 So it's 3.32. I know people might be just taking a bit of a break, but we'll start on the last section. So we're in the home stretch here, and this one really doesn't have a lab component, but it's one that's trying to bring together all of the pieces that we've tried to introduce you to. And I guess it has a couple of opinions and thoughts where metabolomics is useful and how it can play a role. So it's integrating applications and the future of metabolomics. So I'm going to introduce you to a couple of things. I mean we're partly talking about how to go from small molecules to systems biology. We're going to talk about what we've already been able to do is getting lists of metabolites and how to go to pathways, and also to go to something a little beyond that, which is cell models. We're going to learn about some applications of metabolomics to clinical medicine, the use of biomarkers, biomarker discovery, receiver operator characteristic curves, software tool Jeff developed called Rocket, metabolomics in the pharma industry, and then also some new trends in metabolomics. So at the very beginning we talked about how to go from spectra to lists, and we've spent yesterday doing this, and we spent part of today doing this. We spent a little bit of time about going from these lists to get information about pathways. We looked at METPA. We've also looked at some pathway databases. You guys used KEG and METPA or the pathway analysis. We've also learned about the small molecule pathway database. We talked about biopsych, reactome, just another plug again for the small molecule pathway database. It is certainly human or mammalian centric, but it is quite diverse, and hopefully those of you guys took some time to explore it. I found that it does offer some utility, and it allows you to take those lists of metabolites and their concentrations and other data, and hopefully allows you both to interpret and visualize that. And so that's an important step going from, as I say, just these raw lists to something else that's a little more meaningful. From the pathways and metabolite lists, there's another group of people in the metabolomics world who are trying to go to models. And one of the more interesting ones is the work that's being done at UCSD with Bernie Paulson. He's a chemical engineer, and he's been doing what are called metabolic reconstructions. About five years ago, they had Recon 1, Reconstruction 1, and then last year, they did Reconstruction 2. And that's the most comprehensive system biology model or SVML model of human metabolism. They've got about 2,600 compounds, 7,500 reactions, they've got the genes, the proteins, and they've marked it all up into systems biology markup language. They have data, it's relevant, and it's all available on this website, humanmetabolism.org. It was published last year, I think, in nature medicine or something like that. A very impressive collection of work, and this is how people can do fairly detailed metabolic modeling. The other thing you can do with these kinds of models is to do something called Flux Balance Analysis, or FBA. And again, this is something that Bernie Paulson has developed, it's also very popular in Europe in general, less frequently done in North America. But they've done it for single cells E. coli, they've done it for human cells, a number of different systems. And it's allowed people to understand the consequences, so you can actually do in-silico mutations to your metabolic model, see what happens. And in many cases, the prediction is pretty accurate. Now, it's more useful for single cells than for humans, and this is because humans are multicellular, and we have organs, and we have methods for compensating losses. But a single cell like E. coli, it turns out these Flux Balance Analysis are quite predictive, quite useful. They've also been used to help identify missing elements in pathways, missing genes, or missing reactions. There's another tool that was developed in our group a number of years ago, which was the idea of modeling cellular kinetics, both with metabolism, but also aspects of gene regulation. The system's called SIM cell, or simulated cell. You can do metabolic modeling. And essentially what it does is it creates little movies that allow you to model Flux changes in metabolites. Normally to do this, you'd have to solve a whole bunch of partial differential equations, and most of us aren't too quick at solving them. But this uses cellular automata, and if you've ever played around with some of the games like SIMS, or SIMCITY, or other things like that, those use cellular automata to generate or solve the equivalent to stochastic partial differential equations. And so you get very realistic data. So this uses the same idea, and you can model cell signaling, protein interactions, drug interactions, and it has a pretty simple user interface. It's not a web tool, it's a downloadable thing. And we used it to model, in this case, drug, pro-panelol, how it interacts. So on one side is the SMIPDB image. On the right side is the SIMCELL diagram showing these compartments, where we've got different metabolites and gene regulators. The diagram, the SIMCELL diagram is not intended to be pretty. What it then does is allows it to then perform the calculations, and actually you can see in real time as things are drifting in and out, but you can also track the stuff as concentrations and fluxes move up and down over time. And so you get graphs, and in fact, the graphs look like the one on the right, where you can see certain things going, oscillating in some cases, up and down. Okay, so that's an example where we're going from lists to models, and this is part of the thrust behind systems biology. But there are also other things, not so much tied to systems biology or basic science, but what you can do sort of practically with metabolomics. And this is a slide you've seen before. A lot of applications in the environment, a lot of applications in food and beverage applications, a lot in clinical medicine. And so I'm going to talk about some of the applications in clinical medicine, and probably about two thirds of you are doing something involving humans. And the rest of us are humans, so we're obviously concerned about our own health. So part of the example you guys did, the tutorial last night, looking at inborn errors of metabolism. So metabolomics is particularly powerful for that. And you guys were using, I think, my Compound ID database, you could also use the HMDB with the hopefully newly upgraded disease browse tools to allow you to type in lists of compounds or others and then actually associate it with disease. You could type in a whole bunch of mass spec peaks, could also associate that with diseases. And in fact, one of the very earliest applications of metabolomics is by a fellow named Ron Wavers, who's in Holland, who's discovered quite a number of novel metabolic disorders just by looking at NMR and mass spec. And interestingly, many of these things, when you find what the disorder is, actually are treatable. So these are things that, at least to the parents of these children, seem to cause profound disability and life threatening conditions, chronic trips to the hospital. And then when they find out what the disorder is, sometimes it's just simply a nutritional supplement, and they're essentially cured. So you can detect and characterize in rodent errors of metabolism by NMR, by LCMS. We can do identification of diseases once you characterize these things. And again, this is possible through databases like the Human Metabolome Database. There's been some interesting applications where there is metabolomic studies done. This is looking at, this isn't, I should change this, it's a type 1 diabetes, actually. So type 1 diabetes is juvenile diabetes. And it's still a mystery of why it develops. So these are ones where kids become diabetic at age 2, 3, 4. It's not as if they're overdosing on candy, it just happens that something occurs where their pancreas shuts down. They have a large cohort study that's been going on in Finland for a number of years. And they were actually monitoring children who were in families at risk for type 1 diabetes. Finland has the world's highest incidence of type 1 diabetes. And what was interesting is that they found that this particular kid exhibited at age 1 or 2 incredibly high levels of glutamate and gamma-aminobutyric acid. Anyways, why or how, they don't know, but the kid seemed healthy. And then, as these metabolite levels increased, they started saying an antibody appearing to GAD. So this is glutamate, whatever, what's the other part of it, something, something, di-drogenase. But it's an enzyme that is actually produced largely by the pancreas. So the body was producing this enzyme to deal with the high levels of these metabolites. So it's pumping up into this enzyme, this protein. It's increasing, too. So that goes up at age 3 or 4. What happens? Well, there's too much of this enzyme around. So the body develops an antibody to it. So, yes, brings the levels of the enzyme down, but because the enzyme itself is produced by the pancreas, the antibodies attack the pancreas. So the pancreas dies. And the kid becomes diabetic. So this is tracked here, and it's sort of confusing with the pictures. But it's a case where, at least in this case, situation, high levels of common metabolites, for whatever reason, led to an autoimmune disorder that eventually destroyed the pancreas and led to type 1 diabetes. So you can get this sort of information, again, through human metabolism data, because a lot of this stuff is what we track and try to include. It's not always obvious, but if you read about metabolites and look at some of their concentrations and read the references, it does allow you to get that. So that's one example. There's other interesting ones where people were using metabolomics to do bacterial identification. And, in particular, this is looking at pathogens more specifically in urinary tract infections. So this is where you could take a urine sample and essentially add a few substrates, small amounts, lactate or glycerol or nicotinic acid or methionine, and see what happened in the urine sample. And because their different bacteria metabolized these different compounds in different ways, you could actually identify which pathogen was causing the urinary tract infection. And given the issue where people are using too many broad-acting antibiotics, it would be really nice if we could be much more specific. And so this is an example where metabolomics allows you to do this. Now they were using NMR, and NMR is intrinsically insensitive, so potentially if they were using mass spec, which is up to sometimes more sensitive, instead of waiting I think six hours that they were doing, they could probably do it in about six minutes. So this could potentially be a very rapid test for identifying causative organisms with urinary tract infections. So these compounds that they were detecting in urine, the glutamate that they were detecting in this kid with type 1 diabetes, are examples of biomarkers. So biomarkers are hot. They've been hot for a long time. Over the last 30 or 40 years, there's been about 150,000 papers published on biomarkers. Most metabolomics people, if they're doing any kind of biological work, even on animals, even on plants, but also in humans, will do biomarker studies. There are different types of biomarkers. Some biomarkers are diagnostic, so that's telling you whether you have a disease. There are also high blood glucose, 7, 8 millimolar. That tells you that you have diabetes, so that's a diagnostic. They're prognostic biomarkers. So in this case, you may look at a person and you know they're sick, but how sick are they? So there are breast cancer biomarkers that allow you to tell whether this is one that's going to progress, a triple negative breast cancer, which is pretty bad, versus sort of regular breast cancer, which generally is treatable. The most interesting biomarkers are the predictive ones. If we could take a blood sample or urine sample from each of you and tell you what you're going to develop when you turn 70 or 80, it might be useful in the sense that, okay, if you're going to develop diabetes when you're 70, time to change your diet, or if you're going to develop heart disease, well, start jogging now. So these are things where, and it's a whole point of a lot of predictive medicine, genetic tests are the ones that we originally hoped would give us all kinds of insights to that, but these are the ones that are obviously of great interest. There are also ones like markers of exposure and markers of response and toxicity. So exposure is one that's very interesting. Carolina's been doing some very interesting work where looking at hair and metabolomics, and this measures what you've been exposed to over many months. And normally metabolomics only measures what you've been exposed to over just the previous 10 or 20 hours. The FDA, many drug companies are very interested in markers of response for drugs. They'd like to find biomarkers to tell you whether, in fact, the drug is making you better or making you worse. They also like to use and identify biomarkers that would tell you whether you are a candidate for the drug or not. Some people are, quote, allergic to certain drugs. Some people are highly responsive. Some are fast responders or slow responders or fast metabolizers. So these are things also that metabolomics and biomarkers are of interest. If you look at the statistics for biomarkers, the vast majority of approved tests, biomarker tests, are genetic tests. It's over 2,000 that are listed, although if you break those down, it's actually something on the order of about 100 diseases. So it's 2,000 genetic tests, 100 diseases. So it's a lot, but they're essentially for the rarest of rare diseases. So those are around. They're available. They are used. There are some microarray tests. They're a total of five that have been approved, although most of them have not gone very far because they're so expensive. Proteomics, although there are lots of protein tests, like C-reactive protein and glycated hemoglobin and others, tests that are based on doing classical proteomics like the mass spec. There's only one, and that test actually was converted to an antibody test a few years ago, so it's now down to zero. So metabolomics, which is sort of the young gun here, actually has about 150 tests. It's each metabolite tests that are used or approved. And it's a little generous in terms of my definition, but the fact that glucose is a metabolite and we do that, measure that pretty routinely, creatinine is another test. But all the inborn errors of metabolism, all the acylcarnitines, all the amino acid disorders that are checked, there are about 150 metabolites that are measured or routinely measured. And these tests actually allow us to detect, diagnose, characterize, assess about 80% of diseases. So while the vast majority, certainly of the news and information and press that you get about new potential tests is about genes, the ones that actually are doing most of the work, the heavy lifting are metabolite tests. So I'll give you some examples of ones that we've been involved with or colleagues of ours have been involved with. One is an example of metabolomics and a biomarker for pneumonia. So pneumonia is actually a pretty easy disease to diagnose, except you usually diagnose it long after you've had it. So it starts off, you've got a cold, the cold doesn't go away, the cough doesn't go away, you get sicker and sicker. So by the time your death's door you go to the doctor and say, yes, you have pneumonia, so big news. You can characterize someone with pneumonia because you'll see by an x-ray, chest x-ray that they've got pneumonia. But that's no news either because you've been coughing up your lungs for the last week. So it's not particularly useful. What they would prefer actually is to distinguish between pneumonias that are treatable and pneumonias that aren't. Treatable pneumonias are bacterial pneumonias, the untreatable ones are viral pneumonias. Most people, if they're given bed rest and some prophylactic treatment will recover from viral pneumonias, bacterial pneumonias are actually very treatable because you can give them antibiotics. So this particular paper study was actually looking at urine. Urine's a noninvasive test. You don't have to do x-rays. You don't have to do a three-day culture. And they're just simply looking at urine of patients from pneumonia, having pneumonia and distinguishing which ones had not only what type of pneumonia, viral or bacterial, but also distinguishing those who had TB, which is something that people commonly confuse with pneumonia. And what was surprising is that you could distinguish them from urine. I mean, what does urine have to do with your lungs? So this is sort of a head scratcher, but it's still quite striking. And this just is illustrating the different amino acid concentrations. And certainly people with pneumonia and TB have very high levels of branched chain amino acids, but they also have very different levels of lactose and suberate. And because of the different combinations, it is possible to use metabolomics with these six or seven biomarkers to distinguish between viral pneumonia, bacterial pneumonia, healthy, which is trivial, and people with TB. So that's kind of a cool test. And in the developing world, being able to distinguish between bacterial and viral pneumonia and TB is actually very, very important. Another one that we've been involved with was looking at kidney transplants. So kidney transplants, most common transplants done around the world, thousands are done, it saves thousands of lives. A kidney transplant can last for a good long time if it's monitored correctly. And to monitor it basically means once you've had a kidney transplant, you get to go in every few weeks and then get to stick a long needle in your back and take a tissue sample out. That hurts, it's painful, and then it goes to a histologist and they look at it and they look at it again and sometimes they say, oh, it looks like you're rejecting. And about 30% of the time they're wrong. So you end up getting a whole dose of drugs, your face swells up and you feel awful, then finally your organ settles down. They've tried to do biopsy monitoring and through another way. They've tried blood with microarrays, didn't work, they actually had to go to biopsies. So there really isn't a simple, non-invasive test to tell you whether your kidney is malfunctioning or starting to malfunction. So we wanted to look to see if we could use metabolomics with this. And we've got a group in Winnipeg that we've worked with. We found just with a simple urine test there were about half a dozen metabolites that could distinguish between people who were starting to reject and those who were essentially having healthy organs. So red is the normal transplant. They're healthy, nothing to intervene. Green is the individuals who are starting to reject. And it's not as if the organ has fallen out. This is where they're in the early stages and you're trying to identify so you can give them a slightly increased dose of cyclosporin. So this test, as I say, is non-invasive, much more accurate than the histology one and certainly a whole lot less painful. Another area where metabolomics has hit the news, particularly even the last year, is the relationship between metabolites and cardiovascular disease, specifically atherosclerosis. A group with Stan Hazen in Cleveland made a really interesting discovery that was first reported in 2011. They've repeated it and published two more times on different samples and different cohorts. But this is this connection between what you eat, your gut microflora, and heart disease. So everyone knows that if you eat too much fatty food, McDonald's fries every day, lots of eggs and fried food, you will develop cardiovascular disease over time. So you'll become obese, but you'll also have your arteries get thick. But there are some people who are not susceptible to that. They can basically live off fatty food all the time and they just don't seem to get cardiovascular disease. And there are other people who basically are diligent about eating correctly but do develop atherosclerosis. And this is where people have said, well, in fact there is a bacterial connection. People have seen bacteria in plaques. People have seen the fact that there are different families, different backgrounds. There's something, well, first of all, it was purely genetic. Of course, the argument has been it's all cholesterol, nothing but cholesterol. But I think there's been more and more studies essentially disproving that. Cholesterol has almost nothing to do with cardiovascular disease and atherosclerosis. So what these guys found was it had to do with what you ate and what was in your gut and what happened to those products courtesy of your liver. So if you eat fatty foods, you get lots of phosphatidylcholine. Phosphatidylcholine is then cleaved to choline. Choline then goes to the gut and your bacteria. Your gut microflore convert that from choline to betaine to trimethylamine. Trimethylamine then floats around and is converted to trimethylamine oxide and that's called an atherotoxin that causes atherosclerosis. And so this is a picture. Eat your fatty foods and they'll be converted to choline. But if you have the wrong gut microflora, they'll convert the choline to trimethylamine. If you have the right gut microflora, they'll probably leave the choline alone. If you have a well-functioning liver, the liver will take the trimethylamine and convert that to trimethylamine oxide. And they've done studies where they inject trimethylamine oxide into rats. They develop atherosclerosis and they've measured trimethylamine oxide in people. People with the highest levels of trimethylamine oxide have the highest levels of atherosclerosis. So there's this connection between what's in your gut and what you eat. So obviously if you don't eat any fatty foods, generally you're probably going to be okay. But it's partly the probiotic thing, but it's also... most of your gut microflora is determined when you're in your first year of life. And so it stays with you, really, even if you do follow eating lots of probiotics. So the gut microflora is kind of stuck. You're stuck with it. But if you're fortunate and have the good bacteria, then go ahead and eat at McDonald's every day. Another one which has come out, which is Bob Gersten at Harvard, Massachusetts General Hospital. They've done some very elegant studies with the Framingham cohorts and also Swedish cohort. And this is the idea of predicting who will develop type 2 diabetes. So this is the diabetes of eating too much fatty food, being obese, but in fact it's not always the case. There are people I know, and maybe even you know, who are healthy and fit who also develop type 2 diabetes. And this is partly why we figure, oh, diabetes is genetic. Well, actually it's not. Anyways, what they found in the metabolomics, and this is targeted quantitative metabolomics, was that people who did develop diabetes up to 12 to 15 years after, so they were presenting as normal, they were presenting as healthy, fit, normal weight, mid-40s and 50s. When they followed them for 12 years after, the ones who had the highest levels of branched henamy acids and aromatic amino acids had the highest probability of developing diabetes. It's not shown in your textbooks, although I think it is in the small molecule pathway database now. Branched henamy amino acids in particular leucine is an insulin analog. The body uses leucine and isoleucine as insulin. It binds to the mTOR protein which is a master controller for metabolism. So, if you have high levels of leucine, it's equivalent to having high levels of insulin all the time. So, diabetes type 2 is called insulin resistance. So it means that if you have been exposed or just like with your head being exposed to loud sounds all day, eventually you get resistant or you get deaf. Same thing happens with constant exposure to leucine or other branched henamy amino acids. The pancreas just doesn't respond anymore to leucine or to insulin. So when they compared how well this did, they set of 5 amino acids to the current collection of GWAS information, genetic links to diabetes. The metabolites explained anywhere from 5 to 10 times more of the variance or likelihood of developing diabetes than the genes did. Part of the irony is that we've spent probably more than $100 million on GWAS studies for diabetes and this study probably cost $30,000 and they found a whole lot more information. They did a repeat on this with the Swedish group, got the same result and then they expanded their analyses and found that there's one other compound called alpha amino adipic acid and it's essentially doubled the blue bar for metabolomics. So combine these 5 amino acids with alpha amino adipic acid and you're explaining at least 30 times more of the variance than what GWAS does. So these are some examples of people finding metabolite biomarkers that are either predictive or diagnostic or help understand better understand the etiology of the disease. What if you want to find biomarkers and get your name in light? So what you need to do to design a biomarker study is you need just as we talked about a group of at least 30 to 40 samples individuals, so there'd be 30 controls 30 cases, 100 controls, 100 cases. To do this you also want to try and limit the biological variability. We're trying to create biological replicates so for humans we would have much people with age and gender or not perfectly but close. What you'll also need to do is if you're going to collect samples you might get urine or serum but you want to make sure you store them properly. Collect it, do it quickly freeze it quickly. Make sure that you don't add a bunch of junk like polyethylene glycol or lots of EDTA or heparin and other things that's not good. You can add azide, that's helpful. You should freeze it at 80 degrees. That's the coldest you can get. Generally you should try and collect samples at roughly the same time so that you are sort of controlling for diet and other diurnal variations. For biomarker studies and this is why I've talked about this things have to be quantitative so you have to use what we call targeted or quantitative metabolomics. If you can't identify the compound the biomarkers can never make it in the clinic. They'll never be allowed in the clinic. Once you've found your biomarkers and they look promising you have to do another study with probably an equally large cohort and that's called a validation study. And it's sort of silly because in principle if you've done the statistics right you don't need to do this but the practice in medicine is you should always validate and it's probably a good idea because there's always surprises. You can use certainly different technologies. I would advocate using all of them because this allows you to look for as many biomarkers as possible. If you're only using LCMS the problem is you'll tend to find a lot of unknowns many cases you can't quantify them therefore you can't translate them to any clinical utility. On the other hand if you're only using NMR you're only measuring sort of the accessible ones and so you're not going to find maybe a truly unique or completely novel biomarker. So let's say you've used several techniques. You've got your list of metabolites you've got their concentrations you've got people that are age matched, gender matched you've got healthy and control. What do you do? Well the next step is go to something that Jeff wrote which is called Rocket. Enter the data follow instructions and you'll try and find a biomarker profile. So why Rocket? Rocket is short for receiver operator characteristic explorer and tester. Rock, receiver operating characteristic is a plot. It's a way of measuring sensitivity and specificity. It's become the standard way that people measure the quality of biomarkers. It was actually developed during the war World War II tracking how many times enemy objects were being hit by artillery. So just like the Poisson statistics came out of Prussian soldiers dying from horse kicks, rock curves we can thanks British military engineers for. A rock curve, a good one should look sort of like an upside down L or a logarithmic curve. So it's sort of a curve. Completely random rock curve is a straight line with a slope of one. We measure the quality of the sensitivity specificity by the area under the curve. So if the area under the curve is one, that's a perfect rock curve, if the area under the curve is half, that's a random rock curve, that's a straight line, and areas under the curve about 0.75 are what you try and aim for. So these are some examples of rock curves and the area under the curve. So there's a blue rock curve which sort of follows a slope of one. It has an area under the curve of about 0.5 or 0.52. There's the one that's a red rock curve and it has an area of about 0.8 and then there's this purple rock curve and that's about 0.9 or 0.95. So we're plotting, one is false positives or one minus specificity and then on the Y we're plotting sensitivity or true positive fraction. So that's how a rock curve is plotted out. I'm not sure if I've shown that, yeah I probably should show that a little more clearly. So if you look at rock curves for some common medical tests, here they are. So mammograms, some of you might have heard about the news but basically they've decided to stop mammogram screening and the reason is because of that rock curve. In terms of distinguishing benign versus malignant tumors it's random. It's not much better than flipping a coin. So yes they can detect a mass, that's useful, but it really doesn't help you to tell whether it's benign or malignant. Yes? I would just say it's only because the data they're using for the very wide range of calendar years that incorporates many changes in technology but that's why I think the newer ones are getting better. But this is the reason why they've been essentially pushing or having this debate. The PSA test is also one that is pretty marginal and the area under the curve for that is 0.65. Same situation as Michelle brought up, the PSA test has changed so the earliest ones were almost completely random, later ones prostate-specific antigen. So this is a prostate cancer screening. Later ones they finally figured out what they're supposed to be testing for so those are a little more accurate but it's still a case that overall it hasn't done particularly well. So those are some examples of rock curves. If you want to use or calculate a rock curve, you can go to Rocket and just like Metaboanalyst you can click to start. You can upload some samples and submit. You can check your data just like Metaboanalyst. You can do some a variety of quality assessments looking for missing data, interpolate add it, replace missing values. So again, it's a lot like Metaboanalyst that you guys have just done. You guys made it? Yeah. You can do some log transforms and scaling. Again, a lot like Metaboanalyst. You can check your data for normality just like you did for Metaboanalyst. And then you can decide whether you want to just do a classical univariate rock curve a multivariate rock curve a rock curve based model to sort of test which ones are most suitable. So sort of an interactive one. In this case you just use the multivariate rock curve so it will try and use multiple metabolites and determine what's best. And for this particular sample that we chose which I can't remember which one we used you can calculate a rock curve where the area under the curve is around 0.95 0.97 What's interesting and it's important when you're calculating biomarker rock curves is you want to minimize the number of metabolites and maximize the area under the curve. The reason why you want to do that is that if you have very few compounds it's a practical test. If you develop a test that says I need 170 variables that I have to measure that's a very very expensive test to do. If you say I can do this with two or three compounds that's a very cheap test to do and if it gives you essentially the same performance as one that requires 170 then that's the one you want to go with. So you would have a table of concentrations or relative concentrations exactly and you have your cases and your controls so two sets disease or healthy and upload it as it is just like you were doing with your cow data or your breast cancer data and you just look at Kexia and just run it through there. So as I say the analysis is very similar to what you would have done just like you did this afternoon from a table analyst. It doesn't have quite as many options as a table analyst and I think in the future rock it will probably move right into metabolism as 3.0 but on its own it can do a fair bit of standard work. Clean up the data and then you'll get these rock curves. It gives you a confidence interval I don't know if it plots the width of the confidence interval in the curve or not Yeah, it would be kind of messy. So this is just giving you a bunch of different curves and then you can choose one. In this case the one with two metabolites is probably the one you want to go with because that's the smallest number gives you pretty much the same answers just about every one. You could go for the three one it's sort of up to you. So this is where humans have to decide. I can also identify from there because you can click on different tabs identify which ones are the most significant metabolites in this case glycerol and 3-hydroxybutyrate were the most significant metabolites that identified individuals in this case with preeclampsia or who would develop preeclampsia. So this is a tool that allows you to do it and in fact we've used this tool in Edmonton to look at a whole bunch of cohorts to look at cases and controls for a wide variety of diseases where we've looked at blood or urine and I'd like you just again to recall what the performance was for PSA tests and for mammograms. The area under the curves is about 0.53 to 0.65 So these are some rock curves for predicting preeclampsia. So preeclampsia is a condition that pregnant women will develop usually 6th, 7th, 8th month of the pregnancy. It's high blood pressure. It's the leading cause of infant and maternal morbidity and mortality in the industrial world. So it's not really common but it's common enough so that it is something that causes deaths. And the point here is that we're trying to take from a serum sample at 3 months. So this is where the woman is figuring out she's pregnant and is also perfectly healthy and we're trying to identify which ones are going to develop preeclampsia. So in this case on the left it looks like we can identify just a couple of metabolites. We'll give you an AUC of 0.99. That was the glycerol and hydroxybutyrate. Early preeclampsia is much more serious disease. Late preeclampsia is somewhat more common, not quite so serious. It's dealt with just simply with bed rest but eventually it could be preventable if we could predict them. So same thing person at 3 months into their pregnancy blood sample we're able to predict which ones will develop with an AUC of 0.96. In this case the curve or test requires 8 metabolites. Birth defects so these have to do with trisomy, trisomy 21 which is down syndrome, trisomy 18 which is the second most common chromosomal defect. We can detect these now with amniocentesis but that's an invasive procedure. It's a big needle it's I don't know something around the fourth, fifth month or something. So the idea here was it could you detect these conditions at the third month. So again, normal looking pregnancy very early so in the case of trisomy 18 7 compounds from serum allowed us to detect or had an AUC of 91%. Down syndrome was harder the performance just with metabolites alone was about 83 to 85%. When we included age which is something that we normally screen with with Down syndrome the performance for the test was about 90% AUC and it was just with 3 metabolites. The most common birth defect is a congenital heart defect and this is one that we can actually do something about especially if we have enough warning so they are doing heart operations in utero and it's becoming more common so they can repair septal defects and other things but they'd like to be able to identify the babies that have them early enough so they can both prep the mother and the surgery so that they can do it. So the question was can we detect congenital heart defects within the third month of the pregnancy so this is very early on and we were very surprised with this one but again just with 3 metabolites from serum the AUC here was about 98%. You guys were working with some sample data we talked about the cancer catechia one that was one of the data sets we looked at in the finalist so this is real data published a few years ago this was also can we predict who will develop catechia and who won't so it's pretty easy to identify someone with catechia we don't need a biomarker for that but if you have just been newly diagnosed with cancer it would be nice to know are you at risk for developing catechia or not so that's what we were asking and from a simple spot urine sample it looks like we can it's basically four compounds in urine that seem to identify those who are going to be developing cancer catechia and there are potential treatments for it so if you can get it started early then you can prevent people from developing catechia and that could give them many months of life we talked about transplants this is a case where we looked at both adults and more recently pediatric transplants the adults one uses a larger set of metabolites but very accurate more relevant one is actually a pediatric one because those are ones where actually kids reject kidneys much more frequently there are many more problems with them and so they'd really like to be able to monitor identify those at risk and so this one just from a urine sample four compounds gives a very good predictive or semi-diagnostic method for detecting transplant rejection the idea is you don't want to bring kids in every week and put a big needle in their back and pull tissue up so this would be a nice way of just doing the monitoring and then making useful decisions some of you have heard of conditions like chronic fatigue syndrome this is one where people are very very tired all the time bedridden it's been debatable is it a real disease is it not anyone who knows someone with CFS would argue it's definitely a real disease is it something that's a mental disorder or not what was interesting at this is that we could identify people with chronic fatigue syndrome with set of six compounds very interesting distinct metabolic signature for people with chronic fatigue syndrome another one is called eosinophagitis this is a game where kids have difficulties swallowing and eating and it's a condition that can be mistaken for many other conditions and so in order to distinguish it they have to do tissue biopsies so they stick something a big set of scissors down your throat and cut out tissue so again it's not something kids like to do and especially given that the throat is constricted it's very painful it's like a non-invasive way of identifying this disease and distinguishing it from others and so again it looks like urine allows you to distinguish some of these fairly easily heart failure different types of heart failure some are treatable some aren't so there's systolic heart failure and diastolic heart failure different methods to treat them but very hard to distinguish them to distinguish them you have to do probably about a week or two of tests and monitoring so it's expensive people would like to find a simpler test and so the idea is just to look for serum look at metabolites see if you can distinguish systolic from diastolic and again very simple signature for compounds separates these things very very nicely colitis this is a condition ulcerative colitis which is very painful it's grouped with inflammatory bowel disease this one is not just saying do you have it but will you have a flair so people have kind of a lifelong condition with ulcerative colitis but they can live with it except when it flares in that case it's a sort of painful episode of bloody diarrhea for days or weeks and it can be fatal if you could identify people that are about to develop or could develop then there's potentially things you could do to prevent it so we'd like to be able to predict these flares turns out just from blood samples you can do this very very well from urine samples you can do it not quite so well so it's a perfect L so it's it's dead on so it's very striking yeah so this one it's a smaller sample set so we're not seeing the actual confidence interval so yeah it's not perfect because the subset was too too small but it is compelling and it's certainly telling us that there is certainly clear metabolic changes that are happening mostly in the gut microflora that pre-sage an attack of ulcerative colitis relating to the colon colorectal cancer this is a major cause of death in North America also in Asia if you want to be able to treat it early detection is critical but the way that we standardly do it is not so good so in the U.S. they do colonoscopies those are six or seven hundred dollars each one they're pretty painful but they have great lengths to avoid them in Canada we don't do colonoscopies we do fecal occult blood testing people also go to great pains to avoid using those there's about a 5% compliance rate so this is a condition that could be easily treated but people just because they have these obsessions about things going up their colon don't do anything about it so the idea is if you could avoid the colonoscopy or avoid the fecal occult blood test and just go to a urine test you'd have almost 100% compliance so we've been working with a group at the OVA where they have developed a urine test where they can use 13 compounds and distinguish those with polyps from without so this isn't cancer this is polyps so this is pre-cancer that's something you can do about it so this is predictive the test isn't perfect the area in the curve is 0.74 but when you compare it to a fecal occult blood test it's about 3 or 4 times more accurate so these are some examples mostly because people don't want to have or donate fecal matter so it's just it's the yuck factor and so urine tests are preferred ideally a blood test would also be preferred or saliva test but yeah the whole thing is the compliance with it is it tends to reflect a lot more things going on but it's also pretty sensitive too and it turns out all these markers are bacterial microbial so they just have been recycled into the system out into the kidneys so there is a clear microbial that's going on in the polyps or areas around the polyps and then it's just reflected in the urine I've mentioned this issue about GWAS we talked about diabetes and the studies 100 million dollars have gone into diabetes a single GWAS study is anywhere from 1 to 10 million a single metabolomic study can be as little as 100 dollars but a more extensive one is up to a couple hundred thousand because of this false discovery rate with genes because you're measuring thousands and thousands and thousands you have to have very large patient samples because metabolomics were able to do sort of a broader survey with fewer things we don't have this big issue with false discovery rates so we can get away with smaller numbers certainly practical experience tells us that metabolites we can get away with smaller numbers rather than with GWAS studies among the GWAS studies that have been published where they've indicated areas under the curve for specificity sensitivity they have around 51 to 60 percent whereas the ones I was showing you have been hovering around 85 to 99 percent GWAS studies are not used for diagnosis they're used for prediction but you can also use you can use them for prediction but whereas metabolomics I showed you examples where they are diagnosing and predicting I can also argue that many of the metabolomic tests especially if they're three or four compounds can be done for about 10 or 12 bucks if you're trying to do a GWAS test it intrinsically is always going to be expensive because you have to do sequencing or microarrays so that's sort of metabolomics and MetWAS and GWAS applications to pharmaceutical research in terms of drug discovery yes Caroline so the colon one is is being worked through the polyps is going through that the preeclampsia ones we're starting to validate the other ones we're still waiting for more samples to do validation so they're stepping through yeah I mean I think my expectation is that of these maybe two or three we'll actually make it to some practical use the other is I think we'll find that the findings are not as robust as initially found and that's why this validation is still important so statistics tells us we can be confident but human variability is something that statistics doesn't always handle so I think we're going to go through some of these that were handled so drug discovery so the reason why metabolomics got its start and it's originally called metabanomics was in the area of drug discovery and development problem is that drug discovery is extremely expensive huge numbers of patients like success rate for compound identification that phase one is one in a thousand 20% of phase one drugs ever get approved so there's a huge failure rate but metabolomics has found a niche in just about every aspect of drug development in the discovery phase in phase one, two and three trials but also in FDA approvals certainly genomics proteomics and standard chemistry play a key role but the surprising thing is that metabolomics has found its way in just about every aspect and this is just highlighted here and these are slides I got from a colleague in the drug industry but they are showing that metabolomics is used in identifying the targets so just where we discovered all these biomarkers for these diseases these also indicate some potential drug targets people use metabolomics to do some talk screening they do this to help identify markers for efficacy they also help from that which leads to work with they do this with monitoring and biomarkers with preclinical phase early phase toxicity and then once the drugs are approved or getting through late stages they use clinical efficacy and clinical safety markers so this is an example where they use metabolomics to help discover some drug targets and sarcosine and two hydroxyglutarate are examples of which are called oncometabolites compounds that are associated with cancer or caused cancer sarcosine was when it was discovered by a group in Michigan and they looked at people with benign prostate cancer and metastatic prostate cancer and they found this derivative of glycine called sarcosine was substantially elevated they looked at a whole bunch of compounds over many different samples they were looking at tissue samples so it wasn't a blood test but the ones that had high levels of sarcosine had metastatic cancer about 80% of the time and they found them in an early prostate so this suggested sarcosine is not only well it's essentially a prognostic marker predicts whether it's going to progress to a bad form and certainly people who were cancer free didn't have any sarcosine there's an enzyme GNMT that is involved in sarcosine metabolism and when they knocked down that enzyme prostate cancer went from being metastatic to non-metastatic so the metabolite which then allowed them to look at the gene and allowed them to do a genetic study with knocked down suggested a drug target so this is metabolomics not only identifying a biomarker but also identifying a drug target and there's a fairly active program now developing gene MT inhibitors so drug discovery and genomics versus metabolomics most genetic diseases are things that we think are genetic diseases are polygenic diseases meaning they are seen to be caused by many many genes many different snips or mutations and given that the most complex diseases are usually the ones of greatest concern and given the complexity of the diseases we've needed huge cohorts the GWAS studies have allowed us to identify a lot of interesting snips and mutations literally dozens of them there's databases for these and the way that GWAS concept is sold was that we would use this to identify the genes these are then related to what we call drugable genes some aren't some are so these are ones that could serve as drug targets once you identify the drugable genes and you can clone them get the proteins start doing screens identify some drug leads and you can start doing preclinical phase one two and three trials and then you've got your drug so that's the model and that's called gene centered drug discovery and that's the model that pharma industry has been buying into for the last 10 or 15 years here are some numbers because we've been doing this for a while we know that GWAS studies are about 20% successful in finding useful gene targets that cost a lot of money to take a fair bit of time of the genes they identify about one in two are actually drugable of those ones that they can actually clone or produce to get material to do some testing about half of them are cloneable of the targets they can get about 20% can get to past phase one and then or early stage clinical one and then if you go to the phase one two three we know there's about a one in five hundred success rate and then once things are actually approved especially among the recent drugs there's about a 50% chance they'll eventually have to be pulled so if you multiply all these odds all these years all the costs the process of gene centered drug discovery yields about a 0.001% success rate costs it takes about 20 years and costs more than a billion dollars of drug that's one of the main reasons why the drug industry is struggling now in fact their pipeline is completely dried up most of the way they're making their money is consolidating buying each other up buying other drug companies selling off assets most drugs are off patent now and they're starting to feel that this model wasn't the way to go it's also important to remember that not all diseases are genetic when we look at heritability some of the diseases with the best GWAS data only tells us that they are about 20% of them seem to be genetic many cancers that we think is purely a genetic disease at least 20-40% are caused by bacteria or viruses there have been a few review studies articles that coming out 80-90% of diseases either have an infectious origin or acquired or have a rise from cumulative exposure so that suggests that maybe we're looking under the wrong tree so this is the idea about metabolite based drug discovery so in this case you do from tabloid associations just like the sarcosine and cancer just like the pre-eclampsia studies look what's going up or down go through your pathways you guys have all done this if you can figure out if you're altered you can go down and look at things like Brenda which is an enzyme database you can go to sigma or drug bank to see where these particular pathways have inhibitors or activators in many cases a lot of the inhibitors and activators are common compounds some of them are already approved drugs some of them are already essentially supplements once you've been able to identify some of these things or use metabolomics to monitor things so how might that work well we know that metabolomic studies are pretty cheap, pretty fast our experience about 50 to 60% of the studies we undertake give us an interesting hit you guys have spent the last day and a half doing steps one and two so it doesn't take long to sort of figure out what's relevant you can just go into the web there's a bunch of resources to figure out what useful inhibitors there might be obviously if the inhibitors aren't small molecules then you have to do things like antibodies and knockdowns and other things and then because metabolomics is ideally suited for monitoring things you could also monitor what's going on so if you calculate the costs the odds and other things this approach could potentially be much more successful take much less time and cost a whole lot less money is this feasible well we can kind of do a thought experiment we could take some examples of diseases that we are dealing with or have dealt with so let's say all of our sailors kept on coming back from the oceans with the scurvy and we said well what's going wrong we could start testing them and we'd find that they're all low on ascorbic acid and we could say oh well maybe a way to cure this scurvy condition is to give you vitamin C or lines so this was actually figured out 200 years ago but something that was figured out much more recently was neural tube defects and this is why expected mothers are asked to take folate supplements and this use of folate supplements has cut down the incidence of neural tube defects by something like 80% many conditions which were common especially in the early 20th century the anemia, polygra, rickets in the 18th, 19th century have all been dealt with by essentially standard supplements these, if we had the tools could have been picked up using metabolomics the treatments also could have been easily identified as well to appreciate this 120 years ago probably a third of all people had these conditions so they were prevalent very prevalent there are other conditions which we can diagnose and look at people certainly with epilepsy have very low levels of ketone bodies so this is ketogenic diets or very effective treatments for epilepsy we can look at high lactate levels and then treat them with thiamine supplements certainly with PKU we can look for phenylalanine so this is the most common metabolomic test in the world and in fact it is very treatable so these are all examples of where metabolomics or metabolite measurements can and could have been used there's no evidence to cheat I guess so but the point is that we have an obligation and it's more at the level of governments to do these sorts of things and we have fairly active programs certainly for treating the treatable metabolic disorders so these are all examples where conditions can be detected very characteristic metabolites and where there are treatments some of which are relatively inexpensive to perform we can also use metabolomics for studying drug metabolism drug disposition drug excretion with pharmaceutical development so in this case we monitor the systems we don't have to actually sacrifice animals so this is cheap and we can work these studies with rats and so this is becoming quite standard now and this is pioneered largely by Jeremy Nicholson's group and his large pharma has bought into it and these are some examples where they're looking at different drugs and seeing how rats recover with some of these toxins or drug derivatives and you've seen this already these are some of the results they measured and identified different responses and whether they recovered they could also identify which organs were being damaged and even which regions in which organs were being damaged and again they didn't have to sacrifice the animal or open it up so this is a very cheap, rapid way and they're just simply looking at urine and for this case they were using NMR but they could have used any other technique so as I say this gives you a way of matching specific metabolites to specific damaged regions to specific organs and they can do this with kidney damage they can do this with liver damage and so on so these are essentially lookup tables that they can use to figure out drug toxicity they can go to humans and they can start monitoring whether people are taking the drugs or not so we can see who took the dose who didn't whether they missed a dose and so this is important in clinical trial work because obviously you want to have good compliance and if people aren't complying well then the results from that particular patient can't be used so we can have or not avoid alcohol or drugs tabulomics is actually pretty good at being able to identify when people aren't avoiding those we can use metabolomics to figure out what people are eating we can also use metabolomics to figure out what's in foods some are simple foods some are complex and some metabolomics gives us many many hundreds of compounds that we can read we can distinguish between different foods we can also use metabolomics to identify adulteration whether things have been modified these are very strict rules about what can be added to foods particular things like juices and wines the types of sugars that can be added again to sweetened juices illegally or illegally we can also see how metabolomics changes our gut microflora and how that changes composition of blood, urine and we can see certain kinds of biomarkers or metabolites that indicate what food we've eaten so in some respects if you're good enough you can do or perform experiments without worrying about who's eaten what because you can see this and so you can control for these things so there are starting to be increasingly large numbers of food biomarkers gathered and being used as I said before what we eat also influences what grows in our gut and if things are perturbed through either diet or probiotics or disease we can also detect those changes certainly there are many many conditions that we're seeing gut microflora seems to have a very significant role obesity being one and this is paper from seven or eight years ago showing that gut microflora is associated with obesity and we're starting to realize that there are some very distinct nutritional phenotypes for people depending on where you live, your ethnicity we see there's very distinct diets, distinct composition for microflora and these things stay with you this is a study done in Italy where they looked at people there's about 15 different individuals over a long period of time and basically people yes show some variation but your urinary phenotype, your blood phenotype is like a fingerprint it stays with you your whole life it varies but it's distinct and it's unique so every one of these individuals was unique and they didn't crossover or match anyone else in terms of what's ahead from metabolomics I think there's exciting times some of the most interesting stuff is the concept of metabolomic imaging or metabolite imaging so we can do this with MRS or MRI but we can also do it with Maldi so MRS, PET scanning are all examples of metabolite imaging but we can do small cell and tissue imaging with matrix assisted laser desorption ionization and this is an idea where we put a tissue sample on and it's a special matrix several matrices actually being developed in Canada for this specifically and it allows you to look at very low molecular weight metabolites and you can scan across the sample sort of rastering across collecting thousands of mass spectra identifying all of the metabolites and then coloring things another trend in metabolomics is not using GCMS or NMR or LCMS but to go to portable devices handheld devices ISTAT is one example it's a portable pseudo metabolomic device there are also techniques for measuring volatiles called the electronic nose or e-nose and these are very innovative polymeric nano sensors they can distinguish based on the patterns whether something is acetone, benzene or chloroform so these are all examples of trends that are starting to emerge and some new ideas that are developing in metabolomics so while I've been pushing the utility of metabolomics I still think it's important to understand that metabolomics doesn't stand on its own it has to be integrated with genomics and proteomics it is suited for lots of different applications we can use it in modeling we can use it in prediction and the reason why we can do that is because it uses well understood chemical principles and well understood pathways I've given you examples of how metabolomics is being used in environmental science and medical science and agriculture even in petroleum research I think what I've also tried to emphasize is that good experiments are key, they're well designed experiments are key good quality control, quality assurance good understanding the methods the analysis understanding the statistics and why you're doing that those are all important so that's why we have the course I think we've also seen in terms of trends I think metabolite imaging is something very exciting you've seen some examples as well where we're automating techniques in metabolomics whether it's the data analysis, sample collection there's also emerging trends to look for volatile compounds, the electronic noses more gas chromatography most of the flavor associated with foods and many interesting conditions including lung cancer seem to be very detectable and diagnosable you're looking at volatiles I've told you about the trends towards quantitative metabolomics hopefully you guys will buy into that idea smaller handheld kits making things cheaper, faster, better and then trying to go towards at least in the world of clinical work diagnostic and predictive applications it has to move out from the lab into real space so people are actually doing that in terms of context metabolomics is a new kit on the block really only got started in 1999 here we are at 2014 is it going to continue to grow? I hope so certainly we're seeing genomics has been around for long it's not fading, but in terms of innovation it's begun from being something that you would do your PhD into one where it's so routine that it's a commodity sequencing is a commodity proteomics is a similar sort of thing it's sort of reached its zenith about 2010 technologies are largely now become commoditized for it yes, people are still doing their PhDs in proteomics but it is becoming much more standardized I think you've seen from discussion series there's still lots of things to do for metabolomics we've learned a lot and will learn a lot from genomics and proteomics and the technologies and techniques and statistics but I still think there's lots to do in the field so with that I want to thank some of the people who have helped put this together Jeff who's been your TA and diligently stayed here late nights early mornings great Michelle who's been responsible for these programs for the last six years almost put it all together and made sure you guys were well fed and well watered and made for all the arrangements and then we've had sponsorships from Genome Canada OICR, Albert Innovates and many others to keep this going so thank you guys for sticking around and surviving the workshop