NAT MED (IF:87): Microbiome, diet and genetic effects on inter-individual variation in the human plasma metabolome

Introduction The levels of thousands of metabolites in the human plasma metabolome are strongly influenced by individual genetics, diet and the composition of the gut microbiome. In this study, the proportion of inter-individual variation in the plasma metabolome explained by different factors was quantified by assessing 1183 plasma metabolites in 1368 broadly phenotyped individuals from the Dutch Lifelines DEEP and Genome cohorts, characterizing 610, 85 and 38 metabolites as mainly associated with diet, gut microbiome and genetics, respectively. In addition, dietary quality scores derived from metabolite levels were significantly correlated with dietary quality as assessed by a detailed food frequency questionnaire. Through Mendelian randomisation and mediation analyses, the present study revealed a putative causal relationship between diet, gut microbiome and metabolites. In conclusion, characterisation of factors explaining inter-individual differences in plasma metabolome could help in designing approaches to modulate diet or gut microbiome to form a healthy metabolome.

1 Non-targeted plasma metabolites in a Dutch cohort.

In the present study, we examined the plasma metabolome in 1679 fasting plasma samples from 1368 individuals from two LLD subcohorts (LLD1 and LLD2) and the GoNL6 cohort (Extended Data Figure 1 and Supplementary Table 1). the LLD1 cohort was the testing cohort and information on genetics, diet and gut microbiome was available for 1054 participants. In addition, 311 LLD1 subjects were followed up after 4 years (LLD1 follow-up). We also included two independent replication cohorts: 237 LLD2 participants (with genetic and dietary data) and 77 GoNL participants (genetic data only) (Extended Data Figure 1 and Supplementary Table 1). Untargeted metabolomic analysis using flow injection-time-of-flight mass spectrometry (FI-MS) yielded plasma levels of 1183 metabolites (Supplementary Table 2). These metabolites covered a wide range of lipids, organic acids, phenylpropanoids, benzenes and other metabolites (Extended Data Fig. 2a). As we observed weak correlations (absolute rSpearman < 0.2) between 1183 metabolites (Extended Data Fig. 2b), no data reduction was required and therefore follow-up analyses were performed for all metabolites. We verified the identification and quantification of certain metabolites (e.g., bile acids, creatinine, lactate, phenylalanine and isoleucine) by comparing their abundance levels in FI-MS with those previously determined by liquid chromatography-tandem mass spectrometry (LC-MS/MS) or NMR (rSpearman> 0.62; extended data Fig. 2c, d).

2 Metabolic groups reflect dietary quality scores.

In this study, 2854 significant associations (FDRSpearman < 0.05) were observed between 74 dietary factors and 726 metabolites (Fig. 3a and Supplementary Table 5). Theoretically, associations with food-specific metabolites could be used to validate the food questionnaire data. For example, the strongest association we observed was between quinic acid levels and coffee intake (rSpearman = 0.54; P = 1.6 × 10-80; Fig. 3b). Quinine is present in a wide range of different plants, but is particularly high in coffee. Another example is 2,6-dimethoxy-4-propylphenol, which is strongly correlated with fish intake (rSpearman=0.53; P=1.5×10-76; Fig. 3c). Associations between dietary factors and metabolic biomarkers of certain diseases were also tested, according to HMDB annotations, due to the particular presence of this compound in smoked fish. For example, 1-methylhistidine is a biomarker for cardiometabolic diseases (including heart failure) and we observed a significant association between 1-methylhistidine and meat (rSpearman=0.12; P=7.2×10-5), fish intake (rSpearman=0.11; P=3.1×10-4) There was a significant association between 1-methylhistidine levels, which were lower in vegetarians (rSpearman=-0.15; P=9.7×10-7, Figure 3d).

Given the relationship between diet, metabolism and human health, we wanted to know whether plasma metabolomics could predict diet quality. For each Lifelines participant, we constructed a lifeline diet score based on Food Frequency Questionnaire (FFQ) data, reflecting relative diet quality based on the diet-disease relationship. To construct a metabolic model to predict the quality of an individual’s diet, this study used LLD1 as the training set and LLD2 as the validation set. The resulting metabolic model included 76 metabolites, 51 of which were primarily diet-related. The diet scores predicted by the metabolites were significantly associated with the true diet scores assessed by FFQ in the validation set (r2adjusted = 0.27; PF-test = 3.5 × 10-5, Figure 3e). We also tested four other diet scores (alternative Mediterranean diet score, healthy eating index (HEI), protein score and modified Mediterranean diet score) and found that plasma metabolite predicted HEI was also significantly associated with FFQ-based HEI (r2adjusted = 0.23; PF-test = 6.5 × 10-5, Supplementary Table 12).

The bars represent dietary habits and the order of the bars is sorted by the number of significant associations. The colour of the direction of the association differs: orange indicates a positive association, whereas blue indicates a negative association. b. Association between plasma quinic acid levels and coffee intake. c. Association between plasma 2,6-dimethoxy-4-propylphenol levels and frequency of fish ingestion (n=1054 biologically independent samples). d. Differences in plasma 1-methylhistidine levels between vegetarians and non-vegetarians (n=1054 biologically independent samples). P-values from Wilcoxon test (two-sided) are shown. e. Association between plasma metabolomic predicted dietary quality scores (y-axis) and FFQ-assessed dietary quality scores (x-axis) (n=237 biologically independent samples). In b, c and e, each grey point indicates a sample, the dark grey dashed line indicates the linear regression line and the grey shading indicates the 95% CI. in b and c, the strength of association was assessed using Spearman correlation (two-sided; correlation coefficient and P-value) and in e, linear regression (F-test; two-sided; adjusted r2 value and P-value) was used to assess predictive performance.

3 Moderating role of the diet-microbiome in metabolite control.

Next, a mediation analysis was performed to investigate the association between diet, microbiome and metabolites. For 675 microbial traits associated with dietary habits and metabolites (FDR<0.05), we applied a two-way mediation analysis to assess the influence of the microbiome and metabolites on diet. The method established 146 mediated associations: 133 through the dietary influence of metabolites on the microbiome and 13 through the dietary influence of the microbiome on metabolites (FDRmediation<0.05, Pinverse-mediation>0.05; Fig. 6a, b and Supplementary Table 21). Most of these associations were related to the effects of coffee and alcohol on microbial metabolic function (Fig. 6a).

Coffee contains a variety of phenolic compounds that can be converted to hippuric acid by colonic flora. Hippuric acid is an acylglycine that has been associated with phenylketonuria, propionic acidemia and tyrosinemia. We observed that hippuric acid modulates the effect of coffee consumption on the abundance of Methanobrevibacter smithii (Pmediation=2.2×10-16; Figure 6c). It was also observed that hops acid, normally detected in alcoholic beverages, could modulate the effect of beer consumption on the Clostridium methylpentosum iron oxytocin:NAD+ oxidoreductase (Rnf) complex (Pmediation=2.2×10-16; Fig. 6d), an important membrane protein that drives the synthesis of ATP necessary for all bacterial metabolic activities.

An interesting example of dietary effects on metabolites via the microbiome (Fig. 6b and Supplementary Table 21) is the genus Clostridium tumefaciens vSV (300-305 kb), which encodes an ATPase responsible for the transport of various substrates across the membrane. This C. tumefaciens vSV mediated the effect of fruit intake on plasma urolithin B levels (Pmediation = 2.2 × 10-16; Fig. 6e). Urolithin B is a gut microbiota metabolite that protects against myocardial ischaemia/reperfusion injury via the p62/Keap1/Nrf2 signalling pathway. In conclusion, the data from this study provide a potential mechanistic basis for diet-metabolite and diet-microbiome relationships.

a. Parallel coordinate plots showing 133 mediating effects of plasma metabolites significant at FDR<0.05. Shown are dietary habits (left), plasma metabolites (centre) and microbial factors (right). b. Parallel coordinate plots show 13 mediating effects of the microbiome that were significant at FDR<0.05. Shown are dietary habits (left), microbial factors (centre) and plasma metabolites (right). c. Analysis of the effect of coffee intake on horse uric acid-mediated abundance of M. smithii. d. Analysis of the effect of beer intake on the hops acid-mediated C. methylpentosum Rnf complex pathway. e. Analysis of the effect of fruit intake on the effect of C. tumefaciens species (300-305 kb) in vSV-mediated plasma urolithin B. In c-e, grey lines indicate the association between the two factors and the corresponding Spearman coefficients and P values. Red arrows indicate direct mediators and blue arrows indicate reverse mediators.

Discussion.Fasting plasma profiles for 1183 metabolites were generated from 1679 samples from 1368 patients (311 patients with 4-year follow-up data) for whom extensive dietary records, genetic and gut microbiome data were collected in this study, together with systematic dietary, genetic and microbiome association analyses. The results of this study suggest that diet and the gut microbiome play a more important role than genetics in explaining inter-individual variation in metabolism, and that the more variation explained in metabolites, the more stable the metabolites become over time.

Dietary composition is a fundamental resource for the plasma metabolome and a recent study has shown that individual dietary habits predict levels of specific metabolites in plasma, emphasising that the plasma metabolome reflects individual dietary habits. However, whether it is possible to assess an individual’s dietary quality score based on the plasma metabolome remains to be determined. Using a machine learning-based prediction model, dietary quality estimated by an individual’s plasma metabolome was found to be significantly correlated with dietary quality estimated by the FFQ, suggesting that the plasma metabolome reflects dietary quality to some extent.

Dietary components act as substrates for the metabolic pathways of gut microbes, leading to the formation of a range of metabolites that can be absorbed from the gut into the host circulation. Although earlier studies have linked the taxonomic abundance of gut microbes to plasma metabolites, these studies did not capture the specific microbial enzymes responsible for metabolite production, and although this information is needed to link the relevant associations to potential molecular mechanisms, the present study identifies putative metabolic functions of previously unannotated microbial gene sequences. In addition, through bidirectional mediator analysis, we identified hundreds of mediating links that provide insight into diet-microbiome interactions in human metabolic health, as shown for several metabolites previously associated with cardiometabolism and renal disease (e.g. phenol and piperidine acid). Notably, these mediating links primarily suggest that the effects of dietary composition on the microbiome can be mediated through metabolites, highlighting the remarkable selective power of dietary habits in shaping the gut microbiome. However, as these results are largely based on observational data, these associations should be interpreted with caution and future interventions and experimental studies (focusing on specific dietary and microbiome competencies) will be essential to confirm causal relationships.

In addition to diet and gut microbiome, human genetics also serves as a potential determinant of the plasma metabolome. Using this metabolomic dataset, this study not only replicated previously reported mQTLs, but also identified three mQTLs involving three loci previously unknown to be associated with any metabolite. The mQTLs characterised in this study may be associated with cardiometabolism and chronic kidney disease. This study also used genetic variation as a tool for MR to infer a causal relationship between the gut microbiome and metabolites. This analysis suggests that the microbiome may contribute to elevated levels of toxins (bisulfite and 5-hydroxytryptophol) associated with chronic kidney disease and cardiometabolic disease. Thus, the causal relationship between the microbiome and metabolites that has been established in this study reveals the potential metabolic function of gut microbes on human health.

Taken together, the dietary, genetic and microbial associations with plasma metabolites and the causal and mediating links reported in this study provide a comprehensive resource that can guide subsequent studies aimed at designing preventive and therapeutic strategies for human metabolic health.