPE or at least significantly detected to be different in FFPE is far more diverse than what we find in frozen. So here the micro RNA contributing to cluster formation in frozen here the micro RNA contributing to cluster formation in FFPE so we see greater diversity all micro RNA up that are present here or present down here is highlighted in yellow and then when we map the frozen to the FFPE we see that frozen patient samples map to similar cluster as the FFPE for the most part so the orange and green you see that very easily but then for for other patient samples there is some strain from the original cluster so overall FFPE has a weak effect on micro RNA characterization however there is quite a bit more work we need to do to look into the the cause and effect of the increased micro RNA diversity and so and then just just lastly I can very simply sum up the methylation platform we found very little effect of FFPE on methylation studies and so if we just just go right to the pairwise Pearson correlation coefficients this these are representative of what we see and so with DNA methylation there is very high concordance between frozen and FFPE tumor specimens but we need to declare that this is very much dependent of or at least requires the Illumina FFPE restoration protocol as that was best practice already in use when we ran this pilot study so to to close I've shown you results on our optimization of nucleic acid co-isolation we've found the DNA and RNA extracted from FFPE can be employed for multiple state-of-the-art platforms and then to just rehash the slides that I've kind of zipped through SNP6 array had a high false discovery rate exomes were interpretable with artifact caveats methylation worked very well assuming you're using this Illumina FFPE restoration protocol mRNA seek there was very good correlation between FFPE and frozen samples however there is some differential detection of transcripts that needs to be evaluated further and also that this platform utilize the riboseurochemistry and then lastly the mRNA seek systematic increase in the diversity of micro RNA species from FFPE so in conclusion with 18 seconds blinking red we intend to further this work by analyzing the signature of FFPE through multicenter calling and that work is underway right now we think that will enhance the robustness of our ability to characterize this effect we're also working to delineate the influence of tumor heterogeneity in this comparison that we are performing so again this is frozen specimen compared to FFPE from different areas of the tumor and literally before I walked up here I received the the Lego plots from from that heterogeneity analysis specifically looking for CDT artifact in our luad samples in the frozen adjacent portions and we do not see that artifact suggesting that that the results that I've shown today are more likely to be attributed to the effect of FFPE than a spatial heterogeneity difference and then finally we intend to do deeper analysis of the differences between these two preservation methods so that we may identify bioinformatic mechanisms to correct for these artifacts that's it happy to take any questions and the full-time interest let us give the question for this so if your question can discuss later yeah so so as you know the ps3k kindness is one of the most mutated gene across all different cancer type so our next week is Chris Benz from Bach Institute for Research on Aging he will talk about the domain specific ps3k mutation affects different pathway activity across more than 3,000 pancancer