Perspective: Advancing Understanding of Population Nutrient–Health Relations via Metabolomics and Precision Phenotypes

ABSTRACT

Diet and lifestyle are vital to population health, but their true contribution is difficult to quantify using traditional methods. Nutrient–health relations are typically based on epidemiological associations that are assessed at the population level, traditionally using self-reported dietary and lifestyle data. Unfortunately, such measures are inherently inaccurate. New technologies such as metabolomics can measure nutritional and micronutrient profiles in body fluids, providing objective evaluation of nutritional status. A critical step toward accurate health prediction models would be the building of integrated repositories of nutritional measures combining subjective methods of reporting with objective metabolomics profiles and precise phenotypic data. Here we outline a roadmap to achieve this goal and discuss both the advantages and risks of this approach. We also highlight the uncertain associations between the complexity of high-dimensional data generated in ‘omics research (along with the public confusion this may engender) and the rapid adoption of ‘omics approaches by nutrition and health companies to develop nutritional products and services.

Introduction

Diet and lifestyle are associated with a number of health and disease outcomes (1). More importantly, lifestyle changes and medical nutrition therapies can help improve health outcomes, and prevent the onset of a number of diseases (2–4), even if challenging in certain subgroups of the population (5). Yet, with all of these nutritional data at hand, a number of methodological problems in nutrition research still need to be addressed if science is to truly deliver greater nutritional benefits (6, 7).

The 3 main approaches for assessing nutritional intake and/or status include: 1) subjective methods of dietary reporting, 2) objective measures of dietary intake, and 3) analytical profiling of nutrients in body fluids (8–18). First, dietary reporting has remained the predominant method for approximating populations’ eating behaviors and food intakes. However, traditional dietary data collection methods suffer from a number of limitations (8): they are subjective in nature, thereby limiting robustness and reproducibility (9, 11); they reflect food intake, not nutritional status; and they are insufficient to characterize the underlying mechanisms linking exposure to outcome. Second, objective biomarkers of dietary intake do not assess metabolic health status and are still narrowly focused on specific foods or food groups (12, 13). Notably, these objective biomarkers are typically validated against subjective methods of dietary intake, bringing back the questionable element of subjectivity. Third, most analytical panels attempting to measure nutritional status are not typically exhaustive (10, 14–18). In other words, profiling 1 or a few compounds does not necessarily reflect dynamic processes occurring within the human body.

A better understanding of nutritional metabolism could arise if nutritional studies were to transition from traditional measures of intake to more comprehensive measures of metabolic fingerprints within the blood, capturing dynamic exposures in synergy. However, going to the other extreme of measuring every possible nutrient may or may not provide more impactful nutritional knowledge. The happy medium would be to determine the relative importance of the relevant and nutritionally alterable compounds in research settings (e.g., water-soluble vitamins, lipid-soluble vitamins, phenolic compounds, SCFAs, amino acids, and one-carbon metabolism compounds), and then use only the most informative ones to enrich rather than complicate new public health and clinical models, while still gaining a good mechanistic understanding of nutritional processes (in academic research) (19, 20). Notably, developing, applying, and interpreting comprehensive analytical methods comes with a great deal of complexity. Some of the compounds measured may not be in their biologically active forms, whereas others may degrade easily upon storage or be difficult to chromatographically separate, among many other challenges (21).

The composition of a healthy metabolome is not well established (22). Metabolomics analysis of nutritional profiles in longitudinal cohorts could now enable potential associations to be identified between objective nutritional markers and health/disease outcomes. Both observational and interventional comparative studies are needed of nutritionally relevant nutrient profiles in individuals with healthy and unhealthy environmental lifestyle exposures (e.g., diets) as well as health outcomes. This would enable the definition of a healthy baseline nutrient profile, and the homeostatic range of differential responses to environmental cues and interventions.

By contrast, the notion of a healthy diet is generally well established (2, 23). This is largely due to the wealth of epidemiological data on dietary intakes, and complex associations with health and disease states (24). There has been a surge in the numbers of research and direct-to-consumer platforms (e.g., MyFitnessPal, Pinto) that have been developed to give a detailed breakdown of macro- and/or micronutrient composition of foods (25, 26). Yet, an in-depth knowledge of food composition does not necessarily inform on nutritional status in the human body. This is exemplified by a foundational experiment demonstrating that the postprandial change in blood amino acids is different to the amino acid profile of the foods consumed (27). Thus, dietary intake does not necessarily equate to nutritional status (Figure 1).

FIGURE 1.

Nutritional status is the endpoint of complex genetic, dietary, and metabolic interactions affecting health status.

Dietary Intake Does Not Equate to Nutritional Status

The entire farm-to-fork food production process affects micronutrient composition and bioavailability (28–30). Moreover, humans consume complex diets and the combination of food sources (plant compared with animal) within a meal affects micronutrient bioavailability and absorption (31–35). Nutritional status is further complicated by human genetic variants that affect the expression of genes involved in the uptake, conversion, metabolism, and transport of micronutrients, and thus contribute to interindividual differences in micronutrient profiles (36–46). The genetic component is further affected by physiological and pathological conditions [e.g., increased adiposity (47)] that also contribute to impair or alter nutritional status. Modifiers of human nutritional status are diverse and include baseline health and nutritional status (48), age (49), physical activity (50), sex (51, 52), bariatric surgery (53), environmental toxins (54, 55), and drug intake (56, 57). The human microbiome is also capable of producing micronutrients including, but not limited to, vitamin K-2 (menaquinones), folate, cobalamine, pyridoxine, riboflavin, betaine, and vitamin D (58–61). Collectively, the complex interrelations between intake, catabolism, anabolism, and excretion of nutrients illustrate the difficulty of measuring and understanding human nutritional status (Figure 1).

Why Quantify Human Nutritional Profiles?

Nutritional profiling could deepen our understanding of other health measures

Health and nutrition companies are increasingly designing commercial products that attempt to give individuals insights on how their bodies work (62, 63). Examples of these include Habit and Nutrino, which are used to give personalized dietary advice (62, 63). Micronutrient profiles, an important subset of overall nutritional status, are not typically included in such health and nutrition platforms (62–64). Adding micronutrient profiling to these commercial algorithms would enable the measurement of dietary responses that occur, and their functionality within the context of an individual's clinical profile and genotype.

Metabolomics has increasingly been used to objectively characterize nutritional profiles and diet-derived compounds (65). Nutritional metabolomics can fulfil different but complementary gaps in nutrition research, including 1) the development of robust and objective biomarkers of dietary intake to potentially strengthen and validate subjective methods of reporting; 2) the identification of metabolic responsiveness to nutritional interventions; and 3) the characterization of diet-derived metabolic profiles associated with phenotypes or disease outcomes (65). All could contribute to building a better understanding of “healthy” compared with “unhealthy” nutritional status, which may ultimately inform the development of novel products and services.

However, a significant amount of work still needs to be done in research settings before applying these techniques meaningfully in clinical and population settings (Figure 2). This includes 1) the development of assays that exhaustively quantify entire families of nutrients in different body fluids, rather than assaying a few selected compounds (10, 14–18); 2) the characterization of profiles for defined subsets of a population (i.e., subgroup analysis), as well as single individuals over time (i.e., N-of-1 analysis) (66, 67); 3) the assessment of metabolite–health/disease associations; 4) the incorporation of nutrient measures within prediction models of defined health outcomes; 5) the establishment of causality by comparing nutritional profiles between interventions and controls in intervention studies; 6) the characterization of nutrient profiles in different body fluids (14); as well as 7) the cross-correlation and mathematical modeling of nutrient repartition in different tissues from plasma inputs, for tissue-specific analyses (68, 69).

FIGURE 2.

Key actions for applying nutritional profiling in population research.

Anthropometric and clinical markers of health may not be sufficient for approximating current health status

Healthy/unhealthy anthropometric and clinical measures do not always correlate with a healthy/unhealthy nutritional status (50, 70, 71). For example, seemingly healthy athletes often suffer from micronutrient imbalances and deficiencies (50). Similarly, micronutrient insufficiencies have been demonstrated across the BMI spectrum, and in individuals with different metabolic conditions (72). Notably, obese individuals who underwent weight loss on micronutrient-sufficient low-calorie diets (providing >100% of the micronutrient DRIs) still exhibited dramatic decreases in blood micronutrient concentrations (70). Furthermore, an individual's suboptimal dietary intake or lack of physical activity may not be directly reflected in traditional clinical measures. For instance, measuring blood glucose concentrations in nondiabetic individuals who typically exhibit homeostatic glycemic responses, but may have very different lifestyles, may not reflect optimal health, but rather the absence of disease (73).

The initial intervention for metabolic dysfunction is usually weight management through dietary and lifestyle modifications. Popular weight loss diets typically focus on dietary macronutrient composition rather than micronutrient intakes, and thus confer risks of micronutrient inadequacies (74). Even when micronutrient intakes are factored into nutritional planning, the DRIs of micronutrients 1) were developed for healthy nondiseased populations, 2) attempt to prevent overt micronutrient deficiencies and not necessarily optimize health or performance, 3) have not all been revised over time although dietary habits and lifestyles are constantly changing, and 4) have often been developed on the basis of expert opinions rather than robust scientific evidence (70, 75, 76). Moreover, the effectiveness of lifestyle interventions is generally monitored by anthropometric changes, and/or measures of clinical markers associated with disease states (e.g., blood lipids, glucose, and glycated hemoglobin concentrations) (77). Micronutrient insufficiencies are associated with increased metabolic health risks (71), leading to a potential feed-forward cycle of ever-increasing impacts. Therefore, micronutrient profiling of individuals across the BMI spectrum and undergoing different lifestyle interventions could allow a potential tailoring of lifestyle interventions based on dynamic micronutrient changes over time.

Nutrient profiles might improve the power of prediction models of future health using machine learning and personalized approaches

Health risk calculators and predictive models of disease risk are widely used in routine clinical practice (78, 79). Examples of these include cardiometabolic risk calculators (e.g., the Framingham Risk Score, Systematic Coronary Risk Evaluation) (79) and cancer prediction models (e.g., algorithms diagnosing skin nodules) (78). Notably, one of the mechanisms that has been proposed to explain the protective effects of the diet on health outcomes includes, but is not restricted to, gene–nutrient interactions (e.g., epigenetics and gene regulation) (48, 80). Even though this dynamic nature of gene–environment interactions has been extensively reported in the scientific literature, currently, predictive models of disease risk are typically rooted on standard genetic and clinical measures and/or subjective self-reported data (64, 79, 81, 82). Current predictive models suffer from limitations that include their inability to segregate the “nondiseased” or “low-risk” category of people into different “health” categories, to better target “healthy” individuals with a higher predisposition to a certain outcome. The incorporation of metabolomics measures of nutritional status as variables in predictive health models could potentially strengthen their prognostic power.

Conflicting views have arisen regarding variable incorporation in health prediction models. Some have argued that implementation of robust public health strategies may far outweigh the value of precision prediction (19, 20). Although noting the inherent challenges, others have suggested that a systems biology approach will deliver great benefit from a new understanding of intricate interrelated biological mechanisms in their emergent complexity (83–85). Falling into either one of those extremes could be problematic: simple measures may be clinically relevant and important, but ignoring the comprehensiveness of ‘omics measures in health models may restrict our understanding of human biological dynamics.

Continuing to use current prediction models, while concurrently testing some carefully chosen ‘omics measurements, might be the best compromise (78). Notably, advances in machine learning approaches have allowed for a meaningful compilation of high-dimensional data, generated in large cohorts with millions of measurements, into accessible and workable recommendations for the individual (86–88). For instance, a cohort study (n = 800) identified highly interindividual postprandial glycemic responses to nearly 47,000 meals, even after the consumption of the same standardized meal. Thus, traditional methods that exclusively look at the glycemic index/load of foods at the population level may not be relevant to individuals (86). These differential responses are driven by a host of other factors including microbial composition, body composition, time since sleep, and genetic factors (86). Zeevi et al. (86) have deployed machine learning approaches to accurately predict this interindividuality in glycemic responses based on clinical and microbiome features, and further create and test personalized diets based on these predictions. The future of machine learning and high-dimensional data generation and compilation is potentially both exciting and fraught: “Siri, What Should I Eat?” (87).

Several nutrients and diet-derived metabolites have been profiled and proposed as associated with health and disease outcomes (89–92). Examples of these include BCAAs, water-soluble and lipid-soluble vitamins, SCFAs, trimethylamine-N-oxide, phenolic compounds or polyphenols (e.g., caffeic acid, ferulic acid, epigallocatechin gallate), and postbiotics (51, 89–96). Although some of these associations are controversial (97), they illustrate how objective measures of diet-derived or nutritionally alterable compounds can be placed at the intersection of the diet, the microbiota, genetics, and health (98). It may be time to move from asking what people eat to characterizing their (relative or absolute) metabolomics profiles (65).

'Omics in Nutrition Sciences

Nutritional profiling is a subcomponent of “omics” analyses

Mixed “omics” analyses consist of the exhaustive profiling of individuals at different biological levels (84). These biological levels include genes (genomics), non–DNA-based gene expression changes (epigenomics), genetic transcription (transcriptomics), the production of proteins (proteomics), metabolites (metabolomics), and biochemical regulations and interactions (interactomics), in addition to clinical and phenotypic traits (phenomics). 'Omics not only covers endogenous processes but also accounts—or should account—for exogenous exposures affecting our bodies (i.e., the “exposome”) (99). Exposomics typically encompasses environmental inputs such as food, lifestyle, toxins, and medication, among others. Contrary to the “endogenous” 'omics, exposomics are generally measured using subjective methods, such as self-reported medication, diet/physical activity diaries, and 24-h dietary recalls, as well as FFQs or physical activity questionnaires (100, 101). Therefore, adding objective measures of the exposome as well as profiling nutritional status will help build objective repositories and models of interactions between these environments.

Nutrient profiling is a targeted metabolomics approach that quantifies (hence reports in absolute concentrations) the concentrations of interrelated compounds involved in specific metabolic pathways within different body fluids (102). By contrast, untargeted metabolomics profiles the entire metabolic fingerprint of an individual in a discovery mode, reporting compound types/names and relative rather than absolute concentrations (103). Despite being traditionally perceived to be similar, these 'omics approaches enrich studies from 2 complementary angles, and both are needed for improved gene–nutrient–metabolism understanding. Untargeted metabolomics provides a global metabolic snapshot and is useful for hypothesis generation (102, 103). By contrast, targeted metabolomics (applied in nutrient profiling for instance) quantifies concentrations of compounds of interest and is useful in clinical settings. By providing a molecular fingerprint of populations, these big 'omics data can complement the clinical data sets that are used to feed diagnostic algorithms in clinical settings (78, 104). However, realistically there are major challenges in high-throughput 'omics on all levels of analyses, including differences in study designs, the lack of standardization of laboratory protocols, and difficulties in statistical, computational, and biological interpretation (83, 104–109). Some of these issues can be overcome by encouraging the exchange of robustly validated standard operating procedures for different types of 'omics analyses internationally.

Translation of 'omics from laboratory science to clinical and home diagnostics: does the future of 'omics lie outside the laboratory?

In a word, yes. Most portable medical monitoring devices that are currently in use have been developed for diseased populations [e.g., glucose monitoring devices for diabetic patients (110), heart monitoring devices for cardiac patients (111), and blood pressure monitoring devices for hypertensive patients (112)]. However, technological advances have also allowed the miniaturization of large laboratory instruments used in 'omics applications, thus rendering them available for use by the general population (113). An example is the MinION device for high-throughput sequencing (113, 114). This device can be plugged into a laptop computer to sequence a sample's genome, epigenome, or transcriptome in real time and outside a laboratory environment. Semiportable metabolomics devices have also been designed, such as The MassSpec Pen, which profiles tissues in real time to differentiate between cancerous and normal tissues during surgeries (114). The MassSpec Pen allows surgeons to exclusively remove the cancerous part of the affected tissue (114). Moreover, some recently developed electroacupuncture-based devices have claimed the ability to profile 30 vitamins and minerals using a digital pen linked to mobile applications (115). This pen has already raised around half a million US$ of venture capital over a few years, highlighting a market interest in the field. Future areas of portable MS quantification devices using scientifically robust technologies, such as microfluidics (116), must be considered and designed. Advances in sensor technologies and portable near-infrared spectrophotometry, used in plant and pharmaceutical research, could also potentially be adapted for human nutritional research (117–120). Such technologies might be useful for the continuous monitoring and release of nutrients in blood from sensor-vesicle type devices embedded within smart bandages, or synthetic skins (118, 121). In the future, these devices may lead to an improved understanding of human health through real-time monitoring and adjustment of individuals’ responses to different environmental and nutritional stimuli.

The application of machine learning, 'omics technologies, and nutritional metabolomics is rapidly becoming highly commercialized. However, although developing algorithms that calculate what needs to be eaten, how and when, might be the number one priority for health professionals and insurance companies, it may not be the case for the general population. Thus, personalizing diets based on metabolic responses and health benefits might not directly translate to improved health outcomes, no matter how good the technology or underlying data. Attention must also be paid to 1) public acceptance of or opposition to advanced technologies, which has previously been seen in advanced nutritional biotechnology (e.g., genetically modified crops) in low socioeconomic settings (122); and 2) adherence to the generated recommendations, a problem for many nutritional interventions (123). Although personalization has shown benefit in some cases, some evidence suggests that it is high levels of compliance to a diet, rather than the specific diet itself, that may contribute most to weight loss and positive outcomes (124). Thus, the science of behavior may be crucial to capitalizing on the science of micronutrients, and all approaches must be carefully evaluated in those for whom the recommendations are designed.

Technological advances in 'omics should not imply to users that scientists have a full understanding of all of the outcomes measured in 'omics studies. Many of the complexities of human biological processes are yet to be disentangled via 'omics profiling. Further, although these devices and platforms may provide a glimpse of the future, this should not come at the cost of the important social and cultural meaning of food and nutrition.

Concluding Remarks

Nutritional status is a key contributor to health status, and micronutrient profiling complements subjective methods assessing dietary intakes. Untargeted metabolomics profiling should be complemented with targeted metabolomics analyses, quantifying families of interrelated compounds, for implementation in clinical settings. Integrative analyses of different 'omics outputs, including genetic, clinical, and micronutrient data as a measure of nutritional status, may further enrich predictive modeling of health status, and ultimately the design of evidence-based personalized nutritional interventions. However, personalized diets should only be implemented if shown to translate into measurable improvements in health, which may depend on cultural acceptance and individual adherence as much as on the science. Further, the need to distinguish short-term from long-term health benefits will require monitoring over many years. Technological advances coupled with the increased consumer interest in health-related devices will continue to result in a rise in market need of scientifically robust devices commercializing 'omics analyses.

There remains a mismatch between the demands and needs of 3 parties: 1) scientists, who are still trying to make sense of the data generated from advanced 'omics technologies; 2) the general populace, who are both intrigued and overwhelmed by the breadth of (scientific or advertised) nutritional information; and 3) health and nutrition companies, who are trying to maximize their profit. There is still a substantial gap between advanced laboratory technologies used in nutritional studies and direct-to-consumer testing, which will ultimately be the endpoint of nutritional profiling. We are certainly not there yet.

Notes

Perspective articles allow authors to take a position on a topic of current major importance or controversy in the field of nutrition. As such, these articles could include statements based on author opinions or point of view. Opinions expressed in Perspective articles are those of the author and are not attributable to the funder(s) or the sponsor(s) or the publisher, Editor, or Editorial Board of Advances in Nutrition. Individuals with different positions on the topic of a Perspective are invited to submit their comments in the form of a Perspectives article or in a Letter to the Editor.

Supported by Ministry of Business, Innovation & Employment catalyst grant UOAX1611 (The New Zealand-Australia LifeCourse Collaboration on Genes, Environment, Nutrition and Obesity), the New Zealand International Doctoral Research Scholarship 2017 and the Liggins Institute doctoral research scholarship (to SA), Australian National Health and Medical Research Council (NHMRC) Senior Research Fellowship APP1064629 (to DB), and NHMRC Principal Research Fellowship 1160906 (to MW).

Author disclosures: SA, MW, RS, DB, and JO, no conflicts of interest. MK is the Editorial Director at Frontiers Media SA.