Thanks very much, Ken, for that kind introduction. One morning, I was putting together a grant, and I got a phone call, and it was from Ken. And Ken asked me if I could help him out. And so when a former postdoc and a good friend ask you for help, you say, what do you want me to do? And so he wanted me to talk about machine learning. And I said, I have absolutely no clue as to what machine learning even is. So he said, why don't you talk about something you know about? I said, well, OK, I'll talk about how you can find collaborators. Because I'm a physician scientist, but I started as a clinician. And I'm going to tell you an anecdote of how I got into microbiome research, and I'm going to keep it real so that you can see that anybody can do this. If I can do this, you can do this. And it's all about admitting what you can do and what you can't do. So I want to sort of pick up on what Herbert was leaving off in his talk about how do you conquer the unknown. And so as I kind of look through the rearview mirror about what I've learned over the years, I think I have some take-home messages that this is a general roadmap for success. And they're listed on their slides. And I'm going to just go through kind of what that means. So the first thing is know yourself. Review your inventory. You're a doctor, so what have you got to bring to the table? We have patients. You have clinical data. You might have bio samples. Most importantly, you have clinical expertise. And don't poo-poo that. You have a nuanced appreciation of clinical challenges. And that's really important because that helps frame the gap in clinical knowledge. But as Herbert said, when you start wanting to get into an area because it sounds exciting, the tendency is just to say, OK, I have these patients. And I'm going to measure everything. But first, it's impossible to do that. And it's very expensive to do that. So you're better served if you can think before you act and try to define a goal. So what is the question that needs to be addressed? What is the outcome that you want your study to have? And then you have to, after you sit back and pontificate about that, then you have to have a reality check. What do you bring to the table? Other than your patients and your bio samples, what do you already know? Because what you don't know is where you need to get help. And this is another sort of probably oversimplification or stating the obvious. But a lot of us sort of stand around and expect that everybody else knows we're there. And we can't understand why people aren't running to help us. But actually, they've got a day job, too. And they're all very busy. So you have to go beat the bushes and figure out where these people are in your institution. So you've got a clinical problem. You're seeing patients. You've thought about it. You've read. You think you can put a question out there. But then you realize, oh, yeah, it sounds good. But I can't do this or I can't do that. How do you find these other people? Well, in most of your institution, there are institutional clinical and transitional research hubs. And you can access these. You have a dean for research. It's a worst-case scenario. You go knock on somebody's door and just ask them. But most of these are now web-based. And so you can discover this within your own institution. You probably have institutional core facilities. So I would strongly urge you to figure out what they are and what they offer so that you don't have to learn how to do every assay. A lot of this is already available for you. And when in doubt, do what you did when you were on the wards and you didn't know how to do something. You asked somebody else. You asked one of your friends. So informal networking is a great way to figure out what's where and to begin to tap into that. So then you've got the question. You've figured out who you're going to go talk to. But be prepared when you go to that meeting. Because somebody is taking time out of their day to talk to you. And they didn't plan to talk to you. They didn't have you on their radar screen. So they're being nice by just meeting with you. And it's your job to persuade them to participate in your project. And so what's your best chance for success? Well, when you walk into that room, you have to have a clear idea of what it is you want to accomplish and why. And then how can your new friend help you do that? And why should they be excited to do that? So you're making a sales pitch to try to get somebody you've never met before, and they've never met you, interested in working with you. So you have to appear knowledgeable. You have to appear professional. You have to have a clear ask of what you want the other person to do for you, and what they're going to do, and what you're going to do. And then the rubber hits the road. How are you going to pay for it? Because in medicine today, as we all know in the United States, everything is all effort-based. And so if you're putting effort on something, you're not putting effort on something else. So how are you going to pay for this? Not just the cost of the assay, but the other person's time. And then, assuming you can get all that going, now you're going to actually have to work with your new partner. And so how do you make that successful? And Herbert kind of alluded to that at the end of his talk. So often, we collect a sample, and we hand it off. And we expect that all this data is going to come back. And then that's it, right? It's over. It's answered. But in fact, what you realize is the person that you're interacting with has really very little idea about the hypothesis, or the background, or the clinical content. And you have probably no clue about the biostatistics, or how they actually do that. So how are you going to learn that? Well, you have to meet regularly with your new partner so that you can begin to understand their language, and they can begin to understand yours. And it's an iterative process. And as this goes on, it's actually quite rewarding, because you learn things that you never knew about at all. It gives you a new appreciation for the other guy's shoes. And then they get that from you. And over time, you build a very, very good working relationship. Of course, as you're doing this, you have to stay current. Because most of these projects now with big data take a long time to accomplish. And while you're working on your thing, other people in the field are working as well. And so you have to be prepared to adapt. What you started out wanting to do horribly, somebody may have published while you're in the midst of your workup. So how are you going to adapt to that? So those are sort of the general rules. And now I'm going to show you how I got into microbiome workup. And I'm glad that Herbert's sitting in the front row, because we were just talking about an article we wrote for the New England Journal, what, 27 years ago now, Herbert? What was it? It was a year ago. Yeah. And it came out of a lot of this kind of work. So I was just beginning to be a clinician, scientist, physician, investigator. I was interested in fatty liver disease. We were trying to figure out what caused Obi-Obi mice to get NASH. And I had this brilliant idea that maybe you could measure stuff in the gas. It would be the expired air. Of course, it would also be flayed as from the mice. So we collected a sample. And I had found a colleague at Hopkins who had a spectrophotometer. And he could measure all kinds of stuff in air. So I said, great, I'm going to give this stuff to him. And he's going to find isoprostate or something that's going to distinguish the Obi mouse from the wild-type mouse. So when I give him the sample, he comes back and he said, what did you give me? And I said, what do you mean? And he said, it's got tons of ethanol in it. I said, ethanol? My mice weren't drinking ethanol. So then we did it again. And we got the same answer. It was ethanol. I had no idea. So then I went and I talked to an older colleague, Esteban Maze, who was a hepatologist at Hopkins. And he said, well, you know, you can ferment dietary carbohydrate to ethanol. I said, whoa, really? And so then it turned out that what we realized is that the microbiome in the Obi-Obi mice must be different because it was fermenting the chow to ethanol. This is my most famous publication. It is recognized by my nephews, apparently cited in a widely read men's magazine. Hopkins doctor says we have our own internal still. But that then led to other questions. OK, this happens in mice. You know, they eat their stool. So is this really happening in people? So what did we do next? So now I have to get my new friend, who's measured this stuff for me in a little vat of air from mice, to actually bring his machine over into the clinic so we can try to do this in patients. And we did. And what we found is that the more obese patients seem to be making more ethanol in their expired air. So it looks like what we found in the mouse was true in the people. And so then we went back to the mice. So what I'm trying to show you here is that it's an iterative process. We started with something that we found in an animal model. I'm a clinician, we wanted to take it to patients, then we come back to mice. And we showed that if we manipulated the diet of the mice or treat it with antibodies to TNF, and that's how Herbert and I began to know each other, that we could actually modify the phenotype of the mouse. So we could change the fatty liver disease. And then people started to correct these maps of maybe the microbiome was altered and it was somehow causing a change in the gut-liver axis. Now, I want to tell you, that was all before Jeff Gordon became famous, right? And so Jeff published these beautiful papers in PNAS where he actually showed that germ-free mice were leaner than wild-type mice, suggesting that there was something in the microbiome that was facilitating nutrient absorption. And he, of course, has done an enormous amount of work over the years to kind of work out the mechanisms involved in that. And we're not supposed to be talking about the real science of this here, but he's done a phenomenal job. And leading to the idea now that there is an obese microbiome and that it actually contributes to obesity. And that, as you heard earlier, that this can be transferred in mice. And now we know that there's interactions between the liver and the gut. It's not just the microbiome talking to us, but we're actually talking to the microbiome. And for the purposes of people interested in fatty liver disease research, using transfer experiments that you heard a little bit about from the first speaker, that we now know that the microbiome contributes to the pathogenesis of NAFL, NASH, liver fibrosis, and liver cancer. And the general paradigm that people use when they work on this in mice is they give some mice a diet. Some animals get the disease, other animals don't get the disease. They compare the microbiome in the animals that got it versus those that don't. It'll be different. And then they transfer the microbiome from the mouse that got the disease into the one that didn't. And they generally show that they can transfer the disease with the microbiome. And so that's been done with NAFL. It's been done with NASH. And it's been done with liver cancer. And so basically we know that there are rodent models of NAFL and NASH. We started this work back early in the 1990s. And a lot of other people have done similar work. But whether that's actually true in humans gets us to where we are now. We need to take this into people because we know that mice eat their stool. They have different bile acids. So all of these things make us wonder whether all this great discovery really is relevant. And so this was a paper that was published in 2013 where they looked at the stool of children. And they discovered that there were differences in the fecal microbiome in kids that were obese and children that had NASH. But it was a very small sample size. And they only had liver biopsies in the NASH patients. And most of the kids didn't have very much in the way of fibrosis. But interestingly, I've probably covered up the legend there on the bottom of the slide, you can see that the children that had NASH did have more ethanol-producing bacteria. So we thought that this was time to move this into adults. And so while, again, to try to make this real for you and how a real person does this kind of research, remember we started with some observations in the mice and patients. We went back and forth. We get to more work in mice. And now we wanna take this back into people. What do we need to have assembled so we can go back into people? So while all of this was going on, I moved to Duke. And with the help of collaborators, and you see a couple of them down on the slide there, Manal Abdelmalek, but many other people have been involved, we established a biorepository. So here's another thing that you can do as a clinician. You can actually start to bank your samples, but this requires work. You have to figure out who you wanna bank from, what kind of clinical data you wanna store, how you're gonna collect the samples, where you're gonna store them, and get the appropriate regulatory approval. And again, it's very important that you ask a question about what you want to address so that you store samples that can help you address that question. And for us, we wanted to ask what would differentiate patients who are gonna have a bad versus a good outcome of NAFLD? And then we identified tools that we needed to apply to our samples in order to answer that question. And then we needed to find collaborators with relevant expertise. We had a couple of, again, institutes at our institution where there were local collections of experts. And then we convinced them to work with us, and we collaborated to analyze our data set. And basically, we succeeded. We found gold at the end of our rainbow, and we did find some things that distinguished people who had a bad versus a good prognosis from NASH by finding a distinctive transcriptional signature and a pathogenic methylome, and those have been published. There were a lot of other good things that came out of that training, spin-offs, other investigators that we helped model and do other biorepositories, preliminary data for grants and other manuscripts. But most importantly, once we published, people knew that we could do this. And so that attracted other collaborators. And it attracted my friend Jerome Boursier from France who had collected a big bunch of stool from his French NASH patients, and they had all been biopsied. And he said, what if I come over to your lab? Can you help me analyze the microbiome? And again, all I could help him do was get the DNA out of the stool, which made the lab smell really great for about a week. But I also knew some people at Duke who were experts in the microbiome. So once again, I went, knocked on their door, said, here's what I would like to do. Can you help me do this? And they did. So we engaged our core facilities at Duke and they took Jerome's samples, which were 57 people with biopsy-proven NAFLD. They used 16S sequencing. Jerome had already gathered a whole bunch of clinical information. And with that, we were able to both profile the stool, but also they used some of the new technology that you heard about to infer the metagenomics. And this, I think, was the first study in NAFLD that was able to do that. I'm not gonna have time to go through all the results, so they're summarized right here. Basically, what we found was that bacteroides was increased and prevotella was decreased in people with NASH and F2 or greater fibrosis. And that made sense because these two genera are known to be competitors. And we found that ruminococcus increased with F2 or greater fibrosis. And because Jerome had done such a great idea as a clinician, he had all the clinical data, he had collected that carefully, we were able to do multivariant analysis and show that bacteroides, independently of other clinical variables, correlated with NASH and ruminococcus independently associated with fibrosis. So the bottom line is that anybody can do this. Basically, what you have to do is just be thoughtful and purposeful and logical. And most importantly, don't be afraid to say what you don't know. And hopefully, meet people that are as interested and as curious as you are and committed to making discovery. You'll have a great deal of fun working on your projects. You'll meet a lot of good friends and you'll actually learn a lot. So I wanna thank you here and I'm happy to stop. Thank you very much Dr. Diehl for your excellent presentation. So we'll invite Dr. Diehl as well as Dr. Tilg on to the podium here for a panel discussion. So the floor is open to questions from the audience. If you have any questions or comments that you'd like to wish to, you'd like to bring forth. Since I'm on the podium but wasn't a speaker, I'm going to ask each of you, what is your consistent collection of stool? So for my studies, I've been doing whole stool samples and these are frozen immediately upon collection. So that's the minus 80 conservation method, no additional reagent used. I have been, so I wanted to actually bring this up. There are other methods of collection that have been proposed as being almost equivalent to whole stool collection, including stool swabs and rectal swabs. And so maybe you can also comment about how you've been collecting and then also comment on these particular methods. Because I've heard that there is certainly controversy with respect to how well these characterize the whole microbiome. However, they're much easier to collect, practically speaking. So I mean, at the end, it's always a compromise, what is doable. And we also do, for example, refrigerate for degrees to start. So keep the samples as cool as possible. And I think it's doable in many instances. So when you, I think when you decide on a proper study and you say, those are my questions and that's the design, and I want to go for it, and I have to invest, then I think you should go for that. Because all the components we are adding, and I mentioned RNA later, which is considered by many people something very good, but probably it's not that good. So I think the things you're adding could be okay. But I think the original four degrees start could be, it's probably, it's rather safe and good for most. Yeah, actually it was Dr. Borcier who collected the stool for the sample that we analyzed. And I believe he just collected fresh stool and then kept it at four degrees until he could get it at minus 80 degrees. I will say it actually took more time and money to get the stool samples from France to the United States than probably we could have had all the people fly over and give a fresh sample. It would have been easier because it was thought to be a biohazard. But we are trying to do this at Duke now, and it is not trivial to actually get a stool sample from a human being in real time and to keep it consistently refrigerated and frozen. Question. Sorry. Of course, I mean, with swabs it's a little bit different or difficult, and still I think we will use such strategies in the future, accepting that they have certain shortcomings. But there is a lot of information you still can gain, I believe. As opposed to stool swabs and record swabs. Yeah, exactly. I mean, this is something, and it brings us again to consistency. I think when you have samples of 500 patients and it's very consistent, you still can learn a lot. Microphone in the middle. Identify yourself, please. Yeah. This is Dr. Tao from China. I was thinking about that so often that we use the drug for the food production. So how to determine as a confounder to doing this analysis for the microbiota, antibiotics so often used in food production? So how to determine as a confounder? Right. No, I mean, I think the best you can do is try to get information that you think might be relevant for the outcome variable that you want to measure. So in the case of the study that we did, we were most interested in trying to figure out what was the role of the microbiome in perhaps contributing to more advanced fibrosis in NASH. So the data that we collected clinically had to do with other things that we thought would impact on fibrosis in NASH. But we didn't take a detailed food questionnaire, and even if we had, the question is, how accurate would that information have been since the person providing it is the patient, right? So how many people actually know the answer to that question? So it gets down to what we were talking about in the beginning. There's the theoretical of what you would like to do in the best of all possible worlds, but then there's the reality of what you can do. And so can you actually acquire that information? And if so, how? Yeah. Okay. Thank you. Wow. Middle microphone. Yes. I'm Shiv Sarin from New Delhi. We regularly use fecal microbial transplant. And for that, for alcoholic liver disease, and we use that as a fresh isolate. The donor is required after a screening, he would report in the morning, and then the fresh sample is used for FMT. My question to Dr. Tilg and maybe Anna Dale is, have you assessed the metabolome after a storage? Let's say you have a fresh sample, and then you've kept it for 30 days or 15 days. One is metagenome, but has any metabolomic study been done to see how the bacteria which are kept at 80 degrees or 4 degrees, how do they differ in their metabolome? This is, of course, a very important question, and I'm not aware whether this has been investigated precisely in a prospective manner. But coming back to FMT, as you are obviously doing, we have learned that, I mean, irrespective of the metabolomic status, we can, when we froze those samples for at least two years, we can use them with high clinical success. So obviously, overall, whatever works in FMT, we don't know, it's bacteria, metabolites or whatever, it's very stable with freezing at minus 80 degrees. So you can, from that knowledge, which is established in FMT, we can assume that it's very stable. Okay, microphone at the front, please. Just a question to consistency. How important is it to have a standardized diet if you want to compare two groups of patients regarding changes in the microbiome? It probably is important. I can tell you we did not do that in the study that we did, but we subsequently collaborated with Lawrence David, one of the collaborators that I met in the first project, who's very interested in the effect of diet on the microbiome and has published very acute changes with diet modification. So there's no doubt that diet has an effect and it has an effect very quickly. So probably in the future, controlling for diet would be important. Just one important information also regarding drugs, it's not only antibiotics and PPIs. I mean, there was a Nature paper this year, or it was last year, that showed that 20 percent of all drugs used in medicine are affecting the gut microbiome. So I think a detailed drug history is very important for your collection. Microphone at the back. Hi, I'm Nadine. I'm from Australia. I was just wondering if your participants are doing their collection at home and if they're storing in a home freezer for a certain period of time. I think that the patients that Jerome collected did give the sample at home. They put it in the refrigerator and then brought it into the clinic in a cold bag, like a cold lunchbox kind of thing. And then as soon as it came in to us, we put it at minus 80. Is that how you do it? Last question for this panel discussion from the back. Thank you. Ingo Frontier of German Liver Aid. I have a basic question from my understanding. How come fecal transplantation is so successful? Because to my understanding, you have competition between what you transplant and between the existing microbiome. Is this easy to answer? Is it a matter of dosage? I mean, I think it's very much dependent on the clinical situation you're dealing with. So let's say C. diff. We know there is a massive decrease in species. So it's almost, I would not say zero, but there is nothing around. So it's highly effective there because let's say it's easy. There is nothing there and you bring it in and it grows and it works perfectly. So I think that's a very special situation. And in all the other situations, as you are aware, it's at the moment experimental. Of course, when we look, for example, at IBD, let's say ulcerative colitis promising data, but still, of course, immediately we have exactly this aspect of immunity targeting whatever, whatever. So it makes it much more complex. So I would say it's very, very much dependent on the clinical situation. And when you're doing it in an immune-mediated disease where immunity already is exaggerated, it's probably totally different. Thank you. We're having technical difficulties, please hold. Technical glitch here, okay. Actually while we're waiting, maybe I can bring up one question that I'd like to bring to your attention too. So how about shotgun sequencing versus metagenomic sequencing? Can you comment on, so say shotgun sequencing is much cheaper, offers quite a bit more depth than, say, the standard 16s RNA sequencing that's been used, but certainly doesn't offer the same depth as metagenomic sequencing. So do you feel that it achieves a happy medium between the depth of information and cost? It's the compromise, and so it's, I think, a matter of what is in your circumstances technically available, what you can do, what you can afford for what you can pay for. But of course it's, everything is so under evolution, I would believe that within a definite time we will do, of course we will use the most advanced technique because it's then affordable and gives you more information. So I think in a way it's a rapidly evolving field, and I believe very soon we will use probably mainly this technique, or a new one. NME as well. I agree. Again, it comes down to a balance between what you want to do and what you can afford to do and what's available to you.

machine learning

clinical expertise

defining goals

networking

collaborations

microbiome research

gut microbiome