August 23, 2019Preventive Healthcare

The future of nutrition - part 1

Interview with:

Eran Elinav, MD, PhD

Head of the research group at the Department of Immunology, Weizmann Institute of Science.

Dr. Eran Elinav is a professor at the Department of Immunology, Weizmann Institute of Science, and since 2019, the director of the cancer-microbiome division, at the Deutsches Krebsforschungszentrum (DKFZ), Heidelberg, Germany. Dr. Elinav is also a co-author, together with Prof. Eran Segal, of the book, The Personalized Diet, summarizing some of their research on personalized nutrition.  

Your book, The Personalized Diet, reveals the fact that standardized formulas don't work and that there is no one universal, ideal diet for everyone. Your study was one of the largest ever on nutrition and health at the Weizmann Research Institute where you have proven that individuals react differently to the same food. Could you please tell us more about your research on personalized nutrition? 

Dr. Elinav: This research was conducted in close cooperation and collaboration with my friend and colleague Eran Segal who is a mathematician also from the Weizmann Institute. We originally wanted to explore the connection between different nutritional regimens and our gut microbes, which live within us from the moment we are born till the moment we die, and to try and study whether we can learn something about this relationship between food and the microbes and how they regulate our metabolic health. When we read all the literature, when we try to come up with a so-called healthy diet for ourselves, we were quite surprised to find everything that we  do is based on several scoring systems, which give different numbers to foods or food items. Based on these numbers, we performed some mixing and matching and tried to come up with healthy diets. For example, calories are such a grading system which gives different numbers to different foods. An even more widely used grading system is called the glycemic index, which gives numbers to different foods based on their capacity to raise the blood sugar levels in our blood. All of this glycemic index method is based on very small studies in the 1970s, which included usually 10 human volunteers, which were given identical foods, and their blood sugar levels were measured after they ate this identical food. Based on the increase of the blood sugar levels, each food was given a number which is the glycemic index for that food. What was really surprising for us is that when we repeated this experiment, but not with 10 human individuals, but with 1,000 human individuals, and measured their responses to identical foods that they were given, the average spike in blood sugar levels to each food was exactly the same as the index – it was. 

So it is absolutely correct. But when you looked at the individualized level, we saw that some people reacted very differently than others when taking the same exact food. This already intuitively told us that the concept which we were all chasing for so many years of a one-size-fits-all diet can probably not exist, because if your glycemic response to a given food is the opposite of mine, then the same food cannot be included in a healthy diet for both of us.  This formed the basis for a very ambitious project, which we termed the Personalized Nutrition Project, in which we took these 1,000 people and kindly asked them to give us a week of their life. During this week, we collected an unprecedented amount of person-specific data by first sequencing their gut microbes very deeply, making them fill many questionnaires related to their background conditions, dietary habits, and their family history. We measured their height, weight, and took many blood tests, we connected each of these individuals to a continuous glucose monitor, which measured their blood sugar levels every five minutes for an entire week. So from each of these 1,000 individuals, we didn't just get six measurements of sugar as was done in the 70s, but 2,500 measurements, which created a huge database, and we created a smartphone app for this project: in which the people posted everything that they were doing during this week, when they were eating, what they were eating, when they were going to sleep, when they were waking up. Together this created a huge amount of person specific data on each one of the 1,000 volunteers. After this was collected, a very smart group of computational students generated an artificial intelligence-based machine learning-system that created an algorithm that is unique to each participant and can accurately predict a person's response to any given food, even foods that they were never  exposed to, during the week of follow-up. We were very happy to computationally validate the system and to show that it was much superior to what we empirically do today. But then we also put ourselves to the test, we took a new group of volunteers, these were people that already feature prediabetes, these people already have disturbances in the control of their blood sugar levels. 

40 percent of the American adult population is considered prediabetic and that population is at great risk of developing type two diabetes within the next decade of their life.

This is a huge problem, 40 percent of the American adult population is considered prediabetic and that population is at great risk of developing type two diabetes within the next decade of their life. We took this group of prediabetic individuals and put them through our process. But now we asked the computer to generate a set of individualized good and bad diets for each of the participants. The good and the bad diets were the exact same number of calories, so we cannot blame the calories for what we saw. We asked each of the people to eat only their good diets for a week and only their bad diets for another week, while we extensively measured them. We were really surprised to find that during the week of consumption of the good personalized diet, even though in some cases it contained ice cream or candies, or did not contain fruit, which was found to be not good for a given person, all of these prediabetic people improved their blood sugar control considerably. In the bad diet week, all of them made their blood sugar control worse. And this told us that, the system is not only new and counterintuitive, but it can also be implemented in real life. Now, what are we doing following our studies is testing ourselves even further by comparing a large group of prediabetic individuals that are given this personalized diet for a whole year and compare them to another group of prediabetic individuals, which are given the gold standard American Heart Association diet for the same year. We monitor them very extensively looking for long-term changes in their health, such as improvement in their blood sugar control, their fatty liver and so on. This ambitious follow-up study is just about to finish and we will be very excited to study and to test the results that we will be getting. 

This is extremely interesting. So is your algorithm already available or still in the testing phase? 

Dr. Elinav: The algorithm and this entire project that we created and followed was done as a scientific project. It was born out of curiosity, and we did it academically at the Weizmann Institute of Science. Following the publication of the results, and multiple follow-up studies that we also published, the Weizmann Institute outsourced the set up discoveries to a spin-off company called DayTwo, which further developed it as means of making it available to wide populations. The populations that now use this technology no longer have to provide all the huge amount of data that we used during the academic study, because we now know what is important and what is redundant. But it is available currently, both in Israel and in the U.S.  To the best of my knowledge, they're going to be made available in other countries. 

To continue this article please look forward to The future of nutrition - part 2


  • We didn't just get six measurements of sugar as was done in the 70s, but 2,500 measurements, which created a huge database. 
  • Following the publication of the results, and multiple follow-up studies that we also published, the Weizmann Institute outsourced the set up discoveries to a spin-off company called DayTwo.