Between 2013 and 2014, 19 people were voluntarily locked in a clinic for days at a time—not once, but on four separate occasions. They were fed a different, strict diet on each of their three-day-long visits, and were forbidden to exercise. Computer access and visitation were allowed, so long as guests didn’t smuggle in snacks. Subjects turned over all their urine, from morning, afternoon, and night, to researchers.
These participants temporarily sacrificed their freedom to help dietician Gary Frost and colleagues at Imperial College London understand how eating habits influence the relative concentrations of metabolites excreted in urine, and thus how urine could serve as an indicator of a person’s diet, which in the team’s experiment ranged from healthy to gluttonous. Frost’s team anticipated that such metabolomics analyses would provide more-reliable data for nutritionists than the traditional tactic of asking free-roaming subjects what...
“Most people have a really bad memory of what they’re eating. . . . People will deny eating a dessert or forget that they ate a chocolate,” says David Wishart, a biochemist who works on metabolomics and nutrition at the University of Alberta but was not involved in Frost’s study. On the other hand, he adds, “blood and urine don’t lie.”
Blood and urine don’t lie.—David Wishart, University of Alberta
Indeed, Frost’s team was able to use the volunteers’ data to turn profiles of metabolites in urine into a single score, which they can now use to make inferences about the diets of people whose meals they didn’t control. Other researchers seeking objectivity in nutrition research are identifying metabolites that reveal if a person has consumed a specific food.
The approach is not yet perfected or widespread, and researchers are still working to define the metabolites linked to certain dietary qualities or foods. But with careful analyses, scientists are beginning to uncover nuanced information about people’s diets, such as how much milk and cheese they consume, or what kind of brew coffee drinkers drink. As the techniques improve and data on diet-metabolite correlations amass, researchers expect to standardize dietary epidemiology, which seeks to elucidate links between specific eating habits and disease risk. Some even see an opportunity to develop personalized nutrition recommendations to help people boost or maintain physical and mental health. Wishart, for example, is the chief informatics officer at a Vancouver-based company called Molecular You that uses metabolomics and other information to advise customers on eating habits to improve their health.
“There is a high expectation that [metabolomics] will play a leading role in deciphering the interactions between diet and health,” says Cristina Andrés-Lacueva, a nutrition researcher at the University of Barcelona.
Linking metabolites with specific foods
Before scientists can begin to link metabolites with health and disease, they must detail the relevant biomarkers for consumption of diverse foods. In 2019, Wishart and collaborators from Europe and New Zealand wrapped up a project that aimed to identify biomarkers for various foods and drinks, from Coca-Cola to chicken breast to Gruyère cheese. Potential biomarkers include compounds directly derived from those foods or fluctuations in the concentrations of human metabolites or metabolites produced by the gut microbiome. Called the Food Biomarker Alliance (FoodBAll), the study turned up several promising candidates.
For example, the team investigating blood biomarkers for dairy proposed the sugar alcohol galactitol as an indicator for consumption of cow’s milk, and the aromatic compound 3-phenyllactic acid as a signal for cheese ingestion. The FoodBAll collaborators also developed protocols for validating these novel biomarkers and created several online databases to fuel food metabolomics research. (See table below.)
Even beyond the FoodBAll group, once researchers started looking, they found metabolites that reveal striking details about dietary habits. For example, biochemist Augustin Scalbert, of the International Agency for Research on Cancer, is interested in coffee, which comes in various types. It’s also been linked to a range of health effects, so scientists want to know the best biomarkers for what kind, and how much, subjects consume. Scalbert and colleagues compared blood metabolites from 451 individuals from four countries: France, where people often drink espresso; Germany, where drip-filtered coffee is the norm; Greece, where boiled coffee is preferred; and Italy, where an espresso-like shot brewed by a percolator called a moka pot is popular. In France and Germany, the alkaloid trigonelline was the best marker for coffee consumption. But in Greece the best way to estimate someone’s coffee intake was quinic acid, Scalbert says, and in Italy, it was the amino acid derivative cyclo(isoleucyl-prolyl). The results suggest the ideal biomarkers might depend on the population under study.
From Dinner Plate to Dataset
To achieve greater objectivity in nutrition research, which has historically relied on self-reports of what subjects eat, scientists are turning to biomarkers in bodily fluids that reveal details about a person’s diet. Much of the work to this point has involved screens to identify novel markers for specific food items (or even for how those foods are prepared). In some cases, researchers have begun to use markers identified in these screens to correlate diet with health risks.
In some studies that aim to identify metabolites associated with certain foods or diets, scientists tightly control people’s intakes before analyzing their metabolites. More often, they ask subjects what they’ve been eating. People’s bodies will contain molecules from the foods they eat, as well as metabolites made from or in response to those foods, and even metabolites from their microbiota.
Most studies sample blood or urine, but stool, hair, or fingernails might also yield dietary clues. Mass spectrometry allows for highly sensitive analyses of these metabolites in any sample type, even picking up those found at low concentrations. Nuclear magnetic resonance (NMR) provides more reproducible results, but may miss rare molecules.
Results can reveal metabolites that are positively (red; example shown) or negatively (blue) associated with specific foods or correlate with the overall healthfulness of a diet.
Researchers link metabolites with risk of specific diseases
Diet has long been linked to cancer risk. But most studies have simply asked people what they eat, and then tracked later cancer diagnoses.
Like many nutrition researchers, epidemiologists at the American Cancer Society are now seeking specific markers for individual foods to obtain more-reliable data on diet-cancer links. A couple of years ago, Marji McCullough and Ying Wang analyzed 1,186 serum metabolites from 91 food groups and individual items, based on a study of 1,369 women who had filled out food-frequency questionnaires as part of the Cancer Prevention Study II. Correlations abounded: the scientists were able to connect 42 of those foods and food groups to 199 different metabolites, including some novel, not-yet-named biomarkers for coffee and for dark fish, a category that includes sardines and salmon. McCullough and her colleagues hope to validate and apply this profiling method to understand cancer risk in future studies. “The field is still young,” she says.
Scalbert agrees. “We are learning, little by little, to exploit this information and make the best use of it to understand the link between food intake and different foods . . . and different diets and the risk of cancer.”
He was interested in coffee because it is associated with a lower risk for liver cancer and its precursor, chronic liver disease. Liver cancer affects about 33,000 Americans each year, according to the CDC, but a regular coffee habit slashes cancer risk by up to 50 percent. Armed with their knowledge of coffee biomarkers, Scalbert and collaborators investigated banked blood samples collected from male smokers during a trial of nutritional supplements for the prevention of lung cancer in Finland in the 1980s. In this dataset, coffee-drinking was associated with higher blood concentrations of several compounds—including the neurotransmitter serotonin, glycerophospholipids that make up cell membranes, and trigonelline from the coffee beans themselves—and lower concentrations of tyrosine and bile acids.
Then, the researchers compared the data from trial participants who’d later been diagnosed with liver cancer or who died of liver disease before the end of 2012 with the data from volunteers in the same trial who had healthy livers. Those healthy controls tended to have higher levels of the coffee compounds and associated molecules, while tyrosine and bile acids were higher in those who went on to develop liver disease or cancer.
“This is a very nicely done and well-designed study,” says Wishart, who was not involved in that particular piece of research but has collaborated with Scalbert on other projects. He says the results fit with other studies suggesting that consuming coffee can diminish inflammation in the gut while promoting the growth of a beneficial microbiome—factors that then limit the damage caused by compounds such as bile acids and tyrosine.
WHAT’S IN YOUR FOOD?
Researchers involved in the FoodBAll project built several databases to facilitate research on food biomarkers.
FooDB lists information on food components including biochemical makeup, health effects, and color, taste and smell.
70,926 compounds in 797 different foods
Polyphenols are plant compounds with documented health benefits. Phenol-Explorer lists chemical details, which foods contain the compounds, and whether they appear in blood or urine upon consumption.
500 polyphenols found in more than 400 foods
PhytoHub lists polyphenols and other plant compounds, along with their metabolites produced by humans or other animals.
1,200 compounds in more than 350 foods
Food Compound Exchange
FoodComEx is a virtual library of food-derived compounds that different labs possess, so scientists can easily share them for study.
What metabolites can tell us about overall diets
Taking a more generalized view of eating habits, Frost and others are attempting to quantify the overall healthfulness of a diet. In the lock-in study, held at the UK’s National Institute for Health Research/Wellcome Trust Imperial Clinical Research Facility in London, he and his team fed their subjects four different meal plans: Diet 1 included foods considered healthy by the World Health Organization, including whole-wheat cereal, steamed salmon, and grapes. Diet 4 represented the opposite end of the spectrum, with sugar-coated cereal, fried pork sausages, and milk chocolate. Diets 2 and 3 fell in between those extremes.
Within just a few days, the researchers could detect the effects of these menus in the subjects’ metabolomic profiles. For people eating the nourishing Diet 1 meals, 19 metabolites appeared at higher concentrations than they did in the urine of people consuming junk food–heavy Diet 4. One of those metabolites, for example, was hippurate, an indicator of fruit and vegetable consumption. Conversely, nine metabolites were higher in people feasting on unhealthy Diet 4 than those on Diet 1. Carnitine, a biomarker for red meat, was one.
To incorporate these individual signals into a broader assessment, data scientist Joram Posma applied machine learning. He used the biomarker levels from participants on Diet 1 or Diet 4 to train a computer algorithm to predict a person’s diet quality by profiling their urine. Then, the team tested its model on the data from the intermediate diets. Sure enough, the model correctly identified Diet 2 eaters as having relatively healthy metabolomes, and Diet 3 diners as skewing closer to the unhealthy patterns. The team also validated the model in other cohorts from Denmark and the UK. Those with healthier reported diets had metabolomes more similar to those of locked-in subjects who ate Diet 1.
“This paper is interesting as it combines an intervention study with observational studies in two different cohorts,” says Scalbert, who was not involved in the research. “This has often been done for specific foods, but more rarely for whole diets.”
Based on a metabolite profile, the team can now calculate a single Dietary Metabotype Score (DMS) to represent how healthy a person’s eating habits are.2 “We can say where you are on the spectrum,” says Frost.
Crucially, the scores are objective, with no interference from faulty recollections by eaters, adds Isabel Garcia-Perez, a chemist on the project who is now putting the algorithm into practice. In a trial of clients seeing dieticians collaborating on the project, Garcia-Perez will use dietary scores to give providers an idea of a person’s eating habits before they meet, and to determine how closely clients follow, or don’t follow, their prescribed eating plans. She predicts that clients will be more motivated to keep up with the recommended diets if their metabolome provides frequent, reliable feedback on their adherence to the instructions.
McCullough and colleagues are seeking easy markers for broad dietary patterns in blood. They specifically looked for metabolites that would likely indicate scores on four different diet measures: the alternate Mediterranean diet score, the alternate Healthy Eating Index, the Dietary Approaches to Stop Hypertension, and the Healthy Eating Index. Top predictors for high scores on these indices, reflecting a healthy diet, included markers for fish consumption such as the omega-3 fatty acid DHA, and the vitamin carotene from fruits and vegetables.
Ultimately, says McCullough, this kind of research may lead to affordable blood tests that indicate a person’s true dietary patterns—an objective measure that clinicians could use to assess disease risk and advise patients. “That’s far in the future,” she says, “but there’s a lot of potential.”
While nutritionists agree on the basic ingredients of a healthy meal plan—lots of fruits and veggies, for example—the general guidelines don’t speak to differences between individual diners.
“People respond to foods differently,” points out biochemist David Wishart of the University of Alberta. “Particularly with vitamins, there’s a fair bit of variability.” Vitamin processing may vary with age, genetic makeup, and physiology—for example, if someone is obese, fat-soluble vitamins might be stored in fat tissue instead of circulated through the body, he says. Thus, a glass of vitamin C–rich orange juice might in fact provide different vitamin C effects for different drinkers.
That’s why he and others are developing personalized nutrition plans that recommend specific foods, supplements, or even exercises to improve health for individual clients. Wishart is the chief informatics officer for Molecular You, a company in Canada aiming to boost people’s health with detailed prescriptions based on metabolomics and other information. But at this point, some experts caution, it’s unclear how personalizing diets can improve the health of individuals, aside from populations with specific nutritional needs or allergies.
In 2016, researchers from the Weizmann Institute of Science in Israel launched DayTwo, a company focused on blood-sugar control for people with type 2 diabetes and prediabetes. In diabetics, blood glucose spikes after eating can boost their risk for cardiovascular disease.
DayTwo’s approach is based on a study published by its scientific cofounders in 2015. The scientists had monitored glucose levels continually for a week in 800 people as they collectively consumed nearly 47,000 meals (Cell, 163:1079–94, 2015). Using machine learning to incorporate glucose patterns, dietary habits, and other factors, the team developed an algorithm to predict glucose levels after consuming specific meals.
The team then offered 12 new subjects the opportunity to receive dietary recommendations informed by its computer model. Participants worked with dieticians to devise meals that matched their preferences, then underwent glucose-level monitoring for a week while eating those meals to provide input data for the algorithm. The team then used the algorithm to predict “good” diets that avoided spikes, or “bad” diets that created them, for each individual. The subjects followed each of those diets for a week. Sure enough, blood-sugar spikes were higher in 10 people when they were on the “bad” diet. The algorithm performed as well as human experts in identifying whether meals would be “good” or “bad” for an individual, but outpaced the experts in that it could do the same for new meals, without any data on that person’s previous glucose responses to those foods.
Molecular You launched its own metabolomics-based nutrition services to doctors in 2018, and to consumers last year. Company scientists delineated “safe” blood or urine concentrations for a slew of metabolites, based on the scientific literature and other sources. (The US Food and Drug Administration offers guidelines on safe levels of various compounds in foods, but not in bodily fluids.) Molecular You nutritionists combine company algorithms with their own experience to help customers reach those ideal ranges.
A common, though not universal, recommendation is to eat less red meat, says company CEO Robert Fraser. Other tips are more individualized. For example, if people don’t have enough vitamin B, the company might recommend supplements. “There’s a lot of variety in what people need,” says Fraser. “There’s usually something very personal for each person.” In an as-yet unpublished study, the company analyzed about 150 customers who followed their Molecular You plan for 100 days. Those who stuck to the recommendation nudged all of their metabolite values into that “safe range,” Fraser says.
Not everyone is ready to buy in. “I’m not a strong believer of personalized nutrition,” says Augustin Scalbert, a biochemist at the international Agency for Research on Cancer in France. Diet is just so complex, he explains, it will be difficult to prove any interventions are working.
Amber Dance is a freelance science journalist living in the Los Angeles area. Read her work or reach out at AmberLDance.com.