by Y Combinator8/22/2018
David Zeevi is a James S. McDonnell independent fellow at the Rockefeller University Center for Studies in Physics and Biology. He focuses on developing computational methods for studying microbial ecology in the human gut and in the marine environment, and its contribution to human and environmental health.
He was one of the authors on the paper Personalized Nutrition by Prediction of Glycemic Responses.
00:40 – Why did David start working on personalized nutrition?
4:10 – How did the measure the effects of food in their study?
11:20 – How was the study standardized across people?
15:20 – How they measured an individual’s gut microbiome.
16:55 – What is the gut microbiome?
21:30 – Is there an ideal gut microbiome?
22:45 – How do you manipulate your gut microbiome?
24:15 – Fecal transplants.
26:20 – Elizabeth Iorns asks – Does post prandial glucose response regulation track with weight regulation? I.e. can they use their test to determine what individual people should eat or not eat to lose weight?
28:00 – Has this research been turned into a product?
29:00 – Who else worked on this research?
30:00 – How was their predictive algorithm made?
34:40 – Did they end up with any dietary suggestions?
38:20 – Has David changed his own diet?
38:50 – Why fat was vilified.
42:40 – David’s ocean microbiome and other research.
50:30 – Traveling and your microbiome.
56:00 – Trying this out yourself.
Craig Cannon [00:00] – Hey, how’s it going? This is Craig Cannon, and you’re listening to Y Combinator’s Podcast. Today’s episode is with David Zeevi. David is a James S. McDonnell independent fellow at the Rockefeller University Center for Studies and Physics and Biology. He focuses on developing computational methods for studying microbial ecology in the human gut and in the marine environment and its contributions to human and environmental health. He was also one of the authors on the paper, Personalized Nutrition by Prediction of Glycemic Responses. You can find David on Twitter at @DaveZeevi. Here we go. Today we have David Zeevi on the podcast and you are an author on many papers but the paper that I initially contacted you about is called Personalized Nutrition by Prediction of Glycemic Responses. This is a quick summary. People eating identical meals present high variability in post-meal blood glucose response. Personalized diets created with the help of an accurate predictor of blood glucose response that integrates parameters such as dietary habits, physical activity and gut microbiota may successively lower post-meal blood glucose and its long-term metabolic consequences. Why did you start working on this?
David Zeevi [01:14] – We got to see the amazing statistics on metabolic disease in the world. Right now four out of 10 US adults are obese–
Craig Cannon [01:31] – Just to clarify obese means what?
David Zeevi [01:33] – It means a BMI, a body–
Craig Cannon [01:36] – Mass index.
David Zeevi [01:37] – Mass index of over 30, which is actually not that bad but it’s still considered obese by the CDC. That’s for adults. Now, it was about one out of 10 in the 1980s. It progressed massively and this is based both on the World Health Organization and the CDC, Centers for Disease Control. One out of 10 Americans are diabetic, and this is an awful disease, it’s a lot of suffering. It’s a lot of related complications and it’s huge burden not only on people who have the disease but also on healthcare systems. I told you before it’s $250 billion spent on diabetes and it’s related costs in 2012.
Craig Cannon [02:27] – Annually.
David Zeevi [02:28] – It’s a huge deal and it’s widely accepted that nutrition is a major source of these diseases.
Craig Cannon [02:38] – Because diabetes was not nearly as prevalent in for example 1990. Right, it was not nearly as prevalent in the 199O, 1980s, ’70s, it was not as prevalent and neither was obesity. When we came to look at it, we just tried to figure out what are the changes, what are the major changes that were done to our nutrition over the last 30, 40, 50 years or so. And we came up with four, five main changes. First of all, we started consuming much less fat. It was reduced from about 20% of our calories to about 15%. Having fat in your food is tasty. It’s also very fulfilling and everything,
David Zeevi [03:31] – and if you want to, give food a taste without fat, you usually add sugar. Sugar mainly took the place of fat in our diet. There’s a graph I sometimes show in lectures where you see the sugar consumption per capita per year since I think 1700 until today. The crazy fact is that, the annual consumption in 1700 is the daily consumption today. We couldn’t have evolved to undertake to treat this amount of sugar that’s going into our system. The other couple of things that have changed is that we consume much more additives with our food. It’s much less food and much more industrialized. And last thing is that mealtimes changed. We work in shifts, we have electric lights and that changes when we eat and our daily routine in general.
Craig Cannon [04:42] – Gotcha, and so then this study, how are you actually measuring the effects of food intake?
David Zeevi [04:49] – This is also an interesting thing because, so we were thinking that if nutrition did cause this epidemic, what can restore healthy nutrition? When you try to ask what’s healthy nutrition, you can look at popular Time magazine covers for example, and we looked at that. You can see that some of them say that saturated fats are bad for you, some say that fats are good for you and some say that you should be vegetarian, some say that you should eat an Atkins diet, and there’s a famous one which I really like from 1972 that says eating may not be good for you.
Craig Cannon [05:35] – Eating?
David Zeevi [05:36] – Yeah, but we thought as scientists that what you should eat is not a question of trend or fashion or whatever, it’s a scientific question and we want to address it with scientific metrics. We had to choose a metric that was specifically good for this question. And we ended up choosing the blood glucose response. The reason we chose this is that, well, when we eat the carbohydrates in our food, they’re broken down to sugars which are then absorbed by our gut into our blood streams and that causes spikes in our blood glucose levels. These spikes cause insulin secretion from the pancreas which signals the body to store this glucose as fat or as other storage components. This leads to weight gain. Now spikes in blood glucose are also associated with many other metabolic diseases and of course, with diabetes and obesity.
Craig Cannon [06:42] – It leads to weight gain because it is transferred to fat, right? It turns into fat.
David Zeevi [06:47] – Yeah, it turns into fat. Generally not just turns into fat, it also turns into… it gives a boost to the natural mechanism of storage, it causes you to store more. The last thing that was good about glucose responses was that it was very easy to measure. You just connect a small device, continuous glucose monitor, has a tiny needle or a tiny sensor that goes into your body. It’s like probably quarter-inch, something like that, it goes into your body. That measures the glucose levels in your interstitial fluid, that’s the fluid within, between your cells. It’s highly correlated with the glucose in your blood so you get a very accurate measurement of the glucose in your blood, or the proxy for the glucose in your blood every five minutes. You have very high resolution of this metric. Try to think of it. If you now conduct the nutrition study, you can measure weight for example. But weight is very noisy. It’s affected by what you drink, what you eat, that morning, the time of day that you–
Craig Cannon [08:04] – Exercise.
David Zeevi [08:05] – Exercise. You can only measure it once in every long period of time just because it’s very noisy and because it changes very slowly. You can see the effect or the average effect of a diet over a week or two or a month or so. Even though I know some people who now step on the scales every day. But it’s usually recommended to look at every week or so. If you look at blood glucose, you can measure it for every meal. You can just see, get a fast feedback on each and every meal that you ate. That’s what made this blood glucose such a great metric for us. Since it was correlated with so many… so many diseases such as cardiovascular disease, obesity, diabetes and so on. We quickly realized that in order to maintain health or to restore that healthy phenotype, what you need to do is to probably reduce the glucose responses, and that sounded easy. Okay, we just collect a few people and we look at their glucose responses and we find the foods that are good for everyone and–
Craig Cannon [09:24] – And you find the best diet in the world.
David Zeevi [09:26] – And you best find the best diet in the world that would reduce glucose responses and that’s it, and we’re done. But biology and the world is more complicated. What we found is that there were several, usually very small-scale studies that showed that people’s glucose response can be very different from one person to another. Two people eating the same loaf of white bread, one would really spike their glucose and one would really stay flat. That’s true even if you normalize their responses to their responses to glucose. Just to see if, so even foods are not categorically good or bad.
David Zeevi [10:15] – It also depends on the person, and that was shown in very small-scale. We said okay, so… Let’s think of what can affect these glucose measures and, we came up with three main, causes that can affect people’s glucose responses or personal responses. One is genetics which unfortunately we can’t really change. We are what we were wrong with.
Craig Cannon [10:45] – For now.
David Zeevi [10:46] – Yeah, for now, I mean that’s, CRISPR is going to change all that. The second is lifestyle which we all agree should be healthy, active and so on. There’s not a lot to do there, we already know the answer. The third, one that was when we started sort of flying under the radar was the human microbiome which we found to be associated with many diseases, many disorders. If we have time I can tell you a little bit about that. We wanted to create a study that combines all these factors. Nutrition, nutrition as a target, genetics or a proxy for genetics, lifestyle, the microbiome to predict what’s good for people to eat and that’s how we came up with this study.
Craig Cannon [11:40] – You standardized the study, it was something like 800 people? The study was standardized by giving them the same breakfast over the course of a week. Well, there are a couple different breakfasts that you give them.
David Zeevi [11:53] – Tthe first thing we wanted to do in this study is to see, to try to recapitulate the variability that we saw in the small-scale studies. As a controlled way to study variability in people’s responses to food, we replaced their breakfast with a standardized meal that contained either bread, bread and butter, glucose or fructose which had 50 grams of available carbohydrates each and that was to be taken in the morning after the night’s fast without exercising, without eating before that, only drinking, only exercising in two hours after eating the meal because we wanted to get a clean response to the food. What we found is that one person eating the same meal in two different days was very similar to themselves. We had a correlation there of 0.7, 2.77 which is very good considering the noisiness of people. But across people, across the population the variability was huge. People for any given food covered the entire range of responses. They were very reproducible within themselves, you can see a person eating the same loaf of white bread, having two very, no flat responses to glucose. Glucose doesn’t go up after the meal, it doesn’t go rapidly down after that. Other people who were not diabetics, were not pre-diabetics or anything had huge spikes to the same, the exact same loaf of white bread. These people, you couldn’t tell the difference otherwise.
David Zeevi [13:40] – Again it’s not just that one food is categorically worse than other foods. Some people responded, had the highest response to glucose, some people had the highest response to bread and minority had the highest response to bread and butter. Actually, fewer people had a high response to bread and butter than to bread alone.
Craig Cannon [14:05] – The fat is somehow neutralizing it?
David Zeevi [14:08] – Yeah, we think it is.
Craig Cannon [14:11] – Then interestingly it’s not, in the pursuit of the optimal diet, it’s not just that, “Oh, white bread has a lot of sugar, ice-cream also has a lot of sugar, this isn’t good for you, you can’t eat it.” Someone will respond in one way to bread and then differently to ice cream?
David Zeevi [14:29] – Yeah, we saw that exactly. The exact same thing we saw with naturally occurring foods. Some people have high response to rice for example and low response to ice cream, and other people would be the other way around. That’s with the exact same amount of carbohydrates in the food.
Craig Cannon [14:48] – We should clarify. The breakfast was standardized, then they could eat whatever they wanted afterward?
David Zeevi [14:55] – Mm-hmm, but they had to log it. We also gave them an app in which they recorded what and when they ate. Maybe I should say a few things about what we collected in the study? We recruited people, about 800 people. We had them go through a process in which they gave us blood, they filled in questionnaires, both food frequency questionnaire and general medical questionnaires. We had them connected to a continuous glucose monitor as I told you before. That measured their blood glucose every five minutes for the duration of a week and then this week, we also gave them an app which we developed in which they recorded what and when they ate, slept, exercised and so on and the exact amounts of every food in their diet. We also gave them weights to weigh their–
Craig Cannon [15:46] – Oh, you give them a scale?
David Zeevi [15:47] – A scale to weigh their food when they go to, when they eat at home and we gave them some leeway to eat at restaurants as well.
Craig Cannon [15:57] – And did the stool sample element, was that in the original spec of the–
David Zeevi [16:01] – Yeah, we also collected stool samples which we analyzed to see the microbiome in various levels, both which microbes are in there, which genes of the microbes are in there. If I talk later about microbiomes, it’s an amazing ecosystem we know, with thousands of species about as many cells as in the human body–
Craig Cannon [16:29] – Just all in your stomach.
David Zeevi [16:30] – All in your gut. It weighs as much as your brain or a little bit more than your brain. It’s like, people call it like the forgotten organ and not so forgotten now. These microbes have 150 times more genes than are in the human genome. They have about three million genes. They have huge metabolic potential. This metabolic potential can be harnessed or can be accounted for when we’re looking at what people are eating. Yhis is very interesting because unlike genetics, the microbes can be changed. If we figure out a way to change the microbes that are affecting or have a deleterious effect on our health, we can maybe improve people’s health altogether.
Craig Cannon [17:15] – You should both explain like what this gut microbiome is actually for people. Because this word gets thrown around a lot and then you’re talking about changing it, and how would you even go about doing that. For context, let’s give like a proper definition for folks.
David Zeevi [17:32] – Yhe gut microbiome is the ecosystem of bacteria archaea which is also a type of unicellular creature, fungi, viruses and small worms or whatever that we have in and around our body that are not of human origin, that’s the microbiome generally. All of its associated genes and genetic material and so on and so forth. That’s what usually people mean when they say microbiome. As I said before, it’s huge. There’s a lot of cells, there’s a lot of diversity there. There are a lot of genes, and there are more and more related, more and more relations are found between this gut microbiome and many disorders and different outcomes. I can name a few examples. One of my favorite microbiome studies was done in Stanley Hazen’s group in the Cleveland Clinic. They looked at carnitine, which is a compound that is found in red meat. This carnitine is metabolized by the microbiome to form TMA. It’s a compound. TMA is then oxidized in the liver to form TMAO, and TMAO causes a reduce in reverse cholesterol transport and bile acid synthesis. These are long words, but what it essentially means is that it causes atherosclerosis. These two processes, if they it causes atherosclerosis, it causes your arteries to clot. Interestingly, if you remove these specific microbes that metabolize carnitine from the equation, the downstream effects are attenuated as well. This was a major thing for us because this is the first time we saw that the microbiome can affect how each and every one of us responds to nutrition, so it was beautiful.
David Zeevi [19:41] – Another study, by Nan Chin and colleagues in 2014. I’m not sure, maybe it was published in Nature but I’m not sure. They showed that you can accurately detect cirrhosis, liver disease by only looking at your gut microbes and that showed us that the gut microbes can reflect our health status.
Craig Cannon [20:11] – In monitoring what people are eating and their stool samples, you can kind of recompose what their gut microbiome is, right?
David Zeevi [20:20] – Right. Well, you have to measure the gut microbiome as well but you can, maybe you can get some idea on their health status and what they’re eating from the gut microbiome. It’s not only that microbes can reflect your health they can also actively affect your health and there are a few very nice studies by Jeff Gordon’s group at Washington University of St. Louis, especially one that I liked the most from 2013, they took pairs of twins that were discordant for obesity. One twin obese and one twin was–
Craig Cannon [21:07] – These are mice?
David Zeevi [21:08] – No.
Craig Cannon [21:09] – These are people? Okay.
David Zeevi [21:10] – They transplanted their microbiome into germ-free mice. Germ-free mice are mice that are born and raised in sterile conditions and they don’t have a gut microbiome of their own. These mice were transplanted, there microbiomes of twins, one obese and one lean, many pairs of twins. Interestingly, the mice that received the microbes of the obese twin became obese and the mice that received the microbes of the lean twin remained lean after eating the same food and doing the same things. That also showed us that it’s pretty…
Craig Cannon [21:46] – Maybe the logical extension in the sense that every human wants things to be black or white. Were you often asked like, “Okay, is there an ideal gut microbiome? Because rather than the diet maybe we just do the gut microbiome and then we do the transfer and everyone has the same one.”
David Zeevi [22:06] – I’m not sure if there’s an answer or there’s a clear answer. People are trying to study the gut microbiome in health and disease. The thing is that it’s, and this is maybe just my opinion. It’s so diverse that you need a huge sample to study what’s good and what’s bad in a microbiome. Once you get to know the exact effect size of the microbiome on human health and whatever, maybe then you can start asking the question of what is healthy and what is not healthy. We know right now that, we know of some species that are healthier than others or are associated with better health. Generally, a microbiome diversity, a high diversity of the microbiome is associated with healthy hosts. You want to let your kid eat the dirt I guess or have a dog. That’s usually contributing to a healthy microbiome.
Craig Cannon [23:13] – Okay, and so then in the context of Jeff Gordon’s group where they identify maybe a certain bacteria that’s not ideal. What is the process of trying to eliminate it?
David Zeevi [23:27] – I don’t know if I have a good answer for that. There are a lot of ways to affect or to exert an effect on the gut microbiome. You can take antibiotics or very specific antibiotics. You can try and replace this micro by ingesting some sort of probiotic or some sort of microbe that will occupy the same niche as this microbe just to push it out and take over. You can take prebiotics which is some sort of fiber but I’m not sure that people have an idea of the full effects of each and every one of these things. There’s still a lot of study to be done in this field.
Craig Cannon [24:14] – I’ve always wondered. I read a couple of studies before this podcast and I read the book, I Contain Multitudes. But there are so many things out there between like fecal transplants and the pills that you can digest where companies say we have found the optimal probiotic or got biome supplement. In large part, have you found that stuff to be effective or is it just kind of bogus?
David Zeevi [24:42] – I’m not sure I can answer that with confidence.
Craig Cannon [24:45] – Yeah, right. Because specifically that the fecal transplant stuff is the most eye-catching, definitely.
David Zeevi [24:53] – That’s the dark side of microbiome science.
Craig Cannon [24:55] – Exactly. But it has been proven effective for some percentage of people, right?
David Zeevi [25:00] – Fecal transplants have been used, their claim to fame is by treating clostridium difficile infections. That’s a type of infection that takes over your gut. It’s a certain bacterium that takes over your gut. It pushes everything else out. Now when you try to treat it with antibiotics, it usually sporulates, it creates spores and it resists the antibiotic. The antibiotic kills everything else and this thing just takes over all the gut spaces that were left by other bacteria. Usually when you have a c-diff infection, it’s predominantly the most abundant microbe in your gut, and it causes extreme diarrhea and these sort of things. Now when you treat these patients with antibiotics, let’s say it’s not working, so you want to treat them with something else, you want to replace their healthy microbiome and you indeed transplant stool into these people. That transplant works mainly because their microbiome is so depleted and it’s like cultivating an ecosystem in a place where there was none. If you take this ecosystem and you try to transplant it to a person with a healthy ecosystem, that’s not necessarily going to work.
Craig Cannon [26:20] – Right, okay.
David Zeevi [26:21] – But people are making big money out of it. I heard that companies collecting stool out of professional athletes, NFL players, NBA players and so on to transplant to other people. I support that.
Craig Cannon [26:41] – Yeah, I mean however you want to get paid, go for it.
David Zeevi [26:43] – Yeah, exactly.
Craig Cannon [26:45] – We have a couple questions people submitted ’cause they were very curious about this. So Elizabeth Irons from Science Exchange had a couple of questions, one of which was, does postprandial glucose response which is the response that you’re measuring with the glucose monitor, does it track with weight regulation (i.e. can they use, can you guys use their tests to determine what individual people should eat or not eat to lose weight)?
David Zeevi [27:13] – Theoretically, postprandial glucose response is associated with changes in weight just because of a mechanism I told you about, that when we eat things that spike our blood glucose we cause insulin secretion which signals the body to store things as fat and among other things. We haven’t tried and tested it specifically. Our study was a short-term study, even the intervention that we there was a two-week intervention, a good week and a bad week, we can get to that later. But we didn’t do anything that’s longer-term and in order to see differences in weight you need to follow people for months if not years. But choosing the foods that are right for you out of your own diet gives you an advantage, if indeed it does improve your blood glucose and therefore your weight, because you don’t have to change your diet drastically. You only have to eat out of your own diet the foods that are good for you.
Craig Cannon [28:25] – Right, and it could at the very least, steer you away from becoming pre-diabetic.
David Zeevi [28:28] – Exactly.
Craig Cannon [28:29] – Which is another huge concern. We should talk about the follow on stuff but another very common question is who is turning this into a product? Or how is that being done?
David Zeevi [28:41] – There’s the company called DayTwo. You can go to their websites, they’re working on that. What they’re doing is, they did a study similar to ours in which they collected participants and they had them go through this sort of analysis. I’m not sure what they’re doing, I’m not in touch with them or anything, but I think what they do now is they have you fill in a questionnaire and they take a sample of the microbiome and they give you a prediction for each food that you eat if it’s good for you or bad for you.
Craig Cannon [29:21] – You’re doing some computer science stuff as well, right? You built essentially an algorithm from the 800–
David Zeevi [29:29] – Well, it’s a good time as any to say that it’s not just me.
Craig Cannon [29:32] – Yeah, of course.
David Zeevi [29:33] – We’re a huge group of people, you can see on the paper and mostly the person that I’ve worked with closest and this is Tal Korem, who’s going to start a faculty position in Columbia in the fall. So if you’re a potential PhD candidate or a postdoc listening to this podcast, then you can contact him. He’s a very good scientist and under the supervision of Eran Segal. And with the fabulous Adina Weinberger who handled the wet lab and all the samples and everything and made protocols out of, where there were none. It was an amazing group of tens of people and a lot of… And obviously if I try to thank everyone else I’ll forget some.
Craig Cannon [30:22] – There are other people.
David Zeevi [30:24] – But yeah, please download the paper and see for yourself.
Craig Cannon [30:28] – And there’s a cool video you guys made but yeah, keep going.
David Zeevi [30:34] – Yeah, you were asking about the algorithm? We developed an algorithm that was based on people’s metrics, on their… What we first did was to see if these responses to food were associated with any of the other metrics that we found. And we found many associations between the response to standardized meals for example to BMI and to glycated hemoglobin which is the metric for diabetes. We found many, many associations with gut microbes. We said, “Okay, why not try to combine all these signals into something that would predict people’s responses to any given meal?” Just to give you an idea of what people used before we came around to do that, so usually when you think of glucose responses you think of counting carbs. You just take the correlation between the carbs and the meal and if you take the correlation between the carbs and the meal and the postprandial glucose response of the meal, you get a correlation in R of 0.38 which is not a very good correlation. It’s significant because it’s a lot of points. For example, there’s meals in which there is
David Zeevi [31:48] – a huge amount of carbs but not high response to glucose and the other way around also happens. We were set out to fix that, to try and do something better. We built an algorithm on these 800 people we collected. We used booster decision trees on about.. We didn’t predict people would eat the meals, we predict about 40, more than 45,000 meals. We train on a subset of the 800 people and we tested our prediction on the left out cohort and we made sure that person’s meals were not both in training and tests so this thing would be more generalizable. In terms of features, we took the microbiome composition of people including the microbiome genes and microbial growth rates which is from a different, very nice study. We looked at the nutrients in every meal; fat, carbohydrates and so on. But also sodium and other nutrients. Other recorded features; meal times, sleep times and so on. And blood parameters, questionnaires–
Craig Cannon [33:05] – Meaning blood type, that sort of thing?
David Zeevi [33:07] – No, not blood type but for example, cholesterol in the blood or like hemoglobin and these sort of things. Overall, we had 137 features, after feature selection on 40 something thousand meals. We ran this prediction. This prediction got us to an R of 0.68 compared to the previous 0.38. And this R of 0.68 is pretty close to the 0.7 that we get when we look at the same person eating two different meals, the same person eating the same meals in two different days. This is a theoretical upper bound that we almost reached. We then collected 100 additional people that were not used to create the algorithm or anything and we tested this prediction on them.
Craig Cannon [34:00] – Meaning you took a stool sample.
David Zeevi [34:01] – We took a stool sample and we had them go through a week of glucose monitoring. We ignored their glucometer and we tried to use all the data that we collected on them to predict how their spikes would look. And we got an R of 0.7 again which was great. It means that this predictor is generalizable at least for the Israeli public.
Craig Cannon [34:27] – I was wondering that, having not been to Israel, is there a large difference in the types of foods? I don’t know, are you really good at tracking hummus and that kind of stuff?
David Zeevi [34:37] – People ate hummus but people also ate… In Israel people eat that Western diet maybe fortified with more vegetables. One thing I can tell about New York is that it’s harder to find fresh vegetables here, even though there’s the fruit carts that are really nice but, still, not of course so much.
Craig Cannon [35:06] – Were there dietary suggestions that you took away from this? Or did you kind of just step back, for instance, you mentioned fat? This is now a thing that’s much more common people doing. Ketogenic diets, or just adding more fat, fewer carbs. Did you guys walk away with suggestions or did you kind of not choose to make any?
David Zeevi [35:32] – We chose not to make suggestions. This kind of beats the purpose of what we found that, people are very different and anything universal and a universal dietary recommendation would be suboptimal at best.
Craig Cannon [35:51] – There were no foods where consistently they were good.
David Zeevi [35:55] – No.
Craig Cannon [35:58] – I didn’t expect that. We should talk about your bread study because I found that interesting and related where you basically increase the amount of bread someone consumed over, what did you say? From 15 to 30%.
David Zeevi [36:18] – This study spiked another study that was about bread. We collected 20 individuals, we gave them just white bread for a week, we gave them two weeks of washout and then whole wheat bread made in traditional methods and that sort of thing. It was randomized and, some people started with that brand, some people started with this brand. We measured the microbiome along the way and one take-home message from this study is that people’s microbiome changed from this huge consumption of bread. Usual bread consumption over this cohort, of the big cohort that we didn’t offer this 20 people cohort was about 10% of daily calories came from bread. In this study, we upped those to 25, 30% of their calories. Despite this change, this significant change in diet, their microbiomes didn’t change. You can see that their microbiomes remained mostly similar to their own microbiomes and still dissimilar to other people, even though they changed their diet drastically.
Craig Cannon [37:32] – How long were the effects of increasing the bread consumption?
David Zeevi [37:37] – On the microbiome? We didn’t see any effect that we can consider consistent across the population. There were some effects in some people and other effects in other people but there was not a consistent change across people. The effect depends mostly on your initial microbiome composition, and we still need to study how certain things affect your microbiome given your initial microbiome configuration.
Craig Cannon [38:18] – Are there any long-term studies being done now on microbiome and changes in microbiome?
David Zeevi [38:23] – Eran Segal’s group, the group in which I conducted these studies is doing a long-term study on 200 or 300 people. They follow them for six months or a year–
Craig Cannon [38:39] – Doing the same stuff.
David Zeevi [38:40] – Doing similar stuff and I think it’s going to be a very exciting study with very exciting data. It’s going to be beautiful data.
Craig Cannon [38:49] – Spoken like a true nerd. So have you changed your diet because you said you were part of like the beta test basically before the full-on study happened.
David Zeevi [39:01] – I did participate in the bread study.
Craig Cannon [39:04] – Oh, you did. Okay. What have you changed about your diet or have you?
David Zeevi [39:09] – Well, I’m not afraid of dietary fats anymore. That’s one thing that, but it’s not just this study that convinced me. It’s reading the history that convinced me. I can say in a few words why fat got vilified. It all started in the 1950s where a guy named Ancel Keys. He had a notion that, “Okay, something is clogging the arteries, this thing is fat and fat is probably the cause, dietary fat can probably cause this thing.” He supported his claim by looking at six countries. I have it somewhere in my notes. It was Japan, Italy, the UK, Canada, the US, and Australia. He correlated the fat percentage out of the total calories consumed by a person with cardiovascular disease. He saw an almost perfect correlation and that… that led him to get funding for studying other stuff. Now there was data on 22 countries at that time including for example, France that had a huge amount of fat from calories but not a huge amount of cardiovascular disease–
Craig Cannon [40:39] – That didn’t make it into his study?
David Zeevi [40:40] – That didn’t make into the… I don’t know if it’s a study or just something that prompted the study. Anyway, he got very famous, he was on the cover of Time magazine. In 1961, the American Heart Association had a recommendation to decrease fat consumption. This kept going and in 1970s, there was a committee of the Senate called the McGovern committee, that was, a Committee on nutrition and human needs or something like that. It recommended reduction in fat and what came out of this committee was, what’s known today as the food pyramid. Have you seen a food pyramid?
Craig Cannon [41:22] – Yeah. It usually has, it’s like lined up with a lot of bread–
David Zeevi [41:28] – The bread is a foundation. Yeah, and there’s a small portion of fat at the top. This indeed caused Americans and the world over to stop consuming fat and start consuming more carbs. And you can see it, if you look at… and there is something called the NHANES Study and it’s the National Health and Nutritional Examination Survey. They publish something every few years, and if you look at their stats, you can see that people did consume more carbs and less fat. Just when they started consuming more carbs and less fat did this epidemic of obesity and diabetes begin. Now, is this related? Maybe not just this, maybe there are probably other effects including the rise in sugar and high fructose corn syrup and all that and additives to the diet. But that’s probably one of the effects. Just by looking at this experiment done on a billion people, and just by reading the history, I stopped being afraid of dietary fats.
Craig Cannon [42:43] – Right, and you’re fine now.
David Zeevi [42:44] – And I’m fine.
Craig Cannon [42:46] – You were mentioning the research that you’re working up to right now, and I found it very interesting because you’re thinking about the ocean, you’re thinking about bacteria in the ocean. I found this interesting trend in that you’re just seemingly just trying to help people with your studies, your research. The first one being help people lose weight, maintain health, the second one being possibly across the entire environment of carbon dioxide. But could you explain what you’re interested in and what you’re working on in the new study.
David Zeevi [43:18] – In a word, I’m trying to move from a more human oriented view. Instead of looking at the human microbiome and trying to see how it affects human health, I’m trying to look at the ocean or soil microbiome and see how it affects global health. Microbes in the ocean for example are responsible for about 50% of the oxygen that you breathe. They recycle a lot of metabolites. They do a lot of these things and what I’m trying to do is to apply my know-how both in microbiome analysis and in data science, and to combine data that’s publicly available on the ocean or samples that I will collect with other data that’s publicly available on a bunch of other things that you can collect from the ocean and see where it gets me. Maybe seeing which bacteria, which conditions can sequester more CO2 from the atmosphere to see how we can treat pollution in the ocean, acidification of the ocean that causes all the corals to die. That’s the sort of things, this is sort of questions I’m after right now but actually before that, we’re in the process of publishing a different study that still looks into the human microbiome. This is a really interesting one to me because when we were finished with this big study of 800, 900 people, we next started on, our next thoughts were let’s see if we can try to clarify what role the microbiome has in this. Now usually what studies do predominantly is that they either look at a whole bacterium to see if it’s there. They just count the number of microbes that are in your gut,
David Zeevi [45:24] – they do that by taking your stool. They produce DNA out of the microbes and they sequence it. They use a sequencing machine that breaks it down to small pieces and tells you each, and then you can map it and say, for each piece, which bacteria it came from. Or which bacterial gene it came from. What we thought is that this is interesting, but what we really want to see is something that’s bigger than genes but smaller than microbes, smaller than a genome. We want to see regions in the microbiome and how they change within people. We produced an algorithm, I won’t get into it right now but that accurately maps each of these small tiny DNA fragments into a microbe. Some of the mapped to two microbes because bacteria are very promiscuous about sharing DNA.
Craig Cannon [46:17] – Yeah, I didn’t realize that until I read the book, and that was crazy they transfer–
David Zeevi [46:21] – They transfer a lot of stuff. Yeah, it’s really crazy. We wrote a sort of algorithm that would help delineate it a little bit. And then we run another algorithm that would find regions in the genomes of people’s microbe that were either deleted completely, or that are present in a higher copy number. We looked at these regions. We found about 5,000, 6,000 of these regions across the 900 and something people that we looked at. We just compiled all the people from all the studies. These regions were prevalent across all microbes. They were very, very… they were they were all there. And we correlated these regions with metrics of health that we also collected in these studies like BMI, weights, glycated hemoglobin and these sort of things. And what we found is that, we found many correlations, about 100 or more correlations and one specific correlation that we dived into just to see what we can get from this region showed us maybe, or proposed mechanistic connection between the microbiome and human health. So this is like, well it’s a tiny region,
David Zeevi [47:55] – the microbe, probably 1% of the microbe’s genome. Gor people who have this region in the genome of their microbiome are about 15 pounds thinner than people who don’t have this region. We were baffled.
Craig Cannon [48:14] – Yeah, wow.
David Zeevi [48:18] – Now the reason on why we thought that the interesting thing was not microbes and not genes but something in the middle is that we could look at this region and see what genes are there and try to compile them into some sort of a pathway, a metabolic pathway. Apparently, what this region does is it takes… it takes up sugar or sugar alcohols from the gut and in an energy favorable process for the bacteria, it turns it into butyrate. Now, butyrate is a compound that was shown to be very advantageous for the host, because it reduces inflammation and it helps treat in mice I think supplementing their diet with butyrate or adding butyrate to their gut directly really improve their metabolism, the glucose metabolism and so on. This is of course not proof, this is not causality or anything and we’re still set out to prove in order to show it some way. But it could be that these bacteria are enjoying a compound that’s just lying there, they’re producing butyrates. Then the host is enjoying this butyrate. And if this region doesn’t exist, then the host is not enjoying this great butyrate.
Craig Cannon [49:36] – Could you just take supplemental butyrate?
David Zeevi [49:39] – Maybe, I don’t know if it will help you, and it would probably taste awful.
Craig Cannon [49:46] – But I mean for that extra 15 pounds, people would probably do anything.
David Zeevi [49:49] – I don’t think so. I think that that you would gain more from having a bacterium that metabolizes things that you eat and fiber that you eat into butyrate than eating butyrate directly. Another question could be, could you supplement people with this specific region?
Craig Cannon [50:11] – Maybe or some kind of CRISPR situation where you edit. Let’s go back to the ocean studies, what’s coming up next for you?
David Zeevi [50:22] – Coming up next, I’m going to look at microbiota in the ocean and I’m going to look at many layers of data including oil refineries, oil wells and that sort of thing that are situated in the ocean. I’m going to try for example to look for, for genes that metabolize these things, these compounds or metabolize plastic in the plastic islands and Pacific for example. I’m going to also add many other data layers that you can get from NASA just to ask very basic and interesting questions in the ocean microbiome that I’m interested in.
Craig Cannon [51:00] – I remember just a random question. You’ve been in New York you said for a year after you were you in Israel your whole life?
David Zeevi [51:08] – Most of it, for the most part. Have you noticed any changes in your personal microbiome since moving to a new country, new food? Any weirdness, any good things? I haven’t tested the microbiome. I’m vegetarian so I don’t see why it would change so much. I’m not eating any food that is too processed.
Craig Cannon [51:30] – I’ve just heard these explanations of like going to whatever, pick a country. So going to Israel as an American you’re like your stomach is a little off, you’re on a plane, you’re a little weird but it’s been fine for you.
David Zeevi [51:42] – That happens a lot. Eric Alm at MIT, if I’m not mistaken, had a study in which he followed his and the postdoc of his diet for microbiome for a year and they traveled a lot and you can really see changes, differences in the microbiome when traveling. But I think, I’m not sure, I’m trying to, I’m probably… I’m not doing good this–
Craig Cannon [52:11] – It’s okay, this is what I do all the time, you’re fine.
David Zeevi [52:12] – But I think that it bounced back when they got back. You get this, you get this distribution of bacteria in your gut that even when you go someplace else, it changes in abundance but it doesn’t change in presence or absence so it bounces back when you get to a different place.
Craig Cannon [52:33] – What about food poisoning?
David Zeevi [52:36] – That could cause your microbiome to…
Craig Cannon [52:40] – Really?
David Zeevi [52:41] – Change a little bit but also it reinoculates and it stabilizes. We have a lot of things that stabilize our microbiome. Some people think that maybe the appendix is related with that and maybe it stores microbiome for times of distress
Craig Cannon [52:57] – In that event. Interesting.
David Zeevi [52:59] – Yeah, food poisoning, your microbiome is swept out and then the appendix reinoculates your gut.
Craig Cannon [53:06] – Yeah, ‘ecause I was traveling earlier this year and then got food poisoning two hours before the flight back from London. But it was a week or two, I just felt off and I couldn’t explain it. I’m just like looking for cheap answers right now.
David Zeevi [53:20] – That was London. Earlier, you mentioned doing an intervention in the 800 person study, the one you published, what does that actually mean? What we wanted to do is to get a proof of concept just to show that this predicted diets can actually work.
David Zeevi [53:38] – We wanted to see for ourselves. We collected 26 participants, most of them were pre-diabetics. We had them go through a week of profiling like we did with the 800 plus 100 cohort. Then we had them go through a good week that was designed to reduce their blood glucose levels and a bad week that was designed to increase their blood glucose levels. These weeks were followed in random order, there were double blinded and they were isocaloric. They had the same amount of calories for each day, for each breakfast, for each lunch and so on and so forth. People actually didn’t know if they were on the bad week or the good week. Because they were based on people’s meals that they usually eat. Half of the people, about half of the people were predicted, their good and bad weeks were predicted using a predictor. Since we didn’t have anything to compare to, we created our own gold standard which were two researchers, Orly and Daphna who looked at people’s glucose responses during the profiling week for half of the people and just based on their responses, something that’s not available to people usually, they divided their foods into good week and bad week. This is something that can only be done for foods you’ve tested and with a predictor, you can do it for any given food, right? But we wanted something to compare to. This worked perfectly. First of all, we had some foods that were on the bad diets of some people were on the good diets of other people.
David Zeevi [55:17] – For four people for example, pizza was on the bad diet and for two people it was on the good diet.
Craig Cannon [55:22] – Nice.
David Zeevi [55:23] – So you want to hope that pizza is on the good diet.
Craig Cannon [55:26] – You might get lucky.
David Zeevi [55:27] – You might get lucky. Based on this very small sample we have like a 33% chance of getting lucky with pizza. In the bad week, we saw huge glucose peaks for most people. Some that if you were a physician you would look at and you would say this person is a pre-diabetic. These peaks completely normalized during the good week. For some people the difference between the good and bad week were almost two or three folds in the responses to meals. And this was both for the gold standard and the predictor and works the same. We were very happy about that. Since we followed the microbiomes of people every day, you could see consistent changes to the microbiome following a good diet or a bad diet. And these changes were consistent both within people and consistent with the literature showing that, bacteria that increased during the good diet were considered beneficial and that decreased during the good diet or increased during the bad diet were considered deleterious or harmful.
Craig Cannon [56:33] – That’s great. If I wanted to do this study on myself basically, could I just buy a continuous glucose monitor and go for it? I guess I need some kind of way to measure my gut biome.
David Zeevi [56:45] – Well, I guess you need the support of all the other people who participated in the study for the algorithm to work. Right now the best option is either collect 1,000 people or try to open your ears to see if there’s any upcoming studies. Or go to DayTwo, but I’m not trying to give them a promotion or anything.
Craig Cannon [57:09] – You haven’t tried it yet?
David Zeevi [57:10] – I haven’t, no.
Craig Cannon [57:12] – Okay, cool, alright, well thanks so much for your time.
David Zeevi [57:14] – Thank you.
Craig Cannon [57:16] – Alright, thanks for listening. As always you can find the transcript and the video at blog.ycombinator.com and if you have a second it would be awesome to give us a rating and review wherever you find your podcasts. See you next time.
Y Combinator created a new model for funding early stage startups. Twice a year we invest a small amount of money ($150k) in a large number of startups (recently 200). The startups move to Silicon