 This will be, I think, a shorter talk, and it's really more light than heavy in the sense that we're just simply trying to look at where, or I guess more an opinion piece of where I think metabolomics is going. The idea is to sort of inspire you to sort of, I guess, suggest some things where you might want to explore, but also to encourage you to explore. So over the last 10 years, metabolomics has grown quite considerably, and this is just simply doing a PubMed search, looking at terms for metabolomics, metabolomics, metabolome, and metabolome. You could do a PubMed search right now, and you'll get very similar numbers, but it's actually growing exponentially, which is quite something. The question is, where is it going to go after this? Is it going to continue growing? Is it going to flatten out, or is it going to plummet? And how it will proceed over the coming years is actually defined by where there are some critical bottlenecks in metabolomics that would either prevent it or allow it to realize what its real potential is. So I think you guys have already seen the first bottleneck, which is, in the case of trying to do untargeted LCMS, it was a lot of work. It's not particularly automated. I think you can also see that there are real challenges with the level of coverage that we get. I've brought up some of those points in earlier lectures, but you can also see from whether it was the untargeted data or the NMR data or the GCMS data that you're typically working with 50 to a couple hundred compounds that are identifiable. Metabolomics is really expensive to get into. I've mentioned a little bit, to some of you, I guess the Metabolomics Innovation Center, which is a center that I've been running for the last five or six years. NAMA is part of it, and Jeff was involved with it in the early days. It's based in Edmonton. It's distributed into different cities in Victoria and McMaster University in Hamilton and McGill University in Montreal. And all together, there's $30 million of equipment that is in this center. So it's not something that someone can just say, I'm going to do this tonight. It's a lot of work to build out. It's a lot of work to maintain. And so if that's the barrier to get into metabolomics, then it's too big a barrier for most people. There's a central issue in metabolomics, which is about quantification. I've been emphasizing that. And you guys saw this as well, where we had quantified data for the targeted work, but for the untargeted work, all of the data is non-quantified. It's only relative. Another central issue with metabolomics is that people are finding cool things all the time, but are we actually seeing metabolomics being used in the clinic? Are doctors showing up on TV advertising their metabolomic applications? And then I think the other key issue, especially with small molecules, is drug companies are the elephant in the room. They are the ones with huge R&D budgets. They're the ones that often direct and decide where major initiatives should be. They're the ones that really said we should do genomics. They're the ones that said we should do proteomics. So if you can make metabolomics matter to drug companies, that also will help sort of clear some bottlenecks. So those are bottlenecks. So there are some trends that are happening to essentially address those. One area people are working on is automating metabolomics. Another area people are working on are trying to expand the metabolome coverage. Another area is trying to make metabolomics more portable. Another area is obviously moving more towards quantification. Another area of active work is metabolomics moving from the lab to the clinic. Another one is trying to get metabolomics back into drug development and discovery. So I'm going to talk about each of these points over the next 40 minutes or so. And the first one I'll just talk about is move toward automation. So this has already been hinted at. So there are now more and more types of equipment that are automating metabolomics. So I mentioned the juice screener and wine screener. This is something that Broker developed a number of years ago. Well, four or five years ago they've also got one that does lipid analysis that came out last year. So these are NMR instruments. Can't even tell that there's a magnet in some cases. But they're designed for doing very high throughput analysis and very automated analysis of wine and juice. So you can figure out the exact composition, their provenance, where they're from, which country, which province or state or region. They can perform HDL, LDL, VLDL, lipid analysis as well automatically. And it's a walk away turnkey instrument. So you just have to load up samples and you can process hundreds a day. In the case of mass spectrometry, there are kit companies like Biocrities, which will be co-sponsoring tomorrow's symposium. Shimadzu, which is also moving towards this. I believe A.B. Sykes also is trending towards this, where either the vendor or independent companies are selling kits for people to do, relatively automated high throughput quantitative metabolomics. There are also many regional centers, which allow you to just send your samples in and get your answers back. So the Metabolomics Innovation Center, TMIC, is an example. There are six regional centers in the US that allow that. There are regional centers appearing, or national centers in Japan and the UK and in Holland and in France. And then there are companies like Metabolon, also where you can take your samples, send it in and automatically, well, a week later, get your data back. So it does cost money, but there is a trend towards more and more automation. And if you tour some of these facilities, you'll see how they are trending towards automatic metabolomics. We've also seen an example of not just the equipment that's being automated, but also the software. So you guys had a chance to use Bazel, and you could see that it took a couple of minutes. But if you tried to do this manually, it would take all of you anywhere from half an hour to two hours to do some of the stuff with just a single spectrum. So having it work in two to three minutes and having it process 20, 30, or 40 spectra in a reasonable time and quite accurately is, I think, an important advance. We've also been working in the area of GC autofit. You guys have tried that. That's another example of automation. And again, most of you have given it a try. So it's not just the hardware. It's also the software. And it's happening not just in Tmic, the Metabolonics Innovation Center, but it's also happening in other places in North America and in Europe and in Japan where more software is being added. Now, there's still a bottleneck, as you guys saw, with trying to do untargeted metabolomics, but perhaps in a year, perhaps in two years, it'll be as automated as Bazel or GC autofit. I think that would be a really important advance. OK, so that's a little spiel about automated metabolomics. How about expanding metabolome coverage? So this is a picture we've seen before, and it's just sort of highlighting the number of metabolites or features that can be detected with different technologies and their sensitivity limits. So NMR, micromolar sensitivity, maybe about 100 compounds. DCMS, submicromolar sensitivity, maybe about 200 compounds. LCMS, nanosubnanomolar sensitivity, and thousands to tens of thousands of features. So obviously, if we can go to more sensitive instruments, we can cover more compounds, but the dilemma for us, and as you guys saw, is that many of the features that we see through LCMS methods are not identifiable. You guys have put in a bunch of masses. We saw some features that were important. Some of them hit something. Some of them didn't hit something. And in most cases, 85, 95% of the time, the masses don't correspond to anything that are in our database. So the question is, what are these unknowns? And my belief, and I think a growing number of people are believing that these unknowns are actually metabolites of metabolites. These are transformation products of compounds that are already in our bodies or in plant bodies or other things, where other enzymes or other microbes or other processes are transforming and changing them into something else that is not common in other pathways. This is a picture that Augustin has been using in his lectures. It came from a paper that we wrote together about the food metabolome, or essentially exposures. And whether it's exposures that we get from our diet or exposures from drugs or exposures from pollutants, these all constitute either the pollutant metabolome, the drug metabolome, food metabolome, or the endogenous metabolome. And these are quite diverse. And the reason why they're diverse is that the compounds that are at the top are often ones that we already know. We have lists of all the known drug components. We have a pretty comprehensive list of all the chemical components in foods and the food additives. We also have a very comprehensive list of what are released into the environment in terms of chemicals, pollutants. But when these things go through microbial metabolism, house metabolism, whether they're digested or even if they're just processed through UV light or sitting on the water or microbes in the soil, they get transformed. And they get transformed into a green goop that we don't know about. And that's what we see with these increasingly sensitive instruments. So how can we actually characterize these things in order to be able to have continental identification in these compounds? We would have to have authentic standards for them. So whether it's 100,000 unknowns, 500,000 unknowns. And my own calculations suggest it could be up to 5 million unknown compounds. So to be able to synthesize 5 million compounds, to characterize 5 million compounds by NMR, to collect MS and GCMS spectra for 5 million compounds. That would cost anywhere between $5 and $10 billion. So the Human Genome Project was done for about $2 or $3 billion, and so these days no one's going to put that much money into sort of the chemical genome project. So what's the solution? If we can't get these authentic standards, if we're never going to have the chemical libraries, if we're never going to have the reference spectra, what should we do? So a task now, I think, is to move from trying to do it in the slow, tedious way of synthesizing and characterizing pure compounds for libraries to go to in silicoat metabolomics. And this is the idea of performing systematic spectral prediction for the compounds that we think exist or we know exist. So there's lots of compounds, but many of them don't have their spectral spectra collected. Many compounds that we know of at one time, but the spectra were collected 30 years ago and they haven't been updated, and so that's kind of useless. So this goes back to this concept of CFMID, but there's also programs like Metfrag and Metfusion. These are programs that can take a picture of a compound and predict the MSMS spectra or the EIMS spectra. There are also tools that we're working on to predict the NMR spectra of compounds in water, and there are also concerted efforts to try and predict the metabolism of compounds just from their starting components. So whether it's tools to predict spectra or tools to predict metabolites of metabolites, these are coming on stream fairly quickly, and these may in fact become the new reference materials for both the expected compounds as well as their expected reference spectra. So this will be, I think, an important way of expanding coverage of covering those unknown unknowns, of getting assignments to all those mystery peaks that Nama generated from her XMS analysis. We'll see. It may not yet work as we'd hoped, but it is already proving to be useful. There are a number of people that have been using spectral prediction methods to actually identify novel compounds, and there are a number of people who have been starting to use predicted transformations to confirm and also provide hypotheses for novel compounds. So that's, I think, another way of addressing the expanding bottleneck, expanding metabolism coverage bottleneck. The other one, I think, is very challenging for metabolomics. It's challenging for proteomics, it's challenging for genomics. It's expensive to do these things. You can't do it at home. You can't do it in the field. You have to bring samples back into an expensive lab that's very sterile. But one of the things that is happening is that people are moving more and more towards portable devices. So some of you might have Fitbits or have tried them. There are things called sensors and other kinds of tools that are like portable devices to monitor your health, can measure your blood pressure and body temperature, and they can even do some small chemical analyses. And so this concept of trying to democratize metabolomics to convert a 900 or GHz NMR instrument which would fill this room and two stories above us to something that you could hold in your hand is, I think, the best way of democratizing metabolomics. It's something that everyone could afford, everyone could do. It could be in every lab, it could be in the field, it could be in the farm. And so this idea of making a handheld system that could kind of scan up and down your body and tell you what's in you is actually not that absurd. So Qualcomm has just finished or wrapped up their XPRIZE for a tricorder. And it was actually awarded to a family firm in the U.S. and I think the Canadian group finished third in the XPRIZE competition to make essentially handheld medical devices that can measure or do a small amount of clinical chemistry but also measure things like heart rate, blood pressure, temperature, and other things. So, you know, if a family firm working in their basement can win a $5 million XPRIZE, it certainly suggests that the technology is starting to appear for doing miniaturizing of metabolomics. It's possible to do computer electrophoresis on a chip. It's possible to do HPLC on a chip. It's possible to do gas chromatography on a chip. And these are small chips and these are commercially available. It's also possible to get equivalent to electronic noses for small molecule detection. And these, too, are not much larger than a credit card machine. And they can, through various combinations of pattern recognition, distinguish between acetone, benzene, and chloroform. These are chemicals. They're volatile chemicals, but they are metabolites, if you want, or chemicals of utility for metabolomics. So, this is another trend towards miniaturizing chemical analysis. We've been working in Edmonton at the Metabolomics Innovation Center and with colleagues as well on trying to develop protein and RNA-DNA-aptum remediated metabolite sensing. And the concept here is to basically make use of proteins that bind metabolites specifically and to sort of carry out competitive binding assays. So, they're, in fact, whether they're paraplasmic binding proteins or antibodies, quite a number of proteins that are known to bind metabolites. Now, if you tag a metabolite with something like a gold nanoparticle or a fluorescent latex bead, you can actually carry out some kind of competitive assay. You can get compounds that bind, or if you've got the free compound, they can displace the gold nanoparticle from the protein. When you're working with gold, you can actually start doing a bunch of things with relatively low-cost sensors. You can use surface plasma and resonance, surface-enhanced Raman scattering, or you can look at electrical impedance because once the gold is displaced, these things change quite dramatically. So, if you've got proteins that are specific to metabolites and there's lots of them, then you can start making these kinds of handheld sensors. So, working with colleagues at the University of Alberta in electrical engineering, we've been developing this impedance-based metabolite sensor. And Reza has been involved with this work for a while. And so it uses either antibodies or paraplasmic binding proteins, conjugated metabolites, and a very, very small impedance sensor equipped with a very small inexpensive chip. The target is to be able to measure 8 to 10 metabolites at a time with different chips and to have the chips cost about $1 to $2 apiece. So, there are other people looking at miniaturizing metabolomics, but I think this is something that would make it much more accessible to a much wider community. Now, the other thing that I've been emphasizing, and I think some of you have started to appreciate, is the utility of having quantified data for metabolomics. Again, we tried to give you guys a flavor for untargeted metabolomics, which gives you relative quantitation or semi-relative quantification. We've given you GCMS and NMR metabolomics, which gave you absolute quantification. There's a cover of a Trends in Biotechnology article that was actually for the entire journal. It came about three years ago, and it had this title, a specter is hunting metabolomics, specter being ghost, specter of quantification. And two or three years ago at the time when this was published, more than 90% of the published metabolomic studies were basically semi-quantitative, people reporting relative peak areas, relative intensities. Less than 10% of the published metabolomic studies were using absolute quantification. Absolute quantification is key to reproducibility, and since this was published, a major crisis has appeared in all of science, which is the crisis of reproducibility. Most science is not reproducible. And the reason why is largely because people are not being careful with quantitation. They are not doing absolute quantitation. And the one or two fields where things are really reproducible is typically physics, where they're doing precise measurements of mass or the characterization of various leptons and bosons, where they have to be very precise. But in genetics and genomics and transcriptomics and proteomics, people are happy if they just get a set of values, which they report and only works in their lab. And if you go on any other platform, any other lab, you'll get a completely different set of values. Metabolomics has its origins in analytical chemistry and analytical chemistry, historically, was the most quantitative science in all of the fields of science. It was about reproducibility. It was about having standards. It was about reporting absolute concentrations in millimolar and micromolar and molar. It's also important that if you want to do practical applications in clinical work, in environmental work, environmental monitoring, in any form of veterinary work, what you report has to be reproducibly quantitative. So there are efforts with several companies. We've already mentioned Biocrities, Brooker. I've mentioned Canonomics, which is an Edmonton-based company. They've all moved towards quantitative tools for absolute quantitation. There's a number of academic efforts, whether it's things like Bazel and Batman and GC Auto Fit. You guys have tried a few of those. These are also trying to encourage people to do absolute quantitation. And when you compare metabolomics, actually, in terms of its record for quantification, it's actually not so bad. People have been characterizing serum and plasma. They've been characterizing cerebral spinal fluid. They've been characterizing urine for many, many years. And metabolomics, to date, has absolutely quantified, not relative, but absolute, about 200, almost 300 compounds in serum and plasma, 175 in CSF, about 400 compounds in urine. Now, metabolon can claim to have identified, I know, 1,200 compounds. But those are not absolutely quantified. They're relative quantifications. In terms of proteomics, a number of proteins fully quantified is now probably a little over 73. It's closer to about 100. In CSF, there's been 130 proteins fully identified and quantified in urine 63. In transcriptomics, because every measure, whether it's RNA-seq or microarray, is relative, there has actually never been fully quantitative data reported on genes. So, proteomics, which is 25 years old, genomics, transcriptomics, which is 20, 25 years old, and metabolomics, which is usually considered to be about 10 years old, given its late start, it's actually done remarkably well in terms of quantification. But a lot more could be done, because we know that in serum and plasma, there's probably 20,000 metabolites. In CSF, there's also probably 10,000. In urine, there could be 50 to 100,000 metabolites. So, we're only getting partial coverage. So, a lot more could be done, but a lot already has been done towards quantification. If more people were doing quantification, I think we'd do a lot better. So, quantification is a bottleneck that's being cleared, but not entirely. And I think another one that's really important in moving the ball forward is moving metabolomics from the lab to the clinic, or from the lab to practice. It doesn't have to be medical practices. It could be environmental. It could be in the area of veterinary science. It could be the area of plant science. In the case of clinical studies, a lot of them have focused on biomarkers. And over the last 45 years, there've been more than 700,000 biomarker studies or papers that have been published. But of those, fewer than 250 have actually been approved for clinical use. And that includes protein biomarkers, genetic biomarkers, metabolite biomarkers. Interestingly, with proteomics, there hasn't been a protein biomarker approved using mass spec. Almost all of them use ELISA's. There have been five biomarker tests that were used using gene chips or transcriptomics. But one of the things I think a lot of people aren't aware of is that if you're under the age of 25, you've probably had a metabolomic test. And we brought this up earlier, which is remarkable, given that it just doesn't get any traction. People don't mention metabolomics as being sort of the universal test that everyone has had. And I think I also gave you some of these statistics as well, which is metabolomics, even though it's a late comer, is moving into the clinic. And it's more so in the area of clinical chemistry. So if you tally up the total number of metabolites that are measured, at least in Canadian labs, and then are reported, it's a little over 320. If you tally up the total number of genes that are approved for genetic testing in Canada, it's about 130. If you look at the total number of protein tests that use ELISA, it's 108 that are approved in Canada. I've mentioned the gene chips, five. And then there was one proteomic test that was approved, but then it was moved to an ELISA. So it's back down to zero. So metabolomics is moving into the clinic because in fact, a lot of these tests are relatively quantitative. And that's been what's held back transcriptomics. It's what's held back proteomics. In ELISA's, the assays are relatively quantitative. And in genomics, it doesn't necessarily have to be quantification. It's simply say the mutation is an A to a T or a G to a C. And so it's important or not important. But that too, you could say is reasonably quantitative. But given the relatively small investment that's gone on to metabolomics, which is actually less than 1% of the investment that's gone into genomics, and less than 5% to the investment that's gone into proteomics, metabolomics has actually done pretty well. So not only for a small amount of money, but for a relatively short period of time, it's actually made those translation to clinical practice quite well. If we're looking at biomarkers, you can sort of compare how metabolomics does to some of the other methods, but also just in terms of certain conditions to explore. So you guys were working a little bit today with data that we'd had from preeclampsia, looking at mothers at three months into their pregnancy, and then looking to see if we could predict which mothers would actually have preeclampsia later on. And using metabolomics, we saw we were able to get an area under the curve of about 0.94 to 0.96. So that's an example of preeclampsia. You can also do the same for late preeclampsia, looking at serum metabolites. You can get areas under the curve, rock curves for diagnosing heart defects, gain well above 90%. You can even do sort of genetic testing for an infant's three months into gestation to identify which ones have trisomy 21 or trisomy 18. So this is typically before you can do amniocentesis, and this is just simply looking at metabolites. We also had an example where you guys were looking at catexia of people with colon cancer and lung cancer, and when you use that data, it's also possible to get rock curves that are above 80, 85% in terms of predicting which individuals will develop catexia. So if you can predict when people might get a disease, then often you can prevent it. So in the case of preeclampsia, a very simple preventative treatment is to use aspirin. In the case of catexia, changes in terms of dietary consumption of certain fatty acids and certain amino acids also seems to slow it down or prevent it. In terms of diagnosing diseases, metabolomics also is a very good tool. These are some examples again from the team at Group in Edmonton where we've looked at using urine samples to diagnose kidney rejection. So currently, if you're monitoring someone who's had a kidney transplant, you actually have to stick a giant needle in their back and take a chunk of the kidney out to analyze it through histology. So it's pretty painful and it happens a lot, especially right after treatment or if there's been a difficult bout of certain disease. So if you could just have people have a urine sample and determine whether there's a rejection happening, then again, it's easy to diagnose and actually easy to treat. Obviously, urine samples are a lot less painful than a giant needle in your back. Works well for pediatric cases. Also seems to work well with heart failure. There seems to be some metabolic tendencies that have been seen not only by our group but other groups in Australia to diagnose chronic fatigue syndrome, which is very hard to identify. Eosinophilic esophagitis, another one that seems to be diagnosing quite well. And one that we, I guess, are quite proud of, this is involving operations with a company in Edmonton called Botanbolomics Technologies Incorporated. They developed an NMR assay to take urine samples and to distinguish people who have colonic polyps from those that don't. So polyps are the precursors to colon cancer. And currently the best way to characterize whether someone has colon cancer or polyps is to have a colonoscopy. And if you've ever known anyone who's had one or if you've had one yourself, they are rather painful and they're also quite elaborate. So you're basically out of commission for about two days. You have to change your diet before the colonoscopy. You also have to recover for about a day or more after the colonoscopy. And if all gastroenterologists have their way, everyone would have a colonoscopy every two weeks sort of thing. So obviously it's too expensive to do that. Obviously, I mean, the typical colonoscopy is about five or $600 in Canada and the US, it's maybe a thousand. It's a money maker, but that's not exactly how health systems want to work. If you can have just a simple urine test that could pick up the polyps or at least screen the people who seem to likely have the polyps, then it'd be a whole lot cheaper. So we worked with them to convert the urine test which used 12 metabolites to a mass spectrometry-based test that uses three metabolites. That was successfully done. And because it's using multiple metabolites, we can adjust the sensitivity and specificity of the test. So it's now been picked up by several companies in the US which are now using it to screen patients for colonoscopies or not. So this is one of the first examples of a multi-metabolite marker moving into the clinic. Now, those are some trends. We've talked about automation, expanding metabolite coverage, making metabolomics more portable, efforts to quantify moving metabolomics and allow them to clinic. So some of these things are starting to happen. They're trying to give you some examples, but more needs to happen. And I think another area where more needs to happen is trying to get metabolomics into drug discovery and development. The history of metabolomics is that it was first used in drug discovery and development. It started in Imperial College and it was a pitch to drug companies. But relatively quickly, they decided to abandon metabolomics. It was called metabolomics back then, partly because it wasn't quantitative, partly because it wasn't portable, partly because it didn't cover things and partly because it wasn't automated. So now that we're getting closer to automation, expanding the coverage, making things a little cheaper or more portable, but we're quantifying. And there are clear examples of where metabolomics has now moved from the lab and the clinic. I think it's time for the drug companies to start thinking seriously about metabolomics. So if you look at drug discovery and drug development, the cost and time to take a drug from concept to final product is about 10 to 12 years and anywhere from 800 million to $1.2 billion. Interestingly, there are different phases where drug technologies are needed. So the first phase is discovery, typically chemistry's involved. Discovery in phase one analyses is often worked with genomics and proteomics. But in fact, metabolomics is needed just about every phase of the drug development process. And this is illustrated here. So in many cases, people are finding drugs. Through metabolomics, they're finding drug targets. And in some cases, they're finding natural products that seem to be effective in fighting or in some cases causing disease. So once you've got a drug lead identified to metabolomics, then you can also use some of the automation that's been developed to identify toxicity for the drug. So that's can be done on mice and rats. And then there can also metabolomics can be used to assess efficacy biomarkers. So you think the drug is working, well, let's see if it actually changes things. And so you can start looking and using metabolomics to find these efficacy markers. Then you can also use metabolomics to assess preclinical safety or to look at toxicology, particularly as you move from animals to humans. And then as you move to later phases in the drug development, you'll look at clinical safety and clinical efficacy markers. So an example of this would be here where generally when you do clinical trials in phase two and phase three, you recruit a bunch of people and you tell them, take these drugs, but do not take them with alcohol. And everyone will come and say, yeah, I didn't take any alcohol and I've taken the drugs. And then they'll analyze people and they say, well, obviously you didn't get any better or there isn't any response or you're sicker now. And they're stumped. But if they could actually analyze people's blood or urine and to look for things, you could actually see in this case, this person, even though they said they didn't take alcohol, they did. And so this is a way of monitoring compliance for clinical trial patients. Then it's often a case that just one or two outliers in a clinical trial can destroy a billion dollar drug effort. You can also look at individuals in these clinical trials to figure out whether people are slow metabolizers or fast metabolizers. Some people will take drugs and will metabolize them very, very quickly. And in fact, it seems like the drug has no effect. I am a rapid metabolizer of caffeine. So if I try and drink a lot of Coke and coffee right before bed, I will promptly fall asleep. Our other people, I imagine, are slow metabolizers and in fact, a cup of coffee even at four in the afternoon will keep them up all night. So this is something that differs with individuals and it's very much a function of their physiology as well as their genetics. But you can't predict purely from genetics and you can't predict purely from physiology. The best way to see this is through metabolomics. In terms of traditional drug discovery, there's a model that's been developed and used probably the last 10 or 15 years. A lot of techniques are based on doing large scale studies sometimes with genome-wide association studies or GWAS. The idea was to hopefully identify genetic variants that were seeming to cause certain disorders. In many cases, only 10% of disorders or disease actually have a genetic basis or a clear one. So typically the GWAS studies are only leading to about a one in five or one in 10 success rates. They're big, they involve tens of thousands of people so they cost millions of dollars and they often take multiple years. Once they found potentially target gene, then that gene has to be drug-able. In some cases the gene is not drug-able and even if the gene is drug-able then the protein target is not amenable for drug-ing. So usually they lose about half to the drug-ability issue or half to the protein issue. So once they finally isolated the protein target that they want to drug, then they start doing high throughput screening. And the typical success rate for that is about one in five of the protein targets they put into their system leads to potentially high nanomolar affinity drug. One in five success, that screening process can take anywhere from one to five years. But then to take that drug lead and actually try and get it into phase one, preclinical phase two trials, that's where it costs a billion dollars, that's where it costs or takes 10 to 15 years and that's where the actual ultimate success is only one in 500. Once the drug is formally approved, there's about a one, well maybe it's not quite 50% but a reasonably high probability that in fact the drug will be taken off the market because it has or proves to be ineffective or too toxic. So even approved drugs are regularly pulled. So if you add the one in five, one in two or multiply one in two, one in two, one in five, one in 500, one in two, you end up with essentially from a beginning GWAS study to the point where you have a final product of 0.001% of getting a drug to the market. It's gonna take 20 years and cost more than a billion dollars. That's largely why many drug companies have stopped their drug research. It's also why really all we're seeing now are copycat drugs coming from pharma companies or why we're seeing a lot of drug companies merging with each other because they can't sustain this. On the other hand, you could think of trying to use metabolite-based drug discovery. So metabolite-wide association studies, MWAS, actually can be done with a much smaller cohort because metabolites are the canaries of the genome. They're more sensitive. So you can actually do a pretty decent metabolomic study for a couple of hundred thousand dollars. And in our case and I think in many other labs, people have found very good and very useful markers but also probable targets that suggest that if this metabolite is reduced or if this one is increased or if this protein is made so it could increase these levels of metabolite, there would be a positive response. You guys have seen some of how quick it is possible to analyze your data, your metabolomics data. You did it this afternoon. So whether it's a few hours to a few days, you've also done some pathway analyses which help identify some of the proteins or pathways you want to target. You can use some of the databases we've talked about to learn a little bit more about those pathways and processes. And in many cases, if it's a case that there's a shortage of a metabolite or in other cases there's an excess of a metabolite, it's possible to simply substitute for that compound or to find certain types of proteins or enzymes that may modify that metabolite level. If you're wanting to see whether this is actually making a difference, then you can do metabolomics to monitor. So in these cases, it's possible to go from 15 to 20 years and a billion dollars to do something where you've got a possible or probable compound that's maybe even being used in the clinic in less than a couple of years and having excess sex rates of about 15%. And as I said, we've done this with colonic polyps and it's also suggested some other routes for treating a couple of disorders that we've been working on. And there's been some really interesting examples elsewhere, not just in our labs, but in Stan Hazen's group. So this is an interesting story. And they screwed out of the Cleveland Clinic with work that they were looking at with people who exhibited heart disease and atherosclerosis. And it's generally well known that people who have atherosclerosis typically have had high fat diets. They've all their lives, typically they're obese, typically they have a record of consuming fatty foods, lots of cheese, egg, and meat. But not everyone develops that. And there are actually large populations that have very fatty diets but have very low levels of cardiovascular disease. And in particular, over the last 30 or 40 years, people have been very interested in the incidence of heart disease in France and Italy, Spain, and other countries where consumption of eggs, milk, cheese is very high. And total fat consumption is much higher than in many other countries, but cardiovascular disease is very low. So why is it that some people are able to avoid heart disease or atherosclerosis and others, particularly in North America, seem to have high levels of cardiovascular disease? So what they found was that with people with atherosclerosis, they had high levels of trimethylamine oxide in their blood. Now TMAO is also something that you'll find in urine at fairly high levels if you eat fish. So it's a marker of fish. But normally it doesn't persist for long in the blood. And so they found these persistently high levels of TMAO and about 2011 was when they first reported it and they've reproduced this in multiple studies. We've also seen it in others. And then they tried to go back and look at these pathways and to try and understand it. So here's the lead, what's it meaning? So they traced it back and said, well, trimethylamine actually comes back from trimethylamine, trimethylamine oxide. So trimethylamine is a precursor for trimethylamine oxide. Trimethylamine oxide, TMAO is produced in the liver but trimethylamine is produced in the gut. Where does trimethylamine come from? Well, it generally comes from choline or other tertiary amines, which are generally produced from fatty foods, phosphatidylcholine. So here you've got an interesting link between what you eat, what's in your gut, which converts the phosphatidylcholine to choline to trimethylamine, and then what your liver does. So they started out this pathway relatively quickly, just like you guys would do from keg or HMDB or SNPDB. And then they said, okay, where should our drug target be? Well, should we target the liver so it doesn't convert TMA to TMAO? Or should we target the microflora so it doesn't convert choline to TMA? Or should we target people's diets and just tell them to stop eating fats? So the fat drug didn't work, and then they tried the liver drug and that proved to be too toxic. And then they said, well, let's try and look at the bacteria and see if we can change that. So in 2012, the other group started characterizing a particular enzyme that's found only in sulfate-reducing bacteria. And these bacteria are not in everyone, although they tend to be relatively high abundance in people in North America. And the enzyme that's produced by these bacteria is called choline TMA liase. So it converts choline into trimethylamine. So if you have these bacteria in your gut, you will tend to have high levels of TMA. If you don't have these bacteria in your gut, you will not have high levels of TMA. So then they said, well, now we've got our drug target. Can we start screening it to see if we can find compounds that stop TMA liase? So they started dumping all kinds of different things on them, and one of the things they found that really worked very well at stopping TMA liase is something called 3-3-dimethylbutanol. So these are the structures. So there's choline above and there's 3-3-dimethylbutanol. It looks a lot like choline. So in fact, it's a very good inhibitor. So where do you find 3-3-dimethylbutanol? Find a lot of it in extra virgin olive oil, in grapes and in red wine, which are sort of the characteristic features of the Mediterranean diet, which is what you eat a lot of when you're in Italy and France and Spain. And so it seems to have a really interesting effect or potential where A, here's metabolomics, finding a marker, B, identifying a pathway, C, identifying a drug target, D, identifying a potential inhibitor, and then finally pointing out to a rationale why certain populations have relatively low heart disease and why other populations have relatively high levels. So with that, I think I'll wrap up here, but what I've tried to identify are some of the bottlenecks that are limiting metabolomics, the lack of automation, relatively limited coverage, the fact that it's expensive to get into, the fact that historically it's been non-quantitative, the fact that it hasn't really moved from the clinic or at least we didn't know it was moving from the lab to the clinic, and then the potential or the limit applications metabolomics to drug development and discovery. What I've then tried to do I think is then identify areas where in fact it is happening, where you are seeing automated metabolomics, where you are seeing expanded coverage, where you are seeing metabolomics becoming more portable or quantifiable, and where there are plenty of examples of metabolomics going from the lab to the clinic, and where there are nice examples of how metabolomics is helping with drug discovery. And so what I'm hoping is that you could be messengers that in fact metabolomics is not sort of this obscure science that's leading nowhere. And in many respects, in fact, it's probably more automatic than many other omics methods. It's moving more rapidly, expanding more rapidly, that it is the most quantitative form of omics technology, that it is already successfully translated to the clinic over and over and over again, more so than any other omics field, and that it is happening I think the significant impact in the field of drug discovery and development. So be happy that you're actually in the field of metabolomics or that you're interested in. I think it's a burgeoning field and it could take you a long ways if you want to stick with it. So with that, I think we'll wrap up the course and I'll thank everyone for listening in, attending and working really, really hard.