Establishing causality in microbiome studies
Good, afternoon. My. Name is Kumar and Ramamurthy and it is my pleasure to, introduce on, behalf of the lamda, lunch bacteriology. Interest, group professor. Rob Knight from the University of California, San Diego, so. Rob started a scientific, career when he was about nine years old when his parents were postdocs, at the NIH, he. Did his undergraduate, at the University of. Otago in New Zealand shortly, there afterwards and received. His PhD from Princeton University. He. Was a postdoctoral fellow then professor, at the University, of Colorado Boulder for. Many years before recently, assuming, a professorship. At the at UCSD, where he has been for the last three years so. Rob's lab has been at the forefront of, employing and developing, laboratory. And computational, approaches, to studying complex. Microbial, communities, in the environment, in animals, and in humans his, lab has studied microbial, communities and pre agricultural, soils oil, spills. Decomposing. Corpses, and human. Built indoor environments, such as hospitals, offices, and homes, so. The last several years have given rise to the notion the bacterial, cells in the human body outnumber. Our own cells suggesting that human health is intimately, linked to the quality of microbes that we carry in and on us and. Rob's lab has been detailing, how these communities differ by anatomical, location, and change over time as humans mature so. To aid in the massive data collection, efforts that these projects require Rob. And his colleagues have cleverly harnessed, the, power of crowdsourcing to, citizen scientists, he. Is the co-founder of the earth microbiome, project that. Aims to construct, the microbial, map of the planet that describes the genomes protein, space and local meta and metabolic models, for different locations around, the world a part. Of this project is the American gut project, which may be the largest, crowdfunded. Citizen. Science project, in, which the gut microbiome, of thousands of volunteers is analyzed, and made, available to researchers as. Further examples, of his educational, outreach Rob. Gave a TED talk in 2014. Which, later inspired a book written, for the general public and, he hosted and asked me anything session on reddit. If you don't know what that means. Your postback students can explain that to you later, he, has received many awards most, notably he's in each of my early career scientists, and he received the bullshit prize for, creative promise, in 2015, please, join, me in welcoming professor Rob, Knight. Great. Thanks. Thanks kumaradas I'd like to thank you in season and the Lambs, lunch group again for being, wonderful hosts on this visit and for, the for the lamination, to this sort, of series it's, a true, pleasure and true, privilege to be speaking here, so. So. What what, I'm going to talk about is, establishing. Causality in. Microbiome, studies which, is a very very, hot topic at the moment and. Essentially. What I'm going to do is take you through what, is the value of of. Determining. The microbiome, in the first place what. Can we do with high resolution tracking. Of individuals, and then, using obesity, as the case study go, through the pathway from, first looking at an association. All the way through all, the way through establishing. Causality and, what. Are some ways that we could do to accelerate that process, I. Understand. This will see Emmy I don't, think any of these disclosures, are particularly relevant, to today's talk unless your patient, isn't Oilwell or something. But. There. They are nonetheless. So. I'd, like to begin by. I'd. Like to begin by just challenging. A challenging you, know your to think about, what did you see when, you when you looked in the mirror this morning I, saw. An organism, that's just 43%, human and not just because I flew in from the west coast and hadn't had my damn coffee yet but. When, we think about what makes up our bodies each. Of us at, the cellular level has, about 30 trillion human cells according, to the latest estimates and about, 39, trend in microbial, cells so, the error bars on these numbers have shifted somewhat and. Have gotten much narrower but these are about the Centers of the estimates, and this was where that 43 percent human, number comes from, now. You might be thinking well wait, a minute. About the count of different cells shouldn't, we be thinking about this the DNA level since we're now in the 21st, century so let's think about that for a moment each, of us has about 20,000, human genes again, depending on what you want to count exactly as a gene but.
Amazingly, We're fans through the human microbiome project and, other large-scale endeavors, but, the science about microbial, gene catalog is about two to, 20 million, microbial. Genes dwarfing. Our human genomes and by, that measure we're at best one percent human and there's, a lot of excitement at the moment about systems biology even, systems, medicine but, I'd argue it's difficult to do systems anything if you're neglecting 99%. Of that system and perhaps. What's, most exciting is, the microbial, genes that we have performing. The majority, of unique biochemical reactions. Associated, with our bodies and the part of the genome that we can change very easily compared. To changing out human genome. Now. That's, not to say that the human genome wasn't, a remarkable, achievement right, so the, cost of the human genome was estimated, at about 3.8. Billion dollars, and in. 2013, when Obama was launching I was, launching the brain initiative he, estimated that even at that point the return on investment of the human genome project had, been about 140 to one so, this, kind of large-scale scientific, investment. By NIH, can, can, pay enormous dividends, and given. That the microbiome is component of the genome that we can change we, can expect that to be even true of the human microbiome, project so. The human microbiome project again. Was one, of these large-scale NIH. NIH. Initiatives. So. I was, involved in this and in a number of different capacities and one, of the most important, components, of the project was the what, was the baseline. Cohort, study characterizing. The the, healthy human microbiome. And so, the results from us came out and in nature in 2012, characterizing. 242. People had up to 18 sites on the body and up, to three time points and, so we collected a combination, of 16s, ribosomal RNA. Profiling, to read out the name tags of all the bacteria as it were and then, shotgun metagenomics. To get a more. Complete gene catalog and so, we collected four and a half trillion, bases, in this project so, if 1,500, human genome equivalents, which for the time was an absolutely, massive microbiome, data set and. One thing that was one, thing that was both surprising, and gratifying about, this was how rapidly, the, results leaped from the pages of the specialty, scientific, journals out into, the out into the general public consciousness so, both science and nature did, cover stories on the microbiome, in June of 2012 and just, a few weeks later it, was on the cover of Scientific American and just, a couple of weeks after that on the cover of The Economist, so, the stuff was not only a big science but also increasingly.
Big Business. Now. The, great thing about the human microbiome project is, we got an unprecedented, amount, of DNA sequence, data telling. Us about the microbiome, but. That was also the terrible, thing about the project in many ways and just, to illustrate this problem I'm showing you the first file of data from the human microbiome project I'm. Telling you a bit of a lie this is actually the first 0.1%. Of this file and there are there are 17 other, 17,000. Additional, files just like it vs, muscles really an issue right because what we're doing is fundamentally an ecology, project we're, trying to figure out who lives where in the environment, how frequent, are they what. Are they interacting, with what are they doing them there but, from this kind of information I'm guessing that you probably can't even tell that's an oral sample right lateral sequencing. To the sequence, signatures, that would allow me to determine loss and. This was increasingly a problem right now because, both through citizen, science projects like American. Gut that Kimura mentioned in the introduction but, I'll say a bit more about later and also various companies the general public now has a lot of opportunities, to get their microbiome, sequenced and if, you're a clinician this is your worst nightmare right your patients going to come home with, a huge grin on their face and they're going to say hey doc I have some great news for you which is that I had my microbiome, done and now I have this list of a thousand, species that they found with my gut and with all this data surely, you can tell me what's wrong with me right then let's well beautiful minutes we have together and. I mean what are you going to do refer them to your colleagues in psychiatry, for being crazy enough to think that you could do that but. The challenge that we face is to make it not crazy anymore and, to figure, out how to integrate all of that information whether.
It's From a single profile or whether some a same person over time and get something useful and actionable, about it and that's what I'm going to show you in the rest of the talk. So. Anyway the solution, to this big data problem, was, largely provided by Kathy lazybone who I actually retreated, to my lab right, when I started, as a faculty member at Boulder to. Work on type 3 secretion, which was the main thing that my lab was working on at the time and about, a year into the project Kathy, said yeah, you know I'm just not that into type 3 secretion, could I do something that's more interesting with sequence analysis, in the wire mountain trees and so on and I told her well you. Know Kathy versus stuff and phylogenetic, diversity that's, that's kind of interesting but I have to warn you that even even, nan pace is, a National. Academy member, and certified, MacArthur, Genius doesn't. Think that it's really going. To get a lot of climbing whereas type 3 secretion is a really hot area and, I'd hate for you to be working in some sort of scientific backwater, later on if. The type III secretion projects. On the cover of science but, she decided that she wanted to persist. In this project it turned out that she did okay for herself she's, now at the University, of Colorado at Denver which, is the third most highly cited faculty, member there and the. Technique that she introduced, has, now been cited about has. Now been cited over 5,000, times so. I think it's fair, to say that she did a cape for herself which is why it's really important. To listen to your grad students and enable. Them to follow what they really want to do but. The technique that she turned to in order to grapple, with all the state her in some ways is a very old technique taking away from Darwin's, book essentially. Essentially, literally, so, if, you haven't read has worked on the Origin of Species recently, or perhaps ever the, first edition was often considered to be a fairly dense read and that's, true it has 502, pages and only, one figure so people who read it for the pictures are often disappointed. But. But I'd argue that in a way you shouldn't be because one figure that Darwin thought was important, enough to have engraving made our version to stick in his book back, in 1859, was this one the, idea of, the phylogenetic, tree being, a way to get a handle on life's diversity, and Kathy's, insight was to realize that that was just as she the level of whole microbio, microbial. Communities, as, it is for individual, organisms, and try to understand, their complexity, and. So. Back in 2005, we published this, called, Yuna frack exploiting. Evolution, to compare microbial, communities, and, the intuition behind this very powerful but very simple at. All is that if you have two unrelated, communities, so. So. Basically, in these phylogenetic, trees the tips represent, modern sequences, the interior nodes represent, hypothetical, common ancestors, the, idea is if you have two completely. Unrelated communities. You, can never move from the red community, into the blue community, and survive and reproduce and add branch. Lengths that represents, a molecular change so.
All The branches will lead only to the red community, or only, to the blue never to both so, there is different as they can be all the branches are unique we define that distance as one in, contrast. If you have two identical communities, that they're completely sampled, what, you find in the red defined in the blue so, all the branches are purple leading to both communities, none, of it's unique and we define that distance to zero now, for any actual two communities, you'll have some unique branches, and some, shared branches and so some distance between, zero and one based on the phylogenetic tree. So. What's so great you're wondering well. What's so great about that as we can combine it with other Darwin's insight but there is one tree of life though connects everything and so we can take that one tree in paint as many communities, as we want comparing, red to yellow red. To blue yellow to blue without metric, I just showed you summarize. The results in a distance matrix and then use techniques I principal coordinates analysis. To boil all that data down into. A map and, so, when we apply that to the human microbiome project data, set we, got this map here and so we did this using chime that took. It we developed led by Greek paper a so he's doing doing. Doing, doing, a, sabbatical rather, at. NCI at. The moment, to. Take, all the data right off the sequencer, and then process it into, these. Kinds of analogical results. So. You're probably thinking to yourself well I don't, really understand this map any more than I understood the sequences a few slides ago but, just bear with me for a moment while we're seeing on here is that each point represents all, the complexity, of a microbiome, readout from its DNA and we, place we. Were and we projected just down onto one point where we place two points together if they're more similar in terms of their evolutionary, history and we, place them further up if they're more dissimilar in terms of that evolutionary, history. Now. That's also healthy subjects cohort so there's no clinical variables that could make a difference to the microbiome but. You can think of all the things other, space in which these microbiome, samples differ like. How old the subject is what, time of day was something was done whether they're male or female, what. They weigh what, ethnicity they, are all, kinds of different factors like that and, what I'm going to do is I'm get a color by a factor that makes it immediately clear, the main thing that's going on in the system and what, matters is of the different parts of the body are completely, different from one another almost. Defining, different continents, in those matter so. What's, interesting about that has approached the human microbiome project most studies have studied. Different cohorts, at. Different sites of the body with different methods and so understanding, what was truly a difference in the microbiome, this is what was the difference in the cohort or a difference. In the technical, approach was very difficult and thus provided, a unifying. Framework for, understanding these differences. So. Just, to illustrate the point of how distinct, these different body sites are I'm going to highlight the mouth from the gut on the first person and the human microbiome project and so, when I do that you can see that they're in very different places on the map almost, like different continents, but, it's difficult to get a sense of scale and we didn't get that until, we did the earth microbiome, project the, first main paper from which was just published in Nature a couple of weeks ago, where. What we could do is we could we could go out to all of these different locations, on the planet that we obtained from crowdsourcing from the scientific community and we could ask what two sites on earth and just as different from one another in their microbiome, as the mouth from the gutter this one individual, in the HMP and, if you think of your mouth was being kind of like a coral, reef maybe, it's this complex mineralized, structure covered with biofilms, maybe your data spoke to you about from time to time, the, amazing factors, of your mouth is as far from your gut intensive, its microbiome, as, the water in this reef is from the - and the sperry right, there essentially non-overlapping, microbial.
Communities, And it's, amazing to think that a few feet along the length of your body can, make as much of a difference your micro biomes as thousands, of miles across the Earth's surface. So. Now you might be wondering well, so you have this map of what the microbiome, looks like in healthy people, can we actually use this kind of map to track disease states and so, the best example we have of us is what we did with our excretes. And Mike's Sadowski at the University, of Minnesota looking. At see different patient and, so the stars that you see up here those are micro biomes from people with with, castor DMC a cluster. Difficile. Associated, disease and, although their fecal samples you can see that they look nothing like the fecal samples of a healthy person and so, what's going to happen is the four of these patients again again I get a fecal transplant from. The same donor and. So and. So if, you're wondering what a fecal transplant is this. Is Bill Sanborn, his that terribly gastroenterology. About to deliver one and he's currently doing about three a week using using. Hospital grade stool from, a nonprofit called, open biome because, remember the FDA does regulate feces, as a drug although it's difficult to prove that you made it under GMP, conditions. So. So. Anyways so four or four of these patients again to get a fecal transplant from this one donor who as you saw was in the healthy region and this was a really beautiful example, of how we can take your data set and integrate. It into the reference frame provided, by the human microbiome project so. You can often interpret, your own data much better in the context of these larger reference frames and, so what's going to happen is that they're going to get this fecal transplant, and the question is are you even going to see anything at the whole microbiome, level given, the complexity, of the ecosystem, that you're, trying to modify and if, you do see anything how complete is the progression, and how fast, is that isn't going to be fast or slow, isn't. Going to be a straight line or is they going to be some sort of detour along the way and, and. Each frame in this animation which was put together by Antonio dashiki, two very talented computer, scientists, in my lab each. Frame limbless is going to be one day in the lives of these patients and. What you can see is that essentially immediately, all four of them move right up to the healthy state and during, those first two or three days all their clinical symptoms in God and. They stay in remission during, the few months of follow-up where you can see that their microbiomes, are moving around a little bit but, still resembling, less healthy state and, this provides a beautiful example, of how you can spot a problem with the microbiome, and. Then see the microbiome, coming. Back into a healthy configuration, and then track during therapy is he going to stay in that healthy, raishin or not and, so, the challenge that's facing us as a field right now is for, many of the diseases, that have been linked to the microbiome and as you know there's been a huge number of these different diseases that, have come up in different Association, studies, everything. From cardiovascular disease. To colon. Cancer and more, recently liver cancer, to, mature, things you might really not have expected like rheumatoid arthritis and, autism, how, can we identify a problem with the microbiome, and then got, it back into shape whether, it's something as Extreme as fecal transplant or some other way of reshaping the microbiome, such as with antibiotics, or with diet or other things that affect it and so, a really important challenge is, to find the good in the bad places on this map and to look at how people move around in this. So. What, I'm going to show you now is, a little bit about the value of doing high-resolution, tracking, of the microbiome, over time for individuals, and right, now this is expensive, enough that we can only do it with 16s, ribosomal RNA, profiling, which is relatively cheap but, we're increasingly, developing, protocols that allow us to do it for shotgun metagenomics. Which is dropping precipitously in, cost at the moment and, then ultimately what, things like metabolomics with meta proteomics, and so on to get an assessment of function and. And. So and. So one. Consequence, of the HMP was that a lot of people were very excited about getting, their own microbiome, sequenced and so.
Over Thanksgiving, in 2012, Thanksgiving. Being a time and a lot of Americans are thinking about they get for some reason, we. We. Decided, that it would be great to launch a project that used crowdsourcing, and crowdfunding and, citizen. Science to let anyone who wanted to participate in microbiome, research so. The marginal cost to processing, each sample is supported. By a contribution, from the individual, so, that anyone who's interested can claim a pin for themselves unless matter of. Course it turns out that not everyone wants, to know what's in there so these are middle schoolers tearing our lab and learning, that we're going to use lasers and robots to look at their DNA and the bacteria or in their poop all of which is literally true by the way you, can see a certain amount of response heterogeneity, there we don't yet know if that's linked to the microbiome but, sooner or later we'll have the sample size to find out, but. In, all seriousness this, is now one of the larger citizen, science projects that exists so we've we've sequenced over 10,000, samples, and deposited. All the data free immediately, as it goes through QC after it comes off a sequencer, into. EBI so that anyone can. Access it so the idea is that any student any educator any clinician and a researcher, should. Just be able to go to this free open resource with, a free open, processing, pipeline and understand, what, a large number of human microbiome looks like and, what additional kinds, of microbiomes, might be out there to discover. So. So, shortly after I moved to UC. San Diego I met Larry Smarr is, the director of Cal 82, and what we're doing here is looking at the human microbiome project data, set on a 64-bit, in pixel, display wall which is a wall, bigger than the screen that we're looking at but, with a very high resolution display, for. Doing advanced computer visualizations, and, Larry. Said to me you. Know Rob I have a very interesting microbiome, myself, because I'm an IBD patient, how do I get myself onto that kind of map. So. I moved to UCSD a little, under three years ago now and as, many of you know when you move from one institution to another you, expect to get a certain amount of crap from your colleagues and in my case has literally been tree service barks showed. Up in my lab and, as. I mentioned Larry is very interesting, because he's been tracking things about his own body for a long time starting. With his weight so, one dimension, which has been tracking for 17, years and then adding dozens of additional dimensions, will store tests with blood tests then thousands, of dimensions, with snip profiling, and more recently millions of dimensions, with microbiome, profiling, and so I've.
Been Cantly had been collecting his own stool samples for microbiome, profiling, for a while and I, had produced this plot of them and this. Is just calories a color that the family level said the different colors the different families, of bacteria, and don't, worry if you don't see any patterns in there because neither did Larry and I think quite a lot of computer, computational. Effort and personnel effort went, into producing Witten, to producing this diagram, but. When we breathe when we reprocess it with with chime and with you know frak what, we see is this much clearer pattern where, you see this very very clear split between the blue region and the red region this is just rotating the frame around so you can see it in 3d, what. You're going to see as he starts out from one corner of the blue moves, all the way to the blue crosses, over into the, and, then bounces around basically. At random and the red so. You're probably wondering, what does this have anything to do with Larry's actual. Health and so what we can do is we can match that up to his clinical information and what. We can see is this initial shift is caused by antibiotics, that, directional changes, he's moving through the blue region where he's having frequent IBD symptoms from losing weight as. As. As. As all, of that blue region of the graph and then, he changes his medications, you have that relatively rapid transition, between the blue and the red and. Then when he's bouncing around basically, at random in that red region his weight goes back up to a healthy setpoint and, then he stays there and has IBD symptoms are basically gone and, so if it had this high-resolution, view of configuration, space for Larry we, would have been able to tell him as soon as you're in this red region you're. Going to equilibrator, although it'll take a while and we'll, be able to predict where you get where, you get wound up even in advance of your symptoms so, the, potential for doing this kind of thing with high resolution mapping, of an individual's configuration, space and how, it relates to their phenotype has a lot of very exciting potential.
So. You might you might be thinking well you, know that's all very nice for Larry but, can. We actually do this from a larger cohort so, this, is a this, is from what we did with Janet Johnson's group that came out and, came. Out in nature biology. Nature. Microbiology, earlier this year that should be not, 2014, but 2017, and what we're looking at is a larger cohort of IBD. Patients with, different, with. Different phenotypes. And in. Particular the yellow points or ileal Crohn's patients, with surgery, the, red or without without, surgery, and then the green or healthy controls and when I start the trajectory is going what you can see is that the lines that are different colors have. Very different behavior, so you can see a whole lot of movement in the red lines and in the yellow lines which are the ileal Crohn's patients, and what, you can see is that the healthy controls are moving a lot less but, also their range of motion is very restricted, to just one corner of the plot as, we track these individuals over, time and so, what we're seeing is that the healthy micro biomes are the most stable and that the Crohn's patients, especially the ileal Crohn's patients, with resection, are the least stable. So, Yoshiki. Figured, a way of. Modeling us with a healthy play and so basically fitting, a play, in mathematically. Through the healthy individuals and so the top row shows a cartoon, of what's going on the. Bottom line a bottom row shows the actual data where, we measure the distance between each point and its closest, approach to that healthy plane and what, was amazing is when we did that for these different indications we, actually got better classification, of disease state than we could from a calprotectin essay, on the same samples, we calprotectin was currently the best essay from Stahl for. Reading out state relation, to IBD, so, essentially. From looking at how the microbiome, moves we, can get a lot of information that we can't get just from a single sample. So. So. So after so, after those events Larry. Ultimately. Relapsed, into into. IBD and was. Having what, was having increasing. Problems so he went in for a colonoscopy and. What, we're doing here is we're just tracking the diversity, of was microbiome, worth the time well, what you can see is that the colonoscopy has, decreased, his microbial, diversity dramatically. Although, he won't sup over, he. Winds up over a couple of weeks coming. Back to the diversity that he started off at and. And. I'm seeing, this you might be wondering well he comes back to the same amount of diversity but. Does he come back to the scene does, he come back to the same state of the microbiome, and, again when we located in Pico a space what you can see is. With. What the colonoscopy he, basically wanders. Around out here so it's a fairly big change but, then it comes right back to the starting point and. One. Thing we see although I won't go into this too much is that although there's, some change in the overall configuration, of the microbiome, when, you look at it at the level of individual, tax what, you see is huge changes, like order of magnitude changes in, terms of increase or decrease in particular taxa, and it's a lot easier to see this with a lot of time points and convinced the convince, yourself that you know that they're real because, you see relatively smooth change from one time point to another then if you just had a couple of samples from this patient.
The. Other thing is at least taxonomy, plots become a lot more useful and a lot smoother if. You have a sampling interval that's frequent, enough that you can see just a little bit of change one time to one time point to another rather, than seeing everything being completely, different between each time point that you look at. So. So. Following the colonoscopy he, had surgery so basically, he had a resection of an 8 inch section of his large bowel and, what. You can see is that on surgery you have a much more dramatic decrease. In the microbiome, than you do after a microbiome. Diversity, than you do after colonoscopy. So, these, these phylogenetic. Diversity values, looking at the amount of the whole phylogenetic, tree that you cover and so it's going down from a value of about 22, to, a value of about 3 had an absolute scale so, as microbial, diversity is, really promising, and. Again, when we put this on this kind of microbiome, map what. You can see is. What. You can see is that initially, so. This was just taking you through the colonoscopies, so the, orange points are before the colonoscopy, he, has the colonoscopy in red which causes a small deviation that he recovers from and. Then he goes into then. The greenness before surgery so, you'll see him bouncing around there and. Then he has a surgery, which is a really dramatic shift, and. Then when he comes back from surgery he, winds. Up bouncing around in a new state but it's a completely different state from the state that he was in originally, and so. And, so again you might wonder well is this a big enough shift to actually, make him look like a different person and, to answer that question what we have to do is we have to integrate his data with, population, data from a whole lot of different people and, so when we do that with the American gut data frame which, is what you're seeing, which. Is what you're about to see here so that's the same trajectory combined. With hundreds of samples from the American gut project, what. You see emerging is the same pattern that you saw in the human microbiome, project with, the different body sites being different from one another and that's. About to come back now with the same coloring and, what. You can see now and has trajectory, is that the. Let's. See yes, that's just starting now so what you can see in this trajectory gave, us since before the colonoscopy, you're, going to see a small deviation in red with. The colonoscopy, that he comes back from and, then. After. That you get to see them go, through the pre surgery samples, and then, when you see him getting surgery which happens. Now. You, say that he's Travis the entire, space. Of fecal samples from people in American gut so, he's more different from himself just, that day after the surgery than, any two people in the American gut project out from each other and, then you can see that when he's come back he's more or less in the same configuration space, but, he still looks like a different person rather than the same person and so, this, is very this is, fascinating in terms of seeing, that a surgical resection can, make you like like a different person in terms of your microbiome and that that change is stable for a fairly long period. Okay, so the, stuff so, the stuff of tracking is pretty exciting, right because you can really see the effects of different procedures that you're doing but what, we really want to be able to do was not, just track the effect some different things that you can do to your microbiome but, establish, when the microbiome, is causal, and, so. What I'm going to do is I'm going to take you through a case study in, obesity which, I've been working with Jeff Gordon's lab and, a number of others on since.
Since, 2005. So, for 12 years now and this will give you an idea of how far we've come in those 12 years but, also give you some prospects, for how you can accelerate that, timeline a lot in other conditions. So. So, obesity was one of the first conditions, to be linked to the gut microbiome, and progress. And this has been pretty a pretty remarkable so, today for example I, can, tell you with 90% accuracy, whether you're lean or obese based. Solely on a sample of your gut that I processed, for microbiome, sequencing, so. On the one hand this was a cool trick on the other hand you, may be questioning, its its, clinical relevance, on the grounds that I bet you can tell which of these people is obese without, doing any DNA sequencing, at all right so. We don't think it has a lot of commercial potentials, a test for obesity but, on, the other hand if you try to do that exact same classification, task based, on human genes rather than based on microbial, genes you, can only classify someone, with lean or obese with 58 percent accuracy, sorry, 57, percent accuracy based on every snip that's, ever been linked to obesity by G wasps whereas, we can do it with 90% accuracy, from the microbial, genes and I'll tell you a little bit more in a few slides on how exactly we do that. So. So. Another so, it's a lot of. So it's a novice to staff establishing. Causality what. We have to do is we have to go into the internet. Abay otic mouse model and. Let. Me just. Yes. So this so, this is so. So. What you're seeing here is the is, the OB OB obese. Mouse so it's leptin Newton's it's, about three times the mass of a. Wild-type mouse and, then what you're seeing right next to it as its Lin Lin littermate and, and. So and so so Jeff Gordon was well it was initially very excited, about the possibility, of just looking at obesity and thus now and, thus now model and asking, is there a microbiome, difference and so, and, so this was really the first that, this was really the first paper back, in 2005, and PNAS, we. Were able to. Where. We were able to link. A phenotypic, state to the microbiome where, essentially what we were able to do was we were able to show with. We, were able to show that with Yuna frac what. We saw was mostly grouping, by. By the maternal status, of the mother so the mother had the biggest impact on the microbiome, of the offspring later on so this is just colored, by which mother or you had but, then when we looked at the ratios of the different kinds of bacteria that were there what. We saw was a substantial, shift in. The ratio to the major phyla, and bacteria, in the obese OB OB individuals, versus. A heterozygous, a wild-type were to mate, so. Rick followed this up a year, later with. With an observation that that, humans on a weight-loss diet over the course of the year were, able to change their microbiome, substantially. So over this period coming to the coming, to resemble the, microbiome, sub lean controls, and, what was especially exciting, about this is whether it was on a carbohydrate restricted. Diet or a fat restricted, diet the, more weight each individual, lost the, greater the change in the abundance of this particular, group of organisms. So. That started to suggest that there might be a relationship in, humans, we, followed that up in a twin study in, 2008. Comparing. Obese and lean twins for essentially what we were looking at was, if, you look at obese individuals, and lean individuals, and especially if you look at twins who are discordant, for obesity, do you see any systematic, difference between the lien or the obese individuals, and so in les paper we essentially establish that there was there, was a population, level difference, between, between, lean and obese individuals.
And, Then we started to look at the predicted value of. Voe. Bisa T so this, is the data I was telling you about the, this, particular graph is from Lisa 2012. Looking. At the area under the curve from. A model from, a predictive model trained, on snips for, human obesity where, we're, essentially what you get is friendly classifier, accuracy, whereas, Dan, nights he was a very. Talented grad student in my lab at the time now, a faculty member at the University, of Minnesota was. Able to show that if we if, we sweep over different clustering. Deaths for, the 16s, ribosomal RNA, and then look at the error rate for clinic for separating. Lean and obese in that, in. That cohort we. Were able to get 90% classifier. Accuracy, based on the microbiomes. As, opposed to based on the human genes and what's, particularly remarkable. About this is the raw data that that's based on look like this so, what you're looking at is two time points from each individual, you can see if you can see quite a lot of heterogeneity where, where, some individuals change a whole lot whereas, others stay basically, the same over the two time points you, can also see what the data are very noisy in terms of the lean and the obese individuals, having. A lot of similarities, of this whole phylum level and yet from that noisy data it cancels very good classifier that we can use to tell the difference between the lean and the obese individuals. However. The story got more complicated when we started looking at other cohorts, and what, we found is that for inflammatory bowel disease we were able to pull out the same biomarkers. Of. IBD from a lot of different cohorts, analyzed, separately even when those different cohorts have been studied with different methods but, when we try to do the same thing for obesity what, we found us that although we can build predictive models to separate lean from obese individuals, in a number of these different cohorts, because, the technical differences between different studies so, that's what you're seeing here is larger. Than the difference between lynard obese individuals, which, College which. Which are colored, by different symbols within each study it's, very difficult to integrate those studies with each other and the obesity signal, is much subtle than, the IBD signal or they were confident, that it exists, in the individual populations, we, do not see anything that bridges those different populations. So, it's a given that we don't see a consistent signal in different populations a, lot. Of people wondered well is that signal really there and to address that question what we need to do is we need to move into the init, of the nota biotic mouse model where essentially what you can do is you can raise the mice with no microbes whatsoever, of their own and then, inoculate them with microbes that you think you're going to cause the phenotypic difference, so.
When, We come back to. The ob/ob model, although this, was also done for diet induced obesity and as, I'll show you for a different use of the tlr5, mutants, what. We what, you can do is basically basically. Take the. Fecal pellets from the obese and from the lean mouse and then transplant, them into the germ-free mice that have no microbes, of their own and then ask what the result is and sure, enough if you transplant the fecal pellet from a fat Mouse what. You get is what. You get as a resulting, mouse that has substantially, more adiposity and substantially. Higher body mass than, if you instead transplant, the fecal pellet from the lead mouse. So. So. To give an example of us the second system that. We did the sin was the tlr5, over. The tlf 5 Newtons so what the tlr5, mutants like is. As. Toe light received 2 v which is a component of the innate immune system that normally recognizes, bacterial, flagellum, and the. Tlr5, mutants, they're. Considerably. Fatter and both males and females, they're, considerably, more adipose, they. Have larger fat pads higher, triglycerides. Higher. Cholesterol and, even high blood pressure and what. We saw and, say fascination, we the basis for this is primarily, that is, primarily behavior, so, what happens is the tlr5, knockout mice, eat, a lot more than other normal, masters and fascinatingly. That, behavior can actually be transplanted, into, a wild-type Mouse by transplanting the fecal pellets from a tlr5, mutant so, so, essentially what you have is you have a genetic change in the mouse that's, triggering, a change in microbiome, configuration. That you can then transplant, from one Mouse to another and the effectors behavioral, which we thought was really exciting. So so. Essentially, what this shows is that you can establish causality, in mice by transplanting, the fecal. Transyl. Transplantation, from, one Mouse to another so, then the question is does. That only matter in mice or doesn't matter in humans as well so. So, Vanessa or adhara an, immensely, talented grad student to Jeff's lab tried, the following experiment which was going back to the human twin cohort and asking, if you take if, you take a fecal sample from. An obese human individual, and you put that into a mouse what do you get and sure, enough you get a fat Mouse which you do not get if you instead transplant. The fecal community, from a lean individual. So. Then you might have the objection well how do you know it's really the microbiome, that's doing it because you're transplanting a lot of stuff when you transplant a fecal pellet and today, the sweet into this approach called personalized. Culture collections, we're essentially what you do is you take a fecal a fecal, sample and you, culture dozens, or hundreds of strains from that one fecal sample AZ nicly and. Separately from one another and then you mix them together and, then and, then you see if you get the same effect from, that mixture of pure back bacterial. Strains but, you got from transplanting, the whole fecal sample and that's how you can prove that it's, the bacteria that are doing it not, a virus not a fungus another small molecule metabolites not, only of the other stuff that you transplant, when you transplant the fecal community, and. What. Was amazing about this was the beautiful concordance, that we got between the the, culture collections, and, and. The primary fecal transplant, so, what. Those figures showing you is days post colonization. Versus, principle principle component, one of the microbial community, and what. You can see is that for the obese donor in blue the, uncultured. And the kolchin collection, converge. On exactly the same state over two weeks the, same is true for, the. Uncultured versus, the cultured, community. Of the lean donor and so you can see they go in very into very similar microbiome States you, can also see that you have essentially perfect correlation, in terms, of function in terms of the EC numbers and the uncultured community, versus culture and community from the same individual, and so. That's how we can prove that it was the bacteria in the microbiome, that we're doing it on. Top of that now that we have these things computer culture and this was a current phase of our program project grant unless. What. We can start to do is we can start to ask which individual, bacteria, out of that very complex mixture are, the bacteria that are making a phenotypic, difference so essentially we're now equipped to adeline particular microbes, to take out particular, microbes and to.
Try To understand, as a much deeper level who's, doing wash in the community, who can substitute for each other who individually has, a large effect against, any microbiome, background and so on. So. So, one additional aspect of the study that was really interesting whether it was the concept of microbiome, invasion. And. What. We were particularly interested, in is suppose, you have two micro biomes with different phenotypic States in the mouse, what. What happens when you put them together in, the same cage and. Let the microbiomes, battle it out so, what. You're seeing here is the result of one, of the most exciting microbiome. Cage battles, where, we took the where, essentially what we did is we took mice the receiving, the microbiome, from an obese person that. Would normally make them obese and then putting and putting them in a cage together with mice that received a 39, member community, that, was specifically, designed based, on the microbiomes of lean individuals, and. What. You can see and. So what you can see is that the, mice that we see the obesity gene community. Have about a 10 percent change in fat mass there's. No change in fat mass and the ones that receive the microbiome, from the lean individuals, and also. The 39 member community. Can, protect. Those. Mice from, being co-housed with, the obesogenic community. Where. Those mice were, those mice remain obese themselves, so, what's really exciting about this is not only can we change the physiological, state of a mouse using, the microbiome, but, we can actually design, a microbiome, that will protect that Mouse against, exposure from the obesity and community, and. This. Is particularly exciting, when you think about human applications because, when. You take mice that, are inoculated, with, the microbiome, of an obese individual, and those inoculated, with the microbiome, of a lead individual, and you carry house them what, happens is the mouse with the with. The obesogenic microbiome. Slims down but. We think the exact opposite, happens in humans where if you live with other, obese humans you're, much more likely to become biess then, other. Than for the other, than for. The obese, individual, to slim down in that context, so being, able to design a community, that changes weight would be really a big deal. Now. The, last thing I'm going to show you here, in, this section as is looking. At a population-based, study so you might be wondering well so, you can do this for one individual, where you do this huge culture collection approach can, you take the approach of basically doing an, epidemiological. Study and then, figuring. Out from that epidemiological. Study the microbes that are associated with the physiological state, and, then proof of those have a phenotypic, effect so. This is what we did with were three clay Andy Clark and inspector, in the twins UK cohort and, essentially. What we saw there as another cohorts, is that, the human genome has a disappointingly. Small effect on. The microbiome, and that monozygotic, twins are about as, different from one another as dizygotic twins, are and. That that effect is statistically, significant, when you look at hundreds of twin pairs but you have to look at hundreds of twin pairs to find it, however. That's of the whole microbiome, level and if, you instead look for individual, microbes that have a larger fate what, you can do is you can ask are there microbes that are highly correlated in abundance, in monozygotic. Twins, but, models correlated, and dizygotic twins, and so those are the heritable ones and one of the most heritable, microbes in, this cohort was from the Kristensen no AC ate so. So, one thing that was also interesting about about. The Christianson violation, Kristensen. Ohta, is that, they were particularly, correlated. With low BMI, in this cohort and so, so, what Ruth wondered was if you take this organism, that we isolate, that's, highly heritable so, it probably has a biological effect, and we isolate, it from lean people what, happens if you transplant that into a germ-free Mouse is receiving an obesogenic community. That would otherwise make it fat and sure, enough what happens is that the live Kristensen ella was able to slim it down considerably, and, so, again what, this means is that we can take a population-based. Study, we, can find individual, microbes, from that cohort that are associated with physiological. State and then we can use those to modulate, physiological. State in the mouse. So. One so wasn't really one really exciting another thing and especially about all the diversity, that we're finding in the microbiome, as a prospect that we might be able to use it to resolve long-standing puzzle. In the field of nutrition and so, so, one one, one, issue with even the largest nutritional, studies has, been despite. The heroic, amounts of effort that are, put into them say for example in, in in in in this study.
And. The study what you're seeing is a summary of results from a hundred twenty thousand people tracked, over 20 years and then looking at the effect of each food item on weight gain or weight loss that. The problem is that the effect size of individual, food items has generally been exceedingly, modest and, that's, true of the study as well so if you think about your stereotypes, of what do what, the lead individuals eat, and what, to obese. Individuals, eat it, probably won't surprise you a whole lot to learn that with food most associated, with weight loss is yogurt but, then the magnitude of that effect is relatively, small so each serving of yogurt per day led. To weight loss of, minus four-fifths, weight, loss of four-fifths of a pound per, year in this cohort and then if we think about the other end of the scales where, the. Food that's most associated, with weight gain that's. Probably not going to surprise you but that's our, friend the French fryer right but, again. The effect size was astonishingly. Modest, so each additional serving of fries per day led. To a weight gain of 1.7, pounds per year and so what that means is if you were going to eat fries every, day for lunch for a year and you decided to ditch just leave forgo them and eat the yogurt instead after exercising. That about your actor will 365. Times in a row it, might make a difference in two and a half pounds to your weight right, which was perhaps not very impressive, and maybe you'll just give up and leave the fries, but. But. But that's at the whole population level. And then individual, variation, results, is very large as, there's. A lot of you I'm sure know either clinically, or from personal experience. So. The solution to this puzzle. Possibly, because this is looking at glycemic, response rather, than it effects, on weight per se but, one possible solution came, from this very elegant study from, Aaron Allen Arvin Erin Siegel that, came out and sell just. Just a couple of years ago and what. They did was they measured they, measured a whole lot of personal attributes for 800 people including. A number that they hooked up to continue a continuous. Glucose monitors, and what, they did is they measure the. Microbiome, blood yes to detailed questionnaires. And anthropometrics. And, also food diaries and essentially. What they were able to do was they were able to measure the effect of each food on postprandial, glycemic, response and, what. Was amazing about this is, that they found that although the population average results, were within about 1% of the. Values that, that. Were previously published the, individual, responses, were totally different so, for example they found one individual, his blood sugar went haywire every time the age matters when, they cut tomatoes out of their diet their blood sugar was under much better control and, one. Of the one, of the amazing punch lines of the study was, that there, are fairly sizable, subpopulation. Of the people they looked at it was actually better for them to eat ice cream than, it was to eat a bowl of white rice and. And. They could predict essentially, all of that from the microbiome, like none of the other data that they collected had any particular, predictive value and that's amazing, right like I'm letting loose most people had two questions the, first question is is, there a test that I can do to tell if I'm in the rice category, on the ice cream category and the. Answer is yes they spun out a company called d2 right, now that has weeks by far the best if you're Israeli but, you, know it's obviously the sort of thing that needs to be replicated, in other populations but. Then the second or more interesting question is suppose. I find out that I'm in say the rice category, and I want to be in the ice cream category, could, I permanently, alter my microbiome, somehow to, get from where I am into, that alternative, state now, that might seem kind of frivolous if you're talking about dietary, choices or, remember that the microbiome is also have been linked to drug responses, for everything from acetaminophen, to. Digoxin to, the latest checkpoint inhibitors, and if you're thinking about it in terms of could I predict from a microbiome, based test is my, drug go to be effective, and then could I move people from the non-responder, category, into the responder category by shifting their microbiome, permanently, that's a much more interesting question. So. Anyway to recap, where we are at the moment after twelve years of research, the. First step was to figure out what lean and obese mice differ in their microbiome, so establishing. Biological, plausibility, then, the next step was to show that so lean and obese humans, then. That obese humans come to resemble the in humans in their microbiome, when they're on the weight loss diet, then. That we can transplant, from lean or obese individuals. Into, germ-free mice to alter the phenotype, then, that we can do a population-based, study to.
Find Microbial. Strains that ocean that phenotype in mice and. Then then. Finally to show that the microbiome predicts, how you get to respond to food items in humans so, that's what we are after twelve years right, but even after all of us we haven't yet established that microbes can cause while, that microbes can treat obesity, in humans and in, order to do that what we need to do is we need to go into clinical trials that. Are going to do that. Are going to do either fmt or cocktails. And microbes and, then and, then test and then, test whether, they have an effect on the obesity phenotype, in humans and in order to do that you're looking at because, because. The FDA regulates. Either. Stool or microbes. Isolation, from stool was live by a therapeutic, agents what, you're looking at as many many, years of obtaining IMD's, and doing the clinical trials before, we're going to see that kind of data. So. So, what we need to get from where we are currently, towards. That kind of thing is, basically we need suggestions, from population-based, studies, though there's an association, to go after at all. And this is especially important, at the moment because reviewers, have a lot of study sections, are now expecting, that you've established that, this microbiome, difference between. Between between. Healthy controls and whatever phenotype, you're proposing to study and so, in general you need that as preliminary data rather than as proposing that as the first name of your grant and that's. Still an issue because although although, you've had that DNA sequencing, is getting cheaper all the time it's, still not free and so, as, a result one, very efficient way to do this is to take large cohort, studies where, you have samples already being collected and. And, very rich data being collected about the individuals already piggybacking. Your microbiome component, on that and they're making it an open public resource that people can mine for, a preliminary data to establish plausibility. To go after the phenotype. Then, the intervention, studies and rodents have been especially effective so especially when you're transferring the phenotype from humans across the species boundary to animal models that's, very useful first, whooshing of the microwave play an important, role in the biological circuit, and for demonstrating that you can modulate phenotype. In an animal model although, it still doesn't prove causality in humans and a very sobering example, is the example of leptin where, if you have a slim mouth oh sorry if you have an obese Mouse you can probably slim it down with leptin no matter why it's obese whether its genetic or dietary or what but, it doesn't work very well in humans where only about we're. Only a tiny fraction of humans. Who, are obese actually, have a leptin deficiency and unless, they have a lepton deficiency, web summit leptin us and gonna treat them even, though it's an important part of the circuit it's, not the switch. So then prospective, longitudinal studies. And humans are really valuable just, being able to tell did you did, you have the microbiome, shift first or did you have the phenotypic sugar chef first but, ultimately what we're going to need is we're going to need intervention, studies in humans and the regulatory burden for doing those the present is extremely, high at, least here in the United States, and. I'm going together again, to win by just giving you a couple of other examples of, how this timeline can be accelerated, so, it was only in 2013. That, Sarkis mazmanian in, this lab were, able to show that in the maternal. Immune activation model, which. Which. Causes the phenotype resembling, human autism, but. Essentially they were able to show that m.i.a causes. A whole range of defects. And - ranging from communications, deficits, to cognitive, -, a compulsive. Behavior like marble bearing -, GI, dysfunction. Which is what gave them the the clue that, it might be might. Be relevant to human autism and what, they were able to show in, this model is that, an, ocean microbiome, led to the production of a particular chemical for, EPS which, then if you inject, that into normal, pups they will develop many of the same symptoms and if, you use factorial, is fragile asst a microbe, out of the human gut you, could actually rescue, the microbiome, to start with producing four EPS and to reverse a lot of the deficits so basically the, cognition of them the GI deficits, and the, compulsive, behavior but, not the communications, deficits, so, this was a very exciting preclinical.
Model That, was just four years ago and. I'll skip over this in the interest of time and. Then that led to this year a human, trial of the court and slant asking. Where the fecal transplant, just. Like I showed you for seed if could, improve autism, symptoms and, say. So you need to restrain your enthusiasm, a little bit because, we're looking at an open-label trial not an RCT and we're, only looking at we're. Only looking at about, a dozen subjects, but what's amazing about even this very preliminary work is that, is, that during. The ten weeks of fecal transplantation, and then eight weeks after you. See you, see substantial you, see substantial improvements, both, from the GI symptoms and. On the cognitive symptoms and that's. Really very exciting, and shows the potential for doing this kind of thing for a wide range of other indications, not just for sea death, and. In. In terms of in, terms of other things that you might want to look at one. Thing that John Crichton is always very fond of saying is the gut is not Vegas right what happens in the gut doesn't stay in the gut and we've. Recently seen in the same sort of preclinical, work that, we've been looking at for for. Obesity that I showed you the. Same sort of thing looking at microbiomes. From multiple sclerosis patients, where. The only where, only the microbiome, from the multiple sclerosis patient. Will induce ei e and a mouse we. See that very dramatic difference which, is then linked to some microbiome, differences, and then, also, what Sarkis mazmanian script, we've been able to share the same thing for, a model of Parkinson's, disease where. Where. Again what we see is, as we see substantial differences. Between Parkinson's, patients and controls and then, if you have the right genetic model and a mouse and you transplant microbiome, from. From, a Parkinson's patient you'll, release that phenotype whereas, in contrast if you transplant the microbiome from a healthy individual, you suppress the phenotype, so. So. What so, essentially, what I'm saying is that there's a very wide range of disorders, where, the preclinical, data in mice motivates, human trials, and and. And so either doing prospective, longitudinal studies. For tracking or. Doing. Doing, population, studies then, identifying, lead microbes, and then putting. Them back into the preclinical model, or ultimately, doing intervention, studies in humans we're right at the point where that's.
Well--that's III believe scientifically, justified, even. For diseases you might not have thought willing to the microbiome until, very recently. But. A large part of the barrier what we really need to do is we need to make it easy and. One. Thing that one, thing that really inspires me in this respect as is how rapidly cell, phones have spread through the population, so. Tonya yet CNN Co one, of Jeff's, grad students took, this photo at the mo nutrition clinic and in Malawi that we work with on, the. Gates Foundation project, that we had at the time on malnutrition, and although. Food is scarce in this population, they, do for example have cell phones and what's, amazing about this is there are now over 8 billion active cell phones on earth whereas, there's, only about 7 billion people and. Of the poorest billion people, in the world about 20%, of them have their own cell phone so, the CIL's technology that's just about everywhere, and so for example i took this photo in dhaka in 2012. When I was doing a cloud computing workshop, at ICD DRB we're, right, next to where you can buy a sizable hunk of cow you, can also get your cell minutes recharge and perhaps, most remarkably, when I was working with the Hadza hunter-gatherers. In Tanzania in 2014. I met, a man who a couple of weeks before and shot a giraffe with about an arrow that he made himself and traded. The skin and some of the meat to the next group over to the data for, the cell phone that he's holding here and he. Literally gavels. Hunting from the wild from wild bees to, keep a top top with minutes and and, our electricity at the way at the Trading Post that's, about a 10 kilometer walk from there from his village and and. This, is amazing right you have this technology that's penetrated, everywhere, and the. Reason why is the fundamental, human need to communicate but also the cost of the digital signal processing network that you need to set up that. You need to set up cell service has dropped precipitously, in. Price, with Moore's law and with. The decline in cost in computation, but, one thing it's important to remember is something that's getting cheaper even faster than computation, which, is this white line here as a cost of DNA sequencing which. Is this green line and on the time computations, got a hundredfold cheaper, DNA. Sequencing, has got a millionfold cheaper, and if we see another million fold improvement to DNA sequencing, that's where it's going to get out of the lab and into, into, the price point where any, consumer will be able to use that and then, the challenger's user interface, right because your cellphone ha