Scientist Stories: Philip Beales, Functional Genomics

Okay, good evening. Thank you very much, one, for the invitation, and two, for coming. I know it's been a long day. So we're going to turn to something a little bit different from a genetics or a genomics conference, if you like, and think beyond the genome. And we've talked about, obviously, the various terms in the past. Functional genomics, I know, is in the program. But I'd like to introduce to you the term multi-omics, really, because it tells you a little bit more about what is going on inside the tin. These are the key learning points. This is an educational session. I won't insult you by reading them out. But just to begin by saying, well, there have been a lot of expectation since the completion in 2003 of the Human Genome Mapping Project. I think it's fair to say that there has been quite a lag before we begun to see any of those particular applications in health. But there's no doubt that there's been an exponential rise in the last few years in how this is affecting us in all aspects of our health, not least diagnostics, and now moving very rapidly into treatment and obviously more precision medicine. And this takes the, takes the form of not only knowing more about us as whole organisms, but also about affecting earlier diagnosis, developing new systems of making those diagnosis in utero, for example, non-invasive prenatal testing, and of course, better understanding of what drives cancer. But then it's also about how does one apply these in order to think about treatments. And what we would say from that point of view is that we've got this huge opportunity ahead of us to use genomics, but other omics as well, to be able to define biomarkers of disease. We already know about the likelihood of responses, but also there will be a lot of advantages of using multi-omics going forward to understand drug efficacy, but also drug safety as well. I don't need to rehearse, especially for this audience, the way in which genetic variants affect all organisms on this planet, but particularly in humans. Everything from the mild trait differences between individuals—hair color, eye color, skin color, and so forth—but also how we respond to drugs or even the type of food that we eat. The influence on disease susceptibility, and of course those diseases that are highly penetrant, that when a mutation arises leads to a specific disease. However, there are some that argue that there is no such thing as a Mendelian disorder with a full penetrance because there are always modifiers. And what do we mean by modifiers? Well, one of the key examples is that a disease like cystic fibrosis, which everyone felt if you had those particular mutations, you will get that particular disease. However, what we now appreciate is that there is a spectrum, a spectrum of severity, and there is now a whole list of different gene variants which are affecting the severity, or the expression, should we say, of that particular gene mutation. And that brings us to the concept of understanding the presence of individual protective alleles within the genome. And these are no less becoming a significant area of interest as far as pharma is concerned. And many of you will know the example of hypercholesterolemia and the discovery of the proprotein convertase PCSK9, or rather mutations that seem to confer protection on a certain group of the population that fall at the lower end of having levels of LDL that are associated with disease, for example. So it's by studying those folks at the lower end of the Gaussian curve, as you can see, whereby it was deemed that if one were to block PCSK9, it might actually have an effect on its ligand function on the LDLR receptor, thereby upregulating the receptor and therefore clearing cholesterol. And this is now the basis of a number of targets or drugs rather, that are being developed by a number of pharma around the world. And the list is starting to grow, and for example, sclerostin inhibitors in osteoporosis and so forth. And this list will undoubtedly get longer. But looking at genomics is undoubtedly very valuable, but it's not all about genomics, is it? We're only seeing part of the picture here. And Eric Topol has suggested, for example, that we should be embracing the concept of the panerome. And what do we mean by the panerome? Well, it means the fact that we are each individually constituted of these different omics. And really what has changed, I think, in the last few years is not the fact that the omics have changed in any way. They've always been there. Our ability to understand omics or various omes has always been there. But I think what has really changed in the last 10 years or so is the fact that we can now measure these at scale efficiently and to a certain extent cost-effectively as well. And there are also— if you can think of a prefix and stick an -ome on the end of it,, you've got an omic. And, you know, there's foodomics that was proposed in 2009 or so. So what examples have there been? Well, the first example was probably back in 2012, the first one to be published at least, from Chen et al., from Mike Snyder's lab. It was really a proof of principle study with a single healthy individual being studied over a period of time, in fact over a 14-month period of time, and it was Mike Schneider himself. And it was interesting because what they decided to do was take blood samples from him, withdraw or isolate the peripheral white cells, the plasma and the serum, also do serial blood biochemistry as well, and also do his whole genome and try to make sense of this kind this kind of data. What happened? There were two specific events during that 14-month period that was significant. The first was a minor upper respiratory tract infection, and the second was a much more severe lower respiratory tract infection due to RSV, respiratory syncytial viral infection. So when they did the profiling and then the pathway analysis of the proteins and the expressed genes, for example. It was no surprise that they saw upregulation of various immune and inflammatory pathways, such as the interferon gamma and so forth. However, about 3 weeks later, what they noticed on his chemistry profiling was that his glucose had shot up into the diabetes zone. And essentially, he— developed diabetes, and the assumption was that the RSV infection was that trigger. Looking back at the genome data, they also realized that he had a family predisposition towards type 2 diabetes anyway, so maybe this type of infection unmasked that. And so of course, integrated profiling again showed dysregulation of those key glucose regulatory pathways. He has since, or as a consequence of that particular event, changed his lifestyle completely. I'm not going to make any comment about what his lifestyle was like before, but he has subsequently taken up, you know, exercise, various other things, but has also gone on to continually measure his health through activity and glucose monitoring. And this is something not for today's discussion, but it's something that's coming to us and really needs to be thought about as an adjunct to multi-omic and obviously precision medicine. So, turning to sort of more serious— that's unfair— but other studies where it has been proposed that multi-omics can be developed in a high-throughput platform manner. This study from Vanderbilt has essentially been looking at lung cancer, and more specifically, the response of a treatment for lung cancer. So it's an in vitro study looking at the function of cisplatin over a 30-day period and taking measurements of the proteomics, the metabolomics, and the transcriptomics. And trying to make sense of those data going forward. And in a nutshell, they found two things, the first of which was a global mechanism for the effect of cisplatin which wasn't known before, which leads to apoptosis. But also by integrating those various omics data, they were able to go from an existing network of known proteins taken from the KEGG pathways, for example, of 2,560 nodes, but adding— added to that other off-target cisplatin effects. And the proposal within this study is that this increase in information about not only the effect of cisplatin, but also the physiology relating to the effects of this kind of treatment, could improve the selection of candidate therapies going forward. So we were really interested in these type of concepts as far as applying this to the area which I have been studying for over 20 years, and that's the area of rare diseases. You'll all know that rare diseases are actually common when you put them all together. More than 300 million people worldwide have it. But the key fact is there are only treatments for around 5% of these people, and that's clearly one of the things we are particularly interested in trying to move forward toward. So we developed a program a couple of years ago at our institute called the HiFi Program. It started out with 5 diseases. We now have 7 diseases in that, which I'm going to tell you a little bit about. So we probably ought to call it the 7-Up Program. Program, and it's really a pilot study to establish strategies going forward, SOPs and all sorts of things on how you should or how you should not undertake a multi-omic study. And we wanted to answer 3 kind of key questions, if you like. The first was, can we explain phenotypic variability that all of us who do see patients with rare diseases We all know that there are individuals within families who will have exactly a sibling who may have exactly the same mutation as their brother or sister but have a completely different profile of their disease in terms of severity, onset, and so forth. Secondly, could we find biomarkers that could be used going forward, disease progression, treatment response, and all of that kind of thing? And thirdly, could we identify key molecular pathways and modifiers that could lead us to looking at— looking for new targets for therapy. So these are the 7 diseases. I won't go into the details around them because we're only going to focus on a couple of those. But they're relatively— we had a relatively— this is a pilot study, so we had a relatively small number of individuals involved. But these are the 2 diseases I'm going to focus most on. And these are the key aims. So first of all is a form of inflammatory bowel disease that affects infants. It's significantly different from what we might expect in adult forms of Crohn's and ulcerative colitis, very early onset inflammatory bowel disease. And these children have, if not treated or if not treatable, These children have a significantly terrible life in the sense that they're, they're dependent on either total parenteral— well, total parenteral nutrition. And the treatment is, is very, very difficult. So we wanted to see whether we could identify further markers of response to the therapies that are out there, but also to better characterize the pathology of the disease. The second one is Bardi-Beadle syndrome. I'll tell you a bit more about that in a moment. And the purpose of that study was to look at— look for gene modifiers, predictive markers of disease complications such as the renal failure that those patients get, and so on. So what we did for all the patients that were recruited to this study, we took skin biopsies and we took blood samples. And from the blood samples, we isolated plasma serum, the cells from which we extracted DNA, RNA, proteins, and we also collected buckets and buckets of urine. And from those, we applied all these multiple omics. And we used no fewer than 8 different platforms to try to generate the mountain of data. I don't expect you to be able to read these. Oh, actually, it's not that bad. I can't read it. The— this just literally shows you a list of the types of tissues that we have collected and the sample number, but it's better illustrated in this infographic here. So for those 7 rare diseases, just under 200 patients, collecting all those various omics data, which now amounts to around well over 50 terabytes of storage, not to mention the amount of compute that we require as well. So it's really compute intensive and really does require a lot of coordination when it comes to driving these type of projects forward. And therein lies another challenge, is then how do you analyze the data going forward? So we adopted a three-tier approach, the first of which was to do pairwise correlations within the platforms as such, so looking at gene expression, looking at the protein profile of individuals, and looking at genome variants for anything that stood out, and also undertaking burden analysis as well to see whether there were potential modifiers in those. The second tier is then to try to make sense of some of these, perhaps reaching through from the expression data into the protein data, see which— what has been translated, and undertake pathway analysis going forward. And then the tricky bit, which you'll be the judge of whether we've been successful at all, is to integrate and model all of this data in a clinically relevant manner. So here's one example of a heat map of expression data from transcriptomics. This is an inflammatory disorder that we've been looking at as well, just to illustrate the point. And in this situation, we've looked at peripheral blood monocytes and then divided those into T helper cells, CD8 cells, cytotoxic T cells, monocytes, and also B cells. And then look to see what is the profile like in terms of expression before and after treatment of that particular disorder. And we see major upregulation, as we might expect, I guess, in interferon alpha and gamma responses, although the clinicians actually haven't even thought of that particular particular pathway. So that's one thing they've been able to discover. But also we can see that for a subset of these, particularly the T cells, there are a certain set of pathways that appear to be upregulated and therefore are now something we need to look at going forward. And this could represent further targets. We can turn to the plasma and look at the protein profile, for example.. And we can see again upregulation of some of the same pathways that we are seeing in the transcriptomic analysis. And we can go back, take a step back, and look at the genome data that we've collected on these individuals. This is what's called a Circos plot, and it just happens to be set out by chromosome, which isn't necessarily that helpful, but you've got to start somewhere. And it's a bit like an onion. So you've got the variants on the outside, you've got the expression data in the middle layers, and then you've got the protein data in the middle. And you can see that if we take one particular target area, this is agnostic of any kind of function, but say chromosome 1, as you can see here, there are some interesting single nucleotide variants, which also appear to be differentially expressed. Rest as well and reach through to the protein level. So we can systematically go around the genome and start to look for targets in that particular manner as well. So turning back our attention to the very early onset bowel disease in these 20 or so children we looked at, our gastroenterologist at Great Ormond Street Hospital, Neil Shah and Joachim Kammermeier, did endoscopic analysis on these children. And what they did, they took 5 biopsies, or rather more than 5 biopsies, but at 5 different sites throughout the bowel, as you can see here, from the small intestine starting with the duodenum, end of the duodenum around the ileum, into the large intestine, ascending, transverse, and descending colons. And again, transcriptomics was undertaken, proteomics was undertaken, and whole genome sequencing was undertaken as well. And we can do all sorts of treatments of those particular data. We can look at hierarchical clustering. And we can see very clearly that this does cluster. And for example, what we're seeing here, the red and the orange represents the the small bowel, and these green and purple in the middle here represents the large bowel. So probably what you might expect in terms of expression. We can do other fancy things, look at various principal component analyses, and see clustering in that particular way with certain components, but not in others. And we can go and do more gene set enrichments that I showed you earlier on, and this is beginning to show us and reveal some significant targets going forward. For example, what we've been doing here is comparing— because our problem is we cannot find suitable pediatric controls, you're just not going to find children who are going to succumb to a colonoscopy for the hell of it. So Essentially, what we've been able to do, or rather had to do, was to compare inflamed portions of the bowel with non-inflamed. So it's not perfect. But nonetheless, what we do see is significant clustering and upregulation of certain pathways, which again, we need to explore in an ongoing fashion. So the second disease is something completely different, Bardet-Biedl syndrome. It's a so-called ciliopathy. It's a recessive condition, a rare condition, and it results in early onset blindness for these children, varying degrees of learning difficulties, and early onset obesity, which is lifelong and associated with lots of other problems. And the other key feature of the disease is kidney failure, and due to cystic kidney disease. And we now know that there are around 21 different genes which are mutated in these particular patients, and we know that those genes encode for proteins which are important for regulating trafficking within the cilium itself, what's called intrafagellar transport. So we took 18 patients to apply this same multeomic treatment. We looked at— we were hoping to look for gene modifiers and mechanisms of disease and ultimately new targets. So multiple things we've done. For example, we've been able to look at proteomic analysis of the patient's urine versus control to see, can we find any biomarkers within within that particular data set. And sure enough, we have 2 or 3 very strong hits that are up to 5-fold upregulated, which we think are probably likely to be biomarkers that could be associated with renal disease, impending renal disease, in these patients. So that's one aspect. We've looked at the genome data, and we've identified new signals such as Apolin that might be of key interest. It's a G-protein coupled receptor ligand that's expressed in lots and lots of tissues that could be involved in oxidative stress as well. So that could possibly be another target to look at. But what we decided to do was actually have a look in more detail at an in vitro study of cells these are the ones taken from the skin biopsies from these patients, from which we grow fibroblasts. And by serum starving for a few days the cultures, you can force ciliation—cilia formation. As you can see here, there's a little tiny long cilium that's forming on this fibroblast, whereas these guys don't have any cilia at all. And then look to see what is happening in terms of your transcriptome. Omics and hopefully your proteomics. And sure enough, serum starvation does lead to upregulation of these various component parts of the cilium, the so-called intraflagellar transport proteins. But there wasn't a significant difference as far as we've looked at so far between the controls and the Bardi-Beadle syndrome deficient— protein deficient patients. But we do need to dissect this a little more, but it's a starting point. But if we do go and look at the upregulated gene sets, again, we can see— we've already seen those IFT proteins, but if we go and start looking at a different set, which is the other BBS proteins— I mentioned there are 21 of these— we start to see there are signals being upregulated again here. And again, if we look at— if we expand that out to other ciliopathy genes and proteins, we start to see that signal as well. But if we go too wide, we lose that signal. And then the last aspect of BBS that we've looked at is going back to the genome to see whether we can find any ciliary modifiers, or rather modifiers of disease altogether. So what we're seeing here is 15 different cases that have all been genome sequenced. And what we're seeing here is if you have two bands here, you're homozygous for this very common mutation in the BBS1 gene. If it's— if there aren't two, if there's just a single one, it's accompanied by— it's a compound heterozygous situation accompanied by another mutation somewhere in BBS1. So for each of these 15 patients, we know exactly what the driver mutations are, the bi-allelic mutations. The question is, are there other variants in other genes that are related to cilia function that might actually be impacting on those particular phenotypes? So we focused, for example, on these 3 individuals here who are 3 siblings. This pair here are monozygous brothers And this is their older sister, who's actually much more— much less severely affected, should I say. She's still working as a nurse, and she still has some useful vision. These two boys have no vision whatsoever, and, um, and they have a lot more problems in terms of kidney failure and so forth. Question is, can we find something in the genome that might or could possibly be affecting this severity in terms of onset or progression of the disease. So it's a start. We don't know just yet, but we found a couple of variants in PKHD1, a disease that when you have two mutations, significant mutations in that particular gene, you'll end up with something called autosomal recessive polycystic kidney disease. So that's one way of looking at possible modifiers. We expanded the study to look at several other genes as well across the whole of the 15 different patients. And we found variants, or a burden of variants, sat within another gene called TTC21B. It just so happens that we've already published this with Nikko Katsanis's group and Eric Pierce a number of years ago, back in 2011, where we were able to show that single variants were both causal and modifying of other ciliopathies across that particular spectrum. So again, we may have picked up something here that could potentially be a modifier. We can then take all of the data, as we've been trying to do— it's not easy— and see whether we can model networks. This, we've taken transcriptomic and proteomics data and look to see by perturbing some of those particular nodes whether they have impacts on any of these other proteins. And, and, and they certainly do, but that's work for the future, and we need to look at the relevance of these going forward. So it's quite clear that data integration is absolutely key if we're going to understand biological organisms as a whole, and particularly how disease is going to be understood. And it's going to be really important for us to develop ways of combining those data as well as understanding them. But multi-omics data, just collecting it on its own in isolation, is completely useless unless you have very good phenotyping information going forward. And so just as an example, this is a summary of the data we collected for the IBD cohort, and you can see already huge amounts of information, 250 Human Phenotype Ontology terms, and somehow we have to integrate this data going forward. We can do all sorts of fancy plots looking through— this is the BBS cohort, for example— matching groups of patients through to various HPO terms. And again, we then have to take that through and try to combine it with the multi-omics data as we go forward. So how is this happening at scale? Well, it already is. I mean, the study from Lee Hood's group in the Institute of Systems Biology, for example, a number of— a few years ago in 2014, completed an analysis of health of about 100 healthy people where they collected proteomics data, genomics data, epigenetics, microbiomics data, on these individuals. And they concluded that there was not a single person in there who had some kind of actionable opportunity from that data. And he's now obviously taken this out to a much larger study called the 100,000 Wellness Project, which has now been spun out as a company called Harry Vale. And you can argue till the cows come home whether this is something should or should not be done, but it's happening as we speak. And it's also happening at a much larger scale if we think about it. The Precision Medicine Initiative announced by the former president— and a lot of people say please come back— he had the vision that there would be a million participants where multiple omics would be collected to better understand health and disease for individuals going forward. And if anything, multiomics needs to be heading towards the area of precision medicine and to help define targets for new therapies. And of course, pharma is extremely interested in this particular area, particularly so in the area of biomarker discovery. And stratifying patients for better clinical trials going forward. There are a number of challenges, of course. And for us, the challenges were that we had small cohorts. We have a lack of pediatric controls. And we're now trying to correct that by having a national program to try to bring in pediatric controls or several samples, but you're not going to get those for a lot of tissues. That's always going to be an issue. The whole concept of computationally analyzing this data in a comprehensive way is still yet to be tackled. There are no solutions out there. There are companies working on this, but I'm yet to see a really good tool. We will have to see. There's the issue of scalability. Can you scale a lot of these kind of different omics? Because as you know, mass spec takes up most of a room, so very difficult to, to think how that might work. And then of course there's cost-effectiveness. So this is my last slide, simply to say that it's not just about the individual and our ability now to measure all of that biomolecular data that makes us human, but also to interact this information with our environmental data, which we're now all carrying various wearables. And all of this information that's being collected, we need to think about how we can integrate this going forward. So I thank you for your time. I just acknowledge the key people who have done this work, particularly Dan Kelberman, who's been our project manager, who's done an amazing job to coordinate the huge amount of data, as I've shown you, that we've collected, but also to make sure we know where it is. That's a big issue. And finally, just to say that we are working on a solution in partnership with a company called Congenica, which is a genome analysis company, with the UK government, a UK government grant we got a year ago, to try to develop a software solution to bring all this data to the fore and to try and integrate it in a sensible manner. And I'll leave it there. Thanks very much.