It’s been a busy month for ZOE’s science team.
Three of them – our scientific founder Tim Spector, Sarah Berry and Paul Franks – are presenting the first results from PREDICT 1 at the American Diabetes Association (ADA) and American Society for Nutrition (ASN) meetings in San Francisco and Baltimore.
PREDICT 1 is the world’s largest study of personal responses to different foods, carried out with Massachusetts General Hospital and King’s College London.
The biggest finding is that everyone has their own unique response to food – even identical twins like Hugo and Ross in the photo above. This means that everyone is different and there is no one right way to eat.
Why are we going to the trouble of running such a huge research project?
It’s all part of our mission to solve a seemingly simple but surprisingly hard question: what should I eat to be healthier?
Sound interesting? Sign up to our early access mailing list to be the first to know how you can find the foods your body loves.
Searching for clarity
There’s a lot of talk in the nutritional community and the wider world about what a ‘healthy diet’ looks like – much of it conflicting and confusing. AS well as the standard government guidelines there’s keto, paleo, low carb, low fat… the list goes on.
All this advice doesn’t seem to be working.
Around 160 million Americans are overweight or obese, and there are more than 100 million people with nutrition-related conditions like diabetes or heart disease in the US today. In total, one billion people around the world are thought to have metabolism-related health problems.
We wondered if we could use new tools and technology – like stick-on glucose monitors, at home blood collection, food logging apps and machine learning – to measure, understand and even predict what happens within the body when someone eats a meal, known as a nutritional response.
Millions of measurements
To find out, we teamed up with scientists at King’s College London, Massachusetts General Hospital and other leading academic institutions to recruit more than a thousand people (mainly twins) for PREDICT 1.
Here’s how it works.
Each person started the study with a full day in the clinic. We gave them a standardized meal and measured the responses of hundreds of factors in their blood – including sugar, insulin, fats and inflammation markers – at ten time points.
Back home, each participant ate a combination of standardized foods and their own choice of meals over a period of two weeks. As well as logging all their meals, they also took regular blood samples and wore a stick-on monitor to measure their blood sugar level every 15 minutes.
Throughout this period, participants used wearable devices to capture their sleep and exercise patterns, and logged their food intake, hunger, and medication in an app. They also collected multiple stool samples so we could look at the diversity of microbes in the gut, known as the microbiome, which we suspect may play an important role in explaining the variety of individual responses to food.
It all adds up to millions of datapoints:
- 32,000 standardized test muffins eaten
- 28,000 blood tests for fats and other markers
- More than 2,000,000 blood sugar readings
- 132,000 individual meals logged
- Nearly 140,000 hunger readings
- More than 11,000 happiness readings
All this data gives us a vastly more detailed picture of individual response to food than has ever been captured before. For example, we now have probably the largest database of blood fat and glucose responses in the world.
What we found
We were expecting to see differences in personal nutritional responses. However, we were surprised to discover such a wide variation in responses to the same foods, even between identical twins who share all their genes and much of their environment.
These graphs show what happened when we gave 1,000 healthy people exactly the same meal, then measured the levels of sugar (glucose) and fats (triglycerides) in their blood over six hours.
Every line in these graphs represents the results from a single person, showing the wide range of responses. Some participants have large, prolonged increases in blood sugar, which are linked to weight gain and diabetes if they happen regularly. Others have fat levels that peak and linger in the bloodstream hours after a meal, raising the risk of heart disease.
Most of the differences aren’t due to genes. Less than 50% of the variation between people’s sugar responses is due to their genetic makeup, with less than 30% for insulin and less than 20% for fat. Importantly, these lines aren’t just noise in the data: we found that individuals have a repeatable, predictable nutritional response.
We also found that having a bad response to fat couldn’t predict whether someone would be a good or bad responder to sugar – an important observation given that these are the two main sources of energy in our diets.
Strikingly, we also discovered that identical twins shared just 37% of their gut microbes, only slightly higher than the 35% shared between two unrelated people. The fact that our results show that the makeup of the microbiome is mostly independent from underlying genetics is great news, as it suggests that there are opportunities to change it in order to improve health.
What does it all mean?
Our findings tell us that personal differences in metabolism and nutritional response are due to a complex mix of factors including age, gender, ethnicity, sleep and activity, genetics, diet, environment and the microbiome.
The message is clear: when it comes to something as personal as food, we’re all different and there is no one right way to eat.
Yes, there are healthy eating messages that should apply to everyone – such as eating more fibre, and more plant-based and fewer ultra-processed foods – but one-size-fits-all guidelines are too simplistic. If you want to find the foods that work best with your metabolism, then you need to know your personal nutritional response.
Crunching the data
We’re now focusing on taking this data and training our increasingly sophisticated machine learning algorithms to predict how someone will respond to any meal. Next, we’re developing a home-based test and app that will enable anyone to measure their personal nutritional response and get tailored information about the kinds of foods that will work best for them.
Even at this early stage our machine learning predictions about an individual’s personal nutritional response are strongly correlated with real-life measurements. We believe they will get dramatically better as the number of people participating in our studies continues to grow, and as the scientists working with us tease out more of the complex interactions that explain our unique responses.
Content on this blog is licensed under a Creative Commons Attribution-NoDerivatives 4.0 International License. This means you’re free to reproduce it without any changes as long as you attribute ZOE and link back to the original post, but you can’t charge people to read it.