Traditional dietary assessment techniques

For more than 100 years, nutritional research has played an important role for our understanding of diet–health relations in the human body. Nutrition researchers have contributed substantially to meaningful ways of improving health of individuals as well as populations. Valid and reproducible dietary assessment has been the cornerstone of this work. Food diaries, 24 h recalls and FFQ are commonly used self-report dietary assessment methods in human studies of nutrition in relation to health⁽ Reference Willet ^{1 )}. Multiple days, i.e. 7 d, weighed food records, the so-called gold standard dietary assessment method, are often used in controlled dietary studies⁽ Reference Cameron and van Staveren ^{2 )}. However, dietary records are very time consuming and labour intensive for both participants and researchers, and they are not very accurate⁽ Reference Nes, Frost Andersen and Solvoll ^{3 ,} Reference Yuan, Spiegelman and Rimm ^{4 )}. Therefore, they are not commonly used in large studies. As an alternative, 24 h recalls, i.e. a structured interview aiming to collect detailed information about all food items consumed during the preceding 24 h, are more frequently used in research settings. A major drawback of 24 h recalls is that 2–3 recalls are often not sufficient to capture the day-to-day variation of a variety of nutrients and foods such as vitamin A, vitamin C, cholesterol and fish⁽ Reference Willet ^{1 )}. Therefore, the use of 24 h recalls is often considered to be too time- and labour-intensive for use in large studies⁽ Reference Beaton, Milner and McGuire ^{5 )}. For larger studies, self-administered FFQ are most often the method of choice and are generally used to evaluate food consumption during the past 1–12 months. In contrast to food records and recalls, FFQ require an intensive preparation before the method is applied in actual studies, but the administration and processing of a validated FFQ can be done rather efficiently for several thousand participants at once. However, the large supply of available foods, inaccurate estimation of portion sizes, socially desirable answers and errors in food composition tables also make FFQ prone to measurement error⁽ Reference Cade, Thompson and Burley ^{6 –} Reference Johansson, Solvoll and Bjorneboe ^{9 )}. A common concern with self-reported dietary intake is the objectivity and accuracy of self-reported data.

Nutritional biomarkers

The limitations of the present dietary measurement tools have motivated many researchers to search for nutritional biomarkers as a complementary or alternative measure of dietary intake. Objective biological food intake and nutrient intake biomarkers may provide valuable information beyond self-reported food intake data and are particularly valuable when food composition data are not available or limited. Unfortunately, there are still very few well-validated nutritional intake biomarkers, with the exception of urinary nitrogen for protein, potassium and sodium⁽ Reference Jenab, Slimani and Bictash ^{10 )}; carotenoids for vegetables and fruits⁽ Reference Brevik, Andersen and Karlsen ^{11 ,} Reference Al-Delaimy, Ferrari and Slimani ^{12 )}; and n-3 fatty acids for oily fish⁽ Reference Saadatian-Elahi, Slimani and Chajes ^{13 )}. These examples of the well-validated food intake biomarkers are all natural food constituents. Researchers also explore the value of nutritional biomarkers reflecting food additives, nutritional biomarkers that are the result of the digestion or absorption of foods, and endogenous metabolites that are affected by the consumption of certain foods and as such may act as nutritional biomarkers. Previous research has for instance highlighted significant associations between fish consumption and biomarkers of mercury, arsenic and polychlorinated biphenyl⁽ Reference Miklavcic, Casetta and Snoj Tratnik ^{14 ,} Reference Philibert, Schwartz and Mergler ^{15 )}. Moreover, the Dutch National Institute for Public Health and the Environment and the University of Ghent are presently focusing part of their research on mycotoxins, a food contaminant where dietary exposure represents the main exposure⁽ Reference Marin, Ramos and Cano-Sancho ^{16 )}. During the past 3 years, two draft calculators have been developed to study associations between dietary mycotoxin intakes and urinary mycotoxin concentrations. Furthermore, these two research centres are conducting an intervention study in which volunteers are exposed to a single dose of a rapidly eliminated mycotoxin (i.e. deoxynivalenol), after which urine is collected quantitatively at different time intervals up to 24 h after administration.

New metabolomics techniques

Recent developments in analytical chemistry suggest that much more useful food intake biomarkers can and should be identified. New metabolomics techniques provide a unique opportunity. Using metabolomics techniques, it is possible to measure up to thousands of metabolites at once providing valuable information on the food metabolome (i.e. essential and non-essential chemicals derived from foods after digestion and subsequent metabolism by the tissues and the microbiota)⁽ Reference Scalbert, Brennan and Manach ^{17 )}, using plasma⁽ Reference Lawton, Berger and Mitchell ^{18 )}, serum⁽ Reference Psychogios, Hau and Peng ^{19 )}, erythrocytes and leucocytes⁽ Reference Norheim, Gjelstad and Hjorth ^{20 )}, urine⁽ Reference Bouatra, Aziat and Mandal ^{21 )}, saliva⁽ Reference Dame, Aziat and Mandal ^{22 )}, faeces⁽ Reference Deda, Chatziioannou and Fasoula ^{23 )}, cerebrospinal fluid⁽ Reference Wishart, Lewis and Morrissey ^{24 )}, as well as hair⁽ Reference Sulek, Han and Villas-Boas ^{25 )}. Metabolomics techniques do not only allow identification of numerous biomarkers at once, but also to conduct studies where we may account more precisely for food–food interactions. Moreover, these new techniques provide the opportunity to more easily use combinations of biomarkers to predict food intakes⁽ Reference Souverein, de Vries and Freese ^{26 )}. Commonly used platforms for food metabolomics analyses are NMR spectroscopy⁽ Reference Pan and Raftery ^{27 )}, liquid chromatography–MS and GC–MS⁽ Reference Kanani, Chrysanthopoulos and Klapa ^{28 ,} Reference Egert, Weinert and Kulling ^{29 )}. Advantages of NMR include its high reproducibility, short sample preparation time and small interlaboratory variability⁽ Reference Dettmer, Aronov and Hammock ^{30 ,} Reference Laghi, Picone and Capozzi ^{31 )}. Conversely, NMR only allows identification of a limited number of metabolites as compared to liquid chromatography–MS and GC–MS⁽ Reference Dettmer, Aronov and Hammock ^{30 ,} Reference Martin, Dumas and Wang ^{32 )}. MS techniques allow identification of many more metabolites, but tissue samples require more preparation and the overall throughput is lower⁽ Reference Dettmer, Aronov and Hammock ^{30 )}. Given the pros and cons of the different platforms, a multiplatform approach is preferred to identify a broad range of metabolites⁽ Reference Bub, Kriebel and Dorr ^{33 ,} Reference Scalbert, Brennan and Fiehn ^{34 )}.

Between- and within-person variability in food intake biomarkers

During identification of novel food intake biomarkers, one needs to be aware that metabolite concentrations may be influenced by genetic factors, different rates of digestion, absorption, metabolism, and excretion⁽ Reference Brown, Ferruzzi and Nguyen ^{35 )}, smoking⁽ Reference Brude, Finstad and Seljeflot ^{36 )}, body composition⁽ Reference Haugen and Drevon ^{37 )}, lifestyle⁽ Reference Vinknes, Elshorbagy and Nurk ^{38 )}, drug use⁽ Reference van Orten-Luiten, Janse and Dhonukshe-Rutten ^{39 )} and geographical location⁽ Reference Walsh, McLoughlin and Roche ^{40 )}. To illustrate, random plasma ¹³C concentrations have been shown to be highly correlated with cane sugar/high fructose maize syrup consumption of the preceding meal (R ² = 0·90)⁽ Reference Cook, Alvig and Liu ^{41 )}. However, these concentrations only reflect very recent intakes that are probably not detected when studying fasting concentrations. Additionally, although red meat consumption has been significantly associated with creatinine⁽ Reference Cross, Major and Sinha ^{42 )}, creatinine is also a product of muscle catabolism. Red wine consumption has been associated with various resveratrol metabolites⁽ Reference Rotches-Ribalta, Andres-Lacueva and Estruch ^{43 )}. However, also in this example, between-person variability should be considered due to for instance the production of resveratrol metabolites by colonic microflora⁽ Reference Manach, Scalbert and Morand ^{44 )}. Ideally, a robust food intake biomarker is not markedly influenced by such factors, but in any case such aspects need to be taken into account in the validation process of food intake biomarkers. However, a general validation system for food intake biomarkers is still lacking and better guidance for biomarker validation is needed. During the past years, a system to score food intake biomarker quality and validity has been developed, allowing researchers to evaluate potentially interesting food intake biomarkers for standard analytical quality control along with criteria related to biomarker kinetics (dose–response, time–response), metabolic and other host factor effects, food matrices, and specificity for the actual foods (Q Gao, G Praticò, A Scalbert et al., unpublished results). In addition to the aforementioned effort, the University Medical Centre Groningen and Dutch National Institute for Public Health and the Environment are investigating to what extent food intake biomarkers are biased by within-person variation, and if so, whether correction for within-person variation improves the estimate of the proportion of people with less than adequate or excessive intakes.

The Food Biomarkers Alliance

It is clear that food metabolomics is a complex discipline requiring the expertise and facilities in nutrition, metabolomics, epidemiology, clinical science, analytical chemistry, molecular biology, food sciences, bioinformatics and statistics. Therefore, a large group of experts in food metabolomics decided to join forces in the Food Biomarkers Alliance (FoodBAll), which is a Joint Project Initiative ‘A Healthy Diet for a Healthy Life’ (JPI-HDHL). This is a nationally funded collaboration between twenty-four partners from thirteen countries (Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Norway, Spain, Sweden, Switzerland and the Netherlands), aiming to identify and validate food intake biomarkers of commonly consumed foods in Europe that can improve food intake assessments. FoodBAll started its funded activities in December 2014 and will continue for 3 years up to mid-2018 (http://foodmetabolome.org/)^{( 45 )}. FoodBAll aims to systematically explore and validate food intake biomarkers to provide a better assessment of the food intake in different European regions. In order to identify and validate new food intake biomarkers, the FoodBAll team decided to focus on the following aspects of food intake biomarker research: (1) discover novel food intake biomarker by means of extensive systematic literature reviews, observational studies and acute intervention studies; (2) develop a validation scoring system for food intake biomarkers and its application to new biomarkers; (3) explore and validate alternative/less-invasive food intake biomarker sampling techniques; (4) develop tools for identification of food intake biomarkers and online resources to facilitate sharing of food intake biomarker data and resources with the (scientific) community; (5) explore biological health effects of food intake biomarkers; (6) describe the value of food intake biomarkers for authorities and stakeholders. In order to align all the work from many different disciplines, a food intake biomarker was defined as being a specific measurement in a biological specimen accurately reflecting the intake, selective and dose-dependent, of a food constituent or food. In contrast to a food intake biomarker, a nutritional biomarker relates to nutrients and can be any biological indicator of nutritional status linked either to intake, metabolism or effect of a nutrient.

Present actions on novel food intake biomarker discovery

Application of new food metabolomics techniques in (acute) intervention studies

Controlled acute dietary intervention studies are assumed to be less sensitive to confounding by correlated foods and therefore provide an important additional source of information in the field of food intake biomarker discovery. Gibbons and Brennan recently published an extensive overview of intervention studies using a metabolomics approach⁽ Reference Gibbons and Brennan ^{50 )}. In addition, the same research group recently successfully predicted citrus intakes in the Irish National Adult Nutrition Survey by urinary proline betaine using calibration curves obtained from an acute intervention study⁽ Reference Gibbons, Michielsen and Rundle ^{51 )}. A German research group also just published a randomised crossover intervention study on the effect of dietary non-fermentable and fermentable fibre (i.e. cellulose and inulin), and propionate (i.e. major product of the fermentation of fibres) on odd-chain fatty acid concentrations, showing significant effects of inulin and propionate, but not cellulose⁽ Reference Weitkunat, Schumann and Nickel ^{52 )}. To contribute to this research area, seven study centres are presently conducting well-defined standardised short-term intervention studies. All study centres follow the same design, exploring potential food intake biomarkers of fourteen foods, including sugar-sweetened beverage, apple, tomato, banana, milk, cheese, bread, meat/meat products, red meat and white meat, potato, carrot, peas, lentils, beans and chickpeas (Fig. 1). Participants are exposed to a test meal after which urine and blood samples are collected over a time-course up to 24 h with the option to continue to 48 h. Collected samples are analysed by a multiplatform approach, including untargeted liquid chromatography–MS, GC–MS and NMR. The first results of these studies have just been published, which pointed towards lactose, galactose, and galactonate as potential biomarkers of milk intake, urinary 3-phenyllactic acid of cheese intake, and pinitol and trigonelline of soya drinks⁽ Reference Munger, Trimigno and Picone ^{53 )}. Moreover, FoodBAll researchers are conducting metabolomics analyses using previously collected data from intervention studies. FoodBAll has access to fifty-four human intervention studies with biobanks of more than 14 000 men and women from age 8 to 95 years, including up to twenty different foods (e.g. dairy, meat, apples, cocoa, Nordic diet) and more than fifteen different nutrients or compounds (e.g. vitamin C, vitamin B₁₂ and polyphenols). Although intervention studies are assumed to be less sensitive to detect food–metabolite associations confounded by other correlated foods, the highly controlled character of these studies may result in identification of metabolites that are not specific enough for the food under investigation when explored in free-living human subjects (i.e. population-based studies)⁽ Reference Scalbert, Brennan and Manach ^{17 )}.

Fig. 1.

(Colour online) Design of the randomised, controlled, crossover studies as conducted as part of the Food Biomarkers Alliance (FoodBAll) project (adapted from Munger et al. ⁽ Reference Munger, Trimigno and Picone ^{53 )}). Test products are administered in random order and urine samples are collected before and after ingestion of test product during defined intervals (yellow blocks) up to 24 h; 6 and 24 h pools are later prepared in the laboratory.

Bridging dietary biomarkers to health pathways

In order to strengthen evidence for the application of intake/health efficacy biomarkers, future research should aim to relate food intake biomarkers to diet-related disease risk by (1) taking a pathway- and network-based bioinformatics approach and (2) building on existing transcriptomic and metabolomics data⁽ Reference Afman, Milenkovic and Roche ^{62 )}. O'Gorman et al. recently applied this approach in the Metabolic Challenge Study and identified twenty-two lipid biomarkers that were associated with total dietary fat intake⁽ Reference O'Gorman, Gibbons and Ryan ^{63 )}. Five of these twenty-two lipid biomarkers were significantly elevated in participants with increased risk of the metabolic syndrome⁽ Reference O'Gorman, Gibbons and Ryan ^{63 )}. Wittenbecher et al. also applied a metabolomics approach and identified six out of twenty-one potential biomarkers for red meat consumption to be associated with type 2 diabetes⁽ Reference Wittenbecher, Muhlenbruch and Kroger ^{64 )}. A Spanish group observed urolithin A glucuronide as the most discriminant marker of nuts consumption, which in turn showed a significant inverse correlation with abdominal overweight and glycaemia⁽ Reference Mora-Cubillos, Tulipani and Garcia-Aloy ^{65 )}. The FoodBAll consortium is presently conducting additional studies on the relevance and value of transcriptomics and metabolomics data patterns relating dietary intake with health.

Online (food) metabolomics databases

Although many food compounds can be found in human bio specimens and detected in metabolomic profiles, the interest in food metabolomics is relatively new, and annotation of the food metabolome is still very limited as compared to annotation of the endogenous metabolome. Thus, identification and validation of potentially new food intake biomarkers using metabolomics techniques requires knowledge of presently available electronic resources, such as food composition, compound and biomarker databases, libraries of spectra, software tools and various online tools for food metabolome profiles annotation and interpretation. A new public web portal developed by the FoodBAll consortium has been launched in 2017 that provides descriptions, links and tutorials for more than fifty selected databases, spectral libraries, software programmes, and online tools useful for food metabolome profiles annotation and food intake biomarker discovery^{( 45 )}. One of the major online resources is the human metabolome database, which contains detailed data about small molecule metabolites identified in the human tissues⁽ Reference Wishart, Jewison and Guo ^{66 )}. The most comprehensive metabolome database specifically focusing on the food compounds is FooDB (http://foodb.ca/), covering more than 25 000 native food compounds, food additives and man-made compounds also formed during food processing with their dietary sources (>900 foods)⁽ Reference Wishart ^{67 )}. However, online databases are still incomplete, especially regarding the human metabolites of all non-nutrients present in foods. Initiatives are ongoing to manually collect information available in the literature and to add in silico predicted metabolites of all food components. PhytoHub is a compound database focused on food phytochemicals and their human metabolites manually extracted from the literature, initially built by Institut national de la recherche agronomique for nutrimetabolomics studies^{( 68 )}. Besides databases useful for annotation in metabolomics, International Agency for Research on Cancer (France) and University of Alberta developed the Exposome-Explorer database on biomarkers of environmental exposure (http://exposome-explorer.iarc.fr/) in order to compare the performance between biomarkers and their fields of application⁽ Reference Neveu, Moussy and Rouaix ^{46 )}. Exposome-Explorer contains data on many attributes related to 142 dietary biomarkers extracted, such as type of exposure, biomarker concentrations in various human body fluids or tissue and associated analytical techniques, biomarker reproducibility over time, population characteristics, study design and with references to publications. FooDB, PhytoHub and Exposome-Explorer are being further developed within the FoodBall project to better meet all needs for the discovery and validation of new nutritional biomarkers. The FoodComEx ‘Food Compound Exchange’ may be another valuable online source for metabolomics researchers. FoodComEx is a chemical library recently developed by FoodBAll partners that aims to facilitate the sharing of standards of food compounds, their metabolites and other reference materials^{( 69 )}. These standards are essential for the validation or identification of candidate biomarkers in metabolomics studies. An interface has been developed that facilitates the exchange between researchers interested in using a specific compound and researchers owning that specific compound. Presently, about forty rare food-derived metabolites and 1005 commercial compounds are included in this database. The database will be further extended by attracting more contributors and as such more compounds. Furthermore, additional compounds will be produced by using in vitro (microbial fermentation, liver microsomes and isolated enzymes) and in vivo (rodents) systems to ‘metabolise’ food chemicals.