A Conversation with Arul Chinnaiyan
, by NCI Staff
Editor's Note: Born in Ohio to immigrant parents and raised in Chicago and Southeastern Michigan, Arul Chinnaiyan is a Howard Hughes Investigator and Professor of Pathology and Urology. In 2005 he and his lab broke a paradigm of cancer biology when they demonstrated that in addition to blood malignancies a solid tumor, prostate cancer, is commonly driven by chromosomal rearrangements. He has also been a leader in precision medicine, having integrated next-gen sequencing of patient tumor DNA and RNA with treatment decisions starting in 2011. His intriguing findings on RAS are discussed below.
Is the translocation you discovered the most common driver of prostate cancer? I would say it is the most recurrent and dominant driver lesion in prostate cancer. At the time it was quite an exciting discovery, because it was the first recurrent gene fusion or translocation found in a common solid tumor at such high prevalence, suggesting that mechanisms that are at play in the hematologic malignancies are also abundant in prostate cancer. And subsequent to that we and others have found that subsets of other solid tumors are also caused by gene fusions or translocations, such as lung cancer and breast cancers and cholangiocarcinomas, among others.
How did your interest in prostate cancer expand to so many other cancer types?
I began my career using microarray technology to profile prostate cancer samples and being in a pathology department I had good access to well-annotated biospecimens. We had discovered in 2005 that about 50 to 60% of prostate cancers harbor recurrent gene fusions or translocations of TMPRSS2 to ETS gene fusions. But we soon branched out of prostate cancer, because we were sequencing patients with advanced cancers and we began to classify cancers by the various pathways and mutations that they have rather than their tissue of origin. Our clinical sequencing program, MiOncoSeq, has taken us into a number of different types of cancers, so we've published in breast cancer, lung cancer, a number of rare cancers as well, although our home base is in prostate cancer.
How did you get into RAS and RAS genes?
We discovered one prostate cancer cell line and some rare prostate cancers that have a rearrangement of KRAS that we reported in Cancer Discovery. We also discovered that about two to three percent of castrate-resistant prostate cancers harbor fusions of the RAS pathway, both BRAF as well as RAF1, or CRAF, and so these are activating fusions of those pathways that are equivalent to the ETS fusions that we described earlier in prostate cancer. But these are RAF fusions that drive an aggressive form of prostate cancer, but only a subset, maybe only 2 to 3 % or so of castrate-resistant disease. So that was our foray early on with the RAS pathway in terms of prostate cancer.
Are those RAS genes wild-type?
Yes, but the RAF genes are wild-type but truncated, as they lack the N-terminal inhibitory loop, the RAS-binding inhibitory loop.
How did you discover the interaction of RAS with AGO2?
About four years ago we decided we would try a flier project, where we would try to look for protein-protein interactors of RAS that we would identify in cancer cells using mass spec technology. The idea was that if we found some novel interactions we could then maybe design small molecule inhibitors that might inhibit that interaction and potentially affect RAS function. So we pursued it somewhat naively in that way, looking for things that we could potentially discover in terms of RAS biology, but then also that we could down the road maybe therapeutically modulate.
Why weren't you dissuaded from this well-trodden ground? This had been done years before in multiple labs.
Yes it had and that's probably one of the reasons that we had quite considerable pushback in terms of our publishing our paper. We were pulling down and doing mass-spec-based analyses of a number of oncogenic factors, and we had a general approach of going after the endogenous molecule rather than trying to use overexpressed tagged molecules. We happened to use an antibody that bound to the switch 1 domain of RAS, and it was a very good antibody to pull down RAS in an endogenous setting. That antibody happened to block some of the other effectors of RAS including RAF and PI3 kinase, so those could not potentially interact with that molecule. We used that antibody in a series of cancer cell lines, pancreatic and lung and others, and in all the cell lines when we pulled down RAS we were able to pick up the AGO2 interaction as the top hit.
It's quite remarkable that the signal was so strong.
Exactly. That is what pushed us to recapitulate that in multiple different ways using multiple different antibodies in both cell lines as well as in tissues using both antibodies against RAS or antibodies against AGO2. We showed that it was a very strong interaction and that we could maintain the interaction even in high salt and high detergent conditions and so forth. It was a very robust finding, and we think that others missed it for a number of reasons. I think many of the RAS pulldowns were done using epitope tags. We've found that if we put the epitope tag in the wrong spot then it creates some steric hindrance where the AGO2 interaction is less apparent. We think it was missed in AGO2 pulldowns because of the small size of RAS, generally people used a cutoff of 25 kD or so.
The binding itself is not dependent at all on being wild-type or mutant RAS is it?
That is correct. We tried a number of oncogenic phenotypes and any sort of downstream signaling of KRAS was all enhanced by AGO2. If we overexpressed AGO2 that would make RAS more oncogenic and we saw a higher oncogenic signal, and if we took away AGO2 by knockdown or knockout that would inhibit KRAS oncogenicity. We used a number of assays including proliferation assays, colony formation assays, signaling assays, pretty much all different types of readouts of KRAS function. They suggested that KRAS “attempts” to inhibit AGO2 function, and overexpression of AGO2 enhances KRAS function, while when you remove AGO2 this inhibits KRAS functionality.
What sorts of follow-up experiments are you doing?
We're trying to recapitulate our findings in GEMM models, both pancreatic GEMM models and lung cancer KRAS GEMM models. We are going to evaluate the KRAS-AGO2 association and its functional relevance. Those models are now cooking, and of course we hope we will get compelling data for the in vivo relevance of the interaction. We're also doing a number of cell biology experiments to explore how AGO2 impacts KRAS function and how KRAS impacts AGO2 in terms of microarray biosynthesis and other readouts. We've got a lot of experiments exploring the biochemical interrelationship of those two molecules. We're beginning some collaborations with Mark Philips to explore the localization of RAS and AGO2. We are doing some co-crystallization work with fragments of AGO2 that we know bind to KRAS, and we're actually collaborating with the RAS Initiative (Dhirendra Simanshu) to do that. We think co-crystals might shed some light onto the interaction and could lead to ways to drug the interaction therapeutically.
How do you see AGO2 affecting the function of KRAS, and vice versa? AGO2 is predominantly perinuclear, correct?
Yes, exactly. We imagine KRAS could negatively regulate AGO2 through its interaction with the N-terminal wedge domain of AGO2, and that could inhibit AGO2's functional role in microRNA biosynthesis. It's in the right location for that. In terms of KRAS signaling it appears that AGO2 binding to KRAS stabilizes KRAS protein stability. Since RAS and KRAS have a presence in the endomembrane of the endoplasmic reticulum, we imagine that their stabilization by AGO2 in that compartment could function as a sort of depot of RAS proteins in the cell.
Is there any KRAS 4A-4B specificity?
We're just doing those experiments to try to figure that piece out.
How would you describe the culture of your research group? Your interests seem quite broad.
I would say that I've tried to instill a team spirit to the group. We work as teams and a lot of our projects are team driven, where we have different expertise brought to bear, whether it be in the area of bioinformatics and computational approaches to more traditional, wet-bench type approaches. We're pretty much focused on cancer but using multidimensional, “omics”-type data, with the idea that that type of data would have some kind of translational impact, either in developing biomarkers or nominating therapeutic targets. That summarizes our general approach but you're right, the topics we've pursued as part of the Center for Translational Pathology can be quite diverse, but generally leveraging data that we've collected from human patient samples.
We watched your keynote lecture at the Cold Spring Harbor meeting on Precision Medicine from last year on YouTube, and looked back and saw that you first published on MiOncoSeq in 2010. You must have been very early into that.
Yes, we were working on this before they called it “Precision Medicine”, at that time it was pretty much called personalized oncology. We were probably one of the first to actually translate these comprehensive sequencing-based approaches to advanced cancer patients in real-time, primarily because we were using these next-generation sequencing based approaches to discover gene fusions in solid common tumors. We had made many discoveries of gene fusions as well as mutations using the technology, and we thought that, OK, we need to move this technology into clinical use. Then in 2011 we published a paper where we sequenced the first two patients on our clinical protocol where we had permission to return results to physicians. We carried out whole exome, whole genome as well as whole transcriptome on two cancer patients, and set up the general paradigm for how physicians should carry out integrative sequencing on cancer patients. We established precision medicine tumor boards, (then called “sequencing” tumor boards, ) and how the results should be returned and so forth, and since we initiated that program we've sequenced well over a thousand patients with advanced metastatic disease.
There must have been a lot of different professions and specialties at the table in order to make that happen, and it doesn't seem like it could have been easy.
Yes, it was definitely a multidisciplinary effort, where we had to bring expertise not only from the conventional individuals who attend these tumor boards, like the medical oncologist, pathologists, radiologists, but we then brought in experts in the areas of bioinformatics and cancer genetics, cancer biology, bioethics and so forth, that had to come to the table to bring this entire project together. Early on there was tremendous pushback to this type of technology by the treating medical oncologist, because they were skeptical about even the idea of matching mutations to potential therapies, but things have changed quite markedly over time.
Do we understand that you might have been in the White House recently?
Yes, I was part of the contingent of AACR members led by Jose Besalga, the president of AACR, as well as Charles Sawyers, the former president of AACR, and I was formerly on the board of AACR. About 15 of us representing different areas of the country met in Vice President Biden's office in early January to describe our thoughts on cancer research, on precision medicine, and so forth. We were certainly pleased when in early February, President Obama introduced the idea of the moon shot effort on cancer, and that they would attempt to devote over $4 billion to that effort.
How do you feel about 2D vs 3D, organoids and that sort of thing?
I think those are exciting technologies, and we were certainly very much excited when the technology came out with the idea of creating these sort of mini-models of patients for precision medicine, where then you could potentially test drugs and drug combinations ex vivo and then perhaps try that cocktail in patients. But I think more recently we felt that they're not necessarily a very viable technology for real-time precision medicine, because generally the take rate tends to be relatively low, depending on your cancer type, so sometimes 10 to 20%, and then often times the media you need for a particular cancer type can be quite complex and different between cancer types and individuals. And the fact that they take a long time to actually grow out, often times months before they're robustly replicating, so that you can actually use them for experiments and so forth, and by then it may be too late to change your approach.
There's obviously a lot of selection going on too.
Exactly, and that's happening in culture, of course, and what's to say that that's a great representation of what's happening in vivo in a patient.
How did you get interested in long non-coding RNAs?
I think it was basically the data itself that led us there. We were using next-generation sequencing, basically the transcriptomic data, to discover gene fusions, but we figured why not take advantage of this relatively unbiased approach to look at not only coding genes but also at non-coding genes. There was a lot of activity around the smaller non-coding genes, meaning the microRNAs, so we figured why not look at the long non-coding RNAs (lncRNAs) as well, so that's how we initially dipped into it. In the course of our early work we called out and named over a thousand novel long non-coding RNAs associated with cancer. We had a paper in Nature Genetics in 2015 that described a suite of different long non-coding RNAs and how you can identify long non-coding RNAs that were exquisitely lineage-specific as well as in some cases exquisitely cancer-specific. So we're quite excited about these lncRNAs as potential biomarkers that could be detected non-invasively and might be correlated with prognosis. But also we think that a subset of these lncRNAs likely have a role in normal biology, as well as in disease processes. We have found a number of lncRNAs across different cancer types that we're quite interested in following up, not only as biomarkers but a subset might be involved in cancer biology.
Can you describe how your sequencing pipeline has changed over the years? Where are you today and how fast is it? You favor whole exome, we gather.
When we began the approach in 2011 we used a combination of low pass whole genome sequencing at 5X coverage, and whole exome sequencing about 100x coverage of tumor matched with normal and whole transcriptome sequencing using polyA transcriptome assessment. After the first 20 patients we found that in 80, maybe 90 percent of cases we could actually find the actionable or informative type of mutations just by using whole exome of tumor matched with normal as well as transcriptome sequencing. As a result, as we've reached well over a thousand specimens we've been primarily focused around whole exome and transcriptome sequencing approaches.
We have certainly made modifications of the approach over time. Early on it took us six to eight weeks from the time of biopsy to the generation of a molecular report for a patient. From the first thousand patients we figured out a set of 1700 genes that we need to monitor quite deeply. We basically expand all of the exons of those 1700 genes, both tumor matched with normal, and achieve over 500 x coverage. Also we have developed a variation on transcriptome sequencing that we call exome capture transcriptome assessment, where we can basically capture RNA fragments. What's nice about our latest generation approach is that we can use it not only on fresh frozen material, which we began with in 2011, but we can now use FFPE blocks that are available in pathology departments. So we get ultra-deep sequencing of those 1700 cancer-related genes and the turnaround time now ends up being around two to three weeks.
What percentage of your samples turn out to be actionable?
Those analyses are actually underway in our adult cohort, with this thousand samples where we've carried out the sequencing. We just published a summary of the first hundred or so oncology patients that we sequenced in JAMA where we were able to do that basic assessment. For 90 percent of the patients we were able to obtain material that we were able to sequence, and for about 45% of patients we were able to detect something that was clinically actionable, meaning anything in the sequence that could change the management of the patient. One of the remarkable findings was that in about 10 percent of cases we found a pathogenic germ line variant that needed to be returned not only to the patient but also to their family because it could impact their incidence of cancer moving forward. And this is a high percentage because these are generally patients without a family history of cancer. This result has been confirmed by subsequent studies by St. Jude's hospital and published in the New England Journal where they also saw the 10 percent pathogenic germ line variant level.
Have you done MiOncoSeq on all comers? You have a substantial pediatric population too?
Right, although because of the issues around minors we had a separate IRB approved protocol for our pediatric sequencing program. But yes, in general our clinical cancer sequencing effort is open to all comers.
Do you want to say anything about your parents or your family or your work life balance?
My wife is an academic cardiologist at Beaumont Hospital so we lead a busy life and we have two kids, 13-year-old and an 11-year-old girls. Pretty much whatever time that I'm not working I try to spend with them and what they want to do.
Do they travel with you when you're on the road giving talks?
I do sometimes have them join me on talks if they're in exotic locations, and if it coordinates with their school schedule and so forth. It's tough to coordinate at times but sometimes I have done that, so that can be fun.