Phenotypic variation influencing response to nutrition intervention studies

To date, several studies have focused on inter-individual variability to standardised (or semi-standardised) meals such as oral lipid tolerance tests (OLTT) and/or glucose tolerance tests (OGTT), with some examples given in Table 1. Examining glycaemic response first, studies such as Vega-Lopez et al.^{(Reference Vega-Lopez, Ausman and Griffith6)}, demonstrated that the inter-individual variation in response to a glycaemic load is greater than the intra-individual variation, but didn't elucidate further on the factors influencing this variation. Following on from their initial examination of variance in repeated OLTT and OGTT in a healthy population. Morris et al.^{(Reference Morris, O'Grada and Ryan8)} examined factors influencing variation in response to the OGTT across the study cohort, using a statistical clustering of baseline characteristics. Using this method, researchers identified a distinct phenotype or ‘metabotype’ group, which had a significantly different response to OGTT, compared to all other clusters. This group of individuals had the highest BMI, highest circulating TAG, C-reactive protein, c-peptide and insulin levels, as well as the highest insulin resistance (HOMA-IR) score, compared to other clusters^{(Reference Morris, O'Grada and Ryan8)}. van Dijk et al.^{(Reference van Dijk, Venema and vanMechelen9)} noted that exercise levels, preceding the OGTT measurements also had a clear effect on the post-prandial glycaemic response, with subjects glycated haemoglobin levels related to the magnitude of response to exercise. Examining reported variance in response to OLTT, results from Morris et al.^{(Reference Morris, O'Grada and Ryan12)} and Ryan et al.^{(Reference Ryan, O'Grada and Morris7)}, both from the same group and focusing on the MECHE (metabolic challenge) study, noted baseline characteristics similar to those influencing response to OGTT including age, TAG, circulating fatty acids, as well as SNP^{(Reference Ryan, O'Grada and Morris7,Reference Morris, O'Grada and Ryan12)} (Table 1).

Two things stand out from these studies: firstly that there is a need for standardisation of parameters in the measurement of these standardised test meals, in order that results from differing studies are not influenced by procedural differences and can thus be both combined and interpreted correctly. In addition, baseline subject characteristics clearly influence the response, such as age, BMI, circulating TAG, C-reactive protein or insulin levels, for example. This information should be used in two ways: (1) to direct selection of study populations to ensure variance within intervention groups is minimised, and/or controlled for in statistical analysis and (2) to direct personalised or targeted nutrition messages to an identified group, which differ due to differing response to the standardised test meal.

Variation in response to non-standardised meals has also been reported in the literature (Table 1). Childs et al.^{(Reference Childs, Kew and Finnegan11)}, noted a significant difference in sex in response to a 6-month intervention replacing standard margarines and spreads with products enriched with α-linoleic acid, with a greater increase in the EPA content of plasma phospholipids in females compared to males after 6 months. Sex differences such as this are not unique and have been previously reported in other studies^{(Reference Fatima, Connaughton and Weiser26,Reference Garg, Brennan and Price27)} . McMorrow et al.^{(Reference McMorrow, Connaughton and Magalhaes15)}, noted a significant variation in response to consumption of an anti-inflammatory nutritional supplement. The authors of the present paper noted that the supplement modulated adiponectin levels, but not insulin resistance. However, they did note that insulin resistance improved in a sub-cohort of adolescents and concluded that the baseline phenotype of responders was insulin resistant and dyslipidemic with higher insulin, HOMA-IR, HOMA-β, total cholesterol, LDL cholesterol and lower QUICKI (quantitative insulin sensitivity check index), than non-responders. It is interesting to note that the authors of the present paper reported that sex, age, BMI and body composition, were not different between responders and non-responders, which one might assume may be due to the homogeneous selection of their population, whereby only overweight and obese teens between 13–18 years were recruited. Feliciano et al.^{(Reference Feliciano, Mills and Istas28)} also explicitly reported variable response in a study examining varying doses of cranberry polyphenols. The authors examined inter-individual variation by calculating the CV for C _max and area under the curve for each individual plasma metabolite, and demonstrated that CV for C _max was 51 %, and the CV of the area under the curve for the total (sum of all 60) metabolites was 53 %. This varied between metabolites with a CV of 43 % for dihydroferulic acid 4-O-sulfate and 216 % for vanillic acid. The authors note that inter-individual variability of the plasma metabolite concentration was broad and dependent on the metabolite, as had been noted previously^{(Reference Manach, Williamson and Morand29)}. There are a number of factors that are suggested to influence the metabolism and absorption of such metabolites, including sex, genetic polymorphisms of transporters or metabolising enzymes, environmental influences and likely the composition of the gut microbiome^{(Reference Wruss, Lanzerstorfer and Huemer30,Reference Selma, Espin and Tomas-Barberan31)} .

Again, such studies highlight that variation in response to any nutrition intervention will vary, and should be considered in both the interpretation, presentation and reporting of any study results. Recent work carried out by the COST Action POSITIVe, specifically investigated inter-individual variation in response to consumption of plant-based bioactive^{(Reference Manach, Milenkovic and VandeWiele3)}. Whilst several meta-analyses to determine factors influencing variation were conducted, lack of consideration or reporting of such variance in publications meant analysis and conclusions were limited^{(Reference García-Conesa, Chambers and Combet32–Reference González-Sarrías, Combet and Pinto34)}. If factors influencing response are to be more fully investigated, then researchers will need to provide information and clearly address variance in response in their analyses^{(Reference Manach, Milenkovic and VandeWiele3)}.

Genotypic variation influencing response to nutrition intervention studies

Alongside specific phenotypic characteristics influencing response to intervention studies, much work has been conducted which examines and reports the differing response of specific genotypes to various nutrition interventions. Too many to mention all within the present paper, some examples are given in Table 2, and will be discussed below to give a flavour of how genetic variation can influence response. Several studies have examined the effect of various genetic variants on weight loss with participants on various diets (Table 1). Gardner et al.^{(Reference Gardner, Trepanowski and DelGobbo24)} in the DIETFITS study, examined the interaction of a multi-locus genotype pattern and the impact of a low-fat or low-carbohydrate diet on weight loss and found there was no significant interaction or diet-insulin secretion interaction with 12-month weight loss. In contrast, Aller et al.^{(Reference Aller, Izaol and Primo16)} examined the impact of the adiponectin (ADIPOQ) gene (rs1501299 G-T), on weight loss in a two-arm randomised trial with two hypoenergetic diets (high-protein and low-carbohydrate diet v. standard diet) over 9 months. In this study, the GG genotype group, regardless of the diet, the decrease in total cholesterol levels, LDL cholesterol, TAG levels, insulin levels and HOMA-IR levels were higher than T-allele carriers^{(Reference Aller, Izaol and Primo16)}. Likewise, Celis-Morales et al.^{(Reference Celis-Morales, Marsaux and Livingstone25)}, in a sub-analysis of the Food4me study, where participants were randomised to one of four arms: level 0, control group; level 1, dietary group; level 2, phenotype group; and level 3, genetic group, level 3 participants were stratified into risk carriers (AA/AT) and non-risk carriers (TT) of the alpha-ketoglutarate-dependent dioxygenase (FTO) gene (rs9939609). This analysis demonstrated that changes in adiposity markers were greater in participants who were informed that they carried the FTO risk allele (level 3 AT/AA carriers) than in the non-personalised group (level 0) but not in the other personalised groups (levels 1 and 2)^{(Reference Celis-Morales, Marsaux and Livingstone25)}. However, this was not seen with respect to dietary changes in other genetic variants including ApoE^{(Reference Fallaize, Celis-Morales and Macready35)} and methylene tetrahydrofolate reductase (MTHFR)^{(Reference O'Donovan, Walsh and Forster36)}.

Much work has also focused on the impact of genetic variation and response to lipid consumption, examining the impact of genetic variation in the ApoE gene on the metabolic response to consumption of differing fat types (Table 1). For example, Shatwan et al.^{(Reference Shatwan, Weech and Jackson18)} examined the impact of diets high in SFA, MUFA or n-6 PUFA over 16 weeks. Stratifying for ApoE SNP (rs1064725), they reported that only TT homozygotes showed a significant reduction in total cholesterol after the MUFA diet compared to the SFA or n-6 PUFA diets^{(Reference Shatwan, Weech and Jackson18)}. However, one must remember that whilst a study may be examining impact of genotype, that this may not be the only factor influencing response, for example, Caslake et al.^{(Reference Caslake, Moore and Gordon37)}, in a double-blind placebo-controlled crossover study, where the consumption of different amounts of EPA/DHA was examined, found significant genotype interactions in response to the intervention, whereby the greatest TAG-lowering responses (reductions of 15 % and 23 % after 0·7g and 1·8g EPA DHA/d, respectively) were evident in ApoE4 men. Similarly, Chouinard-Watkins et al.^{(Reference Chouinard-Watkins, Conway and Minihane23)}, in a study examining changes in circulating lipid profile following 8 weeks consumption of a high-saturated fat diet with the addition of DHA and EPA, found that ApoE4 carriers were plasma responders to the DHA supplement than were non-carriers but only in the high-BMI group. Again suggesting a genotype–phenotype interaction in response to the intervention.

Variations in folate metabolism have also been well researched with respect to response to consumption of B-vitamins and other nutrients, with much of the work focusing on variations in the enzyme MTHFR^{(Reference McNulty, Strain and Hughes38–Reference Wilson, McNulty and Scott40)}. One of the most interesting papers in this area by Wilson et al.^{(Reference Wilson, Ward and McNulty22)}, examined the response of thirty-one MTHFR TT genotype patients with the risk of CVD. This study, was a 4-year follow on from a study where eighty-three participants representing all three MTHFR 677CT genotypes, were initially recruited to participate in a placebo-controlled riboflavin intervention for 16 weeks. In the initial study, the team found the TT group to be responsive with respect to reduction in blood pressure. To confirm these findings, the follow-up study, which only examined those with the TT genotype proceeded to confirm the effect of consumption of riboflavin (1·6 mg/d for 16 weeks) or placebo on blood pressure, conducted in a crossover style whereby the 2004 treatment groups were placed in opposite intervention groups. This study confirmed riboflavin supplementation produced an overall decrease in systolic and diastolic blood pressure in this genotype group^{(Reference Wilson, Ward and McNulty22)}.

Inter-individual variation and personalised nutrition

With many examples of an inter-individual response to consumption of foods/nutrients published to date, the challenge is now to potentially use this information in an informed and appropriate manner to tailor nutritional recommendations for individuals or groups of individuals, the cornerstone of personalised nutrition.

Firstly, how does inter-individual variation fit into the concept of personalised nutrition. There are many published definitions of personalised nutrition, which vary in their manner and/or depth of personalisation. More recently, definitions recognise that this must be broader, with definitions basing personalised nutritional advice on multiple characteristics, such as that used in Food4me (Grimaldi et al. 2018) which encompasses levels of information layering dietary, phenotypic and genotypic information from an individual^{(Reference Grimaldi, van Ommen and Ordovas41)}. Ordovas et al.^{(Reference Ordovas, Ferguson and Tai42)}, simply described personalised nutrition as ‘an approach that uses information on individual characteristics to develop targeted nutritional advice’, not defining the depth or nature of information required. Finally, more recently Stewart-Knox et al.^{(Reference Stewart-Knox, Gibney and Abrahams43)}, have built on these to contextualise the information by including factors influencing food choice determinants, and considering the framework in which the information would be offered. Examining and understanding variation in response to nutrition interventions is important to further the field of personalised nutrition. Using the examples discussed within the present paper, one might suggest that stratifying based on age, sex, baseline biomarkers and exercise levels with respect to glucose metabolism would be recommended (Table 1). Similarly, fitness level, baseline metabolic markers and identified SNP could also be considered when giving advice on lipid consumption and metabolism. However, there needs to be some caution. Whilst variation was observed in many of these studies, before recommendations could or should be used, confirmation in other studies and cohorts, and a full understanding of the mechanism of the variation needs to be elucidated. Furthermore, the impact of the recommendation based on the variation needs to be determined. Both of these issues were addressed in the recent Food4me project. Firstly, Grimaldi et al.^{(Reference Grimaldi, van Ommen and Ordovas41)} in their paper proposing guidelines to evaluate scientific validity and evidence for genotype-based dietary advice focus on a framework that considers study design, type of gene–nutrient interaction, biological plausibility and the scientific validity of the published evidence. Joining the reported presence of variability with some levels of biological explanation for this, ensures a scientific rigour in the development of tailored recommendations, ensuring that they have sound scientific rationale. Secondly, the findings of the Food4me proof of principle study are also of interest when discussing the use of known variation within personalised recommendations. Examining the impact of personalisation on change in dietary intake, researchers within Food4me undertook a large multicentre study, across Europe, which examined the impact of levels of personalised advice, on change in dietary intake. Following recruitment, interested and eligible participants were randomised to control (general healthy eating guidelines), level 1 (nutritional advice based on diet alone), level 2 (nutritional advice based on diet and phenotype) or level 3 (nutritional advice based on diet, phenotype and genotype). Full details of the study are published elsewhere^{(Reference Celis-Morales, Livingstone and Marsaux44)}. A change in dietary intake and other parameters were examined at baseline (0), 3 and 6 months. Following completion comparisons were made between control and personalisation (levels 1, 2 and 3 together) and then across the levels of personalisation. The researchers found that there was a greater positive change in dietary intake in the personalised groups compared to the control group, but that there was no difference between the levels of personalisation, suggesting that participants are responsive to personalisation but the manner in which the advice was personalised didn't have an effect^{(Reference Celis-Morales, Livingstone and Marsaux45)}. Further examination of response to knowledge of specific genetics variants had similar results^{(Reference Fallaize, Celis-Morales and Macready35,Reference O'Donovan, Walsh and Forster36)} . For example, O'Donovan et al.^{(Reference O'Donovan, Walsh and Forster36)}, examined the impact of an individual's knowledge of their MTHFR 677 genotype, and found that the TT group (risk group), who were given specific advice on to increase consumption of folate (foods) or folic acid (supplement), there was no difference in the change of folate in the diet in this group compared to the non-risk group. Thus suggesting that even knowledge of their risk and a recommendation to increase the consumption of the specific nutrient, did not result in a greater behavioural change^{(Reference O'Donovan, Walsh and Forster36)}. Whilst this pattern has also been found in previous studies, other studies have demonstrated a change with knowledge of risk; however, overall results are mixed^{(Reference O'Donovan, Walsh and Gibney46)}.

This brings about a variation that also needs to be considered. Variation in response to personalised recommendations, not at a physiological level, but at a behavioural level. Consumer studies within the Food4me project explored associations between food choice motives, attitudes towards and intention to adopt personalised nutrition and found that food choice motives such ‘weight control’, ‘mood’, ‘health’ and ‘ethical concern’ had a positive association and ‘price’ had a negative association with attitude towards, and intention to adopt, personalised nutrition^{(Reference Rankin, Bunting and Poinhos47)}. This suggests that underlying health perceptions, food beliefs and other psychological factors will influence variability in response to personalised advice, (which was given to address variability in an individuals’ requirement), thus increasing an additional level of inter-individual variability when considering the response in large population cohorts.