Climate policies and pledges stress the need for drastic reductions in human-generated environmental impacts, including impacts that arise from food production and consumption(1). It is now essential to develop and implement strategies to transition to a more sustainable food system in order to meet climate targets. There are many examples of sustainable diets which primarily include high intakes of plant-based foods such as legumes, fruit, vegetables, nuts, seeds, and whole grains with low-to-moderate intakes of animal-sourced foods, including meat, fish, and dairy(2,Reference Willett, Rockstrom and Loken3) . Often, discourse on sustainable diets primarily focuses on reducing environmental impact – and more specifically – reducing diet-related greenhouse gas emissions(Reference Audsley, Brander and Chatterton4–Reference Payne, Scarborough and Cobiac8). While an essential component of a sustainable diet, this narrow focus may neglect other considerations for what makes a diet truly sustainable, such as ensuring diets are acceptable, accessible and promote good health(2). All of these considerations are linked and are all achievable through carefully considered sustainable diets. For example, observational data show that adherence to more sustainable diets is associated with better health outcomes(Reference Bui, Pham and Wang9,Reference Segovia-Siapco and Sabaté10) . However, no intervention data exist to demonstrate this effect in real-time(Reference Davies, Gibney and O’Sullivan11). To consider evidence linking sustainable diets with health, it’s pertinent to consider food-based dietary guidelines (FBDGs), which can perhaps be used as a proxy for sustainable dietary recommendations. FBDGs also include high intakes of plant-based foods and are designed to promote good health as well as balancing local considerations of accessibility and cultural acceptance, similar to the broad definition and goal of sustainable diets(12). While observational data have linked higher adherence to FBDGs with reduced environmental impacts, the fact remains that current diets are far from aligning with FBDGs(Reference Scheelbeek, Green and Papier13–Reference Leme, Hou and Fisberg15). Despite FBDGs being useful tools for public health nutrition guidance, one of the challenges of meeting FBDGs is the fact that population-level dietary advice can be difficult to adopt or interpret at an individual level. It is also now widely accepted that there is no one-size-fits-all for nutrition advice with many factors including genetic makeup, gut microbiome composition, metabolic profile, anthropometrics, habitual dietary intake, and more underpinning nutritional needs(Reference Brennan and de Roos16–Reference O’Donovan, Walsh and Nugent18). While FBDG aim to support good population health, the one-size-fits-all FBDGs approach cannot be sustainable if it is not effective for all. Therefore, it’s critical to understand how individual characteristics underpin nutrient-related outcomes and how to integrate this knowledge into future nutrition advice to support the transition to sustainable healthy diets.
Metabolic phenotype (metabotype) describes an individual’s metabolic state, is the shaped by genetics and environment and is often an indicator of health status and/or disease risk(Reference Holmes, Wilson and Nicholson17). An individual’s metabolic phenotype will impact how they respond to dietary change, leading to variation that is typically seen when evaluating dietary interventions(Reference Brennan and de Roos16–Reference Gibney19). Ultimately metabolic phenotyping can be used as a tool to identify responders and non-responders in dietary interventions which moves research away from the one-size fits all approach when examining health-related changes(Reference Gibney19). Furthermore, using metabolic phenotyping to tailor dietary advice, often referred to as personalised nutrition, has proved to be effective in eliciting dietary change(Reference Ordovas, Ferguson and Tai20). Therefore, it’s important to review the evidence of personalised nutrition and metabolic phenotypes to better understand how individuals or sub-groups respond to dietary interventions. Such information can be applied to future dietary interventions, including in the transition to more sustainable healthy diets. The aim of this review is to understand phenotypic variation that determines response to components of healthy and sustainable diets and the characteristics that are associated with positive physical and metabolic responses. While there is still much to be learned regarding metabolic phenotypes, previous findings from dietary interventions will be a key step in building better personalised interventions in the future.
Understanding inter-individual variation and metabolic phenotypes
Before considering sustainable healthy diets, it is important to consider existing evidence pertaining to inter-individual variation, and how it can be used in the provision of personalised nutrition. The capacity to alter health status, such as blood lipids or inflammatory markers, has been largely inconclusive with controlled dietary change(Reference Celis-Morales, Livingstone and Marsaux21–Reference Guasch-Ferré, Satija and Blondin23). Studying change in health status over time or metabolic response to an intervention has allowed researchers to conclude that some individuals or groups of individuals will respond better to a particular intervention than others. An early example of this research is a 2006 study from Stella and colleagues which used a crossover design and urine metabolomic profiles to compare metabolic response to vegetarian, low meat, and high meat diets(Reference Stella, Beckwith-Hall and Cloarec24). Metabolomics provides a snapshot of metabolite concentrations in a biological sample and can be used to identify both acute and chronic factors impacting an individual’s metabolic state(Reference German, Roberts and Watkins25). Furthermore, metabolomic profiles can provide insights to what is underpinning any change with respect to clinical outcomes like changes in body composition, blood lipids, or glucose metabolism(Reference Holmes, Wilson and Nicholson17,Reference Brennan and Hu26) . Stella and colleagues reported significant inter-individual variation in response to the diets between the 12 participants(Reference Stella, Beckwith-Hall and Cloarec24). Some individuals showed very little metabolic change while following the three different diets, while others showed greater metabolic responses(Reference Stella, Beckwith-Hall and Cloarec24). This work showed that a relatively homogenous group of individuals will respond differently to the same dietary intervention and highlighted the need to understand inter-individual variation to maximise the potential impact of dietary interventions(Reference Holmes, Wilson and Nicholson17).
Almost 20 years ago, this study described by Stella and colleagues, was one of the first to show inter-individual variation in response to a dietary intervention and one of the first applications of metabolomics in nutrition research(Reference Stella, Beckwith-Hall and Cloarec24). Since then, research has moved towards understanding traits that influence metabolic variation/response to interventions. For example, in one comprehensive study, using a well characterised cohort, researchers reported diet and microbiome having the strongest predictive power with respect to metabolomic derived metabolite concentrations, however it was also noted that several factors were highly correlated(Reference Bar, Korem and Weissbrod27). This multidimensional metabolic regulation demonstrates why it is difficult to see the same response for an entire sample population with the same dietary intervention. With a better understanding of certain factors contributing to variation in metabolic phenotype metrics, various research groups have developed frameworks to group individuals based on shared traits or response. In this literature, there appear to be two common approaches that can be referred to as metabolic phenotyping. One approach uses statistical methods to group individuals based on a set of traits and then explores between-group differences in response to an intervention(Reference O’Sullivan, Gibney and Connor28). The other common approach characterises response-phenotypes and so groups individuals based on shared response to an intervention(Reference Morris, O’Grada and Ryan29). The second approach allows researchers to examine between-group differences in traits to better understand the response-phenotype. Findings from these analyses are valuable for the future of nutrition research not only because of the insights into inter-individual but also because the findings can be applied to future interventions to improve health.
Using the first metabotyping approach described above, O’Sullivan and colleagues (2011), grouped a cohort of self-identified healthy individuals based on a set of metabolic biomarkers to explore if groups responded differently to vitamin D supplementation(Reference O’Sullivan, Gibney and Connor28). The analysis identified one cluster out of five that exhibited positive metabolic response to vitamin D supplementation(Reference O’Sullivan, Gibney and Connor28). As 25-hydroxy vitamin D concentrations increased in this cluster, there was a stepwise decrease in glucose concentrations, indicating that individuals with a particular metabolic phenotype at baseline were more likely to benefit from vitamin D supplementation. Similarly, Riedl and colleagues assessed long-term disease risk in the German KORA F4 and FF4 cohorts across metabolic phenotypes(Reference Riedl, Wawro and Gieger30). Three clusters were created based triacylglycerols concentrations (TAGs), total cholesterol (total-C), high density lipoprotein cholesterol (HDL-C), and glucose with clear differences across the clusters; for example, the healthiest cluster had the lowest TAGs and glucose concentrations and the highest HDL-C(Reference Riedl, Wawro and Gieger30). This cluster was mostly female, with the lowest BMI (25·7 kg/m2), more likely to be active and non-smokers, and had the highest intakes of vegetables, fruits, dairy and fibre(Reference Riedl, Wawro and Gieger30). Consequently, this cluster had the lowest incidence of CVD in a 7- and 14-year follow-up(Reference Riedl, Wawro and Gieger30). In a similar analysis which also included KORA F4 and FF4 cohorts, poor dietary index scores and ultra-processed food intake was associated with varying risk of developing type 2 diabetes mellitus (T2DM) across different metabotypes(Reference Deng, Wawro and Freuer31). Those in the most metabolically unhealthy metabotype had stronger associations between dietary intake and T2DM risk compared to those in healthier metabotypes(Reference Deng, Wawro and Freuer31). The findings demonstrate those in healthier metabotypes may have factors underpinning or buffering health outcomes from poor dietary intake while those in unhealthier metabotypes were more susceptible to disease risk from poor diet(Reference Deng, Wawro and Freuer31). While unfavourable metabotypes may be more prone to disease risk from poor dietary intakes, they may also be more likely to benefit from healthy diets(Reference Rundblad, Christensen and Hustad32). In an analysis from Rundblad and colleagues, higher vegetable intakes were associated with better glucose tolerance in unfavourable metabotypes but not favourable metabotypes(Reference Rundblad, Christensen and Hustad32). These findings suggest that dietary interventions have more influence in metabolically unhealthy individuals compared with metabolically healthy individuals. Ultimately, the metabotype approach has proved to be successful in identifying responders to an intervention which can help to build metabotype-based interventions for healthier lifestyles.
Using the second metabotyping approach described previously, Morris and colleagues grouped individuals based on results of an oral glucose tolerance test and identified four distinct response groups(Reference Morris, O’Grada and Ryan29). When response-phenotypes were compared, characteristics such as age, BMI, waist circumference, and VO2max, were significantly different between groups(Reference Morris, O’Grada and Ryan29). Similarly, McMorrow and colleagues characterised responder-phenotypes based on changes in insulin resistance among overweight/obese adolescents randomised to an anti-inflammatory supplement or placebo(Reference McMorrow, Connaughton and Magalhães33). Approximately 40 % of those in the active intervention group were classified responders(Reference McMorrow, Connaughton and Magalhães33). Further analysis of the response-phenotype revealed higher baseline concentrations of insulin, HOMA-IR, total-C, LDL-cholesterol (LDL-C) and lower quantitative insulin sensitivity check index (QUICKI) compared to non-responders(Reference McMorrow, Connaughton and Magalhães33). Yet it is not always possible to differentiate responders from non-responders using this approach. For example, in a secondary analysis from the Food4me proof-of-principle personalised nutrition study from Livingstone and colleagues, those with greater than 5 % change in outcomes like waist circumference, BMI, and total-C concentrations following a six-month dietary intervention were considered responders(Reference Livingstone, Celis-Morales and Navas-Carretero34). However, the authors were not able to identify any characteristics including demographics, genetics, or health markers that differentiated responders from the rest of the sample population(Reference Livingstone, Celis-Morales and Navas-Carretero34). On the other hand, Kirwan and colleagues assessed metabolic response differently in the same sample population and dietary intervention as Livingstone and colleagues(Reference Livingstone, Celis-Morales and Navas-Carretero34,Reference Kirwan, Walsh and Celis-Morales35) . Instead of using the 5 % change, Kirwan and colleagues grouped participants into quartiles of response based on change in a single parameter, total-C(Reference Livingstone, Celis-Morales and Navas-Carretero34,Reference Kirwan, Walsh and Celis-Morales35) . Those in the quartile with the largest decrease in total-C concentrations were deemed to be responders while those in the lowest quartile were grouped as non-responders(Reference Kirwan, Walsh and Celis-Morales35). Compared with non-responders, responders could be discriminated in a step-wise logistic regression model based on baseline characteristics such as age, total-C concentration, glucose concentration, fatty acid concentrations, and alcohol intake(Reference Kirwan, Walsh and Celis-Morales35). These two pieces of analysis demonstrate the complexity in determining responder characteristics, but also show how this analysis, when successful, can be used to design more targeted interventions or to select out groups of the population who are more likely to respond to a particular intervention. Given the difficulty in achieving uniform metabolic response, different interventions tailored to metabolic criteria may elicit the same health outcomes. These two approaches to metabolic phenotyping described here are useful tools to understanding phenotypic variation. Whether grouping individuals by phenotype to assess differences in response or grouping by response to define what contributes to response-phenotypes, these findings enable researchers to build personalised interventions to improve health.
Using personalised nutrition and metabotyping in nutrition research
Personalised nutrition describes the practice of tailoring dietary advice based on an individual’s traits and have been successful in supporting dietary change and improving health(Reference Ordovas, Ferguson and Tai20). Personalisation can be based on biological or lifestyle factors(Reference Ordovas, Ferguson and Tai20). Biological factors can be any phenotypic and genotypic traits that influence response such as those described as components of metabolic phenotypes(Reference Ordovas, Ferguson and Tai20). Lifestyle factors such as habitual diet, dietary preferences, or food-related skills, will influence the appropriateness and acceptability of dietary advice also(Reference Ordovas, Ferguson and Tai20). The Food4Me RCT was the first large scale proof-of-principle study examining the effectiveness of personalised nutrition advice in achieving greater and longer-lasting dietary change(Reference Celis-Morales, Livingstone and Marsaux21). Participants were randomised to receive either generic healthy eating advice or standardised personalised advice based on the following three levels of personalisation: (1) dietary intake; (2) dietary intake and phenotypic data; or (3) dietary intake, phenotypic data and genotypic data(Reference Forster, Walsh and O’Donovan36). Personalised advice led to greater dietary change and improved diet quality compared to generic advice, but there were no significant differences in outcomes between the three levels of personalisation(Reference Celis-Morales, Livingstone and Marsaux21). Furthermore, there were no significant differences in metabolic biomarkers over the course of the 6-month RCT(Reference Celis-Morales, Livingstone and Marsaux21). These findings were echoed in a similar study design by Hoevenaars and colleagues(Reference Hoevenaars, Berendsen and Pasman37). In their 18-week RCT, participants who received personalised advice based on lifestyle and phenotypic factors improved dietary intakes during the study with no changes in metabolic health biomarkers(Reference Hoevenaars, Berendsen and Pasman37). However, a more recent personalised nutrition intervention from Bermingham and colleagues resulted in some positive metabolic changes(Reference Bermingham, Linenberg and Polidori38). In the 18-week intervention, those who received personalised advice exhibited improvements in TAG concentrations, body weight and waist circumference(Reference Bermingham, Linenberg and Polidori38). Similarly, Aldubayan and colleagues reported improvements in insulin resistance and lipid profile following a 10-week personalised nutrition intervention in the PREVENTOMICS study; however, metabolic changes were also observed in the control group(Reference Aldubayan, Pigsborg and Gormsen39). It is worth noting the two studies reporting metabolic changes recruited more ‘at risk’ participants with higher waist circumference or BMI in the overweight/obese category and lower fruit and vegetable intakes, while studies reporting no change recruited generally healthy adults(Reference Celis-Morales, Livingstone and Marsaux21,Reference Hoevenaars, Berendsen and Pasman37–Reference Aldubayan, Pigsborg and Gormsen39) . Accounting for the fact that phenotypes will exhibit different metabolic response in an intervention is an important step to identifying changes over time.
As research has developed, studies have applied the knowledge of inter-individual variation and metabolic phenotypes at an early stage to generate more targeted personalised nutrition intervention interventions. For example, O’Donovan and colleagues used k-means cluster analysis to identify three clusters in the National Adult Nutrition Survey cohort with significantly different concentrations of TAGs, total-C, HDL-C, and glucose(Reference O’Donovan, Walsh and Nugent18). Metabotype characteristics were used to develop applicable dietary advice(Reference O’Donovan, Walsh and Nugent18). For example, feedback was designed for the most metabolically unhealthy cluster to lower total-C, TAGs and glucose(Reference O’Donovan, Walsh and Nugent18). Dietary feedback applicable to all clusters was created using decision trees, which was then expanded to include feedback related to anthropometrics and blood pressure(Reference O’Donovan, Walsh and Nugent18). To generate dietary advice, an individual would first need to be assigned a metabotype, then their anthropometry and blood pressure would determine their dietary feedback messages(Reference O’Donovan, Walsh and Nugent18). When dietary advice derived from the metabotype decision trees was compared to nutrition advice from a dietitian, 89 % of messages were aligned(Reference O’Donovan, Walsh and Nugent18). This was then further tested and compared to the individualised decision tree method used in the Food4Me study, and results showed 82 % of messages matched across the two methods(Reference O’Donovan, Walsh and Woolhead40). Following on from the work of O’Donovan and colleagues, the metabotype based personalised nutrition advice was further tested and refined by Hillesheim and colleagues prior to being used in a RCT(Reference Hillesheim, Ryan and Gibney41,Reference Hillesheim, Yin and Sundaramoorthy42) . The metabotype framework for personalised dietary advice resulted in increased diet quality and decreased total-C, LDL-C, and TAGs during the 12-week intervention(Reference Hillesheim, Yin and Sundaramoorthy42). The development and testing of personalised feedback based on metabotype provides evidence for the efficacy of personalising dietary advice to different metabolic phenotypes(Reference O’Donovan, Walsh and Nugent18,Reference O’Donovan, Walsh and Woolhead40–Reference Hillesheim, Yin and Sundaramoorthy42) . Past evidence demonstrates that many factors predispose individuals to response and the regulation of these factors will differ between individuals. Understanding metabolic phenotypes and inter-individual variation has been explored in previous studies which show metabolic phenotyping can be a practical approach to identifying disease risk, predicting response, and improving health and diet quality. Ultimately, there is a relationship between response and metabolic phenotype, meaning it will be essential to consider metabolic phenotypes in future personalised dietary advice to make diets truly sustainable.
Metabolic phenotyping and personalised nutrition strategies for sustainable healthy diets
Observational research shows positive associations between aspects of sustainable healthy diets such as high intakes of fruits, vegetables, and plant-based proteins and positive health outcomes(Reference Shan, Wang and Li43). However, based on what we know about inter-individual variation and metabolic phenotyping we cannot assume that these observational relationships will be replicated in dietary interventions or real-life scenarios. Therefore, it is critical that we measure changes in metabolic and health related metrics in response to sustainable dietary guidelines and examine variation in response so that we could potentially account for variation when adapting FBDGs to incorporate sustainability goals. This research group has described a framework for developing personalised nutrition advice for more sustainable healthy diets(Reference Davies, Gibney and Leonard44). While framework testing predicted improvements in diet quality, nutrient intakes and diet-related environmental impact, it is more difficult in this case to test impact on health outcomes, though the results of the RCT are to follow(Reference Davies, Gibney and Leonard44). In the meantime, we can take learnings from some other RCTs that have examined metabolic or health metrics in response to changes in fruit and vegetables or plant-based protein intakes, which as expected are quite mixed(Reference Elsahoryi, Neville and Patterson45–Reference Zhu, Fogelholm and Poppitt48). Building on from the previous section, it is likely that inconsistent results are unpinned by inter-individual variation even if not explicitly addressed in the literature to date. For the purpose of this review, we have selected a collection of studies that describe a very specific type of dietary intervention (polyphenols) and a specific phenotyping approach (urolithin metabotypes) in order to demonstrate potential for metabolic phenotyping in future personalised sustainable healthy diets(Reference González-Sarrías, García-Villalba and Romo-Vaquero49–Reference Meroño, Peron and Gargari52). In the first study, González-Sarrías and colleagues examined metabolic response to polyphenol intakes between urolithin metabotypes(Reference González-Sarrías, García-Villalba and Romo-Vaquero49). Urolithin metabotypes which relates to the type of urolithin that is produced by the gut when metabolising polyphenol compounds(Reference González-Sarrías, García-Villalba and Romo-Vaquero49). The authors identified one metabotype (urolithin metabotype B or UM-B) was more responsive to the polyphenol supplement based on greater urolithin production indicating they were better able to metabolise the polyphenols(Reference González-Sarrías, García-Villalba and Romo-Vaquero49). This in turn resulted in better metabolic responses, such as reductions in total-C, LDL-C and apo B, possibly related to better polyphenol metabolism(Reference González-Sarrías, García-Villalba and Romo-Vaquero49). In another example related to urolithin metabotypes, García-Mantrana and colleagues reported greater changes in gut microbiota composition for the UM-B metabotype(Reference García-Mantrana, Calatayud and Romo-Vaquero50). Further polyphenol and urolithin research conducted among older adults show similar results with improvements in gut microbiota regulation and short chain fatty acid metabolism in UM-B participants(Reference Meroño, Peron and Gargari52). Evidence on urolithin metabotypes shows promising results for understanding inter-individual variation and bringing about improved health outcomes, particularly in those UM-B individuals. However, it will be important to understand how to elicit metabolic response among non-UM-B individuals, and what characteristics or interventions will promote response. Furthermore, the examples described here focus on one particular component of foods and one phenotype. Interventions examining sustainable healthy diets will have multiple concurrent dietary changes which will require more complex analysis and potentially many more phenotypes. While there are many promising trajectories for metabolic phenotyping it is clear that there is a significant body of work to be done to better understand interindividual-variation and response, so that we can potentially incorporate a phenotyping approach when designing a personalised nutrition framework to deliver sustainable healthy diets.
Research to date specifically related to sustainable dietary patterns is quite limited, not yet reproduced, and typically with small samples sizes. The research presented here are some examples of work that has examined inter-individual variation and described the underlying factors that contribute to variation. It will be important to apply these findings to whole dietary change, such as dietary change needed for more sustainable healthy diets. For example, literature may develop to better understand whether food groups work together or in opposition to elicit metabolic response for certain metabotypes. It will be important to take this literature to a more applicable place in human nutrition and relate inter-individual variation to daily life, such as response to eating complex meals throughout the day rather than a specific food. Eventually this information, alongside other characteristics such as diet history or nutrient needs, will build to create personalised nutrition advice and well-balanced diets to support the health of individuals/sub-groups rather than using a population-based approach.
Conclusion
Analysing metabolic phenotypes is useful in understanding how to target specific interventions for subgroups to promote good health. Developing personalised feedback based on metabolic phenotype offers new insights for future research(Reference O’Donovan, Walsh and Nugent18,Reference O’Donovan, Walsh and Woolhead40–Reference Hillesheim, Yin and Sundaramoorthy42) . To date, there are many examples identifying metabolic phenotypes who respond to a particular intervention(Reference O’Sullivan, Gibney and Connor28,Reference Morris, O’Grada and Ryan29,Reference Rundblad, Christensen and Hustad32,Reference González-Sarrías, García-Villalba and Romo-Vaquero49,Reference García-Mantrana, Calatayud and Romo-Vaquero50,Reference Riedl, Hillesheim and Wawro53) ; however, it remains a relatively new science. Research will need to move beyond foods, nutrients or compounds, such as polyphenols, and examine metabolic phenotyping with respect to whole diets(Reference González-Sarrías, García-Villalba and Romo-Vaquero49–Reference Tosi, Favari and Bresciani51,Reference Koutsos, Riccadonna and Ulaszewska54,Reference Mena, Favari and Acharjee55) . The shift towards more sustainable healthy diets will mean changes to intakes of multiple food groups(Reference Davies, Gibney and O’Sullivan11). Increases and decreases in food groups and relevant nutritional compounds will mean broader dietary changes, demonstrating the need for more complex systems-based approaches for metabolic phenotyping. Large scale dietary interventions, incorporating new technologies and machine learning are needed to begin to understand these dynamics and build towards a future where metabolic phenotyping is an integral element for prescribing personalised nutrition advice for sustainable healthy diets.
Acknowledgements
The authors would like to thank the Nutrition Society for inviting the present manuscript as part of the 2024 postgraduate competition.
Authorship
All authors contributed to the design and structure of the present review. K.P.D. drafted the initial manuscript. All authors contributed to reviewing, editing, and writing the final manuscript.
Financial support
The SuHe Guide project is funded through the Department of Agriculture, Food, and the Marine (info@agriculture.gov.ie)/ Food Institutional Research Measure (grant number: 2019R546), and Department of Agriculture, Environment and Rural Affairs (daera.helpline@daera-ni.gov.uk) (grant number 19/R/546).
Competing interests
There are no conflicts of interest.
Open access
