Scientist Stories: Arul Chinnaiyan, Functional Genomics and Proteomics to Study Cancer

Very good.

So thanks everybody for coming today, and it gives me great pleasure to start off today's session. We'd like to— our first keynote speaker for this morning is Professor Arul Chinnaiyan. It's a great honor for me to introduce him. He is a professor of pathology at the University of Michigan and director for the Center of Translational Pathology. He's also a member of the Howard Hughes Medical Institute and also an American Cancer Society research professor. I followed Arul's career from a distance for many, many years, and I think he really embodies the true clinician-scientist, being able to take the most cutting-edge technologies and really to apply them to uncover biological findings, medical findings of great clinical significance. His 2008 paper in Science on the discovery of the TMPRSS2-ERG fusion gene is a classic in the field. It's a paper that I use when I teach my medical students what true translational cancer research is all about. So, and he still continues to push the field moving next-generation sequencing into the clinic, and I think that you will— I think the students here will find his talk really inspirational. So without further ado, can I please call Arul to the stage?

Thank you, Patrick, for the kind introduction. And I'd like to begin by thanking the organizers, Paul and others, for the invitation here. This is actually my first time to mainland China, and I've been stunned by the amazing amount of progress that I've witnessed and the amount of building in terms of structures, but also in terms of scientific programs that I've at least witnessed at some universities that I've visited in Shanghai. So it's been quite quite eye-opening for me.

So what I plan to talk about today is a sort of a nice follow-up, I would say, to what Charlie presented yesterday.

I think Charlie really focused on some of the challenges as well as pitfalls associated with precision medicine or precision oncology approaches. What I'll present is more on the potential promise of precision medicine and precision oncology moving forward. So I think various aspects of this need to be considered as we move these efforts forward.

And certainly it has attracted a lot of attention in terms of various countries beginning to invest heavily in these areas. You've certainly heard of President Barack Obama's initiative in the U.S. to begin to fund and support precision medicine efforts. From my colleagues in Shanghai, I have heard rumors that even the Chinese government is considering a similar effort in precision medicine, potentially at a scale much greater than even the U.S.

in terms of the number of participants involved. So this is certainly an area of tremendous interest as we move forward in terms of medicine really taking advantage of understanding the molecular landscape of human disease and using that information in a rational fashion to treat disease.

So my laboratory and center at the University of Michigan focuses on diverse areas of interest. Some of which are listed on this slide. But what I'll be focusing on today are these two topics. One is on the role of long non-coding RNAs in cancer.

I noticed that there were certainly a number of posters as well as talks really focused around non-coding RNA, so I thought I would highlight a bit of our work in that area. This is probably an area that is maybe in its infancy in terms of its relevance to precision oncology and precision medicine. Medicine, but I think is really going to be the future as we move forward, at least in biomarker development as well as even in therapeutic targeting, and probably is relevant to the precision cancer biology component of this meeting.

The second—

the rest of my presentation will really be focused on establishing clinical sequencing for use in precision oncology and sort of how we began our efforts in that area and how we've been sort of advancing that moving forward.

So just a little bit about long non-coding RNAs.

Certainly over 98% of the human genome is not expressed or does not encode protein coding genes. So one of the major questions in the field moving forward is what is happening in terms of the non-coding component of the genome.

What are long non-coding RNAs, lncRNAs? They're transcribed by RNA Pol II. They generally have a transcript length of greater than 200 nucleotides. They can have multiple exons and undergo alternative splicings. They can also be polyadenylated. They classically have an epigenetic signature of H3K4 trimethylation at promoters as well as H3K36 trimethylations across the gene body. So they can often look quite similar to protein coding genes. What's interesting about long non-coding RNAs is that subsets of them are oftentimes exquisitely lineage specific.

And what we find is that some of these lincRNAs are even disease specific.

Long non-coding RNAs, unlike, for example, microRNAs, which you've probably heard more about, The functional relevance of these lncRNAs are much more mysterious, I would say.

Certainly a lot has been worked out in terms of the biology of how microRNAs work.

But long non-coding RNAs have diverse mechanisms potentially of function, some of which I've listed here. For example, physically interacting with epigenetic complexes, enhancer RNAs to tumor suppressor signaling, RNA processing, RNA-RNA interactions, as well as in potentially microRNA sequestration, essentially functioning like sponges. They've been linked to normal biology as well as disease biology. For example, the Xist long non-coding RNA has been implicated in X chromosome inactivation. Other long non-coding RNAs have been linked to stem cell maintenance, differentiation, and specification into the various tissue lineages. They've even been linked to various human diseases including Alzheimer's disease and what I'll be talking about primarily is cancer. So the long non-coding RNA that's probably received the most notoriety in terms of cancer has been the lincRNA called hotAIR, which is characterized by Howard Chang and John Rinn. This lincRNA was identified as a lncRNA that interacts with the PRC2 epigenetic complex as well as other epigenetic complexes. Hot air has been shown to be associated with poor prognosis. So high levels of hot air in breast cancer are associated again with a more aggressive disease relative to low levels of hot air. And our work really stimulated by those early findings in terms of long non-coding RNAs in breast cancer was to try to begin to explore these long non-coding RNAs in other cancers, potentially prostate cancer as well as other epithelial tumors, taking advantage of next-generation sequencing-based technologies. Before our work, long non-coding RNAs were primarily explored using microarray-based technologies. We basically employed next-generation transcriptome sequencing to try to explore and discover novel long non-coding RNAs across different tissue lineages as well as in cancer-specific long non-coding RNAs. We were using these next-generation transcriptome sequencing-based approaches to really discover gene fusions in common solid tumors, and we were generating hundreds to thousands of samples in terms of transcriptomic data. And so one of our questions was, can we take advantage of this transcriptomic data to basically nominate and identify long non-coding RNAs in the context of cancer. So we basically developed a set of computational tools to nominate long non-coding RNA from transcriptomic data. So we basically developed ab initio approaches to basically assemble the reads that were intergenic in nature to basically call the structure of these long non-coding RNAs. And our first foray into this was in prostate cancer using some of the early generation Illumina technology. We basically profiled prostate cancer progression from benign to clinically localized disease to metastatic disease. We certainly picked up many of the protein coding genes, 70% in this case, but we identified probably on the order of about 20% of these long non-coding RNAs as unannotated transcripts. And we named them, and here is a heat map representation.— red meaning elevated, blue meaning downregulated. And we call this first set of long non-coding RNAs in prostate cancer PCAT for prostate cancer-associated transcripts from PCAT1 to PCAT114, some of which are listed in this particular heat map representation. Our next question was, can we begin to use the latest version of these sequencing-based technologies as well as adapted pipelines to basically explore not only our data, we had assembled about 1,000 transcriptomic datasets. We took advantage at the time when we began this project about 2 years ago or about a year and a half ago, we assembled what was available in the TCGA effort, which was about 6,000 transcriptomic libraries, and basically employed a variety of bioinformatics approaches basically to analyze this data. So this is a pie chart representing the compendium of data that we assembled in terms of transcriptomic data to analyze. As you can see, this is a variety of different cancers that are represented here. And we were able to nominate actually over 50,000 novel long non-coding RNAs from this analysis. So we basically deployed meta-assembly approaches to assemble the transcriptome of about 7,000 samples. Samples that we had assembled at the time. So through this effort, we identified over 8,000 cancer-related long non-coding RNAs as well as a host of long non-coding RNAs that basically defined various tissue types or lineages.

One of the remarkable observations that we had were that these long non-coding RNAs could be exquisitely tissue or lineage specific.

As well as in some cases cancer or disease specific. So here is basically a heat map representation across different cancer types that are represented here. For example, breast, prostate, liver, and so forth. And it's sets of long non-coding RNAs that basically define these different lineages of tissues. When we had matched normal samples, we were then able to actually identify not only long non-coding RNAs that were lineage specific, but also long non-coding RNAs that defined cancer, that were also cancer-specific. And these are listed here as being elevated. So in the published paper, you can actually call out and see the various long non-coding RNAs that we've defined across different tissue types. For example, in lung adenocarcinoma, we call these LACats for associated transcripts. So in some cases, hundreds to thousands of novel long non-coding RNAs that were defined through this analysis. So there's potentially certainly a tremendous wealth of information in terms of biomarker development as well as potentially understanding cancer biology. We certainly believe that at least a subset of these long non-coding RNAs are important in differentiation, in normal differentiation, as well as a subset are likely involved in cancer progression. So we've developed a resource to take advantage of this transcriptomic data.

This is actually an evolution of the Oncomine database that we developed for microarray and genomic data in general.

We call this MyTranscriptome.

This is basically an assembly of next-generation sequencing transcriptomic data, RNA-seq data.

This website is freely available to the public, and you can basically explore this dataset. Right now it's really geared towards of expressed long non-coding RNAs, and it's focused on about 8,000 or so long non-coding RNAs that are represented in our first study. Next generation versions of this database will start to include protein coding genes as well as additional long non-coding RNAs, as well as begin to represent the entire TCGA dataset as well as other data that we've assembled from the public domain. So this is a view of the database itself. For example, in HOTAIR, you can basically pull up HOTAIR and see that in breast cancer, a subset of breast cancers have very high levels of HOTAIR, but you can see that HOTAIR is actually expressed in subsets of other epithelial tumors. In this example, we basically interrogated the database for SLAP1, which is a long non-coding RNA that we've been focused on in terms of cancer biology, in terms of prostate cancer biology. And you can see that this long non-coding RNA is exquisitely specific for prostate cancer relative to other cancers. And we ended up following this particular long non-coding RNA in much more detail. This is just another view of the data, again showing exquisite specificity to a subset of prostate cancers relative to other tissue types. Through a collaboration with a company called GenomeDX, we were able to take advantage of their exon array data to actually show that SLAP1 as a long non-coding RNA was the top prognostic gene in this dataset looking at all protein coding genes as well as all long non-coding genes. So we believe that this is a key long non-coding RNA that will be useful in predicting aggressiveness in prostate cancer. And this is just a representation of one of the datasets where we basically show that high levels of the long non-coding RNA SLAP1 is associated with poor prognosis relative to low levels of this long non-coding RNA.

And this has now been shown in about 5 independent datasets by at least 2 independent groups exploring SLAP1 in this context.

One of the aspects of prostate cancer, of course, is can we basically detect this long non-coding RNA in a non-invasive fashion. So this is just early data to show that we can actually detect this long non-coding RNA, SLAP, in the urine of men with prostate cancer. Again, this is a subset of patients where we can detect SLAP1 as being elevated. These patients tend to have more aggressive forms of prostate cancer, higher Gleason grade, as well as higher risk versions of prostate cancer. And we actually plan to add this long non-coding RNA to existing RNAs that we've been able to monitor in urine. For example, the TMPRSS2-ERG gene fusion, which we couple with the long non-coding RNA PCA3. This is actually already available clinically to detect forms of prostate cancer, aggressive forms of prostate cancer. We plan to basically add SLAP1 into this mix to basically create a multiplex biomarker of prostate cancer aggressiveness. This is just an ROC curve looking at TMPRSS2-ERD combined with the long non-coding RNA PCA3 relative to serum PSA alone, which you get generally an area under the curve of about 0.6. When you begin to combine biomarkers, you can achieve area under the curves approaching 0.8. So what's interesting about SLAP, although we've been studying hundreds to thousands of long non-coding RNAs in prostate cancer, only a very small set of these lincRNAs appear to be associated with biological progression or actually potentially have a role in human disease. And SHLOP1, we believe, appears to be one of those lincRNAs. I won't be able to go into the details about its biological function. Much of this is published, although some of this data is still unpublished. This long non-coding RNA we've shown physically interacts with this Y-SNIF epigenetic complex. We believe that this is how its mechanism of action works. It's actually, we believe, an oncogenic long non-coding RNA, again elevated specifically in prostate cancer. It interacts with this Y-SNIF complex, which is an epigenetic complex involved in nucleosomal remodeling. It's been implicated as a tumor suppressor. SWI/SNF has been shown, the SWI/SNF complex has been shown to negatively regulate the PRC2 complex as an oncogenic epigenetic complex. So we believe that Schlapp1 negatively regulates this tumor suppressor complex which then can negatively regulate PRC2 and we believe that this is how its mechanism of action may be presumed. Similar in some ways to hot air and other link RNAs interacting with PRC2 and mediating its function. In this case, this long non-coding RNA actually physically interacts with this epigenetic complex and negatively regulates its function. So what are we going to do next in terms of the long non-coding RNA space? We will— we continue to analyze public domain RNA-seq data including TCGA. We plan to add again functionality to the My Transcriptome resource so it'll be more useful to the academic community. We're beginning to characterize candidate lincRNAs as tissue biomarkers as well as noninvasive cancer biomarkers, primarily really leveraging the lineage specificity of these long noncoding RNAs. We're pursuing the biological function, again, of a small subset of the most interesting cancer-related lincRNAs including SLAP as well as a series of other lincRNAs across across different tissue types. We're also exploring the idea, especially when we identify a lincRNA that appears to be associated with cancer biology and cancer progression, the idea of therapeutically targeting these long non-coding RNAs using antisense oligo technology. One of the mysteries that we're certainly interested in pursuing is why are these long non-coding RNAs indeed expressed in a lineage-specific fashion? And why are these long non-coding RNAs expressed oftentimes in a cancer-specific or disease-specific function?

What are the global mechanisms of regulation of these long non-coding RNAs?

So this is the team that was really associated with our long non-coding RNA work, and those highlighted in red were quite significant collaborators. And now I will transition my talk to something that's a bit more mature, I would say, in terms of clinical application and translation. You heard a lot of introduction about this area, both from Paul and Charlie yesterday. And we began our effort in about 2010, 2011, and basically established this study that was in translation— Science Translational Medicine with the idea that cancer patients— that we should begin to comprehensively define the mutational landscape of individual cancer patients in real time by procuring biopsies and be able to then return that information to individual patients to guide treatment selection.

And this is what was suggested in this particular study.

And as you heard from Charlie yesterday, there are a number of challenges and pitfalls that we need to consider as we move forward with these efforts. But what I'll be primarily focusing on in this particular study is really what Charlie characterized as truncal driver mutations in terms of the types of mutations that we've really focused on in terms of suggesting potential therapies. Sort of truncal mutations that are high allele fraction, major clonal clonal versions of these mutations. So why do we want to take on a comprehensive approach to understand the molecular landscape of cancer? Generally, tumors have a private combination of mutations. Oftentimes there are rare driver mutations, sort of this long tail of mutations that can be identified. For example, RAF fusions or AKT mutations in prostate cancer. ALK fusions, of course, in lung cancer is a good example, relatively rare. Targeted sequencing assays can miss new drivers. One of the early goals of these efforts, of course, are to enrich our phase 1 and 2 clinical trials rather than basically ascribing patients in a random fashion to these studies.

We want to basically have a better chance that these clinical trials will have a better chance of success by enriching these trials based on molecular information.

We also have the ability to characterize mechanisms of resistance because we basically can profile patients before they get onto a targeted therapy or clinical trial as well as once they progress on that particular clinical trial, suggesting that we can compare the mutational landscape and basically understand potential mechanisms of resistance. And then certainly by a more comprehensive approach rather than a targeted approach, This certainly affords an opportunity for new discoveries to be made. So this has been our— the workflow of our clinical sequencing effort. We were likely one of the first laboratories that established a clinical sequencing effort for advanced cancer patients. And essentially we're enrolling patients that have failed conventional treatments, generally metastatic disease and are looking for potential options in terms of off-label use of approved agents or potentially clinical trials. And the general approach is that, again, these are real-time biopsies of metastases, not archival samples that we're using in most cases. And the basic idea is that we take a sample of the tumor tissue. And compare that to a matched sample of the germline, a buccal swab or a blood sample. So we're basically comparing the tumor genome to a matched normal genome. We carry out integrative sequencing, computational analysis, and then this data is presented to our precision medicine tumor board, which then basically deliberates upon the actionability of these findings, and then we have permission to return these results back to individual patients. So I don't really have to go over this cartoon with you in detail, but this is certainly one of the tenets of precision medicine or precision oncology. The idea that you can begin to match potential mutations with potential targeted therapies. For example, a patient that might have a PIK3CA mutation, you might suggest a PIK3 kinase inhibitor. A patient that has an FGFR1 kinase amplification and FGFR inhibitor and so forth.

This is certainly one of the basic tenets of these precision medicine, precision oncology approaches. So in—

when we initiated our effort in 2010, 2011, what we really wanted to do was be comprehensive about the types of mutations or aberrations that are known to cause cancer. We wanted to capture structural rearrangements in terms of translocations and gene fusions. Copy number alterations, amplifications, deletions, loss of heterozygosity, point mutations, and indels, as well as gene expression alterations in terms of various pathways, outlier genes that might be dysregulated. So to get to that goal, what we established early on was this idea of integrative sequencing. It's really a combination of technologies that really allows you to be comprehensive about the classes of mutations that can cause cancer. This includes low-pass whole genome sequencing, whole exome sequencing of the tumor matched with normal. Generally we're doing about 150 to 200x coverage. And transcriptome sequencing or RNA sequencing. So it's really this combination of technologies that allows you to be relatively comprehensive about the mutational landscape. Of cancer. After we sequenced the first 20 patients in the study, and now we've probably sequenced over 1,000 patients, but in the first 20 patients what we found was that we were actually getting most of the information from the exome data as well as the transcriptome data, and thus reserved whole genome sequencing only for a select set of patients in which we were not identifying potential drivers. And this, of course, also helped us mitigate the cost doing these analyses on individual patients. One of the first things that we faced as we established this protocol in 2010 was whether we could return the results in a clinically relevant timeframe. Certainly a number of advances have occurred since then, but what we were able to show early on was that we can certainly carry out these studies and return the results within a 4- to 6-week turnaround period. From the time of biopsy to pathology assessment, sample prep, sequencing analysis, and return of results. Now with the latest developments in terms of sequencing technology as well as improvements in computational analysis, this turnaround time can certainly be improved to about 1 to 2 weeks. One of the innovations that we introduced in our clinical sequencing effort was the idea of using multidisciplinary precision medicine tumor boards. Again, this is basically the adaptation of the conventional tumor board methodology, but now really including experts in the area of genomics and informatics, clinical genetics, bioethics, and so forth to collaborate with and work with very intimately with medical oncologists, pathologists, and radiologists who are typically attending these tumor boards.

What our precision medicine tumor boards or PMTB boards focus on, and we've now We've now had about 60 precision medicine tumor board meetings.

What they really focus on are the actionability of the findings that we identify. And they generally classify the actionability based on a class 1 aberration, which suggests that an already approved agent could be utilized in an individual patient. A class 2 type of mutation suggests that an investigational agent might be appropriate. So this would be a clinical trial, either a late-stage clinical trial or an earlier stage toxicity study. Class 3 type mutations are generally there's preclinical evidence that a potential mutation might be a driver and there might actually be some probe compounds in development but nothing has been actually tested in humans. And then finally, a Class 4 aberration is of unknown significance.

So this precision—

this multidisciplinary precision medicine tumor board really deliberates on the actionability of the individual findings that we discover in an individual patient. And certainly after our precision medicine tumor board meets, we issue a molecular report to the physician as well as the individual patient. I alluded to this earlier. Certainly this effort of comprehensive sequencing of individual patients has been fruitful in terms of individual discoveries. And these are actually individual patients that were enrolled in our MyOncoSeq clinical sequencing effort that actually led to new discoveries. For example, we discovered the NAB2-STAT6 fusion as a driver fusion for solitary fibrous tumors, and 100% of these tumors are actually characterized by this particular gene fusion. We've had some remarkable responses in individual patients. I'll share a few of them in subsequent slides. But, for example, a pediatric patient that we treated on this study or that we sequenced on this study, 11-year-old girl, had an ETV6-ABL fusion that we detected in a pediatric leukemia.

And we were able to put this patient on imatinib and put this patient into a significant remission.

And we've had numerous anecdotes like that. Like that. We've also discovered rare but targetable FGFR kinase fusions across an array of diverse tumor types, across epithelial tumor types enriched in cholangiocarcinoma, as well as found in breast cancer, prostate cancer, and other cancers. We've discovered common resistance mechanisms, for example estrogen receptor 1 mutations as a common resistance mechanism in ER-positive metastatic breast cancer.

Generally patients we believe that have been treated with aromatase inhibitors appear to have this particular resistance mechanism suggesting that ER signaling still is in play in the context of patients that have become resistant to tamoxifen as well as aromatase inhibitor therapy.

So since our efforts began in 2010, 2011, we've now sequenced over 1,000 patients and one of our Goals over the next year or so is to really be able to sequence every patient that comes to our comprehensive cancer center. But to date, we've sequenced about 1,000 patients in terms of whole exome sequencing combined with RNA sequencing. Over 600 patients, adult patients, about 150 patients in the pediatric oncology space, and over 200 patients in terms of a Stand Up to Cancer, Prostate Cancer Foundation Dream Team effort, which I'll share with you at the end. So in terms of the pediatric space, the workflow is quite similar, but the issues around pediatric oncology are often quite different in terms of consent issues dealing with minors, as well as potential implications to the family when germline events are identified, as well as the lack of potential candidate therapies for the pediatric population. As well as understanding the correct dosages for the pediatric population. So these are multiple challenges that we had to face before getting approval for this independent protocol. But what was intriguing about this set of patients that we sequenced, over 100 patients, was that we were able to work with a single pediatric oncologist and he was able to really pursue the actionability of each of these findings. In terms of our adult protocol, we're generally working with a matrix or syndicate of medical oncologists, and they have variable—

they're variably interested in pursuing the actionability of our recommendations.

But in this case, we were able to, in a more regimented fashion, follow each of our patients. So in this prototype study of about 100 patients, 100 pediatric oncology patients that we sequenced, we looked at hematologic malignancies, as well as solid tumors that are represented here, a nice mixture of cancer types. What we found, and this is just a summary, in terms of driver mutations, we generally found in about 50% of cases a somatic SNV or indel, a gene fusion in about 30% of cases as a driver, homozygous deletions, amplifications, and in over 10% of cases we actually found a pathogenic germline variation that we needed to actually report back to the family in terms of implications for that individual patient as well as their family, suggesting that certainly in a pediatric oncology patient that they should— that genetic counseling should be considered in this context. In red you can see that only a subset of these driver mutations were potentially actionable in nature. So now summarizing the actionability in terms of this first 100 patients that we sequenced. We found that only about 46% of cases in the pediatric population, we were actually able to identify something that was clinically actionable. When that was— and in that case, we were only actually able to act upon about 53% of those cases, primarily because of issues in terms of we got the information too late, The patient was not eligible for that particular clinical trial or they weren't able to travel to that site. So a variety of challenges that come across even when you actually identify a clinically actionable event. But in those patients that we were actually able to act upon, we were able to change therapy in 60% of the patients and provide genetic counseling in about 40% of patients. And in this case, about 80% of the patients that we were able to act upon actually had a clinical benefit in terms of basically a durable response in terms of therapy, as well as potential benefit to the family. About 20% of cases, the limited number that we were actually able to act upon, we did not— the patients did not exhibit a clinical benefit.

Again, this is just an early study that suggests the promise of these precision medicine efforts.

So this is the team that was really involved with this study, a number of individuals in my group as well as our close collaboration with Rajen Mody, who is the head of our Phase I pediatric oncology group at the University of Michigan. The last story that I'd like to share with you is really one of our efforts to begin to take these clinical sequencing efforts to a multi-institutional scale, working with multiple different cancer centers in the context of metastatic castration-resistant prostate cancer. So this study was funded by the Stand Up to Cancer organization as well as the Prostate Cancer Foundation. And one of the surprising aspects of this study is, is that we found at least in metastatic castration-resistant prostate cancer, in over 90% of cases, we were actually able to identify something actionable. So this is the team that I co-lead with Charles Sawyers at Memorial Sloan-Kettering. We've involved a number of key investigators across the country, as well as Johan de Bono at the ICR Royal Marsden, who's been a key collaborator. So in this particular study, we're basically doing real-time biopsies of patients with metastatic prostate cancer, oftentimes bone biopsies or soft tissue biopsies. And we're enrolling at 8 different clinical sites, so it's actually one of the first multi-institutional clinical sequencing efforts. We're actually funded to sequence 500 patients with castration-resistant disease. What I'll be sharing with you today is an interim analysis of about 150 patients that we've sequenced thus far.

We've actually sequenced about 250 now to date, and we're rapidly approaching the 500 number.

But we're enrolling at 8 different clinical sites. The sequencing and analysis actually occurs at the University of Michigan and at the Broad Institute. The data is integrated at Memorial Sloan-Kettering and then visualized through the cBioPortal. We have precision medicine tumor boards at each of our respective institutions, and the central pathology review is carried out at Wellcome Cornell.

So this is really a team effort in terms of involvement across multiple institutions, multiple principals in the development of this study.

So we have about 8 different clinical trials that are associated with patients that are enrolled on this clinical sequencing effort. Patients in the pre-abiraterone, pre'enzalutamide space, as well as patients, metastatic prostate cancer patients in the post-abiraterone, enzalutamide space. Phase.

So this is a listing of the clinical trials that the patients were potentially eligible for.

Again, all metastatic castration-resistant prostate cancer, and the various clinical trials that were associated with this study were really focused around targeting androgen receptor or DNA repair pathways. This is basically a summary of the types of biopsies that we were able to obtain. From these patients. This varied from lymph node, bone, liver, and soft tissue. When we began the study, I think one of the challenges that we needed to approach was whether we would be able to adequately sample bone, because bone is a favorite site for prostate cancer metastases. And what we found was that we were in probably over 80% of cases able to adequately sample bone in terms of these precision medicine efforts. So this basically gives you an overview of the landscape of genomic alterations that we saw in the first 150 patients that we sequenced in a multi-institutional fashion. So individual patients are listed as columns, while individual genes are represented as different rows. And we've grouped them into the various clinically actionable pathways that we defined. So this includes AR-associated aberrations, PI3 kinase pathway aberrations, RAF, Wnt pathway, DNA repair, cell cycle, chromatin modifiers, and others. We also identified that a subset of these patients, about 4 out of 150, had a hypermutated phenotype that was actually attributed to aberrations in mismatch repair. This is potentially intriguing because this set of hypermutated patients could potentially benefit from immunotherapy-based approaches with the idea that the hypermutated phenotype could potentially create neoantigens making them more susceptible to an immunotherapy-based approach. So through this analysis of 150 patients, we certainly picked up the common mutations or aberrations that were known in castration-resistant prostate cancer. Including aberrations in the androgen receptor pathway, p53, p10, as well as the ETS fusions in a high prevalence of patients. The rest of the mutations generally range from about 2 to 10% in prevalence. So this gives you the copy number landscape of this cohort of patients. We certainly picked up known aberrations that were known in castration-resistant disease, for example, the AR amplification. As well as the TMPRSS2-ERG deletion which results in the fusion. But we've picked up some novel associations including PIK3CA and PIK3CB amplifications as well as homozygous loss of ZBTB16 which is a negative regulator of androgen receptor signaling. In terms of the ETS fusions that we described in 2005, we certainly identified that over 60% of cases harbored a driver gene fusion. A majority of them were X-based gene fusions. We also identified RAF fusions, ARS-SPONDIN fusions that were previously characterized in colon cancer, as well as intriguingly PI3 kinase fusions. These were all mutually exclusive driver fusions in these cases. So when we compare metastatic castration-resistant disease with primary hormone-naive disease, we see that certain mutations appear to evolve likely due to treatment pressures. This includes mutation of P53, which is enriched in metastatic castration-resistant disease, AR amplification of course, as well as a number of other genes that appear to be enriched in metastatic progression of prostate cancer. So now let's just drill down into the individual pathways that we explored using this technology. The androgen receptor pathway, of course, was expected to be aberrant in prostate cancer, and certainly in a large percentage of cases, probably on the order of 50% of cases, we actually saw AR amplification. But in addition to AR amplification, we picked up activating mutations of AR. Which are shown here. These are activating, in some cases known, in some cases new mutations that were identified that are potentially activating in terms of AR activity. We also found aberrations that in key components of the androgen receptor pathway, for example, a loss of function of Ncore1 and 2, which is a negative regulator of AR, aberrations in FOXK, A1, which is a pioneer transcription factor, as well as homozygous deletion of ZBT16, which is an AR-regulated gene that negatively regulates AR.

These are all aberrations that appear to converge upon or get activated in terms of castration-resistant disease.

In terms of the PI3 kinase pathway, we found that in about 50% of cases we found an aberration in this pathway. We picked up, of course, the homozygous loss of PTEN, which was known, but we also picked up some novel aberrations in a number of other key components of the pathway, which are represented here. We picked up the activating mutations of PIK3CA, which were known to be rare relatively in prostate cancer, but through this study, we actually discovered activating mutations of PIK3CB. And so these activating mutations of PIK3CB are known to be, are shown to be functionally analogous to the activating mutations of PIK3CA. And they're occurring in key amino acids that are analogous to those that are known to be activating in PIK3CA.

And so this particular finding is exciting because there are certain drug companies that are actually developing small molecule inhibitors that target PIK3CB specifically relative to PIK3CA.

Of PIK3CA. And in addition to both the activating mutations of PIK3CA and PIK3CB, we saw amplifications of PIK3CA and PIK3CB which were discovered in the study as well as recurrent gene fusions of PIK3CA and PIK3CB that lead to massive overexpression of these kinases similar to the X fusions where they basically inherit an androgen-regulated 5' partner that leads to overexpression of PIK3CB shown by RNA-seq data. So these are basically— this is an example of clinical impact on a patient that was enrolled on this study that received basically— this is a patient that was enrolled by Johan de Bono at the Marsden that had a PIK3CB amplification. And you can see the location of the tumor here. This patient was treated with the GSK inhibitor of PIK3CB and showed significant regression of the tumor as well as marked decreases in PSA levels, a 70% decrease in PSA levels after receiving the PIK3CB inhibitor. And this PIK3CB inhibitor was only, appears to only work according to Johan in patients that have an aberration of PIK3CB. And then finally, the last major finding in this study that probably had— has the most clinical impact was that in about 20 to 30% of patients, we actually identified aberrations in the DNA repair pathway. This included aberrations in BRCA2, ATM, and BRCA1 in prostate cancer, again approaching 20 to 30% of patients. BRCA2 was the most enriched in metastatic disease. Both somatic aberrations shown here as well as germline aberrations of BRCA2. And these findings are certainly exciting because it really suggests that these patients could benefit from PARP inhibitors. The PARP inhibitor olaparib, for example, has been approved in ovarian cancer patients with BRCA aberration. This would suggest that prostate cancer patients that have an aberration in DNA repair might benefit from PARP inhibitors.

The other intriguing aspect was that a high prevalence of these patients actually have a germline aberration in BRCA2.

And here are some of the clinical responses of patients again that have been enrolled on our Stand Up to Cancer study. This is again another patient from Johann de Bono's study. This patient had a homozygous deletion of BRCA2 and was treated with the PARP inhibitor olaparib. So this is an imaging study before treatment, widespread metastases shown here, and then after olaparib treatment, significant regression of the tumors based on this imaging study.

And there was a significant decrease of PSA from about 30 to 0 in this particular patient.

And this patient, as well as a series of additional patients, have maintained a durable response to olaparib treatment, again according to Johan. So this is a patient, another Stand Up to Cancer Prostate Cancer Foundation patient that was enrolled by Maha Hussain at the University of Michigan. This patient also had a homozygous deletion of BRCA2. You can see the metastasis to the liver. This patient was then treated on abiraterone plus a PARP inhibitor called, it's the Abbott PARP inhibitor. And you can see the significant regression of this liver metastasis. Post-PARP inhibitor abiraterone treatment and a marked decrease in PSA from 127 to less than 0.1. And we've had a number of these exceptional responders, all of which have had defects in DNA repair and have had significant responses to PARP inhibitors, both in this particular study led by Mahal Hussain, as well as multiple potential examples in the 2-part study led by Johann de Bono. So in terms of driver mutations in castration-resistant disease, we find that when we look at somatic mutations, SNVs and indels, in about 100% of cases we can actually nominate a driver mutation that's at a high allele count in a dominant clone. In about 50 to 60% of cases we can identify a driver gene fusion. Homozygous deletions at about 50%, focal amplifications at about 50%, and interestingly, a pathogenic germline variation in about 10%, over 10% of cases, suggesting that again, patients with metastatic prostate cancer should potentially consider genetic counseling because of the high prevalence of pathogenic germline events in this population. And now when we focus in on the actionability of these findings, what's remarkable is that in over 90% of cases, we're actually able to identify an actionable event in metastatic castration-resistant prostate cancer. This is actually broken up into AR itself, which is present in about 60% of cases, but when you don't include aberrations in AR, which of course were known previously, In over 60% of cases, we can actually identify something clinically actionable in the context of being not related to the AR pathway. This includes actionable events in the PI3 kinase pathway, DNA repair, RAF pathway, CDK inhibition, so this would be cell cycle, the Wnt pathway, as well as pathogenic germline aberrations. And so this is an example of some of the clinically actionable mutations that we identified. For example, DNA repair aberrations shown here could potentially match with PARP inhibitors. PIK3CB being matched to PIK3CB inhibitors. Again, these are just examples from this cohort. Another interesting observation were RNF43 loss, ZNRF3 loss, as well as ARS-pondin fusions, which could potentially be matched with porcupine inhibitors or hedgehog inhibitors, as well as antibodies directed against ARS-pondin. And then finally, the deleterious germline variants would certainly suggest potential genetic counseling for that subset of patients. So what are the highlights of this particular study around castration-resistant disease? What did we find from this study? This is one of the first multi-institutional integrative clinical sequencing efforts. And it suggests that it's certainly feasible in metastatic prostate cancer, metastatic castration-resistant prostate cancer. In about 90% of cases, we were actually able to identify a clinically actionable molecular alteration. In this particular study, we actually identified some novel genomic aberrations, some of which are listed here. In over 20% of cases, probably 20 to 30% of cases, we were actually able to show that suggest that these patients harbor DNA repair pathway aberrations and might— and suggest that these patients might be amenable to PARP inhibitor or platinum-based therapies. And then finally, in 8 to 10% of patients, we identified harbor pathogenic germline findings suggesting that genetic counseling or genetic screening of these patients might be considered similar to what's already being done in the context of BRCA screening in breast cancer. One should be begin to consider this in the context of advanced prostate cancer. So, some final take-home points that I think are just relevant to precision medicine and precision oncology in general. I think it may be possible to provide a relatively comprehensive mutational landscape for an individual cancer patient in terms of exome and transcriptomic analyses. I think it's possible. The costs are going to decrease as well as the turnaround time will improve. Improve. I think it's important to monitor a variety of mutation classes and being sure to monitor point mutations, indels, copy number alterations, fusions, and others. One of the early goals of these studies will basically be to enrich patient populations in terms of clinical trials as well as off-label use of approved drugs. I think the long-term goal is really to move towards combination treatments that are based on the mutational assessment of an individual patient, matching the combination of driver mutations that are identified in an individual patient with a combination of treatments. And then the final point, and I think we'll have a panel discussion on this in a little bit more detail, these types of efforts certainly are multidisciplinary in nature and really represent a team effort of a number of individuals across disciplines including pathologists, genetic counselors, oncologists, radiologists, and bioinformaticians, clinical geneticists, and bioethicists to essentially make this type of program work in a regimented fashion. With that, I'd like to thank you for your attention and certainly highlight the various individuals that are part of our Stand Up to Cancer Prostate Cancer Foundation Dream Team, about 8 different institutions, and 2 key individuals who really pioneered our initial study, Dan Robinson at the University of Michigan and Ellie Van Allen at the Broad Institute, and certainly the contributions of the various teams—over 150 individual participants on this particular project. Thank you for your attention, and if we have time, I'd be happy to take any questions.

So thanks, Arul, for that spectacular demonstration of team science in all its glory. So the floor is open for questions.

Very exciting talk. Thanks. I have two questions. One is that I saw that you have a group pipeline for pediatric patients and a pipeline for adult patients. I was wondering whether you guys see the evidence of showing that whether pediatric patients are significantly different in terms of their genetic background. So that they need a totally different treatment.

Yeah, that's a great question. I think the question was basically what is the mutational difference in the pediatric population relative to the adult patients at least that we were enrolling in our studies. And certainly this was known before our study through TCGA efforts and so forth and we were more or less reconfirming this in terms of the dimensionality of the types of mutations in the pediatric population oftentimes are considerably less than what we're seeing in adult populations in terms of the number of driver mutations that were identified. So I can suggest that in terms of the number as well as diversity of driver mutations tend to be less in the pediatric population relative to adult. And in terms of actionability, we find that the impact, we believe, at least comparing our pediatric patients to our adult patients, we've had more clinical responses in our pediatric patients relative to our adult patients, with the idea that they have potentially less drivers to impact. When you look at, looking at a more evolved adult patient where they have multiple drivers that might be in play, it becomes much more difficult to treat those patients. Thanks.

Second question is that do you think the epigenetic profiling will provide more information in addition to the sequencing, for example, the methylation status and other status of them?

Yeah, I think certainly we're beginning to explore the idea of using DNA methylation analysis as well as histone methylation analysis in a subset of patients. We haven't really found a clinically actionable observation using those technologies yet, but I do think that moving into the future that they will have potential. But we're certainly exploring that in a research setting. Thank you. Charlie?

Thanks, Errol.

In terms of the patients with DNA repair abnormalities, do you see— always hemizygous events or are they affecting both alleles when you see responses to, let's say, PARP inhibitors? Do you actually stratify patients based on biallelic events or is, say, a hemizygous event in ATM sufficient to put a patient on a PARP inhibitor?

Yeah, that's a great question.

I think in a vast majority of the patients that we were seeing the responses, we were seeing a homozygous event. So, for example, with BRCA2, there are many cases where we see a germline event as well as a somatic event, and that was the case in ATM as well as BRCA1.

So have you seen responses in patients without a germline event? So if you just take out those patients with germline, say, BRCA2 mutations? Yes.

You do?

So patients who have biallelic abnormalities in ATM or BRCA? Exactly right.

Great. Very interesting. Thank you. Hi. Thanks for the great talk. So given the large genomic alterations you found in this study, I wonder whether you could comment on efforts that we could take into understanding cooperating mutation events, and similarly whether you can talk about the efforts that we should understand alteration that involves DNA mutations versus changes at the transcriptional level, such as the overexpression of EZH2 has been frequently reported in prostate cancer.

Can you summarize that question again? I—

Sorry, I guess the question is, could you comment on the efforts that we could undertake to understand the cooperating mutation events in many of these alterations, as well as cooperating between DNA mutations versus changes in transcriptome?

Yeah, I mean, I think— so I guess the question was the— maybe the integrative aspect of our approach of using whole exome sequencing of DNA with RNA sequencing transcriptomic analyses.

Oftentimes the things that we were nominating using transcriptomic data were DNA-based events but just not easily appreciated using whole exome-based technologies, such as, for example, gene fusions.

We were often using RNA-Seq-based approaches for that. And we find that that's critically important for the pediatric population where oftentimes the driver aberration is indeed a gene fusion, which is often, of course, going to be missed by standard exome technology. So we do think that it's important to use the transcriptomic data to complement the efforts in terms of whole exome assessment in order to pick up gene fusions certainly, also in terms of potential outlier genes in various pathways that are aberrant using transcriptomic data. You can basically complement the analysis in that context, as well as I think moving forward looking at various gene sets that might suggest pathway activation that can be taken advantage of using RNA-Seq data. But of course it becomes much more complicated using RNA-seq data for that because you have to oftentimes subtract out the admixture of normal stromal events that might be— stromal components relative to the tumor component. Thank you. Joel?

Hi, Joel.

Rul, great progress. As we've tried to go down this road, you made the point that in the recurrent disease you were picking up something like 40% new actionable mutations in the subsequent sequencing. And as we've— okay, so now you've got the actionable mutations. The question I'm getting to is, can you get a drug that actually attacks the actionable mutations, and can you get it sufficiently quickly to be clinically usable? I mean, that's been our major problem and stumbling block in dealing with precision medicine, is we can't get the drugs.

Yeah, that's been the major challenge. And there was certainly evidence in the context of the pediatric study where in only about 50% of cases that we actually identified something actionable that we were actually able to get the drug in terms of either getting the approved agent and using it potentially off-label or being able to enroll that patient onto that clinical trial. And it was particularly a challenge with the pediatric population, primarily because they're oftentimes excluded on many of these adult studies.

In terms of our adult population, I think we've certainly had major challenges, at least certainly in the U.S., in terms of actually obtaining an already approved agent, an FDA-approved agent for another indication, and to use that off-label even though we have genetic evidence suggest that that patient might respond, we've had massive challenges in terms of being able to convince the insurance companies involved to potentially pay for those oftentimes very expensive targeted therapies.

That's been an ongoing challenge that's often been sort of a hit-and-miss process in terms of how lucky we are in terms of convincing individual insurance companies to potentially pay for those efforts.

We have since developed a parallel philanthropic fund to basically try to COVID patients in which their insurance doesn't compensate them for that targeted therapy.

So we often use that fund to try to support those efforts. So I think this is an ongoing challenge moving forward, is really beginning to convince not only convince insurance companies to begin to cover these sequencing-based approaches in the first place, but then also begin to consider paying for these expensive targeted therapies, or at least in the context of early signal-finding studies.

I think we'll just take two more questions and then we'll move on.

So I think Mao, and then over to you.

Yeah, so for your— MyOncoSeq study, instead of just looking at those individual responders, have you collected enough sufficient outcome data to demonstrate it is indeed better than the conventional way to treat patients?

Yeah, so with the MyOncoSeq study, in the adult, and I think I alluded to this earlier, in the adult study that has been more of a challenge. So in the pediatric study we were able to actually collect outcome information and actually follow each of the patients in terms of the actionability of the individual findings because we were dealing with a single oncologist.

The adult study often involved over 50 different oncologists. While we've certainly been able to analyze the data in terms of whether the physicians have followed our recommendations, we haven't necessarily been able to really enforce, I would say, sort of the individual physician to pursue that thought if it's an off-label use of an approved agent or enrollment on a particular investigational study. So it's been a lot more difficult in the adult setting than in the pediatric setting for a number of reasons. In the pediatric setting, we're working with a very motivated pediatric oncologist who's willing to sort of follow up the patient and pursue the investigational agent or somehow obtaining the approved agent. While in the adult setting, it's oftentimes sort of a syndicate of oncologists that we've had to work with.

So perhaps you should do it as like a clinical trial setting. You may even have a controlled arm to demonstrate the clinical utility. Exactly.

I think that that's, I think, the next phase of a lot of the studies that people are beginning to consider is really doing this in a much more regimented fashion. Last question.

Hi. Great talk. Your discovery of the lineage-specific and the cancer-type-specific expression of lincRNAs are very interesting. So I have a technical question. How sensitive is the result to the sample preparation? So is there any problem of the degradation? Can you repeat same result on the FFP samples or any like heavy stained cells or cell lines? Right.

Yeah, so it's certainly very sensitive to degradation and in order to do the RNA-seq analysis, we're using the—

so this analysis is really focused around polyadenylated RNAs..

And it's very much sensitive to degradation and it generally doesn't work in the context of FFPE samples. But we have developed a parallel approach that works quite well in FFPE using capture-based technology, a capture-based RNA-seq that then you can essentially monitor these RNA— these long non-coding RNAs in degraded samples. But the data that I presented here was really using fresh samples with high-quality RNA.

All right, so I think all of you— this is a fantastic talk, so please give me— help me to thank Dr. Chan.