Metabotyping and its role in nutrition research

Abstract

Personalised nutrition is at its simplest form the delivery of dietary advice at an individual level. Incorporating response to different diets has resulted in the concept of precision nutrition. Harnessing the metabolic phenotype to identify subgroups of individuals that respond differentially to dietary interventions is becoming a reality. More specifically, the classification of individuals in subgroups according to their metabolic profile is defined as metabotyping and this approach has been employed to successfully identify differential response to dietary interventions. Furthermore, the approach has been expanded to develop a framework for the delivery of targeted nutrition. The present review examines the application of the metabotype approach in nutrition research with a focus on developing personalised nutrition. Application of metabotyping in longitudinal studies demonstrates that metabotypes can be associated with cardiometabolic risk factors and diet-related diseases while application in interventions can identify metabotypes with differential responses. In general, there is strong evidence that metabolic phenotyping is a promising strategy to identify groups at risk and to potentially improve health promotion at a population level. Future work should verify if targeted nutrition can change behaviours and have an impact on health outcomes.

Introduction

Poor diet quality is a major contributor to chronic diseases such as type 2 diabetes, CVD and various cancers^(1,2) . Despite the well-known association between dietary patterns and diseases, interventions to change dietary habits have had a limited impact on wellbeing and public health outcomes^(3,4) . In recent years, the diverse inter-individual responses to interventions have become apparent and support the need for the development of strategies that are based upon the delivery of advice to the individual^(5–9). Concomitant with this, different strategies have emerged for delivering advice taking personal characteristics into account. Furthermore, studies have demonstrated that personalisation of dietary advice is more effective in promoting improvements in the dietary habits of individuals compared with the general healthy eating advice^(10–12).

Metabolomics is the study of small molecules in biological samples and is a powerful tool in the characterisation of individuals^(13,14) . The set of metabolites in the human body, termed the metabolome, is the product of metabolic reactions influenced by endogenous, lifestyle and environmental factors^(15,16) . Applications of metabolomics in nutrition research have expanded in recent years and it has the potential to contribute to the delivery of personalised nutrition⁽¹⁷⁾. Metabotypes are defined as groups of similar individuals based on combinations of specific metabolites. Thus, individuals within a metabotype have similar metabolic profiles and those in different metabotypes have different profiles^(17,18) (Fig. 1). Metabotypes are often defined using cluster analysis, such as k-means analysis and hierarchical cluster analysis⁽¹⁸⁾. Applications of metabotypes has identified differential response to interventions and have the potential of identifying optimal treatment strategies for individuals. For example, using serum metabolites Palau-Rodriguez et al.⁽¹⁹⁾ identified two subgroups with different degrees of improvement in insulin resistance, total cholesterol (TC), HDL-cholesterol (HDL-C) and uric acid following bariatric surgery. Importantly, the metabolic changes in each cluster were independent of the baseline anthropometric/clinical parameters of the patients and the magnitude of weight loss. Another example identified metabotypes with different lipid responses to fenofibrate⁽²⁰⁾. Similarly, in the field of nutrition science there are several examples of applications of metabotypes in healthy and subjects with chronic diseases for determining metabolically homogeneous subgroups with differential responses to dietary interventions⁽¹⁸⁾. However, the applications are not limited to intervention studies, with the metabotyping approach being developed for the delivery of targeted nutrition^(21,22) . Given the rapid growth of this area, the objective is to review the research conducted on metabotypes related to nutrition research and to identify gaps where further work is needed.

Fig. 1. Overview of the concept of metabotyping for the delivery of personalised nutrition. Intrinsic and extrinsic factors influence the metabolic phenotype of individuals. Groups of individuals with similar metabolic phenotypes are termed metabotypes.

Metabolic phenotyping of longitudinal data to examine associations with cardiometabolic risk factors and diet-related diseases

Longitudinal studies are important tools in the epidemiological setting to investigate the aetiology of a disorder and indicate risk factors or population groups that may be targeted as part of prevention strategies. In fact, within the metabolic phenotype approach, longitudinal studies offer the possibility to study subgroups of individuals (metabotypes) over a period of time and the potential to identify those at higher risk of disease development. A summary of studies examining longitudinal associations of metabotypes with cardiometabolic risk factors and diet-related diseases is presented in Table 1.

Table 1. Summary of studies examining longitudinal associations of metabotypes with cardiometabolic risk factors and diet-related diseases

SBP, systolic blood pressure; DBP, diastolic blood pressure; HDL-C, HDL-cholesterol; CHOP, Childhood Obesity Project; PC, phosphatidylcholines; IGF-1, insulin-like growth factor 1; IGF-BP3, insulin-like growth factor-binding protein 3; IGF-BP2, insulin-like growth factor-binding protein 2.

In order to identify risk profiles for the emergence of the metabolic syndrome (MetS), Ventura et al.⁽²³⁾ assessed a non-clinical sample of healthy non-Hispanic white girls (n 154) in a retrospective analysis with follow-up performed every 2 years from age 5 to 13 years old. Six risk factors for the MetS (waist circumference, systolic blood pressure, diastolic blood pressure, HDL-C, TAG and blood glucose) were used in cluster analysis to determine metabotypes at age 13 years. At age 5 years, the higher MetS risk group had the highest BMI relative to the other groups. Across childhood, both the higher MetS risk and the hypertension risk groups had significantly greater increases in weight and fat mass, while the higher MetS risk group had the highest daily sweetened beverage intake. Findings from this study support the role of metabotypes for identifying individuals at higher risk who could be targeted by clinicians as part of preventive healthcare.

Application of metabotypes to baseline data in longitudinal studies can be very useful in defining at-risk groups which could be targeted for the prevention of undesirable health outcomes. The European Childhood Obesity Project (CHOP), using a Bayesian agglomerative clustering method on twenty-one plasma amino acids and 146 polar lipids, classified healthy infants (n 154) of 6 months of age into twenty metabotypes in order to predict later obesity risk⁽²⁴⁾. Only the four biggest clusters (n ≥ 14) were analysed and at the baseline cluster 3 had the lowest weight, height, free insulin-like growth factor (IGF)-1 and IGF-binding protein 3, and the highest IGF-binding protein 2. The BMI z-score at 6 years of age tended to differ (unadjusted P = 0·07) among clusters, with cluster 3 presenting the highest median and largest proportion of overweight/obese children. These results support the concept that even very young individuals can be clustered according to their inter-individual differences so that the clusters provide insight into later development and health and opportunities for developing more targeted and personalised intervention strategies.

Another notable example employing metabotypes in a prospective cohort is the KORA F4 study in which 1729 adults aged 32 to 77 years were clustered based on BMI and thirty-three biochemical markers⁽²⁵⁾. For each of the three metabotypes identified, the current disease prevalence and the incidence in the follow-up cohort 7 years later was determined. The ‘high-risk’ cluster showed the most unfavourable biomarker profile with the highest BMI and prevalence of cardiometabolic diseases at the baseline as well as the highest incidence of hypertension, type 2 diabetes, hyperuricaemia/gout, dyslipidaemia, all metabolic diseases and all CVD together. This study provides strong evidence that metabotyping is a robust approach for identifying groups of individuals that could be targeted for prevention strategies.

Overall, the derivation of metabotypes in longitudinal studies to predict cardiometabolic risk factors and diet-related diseases is nascent. However, replication of the metabotypes in other populations is a necessary next step. Notwithstanding this, the presented studies make a strong case for the metabotype approach and highlight its potential in identifying groups that could benefit from targeted dietary advice.

Metabolic phenotyping to investigate differential responses to dietary challenges and interventions

Differential responses to dietary interventions are becoming increasingly recognised. Concomitantly, metabolic phenotyping has emerged as a useful tool to examine responses to interventions. In the context of nutrition, health can be defined as the ability of an organism to adapt to challenges⁽²⁶⁾. Challenge tests investigate the disturbance and restoration of homeostasis of an individual using a dietary challenge as a physiological stressor⁽²⁷⁾. In combination with metabolomics, dietary challenges have been used to identify groups of subjects with distinct metabolic phenotypes/metabotypes and unique responses. Table 2 illustrates studies which focus specifically on differential responses of metabotypes to dietary challenges and intervention studies.

Table 2. Summary of studies investigating differential responses of metabotypes to meal challenges and dietary interventions

MMTT, mixed meal tolerance test; RCT, randomised controlled trial; HCA, hierarchical cluster analysis; GI, glycaemic index; PCA, principal component analysis; OGTT, oral glucose tolerance test; OLTT, oral lipid tolerance test; hsCRP, high-sensitivity C-reactive protein; HOMA-IR, homeostatic model assessment for insulin resistance; PC, phosphatidylcholines; BCAA, branched-chain amino acids; HSFAM, high-SFA meal; MMM, mixed Mediterranean-type meal; O-PLS, orthogonal partial least squares; TC, total cholesterol; LDL-C, LDL-cholesterol; HDL-C, HDL-cholesterol; CFU, colony-forming units; 25(OH)D, 25-hydroxyvitamin D.

Krishnan et al.⁽²⁸⁾ investigated the differential responses of metabotypes to dietary challenges. The authors used low- and high-glycaemic index meals in a cross-over randomised trial with healthy overweight women (n 24; aged 20–50 years) to identify response patterns that could provide insight into early subclinical glycaemic disruption. By using blood glucose, insulin and leptin responses to the challenges, individuals were clustered into three metabotypes. While the most populated metabotype presented little deviation from the expected response to the dietary challenges, the two minor metabotypes were suggestive, one of sub-clinical insulin resistance and the other of hyperleptinaemia. In the Metabolic Challenge (MECHE) Study, healthy subjects (n 214; aged 18–60 years) were randomised to one of three groups to receive oral glucose tolerance tests (OGTT) and/or oral lipid tolerance tests (OLTT) and four metabotypes were identified based on their blood glucose response curves to the OGTT (n 116)⁽²⁹⁾. The cluster with the most adverse metabolic profile at baseline presented a reduced β-cell function and differential responses to insulin and C-peptide during the OGTT and OLTT, as well as to glucose and TAG during the OLTT, which characterises this metabotype as at risk. The postprandial metabolic responses to different kinds of bread – refined rye bread, wholemeal rye bread and a control refined wheat bread – were investigated in a cross-over randomised controlled trial (RCT) with healthy postmenopausal women (n 19; aged 61 (sd 4·8) years)⁽³⁰⁾. The clustering of the fasting metabolic profile identified two distinct metabotypes. Women with higher fasting concentrations of leucine and isoleucine and lower fasting concentrations of sphingomyelins and phosphatidylcholines had higher insulin responses despite similar glucose concentrations after all kinds of bread, suggesting higher insulin resistance. In a recent study with data from the NutriTech project, the response to the intervention was only evident following the classification of the individuals into metabotypes⁽²⁶⁾. Healthy subjects (n 72; aged 59 to 64 years) were enrolled to a mixed meal tolerance test (MMTT) before and after 12 weeks targeting moderate weight loss (basal BMI 29·7 (sd 2·7) kg/m²). The intervention group (n 40) consumed a diet that reduced energy intake by 20 %, whereas subjects in the control group (n 32) consumed an average European diet matched to their energy expenditure to maintain body weight. Two metabotypes were reported based on the plasma concentration of metabolites (markers of lipolysis, fatty acid β-oxidation and ketogenesis) during the mixed meal challenge test. Before the intervention, individuals from metabotype B (n 36) showed slower glucose clearance, increased visceral fat volume, higher hepatic lipid concentrations, and a less healthy dietary pattern according to the urinary metabolomic profile when compared with individuals from metabotype A. Following the weight loss (about 5·6 kg), only the individuals from metabotype B showed positive changes in the glycaemic response to the MMTT. Since the metabolite differences found between metabotypes A and B are all closely associated with insulin signalling, metabotype B was considered to be prediabetic with a modestly impaired insulin action. Collectively, all these studies clearly demonstrate that the use of a metabotype approach in conjunction with meal challenges has the ability to characterise individuals into meaningful subgroups which could receive targeted nutrition advice to lower the individual disease risk⁽³⁰⁾.

In contrast to other studies that used the responses to challenges to form clusters, Lacroix et al.⁽³¹⁾ used only fasting metabolic data in a cross-over RCT designed to evaluate the metabolic and vascular effects of a high-SFA meal (HSFAM) and a mixed Mediterranean-type meal (MMM). Age, BMI, glycaemic and lipid parameters were used to cluster healthy men (n 28; 18–50 years) into two metabotypes at baseline. Compared with the healthiest group, the less healthy group showed significantly higher BMI, insulin and homeostatic model assessment for insulin resistance (HOMA-IR), in addition to a less favourable lipid profile and a lower intake of fruit and vegetables (dietary pattern score = 5·1 2 (sd 1·7) v. 3·9 (sd 1·4). Following the meal challenges, the less healthy group experienced a greater significant increase in TAG with MMM and endothelial dysfunction with HSFAM, in comparison with the healthier group. The MMM did not significantly alter postprandial endothelial function in both groups. The authors concluded that the less healthy group would benefit even more from consuming meals representative of a Mediterranean-type diet given its non-deleterious endothelial properties, indicating the potential of cluster techniques to individualise dietary advice.

Application of the metabotype approach has also encompassed dietary interventions that did not involve meal challenges. Wang et al.⁽³²⁾ in a controlled cross-over study with healthy subjects (n 23; aged 36–69 years) identified groups of individuals with differing plasma carotenoid responses to carotenoid-rich beverages. Following 3 weeks of daily intake of watermelon juice (20 mg lycopene, 2·5 mg β-carotene, n 23; 40 mg lycopene, 5 mg β-carotene, n 12) or tomato juice (18 mg lycopene, 0·6 mg β-carotene, n 10), cluster analysis applied to weekly carotenoid responses identified groups of individuals with differential responses. This, in turn, was used to classify individuals as strong responders or weak responders to the carotenoid intake. These findings demonstrate that subgroups of individuals can have differential responses to interventions which could be harnessed in the future to give more precise dietary advice. With respect to employing a metabotype approach for dietary interventions in clinical populations or disease risk factors, two studies are noteworthy. In a sample of high-risk cardiovascular subjects (n 57; aged ≥55 years) a 4-week cross-over RCT identified differential responsiveness to red wine polyphenols⁽³³⁾. At baseline, fasting blood and urinary metabolites and anthropometric parameters were used to cluster individuals in four metabotypes, including a higher-risk cluster and a healthier cluster. Following 28 d of dealcoholised red wine intake (polyphenol content = 733 equivalents of gallic acid/d), concentrations of urinary 4-hydroxyphenylacetate significantly increased in the healthier cluster compared with the higher-risk cluster, indicating a differential response in this cluster. In a double-blind 4-week RCT with healthy subjects (n 135; aged 18–63 years), the effect of vitamin D supplementation (15 mg vitamin D₃ per d) to improve markers of the MetS was only visible after the classification of the sample into metabotypes⁽³⁴⁾. The vitamin D supplementation significantly increased the serum 25-hydroxyvitamin D in comparison with the placebo group, but there was no effect of supplementation on the measured markers of the MetS. Based on thirteen fasting blood biomarkers, one cluster characterised by low concentrations of vitamin D and higher concentrations of adipokines showed a significant decrease in insulin, HOMA-IR scores and C-reactive protein and an inverse relationship between the change in serum vitamin D and glucose. Collectively, these examples clearly present how comprehensive phenotyping may identify subgroups of individuals that can benefit from specific dietary interventions.

The metabotype approach represents a tool through which we can start to understand individual responses to interventions. The ultimate goal will be to harness this information to deliver personalised nutrition.

Harnessing the metabotype approach to deliver targeted nutrition

To the best of our knowledge, there are only two published examples of a framework for the delivery of personalised nutrition using a metabotype approach (Table 3).

Table 3. Summary of studies developing targeted dietary advice solutions for metabotypes through the decision tree approach

TC, total cholesterol; HDL-C, HDL-cholesterol.

In 2015, O’Donovan et al.⁽²²⁾ proposed a framework based on metabotyping using four commonly measured fasting markers of metabolic health (TAG, TC, HDL-C and glucose). Application of the approach in 875 adults resulted in three metabotypes. Individuals in cluster 1 (n 274) had high TC concentrations, individuals in cluster 2 (n 423) had adequate concentrations of all four biomarkers, and individuals in cluster 3 (n 178) had the most unfavourable metabolic profile with high concentrations of TAG, TC and glucose and the lowest concentration of HDL-C. Targeted dietary advice was developed for each metabotype incorporating characteristics of the metabotype and personal traits. In order to test the reliability of the approach to deliver personalised dietary advice, the targeted approach was compared with an individual-based approach manually compiled and delivered by a dietitian for a random sample of participants (n 99). An excellent agreement of 89 % (range 20–100 %) was found between the methods, considering the dietary advice given with the targeted approach in relation to those given with the individual-based approach. The most important strength of this study is the fact that for clustering individuals only four biomarkers of metabolic health routinely measured were used. Furthermore, the approach generated a limited number of decision trees with simple and clear messages which allow the automation of the delivery of personalised dietary advice to individuals who are not high priority dietetic patients or where the access to a dietitian is limited. All these features make the proposed framework easily transferable to a clinical or primary care setting.

Development of this approach for a more diverse population was achieved in a proof-of-concept format with data from seven European countries⁽³⁵⁾. Twenty-seven fasting metabolic markers measured in finger-prick blood samples, including cholesterol, individual fatty acids and carotenoids, were clustered into three metabotypes. Individuals in cluster 1 (n 326) had the highest TC and circulating trans-fatty acids and the lowest omega-3 index, so this cluster was therefore considered the metabolically unhealthy cluster. Cluster 2 (n 433) was labelled the healthy group as individuals in this metabotype had the highest average omega-3 index and total carotenoid concentrations and the lowest total SFA. Individuals in cluster 3 (n 595) had the lowest average TC and highest levels of stearic fatty acid. Decision trees with targeted dietary advice were developed on the metabolic markers (TC, total SFA, omega-3 index and carotenoids), demographics and five key nutrients (salt, Fe, Ca, folate and fibre). The targeted approach was compared with the messages delivered by nutritionists as part of the Food4Me study (n 180) to participants receiving personalised dietary advice. An average match of 82 % at the level of delivery of the same dietary message was found and the agreement was also good by cluster, with an average match of 83 % for cluster 1, 74 % for cluster 2 and 88 % for cluster 3. These results, obtained in a European population from seven countries with diverse cultures and dietary intakes, confirm the metabotype approach as a robust approach to the delivery of targeted dietary advice and its applicability in different populations.

Conclusions and future directions

While metabotyping emerged initially to distinguish individuals with and without diet-related diseases, it has rapidly developed to identify those at metabolic risk and interrogate responses to dietary interventions. With a heightened interest in inter-individual variation in response to interventions, the approach presents an unbiased method of identifying differential responses. The ultimate goals will be to harness the approach for the delivery of personalised nutrition. However, further work is needed in understanding the biological mechanisms underlying the differential responses. We need detailed studies examining the underlying biology responsible for the different metabotypes and deciphering the role of genetics and the microbiome will be important future steps. Building this evidence base will be important for the further development of the metabotype concepts.

The framework comprising the metabotypes and decision trees represents a model for the delivery of personalised nutrition. However, there is a paucity of data demonstrating the impact of this approach on metabolic health parameters. Future studies are warranted to demonstrate that the approach is effective in changing behaviours and health outcomes.