Scientist Stories: Anna Gloyn, Genomics and Causal Mechanisms for Diabetes

I'd like to do is share with you some of the work that we're doing in Oxford where we're trying to capitalize on human genetic discoveries largely coming from genome-wide association studies and to turn these into biological insights and mechanisms for type 2 diabetes. Now, I'll start off with just giving you a little bit of introduction on the collaboration that's presented on this slide. This is a collaboration between Philip Felicity Cormack, who's a wooden block artist, and Katya Mattis, one of the students in our team. And this is Felicity's representation of Katya's DPhil project on CRISPR genome editing of human pancreatic beta cells. And I think it's absolutely beautiful, and I'm shamelessly going to use it throughout my talk as a little break between the vignettes that I'm going to share with you this afternoon. So for those of you who don't work on Type 2 diabetes, let me just point to you why it's such an important disease area to be focusing attention on. There are currently 415 million people affected by diabetes worldwide, and it has an estimated cost in the UK of over £25 billion. It's the largest single cause of amputation, stroke, blindness, end-stage renal disease, and nonalcoholic fatty liver disease. And the big problem that we have is although we can treat diabetes, the treatments that we're currently using in the clinic are not disease-modifying. So there's a clear need for novel and targeted therapeutics. But the major barrier to achieving this is that we have a pretty poor understanding of what causes diabetes above and beyond the fact that the cells in the pancreatic islet are unable to produce and release sufficient insulin, and the cells in your body that need to respond to that hormone are unable to do that optimally. So the consistent focus of my research over the last almost 20 years has been to try and capitalize on experimental genetic variants that we know either cause or increase your risk of developing diabetes. So initially, I worked on monogenic varieties of diabetes, but more recently, I've been trying to use the techniques and expertise I learned in that space and apply them to the field of genome-wide association studies. So in each of these cases, we have genetic variants that we know cause or increase your risk of developing diabetes, and my work is focused on using these as tools to interrogate cellular and molecular mechanisms for pancreatic beta cell dysfunction. I'm also interested in understanding the impact of these variants on human physiology. Ultimately, what I'd like to be able to do is to translate new information on diabetes pathogenesis into more effective treatment strategies, more effective uses of existing therapeutics through patient stratification, and also the detection of biomarkers that help us to aid genetic diagnosis and variant interpretation. So we have pretty good evidence that this is likely to succeed, and that comes from a number of sources, but the one closest to my heart is work that I performed as a postdoctoral researcher in Andrew Hattersley's lab in Exeter, where I identified heterozygous activating mutations in a key component of this machinery that couples glucose metabolism to insulin secretion in the pancreas. Pancreatic beta cell. And through identifying and understanding what these mutations did in children like Lily, we were able to transfer them from insulin injections to oral sulfonylureas with improvements in their glycemic control. So this was an example of not only precision medicine, where understanding the genetic basis of the disease led to a change in how we treated and managed these patients, but it also showed us that human genetics can identify effective and safe therapeutic targets because sulfonylureas are commonly used to treat patients patients with type 2 diabetes. So over the last 10 years, as we've heard at this meeting, there's been tremendous progress in unlocking the biology of lots of complex diseases through genome-wide association studies. But we're faced with a huge challenge because most of the variants that have been identified don't sit neatly in protein coding regions of the gene where we can move rapidly to functional assays. They're sitting in those regions that are poorly annotated, and we've got a huge degree of transcript uncertainty. And this makes it very difficult to unlock the biology. For type 2 diabetes, we've had a lot of success. We've already heard from Anooba at this meeting that my slide's out of date. There are published at the moment over 150 regions of the genome that we know robustly associate with type 2 diabetes risk, and Anooba's told us about 141 more. And most of these, as I say, are in that non-coding sequence that makes it very difficult to move to a functional assay. So what we have learned, though, from the data that's been generated thus far is that the pancreatic islet is absolutely central for mediating the effects of those variants. And we know that because we can study the impact of these associated variants in individuals who don't have type 2 diabetes, and we can understand whether or not they impact on the procedures and processes that influence insulin secretion or whether or not they impact on the processes to do with how your body responds to that insulin. And through studying those, the physiology of these variants, we can see that the vast majority of them impact on pancreatic beta cells. And that's really important because we have a handle on the cell type that we need to be working on and focusing our downstream efforts. So I'd like to give you a flavor of how the work in my team is set up and the sorts of approaches that we're taking in combination to try and unlock the biology from GWAS. So we really ask two fundamental questions. First of all, can we get to those affected transcripts, and can we identify the causal variants? So to understand the affected transcripts through which these GWAS variants are mediating effects on disease, we tackle this using a number of complementary approaches. We might use our known biological insight, looking at biological candidates that are mapping to that region of association. Working with Roger Cox mainly at Harwell, we can generate rodent models and see if they recapitulate the appropriate diabetic phenotypes. We can continue working with Mark McCarthy and Andrew Morris and others to define mapping efforts in human genetics and see if we can get down to smaller sets of variants. We can study these individuals in healthy carriers and look at the impact on physiology. And we can do appropriate cellular studies in relevant cellular models. We're also interested in trying to get to that causal variant, and there we do studies looking at effects on gene expression in relevant tissues. We look at improving the annotation of human pancreatic islets so that we can try and understand what variants might be doing. And as you heard earlier from Jim Hughes and from Wieden Eyelander, we're also interested in looking at chromatin interactions using CAPTUR-C. When we do get at those effector transcripts, we're able to start to do things like gene set enrichment analysis and look for protein-protein interactions and network analysis. And you heard also this morning from Caspar Laje at the Broad about his elegant studies using co-immunoprecipitation and starting to put together pathways. And that's something that we're fortunate to be able to be working with him on at the moment. So the first vignette I really want to focus on is to really tell you something about the efforts that we've been taking over the last 5 or so years to try and identify effector transcripts at GWAS loci by looking at effects of of genetic variants on gene expression. Now, we've heard today already about the publicly available resources, GTEx and ENCODE, which have been phenomenally important in trying to understand effects on gene expression and effects on regulatory annotation. But unfortunately, my cell type of preference, the pancreatic islet, is not represented in either of these efforts. And recognizing this some years back, we started to try and make our own equivalent version in Oxford and put down similar annotations and gene expression dataset. On this absolutely critical cell type for diabetes. So now we have about over 200 human pancreatic islet donors. And in those cases, we've been able to look at islet morphology and function. We've been able to genotype them, look at DNA methylation status. We're increasingly doing ATAC-seq and looking at chromatin state. We've been able to improve our understanding of the annotation of regulatory regions by doing transcription factor binding. And we're doing CAPTUR-C. We've on all of these islets been able to do RNA-seq, originally using Illumina, but we're moving now to using PacBio so that we can start to get a better handle on some of the isoform expression profiles. And we've been dabbling, as many have, in the single-cell transcriptomics field. So this human islet biobank is at the heart of a lot of what we're doing in Oxford, and it is an important collaboration between Mark McCarthy and myself in Oxford with Patrick Rawlsman, also in Oxford, who's a beta cell physiologist and Patrick MacDonald over at the University of Alberta in Edmonton, who is also an electrophysiologist and has access to human islets. So we published at the end of 2005 our first endeavors at trying to identify cis-eQTLs for type 2 diabetes. And this was work that was led by a postdoc in the lab at the time, Martin van de Bunt. And we published this work on the first 118 donors that we had got RNA-seq and Omni 2.5+ exome genome array genotyping on. And what Martin was able to do was to show that in 14 cases at 14 different GWAS loci, the lead SNP at the GWAS locus overlaid an enhancer element that was enriched in pancreatic islets, and there was an association of that SNP not only with diabetes but also with regional gene expression. So in 14 cases, we were able to get at a putative effector transcript at that locus. And these fell into 3 major types of finding. In some cases, we were simply confirming an eQTL that had been picked up in a disease-non-relevant tissue such as blood, and we saw that, for example, with CAM kinase 1 delta. In other instances, we were able to provide experimental evidence for something that was already a compelling biological candidate such as ADSCI5. But in a really interesting subset of cases, we were actually identifying genes that had got no known role in islet cell biology. And this was the first evidence that this might be playing a role in insulin secretion and beta cell function. So what I'd like to do is tell you about one of these in particular, ZMIS-1, which is a transcription factor. As I say, no previous known role in islet biology. And what we were able to do was to firstly establish that the direction of effect was that the type 2 diabetes risk allele also increased expression of ZMIS, so that if you happen to harbor a copy of the type 2 diabetes risk allele, you had more ZMIS expression in your pancreatic islets. So the key question that we needed to know was what is ZMIS and what does it do in a pancreatic islet? And if you have too much of it, does it have a detrimental effect on your ability of your islets to secrete insulin? So working with Patrick MacDonald at the University of Alberta, first of all, we were able to show that ZMIS is indeed expressed in the human pancreatic islets. Islet at the protein level. We can see it in both the pancreatic beta cells and in the alpha cells. And then we wanted to ask, well, what happens if you have too much of your ZMIS in the islet? What will happen to insulin secretion? So Patrick overexpressed ZMIS in human pancreatic islets and was able to show that if you have too much ZMIS, you reduce your ability to secrete insulin. So directionally consistent with our CissE QTL and association with type 2 diabetes. So increasing ZMIS in your— in a human islet reduces insulin secretion. So this is the first evidence here that ZMIS is playing a role in pancreatic islet biology. So encouraged by the findings from just 118 islets, we wondered what would happen if we started to bring larger numbers to the table. So as geneticists, we reached out to our friends who had similar datasets, including Leif Group at the University of Matthias Lund, who had a similar size dataset, Manolis Demetakis in Geneva, who had a subset of around 30, and Francis Collins and Michael Stitzel over in the States, who had a similar size number of islets. And we took all of these islets plus unpublished data, which gave us around 420 human islet donors, which is a pretty large study for this rare tissue. All of these samples have been reprocessed and put that through the same pipeline. And we're now in a case where we have been able to identify further cis-eQTLs. We've got 20 putative affected transcripts across 18 of our type 2 diabetes loci. And we've got 8 putative affected transcripts across 5 loci for glycemic traits. And really interestingly, for the first time ever, we've been able to show the first evidence of expression level for TCF7L2 at this locus that was one of the original loci discovered by Decode back in 2006. And it's been an area which has been hotly debated with evidence suggesting that maybe it was a nearby gene. And for the first time now, we can say there's an effect on TCF7L2, which is really exciting. One of the things that's been discussed today in a number of talks is about the extent to which you can use other tissues to pick up disease-relevant CECUTLs. And one of the things that we've been able to look in here by comparing our data in islets with GTEx is the extent to which our signals would only have been picked up in human islets. And although we see that islets are the most abundant tissue for our CECUTLs, it's certainly fair to say that we're able to pick up many of them in other tissues, with TCF7L2 being a rare exception. So you heard earlier, both first from Jim, who took us through his next-generation capture C method, and then from Vebe, you heard about some of the efforts that we've been taking to try and use this approach to, again, add another layer of data to our islet biobank and to be able to link regulatory variants to the genes through which they are affecting. And Vebe told you this morning about her efforts from the INSPIRE Consortium focusing on 21 promoter regions, and we've now also taken this out on a further 289 SNPs that we've picked up across the GWAS type 2 diabetes catalog. So that's the efforts that we've been taking trying to use our omics type 2 diabetes islet biobank to get at the underlying biology. What about other approaches that we might choose to take that would help us to have certainty over the proteins in play and the mechanisms that are occurring? So another thing that we've been doing is asking the question, are there coding variants at GWAS loci that are associated with type 2 diabetes risk? Because clearly this would give us an immediate advantage because we could start to work on relevant assays. So to do this, we've been part of an international effort to try and get a complete picture of exome variation for type 2 diabetes. And this was the Type 2 Diabetes Genes Consortium led by David Altshuler, Jose Flores, Mike Benkey, and Mark McCarthy. And it was published last year. It was exome sequencing on 13,000 individuals. With around 6,500 with type 2 diabetes and 6,500 healthy controls, and it was drawn from a number of different ancestries. It was supplemented with exon array genotyping with up to 80,000 European individuals. So this delivered a large number of novel signals, including a few that were mapping to coding sequence. And one of them that caught our attention was one that had been first published by Marco Laxo, Karen Mulkey, and colleagues a couple a coding variant in a gene called PAM that had been shown to be associated with a measure of beta cell function in a healthy population. And then Kari Steffensen and the Decode team were able to show in a follow-up study that the same variant not only impacted on beta cell function, but also influenced type 2 diabetes risk. They resequenced the gene and found a rarer variant present in around 1% of the population and were shown that this had an independent effect on type 2 diabetes. Now, the Type 2 Diabetes Genes Consortium found the same results replicating this finding, so we had robust evidence at a statistical level that we had independent coding signals in the PAM gene associated with beta cell dysfunction and diabetes risk. So what do these— what does PAM do? Well, PAM encodes for an enzyme called peptoglycine alpha-amidating monooxygenase. It's the only enzyme in your body that is capable of amidating glycine-extended neuropeptides to increase their biological potential. Potency and activity. And it's really important because in doing this, it can increase the half-life of proteins that are going to impact on biological function. So the first question we wanted to know in the lab is we've got two independent coding variants. What do they do to the protein? Do they increase? Do they cause a loss or a gain of function? So Anne Raimondo, a postdoc in the team, set out to establish novel assays to determine this. She set out a— set up a plate-based assay from scratch to measure PAM activity and was able to show that the 5% variant was associated with reduced ability to amidate a glycine-extended peptide. She was then able to show for the 1% variant that it resulted in an unstable protein both in HEK cells and also in a human pancreatic beta cell line, Endo-C-Betrah-1. So we had two independent coding variants associated with diabetes risk, and both of them showed loss of function, but through two distinct mechanisms. One through reduced catalytic ability to amidate neuropeptides, the other through reduced and mislocalized protein expression. So what is PAM doing in a beta cell? How would it result in beta cell dysfunction, and how would that impact on diabetes risk? So to address that question, Søren Thompson, a PhD student in the group, set about to knock it down in a human beta cell line and to see what happened to the cell's ability to respond to glucose. And if you look here at this— these results, you can see that first of all, when Søren knocked down the PAM gene, he was able to see a decrease in insulin— glucose-stimulated insulin secretion, a decrease in insulin content,. And when Benoit Hastoy, an electrophysiologist in the team, set about to characterize these cells, he could see that there was a delay in insulin exocytosis. So reduced in the first stages and then delayed throughout. So we followed this up at the ultrastructure level by doing some EM and were able to show that there was indeed reduced insulin content in the granules of these beta cells. And we then wanted to know, well, what could be causing this? How is PAM reducing insulin content? So we looked at the secretome in a human beta cell to see what was ammendated, and we're excited to see that chromogranin A was one of the proteins that PAM is ammendating to increase its biological activity. And chromogranin A plays an important role in the insulin granule in packaging the insulin correctly. So we wondered whether or not in beta cells the loss of PAM function was affecting chromogranin A, and then this was influencing insulin granule content. So to test this, we used antibodies specific for the amidated and the non-amidated varieties of chromogranin A, and we were able to establish that both when you catalytically inhibit and when you knock down PAM that you see an increase in the glycine-extended variety of chromogranin A., so that there's less amidation. And then when we knock down chromogranin A in the same cells as we'd knock down PAM, we were able to see exactly the same phenotype. So chromogranin A loss recapitulates PAM loss in the same cell model with reduced insulin secretion and reduced insulin content. So PAM is clearly an important protein in a human pancreatic beta cell. And if you lose it, you get an effect on insulin release, on insulin content, insulin exocytosis. And we think that one of the ways that this is happening is through effects of PAM on chromogranin A, which affects the packaging of insulin within the granules. So that's all very well. That's in a human beta cell model. I think it's pretty compelling evidence. But can we get closer to carriers of the, of the variant? And the answer is yes, we can. We can go back to our omics database for human islets, and we can talk very nicely to our collaborator in Alberta and ask him to have a look at his functional data on the islets that we have genotyped. And when Pat went and had a look at the data that he collected on our genotyped islets, he was able to identify 7 individuals who had the 5% variant, and he was able to match them very carefully for age, gender, and BMI with non-carriers. And then to look at the impact of this variant on exocytosis. And as you can see here, Pat was able to show the same reduction in insulin exocytosis in islets from individuals carrying this 5% variant. And over here, just to remind you, this is the data from our endo-C beta-H1 cells, and you can see that we very nicely mirrored that effect. We have a variant. It's associated with beta cell dysfunction and type 2 diabetes. We can show pretty compelling evidence that there's a direct role for PAM in pancreatic islets. We can show effects on insulin content, secretion, and kinetics. But PAM is actually pretty widely expressed, and there are other ways by which PAM could be influencing insulin secretion. It could be having indirect effects. It could be amidating other peptides in the body that are then traveling around to the pancreas to the islets and are influencing insulin secretion. And there's one elephant in the room, because there's a very well-known peptide to the diabetes community, because we use it— GLP-1 agonists, they're 30% of the non-insulin market for treating patients with type 2 diabetes, and GLP-1 is amidated. And what GLP-1 does is it's released from the L-cells in your gut after you've had something to eat, and it helps to help you reduce your blood glucose levels by increasing insulin secretion and suppressing hepatic glucose production. Now, PAM is expressed in the L cells, and GLP-1 is amidated in the L cells. So we wondered whether or not individuals who have the PAM risk genotype could have a reduced incretin response due to reduced amidated GLP-1, and this could could also contribute to their inability to respond appropriately to glucose. So to test this, we realized we weren't going to be able to do it in cells. We needed to go to a good model system, and my favorite model system is the human. And we're very fortunate in Oxford to have the Oxford Biobank, where we've got over 7,000 individuals who have been consented to be reapproached, having been genotyped already, to take part in further studies. So we looked in the Oxford Biobank for individuals who harbored the 1% variant, and we invited them to come into our clinical research facility for 2 visits. On the first visit, we came in and we gave— they came in and we gave them an oral glucose tolerance test. And on a second visit, we got them to come back in and we performed a matched isoglycemic control. And this is the gold standard measurement way of measuring the incretin response. This was work that was performed by Mahesh, and Mahesh did this for 20 individuals with the PAM 1% variant and 20 age, gender, and BMI-matched controls. And if you look up here, you can see that, first of all, very reassuringly, in an assay that Mahesh developed, he was able to show that if you had the 1% variant that increases your risk of diabetes, you have reduced PAM activity. So nice proof that our recruit by genotype study is working. He then had a look at what happened to the incretin response in the two— in the groups of cases and controls, and he saw no difference on the incretin effect. So the evidence that we have at the moment does not support that the GLP-1 axis has been influenced by the PAM variants, and it looks as if the effects on beta cell function are mediated directly. But of course, there are other tests that we should and will be performing. So moving on, I'd like to sort of shift gears a little bit and show you how we can learn from these experiments. Well, clearly, we've got a model, a human beta cell model that's working very effectively. We've got good cellular phenotyping, but we've only really in this extensive study tackled one of the many GWAS loci. Wouldn't it be lovely if we could do this across the genome or at least across a large number of GWAS variants. So a few years ago, we asked the question, were we in a position where we could start to assay gene function at scale in our human beta cell model? So we wanted to be in a position where we could take the 75 GWAS loci that were known at the time, which if we took a cutoff of 1 megabase either side of the lead SNP would give us those 300 genes at those 75 loci? And could we develop a way of measuring insulin secretion in a way that the Wellcome Trust would fund me to do? In other words, could we drill it down to something that was more like 20 pence rather than £3 data point? We needed this human authentic beta cell model. That was absolutely central to our ability to do this work. And we needed to be able to efficiently knock down the genes. And fortunately, all of these things coincided. At the same time and gave us the ability to address this question. As many of you will appreciate, when you have a big project like this, you need a DPhil student. You need somebody at the start of their PhD who's prepared to invest in a high-risk project. And fortunately, Søren was up for this, and he took the mantle and worked with Daniel Ebner at the Target Discovery Group in Oxford to develop a robust and scalable array-based -based silencing assay using acoustic liquid handling, which allowed us to get the price down of our insulin assay and also to hit that magic spot with our Z-score so that we had confidence in the secretory measurements that we were generating through the high-throughput screen. We wanted to be able to look at a cellular readout under a number of different conditions. We looked at low glucose, high glucose, and we wanted to pharmacologically modulate ourselves so that we could try and interrogate other pathways which might be perturbed by a knockdown of the genes. And we did this with a fairly dirty method with IBMx going through the cyclic AMP pathway and also with tolbutamide that's targeting that KATP channel that I've already mentioned. We also wanted to look at cell count for two reasons. One, so we had a way of normalizing our data, and secondly, so that we could look crudely at any effects effects on cell viability. So Sern set about and he performed this for all 300 genes at the 75 loci. And here I've summarized an enormous amount of work on one slide. What you can see here are our controls. So we obviously wanted to know that we could do this for things where we expected to see a phenotype. So we thought if we knocked down insulin, we should see an effect on insulin secretion. And I'm delighted to say that that was the case. We had here other hits that were coming out from outside of our confidence intervals. And some of them were very notable to those in the field, very reassuring because we were seeing genes that were involved in monogenic varieties of diabetes, which is a really good sanity check to tell us that things were working. So in total, CERN identified 67 hits. 15 of them were for cell counts, 52 for insulin secretory phenotypes. Across those different readouts. So we had 45 genes that were involved in beta cell function at 37 different GWAS loci. And as I mentioned earlier, critically, we were seeing this enrichment for genes involved in monogenic forms of diabetes that we were really pleased to observe. So if I just focus on a couple of highlights from the screen just to illustrate the power, I think, of this approach, firstly, HNF4-alpha. Those of you in the field will know that this was one of the genes that was identified early on as a cause of monogenic diabetes. And at first we were a little concerned because it appeared that if you knocked this gene down, you had an improvement in insulin secretion. And that's not very consistent with diabetes. But we were reassured when we went and looked in the literature and saw an elegant study from Andrew Hattersley and Ewan Pearson that showed that actually babies that are born with HNF4-alpha mutations often present with hypoglycemia and are large because they're secreting too much insulin, which is a growth factor. And the cell model that we're using is actually derived from a fetal pancreatic bud. So the biology here actually makes a great deal of sense. So we're very reassured about that. I want to draw your attention to a novel gene that we were able to pick up through this screen, ARL15. This is a gene that hadn't hadn't previously been known to have a role in the pancreatic beta cell, and it's one that's a member of a family that's involved in vesicle trafficking and biogenesis. So you can imagine what it might be doing within the insulin granule field. And we're now looking at this in more detail, having generated a mouse model with Roger Cox. So that was 300 genes at 75 loci. It's clearly delivered results. We're very encouraged by that, but we also realize that we've got many more loci that we need to look at. So we want to go bigger. So Antje Groetz, who's here at the meeting, has set about her DPhil project, and she's going to do a pooled CRISPR library. We think that our data thus far has really shown that insulin content is a very robust cellular phenotype. So Antje is going to do a FAC sorting-based method. To try and identify genes that influence insulin content genome-wide. And this is underway at the moment, and I'm hopeful that we'll have data to share soon. Okay, so we've thought about effects on gene expression. We've shown detailed cellular and physiological studies at a single locus. We've talked about going genome-wide and being able to assay variation at scale. What about if we don't want to look at just a gene? What about if we want to ask the question about what does the what does this specific allele do? Is there a way that we can do this, and can we do it in an authentic cellular model? So Helena this morning introduced us to human-induced pluripotent stem cells, and she talked about the fact that they are a hope in the field for being able to understand the effect of specific genetic alleles on a cellular phenotype. For us in the diabetes field, one of the things that we're particularly excited about is the fact that they give us a tool to look at what happens during pancreatic development, something at the moment that's very difficult to do outside of the mouse. And as often is the case, the mouse doesn't fully recapitulate the human developmental pipeline and makes it very difficult sometimes to make inferences from the mouse to human. So human iPSCs and being able to differentiate them to human pancreatic islet cell types is a really exciting breakthrough in terms of being able to get at a bit of biology that was currently behind a locked door. So in our group, we're interested in using CRISPR-Cas9 genome editing in human iPSCs as a cellular model to understand the effect of genetic variants on cellular function and beta cell development. So for this to be even possible, we need to have a platform in our lab that allows us to make human beta-like cells in vitro so that we can start to assay function. And this has really been possible due to our involvement in an IMI called Stem Bank with the diabetes component being led by Mark McCarthy in Oxford. Working very closely with Novo Nordisk, Christian Honerer in particular, we've been able to establish a pipeline that allows us to take to take a human iPSC and to differentiate it to something that resembles very closely but not exactly and authentically a human beta cell. And Nicola Beer, a talented postdoc who was in the team and is now heading up her own group at Novo, heading the stem cell engineering group there, set this up for us in Oxford. So as you can see here at the bottom of my slide, we've been able to take our iPS cells which had the pluripotency markers, take them through to definitive endoderm, then through to being able to express the relevant islet transcription factors, and then to see hormone positivity for insulin and for glucagon. We were then able to perform a pretty comprehensive omic characterization at each of the 7 stages of our differentiation pipeline, looking at RNA-seq, ATAC-seq, and DNA methylation. We've then been characterizing our stage 7 beta-like cells in a number of different ways. So Benoit has been doing EM on them to show that they are hormone positive, although I have We see insulin and glucagon present in them. And he's also looked at secreting the cells and been able to show that they are responsive to glucose. So we have something that resembles a beta cell. It's not a perfect model, but it is something that is glucose responsive and has many of the features transcriptomically, morphologically, and characteristically of a beta cell. So we've been trying to use this model now to answer some important questions in the field, and I'm going to just give you a taste of that. I'll just focus on one example to finish up on where we've been using this approach. And this really is in one of the poster childs of type 2 diabetes around the SLC30A8 story. Now, as many of you will know, in the first GWAS that were published back in 2007, there was a common coding variant in SLC30A8 that was shown to be associated with type 2 diabetes risk. And then very excitingly in 2014, colleagues at the Broad, Flannick, David Altshuler, Nicola Beer, and others were able to show that there were rare truncating variants in the gene that were associated with protection for type 2 diabetes. And reaching out to the diabetes community and working closely with Decode, they were able to establish that this met statistical significance. Now, this left us all in a little bit of a worry about how on earth this was working at a cellular level. We had compelling genetic evidence that these loss-of-function variants were protected, but you couldn't really work out what was happening in terms of the biology. There had been a number of studies that had tried to knock down SLC30 in rodent pancreatic beta cells, but these had shown that they reduced insulin secretion, so that wasn't consistent with protection. And then if you looked at a large number of studies that had published knockout mice, global and beta cell specific, there was a very variable phenotype, so you had no consistent consistent biological story to go with the compelling genetic evidence. So it became very important to think about how do we predict whether these loss of function variants in SLC30A8 are protective against diabetes because they are due to a loss of function, or is there a hidden gain of function there? Because if we don't know the direction of effect, you're not going to know whether you need more or less of the sync transporter, and then pharmacologically, you don't know whether you need to make an agonist or an antagonist. So what Nicola set about to do with the help of Fernando was to use genome editing in our human iPSCs to introduce the most common of these truncating variants, R138X, and then to be able to measure the expression of the two alleles so that we could see whether or not both of them were transcribed. So she made the edits in our human iPSC model. She took them through our differentiation protocol. Unfortunately, SLC30A8 is not expressed until stage 7, so we had to go through the whole month to get ourselves to the endpoint. She then extracted the RNA and did digital droplet PCR to be able to measure the two alleles. And in doing so, she was able to show that there was a 30% reduced expression of the type 2 diabetes protective allele. So consistent here with loss of function. So we were able to show that less SLC30A8, which is protective at the genetic level, is due to less protein in the cell. So antagonists are the future in terms of the direction here for pharmacomanipulation of this zinc transporter. Okay, so to summarize, I've taken you through a series of different approaches that the group in Oxford are taking to try and unlock the biology from GWAS. I've given you an introduction to our Human Islet Biobank, told you about the multiple genetic and genomic phenotypes that we've been collating on those samples, and have shown that we can use them as a powerful resource to identify Cissi QTLs, which pinpoint effector transcripts at type 2 diabetes GWAS loci. And I've given you the specific example of ZMIS1. I then moved on to tell you about our work on PAM, where there are two independent coding variants that are associated with type 2 diabetes risk. I showed you our cellular studies where we were able to show effects on insulin secretion and content and exocytosis that we think are mediated through effects of reduced amidation of chromogranin A. I then told you about the efforts that we have taken and our plans for the future to work at scale to be able to interrogate multiple loci in authentic human cellular models. I showed you in published work that we've been able to show 45 genes at 37 GWAS loci that are involved in pancreatic beta cell function. And I finished up by taking you through the pipelines that we've established in Oxford for genome editing and differentiating human iPSC cells. And I've given you the example of SLC30A8 where we've been able to introduce a variant that's protective for type 2 diabetes risk and have been able establish that it causes haploinsufficiency. So if you weren't listening, if you have been checking your emails, if you've been thinking about your beers, this is what I'd like you to walk out the room remembering. That to date, the majority of the GWAS signals that have been identified influence insulin secretion. They have their primary effect through beta cell or islet dysfunction. But if we integrate genetic, genomic, and cellular data, we do have the power to unlock GWAS biology. It's not easy, but it is possible. And I really think that genome editing is going to offer us the opportunity to disentangle the effects of coding and regulatory variants in authentic cell types, which might help us address some of those really challenging questions that we are not able to resolve statistically. Now, the work I've presented today really is a tour de force and has involved a large number of people. Many of you are in the room. I'd like to thank all of you for the opportunity to collaborate and do great science, which I think is absolutely at the heart of being able to deliver mechanisms. Nobody can be an expert in one— in all areas, and we really do need to work together. So thank you to everybody. I hope I've mentioned the key people throughout the talk. This is the group back in Oxford. Huge thanks to them for being back there looking after the endobeta cells, culturing islets so that I can come and do drink beer with you guys. And I'd like to finish up with a slide that makes me feel very warm and fuzzy. These are some of the scientists of the future. These are scientists from Headington Girls' School for British Science Week where we went in and helped them to make beta cells, mainly polyhormonal beta cells. As you can see, lots of multicolor. They're happy to extract your DNA from your strawberries, and here they are making a beautiful human eye Thank you very much for your attention.

Thank you, Anna. That was just wonderful. Questions for Anna? So I did want to ask you about the monogenic forms of diabetes, the HNF4A and that, which early on in the GWAS era when GWAS signals were coming up, people were saying, well, maybe those are just tagging the monogenic forms, but maybe there's something else going on. I wonder if you've used or are going to use some of your editing approaches to try to figure out exactly what it is that's contributing both the common disease signal and the rare disease signal.

Yeah, absolutely. So what I didn't mention for reasons of time is that we've been collecting fibroblasts and making iPSCs from individuals with different varieties of monogenic diabetes, and we're at the moment doing a lot of work on 4-alpha and 1-alpha. And I think understanding how those cells behave when you have a bona fide monogenic variant will provide a good framework for understanding what some of the common variants that Anuber and Andrew have reported on previously, and there are several, and there are particular places where we need to disentangle whether or not they're having independent effects, and it might not be that very nice attractive coding variant, it might be that regulatory one. I'd be excited to start looking into that.

Absolutely. So, there's a question way in the back, so our microphone runners can— Hello. Oh, you already have one.

That's for you.

I'm good. Wonderful, huge amount of information and approach. I got confused at the very last part where when you mutated, made introduced that mutation, and you showed that there was 30% less transcript from that allele, and the conclusion was it was haploinsufficiency.

Is that correct?

Yes. So I'm still not clear how you can say that the remaining transcript that's producing a truncated protein is still not having some neomorphic effect, even though there's overall less transcript.

So that's a good question, and shows where I was rushing at the end and didn't give you the full picture. We were investigating whether or not the transcript had been degraded by nonsense-mediated decay, and the rescue with cyclohexamide is a good indication there that actually the transcript has been degraded, which is why we don't think it makes it to the protein stage. What I can also tell you that I haven't shown today for reasons of time is that we've been able to go on and look at those cells in terms of their cellular phenotype in more detail.. And it's really exciting, in fact, because what we're starting to see emerge is a role between reduced zinc in those cells and KATP channel function. And if you remember right at the start of my talk, I introduced that monogenic variety of diabetes that was all about that KATP channel not responding. And it looks as if the biological story is going to take us back to that KATP channel, and that may be what's happening. So I think I think the picture isn't completely solved yet, but I think we've got a haploinsufficient— it's not there, and I think that loss of the channel is then having an effect through the zinc signaling on gene expression of the KTP channel.

Thank you. Great. And then down here in the front. Collaboration.

Anna, you said that your functional screen turned up 45 genes in 37 loci. Have you looked into those where obviously you're getting multiple genes from the same loci, and do you have independent associations then? Does it look like each one's hitting gene, or are you getting master regulators that affect the regulation of both those genes? Have you tried to pull that apart yet?

Yes, we did, yeah, and that's a good point. So this is where I think having layers of data is phenomenally helpful, and actually one of the cases we were able to marry up our Cissy QTL data with the screen, and then you get clarity because you have a Cissy QTL that indicates it's one of the genes, and the other one is the bystander, if you like. It's also having an effect on the eye, but it's got nothing to do with your signal. So I think this is where, you know, I wouldn't use just the screen. It's a prioritization tool. You take that with the other lines of evidence that you're collecting.

Yes, Ben?

Great talk.

Sorry if I missed it.

I wondered if—

could you say something about the breakdown of variants that perhaps function during differentiation to the mature cell versus things that have function in the fully or somewhat fully differentiated beta cell. Can you get a sense of that by looking at, you know, variants that occur in open chromatin at an early stage versus the late stage?

Yeah, no, that's a really interesting question. So there are two ways that we can get a little bit of insight into it, but not the full picture. One is to think about what happens in monogenic varieties of diabetes where we know that many of the genes in play are transcription factors that are playing that critical role in ensuring that you get your pancreas and your islet cells at the end. The other way I'm quite interested in doing is working with Momoko, Rachel Freethy, and the Egg Consortium and looking at the effect of birth weight, because I think some of the signals that we see there where insulin is a growth factor, we might be able to, again, layer up with the data that we have to give us further insight on whether things are— how early things are taking place, and that might also help us focus on the genes that we should be studying in terms of their impact on differentiation. And there's a really nice technique coming out where you can start to do the genome editing conditionally, so you can start to study things that would be perhaps detrimental to your ability to get through that pathway, but you can introduce them later, which I think, again, will be really helpful in terms of getting at that biology.

Last question here. So that was a great talk. So one of the issues and pushbacks with GWAS functionalization that comes up a lot is sort of the small effect sizes that we observe, and I was wondering if you can comment on what your observations have been and how that's impacted your ability to elucidate mechanism.

Yeah, no, no, no, that's a good point. So I mean, the criticism I get, and I'm sure we all get a lot, is that the effect sizes small, so why would you bother? And I always like to go back again to my KATP channel because there's actually a common variant in that gene that's associated with type 2 diabetes risk, odds ratio of 1.1, 1.2. Now it doesn't matter that that's a small size, that is a bona fide drug target, so you get to the biology. So that's the first argument, let's keep that in touch. The second thing I think is really exciting in some ways because the closer you get to where the defect is in your functional assay, the greater your ability to pick it up. So we did some work early on, on glucokinase regulatory protein, that variant P466L that seems to be associated with just about everything. And we were able to develop a kinetic assay and to get very close to where that defect is. So despite the fact the effect size is very small, if you've got the right assay, you can pick it up. If we'd looked in a noisier system, perhaps if we'd been looking at effects on expression or whatever, we wouldn't have seen it. But if you actually get at where the business is, I think you've— even with a small effect, you've got a good chance of picking it up.

Great. All right, Anna. Let's thank Anna and all of our speakers.