 So, I thought at the point we are now at, it would be appropriate to dig deeper into a specific example and dissect what exactly these publications are able to say about the genetic architecture of a common disease. And the example I'm choosing is diabetes based upon the papers that came out last week and a few others that came out shortly before that. This is actually a disease my own lab works on at NIH. And it's a very tough one, to be sure. As was pointed out by the comment a little bit ago, diabetes has been increasing in its incidence in the United States and elsewhere, and that's not because the gene pool has changed, because the gene pool is unlikely to be any different over a 10-year time period. So instead, this increase in diabetes frequency is clearly on the basis of diet, lack of exercise, and the increased presence of obesity. And you can see it's pretty dramatic in just a 10-year period. And yet not everybody who is sedentary, overweight, and has other diabetes risk factors gets the disease. So there must be some interaction going on here between genetics and environment. And again, while we're unlikely to change the genetic pool, at least not anytime soon, if we could identify who the people were at highest risk, especially early on before signs of a disease appear, that might be a good thing. And if we could identify what the pathways were that are involved in diabetes, which we really don't understand very well, that might give us some new ideas about treatment, because nobody would argue that the treatment of diabetes right now is ideal. It clearly is not. And this disease is responsible for untold morbidity and mortality and expenditure of health care dollars. So what do we know about diabetes in terms of what's really wrong? Well, we don't know all that much. We certainly know that heredity plays a role because the disease tends to run in families. And we certainly know that environment plays a role for the reasons I just went through and the fact that the incidence is increasing so much. And we know that there's, at the physiological level, a combination of not making enough insulin, because the beta cells in your pancreas, which is a place where insulin is made, aren't quite up to the task. But it's not just that. There's also an insulin resistance, and that's particularly as a consequence of obesity, where the insulin you are making isn't quite doing what it should as far as getting glucose out of your blood and into your cells. And the combination, you may be able to compensate for a while for this insulin resistance by revving up your beta cell production, but then the beta cells basically run out of gas, your glucose goes up, and you end up with type 2 diabetes. But that's pretty lousy as far as having a really good idea at the molecular level of what is going on with this disease, and we'd like to be able to do better, especially in order to come up with new treatments. From a geneticist's perspective, this disease has always seemed pretty impenetrable. Family history is a substantial risk factor, but the relative risk to a sibling is only three, so it's not like a big, huge increased risk. So that means that all of the genetic factors together, if we understood them all, would only account for at most a three-fold increase in risk, and probably not even that, because this is also probably shared environment. Your sibling is likely to have had the same experiences you have had in childhood, and therefore some of this relative risk is probably not genes at all. So we knew going into this, this was going to be one of those where those ticking time bombs were mostly not going to go off, and what we were looking for is something much more subtle. We know the environment's a big contributor. People did family linkage studies, which we haven't talked about today, because they were sort of the effort of the last decade that were largely unfulfilling when it came to common diseases like diabetes. The strategy often was that you identified families where there were two affected siblings. You know, affected siblings generally share half of their DNA. That's how it works. If you scan down through the genome and you had hundreds of these affected siblings, and you found a place where they seemed to share a little more than half, that told you, oh, maybe there's a diabetes gene there if they both had that disease. Didn't work very well. A lot of time and effort got spent very underpowered when it came to finding these kinds of answers. So prior to this year, the genes that we were sure were right for type 2 diabetes. One is PPAR gamma. That's a candidate gene. It happens to be the target for a class of diabetes treatments, diabetes drugs, which is how it was identified as a possible candidate. This is interesting, by the way, because if you had not known that and you didn't already have those drugs, then finding this might have led you to try out a drug approach, which we now actually know works. So this is, in a way, a backwards validation of the idea that finding genes involved in common disease gives you ideas about treatment. Here's another backwards validation, KCNJ11. This is the target for the drugs in the sulfonylurea class, things which are commonly given to people with diabetes. Again, has a variation in it, turns out to be a validated one for the disease. Yeah, let me just linger on that a minute, because some people have in their heads that if you found a variation in a gene that's associated with diabetes, and you're going to use that to develop a drug, the drug would only work for the people who had the variations. That's not right. I mean, the variants basically tell you, here is a particular gene and a protein that's involved in the disease. And yes, there's a subtle change in it that may increase or decrease the risk. But even the people with a low risk spelling may very well respond to a drug that's targeted to that particular pathway. These two are good examples of that. So the people who have the high or the low risk versions of these two genes respond to the drugs that are targeted to those particular molecules. And then the third one, which came out about a year ago from decode, was a linkage study where they had studied families and came up with something on chromosome 10. And then they did the fine mapping thing, and they found this totally unexpected gene called TCF7L2, which has a horrible name and is a transcription factor that people thought was involved in T cells. But it turns out it's involved in a lot of things. And lots of groups have validated that since that time. But that's all we knew prior to 2007. So those three, you will see, have now exploded into a set of 10 genes that we know are involved in diabetes in just the last few months. So what was done about that? Well, several efforts were made, I'm going to tell you, particularly about a trio of studies. These are the three that were published last week in science. And that involved three different groups that decided to work together because of the perception this was the geneticist nightmare, and you were more likely to succeed if you shared the experience with your friends. So one group called Fusion, which stands for Finland, the United States investigation of NIDDM, which is the old name for type 2 diabetes. Studied individuals in Finland with lots of laboratory and statistical efforts. Another group in three different places studied primarily Europeans. And then the Welcome Trust group studying people in England. So most of the participants in all three of these studies have what you would consider Northern European background, which is another reason that we thought it might work well to pool the results. So each of those groups, in what you'd call stage one, studied between a thousand and 1,500 cases and between a thousand and 3,000 controls. And used one of these two panels, you heard about these earlier from Larry and Terry, basically, which have a pretty good coverage of the whole genome. So each DNA sample got run on this set of steps. And then after that stage one work had been done, all three groups compared their data to say what looks promising here and then made a decision about what to follow up in a validation set, which is an even larger number of cases and controls. Why didn't we just do everything at once? It would have been too expensive. The idea is stage one, you're studying the whole genome, then you decide where the promising signals are, even though you're not convinced yet. And then in the stage two, you test those and see what holds up. So the part of this that my own group is involved in, Fusion, just to give you some sense of the numbers here. In stage one, 1,161 cases, 1,174 controls. Stage two, about the same numbers. In our instance, some of the cases did have affected family members. Now, why is that relevant? You would guess that if you start with people who have a positive family history, they may have a slightly larger genetic load than the average sporadic case. So if you're looking for genetic factors, you might be a little better off to start with people with a family history than without. In our instance, I don't think it made a lot of difference. Now, you heard about power, and again, you saw some calculations from Terry about this odds ratio, OR, and you saw that in the example from the embargoed study, which I'll come back to in a minute, the odds ratio was 1.23, and yet we still thought that was pretty valuable to discover. You can see with these numbers, even though they look big, and even though if you combine stage one and stage two, your power to detect something is not so good once you drop below an odds ratio of about 1.3. You're going to miss things that are weaker than that. And for a disease like diabetes, we didn't know what to expect. We didn't expect we would find very many that had large odds ratios, because again, remember, the total risk to a sibling is only three, so there can't be that much from a specific single gene that could cause a very large effect. And this is one of the reasons why, in our instance, and I think this played out well, we talked to our colleagues in these two other studies early on and said, let's work together to beef up the numbers, because even with, in this case, 4,824 DNA samples, we weren't confident we had enough power to find all the things we wanted to find. So in this instance, we used this Illumina 317K panel. By the way, you have to be very careful about the quality of the data, and we ended up dropping about 2,000 of those because they didn't meet various standards for accuracy. I won't go into the details of that, but that needs to be in there at any time you're doing one of these studies. And you saw this kind of a diagram before from Terry, and this is what we had at the end of stage one. And you can see all of these points represent individual SNPs. This scale over here again is the negative log base 10 of the p-value. So this would be, in fact, I think I've marked it here. If you were satisfied with a p-value of 0.05, which you obviously would not be, then you would say, wow, everything above that is like something exciting. Well, obviously, that can't be right. And it can't be right because you're doing so many comparisons that you would expect to have lots like this. Remember, we said then you've really got to divide by the number of tests you did, and in that instance, dividing 0.05 by 300,000, you would say, really, to be confident that you found something that's going to hold up, you need this p-value down in the range of 10 to the minus 7, which would be 7 on this scale, and there are no results up there in that range. So despite a fairly decent number of cases and controls for stage one, we didn't prove anything. We had some interesting stuff, but nothing that you could write a paper about or convince yourself was anything more than noise. Now interestingly, in this disease, because we knew of three genes that had previously been documented as involved in diabetes, we could check and see whether they turned up, and if so, where did they turn up? And so sure enough, that TCF7L2, I told you about, those two SNPs right there are smack in the middle of TCF7L2, and they gave us a pretty good signal, although not a convincing one. PPAR gamma, that one down there, and this KCNJ11 down here, in the part where you couldn't be confident of anything, but not down here where you found nothing. So this was somewhat encouraging that, in fact, the study does have good balance between cases and controls. But then we got the groups together, and here you can see, again, the study I just told you about, which is collecting its individuals from Finland, the Diabetes Genetics Initiative of Broad Lund Institute and Novartis, also had some of their patients from Finland, but a different set than ours, and from Sweden and from Poland and from the U.S., and then the Welcome Trust with their cases and controls in England. And you can see now the numbers are getting pretty big, both the stage one and the stage two. And so when you put them all together, this is 32,554 DNA samples available for these three groups to analyze. Now you're getting up there in terms of having real power. And when we put all those results together, compared our stage ones, then chose wisely how to do the follow-up validation ones. Now, again, with the p-value over here, so significance is going to be right about in here, there are clearly some that are getting very strong results, the best one in the whole genome is that one previously discovered last year, TCF-702 p-value 10 to the minus 49, you tend to believe that, but many of these others also very strong. So what are they? So here's the table that you can see in the paper, which basically tells you for each of these now 10 genes for which the overall p-value you see here is very strong, except for this one, which was the previously documented one, so we're going to give it a pass and say we already knew that was right. All of the others are falling into this zone of confidence. The ones in blue are those three we knew about before, and the others at the point that we had this data were new, although as you'll hear, three of them turned up in independent studies by other groups at about the same time. Now interestingly, when you look across the table, you can see examples where one of these three studies got a very strong result and the others didn't. Look at this gene for obesity, FTO. Almost all of the signal is coming from the welcome trust. In the fusion study, we got pretty weak p-value, just 0.01. The broad study got essentially nothing. You know why? Because they actually controlled their cases and controls to have the same BMI. So they basically wiped out the possibility of discovering anything about obesity because they wanted to study diabetes. So they intentionally wiped it out, whereas our study and the welcome trust study didn't do that. We didn't try to match for BMI between cases and controls. There's some other interesting ones where there's quite a discrepancy. One that turned up very strongly in our fusion study on chromosome 11, which is actually a signal where there are no genes. It's in a gene desert, making it rather interesting. Very little evidence in the other two studies. And yet this, we believe, is real. It turns out it's quite close to a signal from another study. So those three papers then came out last week, reporting simultaneously the analysis of these three studies and what we've identified. Just a month before, this paper in Nature Genetics, carried out by Sladek and Philippe Fruguel as the senior author, had conducted a genome-wide association study on type 2 diabetes in French individuals and came up with some similar results, which I'll tell you about. And then this paper in Science Express April 12th reported on this FTO gene associated with adult obesity where they basically took that result that I showed you from the welcome trust study and tried it out on a whole lot of other cases and controls that weren't collected for diabetes. They were collected for obesity and showed that it held up very strongly. So that basically says that some of these got validated independently. The French study found H. Hex and SLC-30A8 completely independently and that's actually makes you feel really good. When you see another group that started with totally different patients and you didn't even know they were doing it and ultimately you come up with the same signals and then the FTO I just mentioned. And last week, at the same time, that the three papers in Science were published, this paper was published in Nature Genetics from Kari Stephenson and the group at Decode. And what do you know? The gene they found, CD-Cal1, well, that's one of the ones that we had identified also. And again, completely independently having no knowledge that the folks in Iceland had come up with the same signal. That was quite a stunner. And when you look carefully at the details, yes, it's not just the same general location of the same gene, it's actually the same SNP in the same haplotype. They're seeing the same signal, which is, again, a very nice way of getting immediate validation from an independent source. So that's our set now of these 10 that predict risk of diabetes, but what are they and what can we do with them? So you've seen these pictures and here is from chromosome three, one of them. This is a gene called IGF-2-BP2 and this is where the scientists really begin to roll up their sleeves and try to say, what is this? This is not a gene that would have been on anybody's list of candidates. Here you can see, again, the kind of diagram you saw before, it's that red dot, which is the SNP that gave us the best signal. And again, if you basically control for that, it makes all the rest go away, which is to say that that really is detecting everything that's present in this site. It's not that you have multiple different SNPs that are giving you independent information. It's all the same signal. If you're not used to looking at these diagrams, these are the kinds of images you'll see in the browser that most postdocs use to investigate what's going on in the genome. So this is a gene that's transcribed, in this case, from right to left. And each one of those vertical tick marks is an exon. So that's exon one. And actually, I believe there's two exons there. So exon one and two are close together. Then there's this big, huge intron. And then there's a bunch of other exons down here. And the signal seems to be kind of like in the intron. And I can tell you there are no variations in the exons that code for protein at all. So whatever this is telling us, it's not something that actually alters the coding potential of this gene. It must be something that alters its regulation in terms of whether the gene is on or off, or how much it's on or off, or when it's on or off in the course of its lifetime. Down here you can see the triangles that you've seen before, two different ways of looking at them. You're probably gonna have to deal with R squared and D prime. There are two different ways of looking at linkage disequilibrium. R squared is like more stringent. So you can see the triangles aren't so big. D prime is all over the place because it's less stringent. But you can see there's lots of correlation here. And you can see all these SNPs that seem to be giving you about the same signal. Well, no wonder they're in this block of linkage disequilibrium marked by that triangle. They are all traveling together. So if one of them gives you a signal, the others are going to also. So again, we have a big research project now. By the way, I think in this instance, you can say it probably is the IGF2-BP2 gene that's responsible for this, even if we don't know quite how. The other genes nearby are quite a distance away. There's this funny little thing here, which I suppose could potentially be involved, although we don't really know what that is and it might not be a real gene. Most likely, this signal is saying this gene is involved in diabetes. But what do we know about it? We know almost nothing about it. It has a name, but we probably ought to be careful about the name saying more than we know. There's a related gene, which is similar to it in its sequence, which we know a little bit about. We don't actually know anything about this one, except it's similar to that one. It's sort of two degrees of separation between the gene and what we actually know about what its function might be. And this is even complicated because it makes a protein that binds to the RNA of another gene called insulin-like growth factor 2, and that is involved in development growth and stimulation of insulin action. But you can see how far away we are from really having a handle on what the function might be. Nonetheless, I would be quite surprised if there are not several pharmaceutical companies that are starting drug discovery efforts right now to try to figure out how IGF-2-BP2 could be used as a target for developing new therapies for diabetes because this tells you this is a molecule that's involved in pathogenesis, and that could be very exciting in terms of a totally new way to approach the risk of disease. This is that zinc transporter that Joe was so excited about. So this is one where the variation, the red dot here, is actually in an axon of this SLC-388, which is really a horrible name, but this is this protein that codes for a transporter. In this one instance, actually, the SNP, which has that number, changes an amino acid, an arginine, to a tryptophan near the end of this gene. Remember, it's transcribed this way, so it's almost at the end of the coding region. And that is an interesting potential target as well because it codes for this protein here called ZMT8, which is responsible to pump zinc into these granules in the cells in the pancreas that actually make insulin. And if you don't have zinc in those granules, you can't secrete your insulin. Insulin has to be complexed with zinc before it can be actually put out into the circulation. So in my dreams, this tells you that maybe large doses of zinc would be a good protection against diabetes. Nobody can tell me that's wrong and finding this discovery makes it slightly more likely to be right. That in fact, especially if you have the risk version of this gene, you're not pumping your zinc quite effectively. And maybe if you had a little bit more of it around, this would prevent this from leading into a diabetes susceptibility. Here's the one that really surprised us with this week's finding. So one of the others that I mentioned to you in that table a minute ago are some SNPs that lie upstream. And let me just show you the diagram first and I'll come back to this. So here is another one of these. This is coming from the Wellcome Trust paper. You can see again the triangles that are showing you the linkage disc equilibrium. The SNPs that are showing you association with diabetes are here. They're kind of in a funny position. They're not like in a nice little peak. They're spread out over a region. But again, if you look down here, you can see that those are all rather tightly associated with each other. And these are not in an area where there is a gene. The nearest genes are called CDKN2A and CDKN2B. This is by the way, not in your folder because I just made these up early this morning after seeing the papers that are in. It's not the data. Yes. So that was all very interesting because again, it says maybe these are in fact regulatory. And maybe they have an effect on these two genes. And what are those two genes? Well, these are really hugely studied genes in the area of, guess what, cancer. CDKN2A and 2B play a role in inhibiting these protein kinases that play a role in the cell cycle. If you have a mutation in one of these, you get hereditary melanoma if you have a knockout loss of function of all things. In the mouse model, though, you can in fact knock out one of these and give rise to diabetes. So there was some connection here. But it was a big surprise. The bigger surprise for me was seeing these two papers, which are the ones that are coming out on Thursday. So here we have two groups. They're not looking at diabetes. They're looking at coronary artery disease, myocardial infarction. And they're doing a genome-wide association study with no preconceived idea of what they're going to find following the same strategy. And what do they find? We're in the same place in the genome. Here's CDKN2B and 2A. They're signal exactly in the same place as where we found an association for type 2 diabetes. And yet these authors, and I don't know any more about this than what I read in their paper, say that they looked carefully to make sure that they were not confounding their results by having their coronary artery disease patients all be diabetics. And they say that was not the case. So they're saying this is an independent risk factor for coronary artery disease. Now I know when they wrote this paper, they didn't know about the diabetes result. When we wrote our paper, we didn't know about the coronary artery disease result. So there's going to be a huge storm of interest here over the next few days as people try to figure out what happened here. How can one region of the genome, and it's not even containing a gene, contain risk factors for both diabetes and heart attack when they don't seem to be heart attack on the basis of diabetes? I think this is a stunner. I mean, this is like the seat of the soul of the genome. It seems like this one place that carries all of that weight for two very common and very dangerous diseases. I never would have guessed that we would end up coalescing zeroing in on the same sort of 50,000 base pairs out of 3 billion that seems to contain that signal. So I don't know. That'll be a fun one to figure out how to write about. But I think that's a big part of the story, is not only did they find a variation that's associated with heart disease, but in the very same place where a week before, three different groups found the same association with a different disease, diabetes. How does that work? I don't know. Finally, because I got the signal some time ago, I wanted to go over this diagram, which is something that appeared in last week's science paper from my group, about trying to make a prediction based on these 10 associations we now have. Can you actually say what the risk is for an individual of getting diabetes if you tested them? So this is reconstructed from our own cases and controls in a way that is perhaps a little obscure, but I think it's accurate mathematically. And without going into the details of how this construction was done, basically what we've done here is to divide people up into 20 bins with equal number of individuals in each bin. And then we have, for each individual, look to see for those 10 places where we know there are variations associated with diabetes. What genotype does that individual have at each of those 10 places? What does that say based upon the odds ratios for each of these particular risks? Their risk would be adding them all up based upon their genotypes at those 10 loci and then putting them into this scale. And then actually going back and looking at our real cases and finding out, is that right? So this is not purely prediction. This is actually observation in those roughly 5,000 cases of diabetes from Finland. And what it says is that if you're in that group that has the lowest number of genetic risk factors from those 10 different regions of the genome, your risk of diabetes is down here around 5%. By the way, in Finland, your risk as an adult of getting diabetes is close to 10%. Whereas if you're in the highest risk category, you've got the biggest collection of spellings of genes that put you at risk, you're up at about 20%. So these 10 genes then give you a range over a four-fold kind of multiplier of your risk of diabetes. And that seems interesting. And potentially, those at the extremes of this distribution might find it valuable to have that information or not. I'm sure Alan will talk more at the end of this about to what extent is providing people with this kind of predictive information something that people will value? Fortunately, we're about to see this genetic non-discrimination legislation pass so that people won't have to worry about it on that grounds. But will people use the information in their own health behaviors if they know they're at high risk? Notice that it's not great, however. A lot of people are in here in the middle, and again, these error bars tell you that you don't get a lot of information if you're in the middle from testing these 10 genes at all. You're still pretty much in the middle. It's at the extremes where you get the most bang for the information. So I hope that's helpful as a way of digging a little deeper into what these studies can do and how they're conducted and what the information might mean. And I'd be glad now to hear questions, yes. Yeah, the overlap of the 10 genes between heart attacks and diabetes, my immediate free association is the metabolic syndrome. So what do you think? Well, that's a great question. So of course, the metabolic syndrome is supposed to include lipid abnormalities, hypertension, diabetes, and obesity. And obviously, heart attack is one of the big risks of metabolic syndrome. And it may well be. There's something like that going on. I know that people have done, interestingly, genome-wide association studies for metabolic syndrome when that was what they started with, and they've come up empty-handed so far. And some people have concluded that maybe the metabolic syndrome isn't a real syndrome at all. It's just a collection of relatively common disorders that are going to happen together in enough people just because they're all common to make you think there's a syndrome when in fact there isn't. That's not been decided. But I think it's a good idea that we start to go back and look and see if you went back to those metabolic syndrome studies and studied this particular version of this chromosome nine finding, would you in fact find that there's a signal there that just got missed? What about inflammatory factors? Do these 10 genes have a function on that? Not much that you can say with any confidence, but a lot of these we really don't know what they do. I don't know what IGF-2-BP-2 does. I don't really know what these do, although they control the cell cycle. That could obviously be important inflammation. CD-Cal-1, I didn't really talk about it. We have no idea what that does. I mean, it is kind of like four degrees of separation between the gene name and any consequence that's really experimentally validated of what it might do. It could certainly be involved in inflammation. This H-hex, that one is probably overstated because that's one of the cases where the signal for association actually stretches across three genes. H-hex is just one of them. It might not be the one. It might be the other two lying next to it. And of course, this chromosome 11 signal where there are no genes at all nearby, presumably this is a regulatory signal for some other distant gene. It could be operating on almost anything as long as it's on that part of the chromosome. So I think, I hope you're getting the sense that this is both good news and bad news. It's good news because it's opening a brand new window into what causes diabetes that we never could have guessed at by some other pathway. It's bad news in that we don't know a lot about any of these, and it's gonna take a while to figure that out. Are we posting the video? Okay, I get that. Yes, Maggie. It's Maggie, can I ask the dumbest question? Not at all. When people do these studies, how can we explain to the public the difference between an acquired mutation that this might be, an inherited mutation? Is there any possibility these people acquired all the same mutations? Very unlikely. I mean, again, when we're looking at SNPs, if they're on the chip at all, it's because they have already been identified as being common variants in the population, which means that they are inherited or they wouldn't have ended up being discovered by HapMap and other approaches. And again, when you look at blood DNA from somebody who's otherwise healthy, that IE they don't have leukemia, they don't have a lot of malignant white cells floating around, the DNA you see is almost to a complete perfect approximation. An indication of the hereditary DNA that was present when their sperm and egg came together. So I would think it's fair to say for all of the genome-wide association studies that you see and read about and report on for common disease, they are looking specifically at hereditary factors. The exception, of course, is cancer, the cancer genome atlas, which we aren't talking about today, is focused specifically on those acquired mutations that cause a cell to grow out of control that happened during your lifetime and are obviously a critical part, as Emily talked about this morning, probably much more critical for the progression to an actual malignancy. That's a different set of technologies and obviously starting with different material, namely tumors instead of blood DNA. Can you talk about in terms of symptoms, phenotype or whatever, where the overlap is in diabetes and myocardial infarction or heart disease? Like what? Yeah, how could this all fit together? Well, I'm a bit puzzled by that, too. Certainly diabetes is a major risk factor for heart attack. If you have either type one or type two diabetes over a period of time, and especially if your glucose control is not absolutely perfect, your risk of atherosclerosis in general and coronary artery disease in particular goes up very substantially and a common morbidity for somebody with diabetes is heart attack or stroke as a result of that. In fact, most of the complications that are maybe most feared in diabetes are because of the consequences to large arteries, whether it's that you've got peripheral vascular insufficiency and end up with amputations or that you've got a heart attack or a stroke. So you could sort of see then how you might have ended up with the same signal if you were studying heart attack and you had not actually been careful about this because a lot of your heart attack patients might turn out to have diabetes and you were really mapping a diabetes gene instead of a heart attack gene. They say in the paper, and I haven't had time to look carefully to be sure exactly how they did this, that they looked at that and they were convinced that the heart attack predisposition they were seeing was not on the basis of a lot of their cases having diabetes, but it was happening anyway in people who had heart attack with no evidence of diabetes at all. So it could be though that at least some fraction of those who developed heart attack had subclinical diabetes. It had never been diagnosed, but they already had some insulin and glucose abnormalities that were sufficient to trigger the whole process of doing damage to large vessels. And I suppose that would be the most likely and most uninteresting explanation for the fact that they landed in the same place doing this genome-wide association study. And I'm sure there will be a furious post hoc investigation by everybody to see if that can account for what's now being reported. Do you think it could account for all of it? Again, looking at the paper, I don't think so. I don't think so. From what I see there, it looks like they already have looked pretty carefully and they can't make sense of it on that basis. Now, I tried last night and in the early this morning to look at those papers and look at the diabetes papers and ask the question, okay, it's the same place on chromosome nine. Is it actually the same haplotype? And I'm not sure it is. Now, that would be really interesting. Don't report this, because I could be really wrong and it's gonna take a lot more time than I had early this morning. Be really interesting if in this region of this regulatory domain, there is one stretch of SNPs, one haplotype that predisposes you to diabetes and a different one that predisposes you to heart attack. That would be wild. But it's still possible until we really take this apart and it's gonna take a few weeks, I think.